Lecture Notes in Physics

Editorial Board
R. Beig, Wien, Austria
J. Ehlers, Potsdam, Germany
U. Frisch, Nice, France
K. Hepp, Zürich, Switzerland
W. Hillebrandt, Garching, Germany
D. Imboden, Zürich, Switzerland
R. L. Jaffe, Cambridge, MA, USA
R. Kippenhahn, Göttingen, Germany
R. Lipowsky, Golm, Germany
H. v. Löhneysen, Karlsruhe, Germany
I. Ojima, Kyoto, Japan
H. A. Weidenmüller, Heidelberg, Germany
J. Wess, München, Germany
J. Zittartz, Köln, Germany

3
Berlin
Heidelberg
New York
Barcelona
Hong Kong
London
Milan
Paris
Singapore
Tokyo
Editorial Policy
The series Lecture Notes in Physics (LNP), founded in 1969, reports new developments in
physics research and teaching -- quickly, informally but with a high quality. Manuscripts
to be considered for publication are topical volumes consisting of a limited number of
contributions, carefully edited and closely related to each other. Each contribution should
contain at least partly original and previously unpublished material, be written in a clear,
pedagogical style and aimed at a broader readership, especially graduate students and
nonspecialist researchers wishing to familiarize themselves with the topic concerned. For
this reason, traditional proceedings cannot be considered for this series though volumes
to appear in this series are often based on material presented at conferences, workshops
and schools (in exceptional cases the original papers and/or those not included in the
printed book may be added on an accompanying CD ROM, together with the abstracts
of posters and other material suitable for publication, e.g. large tables, colour pictures,
program codes, etc.).

Acceptance
A project can only be accepted tentatively for publication, by both the editorial board and the
publisher, following thorough examination of the material submitted. The book proposal
sent to the publisher should consist at least of a preliminary table of contents outlining the
structure of the book together with abstracts of all contributions to be included.
Final acceptance is issued by the series editor in charge, in consultation with the publisher,
only after receiving the complete manuscript. Final acceptance, possibly requiring minor
corrections, usually follows the tentative acceptance unless the final manuscript differs
significantly from expectations (project outline). In particular, the series editors are entitled
to reject individual contributions if they do not meet the high quality standards of this
series. The final manuscript must be camera-ready, and should include both an informative
introduction and a sufficiently detailed subject index.

Contractual Aspects
Publication in LNP is free of charge. There is no formal contract, no royalties are paid,
and no bulk orders are required, although special discounts are offered in this case. The
volume editors receive jointly 30 free copies for their personal use and are entitled, as are the
contributing authors, to purchase Springer books at a reduced rate. The publisher secures
the copyright for each volume. As a rule, no reprints of individual contributions can be
supplied.

Manuscript Submission
The manuscript in its final and approved version must be submitted in camera-ready form.
The corresponding electronic source files are also required for the production process, in
particular the online version. Technical assistance in compiling the final manuscript can be
provided by the publisher’s production editor(s), especially with regard to the publisher’s
own Latex macro package which has been specially designed for this series.

Online Version/ LNP Homepage

LNP homepage (list of available titles, aims and scope, editorial contacts etc.):
[Link]
LNP online (abstracts, full-texts, subscriptions etc.):
[Link]
C. F. Barenghi R. J. Donnelly W. F. Vinen (Eds.)

Quantized
Vortex Dynamics
and
Superfluid Turbulence

13
Editors
C.F. Barenghi
University of Newcastle
Mathematics Department
Newcastle NE1 7RU, United Kingdom

R.J. Donnelly
University of Oregon
Physics Department
Eugene, OR 97403, USA

W.F. Vinen
University of Birmingham
Physics Department
Birmingham B15 2TT, United Kingdom

Cover picture: Tangle of quantized vortex filaments computed in a periodic box

by D. Kivotides, D. Samuels and C.F. Barenghi.
Library of Congress Cataloging-in-Publication Data applied for.

Die Deutsche Bibliothek - CIP-Einheitsaufnahme

Quantized vortex dynamics and superfluid turbulence / C. F. Barenghi ...

(ed.). - Berlin ; Heidelberg ; New York ; Barcelona ; Hong Kong ; London ;
Milan ; Paris ; Singapore ; Tokyo : Springer, 2001
(Lecture notes in physics ; 571)
(Physics and astronomy online library)
ISBN 3-540-42226-9
ISSN 0075-8450
ISBN 3-540-42226-9 Springer-Verlag Berlin Heidelberg New York

This work is subject to copyright. All rights are reserved, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustra-
tions, recitation, broadcasting, reproduction on microfilm or in any other way, and
storage in data banks. Duplication of this publication or parts thereof is permitted only
under the provisions of the German Copyright Law of September 9, 1965, in its current
version, and permission for use must always be obtained from Springer-Verlag. Violations
are liable for prosecution under the German Copyright Law. Springer-Verlag Berlin Hei-
delberg New York
a member of BertelsmannSpringer Science+Business Media GmbH [Link]
c Springer-Verlag Berlin Heidelberg 2001
Printed in Germany The use of general descriptive names, registered names, trademarks,
etc. in this publication does not imply, even in the absence of a specific statement, that such
names are exempt from the relevant protective laws and regulations and therefore free for
general use.
Typesetting: Data conversion by Steingraeber GmbH, Heidelberg
Cover design: design & production, Heidelberg
Printed on acid-free paper
SPIN: 10792065 55/3141/du - 5 4 3 2 1 0
Preface

This book springs from the programme Quantized Vortex Dynamics and Super-
fluid Turbulence held at the Isaac Newton Institute for Mathematical Sciences
(University of Cambridge) in August 2000. What motivated the programme was
the recognition that two recent developments have moved the study of quan-
tized vorticity, traditionally carried out within the low-temperature physics and
condensed-matter physics communities, into a new era.
The first development is the increasing contact with classical fluid dynamics
and its ideas and methods. For example, some current experiments with he-
lium II now deal with very classical issues, such as the measurement of velocity
spectra and turbulence decay rates. The evidence from these experiments and
many others is that superfluid turbulence and classical turbulence share many
features. The challenge is now to explain these similarities and explore the time
scales and length scales over which they hold true. The observed classical aspects
have also attracted attention to the role played by the flow of the normal fluid,
which was somewhat neglected in the past because of the lack of direct flow
visualization. Increased computing power is also making it possible to study the
coupled motion of superfluid vortices and normal fluids. Another contact with
classical physics arises through the interest in the study of superfluid vortex re-
connections. Reconnections have been studied for some time in the contexts of
classical fluid dynamics and magneto-hydrodynamics (MHD), and it is useful to
learn from the experience acquired in other fields.
The second development arises from atomic physics and is the discovery of
Bose–Einstein condensation in confined clouds of alkali atoms. The study of
superfluidity and quantized vorticity is now possible in a wide range of other
systems besides helium II. The rapid progress in this area has given momentum
to the use of the Gross–Pitaevskii Equation or Nonlinear Schroedinger Equation
(NLSE). Researchers have become more aware of the approximations and limi-
tations involved in the NLSE model, but also of its range of validity and great
power of prediction. The use of the NLSE has become more established, and
the NLSE is proving to be a powerful tool for modeling problems such as vortex
nucleation, reconnections and even turbulence.
A further development arises from the results of preliminary theory and ex-
periments in turbulent Helium 3 which suggest that there are significant differ-
ences with turbulence in Helium 4 and these are likely to be explored in the
future.
VI

It is apparent from this background that the contributions to this book come
from investigators with a wide range of backgrounds and expertise: condensed-
matter physics and low-temperature physics, classical fluid dynamics and applied
mathematics, MHD, atomic physics, and engineering (for the applications of
helium II as a cryogenic coolant).
The book is divided into topical chapters. Each chapter begins with one or
two introductory review articles, which are suitable for students and new inves-
tigators interested in entering the field. The introductory articles are followed
by shorter, more specialized papers.
Chapter 1 introduces us to the problem of quantized vorticity and super-
fluid turbulence, and it summarizes the key aspects and problems which are
currently studied. Chapter 2 is devoted to turbulence experiments. Chapter 3
considers the fundamental problem of friction and vortex dynamics. The theory
of superfluid turbulence and the interpretation of the experimental results is the
subject of Chap. 4. Chapter 5 is devoted to the application of the NLSE model
to superfluidity and vortices. Chapter 6 moves away from helium and considers
Bose–Einstein Condensation and vortices in the context of alkali atoms. Chap-
ter 7 is concerned with some aspects of classical turbulence and MHD which
are relevant in the study of superfluid turbulence. Finally, Chap. 8 deals with
Helium 3 and other systems.
We are grateful for the support and encouragement of Professor Keith Mof-
fatt, Director of the Newton Institute, and we would like to thank Tracey Andrew
who helped in the preparation of the manuscripts for publication.

Newcastle, Eugene and Birmingham, Carlo Barenghi

June 2001 Russ Donnelly
Joe Vinen
Contents

Part I Introduction

Introduction to Superfluid Vortices and Turbulence

C.F. Barenghi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1 The Two-Fluid Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Quantized Vortex Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Modelling the Vortex Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1 Microscopic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Mesoscopic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.3 Macroscopic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.1 Turbulent Counterflows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2 Turbulent Coflows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5 Motion of Superfluid Vortices for a Given Normal Fluid . . . . . . . . . . . . . 10
6 Motion of the Normal Fluid at Given Superfluid Vortices . . . . . . . . . . . . 11
7 Fully Coupled Motion of Superfluid Vortices and Normal Fluid . . . . . . . 12
8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Part II Turbulence Experiments

An Introduction to Experiments on Superfluid Turbulence

R.J. Donnelly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 Update on Pipe Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Update on Towed Grid Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1 The Nature of Grid Turbulence in Helium II . . . . . . . . . . . . . . . . . . 20
3.2 Four Regimes of Decaying Grid Turbulence in Helium II . . . . . . . . 23
4 Agenda for the Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1 The University of Oregon 6 cm Wind Tunnel . . . . . . . . . . . . . . . . . . 24
4.2 Wind Tunnels for Model Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Tow Tanks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5 Challenges for the Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.1 The Challenge of Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2 Challenges for Understanding Counterflow Turbulence . . . . . . . . . . 29
VIII Contents

5.3
Challenges for Understanding Periodic
Boundary Layer Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.4 Instrumentation to Detect Vortices Below 1 K . . . . . . . . . . . . . . . . . 33
5.5 The Normal Fluid and the Vortex Tangle . . . . . . . . . . . . . . . . . . . . . 33
5.6 Flow over Blunt Objects, Testing Models such as Submarines . . . 34
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

The Experimental Evidence for Vortex Nucleation in 4He

É. Varoquaux, O. Avenel, Y. Mukharsky, P. Hakonen . . . . . . . . . . . . . . . . . . . 36
1 Single Vortex Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2 Multiple Slips and Collapses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Applications of Superfluid Helium
in Large-Scale Superconducting Systems
S.W. Van Sciver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2 Superconducting Systems That Use He II Cooling . . . . . . . . . . . . . . . . . . 53
2.1 Accelerator Magnet System for LHC . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2 High Field Solenoid for the NHMFL 45-T Hybrid . . . . . . . . . . . . . . 54
2.3 RF Cavity Systems for the TESLA Electron Collider . . . . . . . . . . . 55
3 Application Relevant He II Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.1 Second Sound Pulse Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2 Transient and Steady Transport in the Mutual Friction Regime . . 58
3.3 The He II Energy Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.4 Fluid Dynamics of Forced Flow He II . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5 He II/Vapor Two Phase Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.6 Fountain Effect (Fluid Management) . . . . . . . . . . . . . . . . . . . . . . . . . 62
4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

The Temperature Dependent Drag Crisis

on a Sphere in Flowing Helium II
Y.S. Choi, M.R. Smith, S.W. Van Sciver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2 Experimental Apparatus and Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

Experiments on Quantized Turbulence at mK Temperatures

S.I. Davis, P.C. Hendry, P.V.E. McClintock, H. Nichol . . . . . . . . . . . . . . . . . 73
1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2 Creation and Detection of Vortices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3 The Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
 Contents IX

6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

Grid-Generated He II Turbulence
in a Finite Channel – Experiment
J.J. Niemela, L. Skrbek, S.R. Stalp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Intermittent Switching Between Turbulent and Potential Flow
Around a Sphere in He II at mK Temperatures
M. Niemetz, H. Kerscher, W. Schoepe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
1 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.1 Stable Turbulent Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.2 Intermittent Switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
2.3 Turbulent Phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.4 Laminar Phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

Part III Vortex Dynamics

Vortex Filament Methods for Superfluids

D.C. Samuels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2 Vortex Filament Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.1 The Biot–Savart Law and the Local Induction Approximation . . . 99
2.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
2.3 Meshing of the Filaments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3 Reconnections of Filaments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4 Analysis of the Superfluid Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5 Alternative Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6 Conclusions: What Needs to Be Done . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Introduction to HVBK Dynamics
D.D. Holm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
1 HVBK Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
2 Incompressible Renormalized HVBK Flows . . . . . . . . . . . . . . . . . . . . . . . . 120
3 Rotating Frame Renormalized HVBK Equations . . . . . . . . . . . . . . . . . . . 123
Appendix: Lie-Poisson Hamiltonian Formulation . . . . . . . . . . . . . . . . . . . . . . . 125
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
X Contents

Magnus Force, Aharonov–Bohm Effect,

and Berry Phase in Superfluids
E. Sonin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
2 Gross–Pitaevskii Theory and Two-Fluid Hydrodynamics . . . . . . . . . . . . 132
3 Interaction of Phonons with a Vortex in Hydrodynamics . . . . . . . . . . . . 133
4 Momentum Balance in the Two-Fluid Hydrodynamics . . . . . . . . . . . . . . 135
5 Magnus Force and the Berry Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

Using the HVBK Model

to Investigate the Couette Flow of Helium II
K.L. Henderson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
2 Linear Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
3 Nonlinear Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.1 Infinite Cylinder Assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.2 Unit Aspect Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

Part IV Turbulence Theory

An Introduction to the Theory of Superfluid Turbulence

W.F. Vinen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
2 Counterflow Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
3 Grid Turbulence in Superfluid Helium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3.1 Measurements of the Decay of Vortex Lines,
and the Quasi-classical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3.2 Superfluid Turbulence on Length Scales
Larger than the Vortex Line Spacing . . . . . . . . . . . . . . . . . . . . . . . . . 153
3.3 The Turbulent Energy Spectra in Superfluid Grid Turbulence . . . 154
3.4 Superfluid Turbulence at Very Low Temperatures . . . . . . . . . . . . . . 155
3.5 Dissipation at Higher Temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

Numerical Methods for Coupled Normal-Fluid

and Superfluid Flows in Helium II
O.C. Idowu, D. Kivotides, C.F. Barenghi, D.C. Samuels . . . . . . . . . . . . . . . . 162
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
2 The Self-Consistent Equation of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3 Numerical Methods for 2-D Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
3.1 The Normal-Fluid Flow in 2-D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
3.2 Delta Function Forcing on a Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
 Contents XI

3.3Extrapolation of the Normal-Fluid Flow in the Neighbourhood

of the Superfluid Vortex Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
3.4 Numerical Stability and Time Stepping . . . . . . . . . . . . . . . . . . . . . . . 168
4 Results in 2-D Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5 Numerical Methods for 3-D Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.1 The Free Normal-Fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.2 The Superfluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.3 The Interaction Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.4 Preliminary Results in 3-D Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
6 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

From Vortex Reconnections to Quantum Turbulence

T. Lipniacki . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
2 Vortex Motion Following Reconnection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
2.1 The Case vns = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
2.2 The Case vns = const = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
3 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

Vortices and Stability in Superfluid Boundary Layers

S.P. Godfrey, D.C. Samuels, C.F. Barenghi . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
1.1 The Two-Fluid Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
2 Boundary Layer Vortices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
2.1 Properties of the Vortex Line Solutions . . . . . . . . . . . . . . . . . . . . . . . 186
2.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
3 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
3.1 Linear Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
3.2 Stability Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

Grid Generated He II Turbulence in a Finite Channel –

Theoretical Interpretation
L. Skrbek, J.J. Niemela . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
2 The Spectral Decay Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

Vortex Tangle Dynamics Without Mutual Friction

in Superfluid 4 He
M. Tsubota, T. Araki, S.K. Nemirovskii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2 Vortex Wave Cascade Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
XII Contents

3 Cascade Process in the Vortex Tangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

3.1 Decay of the Vortex Tangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
3.2 Comparison with the Vinen’s Equation . . . . . . . . . . . . . . . . . . . . . . . 203
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

Applications of the Gaussian Model of the Vortex Tangle

in the Superfluid Turbulent He II
S.K. Nemirovskii, M.V. Nedoboiko . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
2 Constructing the Trial Distribution Function . . . . . . . . . . . . . . . . . . . . . . . 206
3 Hydrodynamic Impulse of the Vortex Tangle . . . . . . . . . . . . . . . . . . . . . . . 208
4 Energy of the Vortex Tangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211

Stochastic Dynamics of a Vortex Loop.

Thermal Equilibrium
S.K. Nemirovskii, L.P. Kondaurova, M. Tsubota . . . . . . . . . . . . . . . . . . . . . . . 212
1 Introduction and Scientific Background . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
2 Langevin Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
3 Fokker–Planck Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
4 Possible Violation of Thermal Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . 217
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218

Stochastic Dynamics of Vortex Loop.

Large-Scale Stirring Force
S.K. Nemirovskii, [Link]. Baltsevich . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
2 Analytical Investigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
3 Conservation Laws and Pair Correlators . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
4 Some Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

Nonequilibrium Vortex Dynamics in Superfluid Phase

Transitions and Superfluid Turbulence
H.-C. Chu, G.A. Williams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
2 Quenched Superfluid Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
3 Superfluid Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
4 Three Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
 Contents XIII

Part V The NLSE and Superfluidity

The Nonlinear Schrödinger Equation as a Model of Superfluidity

P.H. Roberts, N.G. Berloff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
2 The Fluid Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
3 Shortcomings of the GP Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
4 Vortices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
5 Superfluid Turbulence; Vortex Line Reconnection . . . . . . . . . . . . . . . . . . . 243
6 Intrinsic Vortex Nucleation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
7 Capture of Impurities by Vortex Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
8 Nonlocal Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256

Vortex Nucleation and Limit Speed for a Flow Passing Nonlinearly

Around a Disk in the Nonlinear Schrödinger Equation
S. Rica . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
1 Introduction and Formulation of the Problem . . . . . . . . . . . . . . . . . . . . . 258
2 Critical Velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
3 Flow Around a Disk via a Janzen–Rayleigh Expansion . . . . . . . . . . . . . . 262
4 Unstable Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
5 The Euler–Tricomi Equation near the Transonic Region . . . . . . . . . . . . 265
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

Vortices in Nonlocal Condensate Models of Superfluid Helium

N.G. Berloff, P.H. Roberts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
2 Applicability of the Generalized Gross–Pitaevskii Model . . . . . . . . . . . . . 269
3 Nonlocal Nonlinear Schrödinger Equation . . . . . . . . . . . . . . . . . . . . . . . . . 271
4 Vortex Nucleation and Roton Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275

Ginzburg–Landau Description of Vortex Nucleation

in a Rotating Superfluid
I. Aranson, V. Steinberg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
2 Spin-Up and Nucleation of Vortices in Superfluid Helium . . . . . . . . . . . . 277
3 Stability of Multicharged Vortices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
4 Nucleation of Vortices by Rapid Thermal Quench . . . . . . . . . . . . . . . . . . 281
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
XIV Contents

Weak Turbulence Theory for the Gross–Pitaevskii Equation

S. Nazarenko, Y. Lvov, R. West . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
1 Motivation and Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
2 Weak Turbulence Theory for NLSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
3 Linear Dynamics of the GPE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
3.1 Without a Condensate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
3.2 With a Condensate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
4 Applicability of WKB Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
5 Weakly Nonlinear GPE Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289

Dissipative Vortex Dynamics and Magnus Force

L.M. Pismen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
1 Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
2 Magnus Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
3 Three-Dimensional Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
4 Failure of Mechanistic Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296

Transition to Dissipation
in Two- and Three-Dimensional Superflows
C. Huepe, C. Nore, M.-E. Brachet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
2 Definition of the System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
3 Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
4 Bifurcation Diagram and Scaling in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
5 Subcriticality and Vortex-Stretching in 3D . . . . . . . . . . . . . . . . . . . . . . . . 301
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303

Part VI Bose–Einstein Condensation

Motion of Objects Through Dilute Bose–Einstein Condensates

C.S. Adams, B. Jackson, M. Leadbeater, J.F. McCann, T. Winiecki . . . . . . 307
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
2 Fluid Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
3 Time-Independent Solutions in the Object Frame . . . . . . . . . . . . . . . . . . 309
4 The Critical Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
5 Vortex Shedding and Drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
6 The Critical Velocity in Inhomogeneous Condensates . . . . . . . . . . . . . . . 314
7 Comparison to Ions in Helium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
 Contents XV

Stability of a Vortex in a Rotating

Trapped Bose–Einstein Condensate
A.L. Fetter, A.A. Svidzinsky . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
1 Time-Dependent Gross–Pitaevskii Equation . . . . . . . . . . . . . . . . . . . . . . . 320
1.1 Equivalent Hydrodynamics of Compressible Isentropic Fluid . . . . . 320
1.2 Thomas–Fermi Limit for Large Condensates . . . . . . . . . . . . . . . . . . . 321
2 Energy of a Vortex in a Large Rotating Trap . . . . . . . . . . . . . . . . . . . . . . 321
3 Small-Amplitude Excitation of a Vortex in a Rotating Trap . . . . . . . . . . 322
3.1 Stability of a Vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
3.2 Splitting of Normal-Mode Frequencies Caused by a Vortex . . . . . . 323
4 Vortex Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
4.1 Dynamics of Straight Vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
4.2 Inclusion of Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
Kinetics of Strongly Non-equilibrium
Bose–Einstein Condensation
B. Svistunov . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
2 Kinetic Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
3 Coherent Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
4 External Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333

Quantum Nucleation of Phase Slips

in Bose–Einstein Condensates
H.P. Büchler, V.B. Geshkenbein, G. Blatter . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
2 Effective Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
3 Finite Size Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340

Part VII Vortex Reconnections and Classical Aspects

Vortex Reconnection in Normal and Superfluids

J. Koplik . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
2 Some Vortex Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
3 The Importance of Reconnection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
4 Reconnection in Normal Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
5 Reconnection in Superfluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
XVI Contents

Helicity in MHD and Hydro Reconnection

A. Brandenburg, R.M. Kerr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
2 Dissipation of Energy and Helicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
3 Interlocked Flux Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
4 Orthogonal Vortex Tubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365

Tropicity and Complexity Measures for Vortex Tangles

R.L. Ricca . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
1 Vortex Structures and Tangles
in Classical and Quantized Vortex Flows . . . . . . . . . . . . . . . . . . . . . . . . . . 366
2 Measures of Tropicity for Vortex Tangles:
Tubeness, Sheetness and Bulkiness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
3 Measures of Geometric Complexity:
Directional Alignment and Writhing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
4 Algebraic Measure of Structural Complexity:
Average Crossing Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
5 Measures of Topological Entanglement:
Kinetic Helicity and Directional Linking . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
6 Relationships Between Complexity Measures and Energy Levels . . . . . . 371
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
The Geometry of Magnetic and Vortex Reconnection
G. Hornig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
2 Magnetic Reconnection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
3 Vortex Reconnection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379

Current-Sheet Formation
near a Hyperbolic Magnetic Neutral Line
B.K. Shivamoggi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
2 Current-Sheet Formation at a Hyperbolic Magnetic Neutral Line
in a Stagnation-Point Plasma Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
3 Effect of a Uniform Shear–Strain in the Plasma Flow . . . . . . . . . . . . . . . 385
4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387

Nonlocality in Turbulence
A. Tsinober . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
1 Introduction and Simple Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
2 Different Aspects of Nonlocality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
2.1 Direct Coupling Between Large and Small Scales . . . . . . . . . . . . . . 392
3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
 Contents XVII

Part VIII Helium 3 and Other Systems

Quantized Vorticity in Superfluid 3 He-A:

Structure and Dynamics
R. Blaauwgeers, V.B. Eltsov, M. Krusius, J. Ruohio, R. Schanen . . . . . . . . 399
1 Superfluid 3 He . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
2 Order-Parameter Texture and Superflow in 3 He-A . . . . . . . . . . . . . . . . . . 401
3 Double-Quantum Vortex Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
4 Vortex Sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
5 Dynamic Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
6 Summary and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419

Vortices in Metastable 4 He Films

R. Blossey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
1 Wetting Properties of 4 He on Weak-Binding Alkali Metals . . . . . . . . . . . 421
1.1 Wetting Transitions of Liquid Helium . . . . . . . . . . . . . . . . . . . . . . . . 421
1.2 Interface Model Description of Wetting Transitions . . . . . . . . . . . . . 422
2 Lifetime of an Undercooled Film . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
3 Application to 4 He/Cs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427

Quantum Hall Effect Breakdown Steps and Possible Analogies

with Classical and Superfluid Hydrodynamics
L. Eaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
2 Model and Comparison with Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . 430
3 Analogies with Classical and Quantum Fluids . . . . . . . . . . . . . . . . . . . . . . 434
4 The Breakdown Steps and Their Relation
to Other Types of QHE Breakdown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
Atomic Bose Condensate with a Spin Structure:
The Use of the Bloch State
H. Kuratsuji . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
2 Order Parameter and Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
3 Hydrodynamical Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
4 Vortex State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
4.1 The Profile of a Single Vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
4.2 Vortex Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
XVIII Contents

Quantum Dynamics of Vortex–Antivortex Pairs in a Circular Box

V. Penna . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
2 Canonical Quantization of Planar Vortices . . . . . . . . . . . . . . . . . . . . . . . . . 446
2.1 The Spectrum of Unbounded Vortex Pairs . . . . . . . . . . . . . . . . . . . . 447
3 Pair Quantum Dynamics in a Circular Box . . . . . . . . . . . . . . . . . . . . . . . . 448
3.1 Spectral Structure of Low Energy States . . . . . . . . . . . . . . . . . . . . . . 449
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451

Index of Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453

List of Contributors

C.S. Adams R.J. Donnelly

Physics Department Physics Department
University of Durham University of Oregon
Durham DH1 3LE, UK Eugene, OR 97403, USA
[Link]@[Link] russ@[Link]

C.F. Barenghi
University of Newcastle L. Eaves
Mathematics Department School of Physics and Astronomy
Newcastle upon Tyne, NE1 7RU University of Nottingham
UK Nottingham NG7 2RD, UK
[Link]@[Link] [Link]@[Link]

N. Berloff
Mathematics Department A. Fetter
University of California Geballe laboratory for Advanced
Los Angeles, CA 90095-1555, USA Materials
nberloff@[Link] Stanford University
Stanford, CA 94305-4045, USA
R. Blossey fetter@[Link]
Department of Physics
University of Essen
45117 Essen, Germany K.L. Henderson
blossey@[Link] Faculty of Computer Studies and
Mathematics
A. Brandenburg University of the West of England
NORDITA Bristol BS16 1QY, UK
Blegdamsvej 17 [Link]@[Link]
2100 Copenaghen, Denmark
brandenb@[Link]
D.D. Holm
H.P. Buechler Theoretical Division
Theoretical Physics Mail Stop B284
ETH Los Alamos National Laboratory
8093 Zuerich, Switzerland Los Alamos NM 87545, USA
buechler@[Link] holm@[Link]
XX List of Contributors

G. Hornig M. Leadbeater
Dept. Theoretical Physics IV Physics Department
Ruhr-Universitaet Bochum University of Durham
44780 Bochum, Germany Durham DH1 3LE, UK
gh@[Link] [Link]@[Link]

C. Huepe T. Lipniacki
James Frank Institute Institute of Fundamental Technologi-
University of Chicago cal Research
5640 S. Ellis Avenue Świȩtokrzyska St. 21
Chicago, IL 60637, USA 00-049 Warsaw, Poland
cristian@[Link] tlipnia@[Link]

O. Idowu P.V.E. McClintock

Center for Turbulence Research University of Lancaster
Stanford University Physics Department
Stanford, CA 94305-3030, USA Lancaster LA1 4YB, UK
idowu@[Link] [Link]@[Link]

D. Kivotides
S. Nazarenko
Mathematics Department
Mathematics Institute
University of Newcastle
University of Warwick
Newcastle NE1 7RU, UK
Coventry CV4 7AL, UK
[Link]@[Link]
snazar@[Link]

J. Koplik
Levich Institute, T-1M S.K. Nemirovskii
City College of New York Institute of Thermosphysics
New York, NY 10031, USA 630090 Novosibirsk, Russia
koplik@[Link] nem@[Link]
koplik@[Link]
J.J. Niemela
M. Krusius Physics Department
Low Temperature Laboratory University of Oregon
Helsinki University of Technology Eugene, OR 97403, USA
02015 HUT joe@[Link]
Finland
krusius@[Link] M. Niemetz
Institut für Experimentelle
H. Kuratsuji und Angewandte Physik
Department of Physics Universität Regensburg
Ritsumeikan University 93040 Regensburg, Germany
Kusatsu City 525-8577, Japan [Link]@
kra@[Link] [Link]
 List of Contributors XXI

V. Penna B.K. Shivamoggi

Dipartimento di Fisica University of Central Florida
and INFM Orlando, FL 32816, USA
Politecnico di Torino ijjmms@[Link]
[Link] Duca degli Abruzzi 24
10129 Torino, Italy E. Sonin
penna@[Link] Racah Institute of Physics
Hebrew University of Jerusalem
L.M. Pismen Givat Ram,
Department of Chemical Engineering Jerusalem 91904, Israel
and Minerva Center for Nonlinear sonin@[Link]
Physics of Complex Systems
Technion - Israel Institute L. Skrbek
of Technology KFNT MFF UK
32000 Haifa, Israel Charles University
V Holesovickach 2
S. Rica 180 00 Prague 8, Czech Republic
CMM CNRS UCHILE skrbek@[Link]
Av Blanco Encalada 2120
Santiago, Chile B. Svistunov
rica@[Link] Russian Research Center
rica@[Link] Kurchatov Institute
123182 Moscow, Russia
L.R. Ricca svist@[Link]
Mathematics Department
University College London A. Tsinober
Gower Street Faculty of Engineering
London WC1E 6BT, UK Tel Aviv University
ricca@[Link] Tel Aviv, Israel
tsinober@[Link]
P.H. Roberts
Mathematics Department M. Tsubota
University of California Department of Physics
Los Angeles, CA 90095-1555, USA Osaka City University
roberts@[Link] Osaka 558-8585, Japan
tsubota@[Link]
W. Schoepe
Institut für Experimentelle und S.W. Van Sciver
Angewandte Physik National High Magnetic Field
Universität Regensburg Laboratory
93040 Regensburg, Germany Florida State University
[Link]@ Tallahassee FL 32310, USA
[Link] vnsciver@[Link]
XXII List of Contributors

É. Varoquaux University of Birmingham

CNRS-Université Paris-Sud Birmingham B15 2TT, UK
Laboratoire de Physique des Solides [Link]@[Link]
Bâtiment 510
F-91405 Orsay Cedex, France G. Williams
varoquaux@[Link] Department of Physics and Astronomy
University of California
W.F. Vinen Los Angeles, CA 90095, USA
Physics Department gaw@[Link]
Introduction to Superfluid Vortices
and Turbulence

Carlo F. Barenghi

Mathematics Department, University of Newcastle, Newcastle NE1 7RU, UK

1 The Two-Fluid Model

My aim in this article is to introduce the basic properties of quantized vortex lines
in Helium II and summarize the main experimental observations of superfluid
turbulence. Then I shall discuss a selection of the theoretical methods used to
study quantized vorticity and turbulence and the results obtained using these
methods.
The liquid state of 4 He exists in two phases: a high temperature phase called
Helium I, and a low temperature phase, called Helium II. The two phases are
separated by a transition called the lambda transition, which occurs at the crit-
ical temperature T = Tλ = 2.172 K at saturated vapour pressure and marks
the onset of Bose Einstein condensation (BEC) and quantum order. The phe-
nomenon of BEC is described in the article of Stringari. Helium I is a classical
fluid which obeys the ordinary Navier - Stokes equations. Hereafter the focus of
attention is only Helium II.
A simple, phenomenological model which explains the motion of Helium II
is the two - fluid theory of Tisza and Landau [1]. In this model Helium II is
described as the intimate mixture of two fluid components which penetrate each
others, the normal fluid and the superfluid. Each fluid component has its own
density and velocity field, ρn and vn for the normal fluid and ρs and vs for
the superfluid. The total density of Helium II is ρ = ρn + ρs . The superfluid
component is irrotational, and, since it carries nor entropy nor viscosity, is similar
to a classical, inviscid Euler fluid. The normal fluid component is a gas of thermal
excitations called phonons and rotons depending on the wavenumber. The normal
fluid carries the entire entropy and viscosity of Helium II and is similar to a
classical, viscous Navier - Stokes fluid.
The relative proportion of normal fluid and superfluid is determined by the
absolute temperature T . At T = 0 Helium II is entirely superfluid: ρs /ρ = 1 and
ρn /ρ = 0. If the temperature is increased the superfluid fraction decreases and
the normal fluid fraction increases, until, at T = Tλ , Helium II becomes entirely
normal: ρs /ρ = 0 and ρn /ρ = 1. The temperature dependence of the two fluid’s
fractions is nonlinear: for example ρn /ρ drops from 100 percent at T = Tλ to
55 percent at 2.0 K and to 7.5 percent at 1.4 K, and is effectively negligible at
temperatures below 1 K.
The two - fluid model explains many observed phenomena. Among them
it is worth mentioning second sound and thermal counterflow because they are
important in the study of turbulence. Ordinary sound in Helium II is called first

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 3–14, 2001.

c Springer-Verlag Berlin Heidelberg 2001
4 C.F. Barenghi

sound. A first sound wave is an oscillation of density ρ and pressure P in which

temperature T and entropy S remain almost constant and vn and vs move in
phase with each others. A second sound wave, on the contrary, is an oscillation
of T and S in which ρ and P remain almost constant and vn and vs move in
antiphase. Second sound is used to detect quantized vorticity.
Thermal counterflow is Helium II’s special way to transfer heat. Consider a
channel which is closed at one end and open to the helium bath at the other
end. At the closed end a resistor dissipates a known heat flux W . In an ordinary
fluid, such as water or Helium I, heat is transferred away from the resistor by
conduction, provided that one is careful to prevent convective motion, so the
heat flux W is proportional to the temperature gradient ∇T and there is a well
defined thermal conductivity at small W . In Helium II the heat is carried by
the normal fluid away from the resistor, W = ρST vn . Because of the presence
of the closed end, however, the mass flux is zero, j = ρs vs + ρn vn = 0, so some
superfluid must flow toward the resistor to conserve mass, vs = −(ρn /ρs )vn . In
this way a counterflow vns = vn − vs is generated which is proportional to the
applied heat flux, vns = W/(ρs ST ). If W exceeds a critical value, then superfluid
turbulence is generated. Turbulence limits the heat transfer properties of Helium
II, so it is relevant in the engineering applications, as described in the article of
Van Sciver.

2 Quantized Vortex Lines

The quantization of the circulation of the superfluid makes the hydrodynamics of
helium II particularly interesting [2]. Superfluid vortex lines appear when helium
II rotates or moves faster than a critical velocity. This vortex nucleation process
has been the subject of many investigations and is described in the articles of
Adams and Rica. Superfluid vortex lines can be spatially organized (laminar
vortex flows) or disorganized (turbulent vortex tangles). The key property of a
superfluid vortex line, discovered by Onsager and developed by Feynman, is that
the circulation is quantised, that is to say

h
vs · dl = , (1)
C m

where C is a circular path around the axis of the vortex. The ratio Γ = h/m of
Plank’s constant and helium’s mass is called the quantum of circulation and has
value Γ = 9.97 × 10−4 cm2 /sec.
The simplest way to create superfluid vortex lines is to rotate a cylinder
filled with Helium II at constant angular velocity Ω. Provided that Ω is large
enough, superfluid vortex line appear and form on ordered array of areal density
n = 2Ω/Γ , all vortex lines being aligned along the axis of rotation. In this way
the superfluid mimics the vorticity 2Ω of the solid body rotating normal fluid,
each vortex line contributing one quantum to the total circulation.
 Superfluid Vortices and Turbulence 5

3 Modelling the Vortex Lines

3.1 Microscopic Model
Using cylindrical coordinates r, φ, z it follows from equation (1) that the super-
fluid velocity around the axis of a vortex line is vs,φ = Γ/2πr, hence it diverges
as r → 0. To understand what happens as r → 0 it is instructive to use the
nonlinear Schroedinger equation (NLSE), which is discussed in the article of
Roberts. The NLSE models a gas of bosons which interact with each others via
a delta function repulsive potential of strength V0 :

∂ψ 2 2
i =− ∇ ψ − mEψ + V0 ψ|ψ|2 , (8)
∂t 2m
Here E is the energy per unit mass and  = h/2π. The wavefunction ψ can be
written as ψ = AeiΦ in terms of an amplitude A and a phase Φ. In this way
one can define the condensate’s density ρBEC = mA2 and velocity vBEC =
(/m)∇Φ; the last relation confirms that the superfluid is irrotational as en-
visaged by Landau. This transformation establishes the hydrodynamics of the
model: equation (8) is equivalent to a continuity equation and an Euler equation
(modified by the so called quantum pressure term).
The NLSE has a vortex solution: if Φ is the azimuthal angle φ then we have
vBEC,φ = Γ/(2πr) which is the Onsager - Feynman vortex. Substitution into
the NLSE yields a differential equation for ρBEC . One finds that ρBEC tends
to the bulk value m2 E/V0 for r → ∞, and that ρBEC → 0 for r → 0. The
characteristic distance over which ρBEC changes from its bulk value to zero is
a0 ≈ 10−8 cm. This distance is called the vortex core parameter. We conclude
that the superfluid vortex line is hollow at the core. Geometrically, a vortex
line transforms the volume occupied by the superfluid into a multiply connected
region.
Hereafter we identify ρBEC with ρs at absolute zero and vBEC with vs . It
must be noted that this identification is convenient from the point of view of
having a simple hydrodynamics model but is not entirely correct. The reason is
that Helium II is a dense fluid, not the weakly interacting Bose gas described
by the NLSE, so the condensate is not the same as the superfluid component.
One should compare the case of Helium II (in which the NLSE is a rather
approximate model) with the case of BEC in clouds of trapped alkali atoms (in
which the NLSE is a better model because the bosons’ interaction is weaker).
Another drawback of the NLSE is that it fails to describe the observed dispersion
relation of Helium II at high momenta. If one studies small oscillations of the
uniform solution of the NLSE and interprets them as thermal excitations, one
finds a dispersion relation E = E(p) in which the energy E is proportional
to the momentum p at small p (phonons) and then becomes quadratic in p at
high p (free particles). The spectrum of excitations observed in Helium II is
different, because the phonon part is followed by the rotons’ minimum. Despite
these shortcomings, of which one must be aware, the NLSE is much used as a
convenient hydrodynamical model of Helium II at T = 0.
6 C.F. Barenghi

3.2 Mesoscopic Model

An important feature of the NLSE model is that it makes visible what happens
on the microscopic scale (≈ 10−8 cm) of the vortex core parameter. This makes
the NLSE a useful tool to investigate phenomena such as vortex nucleation
and vortex reconnections. There are however other problems in which one is
concerned with the behaviour of a large number of vortex lines. In these problems
the resolution of what happens on the microscopic scale is not necessary and can
even be a waste of computer resources. It is more appropriate to use the classical
vortex dynamics model.
The vortex dynamics model was pioneered by Schwarz [3] and consists in
representing a vortex line as a curve s = s(ξ, t) in three-dimenensional space,
where ξ is arclength and t is time. We call s = ds/dξ and note that the vectors
s, s and s × s are perpendicular to each others and point along the tangent,
normal and binormal respectively. To determine the equation of motion of s
we must identify the forces acting upon the line: the Magnus force fM and, at
nonzero temperature, the drag force fD .
The Magnus force arises when a body with circulation about it moves in
a flow: the circulation creates an increased total velocity of fluid on one side,
which results in excess pressure from the other side. Since the key ingredient is
the circulation rather than the details of the body, we apply the concept to a
vortex line and write
fM = ρs Γ s × (vL − vs,tot ), (2)
where vL = ds/dt is the velocity of the line in the laboratory frame and vs,tot is
the total velocity of the surrounding superfluid, also in the laboratory frame. The
velocity vs,tot consists of two parts: any superfluid velocity applied externally
and the self - induced velocity of the vortex line.

vs,tot = vs + vi , (3)

The self induced velocity vi describes the motion which a vortex line induces
onto itself because of its own curvature and is determined by the Biot - Savart
(BS) law 
Γ (z − s) × dz
vi (s) = , (4)
4π |z − s|3
The Biot - Savart law is sometimes replaced by the Local Induction Approxima-
tion (LIA), which is
vi ≈ βs × s , (5)
where β = Γ/(4π)log(1/(|s |a0 )). Since |s | = 1/R where R is the local radius
of curvature, we have
Γ R
vi ≈ ln( )b̂, (6)
4π a0
where b̂ is the binormal.
The drag force fD arises from the mutual friction between the superfluid
vortex lines and the normal fluid [4]. Normal fluid flowing with velocity vn past
 Superfluid Vortices and Turbulence 7

a vortex core exerts a frictional force fD per unit length on the superfluid in the
neighborhood of the core given by

fD = −αρs Γ s × [s × (vn − vs,tot )] − α ρs Γ s × (vn − vs,tot ), (7)

The dimensionless parameters α and α are temperature dependent and are of-
ten written in terms of mutual friction coefficients B and B  defined by α =
ρn B/(2ρ) and α = ρn B  /(2ρ). The values of B and B  are known from experi-
ments. Samuels and Donnelly [5] showed that, at least in the high temperature
range in which most experiments are performed, the friction arises from the
scattering of rotons from the velocity field of a vortex line. The calculation of
the mutual friction parameters over the entire temperature range is still an open
question and has subtle aspects, as explained in the article by Sonin. An impor-
tant effect of the mutual friction is that it modifies the propagation of second
sound. By measuring the second sound attenuation one can determine the su-
perfluid vortex line density L0 , defined as the length of vortex line per unit
volume.
Now that fM and fD are identified we can make use of the fact that the sum
of all forces is zero as the line’s inertia is negligible:

fD + fM = 0, (8)

Hence, solving for ds/dt, we obtain Schwarz’s equation

ds
= vs + vi + αs × (vn − vs − vi ) + α (vn − vs − vi ), (9)
dt
An algorithm to numerically simulate the time evolution of any arbitrary
configuration of vortex lines can be developed on the basis of Schwarz’s equa-
tion and is described in the article of Samuels. Here it suffices to say that an
initial vortex configuration is discretized into N points. The time evolution of
each point is calculated using (9), given externally applied fields vs and vn and
given the temperature T , which determines the friction coefficients α and α .
The transverse part of the mutual friction, proportional to α , is smaller and is
sometimes neglected. The number of points N and the time step must be allowed
to vary during the evolution to take into account the appearance of regions of
high or low curvature. Note that, if one uses the BS law, the computational time
is proportional to N 2 , while, if one uses the LIA, this time is only proportional to
N . Numerical simulations of vortex tangles based on the BS are therefore compu-
tationally expensive. However the use of the LIA can give misleading results [6].
Finally the numerical simulation must be able to perform vortex reconnections
when two vortex lines come sufficiently close to each others. This process is an
arbitrary assumption in the context of the dynamics of vortex filaments, but is
justified by a microscopic calculation [7] performed using the NLSE, as explained
in the article by Koplik.
8 C.F. Barenghi

3.3 Macroscopic Model

Besides the NLSE model (in which the vortex core is visible) and the vortex
dynamics model (in which the core is not visible but the vortex line is) there
is a third macroscopic model in which the individual vortex lines are not visi-
ble and Helium II is considered as a continuous vortex flow. The third model,
called the HVBK model[8] [9], is useful to describe laminar flows in which the
vortex line are spatially organized. Examples are solid body rotation, flows in
an rotating annulus or cavity, and Taylor - Couette flow. The HVBK model is
a generalization of Landau’s equations to include the presence of vortices. The
fluid particles of the model are assumed to be large enough to be threaded by
many vortex lines which are aligned in the same direction. In this way the indi-
vidual vortex lines are not visible, the superfluid is treated as a continuum and
we can define a macroscopic, nonzero superfluid vorticity ωs , despite the fact
that, microscopically, the superfluid velocity field obeys ∇ × vs = 0. Clearly the
HVBK equations are valid only if the vortex lines are organized spatially and not
randomly oriented, and if the length scales of the flow under consideration are
much bigger than the average separation between the vortex lines. An example
is the simple case of Helium II inside a rotating cylinder, for which ωs = 2Ωẑ.
The incompressible HVBK equations are
∂vn 1 ρs ρs
+ (vn · ∇)vn = − ∇P − S∇T + νn ∇2 vn + F, (10)
∂t ρ ρn ρ
∂vs 1 ρn
+ (vs · ∇)vs = − ∇P + S∇T + T − F, (11)
∂t ρ ρ
where we have defined
ωs = ∇ × vs , (12)

B B
F= ω
 s × [ωs × (vn − vs − νs ∇ × ω
 s )] + ωs × (vn − vs − νs ∇ × ω
 s ), (13)
2 2
ω
 s = ωs /|ωs |, (14)
T = −νs ωs × (∇ × ω
s ), (15)
νs = (Γ/4π) log(b0 /a0 ), (16)
The quantities F, T and νs are respectively the friction force, the tension
force and the vortex tension parameter, and b0 = (2ωs /Γ )−1/2 is the intervor-
tex spacing. Note that νs has the same dimension of a kinematic viscosity, but
physically it is very different: it is related to the ability of a superfluid fluid
particle to oscillate because of the vortex waves which can be excited along the
vortex lines threading the fluid particle itself. Note that without F and T the
HVBK equations are formally the same as the original two - fluid equations of
Landau. The HVBK equations have interesting limits. If T → Tλ then ρs → 0 so
the normal fluid equation (10) becomes the classical Navier - Stokes equation. If
T → 0 then ρn → 0 so the superfluid equation (11) describes a pure superflow;
 Superfluid Vortices and Turbulence 9

by setting Plank’s constant equal to zero we have then νs = 0 and the pure
superflow equation becomes the classical Euler equation.
The HVBK model has been used with success to study the transition from
Couette flow to Taylor vortex flow [10]: Barenghi’s predictions [11] of the critical
Reynolds number of the transition and its temperature dependence were con-
firmed by the experiments [12]. Taylor - Couette flow has also been studied in
the nonlinear Taylor vortex flow regime [13] [14] and the results are in agreement
with the observations, providing a further test of the theory. These results are
described in the article by Henderson.

4 Turbulence
Superfluid turbulence manifests itself as a tangle of vortex lines and can be
generated in many ways. Turbulent thermal counterflow was the first turbulent
flow which was studied in detail in a series of pioneering papers by Vinen [15].
Since this flow has no classical analogy it deserves a separate discussion. Other
ways to generate turbulence are more classical in character, and we refer to them
as turbulent coflows.

4.1 Turbulent Counterflows

As said in the first section, laminar counterflow breaks down if the heat flux W
exceeds a critical value Wc . Corresponding to Wc there is a critical counterflow
velocity vns = vc1 = Wc /(ρs ST ). At this critical velocity a vortex tangle appears.
The simplest way to characterize the tangle is to measure or compute its vortex
line density L0 . A great number of measurements of L0 were performed in pipes
2
and channels and showed that L0 = γvns for vns > vc1 where γ is some temper-
ature dependent parameter. Further measurements in circular pipes indicated
that there exists a second critical velocity vc2 at which the vortex line density
L0 becomes suddenly larger. The region of weak turbulence vc1 < vns < vc2 and
the region of strong turbulence v > vc2 are called the T-1 and the T-2 turbu-
lent states respectively [16]. On the theoretical side, the numerical simulations of
Schwarz [3] based on the vortex dyanmics approach confirmed the existence of
a self - sustaining vortex tangle driven by a constant, spatially uniform vns , and
gave values of L0 consistent with the observations in the T-2 state. The nature of
the first turbulent states and the physical meaning of the transition at vns = vc2
were a puzzle until Melotte and Barenghi [17] showed that the transition from
the T-1 to the T-2 state is related to the onset of normal fluid turbulence (see
section 6).

4.2 Turbulent Coflows

Turbulence can also be induced in more traditional ways by driving a mass flow,
spinning discs or propellers, towing a grid or a sphere, using shocks, ultrasound,
jets and rotating cylinders (Taylor - Couette flow). Considering all these results
10 C.F. Barenghi

together, the general trend is that the slow, laminar flow of Helium II, with or
without vortices, tends to be rather different from the flow of a classical fluid,
but when Helium II moves fast and is driven turbulent it seems to behave like a
classical turbulent flow.
To characterize the turbulence we use the Reynolds number Re = U L/ν
where L is the length scale, U the velocity scale and ν the kinematic viscosity.
Examples of the observed classical features of Helium II turbulence are the fol-
lowing. Mass flow rates and pressure drops at Re ≈ 106 can be well described
by using classical relations for high Reynolds number classical flows [18]. Ex-
periments on large scale turbulent vortex rings at Re ≈ 4 × 104 detect normal
fluid vorticity and superfluid vorticity moving together as a single structure [19]
Experiments on turbulent Taylor - Couette flow at Re ≈ 4×103 show the typical
structures of classical turbulent Taylor - Couette flow [20]. Experiments on the
decay of superfluid vorticity created by towing a grid show that the decay in
time obeys the same laws as of the decay of classical turbulence [21]. Moreover
the decay appears to be independent of temperatures in the explored range,
from the lambda region down to 1.4 K, where the normal fluid fraction is only
7.5 percent. Experiments on turbulence created by rotating blades [22] show the
classical Kolmogorov −5/3 power spectrum in the temperature range explored,
from the lambda region down to T = 1.4 K again. Finally experiments on the
drag on a moving sphere (Re ≈ 105 ) show the same drag crisis observed in a
classical fluid [23].
The temperature independence of these observations is interesting. The nor-
mal fluid must be responsible for these classical aspects, but the dynamical im-
portance of the normal fluid should be related to the fraction ρn /ρ, so it should
be negligible at temperatures as low as 1.4 K. Since a large number of quantized
vortex lines must be present in these turbulent flow, it is speculated that they
are able to lock together the two fluid components of Helium II into a single
fluid which behaves somewhat like a classical turbulent fluid. This is a topic of
much current interest‘[24] [25] and is discussed in the article of Vinen.

5 Motion of Superfluid Vortices for a Given Normal Fluid

In the next two sections we discuss some selected examples about the interac-
tion of superfluid vortices and the normal fluid. The numerical calculations done
by Schwarz using the vortex dynamics model (section 3.2) were concernd with
thermal counterflow. The calculations were successful; above all, they gave great
insight into the nature of turbulence, confirming the existence of a vortex tangle,
and brought into attention the issue of vortex reconnections. However Schwarz’s
approach had an important limitation: it assumed that the quantity vn − vs
is constant in time and space. The work of Schwarz was followed up by other
calculations which made different assumptions about the normal fluid: uniform
flow [26], Poiseuille flow [26] [27], a Gaussian vortex [28], ABC flows [29]. All these
calculations were kinematic in character, because the driving field (vn for mass
flow and vn − vs for counterflow) were imposed at the beginning of the calcula-
 Superfluid Vortices and Turbulence 11

tion and never changed, neglecting the back reaction of the superfluid vortices.
Not surprising, the vortex tangles calculatedusing different driving fields looked
different from each others. The success of the original calculation of Schwarz in
reproducing a vortex line density L0 consistent with the experiments in the T-2
state was probably due to the fact that the uniform vn − vs profile used by
Schwarz modelled well the average, flattened turbulence profile in the channel
flows under consideration.
Despite this limitation, kinematic calculations are clearly useful and can shed
light on important physical mechanisms. A particularly interesting mechanism
which is relevant to turbulence is the Ostermeier- Glaberson instability. This
is an instability of Kelvin vortex waves which takes place if the component of
the normal fluid velocity in the direction parallel to the vortex lines exceeds a
critical value. The instability was first observed by Cheng, Cromar and Don-
nelly [30], but is was Ostermeier and Glaberson [31] who explained it and it
was Samuels [28] who realized its importance in turbulence. Another physical
mechanism, which is apparent in the kinematic numerical calculations [28] [29]
is vorticity matching: once the Ostermeier - Glaberson instability has generated
superfluid vortex lines by extracting energy from the normal fluid, then the vor-
tex lines become attracted to the regions of concentrated normal fluid vorticity.
Therefore, although the local superfluid velocity pattern in the bundles is very
complicated, the averaged vorticity ωs is similar to the vorticity of the driving
normal fluid. A further interesting application of the Ostermeier - Glaberson
instability is that it creates a damping length scale  for superfluid turbulence:
superfluid structures at length scale smaller than  will lose energy to the normal
fluid and be dissipated [25].

6 Motion of the Normal Fluid

at Given Superfluid Vortices
The kinematic approach can also be used to study the behaviour of the nor-
mal fluid given the superfluid vortices rather than viceversa. An example is the
calculation of Melotte and Barenghi [17] who studied the stability of normal
fluid motion induced by heat transfer in the presence of a uniform tangle of
superfluid vorticity of density L0 . The governing equation is a Navier - Stokes
equation modified by the introduction of a friction term
∂vn 1 ρs 1
+ (vn · ∇)vn = − ∇P − S∇T + νn ∇2 vn + F, (20)
∂t ρ ρn ρn
where
Bρs ρn
F= gωs (vs − vn ), (21)
2ρ
ωs = Γ L0 , and, assuming an isotropic tangle, g ≈ 2/3. Melotte and Barenghi’s
stability calculation determined the critical vortex line density at which the
normal fluid’s profile becomes turbulent. The results are in agreement with the
12 C.F. Barenghi

measurements of the transition from the weak T-1 state (in which the superfluid
is turbulent but the normal fluid is not) to the strong T-2 state (in which both
fluids are turbulent).
The same kinematic approach has been used to study how the superfluid
vortices affect the stability of normal fluid in channel flows. This modified Orr -
Sommerfeld problem[32] is described in the article by Godfrey.

7 Fully Coupled Motion of Superfluid Vortices

and Normal Fluid
The limitations of the kinematic approach can be overcome if one allows the
normal fluid and the superfluid vortices to determine each others self - consis-
tently during the evolution. Essentially one has to combine the (Lagrangian)
calculation of the superfluid vortices based on vortex dynamics (section 2) with
with the (Eulerian) calculation of the normal fluid based on a modified Navier
- Stokes equation (section 5). However, since the velocity field vn which is com-
puted is local, the equation of Schwarz (9) requires a modification[33]. At the
place of (9) one has

ds
= vs + h1 vsi + h2 s × (vn − vi ) − h3 s × s × vn , (22)
dt
where vi is given by (4). In the absence of friction we have h2 = h3 = 0 and
h1 = 1. The normal fluid is determined by a modified Navier - Stokes equation
like (20), but now F is obtained numerically by considering the friction force on
the normal fluid per unit length of superfluid vortex line

f = Ds × s × (vn − vL ) + Dt s × (vn − vL ), (23)

where D and Dt are mutual friction coefficients, and summing the contribution
of each segment of s that falls within the computational grid cell of the normal
fluid.
The drawback of the fully - coupled approch is the computational cost. A
useful compromize is to implement it in two dimensions x, y neglecting the z
dependence: in this way the vortex lines becomes vortex points (the intersection
of vortex lines with the plane z = 0) and the normal fluid can be obtained
easily using the stream function - vorticity formulation. It is found that a single
superfluid vortex point induces an elongates normal fluid jet [34] whose intensity
depends on the temperature, hence creating a dipolar vorticity structure in the
normal fluid. A similar calculation in three dimensions performed by Kivotides,
Barenghi and Samuels [35] showed that a superfluid vortex ring creates around
itself a normal fluid structure which consists of two coaxial vortex vortex rings
of opposite polarity.
Since the normal fluid is accelerated by the superfluid vortices by friction
which depends on the difference between vn and vL , the normal fluid speed
cannot exceed the vortex lines’ speed. If one considers the typical vortex line
 Superfluid Vortices and Turbulence 13

density L0 of experiments and estimates vL ≈ Γ/2πδ using the average inter-

−1/2
vortex spacing δ ≈ L0 , one finds an upper bound for vn , using which one
estimates the normal fluid Reynolds number to be in the range from 1 to 10, too
small for turbulence[36].
This conclusion refers only to the ability of vortex line to make the normal
fluid turbulent by directly stirring it. A second mechanism in which the vortex
line can make the normal fluid turbulent is by causing instabilities in a mean
flow, as showed in section 6. Finally normal fluid turbulence can also be induced
directly by the boundaries.

8 Discussion

I have mentioned only a selection of topics of current interest in the study of

quantized vorticity. Much work is in progress. It is clear that the NLSE will
be used more and more to understand the details of fundamental processes:
an example is vortex reconnections[37]. Similarly, the use of vortex dynamics
simulations in the absence of friction is proving useful to understand processes
such as the turbulence cascades of vortex waves [38]. At finite temperatures,
fully coupled numerical simulations are proving useful to understand the effects
of the reconnections on the normal fluid[39] and hopefully will be able in the
near future achieve contact between theory and experiment.

References
1. L.D. Landau and E.M. Lifshitz: Fluid Mechanics, 2nd edn. (Pergamon Press, Lon-
don 1987)
2. R.J. Donnelly:Quantized Vortices in Helium II, Cambridge University Press, Cam-
bridge (1991).
3. K.W. Schwarz: Phys. Rev. Lett. 49, 283 (1982); Phys. Rev. B 31 5782 (1985); Phys.
Rev. B 38 2398 (1988)
4. C.F. Barenghi, R,J, Donnelly and W.F. Vinen: J. Low Temp. Phys. 52 189 (1983)
5. D.C. Samuels and R.J. Donnelly: Phys. Rev. Lett. 65 187 (1990).
6. R.L. Ricca, D.C. Samuels and C.F. Barenghi: J. Fluid Mech. 391 29 (1999)
7. J. Koplik and H. Levine: Phys. Rev. Lett. 71, 1375 (1993).
8. H.E. Hall and W.F. Vinen: Proc. Roy. Soc. London A 238, 215 (1956).
9. R.N. Hills and P.H. Roberts: Arch. Rat. Mech. Anal. 66, 43 (1977).
10. C.F. Barenghi and C.A. Jones: J. Fluid Mech. 197 551 (1988)
11. C.F. Barenghi: Phys. Rev. B 45, 2290 (1992)
12. C.J. Swanson and R.J. Donnelly: Phys. Rev. Lett. 67 1578 (1991)
13. K.L. Henderson, C.F. Barenghi and C.A. Jones: J. Fluid Mech. 283 329 (1995).
14. K.L. Henderson and C.F. Barenghi: Phys. Lett. A 191 438 (1994).
15. W.F. Vinen: Proc. Roy. Soc. A 240, 114 (1957); ibidem 240, 128 (1957); ibidem
242, 493 (1957); ibidem 243 400 (1957).
16. J.T. Tough: ‘Superfluid turbulence’. In: Progress in Low Temperature Physics, vol.
VIII, ed. by D.F. Brewer (North Holland, Amsterdam 1987) p. 133.
17. D.J. Melotte and C.F. Barenghi: Phys. Rev. Lett. 80 4181 (1998).
14 C.F. Barenghi

18. P.L. Walstrom, J.G. Weisend, J.R. Maddocks and S.V. VanSciver: Cryogenics 28
101 (1988).
19. H. Borner, T. Schmeling and D.W. Schmidt: Phys. Fluids 26 1410 (1983)
20. F. Bielert and G. Stamm: Cryogenics 33 938 (1993)
21. M.R. Smith, R.J. Donnelly, N. Goldenfeld and W.F. Vinen: Phys. Rev. Lett. 71
2583 (1993)
22. J. Maurer and P. Tabeling: Europhysics Lett. 43 29 (1998)
23. M.R. Smith, D.K. Hilton and S. V. VanSciver: Phys. Fluids. 11 751 (1999)
24. W.F. Vinen: Phys. Rev. B 61 1410 (2000)
25. D.C. Samuels and D. Kivotides: Phys. Rev. Lett. 83 5306 (1999)
26. R.G.K.M. Aarts and A.T.A.M. deWaele: Phys. Rev. B 50 10069 (1994)
27. D.C. Samuels: Phys. Rev. B 46 11714 (1992)
28. D.C. Samuels: Phys. Rev. B 47 1107 (1993)
29. C.F. Barenghi, G. Bauer, D.C. Samuels and R.J. Donnelly: Phys. Fluids 9 2631
(1997)
30. D.K. Cheng, M.W. Cromar and R.J. Donnelly: Phys. Rev. Lett. 31 433 (1973)
31. R.M. Ostermeier and W.I. Glaberson: J. Low Temp. Phys. 21 191 (1975).
32. S.P. Godfrey, C.F. Barenghi and D.C. Samuels: Physica B 284 66 (2000); Phys.
Fluids 13 983 (2001).
33. O.C. Idowu, D. Kivotides, C.F. Barenghi and D.C. Samuels: J. Low Temp. Physics,
120 269 (2000).
34. O.C. Idowu, A. Willis, D.C. Samuels and C.F. Barenghi: Phys. Rev. B 62 3409
(2000)
35. D. Kivotides, C.F. Barenghi and D.C. Samuels: Science 290 777 (2000).
36. C.F. Barenghi, D. Kivotides, O. Idowu and D.C. Samuels: J. Low Temp. Physics,
121 377 (2000).
37. M. Leadbeater, T. Winiecki, D.C. Samuels, C.F. barenghi and C.S. Adams: Phys.
Rev. Lett. 86 1410 (2001).
38. D. Kivotides, C. Vassilicos, C.F. Barenghi and D.C. Samuels: Phys. Rev. Lett. 86
3080 (2001).
39. D. Kivotides, C.F. Barenghi and D.C. Samuels: to be published in Europhys. Lett.
(2001).
An Introduction to Experiments
on Superfluid Turbulence

Russell J. Donnelly

Cryogenic Helium Turbulence Laboratory, Department of Physics,

University of Oregon, Eugene, Oregon 97403

1 Introduction
A description of the experimental background of superfluid turbulence was as-
signed to me for this lecture. Superfluid turbulence, or as some call it, quantum
turbulence, has been an active field of physics since the 1950’s. The field was
pioneered experimentally and theoretically by Joe Vinen and as such is approach-
ing a half century in age. It is safe to say that with few exceptions the results
are unknown to those investigators who are interested in classical turbulence,
that is the kind of investigation which has been pioneered by Taylor, Landau,
Kolmogorov and others.
It has only recently been realized that liquid helium I, liquid helium II and
cryogenic (critical) helium gas are attractive candidates for investigating classical
turbulence problems, and in the process many have decided to look at the kinds
of challenges encountered in using helium II, that phase of liquid helium which
exhibits superfluidity.
In preparing this talk I had hoped to cover, however briefly, the entire corpus
of experimental work on the subject. It soon became evident that there was
too much material by far than could be covered in a single one-hour lecture.
Fortunately, I had recently written a review article in honor of Joe Vinen’s
retirement from the University of Birmingham [1] and Skrbek, Niemela and
myself had written a second paper for the same occasion [2]. These papers contain
virtually all the known results on cryogenic fluid mechanics. This talk will instead
concentrate mostly on future directions, and results obtained since the articles
cited above were written.

2 Update on Pipe Flow

Cryogenic helium is of significant value in generating, and studying the highest
possible Reynolds and Rayleigh number flows under controlled laboratory condi-
tions, primarily due to its extremely low value of kinematic viscosity. Pipe flow
is an example where the highest Reynolds numbers can be achieved by using
helium I. Our apparatus is shown in Fig.1. The pipe is a polished stainless steel
tube 4.67 mm in diameter and 25 cm long. The working fluid is drawn into the
bellows by moving the push rod at the top upward. At a preset time the rod
is pushed down at constant velocity and discharges a known mass of liquid per

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 17–35, 2001.

c Springer-Verlag Berlin Heidelberg 2001
18 R.J. Donnelly

Push Rod

Bellows 10 cm
Pump LHe

~
~
P
{

Smooth 25 cm
Pipe

Fig. 1. Bellows driven flow apparatus for measuring the friction factor in a smooth pipe.
The advantage of cryogenics can be appreciated by comparing the size of this apparatus
with the famous apparatus of Nikuradze built in the early 1930’s and weighing many
tons. The highest Reynolds numbers achieved are about the same in both Nikuradze’s
and our apparatus.

unit time. Flow velocity is limited by the power of the drive motor. A sensitive
capacitance manometer has been developed to measure the pressure gradient
in all working fluids and is located 38 diameters down stream of the inlet. The
in-situ differential pressure gauge was specifically designed to avoid mechanical
strains at low temperature, which can overwhelm the 0.05 Pascal resolution of
the device [3].
This experiment, primarily the effort of Chris Swanson of the University of
Oregon, with help from Gary Ihas, visiting from the University of Florida, has
revealed several subtleties in such measurements, including the need for high
purity of helium, fine stability of temperature and pressure, and care to maintain
a time-independent state. Agreement with data at room temperature using air
and water as a fluid is quite good.
The primary advantage of this apparatus is that it accommodates any non-
reactive fluid at a temperature between 1 K and 300 K, allowing unprecedented
ranges of Reynolds numbers to be obtained. For example, we have used room
temperature SF6 , air, and helium gas, as well as cryogenic helium I at 4.2 K
to measure the friction factor from Reynolds number 10 to 5,000,000, nearly
six decades of Reynolds numbers. The fact that the results agree with standard
engineering data from many sources attests to the utility of this new approach.
 An Introduction to Experiments on Superfluid Turbulence 19

Linear Servo Motor

Motor Control
and Counter

16 bit ADC

200MHz Pentium
GPIB, Labview
G
P
I
B
Function Vacuum Tight Sliding Seal
B Generator
U Vacuum
Pump
S

Lock-in
Amplifier

Amplifier
60 mF Grid Generated
Turbulence
Grid

Second Sound
Bias Voltage Receiver
100 Volts DC
Second Sound
Transmitter

Second Sound
LR-110 Standing Wave
Resistance
Bridge
Germanium
Thermometer 1 cm x 1 cm x 29 cm
Channel
Heater
LR-130
Temperature
Controller

Fig. 2. Schematic diagram of the University of Oregon towed grid apparatus in helium
II. Turbulence is generated by sweeping the grid upward, and is observed by measur-
ing the attenuation of second sound in the presence of the turbulence which can be
interpreted to give the vorticity in the superfluid.

It would be of interest to extend these results to helium II, where other

evidence suggests the results will be about the same [4]. However this is yet to
be accomplished in a satisfactory way.
20 R.J. Donnelly

3 Update on Towed Grid Experiments

Superfluid turbulence in counterflow turbulence takes the form of a tangle of
quantized vortex lines, a frictional drag between the moving normal fluid and
the cores of the lines (mutual friction) serving to maintain the turbulence. Coun-
terflow turbulence has no classical analogue. Other types of superfluid turbulence
have been studied over the years, but only recently has there been an experimen-
tal study of superfluid turbulence behind a steadily moving grid. This turbulence
is analogous to the much studied, and fundamentally important, case of homo-
geneous isotropic turbulence created by flow through a grid in a wind tunnel.
The towed grid apparatus, shown in Fig.2, consists primarily of a 1 cm x
1 cm square tube fitted with a sliding grid with about 65 % porosity. The grid
is swept upward at a preset time and creates a slug of turbulence of quite high
vorticity filling the channel. The probe is a pair of second sound transducers
which measure the attenuation of second sound owing to quantized vorticity in
the channel.
The results obtained with this apparatus have been fascinating and a series
of publications have come from this apparatus which show, among other things,
that the decay of the turbulence can most easily be understood treating the decay
as if it were from a classical fluid having the viscosity of the normal fluid, but the
density of the total fluid [5,6,7,8]. The physics underlying this phenomenon has
been carefully discussed by Vinen [9]. His point of view is also further developed
in a lecture at this conference [10].
Two new experiments have been performed which had not appeared in print
at the time of this conference. A brief description of both follows.

3.1 The Nature of Grid Turbulence in Helium II

A detailed experimental and theoretical study of superfluid grid turbulence was
carried out by Stalp et al [8]. As an observational technique they used the at-
tenuation of second sound, which serves to measure the length of vortex line
per unit volume, L, in the superfluid component. They showed that a classical
model could explain the observed time decay of line length. According to this
model superfluid helium behaves as a single fluid with a kinematic viscosity ν  .
The rate of dissipation of turbulent energy per unit mass has the classical form

ε = ν  ω 2  = ν  κ2 L2 (1)

where ω 2  is the mean square vorticity in the coupled fluids, and the identifi-
cation of ω 2  with κ2 L2 is discussed in [9]. Therefore according to the model
second sound measures ω 2 , as a function of time. After an initial time interval
the turbulent energy spectrum has the classical Kolmogorov form

E(k) = C2/3 k −5/3 (2)

within a range of wave numbers ke < k < kη . Energy is being continually

transferred in a cascade from lower to higher wave numbers by inertial forces,
 An Introduction to Experiments on Superfluid Turbulence 21

dissipation by viscosity taking place eventually at k ≥ kη = (ε/ν  )1/4 (the

3

Kolmogorov wave number). C is the Kolmogorov constant and is equal to about

1.5 for a classical fluid, and ke is the wave number of the energy-containing
eddies. ke (t) decreases until it saturates at a value equal to approximately the
inverse of the channel width.
The model made predictions that were in agreement with the experiments
reported in [5,6,7] in the temperature range studied, from the λ-temperature
down to about 1.3K; however, although ν  was shown to be of order μ/ρ, where
μ is the viscosity of the normal fluid and ρ is the total density of the helium, no
careful analysis of the temperature dependence of ν  was carried out.
It turns out that within the inertial range of wave numbers (ke < k < kη )the
relevant length scales are significantly larger than the expected spacing of quan-
tized vortex lines, which turns out to be close to kη −1 . Mutual friction is then
sufficient to keep the two fluids locked together on these length scales, the locked
fluids behaving as a single turbulent fluid in which there is negligible dissipa-
tion. Therefore, a Kolmogorov spectrum forms, as in a classical fluid, provided
of course that energy is dissipated by some means at or beyond the Kolmogorov
wave number kη . The model requires that this dissipation be described by equa-
tion (1). The fact that the vortex line spacing is of order kη −1 implies that in
reality the two fluids cannot remain locked together for wave numbers of order
or greater than kη . It follows that dissipation must be due to a combination of
viscous dissipation in the normal fluid and mutual friction, and it is therefore
not obvious that an equation of the form (1) should hold [9]. It was predicted
in [9] that the resulting rate of dissipation would be given by a formula similar
to (1), with a value of ν  that is weakly dependent on temperature and of order
μ/ρ as measured above 1 K.
We conclude that energy dissipation in grid turbulence is likely to be de-
scribed correctly by (1), in spite of the fact that the mechanism of energy dissi-
pation in the superfluid is quite different from that in a classical fluid. The precise
value of ν  depends on the dissipative mechanisms operating in the turbulence.
These mechanisms operate on a length scale of order the vortex line spacing,
and are therefore strongly influenced by quantum effects in the superfluid.
We see that the parameter ν  , which is analogous to the kinematic viscosity
in a classical fluid, is fundamentally important in superfluid turbulence. The
experimental technique was described in [5,6,7]. Data analysis is carried out for
that period of the decay during which the vortex line density falls as t−3/2 , when
the theory of the classical model gives
d
L(t) = √ (3C)3/2 t−3/2 (3)
2πκ ν 
where d is the width of the channel through which the grid is towed. At times
earlier than this period the scale of the energy-containing eddies has not yet sat-
urated at the channel width (see section 3.2 below) and the decay is slower.
At later times the inertial range of wave numbers becomes narrow and the
22 R.J. Donnelly

-4
2 .0 x 1 0

-4
1 .5 x 1 0
/s )
2
ν ' (c m

-4
1 .0 x 1 0

ν (T )

-5
5 .0 x 1 0
1 .0 0 1 .2 5 1 .5 0 1 .7 5 2 .0 0 2 .2 5

T e m p e ra tu re (K )

Fig. 3. The effective kinematic viscosity as a function of temperature, as measured

from experiment.

decay rate increases. The length of line is obtained from second sound resonance
measurements using the relation
16Δ0 A0
L(t) ≈ ( − 1) (4)
κB A(t)
where Δ0 is the full-width at half maximum for the second-sound (power) res-
onance curve in the absence of vortex lines (at the 50th harmonic), B is the
mutual friction constant obtained from measurements on uniformly rotating he-
lium [10], and A(t) and A0 are respectively the peak second sound amplitude
with and without vortex
 lines present. Using (4) and (5) we see that the ex-
periment measures C 3 /ν  . Note that Maurer and Tabeling [11] performed an
experiment with rotating disks in liquid helium in which they were able to mea-
sure the Kolmogorov spectrum by means of pressure fluctuations using a probe
inserted at a selected location in the flow and connected to a quartz pressure
transducer. Their results showed that the Kolmogorov spectrum could be ob-
served in both helium I and helium II (2.3K, 2.0K and 1.4K) and reached the
important conclusion that the Kolmogorov constant C is the same above and
below the lambda transition. With that information, we see that our experiment
determines ν  providing we know C. The Kolmogorov constant is taken to be
1.5 at all temperatures, which is the accepted classical value [12]. The resulting
values of ν  are shown as a function of temperature in Fig. 3. The error bars
 An Introduction to Experiments on Superfluid Turbulence 23

Fig. 4. The decaying helium II vorticity measured at T=1.3 K for the indicated ReM .

include contributions due to uncertainties in the values of B, and the frequency-

dependence of B was taken into account [13].
The experimental results reported here relate to temperatures greater than
1.1 K. There is a clear need to extend the temperature range downwards, es-
pecially because the theory then makes interesting predictions. The only exper-
iments so far reported on superfluid turbulence at these low temperatures are
those of Davis, Hendry and McClintock [14].

3.2 Four Regimes of Decaying Grid Turbulence in Helium II [15]

Attenuation of second sound in helium II has been used to observe up to six

orders of magnitude of decaying vorticity displaying four distinctly different
regimes of decaying grid turbulence in a finite channel. A purely classical spectral
model for homogeneous and isotropic turbulence describes most of the decay of
24 R.J. Donnelly

helium II vorticity in the temperature range 1.2K < T < 2K. The four regimes
switch successively as the energy containing and dissipative Kolmogorov length
scales gradually grow during the decay, finally both being saturated by the size of
the channel. In Fig.4 each curve represents an average of three individual decays.
As the decay curves tend to collapse on the universal curve, we shifted them for
clarity by a factor of two downwards, the uppermost remaining unchanged. The
early part of the vorticity decay displays a power law with exponent -11/10 (see
left inset, showing normalized data forReM = 104 ) and later −5/6 (see right
inset, showing normalized data forReM = 1.5 × 105 , 2.5 × 104 and 5 × 103 ). After
saturation, typically several orders of magnitude of decaying vorticity closely
follow the power law with exponent −3/2, represented by the thick solid line.
For the first time we report the fourth and last regime - a late exponential
decay - not shown separately in Fig.4. The nature of this last decay is not
understood at present.
This experiment explores over 8 orders of magnitude of decaying turbulent
energy, an impossible task for a wind tunnel, which would need to have a 1000
km test section to observe the same thing.

4 Agenda for the Future

4.1 The University of Oregon 6 cm Wind Tunnel

So far we have not had a continuous flow facility to work on. About two years ago
we decided to build a small wind tunnel using critical helium gas as the working
fluid. The advantage of critical helium gas is the enormous range of properties
which can be reached by adjusting the pressure and temperature. Flow velocities
available depend on the density. At low densities flow velocities up to 1m/s are
possible, at high densities velocities to 30cm/s are available. One of the optimal
operating points for high Re flow will be at 4 bar pressure and 6 K at which
the kinematic viscosity is 3.1 × 10−8 m2 /s. Mesh Reynolds numbers of 30,000 to
100,000 will be available, and corresponding microscale Reynolds numbers will
range from 150 to 280.
We have recently successfully operated this tunnel at 6.5 K and mesh Reynolds
numbers around 1500. The grid generated turbulence is probed with 10 micron
diameter cryogenic hot wire anemometer also developed in our laboratory. These
hot wires are observed to obey King’s law relating velocity and voltage, and pre-
liminary velocity time series show standard statistical features. Note that this is
the first cryogenic tunnel to operate below liquid nitrogen temperature.

4.2 Wind Tunnels for Model Testing

We have given considerable attention to the conceptual design of larger wind tun-
nels designed to be useful for model testing. These have been discussed recently
in [2] and do not need repeating in this article. The importance for superfluid
turbulence of these devices is the possibility of operating them in helium II.
 An Introduction to Experiments on Superfluid Turbulence 25

Fig. 5. Sketch of the University of Oregon 6 cm wind tunnel.

The acoustic characteristics of models being tested in such a tunnel is very

important, and unexplored at the present time. Acoustics is a very large field
having to do with the generation and interaction of sound with mechanical struc-
tures. It is of particular importance for submarine testing.

4.3 Tow Tanks

Liquid helium offers much promise for tow tank design, a subject also reported
in some detail in [2]. Again the importance for superfluid turbulence is the pos-
sibility of operating with helium II as the working fluid.
26 R.J. Donnelly

5 Challenges for the Future

5.1 The Challenge of Instrumentation

Although helium I is a Navier-Stokes fluid, the possibility of generating ever

higher Reynolds numbers carries with it ever decreasing Kolmogorov lengths. Of
primary concern is the smallness of the flow structures such as turbulent eddies
and boundary layers. Consider flow through a smooth pipe at high Reynolds
number. To be definite let us take a pipe 10 cm in diameter using liquid helium
having a kinematic viscosity ν = 2×10−4 (cm)2 flowing at a mean velocity u , and
density ρ = 0.146gm/(cm)3 . For turbulent flow a viscous sublayer is generated
at the walls
 which scales with the friction velocity . The friction velocity is given
by u∗ = τ /ρ , where τ , the wall stress, is τ = λρu2 /8 and λ is the empirical
friction factor [16]. The corresponding length scale is
ν
y= (5)
u∗
and the viscous sublayer has a thickness of a few times y. This y can be thought
of as the scale of the smallest eddies in the turbulent flow. Table 1 outlines
some typical values for helium I. Note that in helium I we can in principle use
velocities of flow up to the velocity of sound before encountering shock waves.
We cannot use such large velocities in helium II because second sound shock
waves will intrude at about an order of magnitude lower velocity. The speed of
sound in the gas is lower than in the liquid, hence the highest Reynolds number
generated with helium should be generated with helium I.
It can be seen that the length scales can range from microns to Angstroms.
For comparison with theories it is important to be able to probe the flow at the
smallest scales. The last entry for y in table 1 is 42 Angstroms. This should be
compared with the cube root of the atomic volume of liquid helium which is 3.6
Angstroms. Kolmogorov lengths in the Angstrom ranges, therefore, can approach
conditions where the continuum hypothesis for the Navier Stokes equation is
going to be in question. Probing such flows will be a major instrumentation
challenge. We must learn to build not only micron sized, but even nanometer
sized transducers in order to take advantage of the new range of high Reynolds
numbers afforded by cryogenic helium.
Another experimental challenge is to visualize the normal fluid. This problem
has always been with research with helium II, and not a great deal of progress has
been made except in specialized cases [18]. Perhaps the most promising avenue
for visualization will be PIV (mentioned below).
We now summarize briefly the diagnostic tools available for cryogenic fluid
mechanics. These are discussed in a recent review [17].
• The measurement of average flow velocity can be measured in many instances
by means of acoustic flowmeters or venturi tubes.
• Flow velocity at a point can be measured in convecting helium gas, for
example, by means of bolometer fluctuation correlations.
 An Introduction to Experiments on Superfluid Turbulence 27

• Hot wires, Laser Doppler Velocimeters (LDV) and Particle Image Velocime-
ters (PIV) should all work in principle in helium gas, helium I and helium
II.
• RMS vorticity can be measured in helium II by second sound attention in
both open flows and counterflows. There is some speculation that chemical
potential probes would give local information on vorticity fluctuations, but
this has yet to be implemented [17].
• Temperature gradients can be measured by means of standard germanium
thermometry and pressure gradients, for example, by means of capacitance
manometers [3].
• Lift and drag on models can best be measured using Magnetic Suspension
and Balance Systems (MSBS), discussed in [18] by Britcher. Of course in a
cryogenic environment the magnets can be superconducting.
• Wall stress gages can be fabricated for work in helium gas and helium I.
• Ion trapping can be used to measure vorticity in helium II [25].

Table 1. Viscous Sub Layer Thickness at Various Reynolds Numbers

u Re λ τ u* y
(cm/sec) erg/cm3 cm/sec (cm)
10 5 × 105 0.0132 0.0240 0.405 4.93 × 10−4
102 5 × 106 0.00898 1.64 3.35 5.97 × 10−5
103 5 × 107 0.00649 118.4 28.48 7.02 × 10−6
104 5 × 108 0.00489 8,940 247 8.08 × 10−7
2 × 104 1 × 109 0.00453 33,100 476 4.20 × 10−7

Hot wire anemometers are resistive self-heating devices which balance heat lost
to flow, which depends on the fluid velocity. Calibration provides the correlation
between fluid velocity and electrical power supplied. Standard hot wires have
d ∼ 5 microns and L ∼ 1000 microns. Standard materials (e.g. platinum) used at
room temperature are insensitive at low temperature. Our cryogenic hot wires are
made on a quartz fiber of diameter 10 microns. They consist of an evaporated Au-
Ge film of thickness of order 3000Å. A small sensitive region in middle of the fiber
is defined by masking the fiber and evaporating a metallic film over it as shown
in Fig.7. Masking is achieved by laying a small diameter fiber perpendicular to
the first fiber. The metal film provides electrical contact to the sensitive region.
Figure 8 shows how the small sensors are mounted. Sensor support dimen-
sions follow the “rule of 10”: they are placed ten times (more or less) their
characteristic dimension away from the sensitive region. Electrical contact to
the fiber is made through stainless steel wires, which are isolated from the brass
ring by epoxy. The fiber is epoxied to the wires using an electrically conducting
epoxy.
We have recently tested a few prototype 10 micron sensors in the wind tunnel
of Fig.5. A helium gas flow at 7 K was generated and we applied a relatively large
sinusoidal voltage to overheat one of the sensors and measured the temperature
28 R.J. Donnelly

Fig. 6. A hot wire anemometer.

Fig. 7. Principle of the cryogenic hot wire anemometer developed at the University of
Oregon.

of a nearby sensor. The characteristic heat signature at twice the input frequency
was clearly seen on the monitoring sensor. Furthermore, we found that the signal
to noise ratio to be the same as that predicted by our noise analysis. Thus we
feel confident that the sensors are working according to our expectations.
 An Introduction to Experiments on Superfluid Turbulence 29

Fig. 8. Mounting the 10 micron hot wire anemometer

We have made substantial strides in detector development. Our 10 micron

detectors are to be contrasted with the bolometers we use in our large convection
apparatus, which are 250 microns on side. But as we see from Table 1, we need
to push our detectors down in size by orders of magnitude. Such advances are
likely to depend on optical techniques.

5.2 Challenges for Understanding Counterflow Turbulence

Counterflow experiments have a history of many decades, but much remains to

be accomplished [1]. The brief outline below suggests some directions needing
exploration.

Comparing the decay of towed grid turbulence and counterflow tur-

bulence. We need to understand the differences between the decay of towed
grid turbulence and the decay of counterflow turbulence after the counterflow
is turned off. Fig.9 below shows decay of both types of turbulence in the same
apparatus, beginning at about the same vorticity and demonstrating nearly iden-
tical decay at long times. The experimental channel was identical in both cases
except the grid was removed for the counterflow experiment. The form of the
upper curve has been discussed by Schwarz and Rozen [19] and by Smith [5].
The decay curves coincide at long times and correspond exactly to the decay of
classical turbulence (see Section 2.2 above).

The first critical velocity. What accounts for the first critical velocity in
counterflow and its temperature dependence? This phenomenon is likely a case
of extrinsic nucleation, but lacks any quantitative explanation.
30 R.J. Donnelly

Fig. 9. An illustration of the dramatic differences in decay of two turbulent flows of the
same initial vorticity produced by a counterflow (upper curve) and towed grid (lower
curve).

The TI-TII transition. The TI-TII transition needs further attention. We

have Melotte and Barenghi’s insight which suggests that TI corresponds to tur-
bulence in the superfluid, and TII marks the onset of disturbances in the normal
fluid. Their linear stability analysis gives the right order of magnitude for the
transition, but does not capture the temperature dependence.

The Tough classification. Why do large and small aspect geometries in coun-
terflow exhibit such different behavior?

Combined rotation and heat flow. Combined rotation and heat flow is a
relatively new area of investigation. Prior to the investigation discussed in this
 An Introduction to Experiments on Superfluid Turbulence 31

section it was assumed (from earlier experiments) that the ordered array of vor-
tex lines produced by steady rotation and the disordered tangle produced by
counterflow preserved their identities in a combined experiment. Measurements
at Oregon with improved sensitivity by Barenghi, Swanson and Donnelly [20]
showed that the picture just described is far from true. The observations con-
sisted of measuring the amount of vortex line present owing to counterflow or
rotation alone using second sound attenuation, and comparing the observed line
density with what would be expected if the two sources of vorticity simply added.
The results are complicated, but appear to be relatively simple in two limits:
(i) Limit of large line densityLH due to heat, slow rotation. Here the effect of
rotation is not simply to add line densityLR = 2Ω/κ. Instead the tangle appears
to be polarized to accomplish the rotation. The effective polarization increases
with rotation Ω by analogy to a gas of magnetic dipoles in a magnetic field.
The results scale with LR /LH by analogy to μH/kT . Thus rotation appears to
produce alignment in the tangle, as does a magnetic field for dipoles, and LH
appears to play the role of disordering heat bath in the statistical mechanics of
superfluid turbulence. Indeed, it takes very little polarization of a dense tangle
to produce rotation at the relatively small angular velocities of the apparatus.
(ii) Limit of fast rotation and small axial heat flux. Any rotation eliminates
the critical velocity vc . In this limit two critical counterflow velocities appear,vc1
1
and vc2 , which scale as Ω 2 .
We might speculate here that the first critical velocity appears to correspond
to the Donnelly-Glaberson instability, excitation of helical waves by the coun-
terflow on the vortex lines induced by rotation [1]. The second appears to be
a transition to turbulence, with the rotation-induced array becoming a vortex
tangle. A more formal investigation of these effects is likely to be rewarding.

5.3 Challenges for Understanding Periodic

Boundary Layer Experiments
The decay of torsional oscillations of various pendulums was a veritable ”cottage
industry” in low temperature physics in the 1950’s. The overall situation was
reviewed by Donnelly and Hollis Hallett in 1958 [21]. The situation can be ap-
preciated by looking at the damping of oscillations of liquid helium in a U-Tube,
as reported by Donnelly and Penrose [22] and reviewed in [21].
Figure 10 summarizes the data. In helium I the damping is independent
of amplitude up to a critical amplitude (or velocity) called ht . In helium II,
however, the results show damping accounted for by the Landau two fluid model
below a temperature-dependent critical amplitude called hc , a steady increase
in damping up to an amplitude hn , and a further critical amplitude, also named
ht . The region of damping between hn and ht seems to be accounted for by the
Navier-Stokes equation. The authors assumed that the region between hn and ht
was characterized by an increasing coupling between the two fluids. The details
of these observations are unclear even today.
What causes the two fluid model to break down at a critical amplitude hc ?
What happens at ht ? What would happen if experiments (including the pendu-
32 R.J. Donnelly

Fig. 10. The variation with amplitude of the damping of gravity oscillations of liquid
helium in a U-tube at a period of 0.94 sec. After Donnelly and Penrose [22] and Donnelly
and Hollis Hallett [21].

lums) were ramped up in amplitude instead of starting suddenly at high ampli-

tude and allowed to decay. Oscillating cylinders and spheres are likely best for
such novel experiments.
 An Introduction to Experiments on Superfluid Turbulence 33

Fig. 11. Apparatus constructed by Davis, Hendry and McClintock [14] to measure the
decay of superfluid turbulence at 70 mK.

5.4 Instrumentation to Detect Vortices Below 1 K

Apparatus constructed by Davis, Hendry and McClintock [14] is the only one
to address the question of the nature of superfluid turbulence in the absence
of normal fluid experimentally. They arranged to generate turbulence by means
of an oscillating grid structure, and to examine it by trapping of negative ions
coming from a field emission tip on the vortex tangle. Their results indicate a
temperature independent decay process. The absolute line density of the vortices
remains unknown.
Second sound will not work when the temperature is too low because the
normal fluid has disappeared. Ion capture may be difficult because of lack of
knowledge of the dynamics of capture at very low temperatures, and experimen-
tal evidence that the capture cross section is decreasing at lower temperatures
[23]. Samuels and Barenghi [24] have suggested that there is enough heat energy
in a vortex tangle to be measured calorimetrically when the tangle has decayed.
Another possibility might be to design some sort of phonon (or roton)detector
fast enough to follow the decay of turbulence from a towed or oscillating grid.

5.5 The Normal Fluid and the Vortex Tangle

Understanding the relationship between the normal fluid and the vortex tangle
was an important topic of discussion at this conference. Theoretical progress on
34 R.J. Donnelly

the self-consistent interaction between the normal fluid and quantized vortices
will likely lead to the need to develop measurement techniques which will give
more information than just the RMS line density L.

5.6 Flow over Blunt Objects, Testing Models such as Submarines

Turbulence can easily be generated by high Reynolds number flows over sim-
ple objects like cylinders and spheres. Here the flows are shear flows and are not
homogeneous and isotropic. Indeed the whole subject of the flow over experimen-
tal models is mostly in the future. However Van Sciver has reported preliminary
measurements of flow over a sphere in a companion article in this volume [24].
We look forward to more complete measurements and interpretation as this work
progresses.

Acknowledgements
First of all I am grateful to Keith Moffatt, Director of the Newton Institute, for
the opportunity to have this remarkable gathering here. I am indebted to Joe Vi-
nen, Steve Stalp, Ladislav Skrbek, Carlo Barenghi, David Samuels, Renzo Ricca
and Peter McClintock for many useful discussions. My research is supported by
the National Science Foundation under grant DMR-9529609.

References
1. R. J. Donnelly, J. Phys Condensed Matter, 11, 7783-7834 (1999).
2. L. Skrbek, J. J. Niemela and R. J. Donnelly, J. Phys Condensed Matter, 11, 7761-
7782 (1999).
3. Chris J. Swanson, Kris Johnson and Russell J. Donnelly, Cryogenics 38, 673-677
(1998).
4. P. L. Walstrom, J. G. Weisend II, J. R. Maddocks and S. W. Van Sciver, Cryogenics
28, 101 (1988).
5. M.R. Smith, R. J. Donnelly, N. Goldenfeld and W.F. Vinen, Physical Review
Letters 71, 2583 (1993).
6. M.R. Smith “Evolution and Propagation of Turbulence in Helium II”. PhD Dis-
sertation, University of Oregon (1992).
7. S. Stalp, “Decay of Grid Turbulence in Superfluid Helium”. PhD Dissertation,
Physics, University of Oregon (1998).
8. S.R. Stalp, L. Skrbek, R.J. Donnelly, Phys. Rev. Lett., 82, 4831 (1999)
9. W. F. Vinen, Phys. Rev. B61, 1410 (2000)
10. W. F. Vinen “An Introduction to the Theory of Superfluid Turbulence” Paper in
this volume
11. J. Maurer and P. Tabeling, Europhys. Lett., 43, 29 (1998).
12. K.R. Sreenivasan, Phys. Fluids, 7, 2778 (1995).
13. C.F. Barenghi, R.J. Donnelly, W.F. Vinen, J. Low Temp. Phys., 52, 189 (1983).
14. S. I. Davis, P. C. Hendry and P. V. E. McClintock, Physica B , 280, 43 (2000).
15. L. Skrbek, J. J. Niemela and R. J. Donnelly, Phys. Rev. Lett., 85, 2973 (2000).
16. H. Schlichting, Boundary Layer Theory, 7th ed, McGraw-Hill (1979).
 An Introduction to Experiments on Superfluid Turbulence 35

17. C. F. Barenghi C. E., Swanson, and R. J. Donnelly, J. Low Temp. Physics, 100,
385 (1995).
18. R. J. Donnelly (Editor), Liquid and Gaseous Helium as Test FluidsSpringer-
Verlag, 1991). In High Reynolds Number Flows Using Liquid and Gaseous He-
lium(Springer-Verlag, 1991). Also R. J. Donnelly and K. R. Sreenivasan (Editors)
Flow at Ultra-High Reynolds and Rayleigh Numbers(Springer-Verlag, 1998)
19. K. W. Schwarz, and J. R. Rozen, Phys. Rev. B , 44, 7563 (1991).
20. C. F. Barenghi C. E., Swanson, and R. J. Donnelly, Phys. Rev. Lett. , 50, 190
(1983).
21. R. J. Donnelly and A. C. Hollis Hallett, Annals of Physics, 3,320 (1958).
22. R. J. Donnelly and O. Penrose, Phys. Rev. , 103, 1137 (1956).
23. R. M. Ostermeier and W. I. Glaberson, Phys. Lett. 49A, 223 (1974).
24. D. Samuels and C. Barenghi, Phys. Rev. Lett. 81, 4381,1998.
25. R. J. Donnelly,Quantized Vortices in Helium II,(Cambridge University Press,
1991).
26. S. W. Van Sciver, “The Drag Crisis on a Sphere in Flowing HeII” Paper in this
volume.
The Experimental Evidence for Vortex
Nucleation in 4He

Éric Varoquaux1 , Olivier Avenel2 , Yury Mukharsky2 , and Pertti Hakonen3

1
CNRS–Université Paris-Sud, Laboratoire de Physique des Solides,
Bâtiment 510, F-91405 Orsay Cedex, France
2
Commissariat à l’Énergie Atomique, Service de Physique de l’État Condensé,
Bâtiment 772, Centre de Saclay, F-91191 Gif-sur-Yvette Cedex, France
3
Low Temperature Laboratory, Helsinki University of Technology,
02150 Espoo, Finland

Abstract. This update on the problem of vortex nucleation in superflows through

micro-apertures follows the recent reviews of the subject by Zimmermann [1] and one
of the authors [2]. Recent developments of the model of vortex nucleation involving a
vortex half-ring are assessed with an emphasis on the statistical properties of the critical
velocity transition. The occurrence of collapses and multiple slips is then discussed in
relation with the problem of vortex nucleation.

1 Single Vortex Nucleation

Thermally assisted vortex nucleation. When superflow is forced through
a micro-aperture and the critical velocity is reached, energy dissipation occurs
by phase slippage. Phase slips by 2π take place according to a scenario in which
1
/2 –rings nucleate at the wall of the aperture, at a site where the critical velocity
vc is exceeded [3,4,5,6]. A remarkable feature of the critical velocity for phase
slips is its dependence on temperature, which is very nearly a straight line from
about 2 K down to ∼ 0.15 K:
vc = v0 (1 − T /T0 ) . (1)
The value of T0 varies somewhat with the micro-aperture but lies in the vicinity
of 2.5 K. A sample of the experimental data obtained in various laboratories is
shown in Fig. 1.
Since superfluid 4 He is nearly fully in its ground state below 1 K, such a
large temperature dependence gives a strong indication that an Arrhenius-type
process must be taking place. The end products of this nucleation process are
vortices because phase slips by 2π are observed [7]. The point of view taken here
is that these vortices are nucleated (i.e., created ex nihilo). However, it has been
suggested that solitons [8] or bubbles [9] can appear as intermediate states in the
nucleation process. Other mechanisms based on pure fluid-mechanical motion
such as vortex mills [10,11], which will be discussed in Sect.2, are difficult to
reconcile with the large temperature dependence of vc below 1 K.
Thermal nucleation implies the existence of an energy barrier to the spon-
taneous formation of vortices which is overcome with the help of thermal fluc-
tuations. As it can be surmised that a critical velocity would be reached even

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 36–50, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 The Experimental Evidence for Vortex Nucleation 37

Fig. 1. Critical velocity, normalised to the zero temperature linear extrapolation value
v0 , versus T , in Kelvin: () [5], () [1], for ultra-pure 4 He. The curves are computed
from the half-ring model for a0 = 2.2, 3.2, 4.5, 6.0 Å and are normalised to match the
experimental value at 0.5 K. The inset shows the influence of 3 He impurities on vc , (◦)
3 ppb, () 45 ppb.

in the absence of thermal (or quantum, as discussed below) fluctuations, this

energy barrier vanishes for a finite value vc0 of the critical velocity. The hydro-
dynamic instability threshold vc0 at which vortices appear spontaneously has
been shown to occur in numerical simulations of flows past an obstacle using the
Gross-Pitaevskii equation by Frisch et al. first [12] and by others [13,14]. Close
to the point where it vanishes, i.e. for v vc0 , a well-behaved energy barrier
takes the form [5]:
 3
2 v2 2
Ea = E J 1 − 2 , (2)
3 vc0
where EJ depends on temperature and pressure.
The nucleation rate for thermally activated process is given in terms of the
activation energy by Arrhenius’ law:

ω0 Ea
Γ = exp − , (3)
2π kB T

ω0 /2π being the attempt frequency. In experiments performed in a Helmholtz

resonator, such as those shown in Fig. 1, the velocity varies periodically at the
resonance frequency as vp cos(ωt) and the probability that a phase slip takes
38 É. Varoquaux et al.

place during the half-period ωti = −π/2 , ωtf = π/2 is

  tf
p = 1 − exp − Γ (P, T, vp cos(ωt ) dt
ti

ω0 −2πkB T Ea
= 1 − exp − exp − . (4)
2πω vp ∂Ea / ∂v|t=0 kB T

Eq.(4) stems from an asymptotic evaluation of the integral at the saddle point
t = 0. The accuracy of the asymptotic evaluation (4) becomes questionable as
T → 0 where the energy barrier vanishes. It has been checked by direct numerical
integration for typical cases and found to be quite satisfactory.
The critical velocity vc is defined as the velocity for which p = 1/2. This
definition is independent of the experimental setup, except for the occurrence in
(4) of the natural frequency of the Helmholtz resonator ω. The implicit equation
for vc reads:

ω0 −2πkB T Ea (P, T, vc )
exp − = ln 2 . (5)
2πω vc ∂Ea / ∂v|vc kB T

If Ea is given by (2), (5) can be solved analytically [5]:

  23 1
2
3 kB T
vc = vc0 1− γ , (6)
2 EJ

where, with logarithmic accuracy, γ ln(0.1ω0 /ω). It is immediately apparent

that the empirical linear dependence of vc on T , (1), is not well reproduced by
(6); this dependence could be satisfied by the following functional form for Ea ,

Ea = Ee (1 − v/vc0 ) , (7)

which is quite different from (2).

Thus, the temperature dependence of vc , (1), strongly suggests the existence
of a thermally activated nucleation process, but a well-behaved energy barrier
such as (2) leads to too fast an increase of vc as T → 0.

Statistical width of the critical transition. The velocity vc of each critical

event is a stochastic quantity. Its statistical spread can be characterised by the
‘width’ of the probability distribution defined [5,15] as the inverse of the slope of
−1
the distribution at vc , (∂p/∂v|vc ) . This critical width is found to be expressed
by:
    −1
2 1 1 ∂ 2 Ea  ∂Ea  1 ∂Ea 
Δvc = − + + . (8)
ln 2 2 vc ∂v 2 vc ∂v vc kB T ∂v vc
 The Experimental Evidence for Vortex Nucleation 39

Fig. 2. Vortex nucleation rate and phase slip probability in terms of the mean velocity
in the micro-aperture expressed as phase winding numbers [16] in 4 He at 12 mK, 0.6
bar, with 100 ppb of 3 He impurities: frame (a) top, time spent, in seconds (, left
scale), and number of phase slips (◦, right scale) per velocity bin of size 0.1; frame
(a) bottom, slip rate in s−1 as a function of flow velocity, in units of 2π; frame (b),
cumulative probability obtained from the histogram of the slip velocities. The plain
curve is a least square fit to the data with the functional form corresponding to (4).
The value p = 1/2, shown by the arrow, defines the critical velocity (here, vc = 59.68).

At low temperatures and large critical velocities, the quantity in curly brack-
ets in the right hand side of (8) is small with respect to the last term so that the
 −1
width is simply expressed as Δvc = −(2/ln 2) kB T ∂Ea /∂v|vc . Thus, the
statistical width is an approximate measure of the inverse of the slope of Ea in
terms of v.
This quantity is derived from p, itself obtained by integrating the histogram
of the number of nucleation events ordered in velocity bins. This procedure is
illustrated in Fig. 2: p shows an asymmetric-S shape characteristic of the double
exponential dependence of p on v, (4), a consequence of Arrhenius’ law, (3),
being plugged into a Poisson probability distribution. The observation of this
asymmetric-S probability distribution constitutes another experimental clue for
the existence of a nucleation process.
The critical transition of vc displays a measurable width, shown in Fig. 3,
which implies (as does the T -dependence of vc ) that the energy barrier is neither
very high compared to kB T nor very steep in terms of v. Fig. 3 contains data from
several groups. These data are somewhat scattered, especially above 0.5 K, but
they do show overall agreement within this scatter. They provide experimental
input on ∂Ea /∂v|vc and put tight limits on the theoretical models discussed
below.
It is also possible to measure the nucleation rate Γ directly as a function
of v. This quantity is the ratio, for a given velocity bin, of the number of slips
which have occurred at that velocity to the total time spent by the system at
the same velocity. The procedure is illustrated in Fig. 2. The slope of Γ (v) yields
∂Ea /∂v|vc ; the value of Γ at vc gives a combination of ω0 and Ea (vc ). Non-linear
40 É. Varoquaux et al.

Fig. 3. Statistical width of the critical velocity transition, normalised to the linear
extrapolation limit at T = 0, v0 , in terms of temperature: () [5], () [1], (×) [17] .

fits of p to (4) give estimates of ω0 , EJ and vc /vc0 which are independent of the
specific features of a given model. These estimates compare well with those of
the more precise analysis described below.

Quantum tunnelling. Below 0.15 K, vc ceases abruptly to vary with T , as

seen in Fig. 1. For ultra-pure 4 He (less than 1 part in 109 of 3 He impurities),
vc (T ) remains flat down to the lowest temperatures (∼ 10 mK). The crossover
from one régime to the other is very sharp. At the same crossover temperature
Tq , Δvc also levels off sharply. It is believed on experimental grounds that this
saturation is intrinsic and is not due to parasitic interferences [5].
If the nucleation barrier were undergoing an abrupt change at Tq , for instance
because of a bifurcation toward a vortex instability of a different nature [8], in all
likelihood Δvc would jump to a different value characteristic of the new process
(presumably small since vc reaches a plateau). Such a jump is not observed
in Fig. 3. Furthermore, vc levels off below Tq , which would imply through (5)
that Ea becomes a very steep function of v, but Δvc also levels off, which,
through (8), would imply the contrary. We are led to conclude [18] that, below
Tq , thermally-assisted escape over the barrier gives way to quantum tunnelling
under the barrier. In which case the attempt frequency is related to the crossover
temperature by
ω0 = 2πkB Tq . (9)
Relation (9) holds for barriers of the form (2) even in the presence of moderate
damping [19].
 The Experimental Evidence for Vortex Nucleation 41

In the quantum régime, the tunnelling rate for a cubic-plus-parabolic po-

tential well [20] can be used in place of (3) to compute vc and Δvc . The cor-
responding formulae are obtained by the following substitution in (3), (5) and
(8):
 1
ω0 ω0 Ea 2 Ea 36 Ea
a) =⇒ 864π , b) =⇒ .
2π 2π ω0 kB T 5 ω0
This procedure yields the smooth, if rapid, transition in vc and Δvc observed
at Tq . The value of ω0 given by (9) (ω0 /2π = 2 × 1010 Hz for Tq =0.147 K) is
consistent with the attempt frequency appropriate to the thermally-activated
régime [21] and that found directly from the fits to the probability p as shown
in Fig. 2. Furthermore, it agrees (for a0 = 5 Å) with the eigenfrequency of the
highest Kelvin mode that a vortex filament in 4 He can sustain, ω + = κ/πa20 =
ω0 where κ is the quantum of circulation. Thus, assuming vortex nucleation
by quantum tunnelling below Tq is fully consistent with the thermally-assisted
nucleation régime that prevails above Tq .

The 1/2 –ring model. Described in full in [5], this model has a long history,
following the work of Volovik [22], itself based on the theory of homogeneous
nucleation of vortices by Iordanski, and by Langer and Fischer [23]. It has been
very successful in accounting for vortex nucleation by ions [24]. In essence, the
model is based on the hydrodynamics of an Eulerian fluid assumed valid to scales
of the order of the vortex core diameter a0 , that is down to atomic size in 4 He.
Nucleation is assumed to occur on an asperity on the walls of the micro-aperture
where i) the flow velocity is largest, ii) the superfluid density is depressed. The
perturbed volume over which nucleation takes place is, on heuristic grounds,
of the order of a30 . The radius of the nucleated half-ring when it escapes to
the bulk fluid turns out of the order of 15 Å. As a consequence, the size and
shape of the asperity can be neglected in a first approximation; the asperity
only serves the purpose of breaking translational invariance. The bases of the
1
/2 –ring model have been re-examined critically by Sonin [25] and by Fischer
[26]. It can also be mentioned at this point that Nore et al. [27] have shown by
numerical simulations of 3D flows past an obstacle in the Gross-Pitaevskii model
that vortex filaments do evolve spontaneously into one, or possibly a few, half-
rings which thus constitute the preferred configuration of small vortices close to
walls.
The 1/2 –ring model is amenable to an expansion of the form (2) but the full
expression of Ea can straightforwardly be evaluated numerically and substituted
into the expressions of vc and Δvc , (5) and (8) [5]. The energy barrier impeding
vortex nucleation is found to vanish at
κ
vc0 = 0.432 . (10)
2πa0
This value, obtained for an Eulerian fluid, is to be compared with that obtained
by Rica for the Gross–Pitaevskii model, vc0 = 0.262κ/2πa0 [28], comparison
which points up quantitative inadequacies in the 1/2 –ring model outlined above.
42 É. Varoquaux et al.

Comparison with experiment and discussion. The outcome of the numer-

ical evaluation of vc (T ) and Δvc (T ) is plotted in Figs. 1 and 3 for various values
of the core parameter a0 which is the only free parameter in the problem. The
temperature dependence of a0 and ρs is taken from refs.[29] and [30] respectively.
In the bulk, the low temperature value of a0 1Å . At a wall boundary, it is
at least of the order of the static penetration depth (2.5 Å) [29], and probably
rather higher because of the large applied flow velocity [31].
The temperature dependence of the computed velocity shown in Fig. 1
matches that of the data above 0.5 K but markedly bends upward below that
temperature; the larger a0 , the more pronounced the bending. As noted before,
this behaviour comes from the form (2) of the energy barrier and is therefore
built into the model.
It is apparent in Fig. 3 that the width at high temperature is in reasonable
agreement with the smaller values of a0 (i.e. ∼ 3 Å), while the low temperature
end calls for larger values (∼ 4.5 Å). This trend is the same as for the critical
velocity in Fig. 1, and, as noted above, it can certainly be expected that a0
would increase close to the wall, hence with increasing velocities and decreasing
nucleated vortex size [3].
The critical velocities scale as the inverse of a0 (This scaling does not appear
in Fig. 1). An estimate of the absolute, local, value of vc , i.e., the value at the
nucleation site, has been made using 3 He impurities as local probes [32]. The
presence of a tiny concentration of 3 He quasiparticles strongly affects the critical
velocity, as shown in the insert of Fig. 1, because quasiparticles condense on the
vortex core and lower its energy. The local concentration of 3 He depends on
the local pressure, itself governed by the velocity according to the Bernoulli
equation. Detailed measurements of this effect have led to a value of ∼ 22 m/s
for the local nucleation velocity on the quantum plateau. This value is to be
compared with the mean flow velocity in the aperture which is measured in the
experiment and which is of the order of 5 m/s. Using the curves in Fig. 1 and
(10), the corresponding value of a0 is found to be ∼ 2.65 Å.
Thus the values of the vortex core parameter a0 which account for the various
experimentally measured quantities, the magnitude and the temperature depen-
dence of vc and of Δvc , agree with one another within a factor of about two
in spite of the blatant oversimplifications of the 1/2 –ring model. Hence, taking
account of the details of the nucleation site geometry and chemical environment,
of the exact nature of the boundary layer at the wall and of the properties of
the small nascent vortex cannot be expected to vastly improve the description
of vortex nucleation given above. On another hand, refinements are needed, es-
pecially for the conceptual grounds underlying the model since it is known that
very small vortices do not even exist [33].

Discussion of the 1/2 –ring model. There have been several attempts to
reformulate the 1/2 –ring model, both to put it on firmer theoretical grounds and
to possibly obtain a better description of the experimental results [1,3,25,26,34].
The 1/2 –ring model main inadequacies lie at the high velocity (small vortex)
 The Experimental Evidence for Vortex Nucleation 43

end. Burkhart et al. [3] have introduced corrections to the vortex energy and
momentum due to the proximity of a boundary as computed with the Gross-
Pitaevskii equation. These corrections go in the right direction but cannot be
considered as a full reformulation of the small-vortex-at-a-wall problem.
A reassessment of the model, due to Sonin [25] and aiming in particular at
a quasiclassical reformulation of vortex quantum tunnelling, yields the following
elegant analytical form for the energy barrier
  2
κ3 ρs κ
Ea = ln
32 π v 4π va0

in the thermally activated régime and a corresponding expression for quantum

tunnelling. This energy barrier does vanish for v ∗ = κ/4πa0 but quadratically in
1 − v/v ∗ . This functional dependence is further away from the empirical relation
(7) than the low temperature limit of the 1/2 –ring model (2). Consequently, and
as also shown by direct numerical computation, the curvature of vc (T ) at low T is
even more pronounced. An even more serious discrepancy is found for the width
Δvc : the computed value is much too large for any reasonable value of a0 . These
results may look surprising since the work of [25] starts from the same premises
as the conventional 1/2 –ring model. The difference has been checked numerically
to reflect the cost of the approximations made to obtain the analytical form of
Ea [35].
Sonin also discusses the influence of the geometry of the asperity but there
is no direct evidence from experiment that this geometry plays an important
rôle. On the contrary, the fact that the slope of vc (T ) with temperature, that is
T0 in (1), is found to be the same in quite different apertures shows that vor-
tex nucleation is not very sensitive to the details of the geometry. It is a little
more puzzling that the crossover temperature Tq from quantum tunnelling to
thermal activation has been found independent of cooldowns for two different
apertures at Saclay, although a somewhat different value of Tq has been reported
by the Berkeley group [36]. The overall agreement between various experiments
basically indicates that finer specific details are not very relevant and that the
simplifications of the 1/2 –ring model are reasonably well founded. It nonethe-
less remains that the nature and geometry of a typical nucleation site are quite
undetermined and that the enhancement factor between the mean aperture ve-
locity and the velocity at the nucleation site is not under control, as shown in
particular by Shifflett and Hess [37].
Theoretical fits comparable to those in Figs. 1 and 3 have been obtained by
Zimmermann et al. [1] with a variant of the 1/2 –ring model in which a0 is kept
fixed to the Hills-Roberts value but additional parameters are introduced to force
the vortex energy and momentum to go to zero for vanishing vortex sizes faster
than the classical vortex expressions. Also, the attempt frequency ω0 /2π and vc0
are treated as fitting parameters. These authors find that ln(ω0 /2π) should be
comprised between 15 and 25 (3.3 × 106 and 7.2 × 1010 Hz for ω0 /2π). The latter
value is compatible with vc0 = 26.5 m/s, which is in fair agreement with the
value of 22 m/s found in [32]. Thus, different approaches give converging results
44 É. Varoquaux et al.

and the 1/2 –ring model can be said to give a semi-quantitative description of
vortex nucleation when single phase slips are involved.

2 Multiple Slips and Collapses

Single phase slips are observed in experimental situations which may be loosely
characterised as ‘clean’, that is, for uncontaminated apertures of relatively small
sizes (a few micrometres at the most), with low background of mechanical and
acoustical interferences, etc ..., and with probing techniques which do not man-
handle the superfluid, namely, with low frequency Helmholtz resonators. When
these conditions are not met, flow dissipation occurs in a more or less erratic
manner by large events - giant slips or ‘collapses’ of the superflow. This last
situation is quite commonly met in practice, as discussed in [38].
Collapses constitute an apparent disruption of the vortex nucleation mech-
anism described in the previous section. Their properties have been studied in
phase slippage experiments and are reviewed in this section, together with possi-
ble mechanisms for their formation. It is likely that these events provide a bridge
between the ‘clean’ single phase slip case and the usual situation of critical ve-
locities which are temperature-independent below 1 K and which depend on the
channel size d according to vF ∼ (1/d) ln(d/a0 ) [39]. It is also possible that they
take part in the build-up of vortex tangles forming superfluid turbulence [40].
Neither problems are fully resolved at present.

The two types of large slips. Examples of multiple slips can be seen in Fig. 4
which shows the peak amplitude chart of a two-aperture resonator at 12.5 mK,
24 bars, in a 100 ppb 3 He in 4 He sample. The very large amplitude drop shown
in Fig. 4 and in the insert is rare (one in 104 to 105 slips) under the conditions of
this particular experiment. This type of events, called in [38] ‘singular’ collapses
and discussed further below, may occur at velocities much below the vortex
nucleation threshold (down to at least a third of vc ).
Besides the usual single slip pattern, there appears in Fig. 4 occasional double
slips (i.e. involving phase changes by 4π) and infrequent triple slips. Raising the
temperature to 80 mK, again for this particular cooldown, causes these multiple
slips to occur much more frequently and to involve more circulation quanta
on the mean. Lowering the pressure to 0 bar results in an almost complete
disappearance of multiple slips at all temperatures. These features are described
in detail in [38].
Some degree of understanding of the formation of multiple slips can be gained
by plotting the mean value of the phase slip sizes, expressed in number of quanta,
against the flow velocity at which the slips take place [41]. This flow velocity is
close to the critical velocity for single phase slips, i.e. the nucleation velocity; it is
varied by changing the temperature,
  the pressure, the resonator drive level. Such
a plot is shown in Fig. 5 for n+ , i.e. in flow direction conventionally chosen as
the + direction. Slips in the opposite (−) direction behave qualitatively in the
same manner but the phenomenon displays a clear quantitative asymmetry. As
 The Experimental Evidence for Vortex Nucleation 45

Fig. 4. Absolute peak amplitudes at successive half-cycles of the resonance, normalised

to the amplitude drop of a single slip versus time expressed in half-cycle number. The
half-period is 31.8 milliseconds. Individual measurements cannot all be resolved on
the scale of the figure. The top trace shows a succession of amplitude drops which
correspond, for the largest part, to succession of phase slips by 2π of opposite sign,
with occasional larger slips. When no slip occurs during the half-cycle, the resonance
amplitude grows under the action of the electrostatic drive continuously applied to the
resonator. The large feature around the 2000th half-cycle is a ‘singular’ collapse. The
insert shows the details of this particular collapse, • for positive peaks, ◦ for negative
peaks.

can be seen in Fig. 5, the mean slip size decreases on either side of the quantum
plateau, as does the nucleation velocity, but increases with pressure, contrarily
to the nucleation velocity which decreases with increasing pressure. It appears
clearly that the magnitude of the superflow velocity does not directly control,
by itself, the occurrence of multiple slips. This implies, as will be discussed fur-
ther below, that the phenomenon under study is not purely hydrodynamical in
the bulk of the fluid but involves some complex interplay with the boundaries.
As shown in Fig. 5, the velocity threshold for the appearance of multiple slips
depends
  on hydrostatic pressure; in fact, the P -dependence of the upturn of
n+ vs v exactly tracks that of the critical velocity for single phase slip nucle-
ation. This indicates that multiple slips appear because of an alteration of the
nucleation process itself.
The pattern of formation of multiple slips changes from cooldown of the cell
from room temperature to cooldown but remains stable for each given cooldown.
46 É. Varoquaux et al.

Fig. 5. Mean size of (positive) multiple slips vs velocity in phase winding number in
nominal purity 4 He (100 ppb 3 He): () pressure sweep from 0.4 to 24 bars at 81.5
mK (all even values of P , and 0.4, 1, 3, 5, 7 bars) - () temperature sweep at 16
bars - (◦) temperature sweep at 24 bars - (∗) drive level sweep at 24 bars, 81.5 mK
- () temperature sweep at 0 bar. For the temperature sweeps, from 14 to 200 mK
approximately, v first increases, reaches the quantum plateau and then decreases, as
shown in the insert of Fig. 1. Lines connect successive data points in the temperature
and pressure sweeps.

It seems to depend on the degree of contamination of the cell, degree which can-
not easily be controlled experimentally. The detailed microscopic configuration
of the aperture wall where nucleation takes place probably plays an major rôle
in multiple slip formation. Multiple slips are different from ‘singular’ collapses
and the underlying mechanisms responsible for both phenomena are bound to
be different as will be discussed below.

Remanent vorticity and vortex mills. Remanent vorticity in 4 He, which had
been long assumed, has been shown directly to exist by Awschalom and Schwarz
[42]. This trapped vorticity, according to Adams et al [43], either is quite loosely
bound to the substrate and disappears rapidly, or is strongly pinned. To account
for these observations, Schwarz has proposed the following formula, based on
numerical experiments, for the velocity at which vortices unpin,
κ b
vu  ln( ) ,
2πD a0
D being the size of the pinned vortex and b being a characteristic size of the
pinning asperity. Thus, vortices pinned on microscopic defects at the cell walls
can in principle exist under a wide range of superflow velocities.
 The Experimental Evidence for Vortex Nucleation 47

It has been suggested by Glaberson and Donnelly [10], in connection with the
critical velocity problem in an aperture, that imposing a flow to a vortex pinned
between opposite lips of the aperture would induce deformations such that the
vortex would twist on itself, undergo self-reconnections, and mill out free vortex
loops. As shown by numerical simulations of 3D flows involving few vortices
only [44], vortex loops and filaments are stable even against large deformations.
Vortices are not prone to twist on themselves and foster loops. It takes the
complex flow fields associated with fully developed vortex tangles to produce
small rings, as discussed at the Workshop [45,46]. And it takes some quite special
vortex pinning geometry to set up a vortex mill that actually works; Schwarz
has demonstrated by numerical simulations that a vortex pinned at one end
and floating along the flow streamlines with its other end free moving on the
wall develops a helical motion, a sort of driven Kelvin wave, and reconnects
sporadically to the wall when the amplitude of the helical motion grows large
enough [47]. This helical mill does churn out fresh vortices.
The above remarks on the stability of vortex loops or half-loops in their course
make it unlikely that the multiple slips be due to the production of small rings
by a vortex after having left the vicinity of the nucleation centre. Furthermore,
such a purely hydrodynamical process would depend on the velocity of the flow
only, contrary to the results shown in Fig. 5. What appears more likely is that
multiple slips are produced by a transient vortex mill of the helical type suggested
by Schwarz operating very close to the nucleation site. The pinning of the mill
vortex and its subsequent release would take place immediately after nucleation
when the velocity of the vortex relative to the boundary is still small and the
capture by a pinning site easy. This process depends on the precise details of the
pinning site configuration and of the primordial vortex trajectory, factors which
allow for the variableness of multiple slips on contamination and pressure (i.e.
single phase slip nucleation velocity) [48]. That pinning does take place close to
sites where vortices are nucleated is shown below.

In-situ contamination by atomic clusters: pinning and collapses. In a

series of experiments conducted at Saclay [49,50], it was found that heavy con-
tamination of the cell by atomic clusters of air or H2 caused numerous collapses of
the ‘singular’ type to occur. The peak amplitude charts of the resonator became
mostly impossible to interpret, except in a few instances where two apparent
critical velocities for single slips were observed. The higher critical velocity cor-
responded to the one observed in the absence of contamination. The lower critical
velocity is thought to reveal the influence of a vortex pinned in the immediate
vicinity of the nucleation site. This vortex induces a local velocity which adds to
that of the applied flow and causes an apparent decrease in the critical velocity
for phase slips. Because of this change, the presence of the pinned vortex could
be monitored, the lifetime in the pinned state and the unpinning velocity could
be measured, yielding precious information on the pinning process.
48 É. Varoquaux et al.

A observation about vortex nucleation which comes from these experiments is

rather straightforward: existing vortices, either pinned or moving, do contribute
to the nucleation of new vortices at the walls of the experimental cell.
Another observation is that a large number of unpinning events were tak-
ing place at an ‘anomalously low’ unpinning velocity. A parallel can be made
[50] with singular collapses, which may also occur at ‘subcritical’ velocities and
which were also quite numerous, suggesting that the two effects have a common
cause. Noting furthermore that pinning and unpinning processes were also quite
frequent in these experiments, releasing a fair amount of vagrant vorticity, it
appears quite plausible that both singular collapses and low velocity unpinning
events are caused by vagrant vortices hopping from pinning sites to pinning sites,
eventually passing by close to the pinning centre or the nucleation site, and giv-
ing a transient boost to the local velocity which pushes a pinned vortex off its
perch or causes a burst of vortices to be shed.
As shown in this section, the capability to observe the behaviour of single vor-
tices in resonator experiments, interpreted with the hindsight provided by finely
tuned numerical simulations, leads to a partial understanding of the phenomena
of pinning, multiple slips and collapses which are taking place on a nanometric
scale at the rims of the micro-aperture in strong superflows. Vagrant vortices
reentering the region where the vortex nucleation site lies, nanoscopic vortex
mills à la Schwarz, offer explanations for singular collapses and for multiple slips
respectively which account for the detailed signatures of these effects. Obtaining
a clearer and better grounded picture would require a knowledge of the mi-
crostructure of the nucleation site and its environment which is not available at
present.
A large number of results on topics relevant to this work were presented at
the QVD workshop to which the authors feel unable to refer appropriately here
and to which they hope that other contributions in this volume will give fuller
coverage. One of them (E.V.) wishes to acknowledge the hospitality of the Isaac
Newton Institute where part of the manuscript was prepared.

References
1. W. Zimmermann Jr., C.A. Lindensmith, J.A. Flaten, J. Low Temp. Phys. 110, 497
(1998), and references therein
2. E. Varoquaux, in: Topological Defects and the Non-Equilibrium Dynamics of Sym-
metry Breaking Phase Transitions, ed. by Y.M Bunkov, H. Godfrin (Kluwer Aca-
demic, The Netherlands 2000) p. 303 and references therein
3. S. Burkhart, M. Bernard, O. Avenel, E. Varoquaux: Phys. Rev. Lett. 72, 380 (1994)
4. W. Zimmermann, Jr: J. Low Temp. Phys. 93, 1003 (1993)
5. O. Avenel, G.G. Ihas, E. Varoquaux: J. Low Temp. Phys. 93 1031 (1993)
6. E. Varoquaux, O. Avenel: Physica B 197, 306 (1994) and references therein
7. W. Zimmermann, Jr.: Contemporary Phys. 37, 219 (1996)
8. C. Josserand, Y. Pomeau: Europhys. Lett. 30, 43 (1995)
9. C. Josserand, Y. Pomeau, S. Rica: Phys. Rev. Lett. 75, 3150 (1995)
10. W.I. Glaberson, R.J. Donnelly: Phys. Rev. 141, 208 (1966)
 The Experimental Evidence for Vortex Nucleation 49

11. K.W. Schwarz: Phys. Rev. Lett. 64, 1130 (1990)

12. T. Frisch, Y. Pomeau, S. Rica: Phys. Rev. Lett., 69,1644 (1992) C. Josserand, Y.
Pomeau, S. Rica: Physica D 134, 111 (1999) S. Rica: this Workshop
13. B. Jackson, T. Winiecki, M. Leadbeater, J.F. McCann, C.S. Adams: this Workshop,
and references therein
14. P.H. Roberts, N. Berloff: this Workshop, and references therein
15. W. Zimmermann Jr., O. Avenel, E. Varoquaux: Physica B, 165&166, 749 (1990)
16. Phase winding numbers are obtained from velocities in cm/s by multiplication by
lh /κ, the ‘hydraulic’ length lh characterising the geometry of the aperture. For a
phase slip by 2π, the phase winding number changes by one unit and the trapped
circulation in the resonator loop by one quantum.
17. J. Steinhauer, K. Schwab, Yu. Mukharsky, J.C. Davis, R.E. Packard: Phys. Rev.
Lett. 74, 5056 (1995) J. Low Temp. Phys. 100, 281 (1995). The criticism of the
1
/2 –ring model made in these references has been refuted by E. Varoquaux and O.
Avenel: Phys. Rev. Lett. 76, 1180 (1996)
18. G.G. Ihas, [Link], R. Aarts, R. Salmelin, E. Varoquaux: Phys. Rev. Lett. 69,
327 (1992)
19. V.I. Mel’nikov: Phys. Reports 209, 1 (1991)
20. A.O. Caldeira: PhD Thesis, quoted in A.O. Caldeira, A.J. Leggett: Ann. Phys.
(N.Y.) 149, 374 (1983)
21. E. Varoquaux, M.W. Meisel, O. Avenel: Phys. Rev. Lett. 57, 2291 (1986)
22. G.E. Volovik: Sov. Phys. JETP Lett. 15, 81 (1972)
23. J.S. Langer, J.D. Reppy: Prog. Low Temp. Phys., Vol. 6, ed. C. J. Gorter (North-
Holland , Amsterdam 1970)
24. C.M. Muirhead, W.F. Vinen, R.J. Donnelly: Phil. Trans. Roy. Soc. A 311, 433
(1984) Proc. R. Soc. London A 402, 225 (1985)
25. E.B. Sonin: Physica B 210, 234 (1995)
26. U.R. Fischer: Phys. Rev. B58, 105 (1998) Physica B 255, 41 (1998)
27. C. Nore, C. Huepe, M.E. Brachet: Phys. Rev. Lett. 84, 2191 (2000)
28. S. Rica: Physica D (to be published) and this Workshop
29. R.N. Hills and P.H. Roberts: J. Phys. C 11, 4485 (1978) P.H. Roberts, R.N. Hills
and R.J. Donnelly: Phys. Lett. 70A, 437 (1979)
30. J. Maynard: Phys. Rev. B 14, 3868 (1976)
31. Numerical simulations in the framework of the Gross-Pitaevskii equation presented
at the Workshop by C. Adams and N. Berloff illustrate the deformation of the
boundary layer with velocity quite vividly. It has also been shown in the same
framework that the vortex core radius diverges at a wall [3].
32. E. Varoquaux, G.G. Ihas. O. Avenel, R. Aarts: Phys. Rev. Lett. 70, 2114 (1993)
33. C.A. Jones, P.H. Roberts: J. Phys. A: Math. Gen. 15, 2599 (1982)
34. F.V. Kusmartsev: Phys. Rev. Lett. 76 1880 (1996)
35. The form of the logarithmic term in the expression for the energy barrier cannot
be considered as accurate (E. Sonin, private communication)
36. J.C. Davis, J. Steinhauer, K. Schwab, Yu. Mukharsky, A. Amar, Y. Sasaki, R.E.
Packard: Phys. Rev. Lett. 69, 323 (1992)
37. G. M. Shifflett, G. B. Hess: J. Low Temp. Phys. 98, 591 (1995)
38. O. Avenel, M. Bernard, S. Burkhart, E. Varoquaux: Physica B 210, 215 (1995)
39. [Link], W. Zimmermann Jr., O. Avenel: in Proc. NATO workshop on Exci-
tations in 2D and 3D Quantum Fluids, ed. by A.F.G. Wyatt, H.J. Lauter (Plenum
Press, NY 1991) p. 343 and references therein
40. K.W. Schwarz: Phys. Rev. Lett. 50, 364 (1983)
50 É. Varoquaux et al.

41. E. Varoquaux, O. Avenel, M. Bernard, S. Burkhart: J. Low Temp. Phys. 101, 821
(1995)
42. D.D. Awschalom, K.W. Schwarz: Phys. Rev. Lett. 52, 49 (1984)
43. P.W. Adams, M. Cieplak, W.J. Glaberson: Phys. Rev. Lett. 78, 3602 (1985)
44. K.W. Schwarz: private communication to E.V.; the 3D simulations presented at
the QVD Workshop by C. Adams also show that vortex loops can undergo severe
deformations and not break apart (to appear in Euro. Phys. Lett.)
45. B.V. Svistunov: Phys. Rev. 52, 3647 (1995)
46. M. Tsubota, T. Araki, S.K. Nemirovskii: Phys. Rev. B (in press) and this Workshop
47. K.W. Schwarz: Phys. Rev. Lett. 64, 1130 (1990)
48. W. Zimmermann: Jr., J. Low T. Phys. 91, 219 (1993) J. Flaten: PhD Thesis,
Univ. of Minnesota (1997, unpublished). The large dissipation events reported by
these authors apparently fall in yet another category than the multiple slips and
the singular collapses discussed here as they persist over a number of resonator
periods.
49. P. Hakonen, O. Avenel, E. Varoquaux: Phys. Rev. Lett. 81, 3451 (1998)
50. E. Varoquaux, O. Avenel, P. Hakonen,Yu. Mukharsky: Physica B 255, 55 (1998)
Applications of Superfluid Helium
in Large-Scale Superconducting Systems

Steven W. Van Sciver

National High Magnetic Field Laboratory, Florida State University, USA

Abstract. The application of superfluid helium (He II) in large-scale superconducting

systems in reviewed. For most of these systems, He II is the coolant of choice, com-
pared to alternative low-temperature helium cooling methods, for its combination of
excellent thermal transport properties at reduced temperature. These advantages come
at some cost to system designers as the hardware associated with delivering He II is
more complex. The paper provides overviews of several major superconducting systems
that utilize He II cooling in terms of the relevant thermal fluid properties. Included
are issues of transient and steady-state heat transport, high Reynolds number flows,
and two-phase phenomena. An effort is made to show the connection between the fun-
damental superfluid properties and the need of the application. Areas where further
understanding of the superfluid state could benefit applications are also discussed.

1 Introduction
Liquid helium has a wide range of uses in low temperature technology today. The
properties of liquid helium that make it particularly valuable are its low temper-
ature, persistence of the liquid state to the lowest achievable temperatures and
the existence of the superfluid state with its associate unique transport phenom-
ena. These properties combined with an ever-improving cryogenic engineering
infrastructure have allowed the development of a number of large-scale systems
that use liquid helium cooling.
There are primarily two classes of large-scale applications for liquid helium:
1. Large superconducting systems such as magnets or RF cavities for high-
energy physics accelerators, fusion systems and other magnet facilities.
2. Large space-based instruments for infrared astronomy and other fundamen-
tal studies.

Each of these applications has a unique set of requirements that in turn place
requirement on the coolant, which determines the preferred state for the liquid
helium. In many cases, the best choice for the coolant is superfluid helium (or
He II) typically in the temperature range 1.4 K to 2.1 K. The decision to select
He II as a coolant is usually driven by a combination of the lower temperature
and beneficial heat transport characteristics.
The present paper begins with an overview of the basic design features and
operating characteristics employed in several large-scale He II cooled applica-
tions. The motivation for He II cooling, the unique design features, and the

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 51–65, 2001.

c Springer-Verlag Berlin Heidelberg 2001
52 S.W. Van Sciver

required engineering database are reviewed in the context of these applications.

In the subsequent section, the current understanding of He II as an engineer-
ing fluid is presented followed by comments about areas where further work is
needed.
By far the largest and most varied application for He II cooling is in the area
of large-scale superconducting magnet systems. These include magnet systems
for high energy physics, fusion experiments as well as other specialized magnet
facilities. In addition, there are several large-scale electron accelerators that uti-
lize superconducting RF cavities and are also cooled by He II. Space does not
permit a detailed discussion of all these applications, so the present discussion
will concentrate on three such systems, which represent the range of technical
demands placed on He II cryogenics:

1. The dipole magnet system for the Large Hadron Collider (LHC) which
is under construction at the European Organization for Nuclear Research
(CERN) laboratory in Geneva, Switzerland.
2. The recently completed outsert superconducting magnet of the 45-T Hybrid
Magnet at the National High Magnetic Field Laboratory in Tallahassee,
Florida.
3. The RF cavity system for the Teravolt Electron Synchrotron Linear Ac-
celerator (TESLA) under design at the Deutsches Elektronen-Synchrotron
(DESY) laboratory in Hamburg, Germany.

There are a number of advantages and some disadvantages to the use of

He II for superconducting systems. The first and most obvious advantage is
simply that the lower temperature provides improved superconductor properties.
For example, NbTi superconductors, which are used in more than 95% of the
operating superconducting magnets today, have substantially improved current
densities with operation near 1.8 K as compared to the normal 4.2 K operation in
atmospheric He I. This improved performance allows the achievement of higher
magnetic fields. For example, a standard 8 T NbTi laboratory magnet will often
achieve 10 T when operated at 1.8 K. Similarly, niobium RF cavity used in
electron accelerators have losses which decrease significantly with temperature
justifying operation near 2 K.
The other, somewhat more subtle, advantage to the use of He II cooling in
superconducting systems is in the uniquely high effective thermal conductivity
of the fluid; a property which provides improved thermal stability to the super-
conducting system. In application, superconductors dissipate negligible amounts
of heat during steady current operation. However, there are number of special
conditions that can lead to power dissipation in the superconductor. To mini-
mize this effect, the superconductor is co-processed with a normal metal (usually
copper) that acts as a shunt for current should conditions prevent the supercon-
ductor from carrying the full current. The heat necessary to bring about the
momentary normalization of the superconductor can come from a variety of
sources (dB/dt, mechanical motion) and must be guarded against. Efficient heat
removal from the conductor is therefore a critical aspect of good reliable design.
 Applications of Superfluid Helium 53

Since He II can transport large heat fluxes (compared to that of He I), its sta-
bilizing effect on magnet systems can be very beneficial. The understanding of
the relevant thermal processes in He II has been a subject of numerous recent
experimental investigations.
The principal disadvantage to the use of He II cooling in large superconduct-
ing systems originates from the increased complexity and cost of the cryogenic
system. In very simple terms, the cost of refrigeration scales with the Carnot
factor, which at low temperatures goes as T−1 . Therefore, a unit of power de-
posited in a He II cryogenic system operating at 1.8 K will require a refrigeration
system 2.3 times larger than the equivalent system operating at 4.2 K. Since the
cost of refrigeration systems scales monotonically with power, the advantages of
superconductor performance and stability must outweigh this cost issue. Thus,
the selection of a He II cooling system is only warranted in cases where the
overall system design is optimized by this approach. What follows are three such
examples.

2 Superconducting Systems That Use He II Cooling

2.1 Accelerator Magnet System for LHC

The LHC accelerator currently under construction at CERN will, when complete,
operate the world’s largest He II cryogenic system [1]. In this case, all the main
ring magnets (dipole and quadrupoles) are made with NbTi superconductor and
cooled to 1.9 K in a He II bath pressurized to atmospheric pressure. The dipole
magnets, shown in cross section in Fig. 1, operate at 8.3 T, which is higher than
previously built accelerator dipoles such as for the SSC and therefore require the
lower temperature and improved thermal stability that He II operation affords.
There are eight refrigeration plants around the 26.7 km circumference of the
accelerator, each cooling one 45◦ sector. The 1.8 K refrigeration capacity for
each sector is between 2.1 and 2.4 kW.
The system uses pressurized He II to cool the magnets, which reduces the pos-
sibility of trapped vapor within the windings of the coil. The use of pressurized
He II to cool magnets was first demonstrated on a large scale in a development
program for the Tore Supra tokamak plasma experiment in France [2]. The ap-
proach requires the use of a low pressure of He II heat exchanger as part of
the installation. In the case of LHC, the heat exchanger is a corrugated pipe
within the magnet vessel and immersed in the He II reservoir. A schematic of
this system is shown in Fig. 2. The heat exchanger is partially filled with satu-
rated He II. The two phase He II heat exchanger provides a much more efficient
heat removal method than can be obtained through counterflow heat transport
in the bulk fluid. Although heat transport is very efficient in bulk He II, the long
distances between refrigeration stations would demand very large He II cross
sections to minimize the temperature gradient.
54 S.W. Van Sciver

Fig. 1. Schematic of LHC dipole magnet

Fig. 2. Cooling system for LHC dipole magnets with two-phase heat exchanger

2.2 High Field Solenoid for the NHMFL 45-T Hybrid

The 45-T hybrid is a large combined superconducting and resistive magnet re-
cently put in service at the NHMFL [3]. The superconducting portion of this
magnet is a 14 T solenoid with a 710 mm cold bore operating at 10 kA. The
outer coil of this magnet is made with NbTi conductor, while the high field
inner sections use Nb3 Sn for its superior high field characteristics. The conduc-
tor in this magnet is a Cable-in-Conduit type (CICC), cf. Fig. 3, consisting of
a superconducting stranded cable in a rectangular steel jacket. This is an ad-
vanced conductor design developed for leading edge magnet systems. The liquid
helium coolant, which is contained within the jacketed conductor, absorbs the
transient heat generation from the conductor and conducts it to the surrounding
 Applications of Superfluid Helium 55

path. The NHMFL hybrid outsert is the first CICC magnet that employs He II
cooling. As in the case of LHC, He II cooling is applied to achieve the highest
current densities in the conductor, while providing a good thermal stabilizing
environment.

Fig. 3. Cable in Conduit conductor for hybrid magnet

A schematic cross section of the 45-T hybrid magnet system is shown in

Fig. 4. The coils are located in the magnet vessel, which is separate from the
He II refrigeration system and cooled by counterflow through a horizontal duct.
Heat extraction occurs at a set of saturated heat exchangers located in the
supply cryostat. These heat exchangers are supplied from the upper reservoir
of the cryostat with the vapor flow going through room temperature vacuum
pumps. The total steady state heat load on this system is less than 10 W at
1.7 K [4]. Thus, the scale of the refrigeration system is modest compared to the
LHC.

2.3 RF Cavity Systems for the TESLA Electron Collider

The TESLA electron collider is one of the proposed major accelerator systems
that will follow the completion of the LHC [5]. This system is a linear electron
collider as compared to a circular LHC machine, which accelerates protons. Lin-
ear machines do not require the large number of dipole bending magnets that
are the signature of the hadron colliders. Rather the electron colliders depend
on high gradient RF cavities for accelerating the beam and use a relatively small
number of magnets for beam steering and focusing. The RF cavities can be either
56 S.W. Van Sciver

Fig. 4. Schematic of 45-T hybrid magnet

made from copper and operate at room temperature or from superconducting

niobium operating near 2 K. This latter approach has been demonstrated in
the Continuous Beam Accelerating Facility (CEBAF) at the Thomas Jefferson
Laboratory in Virginia. The use of niobium cavities provides higher accelerating
voltages and lower overall losses compared to operation at higher temperature
(e.g. 4.2 K).

Fig. 5. Schematic of TESLA cooling system

A schematic of the cryogenic system for TESLA [6] is shown in Fig. 5. It

consists of a long cryostat, which contains the RF cavities, filled with saturated
He II typically at 2 K. Unlike superconducting magnets, large local disturbances
are not expected to occur in RF cavities, so the complexity of a pressurized He II
system is not required. The RF cavity cryostats, cross section shown in Fig. 6,
are maintained at constant temperature over a long length by linking them to
a common vacuum line, which may be partially filled with liquid. Thus, the
 Applications of Superfluid Helium 57

principal technical issue associated with the design of the He II system pertains
to the behavior of two-phase He II/vapor in near horizontal channels of relatively
large diameter.

Fig. 6. Schematic of TESLA cavity cryostat

3 Application Relevant He II Properties

The application of He II cooling to large systems requires an understanding of

the fundamental properties of this unique fluid. Of greatest interest are those
properties that affect engineering design of the system. In some cases, the strong
demands are placed on the He II coolant and it is only through practical inves-
tigation combined with basic understanding that one can apply He II cooling
in an optimal way. The present section reviews current practical knowledge of
He II as it affects applications. Some discussion of the need for further work is
also given.

3.1 Second Sound Pulse Transport

The transport of heat pulses in He II is primarily of relevance to the stability of

superconducting systems (mostly conductors) in He II. This is a very complex
problem, the solution of which continues to elude scientists and engineers. The
primary reason for this problem comes from an incomplete understanding of the
source and characteristics of the heat deposition spectrum that originates from
58 S.W. Van Sciver

a working superconducting magnet. Such disturbances come from a variety of

sources (AC losses in the conductor or mechanical motion of the winding) and
can vary considerably from magnet to magnet. Thus, most of the effort is placed
on understanding the transient thermal response of the He II coolant. One would
like to understand the development of the fully turbulent state in He II and the
limits of heat transfer. Nemirovskii and Tsoi [7] have presented this problem in
the form of a regime map, which qualitatively defines the transition of He II
from the ideal superfluid state through the development of the fully developed
turbulent state to the limit of heat transfer film boiling. For short times, the heat
transport is by a burst of second sound pulse, which leads to the development of
turbulence. This second sound pulse can be of benefit to the operation of very
high current density magnets, which sometimes experience very short duration
intense disturbances.
Several experiments have reported on intense heat pulse propagation in He II
[7,8]. Shimazaki et al. [8] measured the shape and heat content of thermal shock
pulses in a counterflow He II channel. Fluxes as high as 40 W/cm2 for up to
1 ms duration were applied and shown to propagate at second sound velocities.
Turbulence generated in the He II followed the pulse an effect that limited the
maximum energy transported to about 10 mJ/cm2 . This is a large energy de-
position for high field compact superconducting magnets and thus could affect
design and performance. However, one would like to better understand how this
process transforms into the fully developed turbulent state, which is discussed
in the next section.

3.2 Transient and Steady Transport in the Mutual Friction Regime

In many large-scale superconducting systems, the He II can be assumed to be
fully turbulent and the heat transport governed by the mutual friction regime.
There are several features of these systems that support this assumption. First,
most of these systems have relatively large channels and He II reservoirs, so
that the He II already contains a considerable amount of turbulence. Also, large
magnets have slower thermal response times, typically 10 ms to 1 s, so that
propagation of second sound pulses can only be a small portion of the total flux.
This assumption is the design basis of most large-scale superconducting systems
that are cooled with He II.
With the assumption of fully developed turbulence, one can then apply the
mutual friction form of the two fluid equations to the general problem of heat
transport in the He II [9]. Thus, the temperature gradient in the fluid is given
by the relationship,
 1/3
−1 dT
q=− f (1)
dx
where f = ρ3Aρ n
4 3 and A is the usual Gorter Mellink parameter. The interesting
ss T
development related to this process is the realization that (1) can be applied
much like Fourier’s Law for conduction in materials, although the form is pro-
portional to the cube root of the temperature gradient. Thus, slowly varying
 Applications of Superfluid Helium 59

heat transport processes should have a diffusion-like character and be governed

by the He II diffusion equation,
 1/3
∂T ∂ −1 ∂T
ρCp = f . (2)
∂t ∂x ∂x
This formulation has been successful at treating a wide variety of transient
heat transport problems related to superconducting magnet systems such as the
thermal stability of a composite superconductor.
In a superconducting composite conductor, the power generated in the con-
ductor is a function of time. For short times, the power usually peaks as the
current distributes within the conductor and the initial heat deposited diffuses
away. After the peak, the power usually levels off and corresponds to the steady-
state joule heat in the copper stabilizer. A simplified form for the heat generated
in the conductor is shown by the cross-hatched portion in Fig. 7. To ensure that
the conductor returns to the superconducting state, it is necessary to have the
heat transfer to the He II exceed the heat generated. The highest level of reli-
ability is obtained by requiring that the Joule heat never exceed the maximum
steady-state heat flux in the He II. However, to optimize the design, it is really
only necessary to have the integrated average heat transfer exceed the genera-
tion. This condition, which was first proposed by P. Seyfert et al. [10], is shown in
Fig. 7. The curve represents the maximum heat transport by the He II governed
by the diffusion process, see (2). As long as the excess heat generation, area A,
does not exceed the excess heat tranfer, area B, the magnet should be stable.

A H e a t T ra n s fe rre d to H e II

o w e r (W )

B
I2R
H e a t G e n e ra tio n

T im e
Fig. 7. Stability criterion for He II cooled conductor

3.3 The He II Energy Equation

Some applications of He II require forced circulation through a closed loop. This
approach adds complexity to the system, but can provide significantly enhanced
60 S.W. Van Sciver

heat removal particularly in distributed or remote systems. For example, such a

forced flow system was considered for the 45-T hybrid outsert magnet, but not
accepted in the final design due to the additional complexity and cost. Adding
a net fluid velocity to He II provides some advantage in design as well as intro-
ducing a new variable into the analysis of the fluid behavior.
As in the case of static He II applications, if one assumes that the He II is
governed by the fully developed turbulent mutual friction process, then forced
flow He II heat transport should obey a modified energy equation. This equation
has the form [11],
 1/3
∂ −1 ∂T 1 ∂p ∂T ∂ρ ∂T
f − − ρuC −u = ρC (3)
∂x ∂x ρs ∂x ∂x ∂x ∂t

where the pressure gradient terms take into account the frictional losses and
change of internatl energy with pressure. As before, this relationship has success-
fully modeled the behavior of heat transport in He II for a variety of experimental
systems. An interesting observation about this formulation is the fact that He II
will display the Joule Thomson effect when experiencing a pressure drop through
an insulated tube [12]. The result will be an increase in temperature, which will
generally have a negative impact on applications.
An example of heat transport in forced flow He II is displayed in Fig. 8.
The flow is from left to right and a heat pulse is deposited at x = 0. The
plot displays calculated and experimental time-dependent temperature profiles
at different locations along a channel containing He II. Note that the peak in
the temperature profile broadens due to thermal diffusion as time progresses
and the pulse propagates. Also, the location of the peak moves at approximately
the speed of the fluid, suggesting a rough method of measuring fluid velocity in
such systems. At these high velocities, the Joule Thomson effect contributes by
producing a gradual increase of the fluid temperature as seen by the increase in
the baseline temperature as the fluid moves through the tube.
From the application viewpoint, the steady-state and transient heat transport
characteristics a fully turbulent He II can be understood in terms of the He II
diffusion and energy equations. There are limits to this representation and these
limits need to be explored. Work is continuing on the development of turbulence
and heat transport at very high velocities. This seems to be an area where some
fundamental work would be able to contribute.

3.4 Fluid Dynamics of Forced Flow He II

Technical interest in forced flow He II has grown out of a number of applica-

tions. As mentioned previously, the LHC accelerator magnet system and the
NHMFL hybrid magnet both had early designs that utilized forced flow He II
cooling. In addition, several space-based infrared telescopes, including the cur-
rently underway Space Infrared Telescope Facility (SIRTF), have evaluated ways
of circulating He II. The fundamental question that arose out of these studies
 Applications of Superfluid Helium 61

Fig. 8. Pulsed heat transport in forced flow He II (from [17])

was whether the He II pressure drop and associated fluid dynamic properties
would be unique. Since that time, a considerable body of research has come to
support the notion that the dynamic behavior of forced flow He II is essentially
similar to that of classical fluids. Thus, He II when flowing through tubing dis-
plays a pressure drop that can be described by classical correlation [13]. This
result has been supported by theoretical investigations that suggest the two flu-
ids are coupled through turbulent interactions and thus flow together. Figure 9
displays recent measurements of the friction factor for He II flowing in a 10 mm
ID tube at Re>107 (u> 10 m/s). These data continue to support the previous
observations.
Forced flow He II at high Reynolds number has more recently become inter-
esting as a test fluid for basic fluid dynamic investigations [14]. The fluid has a
very small kinematic viscosity allowing high Reynolds numbers to be achieved
in sub-sonic flows. A question that continues to be raised is to what extent can
one ignore superfluid effects. This is still an open issue and a subject for further
investigation. In a companion paper at this conference, we report on drag coeffi-
cient measurements for a sphere in flowing He II [15]. The measurements suggest
a temperature dependence to CD not seen in normal fluids and this dependence
appears to correlate with the normal fluid density, ρn . Clearly, this topic needs
further experiment supported by theoretical analysis.
62 S.W. Van Sciver

4 .1 m s t r a ig h t
4 .6 m lo o p
f 2 m c o p p e r lo o p
1 .2 m s t r a ig h t
V o n K a r m a n - N ik u r a d s e
C o le b ro o k
0 .0 1

0 .0 0 1
5 6 7 8
1 0 1 0 1 0 1 0
R e
Fig. 9. He II friction factor for pipe flow at high Reynolds number

3.5 He II/Vapor Two Phase Flow

An area that has received recent investigation driven by the needs of large accel-
erators such as LHC or TESLA is two phase He II/vapor flow. In many of these
large systems, there are tubing sections partially filled with He II and in near
horizontal configuration. This operating condition leads to some interesting phe-
nomena due to the existence of the free surface. As is in the case of single phase
He II systems, the heat transfer and fluid dynamics behavior are of interest to
designers. Further, because of the free surface and relative velocity between the
liquid and vapor phases, there are a variety of flow stability issues that need to
be addressed. These are topics of current investigation [16,17,18]. Modeling these
systems is complex due to the interaction between the two phases [19,20]. To
date, most studies have treated the He II as a classical fluid with heat transport
character governed by the He II energy equation. Clearly, this fluid system is
complex and further experimental and theoretical work is needed.

3.6 Fountain Effect (Fluid Management)

The thermomechanical fountain effect in He II has been considered during the
design of a variety large scale applications. The most successful application of this
effect occurred with the Superfluid Helium On Orbit Transfer (SHOOT) experi-
ment which flew on the Space shuttle in 1993 [21]. This experiment demonstrated
the ability of a large-scale fountain pump to tranfer He II in micro-gravity condi-
tions. Also, this experiment as well as most other space-based He II systems used
a porous plug phase separator to contain the liquid helium within the dewar and
vent the vapor. The porous plug phase separator uses the heat of evaporation to
remove the small internal heat generated in the dewar. The temperature differ-
 Applications of Superfluid Helium 63

ence between the He II bath and exciting vapor provides the thermomechanical
pressure head to hold back the liquid.

Fig. 10. Schematic of method to use porous plugs to manage He II fluid level

The use of the thermomechanical fountain effect is much more limited in

large-scale superconducting systems. It has been considered in the design of
space-based magnet systems for similar reasons to its use in infrared telescope
technology, i.e. mostly as a phase separator. The fountain effect may also be
useful as part of the fluid management system for ground-based accelerator sys-
tems containing horizontal He II/vapor two phase flows. One concern in these
systems is “dry out”of the piping due to the pressure and temperature gradient
along the channel. Higher temperature means higher vapor pressure, which will
depress the liquid level compared to that down stream. Thus, the liquid in the
channel will slope upward in the downstream direction if there is no net liquid
flow. One possible method to overcome this concern is to use the fountain effect
to return and circulate He II to the upstream side of the channel. A schematic
of how this process can be achieved is shown in Fig. 10. The porous plugs are
part of a parallel fluid handling system. The increase in temperature on the up-
stream side would have the effect to force liquid through the porous plug thus
raising the level and preventing dry-out. There is no current application for this
concept, but some technical sub-components demonstrated the approach [22].

4 Conclusions
Superfluid helium has become an engineering fluid for a number of technical
applications in superconductivity and space-based instrumentation. A consider-
able volume of practical data has been accumulated through the development of
these systems. Steady-state and slowly varying transient thermal processes can
be described to be a diffusive process much like conduction. Forced flow at rela-
tively high velocities appears to obey clssical correlations. There are a number of
areas where further study is needed. We do not have adequate understanding of
the development of turbulence particularly at high heat fluxes. Also, two-phase
64 S.W. Van Sciver

flow is an entire subject that has only recently been under investigation. Future
applications will no doubt require additional investigation.

Acknowledgements
The National High Magnetic Field Laboratory is jointly funded by the National
Science Foundation and the State of Florida.

References
1. P. Lebrun, Cryogenics for the Large Hadron Collider. IEEE Trans. On Applied
Super., Vol 10, 1500 (2000)
2. G. Claudet and R. Aymar, Tore Supra and He II Cooling of Large High Field
Magnets, Adv. Cryog. Engn. 35A, 55 (1990)
3. J.R. Miller, et al., An Overview of the 45-T Hybrid Magnet System for the
NHMFL, IEEE Trans. On Magnetics, Vol 30, 1563 (1994)
4. S.W. Van Sciver, et al., Design, Development and Testing of the Cryogenic System
for the 45-T Hybrid, Adv. Cryog. Engn. Vol 41, 1273 (1996)
5. R. Brinkmann, et al., Conceptual Design of a 500 Gev e+e- Linear Collider with
Integrated X-Ray Laser Facility, DESY 1997-048 ECFA 1997-182, May 1997
6. G. Horlitz, et al., The TESLA 500 Cryogenic System and He II Two Phase Flow,
Cryogenics Vol 37, 719 (1997)
7. S.K. Nemirovskii and A.N. Tsoi, Transient Thermal and Hydrodynamic Processes
in Superfluid Helium, Cryogenics Vol 29, 985 (1989)
8. T. Shimazaki, M, Murakami and T. Iida, Temperature measurement in Transient
Heat Transport Phenomena Though a Thermal Boundary Layer in High Vortex
Density, Adv. Cryog. Engn. Vol 41, 265 (1996)
9. S.W. Van Sciver, Chap. 10: Helium II (Superfluid Helium), in: Handbook of Cryo-
genic Engineering, J.G. Weisend II (ed.). Taylor & Francis (1998)
10. P. Seyfert, Practical Results on Heat Transfer in Superfluid Helium, in: Stability
of Superconductors in He I and He II, IRR Commission A 1/2 (1981), pp. 53-62
11. B. Rousset, Pressure Drop and Transient Heat Transport in Forced Flow Single
Phase He II at High Reynolds Number, Cryogenics Vol 34 supplement, 317 (1994)
12. P.L. Walstrom, Joule Thomson Effect and Internal convection Heat Transfer in
Turbulent He II Flow, Cryogenic Vol 28, 151 (1988)
13. P.L. Walstrom, et al., Turbulent Flow Pressure Drop in Various He II Transfer
System Components, Cryogenics Vol 28, 101 (1988)
14. R.J. Donnelly, Ultra High Reynolds Number Flows Using Cryogenic Helium: An
Overview, in: FLow at High Reynolds and Rayleigh Numbers, R.J. Donnelly and
K. Sreenivasan (eds.), Springer (1998)
15. M.R. Smith, Y.S. Choi and S.W. Van Sciver, The Temperature Dependent Drag
Crisis on a Sphere in Flowing He II, (this publication)
16. P. Lebrun, et al., Cooling Strings of Superconducting Devices Below 2 K: The
He II Bayonet Heat Exchanger, Adv. Cryog. Engn., Vol 43, 419 (1998)
17. B. Rousset, et al. Behavior of He II in Stratified Counter-Current Two Phase
FLow, in: Proceedings of the ICEC17, Institute of Physics Publishing (1998) pp.
671-674
18. J. Panek and S.W. Van Sciver, Heat Transfer in a Horizontal Channel Containing
Two Phase He II, Cryogenics Vol 39 (1999)
 Applications of Superfluid Helium 65

19. L. Grimaud, et al., Stratified Two Phase Superfluid Helium Flow, Cryogenics Vol
37 (1997)
20. Y. Xiang, et al. Numerical Study of Two Phase He II Stratified Channel with
Inclination, IEEE Trans on Applied Super. Vol 10, 1530 (2000)
21. M. DiPirro and P.J. Shirron, The SHOOT Orbital Operations, Cryogenics Vol 32
(1992)
22. J. Panek, Y Zhao and S.W. Van Sciver, Liquid Level Control Using a Porous Plug
in a Two Phase He II System, Adv. Cryog. Engn., Vol 43, 1401 (1998)
The Temperature Dependent Drag Crisis
on a Sphere in Flowing Helium II

Yeon Suk Choi2 , Michael R. Smith1 , and Steven W. Van Sciver1,2

1
National High Magnetic Field Laboratory, Florida State University, Tallahassee,
FL 32310, USA
2
Mechanical Engineering Department, FAMU-FSU College of Engineering,
Tallahassee, FL 32310, USA

Abstract. In a previous paper, we reported observing a drag crisis on a sphere in

flowing He I and He II. Data in He II suggested a possible temperature dependence
to the critical Reynolds number, as well as the magnitude of the crisis. In this paper,
we explore temperature dependence more completely. Dynamical similarity arguments,
which lead to Reynolds number scaling in the case of the Navier–Stokes equations, are
applied to the two-fluid equations. The result is a modified Reynolds number involving
the factor 1-δ, where δ ≡ ρs /ρ. The ramifications of this argument, together with other
possible scaling relationships, are discussed. Data and critical Reynolds numbers are
plotted for several temperatures between 1.6 K and 2.0 K. Results appear to agree well
with the proposed scaling for He II.

1 Introduction

The low kinematic viscosity (ν = η/ρ, where η and ρ are dynamic viscosity
and total density, respectively) of liquid helium makes it an attractive fluid
for modern dynamical similarity studies, where one wishes high Reynolds num-
bers (Re=Ud/ν, where U and d are the characteristic velocity and dimension
of the flow field) without transonic effects. While research suggests that helium
above 2.176 K (He I) behaves as a classical fluid, He II (the liquid phase below
2.176 K) is a quantum fluid with a wide range of non-classical macroscopic prop-
erties. Still, studies have found that classically generated turbulence in He II may
behave classically in certain experiments [1,2,3]. Uncertainties about the micro-
scopic character of classically generated turbulence in He I and He II motivated
the previous work [4,5] in which the form drag on a sphere in flowing He I and
He II was calculated from the observed pressure distribution over the surface.
If the critical Reynolds number for the drag crisis in He II is temperature
dependent, then the dimensionless equations of motion for the two fluid system
must scale with other dimensionless parameters, in addition to or instead of the
Reynolds number. In the classical fluid dynamics of ordinary fluids (including
He I), dynamical similarity and scaling arguments for expressing experimental
data in terms of Reynolds number, coefficients of drag, lift and so on, spring
rigorously from non-dimensionalizing the Navier–Stokes equations. Expressing
He II data in terms of an effective Reynolds number however, has been more
of an empirical convenience. Perhaps one reason for this is the empirical nature

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 66–72, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 The Temperature Dependent Drag Crisis 67

of the two-fluid equations themselves, and the sometimes unspoken question

of whether the current set of equations accurately represent all aspects of the
physics. In the spirit of investigation however, we present one such analysis as a
framework for discussing our data. [6]
The generally accepted form for the two-fluid equations in one dimension [7]
is shown in (1) and (2), where we have restricted the problem to one dimension
for simplicity. The entropy terms describe the exceptionally high effective ther-
mal conductivity of the system through a mechanism called counterflow. The
equations for the superfluid and normal fluid components are coupled through
the velocity difference vns = vn − vs , and the mutual friction term Fns . One
expression for the mutual friction is shown in (3), where α is a temperature
dependent parameter. The local line length per unit volume of quantized vortex
lines existing in the superfluid component is given by L, and K is the quantum
of circulation around a vortex line.
2
dνs ρs dP dT ρs ρn dνns
ρs =− + ρs S + − Fns (1)
dt ρ dx dx 2ρ dx

2
dνn ρn dP dT ρs ρn dνns d 2 νn
ρn =− − ρs S − + Fns + η 2 (2)
dt ρ dx dx 2ρ dx dx
2
Fns = − αkρs Lνns (3)
3
Now, we define non-dimensional quantities in terms of representative veloci-
ties, distances and temperatures as follows;

x0 = x/D t0 = U · t/D
L0 = L · D2 T0 = (Tλ − T )/Tλ
νn0 = νn /U S0 = Tλ S0 /U 2
νs0 = νs /U P0 = P/ρU 2

where the zero subscript refers to the dimensionless quantity and Tλ = 2.176 K is
the lambda point. Upon substituting these quantities into (1-3), and rearranging
terms a bit, we arrive at;
2   2
dνs0 dP0 dT0 αk 1 − δ dνns0
=− − S0 + 3 · L0 νns0 + (4)
dt0 dx0 dx0 UD 2 dx0

   2    2
dνn0 dP0 δ dT0 3 αk δ dνns0
=− + · S0 − · L0 vns0 − (5)
dt0 dx0 1−δ dx0 UD 2 dx0
1 1 d2 νn0
+ ·
1 − δ Re dx20

The terms involving the temperature gradient drive counterflow, in which the
two fluids flow in opposite directions, and heat is transported within the system.
68 Y.S. Choi, M.R. Smith, and S.W. Van Sciver

Note that the effect of a thermal gradient upon the normal fluid component
(Eq. 6) becomes larger than 1 − δ → 0 where δ ≡ ρs /ρ. This is comforting,
since it agrees well with our expectations. Furthermore, the classical equations
of motion are recovered in the limit δ → 0 . From a more general perspective,
Eqs. (4) and (6) describe a wide spectrum of physics, which goes beyond fully
developed co-flowing turbulent fields. The driving terms for such non-classical
phenomena as counterflow however (we include for consideration at this point
terms involving vns0 ), must affect the dynamics of all flows at suitably small
length scales. Since the co-flowing condition (vns0 = 0) generally depends upon
mutual friction, it must depend upon vns0 = 0 across some range of length scales
within the flow.
In addition to the many factors of (1-δ) in (4) and (6), which underscore
the strong role of temperature in the dynamics of He II, there is one additional
dimensionless quantity, UD/(2/3)αk, which is tied into the mutual friction. Al-
though we will be discussing co-flowing turbulence, where vns0 may zero, L0 may
still be quite large. Thus, the term on the whole may fluctuate dramatically on
a local scale.
For the sake of the argument at hand however, consider the case where vns0
is strictly zero (fully coupled superfluid turbulence), with zero temperature gra-
dient. Then at suitably large length scales, we are left with
dvs0 dP0
=− (6)
dt0 dx0

dvs0 dP0 1 1 d2 vn0

=− + · (7)
dt0 dx0 1 − δ Re dx20
The most important point to notice is that we never truly recover classical
equations of motion, except in the limit as δ → 0. The viscous term is driven
by (1-δ) · Re, which provides a mechanism through which the effective critical
Reynolds number associated with the drag crisis might scale with temperature.
At first glance, this appears to pose a dilemma. A large body of data characteriz-
ing skin friction in pipes scales nicely with Reynolds number, alongside classical
data, yet without any apparent temperature dependence. One possible explana-
tion for this may lie in the fundamental difference between pipe flow, and flow
over a bluff body, such as a sphere. Pipe flow is a fully developed turbulent field,
where the two fluids largely flow together [1], except at very small length scales
and within a very narrow region close to the wall. Here, the non-slip condition
on the normal fluid provides the observed pipe friction and associated pressure
drop. Flow over a sphere however, possesses a stagnation point on the leading
edge, from which the two fluids accelerate under different boundary conditions,
with a non-zero velocity difference. This brings into play the other terms in (4)
and (6). Thus, we expect δ, and thereby temperature, to play a greater role in
flows where the two fluids are allowed to have substantially different velocity
fields (vns = 0).
As a final preliminary speculation, note that the viscous term in (7) becomes
quite large as 1-δ → 0, similar to inviscid flow (Re → 0). Thus, it may be that
 The Temperature Dependent Drag Crisis 69

the drag crisis itself changes character and magnitude as the normal component
vanishes (ρn → 0). Ultimately in this limit, one expects to recover some sort of
potential flow.

2 Experimental Apparatus and Protocol

The apparatus was described in a previous publication [4]. A 10 mm diameter
sphere is suspended upon a strut oriented perpendicular to the oncoming flow.
A single pressure tap located on the surface of the sphere is used to map out the
pressure distribution on the azimuth connecting the upstream and downstream
points by rotating the sphere/strut. This pressure distribution is then integrated
to yield the form drag.
The principal difference from the previous work is a matter of protocol. Drag
versus Reynolds number were taken at several intermediate temperatures be-
tween 1.6 K and 2.0 K in order to determine the critical Reynolds number for
the drag crisis as a function of temperature. Additionally, drag and pressure
distribution were observed at different temperatures for fixed Reynolds number.

3 Results and Discussion

The measured pressure distribution in He II at 2.0 K, expressed in terms of the
coefficient of pressure, CP = (P(Θ) − P0 )/ 12 ρU2 , is shown in Fig. 1. Points which
comprise the curve are the result of averaging a series of individual measure-
ments. Since many of the important dynamics are dependent upon the equatorial
velocity (Θ = 90◦ ), the Reynolds number was calculated based upon the mean
velocity in this smallest cross-section. In Fig. 1, the Reynolds number spans the
range from 1.1 × 105 to 7.8 × 105 . Profiles for Reynolds number at or above
1.8 × 105 show supercritical behavior with the boundary layer separation occur-
ring at approximately 100 degrees. The data corresponding to the two lowest
Reynolds numbers exhibit variations which may be due to transition, together
with very low signal levels.
Assuming azimuthal symmetry of the pressure distribution with respect to
the oncoming flow, together with the spherical shape of the surface, we may
integrate the coefficient of pressure to calculate the coefficient of drag directly.
For the discrete data here, this is easily performed by a summation. Figure 2
shows drag coefficients for our experiment, together with published results for
the smooth sphere, shown as the solid line. Error bars are derived from the
statistical scatter in the individual measurements which comprise the points on
the curve in Fig. 1. The lower Reynolds number data have larger error bars since
these measurements correspond to lower signal level. Although He II exhibits a
drag crisis at approximately the same Reynolds number as He I, the variability
in the He II data led us to speculate about a temperature dependence to the
turbulent transition within the boundary layer.
The results displayed in Fig. 2 suggest that the drag coefficient in He II in-
creases with decreasing temperature. For fixed Reynolds number, the coefficient
70 Y.S. Choi, M.R. Smith, and S.W. Van Sciver

Fig. 1. Pressure distribution for various Reynolds number at 2.0 K

Fig. 2. Drag coefficient vs. Reynolds number

is the largest at 1.6 K decreasing monotonically to the value in He I. Further,

the minimum in the drag coefficient occurring just above the transition also in-
creases with decreasing temperature. These two observations can be evaluated
in more detail by careful study of the data.
Figure 3 is a plot of the drag coefficient versus temperature for fixed Reynolds
number. Below the lambda transition, the coefficients increase approximately
linearly although not with the same slope for each case. It is also interesting
 The Temperature Dependent Drag Crisis 71

to note that the extrapolation of the measured He II drag coefficients to Tλ

appear to coincide with the approximately constant values measured in He I.
Although these data need further confirmation, it certainly appears that the
drag coefficient is temperature dependent in He II.

Fig. 3. Drag coefficient vs. temperature

To test the two fluid equation scaling arguments, we plot in Fig. 4 the drag
coefficients versus Re (1-δ), as suggested by [5]. The open symbols correspond
to He I data and the closed symbols are for He II.
Comparing Fig. 4 to Fig. 2, we note that the scaling shifts the minimum in
the drag coefficient to approximately the same value of the modified Reynolds
number for each temperature. That value, Re (1-δ) ≈ 2.3 × 105 , is in reasonable
agreement with the minimum in the classical drag coefficient curve at approxi-
mately Re = 3 × 105 .

4 Conclusion

We have measured the pressure distribution over the surface of a sphere in flowing
He II as a function of Reynolds number. The He II data shows clear evidence
of a drag crisis at approximately the same Reynolds number. The coefficients of
drag have a minimum value at same modified Reynolds number based on scaling
the He II two fluid equations, Re ×(1 − δ) ≈ 2.3 × 105 . The variability in the
He II data confirms a temperature dependence to the turbulent transition within
boundary layer.
72 Y.S. Choi, M.R. Smith, and S.W. Van Sciver

Fig. 4. Drag coefficient vs. modified Reynolds number

Acknowledgments
We wish to thank the National High Magnetic Field Laboratory and the De-
partment of Energy, Division of High Energy Physics for their financial support.
Thanks to David K. Hilton for useful conversation and Scott Maier for technical
assistance.

References
1. P.L. Walstrom, J.G. Weisend II, J.R. Maddocks and S.W. Van Sciver, Turbulent
pressure drop in various He II transfer system components, Cryogenics 28, 101,
1988.
2. D.C. Samuels, Velocity matching and Poiseuille pipe flow of superfluid helium,
Phys. Rev. B 46, 11714, 1992.
3. C.F. Barenghi, D.C. Samuels, G.H. Bauer and R.J. Donnelly, Numerical evidence
for vortex-coupled superfluidity: Quantized vortex lines in an ABC model of tur-
bulence, Phys. Fluids 9, 2631, 1997.
4. M.R. Smith and S.W. Van Sciver, Measurement of the pressure distribution and
drag on a sphere in flowing He I and He II, Advances in Cryogenic Engineering,
Vol 43, 1473, 1998.
5. M.R. Smith, D.K. Hilton and S.W. Van Sciver, Observed drag crisis on a sphere in
flowing He I and He II, Physics of Fluids, Vol. 11, No.4, 1999.
6. A similar analysis was originally presented in Evolution and Propagation of Tur-
bulence in Helium II, Ph.D. Thesis, M.R. Smith, 1992, University of Oregon.
7. R.J. Donnelly, Quantized Vortices in Helium II, Cambridge, New York, 1991.
Experiments on Quantized Turbulence
at mK Temperatures

S.I. Davis, P.C. Hendry, P.V.E. McClintock, and H. Nichol

Department of Physics, Lancaster University, Lancaster LA1 4YB, UK.

Abstract. An experiment to investigate the free decay of quantized turbulence in iso-

topically pure superfluid 4 He at millikelvin temperatures is discussed. The vortices are
created by a vibrating grid, and detected by their trapping of negative ions. Preliminary
results suggest the existence of a temperature-independent vortex decay mechanism be-
low T ∼ 70 mK.

1 Background
The renaissance of interest in the turbulent hydrodynamics of HeII has led to
the realisation that, in many respects, it exhibits unexpected similarities to anal-
ogous flows in classical fluids at high Reynolds number [1,2,3]. Unlike a classical
fluid, HeII is well described by a two-fluid model, with a normal (dissipative)
component mutually interpenetrating with a superfluid (inviscid) component
with quantized circulation
  
h
κ = vs .dl = n (1)
m4
where the integral is taken around a loop enclosing the vortex, vs is the superfluid
velocity, m4 is the 4 He atomic mass and the quantum number n is an integer.
The flow properties of HeII in an open geometry are dominated by singly-
quantized vortex lines [4], linear singularities around which the superfluid flows
at tangential velocity vs . Unless velocities are kept extremely small, the liquid
flowing through a tube becomes filled with a tangled mass of such vortex lines.
Because of their quantization, they represent a particularly simple form of tur-
bulence. In that the vortex cores can be considered as part of the normal fluid
component, but the encircling superflow field in accordance with (1) is of super-
fluid component, vortices provide a weak coupling (mutual friction) between the
two components. So it is not at all clear, at first sight, why this complex liquid
system should ever behave like a single-component classical fluid. The question
has recently been discussed in considerable detail by Vinen [3]. One of the aims
of the present project is to establish the properties of the turbulent liquid when
it really does consist of just a single component, i.e. in the low temperature limit
where the normal fluid density is negligible.
In the conventional scenario at higher temperatures 1 < T < Tλ , a vortex
tangle can be maintained by the work done by the driving force, which could be
e.g. a pressure or temperature gradient for bulk flow, or for thermal counterflow

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 73–79, 2001.

c Springer-Verlag Berlin Heidelberg 2001
74 S.I. Davis et al.

(in which the normal and superfluid components move in opposite directions),
respectively. On removal of the driving force, the tangle decays according to the
Vinen [5] equation
dL  2
= −χ2 L (2)
dt m4
where L is the length of vortex line per unit volume and χ2 is a (weakly
temperature-dependent) dimensionless constant. The physical mechanism driv-
ing the decay has been discussed by Schwarz [6], who concluded that it involves
crossings and consequent reconnections of lines. The rapid self-induced motion
of the resultant sharp cusps through the normal fluid is dissipative, and causes
the rounding off of cusps with consequent line shrinkage. The presence of the
normal fluid component is thus a key component of the decay mechanism.
Details of the decay process, e.g. the existence of two distinct decay rates
[5,7], are not at all well understood, partly because it sometimes seems to in-
volve turbulence in the normal fluid component as well as in the superfluid. The
experiment that we describe below avoids this complication. Vortex decay is in-
vestigated at sufficiently low temperatures (in the mK range) that normal fluid
component as such is absent. Under these conditions, it is far from obvious how
the tangle will decay, but it appears that there are two possibilities –

1. Essentially the same decay process occurs as above 1K, but now driven by the
dilute phonon gas rather than by normal fluid, or
2. A new decay mechanism comes into play.

If the former applied, decay times would be expected to become exceedingly

long as T decreased, given that the vortex-phonon cross section is tiny [8] and
that the phonon density is falling as T 3 . The question is difficult to settle on
theoretical grounds, and best resolved by experiment.

2 Creation and Detection of Vortices

Above 1K vortices can conveniently be created [4] by thermal counterflow in
which a thermal gradient causes the two fluids to flow in opposite directions
above a critical velocity. Vortices are usually detected [4] by their attenuation
of second-sound, an entropy-temperature wave in which the two fluids move in
antiphase. Neither of these techniques can be used at mK temperatures, because
there is no normal fluid component. New techniques have therefore been required.
To create vortices we have employed a resonantly-excited circular grid [9].
The expectation was that, in the absence of normal fluid, the amplitude of
vibration would grow until a critical velocity was attained, after which the energy
drawn from the exciting field would be converted to vorticity.
To detect the vortices we have observed the attenuation of ion signals caused
by trapping of the ions on vortex cores as the ion clouds pass through the vor-
tex tangle. An essential requirement of the experimental design is the necessity
of preventing the ions from themselves creating vortex rings – converting to
 Quantized Turbulence at mK Temperatures 75

F ie ld
e m is s io n tip
T o p e le c tro d e
O s c illa tin g
H V g rid
B o tto m e le c tro d e

F ris c h g rid C o lle c to r

Fig. 1. The experimental arrangement (schematic). Some of the ions created by the
field emission tip may get trapped on vortices created previously by the oscillating grid,
thereby reducing the signal arriving at the collector. The perforated top and bottom
electrodes complete the double capacitor needed to excite oscillations of the grid, and
the Frisch grid screens the collector from the approaching charge.

large slowly-moving charged-vortex-ring complexes, and thus being lost from

the signal. An ion in HeII at mK temperatures, in the absence of normal fluid
component, will accelerate steadily under the influence of any electric fluid, how-
ever small, until it attains a critical velocity. It is necessary to ensure that this is
the Landau critical velocity vL , and not the critical velocity for vortex creation.
This can be accomplished by choosing negative (rather than positive) ions [10],
by applying pressures P > 11 bar [10], by using isotopically pure 4 He [11], and
by ensuring that the electric field is not too large [12]. Under these conditions
the ions do not create vortex rings but, rather, travel freely through the liquid
with an average speed slightly in excess of vL [13].
The mechanism by which a rapidly moving ion can get trapped on a vortex
under these conditions is unclear. The vortex presents an effective potential [14]
and, above 1K, normal fluid dissipation enables the ions to lose enough energy
to get trapped on the vortex core. In the absence of normal fluid, and with the
ion moving almost ballistically, like a free particle in a vacuum, we hoped that
the ions would lose enough energy to be trapped by exciting vortex waves.

3 The Experiment
The electrode structure used for the experiment is shown schematically in Fig. 1.
The operating procedure was performed in two stages. First, a high constant
voltage (usually 500 V) was applied to the vortex-generating grid and a periodic
driving voltage of ±270 V was applied to one of the adjacent plates. The drive
76 S.I. Davis et al.

was maintained for several seconds, to build up a tangle of vorticity. Secondly,

the potentials on the electrodes were adjusted so that the electric field would
draw ions from the field-emitter to the collector. The field-emitter was then
pulsed to create an ion cloud, which travelled down the cell. As it passed the
Frisch screen-grid it induced a signal in the collector which was amplified and
recorded using a Nicolet NIC-80 data processor. The sequence was then repeated,
ensemble-averaging the collector signals to enhance the signal/noise ratio.

4 Preliminary Results
Figure 2 shows a typical sequence of ion signals. The first signal is a reference,
recorded before the grid had been vibrated, and the others show how the signal
gradually recovered after the grid vibration had been completed. It is evident
that there is significant attenuation as a result of the grid oscillation – demonstat-
ing immediately that the technique described in the preceding sections enables
us both to create and to detect vorticity in the mK temperature range.

4 0 0 0

3 0 0 0

2 0 0 0

1 0 0 0

0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0

Fig. 2. A set of collector signals. The duration of each of them is ∼ 200 μs, and a
period of 1.5 s in real time separates each signal from its neighbour. The first signal is
for reference, recorded before the grid was vibrated.

In Fig. 3, the signal heights are plotted as a function of time for several
temperatures T . It appears that for T < 70 mK nothing changes, within ex-
perimental error: the decay mechanism is apparently temperature-independent
within this range.

5 Discussion
Our preliminary data are too scattered for us to be able to draw definite con-
clusions about the form of the decay and there is, in any case, no theoretical
 Quantized Turbulence at mK Temperatures 77

4 6 8 1 0 1 2 1 4 1 6 1 8

Fig. 3. Signal amplitudes, showing their evolutions as a function of time t, for several
temperatures.

form with which to compare. Nonetheless a plot of ln(S0 − S) against time t,

where S0 and S are repectively the unattenuated and attenuated signals, pro-
duces what is definitely a curve, showing that the decay is non-exponential. It
is also interesting to compare with the form of decay that occurs above 1 K in
the presence of normal fluid component. By integration of the Vinen equation
(2), it is straightforward to demonstate that
  −1
S0 A  κ 
ln = χ2 t + L−1i (3)
S κd 2π
where A is a constant containing the ion-vortex trapping cross-sections, Li is
the initial vortex line density at t = 0, and d is the length of the vorticity-
filled region. Thus a plot of [(ln(S0 /S)]−1 against t would be expected to yield
a straight line. It does so, within experimental error, as shown in Fig. 4.
It is interesting to speculate on the physical nature of the low temperature
T –independent decay mechanism inferred from the results of Fig. 3. The fact
that the decay is non-exponential would appear to rule out spontaneous decay
processes, e.g. where there was a constant probability per unit time that any
given element of vortex might emit a phonon and become incrementally shorter.
But the T –independence strongly suggests that the phonon gas plays no role,
given that the phonon density is falling as T 3 within the range of interest. One
possibility, perhaps, is that phonons are emitted during reconnections, leading
to line-shrinkage. Simulations by Tsubota et al [15] suggest that reconnections
are indeed the key to the problem, but that the resultant decay arises because
78 S.I. Davis et al.

1 2

1 0

0
4 6 8 1 0

Fig. 4. Plot of a typical set of data to test the applicability of equation (3). S and S0
are respectively the amplitudes of a signal and of the reference signal.

of the strongly kinked lines that are produced, resulting [3] in Kelvin waves and
a cascade of energy towards smaller and smaller length scales until it is radiated
as sound.
A difficulty in interpreting the results is that the absolute vortex line densities
are unknown, because the ion-vortex trapping cross-section is unknown under
the conditions of the experiment. We can obtain a very approximate estimate
from the measured linewith of the grid resonance which, for typical electrode
potentials and driving amplitudes implies that the energy dissipation of the grid
is (0.4±0.2)μW; insertion of this value in (2), on the assumptions that all of the
dissipation goes into vortex creation, χ takes the same value as above 1K and
that the vortex tangle remains mostly between the electrodes, yields a steady
state line density of ∼1010 m−2 . Any or all of the assumptions could be in error,
however, and the only unambiguous way to clarify the situation will be through
direct measurement of the ion-line trapping cross-section in a rotating cryostat
where the line density is known precisely.

6 Conclusions
In conclusion, we would emphasize the preliminary character of these results,
and the large number of unknowns. As pointed out above, we know nothing
about the spatial distribution of the vorticity, although we imagine that it stays
mostly between the top and bottom electrodes of the triple capacitor in Fig. 1.
 Quantized Turbulence at mK Temperatures 79

Nor do we know anything about the nature of (possible) temperature depen-

dences of the vortex-generation and ion-trapping mechanisms. Our ignorance of
vortex/ion trapping cross-section represents an even more serious lacuna, be-
cause it means that we cannot calculate absolute vortex line densities from the
ion signal attenuation.
Nonetheless, we can tentatively conclude, first, that it is possible to generate
and detect vorticity in HeII at mK temperatures and, secondly, that the decay
of quantized vorticity becomes temperature-independent below T ∼ 70 mK.

Acknowledgements

It is a pleasure to acknowledge helpful discussions with C F Barenghi, R J

Donnelly, L Skrbek and W F Vinen. The work was supported by the Engineering
and Physical Sciences Research Council (UK).

References
1. S.R. Stalp, L. Skrbek, R.J. Donnelly: Phys. Rev. Lett. 82, 4831 (1999)
2. C.F. Barenghi: J. Phys.: Condens. Matter 11, 7751 (1999)
3. W.F. Vinen: Phys. Rev. B 61, 1410 (2000)
4. R.J. Donnelly: Quantized Vortices in He II (Cambridge University Press, Cam-
bridge 1991)
5. W.F. Vinen: Proc. R. Soc. A 242, 493 (1957)
6. K.W. Schwarz: Phys. Rev. B 18, 245 (1978); 31, 5782 (1985); 38, 2398 (1988)
7. K.W. Schwarz, J.R. Rozen: Phys. Rev. Lett. 66, 1898 (1991)
8. A.L. Fetter: Phys. Rev. A 136, 1488 (1964)
9. M.I. Morell, M. Sahraoui-Tahar, P.V.E. McClintock: J. Phys. E: Sci. Instrum. 13,
350 (1980)
10. L. Meyer, F. Reif: Phys. Rev. 123, 727 (1961)
11. R.M. Bowley, P.V.E. McClintock, F.E. Moss, P.C.E. Stamp: Phys. Rev. Lett. 44
161 (1980)
12. R.M. Bowley, P.V.E. McClintock, F.E. Moss, G.G. Nancolas, P.C.E. Stamp: Phil.
Trans. R. Soc. Lond. A 307, 201 (1982)
13. D.R. Allum, P.V.E. McClintock, A. Phillips, R.M. Bowley: Phil. Trans. R. Soc.
Lond. A 284, 179 (1977)
14. R.J. Donnelly, P.H. Roberts: Proc. R. Soc. Lond. A 312, 519 (1969)
15. M. Tsubota, T. Araki, S.K. Nemirovskii, J. Low Temperature Phys. 119, 337
(2000)
Grid-Generated He II Turbulence
in a Finite Channel – Experiment

J.J. Niemela, L. Skrbek, and S.R. Stalp

Cryogenic Helium Turbulence Laboratory, Department of Physics,

University of Oregon, Eugene, OR 97403, USA

Abstract. We present experimental data on decaying turbulence, generated by tow-

ing a grid through a stationary sample of He II. We describe in detail the experimental
apparatus and physical principles that allow observation of up to six orders of magni-
tude of decaying vortex line density over three orders of magnitude in time using the
second sound attenuation technique.

1 Introduction
Superfluid turbulence has long been an area of study, with an emphasis largely on
flows created by applying a heat current in He II; i.e. on thermal counterflow[1].
It is an advantage that, experimentally, counterflow turbulence requires no mov-
ing parts for its generation. The connection of this type of flow with classical
turbulence, however, is not obvious. More recently, it has been of some interest to
explore this connection by generating turbulent flows in He II in a similar manner
as for classical fluids. In particular, turbulence created in the wake of a grid can
create nearly homogeneous and isotropic turbulence (HIT)[2], and application
of this procedure to He II has led to new insights and a few surprises[3,4,5]. In
particular, a deep similarity appears to exist between grid turbulence in classical
fluids and in He II, a quantum fluid.
In this article, we focus on the experimental apparatus and techniques used
to generate grid turbulence in He II, and discuss the observed decay of the vor-
tex line density behind a towed grid. Measurements of second sound attenuation
allow detection of up to six orders of magnitude of decaying vortex line density
L, which can be converted into roughly eight orders of magnitude of decaying
turbulent energy[6] - at present hardly a feasible goal for any laboratory experi-
ment on classical turbulence. In a companion article in this book, we will further
interpret the data in terms of classical hydrodynamics. The underlying quantum
nature of this experiment has been recently discussed by Vinen [7].

2 Experimental Setup

The schematic of the experimental apparatus is shown in Fig.1. The turbulence

is generated by towing a grid through a stationary sample of He II. We use a
65% open brass monoplanar grid of rectangular tines, 1.5 mm thick, with a mesh
size, M (tine spacing) of 0.167 cm[3,4]. The grid is attached to a central stainless

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 80–86, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Grid-Generated He II Turbulence 81

Linear Servo Motor

Motor Control
and Counter

16 bit ADC

200MHz Pentium
GPIB, Labview
G
P
I
B
Function Vacuum Tight Sliding Seal
B Generator
U Vacuum
Pump
S

Lock-in
Amplifier

Amplifier
60 mF Grid Generated
Turbulence
Grid

Second Sound
Bias Voltage Receiver
100 Volts DC
Second Sound
Transmitter

Second Sound
LR-110 Standing Wave
Resistance
Bridge
Germanium
Thermometer 1 cm x 1 cm x 29 cm
Channel
Heater
LR-130
Temperature
Controller

Fig. 1. Schematic of the experimental apparatus

steel pulling rod of diameter 0.24 cm which exits the cryostat via a pair of tight
sliding seals. The space between the seals is continually evacuated to prevent the
introduction of impurities inside the cryostat during the experiment. Above the
cryostat, the rod is attached to a computer controlled linear servo motor that
positions the grid with 0.01 cm accuracy and provides the towing velocity, vg ,
up to 2.5 m/s. This enables the exploration of a wide range of mesh Reynolds
numbers 2 × 103 ≤ ReM = vg M ρ/μ ≤ 2 × 105 , where μ is the dynamic viscosity
82 J.J. Niemela, L. Skrbek, and S.R. Stalp

of the normal fluid and ρ the total density. It is worth noting that the linear
servo motors, developed at the University of Oregon, accomplish this precise
positioning without the problems associated with electrical switching noise and
mechanical resonances that are characteristic of the more commonly used stepper
motors.
The channel of square cross section is 29 cm long and has a width d = 1 cm.
It was manufactured by an electroforming process with a tolerance of 25 μm and
a surface roughness less than 0.5 μm. The channel is suspended vertically in the
helium cryostat and during the measurement is totally submersed in superfluid
helium, which enters the channel via 16 holes, 500 μm in diameter, placed near
the top end. Except for these, the channel is leak-tight. There is no free surface of
the liquid inside the channel during the experiment, and second sound coupling
to the free surface of the bath is minimized by the use of the small diameter
holes for mass transfer between the channel and the bath.
The temperature is measured and controlled via a germanium resistance
thermometer and heater placed in the He II bath. The thermometer is calibrated
against the saturated vapor pressure with a temperature accuracy better than 1
mK and a resolution within 10 μK[4]. Details of possible temperature fluctuations
inside the channel that might occur during and just after pulling the grid are
unknown, as the thermometer is located outside the channel and its response
time is about 100 ms. For maximum cooling, the main bath is pumped via a 20
cm diameter line connected to a 300 CFM vacuum pump and a 1400 CFM roots
blower. This enables bath temperatures down to 1 K to be reached easily. It
is difficult, however, to assure temperature stability below about 1.1 K, so this
marks the lowest reliable temperature for the experiment. Experiments close to
the lambda temperature (above 2 K) require special care (see below) mainly due
to the strong temperature dependence of the second sound velocity, u2 .
To probe the quantized vortex line density resulting from pulling the grid,
we excite and detect second sound using vibrating nuclepore membranes 9 mm
in diameter mounted flush on opposing walls of the channel. These 6 μm thick
polycarbonate membranes have a dense distribution of 0.1 μm holes and on
one side is evaporated an approximately 95 Å thick layer of gold, which is then
pressed against the channel wall. This gold layer forms one electrode of a ca-
pacitor transducer, with the other being a brass electrode pressed towards the
opposite side of the membrane from outside the channel. Applying an ac signal
(∼ 0.3 − 1 VP P ) in addition to a dc bias (∼ 100 V) results in an oscillatory
motion of the membrane. In He II, the normal fluid is clamped by viscosity in-
side the small holes of the membrane, while the superfluid component passes
freely, thereby exciting second sound, an entropy wave in He II. Directly across
the channel the second sound wave produces a corresponding oscillation of the
other membrane (i.e., the receiver) and the induced signal is input to a lock-in
amplifier referenced to the transmitter frequency. The channel acts as a second
sound resonator. The excitation amplitude is adjusted to be the upper half of
the linear response range and the n-th harmonic- with n about 50- of the funda-
mental frequency is used: typically 30 − 40 kHz. A Lorentzian resonance peak is
 Grid-Generated He II Turbulence 83

obtained, having a full width at half maximum that is temperature dependent

and typically reaches values of Δ0 = 20 − 500 Hz without quantized vortices in
the channel.
To understand the use of second sound in detecting the vortex line density
and its relation to averaged rms vorticity, we consider the seminal work of Hall
and Vinen[8]. In experiments with a rotating container of He II, they observed
an excess attenuation of second sound in a direction perpendicular to the rota-
tion axis due to the presence of quantized vortices, αL = BΩ/2u2 . This extra
attenuation resulted from the scattering of the elementary excitations - normal
fluid - by the vortex lines and was absent for second sound propagating parallel
to the rotation axis. Here B is the dimensionless mutual friction parameter (B
generally depends both on temperature and frequency[9]) and Ω denotes the
angular velocity of rotation. It is now well known that the rotating bucket of He
II displays, on average, the same shape of the surface meniscus as any classical
fluid, since the superfluid mimics solid body rotation by creation of a lattice of
rectilinear quantized vortices aligned in the direction of the rotation axis. In this
case, the vorticity ω = 2Ω = κL, where κ is the circulation quantum (κ = h/m4 ,
where h is Planck’s constant and m4 the mass of the helium atom) and L is the
total length of the vortex line per unit volume. It is often assumed that this rela-
tion between ω and L holds in general. By considering a second sound resonance
as an infinite series of reflected waves in a rotating cavity, the extra attenuation
(in a limit of small attenuation) due to quantized vortices becomes [4]
 
BκL πΔ0 A0
αL = = −1 (1)
4u2 u2 A
where A and A0 are the amplitudes of the second sound standing wave resonance
with and without vortices present, respectively. We can extend this formula
to the case of a homogeneous vortex tangle, taking into account that vortices
oriented parallel to the second sound propagation do not contribute to the excess
attenuation. Then we have[3]
 
16Δ0 A0
L= −1 (2)
Bκ A

It can be shown[4] that for arbitrary attenuation one has to use the more
general formula
 
8u2 1 + p2 P + 2p2 P + p4 P 2
L= ln √ (3)
πBκd 1 + P + 2P + P 2

where p = A0 /A and P = 1 − cos(2πdΔ0 /u2 ) that for small dΔ0 /u2 reduces
to (2). It is essential to use formula (3), as using the approximate formula (2)
at some experimental conditions leads to results that are more than an order of
magnitude off!
It is important to consider the time response of the measuring system with
regard to the finite velocity of second sound. Our calculations show that in most
84 J.J. Niemela, L. Skrbek, and S.R. Stalp

1 0
6
T = 1 .7 5 K

) 5
-2

1 0 p o w e r -3 /2
v o r te x lin e d e n s ity ( c m

4
1 0

3
1 0

2
1 0

1
1 0
0 .1 1 1 0 1 0 0
tim e ( s )
Fig. 2. The log-log plot of the decaying vortex line density versus time after grid passes
2 mm above the measuring volume. Each decay curve represents an average of three
identical pulls. The decay curves, in order, correspond to ReM = 2×105 (the uppermost
one), 1.5 × 105 , 105 , 5 × 104 , 2.5 × 104 , 104 , 5 × 103 and 2 × 103 . For each ReM , the
decaying vortex density displays an inertial range with power law exponent -3/2.

cases (except close to the lambda temperature) there is negligible error intro-
duced into the deduced vortex line density, for the following reasons. Immediately
after the grid is towed through the measuring volume the quality factor is very
low, of order unity, and the detecting system can be described rather as a second
sound pulse technique with the characteristic time response given by the time
of flight d/u2 ∼ 10−3 s, where u2 ∼ 20 m/s [10]. As the turbulence decays, the
characteristic time constant increases with the (temperature dependent) qual-
ity factor. Without the vortex tangle, the typical linewidth of the second sound
resonance is 20-500 Hz, the typical frequency used is 30-40 kHz, so the quality
factor reaches 60-2000 and the time response gradually rises to about 0.1-1 sec
at the very end of the decay, where it constitutes an error of less than 1%.
There are also other time restrictions than the finite velocity of second sound.
As it takes a time τg ∼= d/vg to tow the grid through measuring volume, we use
only that data obtained on the time scale longer than τg and also exceeding 8τLI ,
where τLI is the time constant of the lock-in amplifier used for detection of the
amplitude of the second sound signal. Another time scale restriction involves
consideration of how soon the flow can be assumed as nearly HIT [6]. It was
discussed in[11] that a physical criterion to estimate this time leads to about
1-2 turnover times of the largest eddies present in the flow, i.e., those of the size
of the channel. It follows that the minimum time needed to assume nearly HIT
conditions is about 1-2 widths downstream from the grid, or using the Taylor
 Grid-Generated He II Turbulence 85
-4
2 .0 x 1 0

-4
1 .5 x 1 0
/s )
2
ν ' (c m

-4
1 .0 x 1 0

-5
5 .0 x 1 0
1 .1 1 .2 1 .3 1 .4 1 .5 1 .6 1 .7 1 .8 1 .9 2 .0 2 .1 2 .2
T e m p e ra tu re (K )
Fig. 3. The effective kinematic viscosity of the superfluid turbulence ν  deduced from
vorticity decay data at various temperatures, and assuming a value of the Kolmogorov
constant consistent with classical experiments. The dashed line represents ν = η/ρ
which has a dissimilar temperature dependence indicative of quantization effects, where
η is the normal fluid viscosity and ρ is the total density of helium II.

frozen hypothesis, the time it takes the towed grid to pass 1-2 widths of the
channel.
The data acquisition process can be briefly described as follows. The cryostat
is filled with liquid helium, the bath being pumped and the temperature con-
trolled to a desired value. With the grid “parked” towards the top of the channel,
the second sound resonance curve is measured and fitted to a Lorentzian, giving
the value of Δ0 . The grid is then lowered to the bottom and after a necessary
waiting time (typically 2 minutes) pulled through the channel such that its veloc-
ity is constant and equal vg for at least 5 cm below and above the experimental
volume. It is then slowed down and smoothly ”parked” against the top again.
The data acquisition is triggered when the grid passes a predetermined position
2 mm above the measuring volume, as determined by a signal from a photodiode.
Typically 100 s of data are recorded at a rate of 100 Hz from the output signal
of the lock-in amplifier, representing the recovering second sound standing wave
amplitude, A(t). After another waiting period to ensure that the turbulence has
decayed down to a negligibly low level, the reference amplitude A0 is read. The
decaying vortex line density is then calculated using formula (3). It implicitly
assumes that there is no additional attenuation of the second sound that might
be caused by a turbulence in the normal fluid.
Since the second sound is transmitted and detected via membranes on oppos-
ing sides of the channel, we obtain information from a measuring volume of order
d3 ∼
= 1 cm3 . It is this natural integration that bypasses tedious statistical analysis
involved in any local velocity measurements in conventional turbulence, provides
86 J.J. Niemela, L. Skrbek, and S.R. Stalp

enormous sensitivity and unprecedented dynamical range of the method, making

it very useful and complementary to classical turbulence studies.
Examples of decay data and the effective kinematic viscosity of the superfluid
turbulence ν  , are shown in Figs. 2 and 3 respectively. The decay curves represent
up to six orders of magnitude of decaying vortex line density over three decades
of time. The overall form of the decay does not change with temperature. At any
temperature, after some initial period depending on ReM , the decay curves tend
to collapse and display the power law decay L ∝ t−3/2 [5,12]. We note if vorticity
and L are assumed to be related in a similar manner as discussed above for the
rotating bucket experiments, then this feature and others are understandable in
terms of a classical spectral decay model[6,11,12], further discussed in a compan-
ion article in this book. Quantum effects show up in the measured temperature
dependence of the effective kinematic viscosity (see Fig. 3). Finally, we have had
many useful discussions at this workshop concerning complementary measure-
ments to probe the normal fluid turbulence and/or energy dissipation directly,
and also extending the protocol to near zero temperature to complement the
work of McClintock’s group reported here. These ideas are presently receiving
attention.

Acknowledgements

The towed grid He II experiment has been developed at the University of Oregon
over many years. We would like to acknowledge R.J. Donnelly, W.F. Vinen and
M.R. Smith for their valuable contributions to the conception and design of this
experiment. This research was supported by NSF under grant DMR-9529609.

References
1. R.J. Donnelly: Quantized vortices in helium II. Cambridge University Press (1991)
2. G. Comte-Bellot, S. Corrsin: J. Fluid Mech. 25, 657 (1966); 48, 273 (1971)
3. M.R. Smith: Evolution and propagation of turbulence in helium II. PhD Thesis,
University of Oregon, Eugene (1992)
4. S.R. Stalp: Decay of grid turbulence in superfluid helium. PhD Thesis, University
of Oregon, Eugene (1998)
5. M.R. Smith, R.J. Donnelly, N. Goldenfeld, W.F. Vinen: Phys. Rev. Lett. 71, 2583
(1993)
6. L. Skrbek, J.J. Niemela, R.J. Donnelly: Phys. Rev. Lett. 85, 2973, (2000)
7. W.F. Vinen: Phys. Rev. B 61, 1410 (2000)
8. [Link], W.F. Vinen: Proc. Roy. Soc. London A238, 204 (1954); 238,215(1954)
9. C.F. Barenghi, R.J. Donnelly, W.F. Vinen: J. Low Temp. Phys. 52, 189 (1983)
10. R. J. Donnelly, C. F. Barenghi: [Link]. Chem. Data 27, 1217 (1998).
11. L. Skrbek, S.R. Stalp: Phys. Fluids 12, 1997 (2000)
12. S.R. Stalp,L. Skrbek, R.J. Donnelly: Phys. Rev. Lett. 82, 4831(1999)
Intermittent Switching Between Turbulent
and Potential Flow Around a Sphere
in He II at mK Temperatures

Michael Niemetz, Hubert Kerscher, and Wilfried Schoepe

Institut für Experimentelle und Angewandte Physik, Universität Regensburg,

D-93040 Regensburg, Germany

Abstract. Intermittent switching between potential flow and turbulence is observed

with an oscillating sphere in HeII below 0.5 K, where there is no normal fluid component
and no viscosity. The remaining dilute phonon gas is in the ballistic regime and therefore
turbulence in the pure superfluid can be investigated. The amplitude of the driven
oscillations is a measure of the damping which in case of potential flow is due to
residual ballistic phonon scattering or, when the flow is turbulent, is due to a large
nonlinear turbulent drag. In an intermediate range of driving forces the flow is observed
to be unstable, intermittently switching between both patterns. We have investigated
this phenomenon down to 25 mK and have made a statistical analysis of the time series
measured at various constant driving forces and temperatures. We obtain a temperature
independent probability density for switching, different for both directions. We find a
regime of metastable laminar flow whose lifetime is limited by natural radioactivity
and cosmic rays.

1 Experiment
The experimental setup used in our investigations consists of a ferromagnetic mi-
crosphere (radius r = 124 μ m, m = 27 μ g) suspended between two niobium elec-
trodes by superconducting levitation. While cooling the horizontally arranged
electrodes forming a parallel plate capacitor (distance d = 1 mm, diameter 2 mm)
a dc-voltage of several hundred volts is applied to the capacitor giving rise to
an electric charge at the surface of the electrodes and the sphere. As the elec-
trodes become superconducting, the magnetic sphere is repelled and levitates at
an equilibrium position between the two electrodes. Horizontal stability is pro-
vided by trapped flux in the electrodes. As the sphere carries an electric charge
q vertical oscillations (f ≈ 150 Hz) can be excited by applying a resonant ac
electric field to the capacitor corresponding to a force F = q · Uac /d on the
sphere. The oscillations can be detected by measuring the current I = v · q/d
induced in the electrodes of the capacitor by the moving charge. The space be-
tween the superconducting electrodes is filled with pure 4 He (3 He concentration
≤1 ppb). This setup provides a simple geometry without disturbance by mechan-
ical suspension elements and a low background dissipation (Q-factors above 106
are achieved when the cell is evacuated). We measure the velocity amplitude of
the oscillating sphere for different driving forces at different temperatures. For
a more detailed description of the experimental technique and its applications,
see [1–4].

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 87–94, 2001.

c Springer-Verlag Berlin Heidelberg 2001
88 M. Niemetz, H. Kerscher, and W. Schoepe

2 8 3 5
3 0

2 5

2 6 2 0
s ta b le 1 5
la m in a r 1 0

2 4 flo w 5
0
v (m m /s )

in te r m itte n t 0 2 0 0 4 0 0 6 0 0 8 0 0

s w itc h in g
2 2

v c 2
2 0
v c 1
F c
s ta b le tu r b u le n c e
1 8
0 5 0 1 0 0 1 5 0
T = 3 0 0 m K F (p N )

Fig. 1. Three regimes of flow around the sphere: For low driving forces there is stable
laminar flow (•), resulting in a linear dependence v(F ) = F/λ. For large driving forces
there is stable turbulent flow () resulting in large dissipation and low oscillation
amplitudes. (The solid line is a fit of the turbulent v(F ) dependence derived in the text.)
Between these two regions there is a regime where both flow patterns are unstable and
which extends from vc1 over an interval Fc of driving forces to vc2 where turbulence
becomes stable. In this regime the system switches between laminar and turbulent
flow intermittently. The inset shows a larger range of driving forces and velocities. The
enlarged region is marked by a box

2 Results
The inset of Fig. 1 shows a typical result of the experiment. For small driving
forces the flow around the sphere remains laminar corresponding to a linear drag
force λv, which is given by ballistic scattering of phonons (left regime in Fig. 1).
For large driving forces there is stable turbulent flow instead, accompanied by
large nonlinear dissipation. Between these two regimes we find an interval of
driving forces, where neither laminar nor turbulent flow is stable but where
instead the fluid switches between these states intermittently.

2.1 Stable Turbulent Flow

Fully developed turbulence in a classical fluid causes a drag force
|Fd | = γv 2 , γ = Cd πr2 /2 (1)
(Cd ≈ 0.4) on a sphere moving through the fluid. As we perform an ac experiment
where v(t) = v sin(2πf t) we employ the energy balance between the energy input
per half period T /2 at resonance and the dissipation


T /2 
T /2

F sin(ωt) v sin(ωt) dt = Fd (v(t)) v sin(ωt) dt (2)

0 0
 Intermittent Switching Between Turbulent and Potential Flow 89

1 0 0
Fig. 2. Turbulent flow around
8 0 the sphere: Data points taken
up to one hundred times larger
v (m m /s )

6 0 driving forces than in Fig. 1

are well described by the tur-
4 0 bulent drag force introduced
in the text. The shift of the
2 0 T = 1 0 0 m K apex of the parabola towards
negative F values can clearly
0 be observed
0 2 4 6 8 1 0 1 2
F (n N )

to find the corresponding F (v) dependence F (v) = 8γv 2 /3π. In Fig. 2 a larger
range of driving forces is shown, providing a more complete view of the turbulent
regime. The solid line is a fit, using a similar drag force as in (1) but reduced by
a constant value:
|Fd | = γ(v 2 − v02 ) = γv 2 − F0 . (3)
By applying the energy balance (2) this drag force corresponds to
 
8γ 3 2
F (v) = v − v0 = γ  (v 2 − vc1
2 2
) (4)
3π 2

which perfectly fits the data points obtained in the superfluid (Fig. 2). A similar
drag force has been calculated for turbulent flow of a two dimensional dilute
Bose-Einstein condensate around a cylinder [5]. Even if these calculations are
not directly applicable to liquid helium, there might be a similar mechanism in
both cases for the reduced drag force. In our case turbulent drag exists only
above a critical velocity vc1 = 19.4 mm/s (s. Fig. 1).

2.2 Intermittent Switching

In contrast to classical fluids there is no smooth transition from laminar flow to
fully developed turbulence in the superfluid, as reported earlier [2]. Instead, the
fluid switches between laminar and turbulent flow intermittently. We recorded
the velocity amplitude of the oscillations as a function of time while holding tem-
perature and driving force constant. Figure 3 shows three sections of 13 minutes
out of 4 h time series obtained for different driving forces at 300 mK. At the
lowest driving force (series (a)) we observe quite long laminar phases, where
the velocity amplitude exponentially approaches the equilibrium value given by
phonon drag and driving force. The laminar phases are interrupted by turbulent
phases intermittently and the velocity amplitude drops to the lower value vt
given by the large nonlinear turbulent drag. When using a larger driving force
(series (b)) we find a steeper increase of the velocity amplitude corresponding
to a higher equilibrium value that is not reached any more, as the lifetimes of
90 M. Niemetz, H. Kerscher, and W. Schoepe

3 0 a ) 4 7 p N
2 8
2 6
2 4
2 2
2 0
1 8

3 0 b ) 5 5 p N
2 8
v (m m /s )

2 6
2 4
2 2
2 0
1 8

3 0 c ) 7 5 p N
2 8
2 6
2 4
2 2
2 0
1 8

tim e ( 1 0 0 s / d iv is io n )
Fig. 3. Random sections of 800 s out of a 4 h time series of the oscillation velocity
amplitude for three different drives at 300 mK. Series (a) was taken at the lowest
driving force, showing the velocity increase during laminar phases and its exponential
saturation at a level given by phonon drag and driving force. Laminar phases are
interrupted intermittently by turbulent phases accompanied by a sharp drop of the
oscillation velocity amplitude to an equilibrium value given by the large turbulent drag
and driving force. Series (b) was taken using a larger driving force, resulting in a higher
saturation value (that is not reached any more) and slightly longer turbulent phases.
At the largest drive (series (c)) there is turbulent flow most of the time, interrupted
by laminar phases

the laminar phases are shorter than in series a. Taking a closer look at the life-
times of the turbulent phases shows that they in turn are longer than in series
a. Increasing the driving force further results in turbulent flow most of the time,
interrupted by short laminar phases. As the switching between both flow pat-
terns is intermittent, the results are discussed in terms of a statistical analysis
of the velocity amplitudes reached during laminar phases and their lifetimes as
well as the lifetimes of the turbulent phases. We apply reliability theory [6] for
the evaluation of the time series.

2.3 Turbulent Phases

A typical distribution of lifetimes of turbulent phases is shown in Fig. 4a, where

the number P of turbulent phases exceeding a certain lifetime t is plotted versus
t. The data points are described very well by a straight line in the semilogarithmic
plot, corresponding to exponentially distributed lifetimes exp(−t/μ). The slope
of the straight line gives the mean lifetime μ. We have performed this analysis for
many time series at different temperatures ranging from 28 mK up to 400 mK and
 Intermittent Switching Between Turbulent and Potential Flow 91

1 0 0 0
a ) b )
1 0 0
n u m b e r o f tu r b u le n t p h a s e s

1 0 0

μ (s )
1 0 1 0 3 2 m K
1 0 0 m K
T = 3 0 0 m K 2 0 0 m K
F = 5 9 p N 3 0 0 m K
4 0 3 m K
1
1
0 1 0 2 0 3 0 4 0 5 0 2 0 3 0 4 0 5 0
t (s ) F -λ v t (p N )

Fig. 4. Analysis of the lifetimes of the turbulent phases. (a) The number of turbu-
lent phases living longer than a certain time t shows an exponential decay (solid line)
corresponding to exponentially distributed lifetimes. The reciprocal slope of the distri-
bution in the plot gives the mean lifetime μ. (b) Mean lifetimes of turbulent phases for
different driving forces and temperatures. There is no temperature dependence, but a
strong dependence on the driving force. The solid line is a fit of a fourth-power law
divergence of the mean lifetime μ at a critical driving force value

in a wide range of driving forces and have always found exponentially distributed
lifetimes. A comparison of the mean lifetimes obtained is shown in Fig. 4b. The
values of μ are independent of temperature and collapse to an universal drive
dependence if the strongly temperature dependent laminar drag λv (where v = vt
is the velocity amplitude of the turbulent phase) is subtracted from the external
driving force. The mean lifetime of turbulent phases increases with the driving
force and diverges at a critical value Fc = 54 pN approximately with a fourth-
power law and stays infinite at larger drives corresponding to turbulent velocities
above vc2 , see Fig. 1. The power dissipated at Fc is 0.6 pW, corresponding to the
production of ≈ 1 mm vortex lines per half period. This is equivalent to producing
a vortex ring with a diameter of ≈ 1.4 times the diameter of the sphere.

2.4 Laminar Phases

A similar analysis can be performed for the laminar phases, but now it is conve-
nient to analyze first the amplitudes reached during laminar phases. Figure 5a
shows the number of laminar phases which exceed a given amplitude Δ v = v −vt .
The data are perfectly described by a parabola in the semilogarithmic plot,
P (Δ v) = P (0) exp −(Δ v/vw )2 , which means that the probability density func-
tion of the amplitudes corresponds to a Weibull distribution [6]. This result
holds for all driving forces and temperatures and the fitting parameter vw is
constant (Fig. 5b). The results for a given experiment fit well into a 10% band-
width around a mean value. In order to extract the probability of breakdown
of the metastable laminar flow it is necessary to analyze the lifetimes t of the
laminar phases, i.e. P (t) = P (Δ v(t)). The failure rate Λ(t) (i.e. the probability
92 M. Niemetz, H. Kerscher, and W. Schoepe

7
2 8 m K
a ) b ) 1 0 0 m K 2 0 0 m K
3 0 0 m K 4 0 3 m K
1 0 0 6
n u m b e r o f la m in a r p h a s e s

+ 1 0 %

(m m /s )
5
4 .7 9
1 0

w
v
-1 0 %
4
T = 3 0 0 m K
F = 5 5 p N

1 3
0 2 4 6 8 1 0 1 2 0 1 0 2 0 3 0 4 0 5 0
Δ v (m m /s ) F -λ v t (p N )

Fig. 5. Analysis of velocity amplitudes reached during laminar phases. (a) The number
of laminar phases having an amplitude larger than Δ v shows a quadratic dependence
on Δ v in this semilogarithmic plot (solid line), corresponding to a Weibull probability
density of velocities. (b) Fitting parameter vw for different driving forces and temper-
atures. We find no systematic dependence on these parameters. The values vary in a
10% bandwidth around a mean value of 4.8 mm/s

of breakdown after the laminar phase has survived for a time t) is given by [6]
 2
d d Δ v(t) 2
Λ(t) = − ln P (Δ v(t)) = = 2 Δ v Δ v̇ . (5)
dt dt vw vw
Because the velocity amplitude is exponentially approaching an equilibrium
value with a time constant τ = 2m/λ, this result has several implications: First,
as Δ v̇ is constant for small values of Δ v (or t  τ ), the failure rate is increasing
with the velocity amplitude of the sphere. Second, with the velocity reaching the
equilibrium value (or t  τ ), Δ v̇ goes to zero, and so does Λ(t). This means that
the laminar phase, having survived for many τ and having reached its stationary
velocity amplitude, will live forever although the velocity is clearly above vc1 . Ex-
perimentally, however, this is not exactly true. This can be seen in Fig. 6, where
the probability distribution for large lifetimes is shown. The decrease slows con-
siderably down for lifetimes greater than approximately 100 s (corresponding to
3τ at 300 mK), but the distribution does not approach a constant value. Instead,
a constant failure rate is found, leading to a mean lifetime of 25 min (indicated by
the straight line in Fig. 6). Obviously, there must be another mechanism causing
the breakdown of those long lived laminar phases. We can exclude mechanical
vibrations or acoustic noise to be the origin, as we tried to destroy such long
laminar phases by slamming the door, jumping on the floor, or even refilling
helium into the cryostat. But placing a small radioactive source (60 Co, 74 kBq)
outside the dewar, had a dramatic effect on the lifetimes. As can be seen from
Fig. 6, the mean lifetime changes by a factor 8.3 from 25 minutes to 3.0 minutes.
We have measured the dose rate of the source at the position of the measuring
cell inside the cryostat (taking into account a measured 20% loss in the dewar
 Intermittent Switching Between Turbulent and Potential Flow 93

walls) to be 440 nGy/h (± 5%). Comparing this value with a measured dose rate
due to natural background radiation in our laboratory of 50 nGy/h (± 10%),
which is typical for our area, we obtain an increase of the dose rate due to the
source by a factor of (440 + 50)/50 = 9.8 This compares well with the ratio of
lifetimes of metastable laminar phases obatined in the experiment. Therefore, it
is obvious that natural background radioactivity limits the lifetime of metastable
laminar phases above the critical velocity vc1 . This effect may be attributed to
local vorticity generated by ions produced by radiation, inducing the breakdown
of the metastable laminar flow around the sphere.

3 Conclusion
The turbulent flow of the superfluid causes a drag force on the sphere which
is very similar to turbulence in classical fluids except for a constant offset in
analogy with a dilute Bose-Einstein condensate. Furthermore, we have found
that the transition from potential flow to turbulent flow around a sphere in
superfluid 4 He at mK temperatures occurs by intermittent switching between
both flow patterns instead of the gradual transition observed in viscous fluids. A
statistical analysis of this switching phenomenon has shown that the lifetimes of
turbulent phases diverge at a critical driving force. The probablity for breakdown
of the laminar phases has been obtained. Finally, there exist metastable laminar
phases above the critical velocity vc1 . Their lifetime is limited only by natural
background radioactivity [7].
We have repeated our experiment several times by heating the measuring cell
above Tc of niobium in order to prepare a new levitating state of the sphere. All
our observations were reproducible. The three quantities vc1 , vw and Fc which
we expect to be affected by the properties of the sphere have standard deviations
of 8%, 13% and 15%, respectively, probably due to asymmetries of the surface
of the sphere.
We understand now why above 0.5 K the intermittent switching changes into
the hysteretic behavior observed earlier [1,2]: the velocity increases Δ v become

Fig. 6. Influence of radioac-

1
d r iv in g fo r c e = 4 7 p N tivity on the lifetime of lam-
T = 3 0 0 m K inar phases. By adding a ra-
n o r m a liz e d n u m b e r o f la m in a r p h a s e s

dioactive source the mean life-

time for long living lami-
nar phases (t τ = 31 s
n a tu r a l b a c k g r o u n d r a d ia tio n
at 300 mK) is reduced from
0 ,1 25 minutes to 3.0 minutes.
The mean lifetimes were ob-
tained by analyzing the ap-
6 0
a d d itio n a l C o s o u rc e proximately exponential tail
at large lifetimes (straight
0 2 0 0 4 0 0 6 0 0 8 0 0 1 0 0 0 1 2 0 0 lines)
t (s )
94 M. Niemetz, H. Kerscher, and W. Schoepe

very small at higher phonon drag which implies a very low failure rate of the
laminar phase. But if it fails (i.e. when the driving force F is largely increased)
the following turbulent phase is stable because the critical drive is exceeded.
With our present results in superfluid 4 He it appears extremely promising to
extend these experiments to superfluid 3 He in order to investigate the transition
to turbulence in this very different quantum fluid.

References
1. J. Jäger, B. Schuderer, and W. Schoepe: Physical Review Letters 74, 566 (1995).
2. J. Jäger, B. Schuderer, and W. Schoepe: Physica B 210, 201 (1995)
3. P. Eizinger, W. Schoepe, K. Gloos, J.T. Simola, and J.T. Tuoriniemi: Physica B
178, 340 (1992)
4. M. Niemetz, W. Schoepe, J.T. Simola, and J.T. Tuoriniemi: Physica B 280, 559
(2000)
5. T. Winiecki, J.F. McCann, and C.S. Adams: Physical Review Letters 82, 5186
(1999)
6. B.V. Gnedenko, Yu.K. Belayev, and A.D. Solovyev: Mathematical Methods of
Reliability Theory (Academic Press, New York, 1969)
7. M. Niemetz, H. Kerscher, and W. Schoepe: [Link]
mat/0009299 and to be published
Vortex Filament Methods for Superfluids

David C. Samuels

Dept. of Mathematics, Univ. of Newcastle, Newcastle upon Tyne, NE1 7RU, UK

1 Introduction
Vortex filaments are an idealized form of rotational flow where the vorticity
is confined to a small core region, of radius a, around a one dimensional line
embedded in the three dimensional flow. Outside of this core region the flow is
potential. When the dynamics of the core size are not important these objects
may also be referred to as vortex lines.
In classical fluid mechanics, by which I mean solutions of the Navier-Stokes
or Euler equations, vortex filaments are a useful tool for understanding the
geometry and dynamics of a flow. But after an initial popularity in the early
1980’s [1] [2] [3] [4] [5], the use of vortex filament methods fell out of favor
in classical fluid mechanics. Though there has been some slight resurgence in
this method recently [6] [7] due to the rapidly increasing computational power
available and the development of new computational algorithms, direct numerical
simulations and large eddy simulations have become the methods of choice for
calculating the motion of fluids. One reason for the decreased use of vortex
filament methods is that while they give a clear and intuitive understanding of
a flow through the easy visualization of the vortex filaments, this representation
was often just a rough cartoon of the true flow. Vortex filaments are only a
convenient idealization in a classical flow. The vorticity in a realistic classical
flow rarely takes the form of clearly discrete vorticity filaments.
But in superfluids like helium II vortex filaments are real [8]. Due to the
quantization of circulation, vorticity in a helium II flow can only exist within
vortex filaments with a core size of a. Since this core size is very small in helium
II, about 1 Angstrom, the thin-core vortex filament idealization is actually a
very accurate description of the true superfluid flow.
There are a few special qualities of the superfluid vortex filament [8] that
make it even simpler than the standard model of a vortex filament in a classi-
cal fluid. The circulation around each superfluid vortex filament in helium II is
set by quantum mechanics to be an integral multiple of κ = 9.97x10−4 cm2 /sec.
Since an n quantum vortex filament contains more energy than n single quan-
tum vortices, it is generally assumed that only single quantum vortex filaments
will be commonly observed. Thus all helium II superfluid vortices have identical
circulation. The core size a is also determined by quantum mechanics and is
closely related to the concept of a healing length, the length scale required for a
wave function to change from its bulk value to zero. In the classical fluid vortex
filament the core size is an important variable and much of the complications,

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 97–113, 2001.

c Springer-Verlag Berlin Heidelberg 2001
98 D.C. Samuels

and the interesting dynamics, of classical vortex filaments are due to changes in
the core size along a filament [9]. Without a variable core size, the dynamics of
a quantized vortex filament are much simpler than those of its classical counter-
part. Of particular importance is the effect of the constant core size on vortex
stretching. When a classical vortex filament is stretched, the core size shrinks
and the local vorticity rises along the filament. When a quantum vortex filament
is stretched, it can only elongate with no change in core size. Due to this dif-
ference, many prefer to reserve the phrase “vortex stretching” for the classical
process, and will say that vortex stretching does not occur for individual quan-
tized vortex lines. One should note, however, that a bundle of parallel quantized
vortex filaments can undergo vortex stretching, when the core size is interpreted
as the radial size of the bundle of filaments.
Many of the complications of classical vortex filament calculations have to
do with vortex stretching and with other core dynamics which are assumed
negligible in helium II quantized vortex filaments. We thus have a much simpler
system to deal with, both in terms of physical behaviour and in terms of the
computation of the motion of the filaments. My discussion in this paper will be
concerned directly with the superfluid vortex filaments of helium II, and when
necessary I will take the values of constants to be those of helium II.

2 Vortex Filament Motion

We represent the superfluid vortex filament by a series of mesh points distributed
along the centerline of the vortex filament. The motion of the vortex filaments
is calculated as the motion of each of these mesh points.
Since the vortex core size is on the atomic scale, it is usually assumed that
the effective mass of the vortex is negligible and the motion of the vortex can
be found by setting the sum of the forces at each point on the vortex filament
to be zero and solving the resulting equation for the velocity of the vortex line
[8]. From these procedure we get the basic equation of motion for the superfluid
vortex filament,

dS/dt = V S +V I +αS  ⊗(V N −V S −V I )−α S  ⊗[S  ⊗(V N −V S −V I )] (1)

where S is a position of a point on the vortex line, V S is the local superfluid

flow due to all non-vorticity sources, V I is the superfluid velocity due to the
presence of the vortex filaments (the induced velocity). S  is the derivative of
S with respect to arclength along the vortex line. V N is the local normal fluid
velocity and α and α are temperature dependent mutual friction parameters. It
is common to claim that α > α and then neglect the last term in the equation
of motion. However, this inequality is not true for temperatures very near the
lambda transition and there should also be the more general concern that since
these α and α terms in (1) are at right angles to each other it does not strictly
matter that one term may be of smaller magnitude than the other. The α term
produces a qualitatively different type of motion of the vortex line than does the
α term, and it should therefore not be discarded in an offhand manner.
 Vortex Filament Methods for Superfluids 99

The two fluid model presents the motion of helium II as a superposition

of two fluids, the normal fluid and the superfluid. The vortex filaments that
concern us here are vorticity structures of the superfluid. They interact with
the normal fluid through the mutual friction force, and the motion of these
vortex filaments can be strongly affected by the flow in the normal fluid. The
reverse is also true; the presence and motion of the superfluid vortex filaments
can strongly affect the flow of the normal fluid. This coupled interaction between
the two fluids is the subject of many of the articles in this volume. Calculating
the fully coupled motion of the entire two-fluid system is a very complicated
problem (which is discussed more fully in the article by Idowu et al). While
work on the fully coupled calculations is just beginning, most superfluid vortex
calculations up to this time have only been concerned with the motion of the
superfluid vortices without considering the response of the normal fluid flow
to the mutual friction force. This kinematic approach requires us to define a
normal fluid flow, with spatial and possibly even time dependence, and use (1) to
calculate the motion of the vortex line due to its self induced motion, advection
in the superfluid potential flow VS and the response to mutual friction with
the normal fluid flow VN . This is the problem that we will consider in this
article. The vortex filament techniques discussed here are also needed for the
fully coupled calculations described elsewhere in this volume.

2.1 The Biot–Savart Law and the Local Induction Approximation

The first difficulties in the vortex filament method come from the definition of
the term V I in the equation of motion. This term represents the advection of
the point on the vortex filament by the velocity field due to all the superfluid
vortex filaments present in the flow. This advection velocity at the point r is
given by the Biot-Savart law.

κ (S − r) ⊗ dS
V I (r) = (2)
4π |S − r|3

The line integral is taken over the entire length of vortex filament present in
the fluid. The Biot-Savart law can be used to find the flow at any point r in
the fluid, but the immediate difficulty with this equation comes from its use
in calculating the advection velocity in (1). In this case, the point r lies on
the vortex line and the Biot-Savart law contains a singularity at that point.
To heal this singularity we must include some aspect of the core structure of
the vortex filament. Many different methods of de-singularizing the Biot-Savart
law have been used in classical fluid dynamics [10] [11] [12] [13], where the core
dynamics of the filaments can often be quite complicated and important, but in
superfluid vortex methods I have only seen one method used [14]. This method
breaks the line integral into two parts; an integral over a local neighborhood
around the point r (this contains the singularity), and the rest of the Biot-Savart
integral (this part is non-singular). Then the local integral must be replaced by
an algebraic approximation that takes into account the core size of the filament.
100 D.C. Samuels

It is convenient to split up the Biot-Savart integral using the local meshing of

the vortex line. In calculating the motion of the J’th mesh point on the filament,
the local section of the integral is taken to be section containing the mesh points
J − 1, J, and J + 1. By calculating the Biot-Savart integral over a circular vortex
ring excluding the section of line between the mesh point J − 1 and J + 1, and
then subtracting this value from an analytic expression for the velocity of an
ideal vortex ring, the remainder can be identified as the Biot-Savart integral
over the section of the vortex line from J − 1 to J + 1. Thus this section of the
Biot-Savart integral (containing the singularity) is replaced by the term
  
κ   2 l+ l −
V local = (S ⊗ S ) ln . (3)
4π aef f

S  is the unit tangent vector to the vortex line at the J’th mesh point and S  ,
the second derivative of the line with respect to arclength, is the local curvature
vector. The length scales l+ and l− are distances between the J’th mesh point
and the J + 1 and J − 1 points respectively. The parameter aef f is an effective
core size since numerical constants of order one have been absorbed into this
parameter. This choice for the local part of the Biot-Savart integral gives the
correct velocity for a planar vortex ring. Defining the velocity of a planar vortex
ring requires that you specify some details about the core structure of the vortex
(hollow core vs solid body rotation, for example). Different choices of the core
structure will alter the log term in (3), but unless you are calculating flows on
very small length scale, on the order of the core size, then these details make
negligible changes to the value of (3).
Once you prevent the singularity by splitting the Biot-Savart law into two
sections you are still left with the difficulty of calculating the integral over the
rest of the vortex line. In a simulation, we only know the positions of the discrete
mesh points along the line, so the mesh points must be joined in some numerically
convenienent manner to calculate the integral between the mesh points. Any
interpolation method through the mesh points will do the job. I prefer to use
the simplest method of integrating over piece-wise linear vortex line segments
between the mesh points. The velocity at a point r (which may be a mesh point
on the vortex line or may be any point in the volume of the fluid) due to the
Biot-Savart integral over the segment of vortex line between the J and J + 1
mesh points is
 SJ+1
κ (S − r) ⊗ dS
V segment (r) = . (4)
4π SJ |S − r|3
Taking this integral over a straight line between the two mesh points gives

κ (|RJ | + |RJ+1 |)(RJ ⊗ RJ+1 )

V segment (r) = (5)
4π |RJ ||RJ+1 |(|RJ ||RJ+1 | + RJ · RJ+1 )

where RJ = SJ − r and RJ+1 = SJ+1 − r. Other interpolation methods will

have more complicated results for the Biot-Savart integral.
 Vortex Filament Methods for Superfluids 101

One of the reasons that vortex filament methods have fallen out of favor in
classical fluid dynamics is the fact that the calculation of the Biot-Savart integral
requires order N 2 operations over a mesh of N points. The calculation of this
integral is computational expensive whenever a large number of mesh points are
needed to represent the vortex lines. This calculation is often not practical, even
with the speed of today’s computers, if the amount of vortex line is large, or if
the fine detail of the vortex line shape needs to be captured using a very fine
meshing. And these are often exactly the cases that are of the most interest to
us.
One way around this difficulty is the Local Induction Approximation, (LIA)
[15]. In this method we keep the local term in the induced motion (3) and neglect
completely the non-local Biot-Savart integral. This is typically done with some
minor adjustments to the log term so that we have
κ  2R 
V I,LIA = (S  ⊗ S  ) ln , (6)
4π aef f
where R is a length scale of the filament. R may be taken as a constant, such as
the length scale of the computational box, or it may be taken as the local radius
of curvature of the vortex line. In the first case, the log term is a constant, and
it is often absorbed into a non-dimensional timescale. In the second case, the log
term will vary along the filament and with time, but unless you are capturing a
very wide range of curvatures with your mesh, or the curvatures are only about
a factor of 10 larger than the effective core size, the log term is nearly constant.
LIA is a very convenient approximation. It is simple and easy to calculate.
The time required to calculate (6) increases only linearly with the number of
mesh points. The interpretation of (6) is simple: the induced velocity is in the
direction of the local binormal S  ⊗ S  and is inversely proportional to the
local radius of curvature. It is commonly used in analytic investigations of the
properties of vortex filaments. And LIA correctly describes the motion calculated
by the Biot-Savart law of simple vortex line geometries such as the planar vortex
ring and low amplitude vortex waves. But it is a very severe approximation to
the true equation of motion!
LIA clearly works very well for calculating the motion of single vortex lines
which do not loop around so that sections of the filament far apart in arclength
are actually close to each other. Also it probably works well in a random vortex
tangle, where the non-local part of the Biot-Savart law may tend to cancel out
when integrated over the random vortex lines. LIA will not work well for vortex
configurations which tend to have parallel, or antiparallel, sections of vortex
filament near one another (for example, vortex knots [16]). LIA will not work well
for flows which tend to develop any alignment of vorticity. Unfortunately, aligned
superfluid vorticity does tend to be formed by flows at non-zero temperatures
where the normal fluid has some local vorticity structure (the superfluid vortices
tend to align with the normal fluid vorticity [17]). In the case of turbulent flows, I
personally doubt that the LIA captures enough of the physics of the interactions
between the vortex filaments to give much useful information on these flows,
though others would debate this point.
102 D.C. Samuels

Some of the qualitative differences between the motion of vortices under LIA
and the full Biot-Savart law should be pointed out. A superfluid flow calculated
by LIA can never develop any rotational flow, aside from the rotation around a
single filament. In an LIA calculation, the energy of the flow is just an energy
per unit length of the vortex lines and when there is no mutual friction (at zero
temperature) the length of vortex line does not change with time (without further
assumptions to the model, such as phonon emission). Under the Biot-Savart law,
the kinetic energy in the flow can be quite difficult to calculate, and even at zero
temperature the length of vortex line can change while the kinetic energy remains
constant. In a LIA calculation, the physical effect of vortex stretching cannot
occur while it can occur in a Biot-Savart calculation whenever there are even
just two approximately parallel vortex lines.
With the vortex filaments represented by N mesh points, the vortex equation
of motion (1) becomes N coupled, first order ordinary differential equations. This
coupled set of ODEs can be solved by any general method. I prefer to use a fifth
order Runge-Kutta method with an adaptive stepsize. An adaptive method like
this calculates the result of one time step in two different ways (one fourth order
method and one fifth order method) and compares the results at each of the N
mesh points to estimate the error. An allowed error range is defined and if the
estimated error lies below this allowed error range for all N points, the time step
is increased, while if the estimated error is above the allowed range for any of the
N mesh points, the time step is decreased and that timestep is recalculated. This
is a sturdy method, capable of automatically handling the rapid changes in vortex
line velocities which can occur, often as a result of vortex line reconnections. It is
not a very efficient method however. Whenever any region of very high curvature
(or a close approach of two vortex filaments under the Biot-Savart law) occurs
then the timestep for the entire vortex tangle can drop significantly. Some type of
‘multi time step’ method, where the motion of fast moving sections of the vortex
line could be calculated using more time steps than the slow moving sections of
the vortex, would would be far more efficient, particularly when the equation of
motion is as expensive to evaluate as is the Biot-Savart law.

2.2 Boundary Conditions

There are three basic types of boundary conditions used in vortex filament cal-
culations. In order of increasing complexity these are: an infinite fluid, periodic
boundary conditions, and solid boundaries (no penetration of the superfluid
through the boundary). In an infinite fluid calculation no boundary conditions
are required in the calculations, as long as you set up the initial conditions as
closed vortex loops. Many flow quantities, such as the total kinetic energy, have
simple line integral definitions that only work in the infinite fluid simulations,
where the vorticity is zero at spatial infinity. This is a good choice of bound-
ary conditions for studying simple closed loop vortex structures, such as vortex
knots, but it is not a good choice for complicated vortex tangles.
Periodic boundary conditions work well for vortex tangles since they allow a
homogeneous flow to develop and allow simple definitions of averaged quantities
 Vortex Filament Methods for Superfluids 103

such as the vortex line density. Periodic boundary conditions are easily pro-
grammed for LIA simulations, though care must be taken at the places where
neighboring mesh points along a filament lie on opposite sides of your periodic
boundary. But these problems are easily handled. The main difficulty with peri-
odic boundary conditions comes from the Biot-Savart calculations. In principle,
periodic BCs make the Biot-Savart integral infinitely long. In practice, this inte-
gral must be cut off at some arbitrary point. The costs of periodic BC in Biot-
Savart calculations can be staggering. For a three dimensional periodic cube,
including the first layer of periodic vortices around the central cubic volume will
increase the number of vortex mesh points by a factor of 27. Including even this
level of periodicity is usually prohibitively costly. To deal with problem I will
sometimes define a buffer layer around the central cubic volumn, with a width
of 1/2 to 1/4 of the width of the computational volume, and I will only include
in the Biot-Savart calculation the periodic vortex filaments that fall within this
fairly thin buffer zone. This type of method actually only ensures that vortex
filements moving through one side of the periodic volume and re-entering on the
other side, experience a more smoothly varying velocity field during the transi-
tion. Ideally, this type of problem could be removed by an analytic method to
calculate the velocity field due to this infinitely repeating vorticity distribution.
While this has been done in two dimensions, I know of no such method in 3D.
In LIA, solid boundary conditions are quite easy to implement [14]. In this
case, for any vortex filaments which end on the solid surface we must set the
normal vector at the ends of the vortex filament equal to the local normal vector
of the solid surface. This is the only condition needed no matter how complex the
boundary geometry. In Biot-Savart calculations we must use image vortices for all
the vortex filaments in the flow. For a single flat boundary, these image vortices
are simply vortices in the mirror image positions, and with their orientation
reversed. For a single flat surface this doubles the length of the Biot-Savart
calculations, which is not a terribly onerous increase. But for multiple solid
surfaces, you must then include the images of images, ad infinitum (just as
you must do in electrostatic calculations). This naturally introduces the same
computational difficulties as in the periodic BC case, and it must be solved in
the same manner, either by cutting off the image calculations at some arbitrary
number of images or by developing an analytic summation method to collapse
the infinite images to a reasonable calculation. I know of no one who has done
the latter for a 3D vortex flow, though the tools may exist in the electrostatics
literature for us to use.

2.3 Meshing of the Filaments

Since the CPU time cost of a vortex simulation rises rapidly as the number of
mesh points increases it is important to manage this number carefully. As the
vortex filaments are locally compressed or elongated the spacing between the
mesh points will change with time. New mesh points will need to be inserted on
some sections on the filament to keep a certain mesh resolution, and to keep the
number of mesh points at a minimum some mesh points will need to be removed
104 D.C. Samuels

from other sections of the filament as they move too close to one another (close
in terms of arclength). The criteria that you choose for remeshing depends on
the behaviour that you are trying to capture with a specific simulation.
For this discussion, let the arclength distance between two neighboring mesh
points, J and J + 1 be represented by δJ,J+1 . The simplest remeshing criteria is
to attempt to keep an approximately uniform distance between the mesh points.
In this method, the mesh point J would be removed if δJ−1,J + δJ,J+1 < δ and a
mesh point would be added between the J and J + 1 points if δJ,J+1 > 2δ. This
will keep all the distances along the mesh roughly in the range δ/2 to 2δ. This
is the most straightforward meshing method, but it suffers from the limitation
that it can only resolve structures on the vortex filament of scale δ or larger.
This method is best if you are only interested in the vortex behaviour at one
particular length scale.
If you want to calculate the development of vortex filament structures over
a wide range of length scales then you must use a more complicated remeshing
method. We would like to have more mesh points in the parts of the vortex
filament with high curvature and fewer in regions of low curvature. This can be
done by interpreting the meshing length scale δ as a variable, proportional to the
local radius of curvature. Typically, I will take δ = 2πR/32, where R is the local
radius of curvature of the filament. This will mesh a planar ring, at any length
scale, with approxamately 32 mesh points. Of course, there must be a limit on
the range of this meshing. This is done by setting a range of R over which this
remeshing will be used, with minimum and maximum mesh lengths to be used
outside this range. With this method it is possible for the mesh to represent a
range of two or three orders of magnitude in radius of curvature without having
an unreasonably large number of mesh points. It may be possible to extend
this range by another order of magnitude by setting the meshing length δ to
decrease slower than linearly with R. Techniques like this are used in 2D vortex
simulations in the meshing of the surface of vorticity blobs [18].
However you choose to remesh, removing mesh points poses no problems but
in adding mesh points you must make a choice on the position of the new mesh
point. Once again, the simplest choice is to use a piece-wise linear interpolation,
and simply place the new point at the center of the straight line between the J
and J + 1 point. But if you do this, then you are likely to change the first and
second derivatives, S  and S  at the points J and J +1. Worse than this, the new
point would be introduced with zero local curvature, almost certainly causing a
rapid (and artificial) change in curvature along the filament. For adding mesh
points a better interpolation must be used, and almost any interpolation will
perform well as long as it is not piece-wise linear. I use a method that inserts
new mesh points at a position where the curvature vector, S  , at the new mesh
point is the average of the curvature vectors at the two neighboring points, J
and J + 1. This prevents the intruduction of sudden jumps in the vortex line
curvature, errors which would radiate vortex waves and affect more than just
the local segment of the vortex line.
 Vortex Filament Methods for Superfluids 105

For ease of explanation in this paper I have been labelling neighboring mesh
points with consecutive number, J and J + 1 for example. Any remeshing natu-
rally will destroy this order, and consecutive numbering is not practical. Instead,
a data structure should be used that has for each mesh point, the labels of the
two neighboring mesh points. Then a remeshing involves only the addition or
deletion of a mesh point and the necessary changes in this data structure for the
neighboring mesh points.
One further remeshing routine should be used. Occasionally, very small vortex
rings may develop. These may be formed by the decay through mutual friction
of larger vortex loops. Or they may be generated by the pinching off of a small
loop in a reconnection event. However they are formed, these small loops should
be removed from the calculation. By definition, the curvature of these loops will
be very large, and thus their motion will be very fast. With an adaptive stepsize
technique, the code will automatically drop the time step to a very small value
to attempt to accurately calculate the rapid motion of these small loops. If you
are not using an adaptive stepsize, then the motion of these small loops will not
be accurately calculated, possibly leading to the sudden and artificial expansion
of the small loop. In either method these small loops must be removed from the
calculation. It is a good idea to keep a record at least of the length of vortex line
removed from the simulation by this routine.

3 Reconnections of Filaments
When two classical Navier-Stokes vortex filaments cross, viscous effects will re-
connect the vortex filaments. Lacking these viscous effects, the crossing of two
classical Euler vortex filaments is believed to lead to a singularity and the break-
down of the Euler equations [19]. This singularity formation is not observed in
calculations of the crossing of superfluid vortex filaments, where vortex recon-
nection occurs without the need for viscosity [20]. In this way, superfluid vortices
are clearly not Euler vortex filaments. The difference lies in the core structure
of the superfluid vortices. At the centerline of each vortex filament, quantum
mechanics requires the density of the superfluid (given by the amplitude of the
ground state wave function) to go to zero and the vorticity of the flow is singular
along this centerline.
Studies of superfluid vortex filament reconnections must be done with a quan-
tum theory calculating the evolution of the wavefunction of the superfluid ground
state. A more detailed discussion of these calculations can be found in the articles
by Roberts and by Adams in this volume. These studies have traditionally been
done using a Non-Linear Schrodinger Equation (NLSE). It is recognized that
the NLSE is probably a poor representation of helium II superfluid because the
dispersion curve calculated from the NLSE contains only phonons and no rotons.
The NLSE does, however, seem to be a very good representation of the new Bose
condensed alkali atom superfluids. Without a fundamental microscopic theory
of helium II superfluidity, we must consider the results of NLSE calculations,
keeping in mind the possible errors of these calculations. The NLSE calculations
106 D.C. Samuels

clearly show that when two superfluid vortex lines cross, a reconnection event
occurs and the topology of the vortex lines changes [20]. This behavior must be
included as a basic assumption in the superfluid vortex filament model since it
is not a consequence of the equation of motion. This discussion will be made in
terms of the crossing to vortex filaments in the bulk of the fluid. In simulations
with boundaries, reconnections of filaments with the boundaries must also be
considered, but these reconnections may be easily detected as vortex segments
approach and eventually hit the boundaries so we need not discuss them further
here.
The reconnection assumption leaves us with the problem of detecting the
crossings of 1-D lines moving through 3-D space, when we only know the posi-
tions of a finite number of mesh points along these lines at discrete time intervals.
Though the assumption can be quite simply stated, it has proved very difficult
to develop any satisfactory algorithms to implement this. The first reconnec-
tion algorithm, due to Schwarz [21], was a simple, intuitive one: vortex lines are
reconnected whenever two mesh points come within a pre-defined distance Δ
of each other. While this algorithm is very simple to use, the objections to it
are many and strong. This algorithm will obviously trigger a reconnection event
early, since a true reconnection event does not occur until the distance between
the lines is on the order of a core radius (practically zero in filament simulations).
Sometimes, a large section of vortex filament, including many mesh points, will
move within the reconnection distance triggering a series of reconnection events
resulting in the formation of a number of small vortex loops where only one
crossing event would actually have occured. Even worse, in some cases this al-
gorithm will trigger a reconnection where no crossing of the vortex lines would
actually occur. This error happens when vortex lines pass by each other, without
intersection but within the distance Δ. This may at first seem to be a rare and
thus negligible event, but it is not. We now know that vorticity concentrations
in the normal fluid generate bundles of well-aligned superfluid vortex filaments
at the center of the normal fluid rotation [17]. These superfluid vortex filaments
may be closer to each other than the defined reconnection distance Δ but they
cross only rarely. With the Δ reconnection algorithm, these vorticity bundles
will reconnect wildly.
The Δ reconnection model also introduces a new and completely artificial
length scale to the simulation; the reconnection length scale Δ. If you set Δ too
large, you will have many spurious reconnections. If you set it too low, you will
miss many real reconnections, as vortex lines jump past one another in the finite
time step, plus you will still have some spurious reconnections of nearly aligned
vortices. Some of the problems of this algorithm can be helped (though not
eliminated) by making the reconnection length scale variable along the vortex
filament and setting it equal to the local mesh size (a natural choice). This
refinement to the algorithm still suffers from the separate problems of premature
and also spurious reconnection.
A different reconnection detection algorithm, the one that I currently use,
is based on the idea that a vortex filament reconnection event is a dissipative
 Vortex Filament Methods for Superfluids 107

event, even in a superfluid. The reconnection is assumed to involve the emission

of phonons and possibly rotons, thus converting superfluid energy to heat. With
this in mind, we assume that reconnections can only occur if they are accom-
panied by a loss of energy in the superfluid. Identifying vortex line energy with
vortex line length (an exact proportionality in LIA, but only approximate for
the Biot-Savart law), we reconnect the vortex lines when the reconnection will
directly lower the line length. The reconnections that occur in this algorithm
still happen when the filaments are slightly less than a local mesh length apart,
so it still suffers from the premature reconnection problem. But the generation
of additional small vortex loops in a reconnection rarely occurs in this algorithm
and spurious reconnections from near misses and between aligned vortices are
almost eliminated.
It is worth taking care with the reconnection algorithm. Differences in the
implementation of reconnections [22] [23] [24] were initially blamed for inconsis-
tancies between the vortex tangle simulations of Schwarz and Buttke in the late
80’s, though the more serious error of the introduction of an unphysical ’mixing
step’ by Schwarz was eventually determined to be the prime cause of the dis-
agreement [25]. We must admit that the reconnection algorithms in use today
still leave much to be desired and can be improved. A reconnection is a major
event, drastically changing the vortex filament motion, curvature distribution,
and topology. The close approach and eventual crossing of two vortex filaments
create superfluid flow structures on the full range of length scales, down to the
core size scale. And the motion of a large section of the vortex filament can
be altered by a reconnection as the initially localized, but extreme, change in
curvature radiates out from the reconnection point as vortex waves. It is still
worthwhile to carry out detailed investigations of individual vortex reconnec-
tions [26] (see Lipniacki’s contribution to this volume) in order to develop better
reconnection models for use in both the filament simulations and in analytic
theories of the averaged quantities in superfluid turbulence.

4 Analysis of the Superfluid Flow

Once you have developed and tested your vortex filament code you are faced
with the analysis problem common to all 3-D fluid dynamics simulations: how to
make sense of all that mass of data. Though the problem is the same, the solution
must be different for vortex filaments due to the very different nature of the data.
Many quantities that can be calculated easily from the 3-D velocity fields of a
classical fluid dynamics simulation are actually very difficult to calculate with
vortex filament methods. One example of this is kinetic energy. Conversely, there
are at least a few things (such as flow topology) that are easier to calculate in a
vortex filament simulation than in a standard fluid dynamics simulation.
The simplest measurable quantity in a vortex filament simulation is the total
length of vortex line. In a finite
 volume V this is measurement gives you directly
the vortex line density L = ( dξ)/V , the length of vortex line per unit volume
where the integral is taken over the arclength ξ of the vortex filaments. This is
108 D.C. Samuels

the fundamental averaged quantity of a vortex tangle and many of the analytic
model of superfluid vortex turbulence have been based on this quantity alone.
These models are differential equations for the evolution of the line density as a
function of the average normal fluid velocity and some temperature dependent
parameters. The most basic of these models is the Vinen equation [27]

dL χ2 κ 2
= χ1 αVns L3/2 − L , (7)
dt 2π
where χ1 and χ2 are temperature dependent parameters and Vns is the average
relative velocity between the normal fluid and the superfluid. The standard in-
terpretation of the Vinen equation is that the first term represents the growth of
L due to mutual friction and the second term represents the decay of the tangle
due to reconnection events, though other interpretations of these terms are have
been made [28].
Now let us define some measure of the isotropy of the tangle. Very little work
has been done on measures of isotropy in vortex tangles. One set of measures
defined by Schwarz [28] defines three quantities which measure the isotropy of a
tangle relative to two perpendicular unit vectors, r̂ and r̂⊥ .

1
I = [1 − (S  · r̂ )2 ]dξ (8)
VL

1
I⊥ = [1 − (S  · r̂⊥ )2 ]dξ (9)
VL

1
I = (S  ⊗ S  ) · r̂ dξ (10)
V L3/2
In flows with a unidirectional normal fluid flow the unit vectors r̂ and r̂⊥ are
usually taken to be parallel and perpendicular to the direction of the normal
fluid flow. The interpretation of these three isotropy measures is discussed in
[28]. Other isotropy measures can be defined, such as the length of line in a
projection along a vector r̂
 
1
J(r̂) = 1 − (S  · r̂)2 dξ. (11)
VL

It is possible that more refined versions of (7) could be defined using these
isotropy measures as well as L.
The local rate of extension (or compression if negative) of the vortex line is
defined by
r(ξ) = −S  · VL (12)
where VL is the vortex line velocity at the arclength position ξ. Integrating
r(ξ) over the arclength gives the instantaneous rate of change of the vortex line
density. 
dL 1
= rdξ (13)
dt V
 Vortex Filament Methods for Superfluids 109

Simulations show that even in a steady state vortex tangle the line length density
has strong fluctuations, so one must be careful in interpreting instantaneous
measures such as (13).
It is tempting to define a whole range of quantities averaged over the vortex
tangle. One that must be mentioned is the averaged curvature

1
C= |S  |dξ, (14)


where  = dξ is the total vortex filament length. This quantity can be used
to define an average length scale of the tangle, 1/C, and an average velocity
scale κC. Following the evolution of this quantity with time can detect if the
vortices are ‘crinkling’ (increasing C) or smoothing (decreasing C). But I urge
against an over-reliance on averaged quantaties. It is easy enough to measure the
full distribution of values of quantities such as the curvature in the tangle, thus
measuring the full range of behavior of the vortices and not just some average
value. Again, little has been done with this approach.
But what about the more traditional fluid mechanics measures such as kinetic
energy and linear and angular impulse? While these are quite easily determined
in a simulation where the velocity is defined on a grid extending throughout the
fluid, they are actually quite difficult to measure in a vortex filament simulation.
Purely for diagnostic purposes, one can define a 3D grid and calculating the
velocity on the grid by the Biot-Savart law (2) and then such quantities can be
easily measured. But it is important when doing this to check the convergence
of your results with grid resolution. The 1/r velocity field of the vortex lines can
make resolution of the energy and momentum difficult. It is possible to avoid
this problem by using definitions of these quantities in terms of the line integrals
over the vortex filaments [29]. Some useful integrals are: kinetic energy

T = κρs V s · (S ⊗ dS), (15)

linear impulse 
1
I= κρs S ⊗ dS, (16)
2
angular impulse 
1
A = κρs S ⊗ S ⊗ dS, (17)
3
and helicity 
J =κ V s · dS. (18)

These integrals are limited to flows with no vorticity at infinity, and the integrals
must be taken over closed vortex loops.
While simple things such as kinetic energy can become quite difficult to mea-
sure in vortex filament methods, there are some qualities of the flow which are
simpler to define and measure in filament methods. There has been much re-
cent interest in the topology of vorticity in Euler flows, interest which is mainly
110 D.C. Samuels

generated by the possibility of producing singularities in these flows in finite

time [19]. Remembering that the vorticity in a classical Euler fluid may be dis-
tributed thoughout space (not just as vortex filaments!), the topology of the
vorticity field can be difficult to define, or picture, easily. But since the vorticity
of the superfluid is confined to discrete vortex filaments, vortex topology is a
natural and simple quality in a superfluid [16]. In an Euler flow, the topology
of the vorticity is conserved as the flow develops because filament reconnections
are not allowed, and the conservation of topology can lead to strong constraints
on the development of the flow [30]. But since superfluid vortices can reconnect,
the topology of a superfluid vortex is not conserved, and may change with every
reconnection. Despite this complication, it is possible that by determining the
topology (knotted and linked loops) of a vortex tangle we can develop a mea-
sure of the complexity of a tangle, a quality that is missing from all averaged
measurements on the vortex tangle.

5 Alternative Approaches
The computational cost of calculating the full Biot-Savart law is high, due to
its non-locality. The usefulness of local induction approximation calculations is
suspect because of its complete locality. Is there not a middle way that contains
some, but not all, of the non-locality of the Biot-Savart integral? The recent
development of fast multipole methods seems to fill this need [31] [6]. Originally
developed for simulations of large numbers of gravitationally interacting parti-
cles, these methods can lower the computational time of pairwise interactions
between particles from order N 2 to order N log(N ). In these methods the inter-
acting particles are grouped into clusters, and then into clusters of clusters, et
cetera, in a hierarchical tree structure. The interactions between the clusters are
calculated by a finite multipole series expansion, greatly decreasing the number
of calculations needed while retaining the primary non-local effects. The general-
ization from the original gravitational interaction to a vortex motion calculation
is straightforward and these new methods have fueled much of the recent resur-
gence (however slight) in vortex methods for classical fluid dynamics. These
methods work best in cases where the fluid vorticity has some strong spatial
structure. The drawback to these methods is the high overhead cost associated
with the determination of the hierarchical tree data structure, and the necessity
of rebuilding this data structure as the vortices move, and groups of vortices
break up or coalesce. For a moderate number of vortex mesh nodes, up to a few
thousand in my experience, the Biot-Savart calculations are still faster than the
fast multipole methods, though this limitation may become outdated soon as
new implementation algorithms are being developed. For simulations af actual
superfluid turbulence, where we must reproduce a wide range of spatial scales in
the superfluid vorticity (requiring a large number of mesh points) and include
the non-local vorticity interactions of turbulence (ruling out the LIA approach)
we must use some non-local approximate method in our simulations, and the
fast multipole methods are currently our best hope.
 Vortex Filament Methods for Superfluids 111

Yet another approach is to drop the idea of vortex filaments completely and
use a locally averaged set of equations for the superfluid vorticity field, ωs . These
approaches use the Hall-Vinen-Bakharevich-Khalatnikov (HVBK) equations (or
modifications of these equations) for the coupled evolution of continuous vorticity
fields in the superfluid and the normal fluid. Since this approach does not use
vortex filaments it lies somewhat out of the scope of this article, but is dealt
with in the articles by Henderson and Holme in this volume.

6 Conclusions: What Needs to Be Done

The primary research challenge before us today is to develop a detailed under-
standing of the interaction between the normal fluid and the superfluid [32][33],
particularly in turbulent flows. In terms of simulations, this presents us with the
double task of calculating the turbulent flow of the normal fluid (including the
forcing on the normal fluid due to mutual friction, see the paper by Idowu et
al, this volume), and simultaneously calculating the motion of a large length of
superfluid vortex lines. The simulation method must allow for the possibility of
the formation of organized structures in the superfluid vorticity, so non-local vor-
tex interactions must be included. The solution of the full superfluid turbulence
problem will be difficult and will take some time, both CPU time and time to
develop an understanding of the problem. But there is much to be learned from
simpler studies, short of the full turbulence solution. In kinematic simulations
(simulations where the normal fluid velocity is given and only the superfluid mo-
tion is calculated) the most complex normal fluid flow that has been used so far is
an ABC flow [34]. We could learn many things from kinematic simulations with
more complicated normal fluid flow structures. We should take a hint from the
closely related field of magnetohydrodynamics and investigate the development
of superfluid vorticity driven by a turbulence model to represent the normal fluid
velocity. These types of studies would give us valuable information on the reac-
tion of the superfluid to complex normal fluid flows. Does the superfluid vortex
tangle lack structure (as it does under a uniform VN ) or do structures develop
in the superfluid vorticity? If so, are there any relationships between the flow
structures in the normal fluid flow and those developing in the superfluid? What
are the scaling laws for kinetic energy in the turbulent superfluid? How do the
velocity correlation functions in the superfluid compare to those in classical tur-
bulence? We could have at least preliminary and approximate answers to these
questions from kinematic simulations. In fully coupled simulations (solving for
the motion of both VN and the superfluid vorticity) we must investigate the
possible changes in the normal fluid turbulence itself due to the mutual friction
forcing from the superfluid. This problem bears a resemblance to turbulence
with polymer additives, where the Navier-Stokes turbulence is clearly affected
by the 1-D forcing from the polymers.
A not-entirely separate issue is the problem of superfluid turbulence at zero
temperature [35] (see the articles by Tsubota and McClintock, this volume). In
studies of this type we need a firmer understanding of the loss of energy by
112 D.C. Samuels

phonon and roton emission from reconnecting vortices [36] and possibly even
from just the motion of vortices. Does this ‘zero temperature dissipation’ clearly
affect the properties of the vortex tangle? Is it enough to produce an inertial
range in the pure superfluid turbulence? Can we develop an understanding of
the energy cascade in a pure superfluid vortex tangle? I say that this is a not en-
tirely separate issue from that of the fully coupled helium II turbulence because
we need to understand the behaviour of the fully coupled turbulence in both the
high and low temperature limits. In the high temperature limit, helium turbu-
lence must surely be classical Navier-Stokes turbulence after the lambda tran-
sition temperature is passed and some experiments show no detectable change
in the measured turbulence quantities as this transition temperature is passed.
Pure superfluid turbulence is the low temperature limit of fully coupled helium
II turbulence and there is no reason not to expect this transition to also be
very smooth. Helium II turbulence experiments show little, if any, temperature
dependence down to approximately 1.4 Kelvin [37], where the normal fluid rel-
ative density and the mutual friction interaction both become quite small. An
understanding of both classical Navier-Stokes turbulence and pure superfluid
turbulence will be helpful in developing our understanding of coupled helium II
turbulence.

References
1. A. J. Chorin: SIAM J. Sci, Stat. Comput. 1, 1 (1980)
2. A. Leonard: J. Comput. Phys. 37 289, (1980)
3. B. Couet, O. Buneman, A. Leonard: J. Comput. Phys. 39 305, (1981)
4. A. Leonard: Ann. Rev. Fluid Mech. 17 523 (1985)
5. W. T. Ashurst, E. Meiburg: J. Fluid Mech. 189 87 (1988)
6. G-H. Cottet, P. D. Koumoutsakos: Vortex Methods (Cambridge University Press,
Cambridge, 2000)
7. R. Cortez: J. Comput. Phys. 160 385 (2000)
8. R. J. Donnelly: Quantized Vortices in Helium II (Cambridge University Press,
Cambridge, 1991)
9. V. M. Fernandez, N. J. Zabusky, V. M. Gryanik: J. Fluid. Mech. 299 289 (1995)
10. T. Y. Hou: Math. of Comp. 58 103 (1992)
11. R. Klein, O. Knio: J. Fluid Mech. 284 275 (1994)
12. M. F. Lough: Phys. Fluids 6 1745 (1994)
13. R. Klein, L. Ting: Appl. Math. Lett. 8 45 (1995)
14. K. W. Schwarz: Phys. Rev. B 31 5782 (1985)
15. R. L. Ricca: Fluid Dyn. Res. 18 245 (1996)
16. R. L. Ricca, D. C. Samuels, C. F. Barenghi: J. Fluid Mech. 391 29 (1999)
17. D. C. Samuels: Phys. Rev. B 47 1106 (1993)
18. D. G. Dritschel, N. J. Zabusky: Phys. of Fluids 8 1252 (1996)
19. R. B. Pelz: Phys. Rev. E 55 1617 (1997)
20. J. Koplick, H. Levine: Phys. Rev. Lett. 71 1375 (1993)
21. K. W. Schwarz: Phys. Rev. Lett. 49 283 (1982)
22. T. F. Buttke: Phys. Rev. Lett. 59 2117 (1987)
23. K. W. Schwarz: Phys. Rev. Lett. 59 2118 (1987)
 Vortex Filament Methods for Superfluids 113

24. T. F. Buttke: J. Comp. Phys. 76 301 (1988)

25. K. W. Schwarz: J. Comp. Phys. 87 237 (1990)
26. T. Lipniacki: Eur. J. Mech. B - Fluids 19 361 (2000)
27. W. F. Vinen: Proc. Roy. Soc. Lond. A 242 493 (1957)
28. K. W. Schwarz: Phys. Rev. B 38 2398 (1988)
29. P. G. Saffman: Vortex Dynamics (Cambridge University Press, Cambridge, 1992)
30. H. K. Moffatt, A. Tsinobar: Topological Fluid Mechanics (Cambridge University
Press, Cambridge, 1990)
31. L. Greengard, V. Rokhlin: J. Comp. Phys. 73 325 (1987)
32. D. Kivotides, C. F. Barenghi, D. C. Samuels: Science 290 777 (2000)
33. O. C. Idowu, A. Willis, C. F. Barenghi, D. C. Samuels: Phys. Rev. B 62 3409
(2000)
34. C. F. Barenghi, D. C. Samuels, G. H. Bauer, R. J. Donnelly: Phys. Fluids 9 2631
(1997)
35. B. V. Svistunov: Phys. Rev. B 52 3647 (1995)
36. M. Leadbeater, T. Winiecki, D. C. Samuels, C. F. Barenghi, C. S. Adams: Phys.
Rev. Lett. 86 1410 (2001)
37. J. Maurer, P. Tabeling: Europhys. Lett. 43 29 (1998)
Introduction to HVBK Dynamics

Darryl D. Holm

Theoretical Division and Center for Nonlinear Studies, Los Alamos National
Laboratory, MS B284, Los Alamos, NM 87545

Abstract. We review the Hall-Vinen-Bekarevich-Khalatnikov (HVBK) equations for

superfluid Helium turbulence and discuss their implications for recent measurements
of superfluid turbulence decay.
A new Hamiltonian formulation of these equations renormalizes the vortex line
velocity to incorporate finite temperature effects. These effects also renormalize the
coupling constant in the mutual friction force between the superfluid and normal fluid
components by a factor of ρs /ρ (the superfluid mass fraction) but they leave the vortex
line tension unaffected. Thus, the original HVBK form is recovered at zero tempera-
ture and its mutual friction coefficients are renormalized at nonzero temperature. The
HVBK equations keep their form and no new parameters are added. However, a tem-
perature dependent trade-off does arise between the mutual friction coupling and the
vortex line tension.
The renormalized HVBK equations obtained via this new Hamiltonian approach
imply a dynamical equation for the space-integrated vortex tangle length, which is the
quantity measured by second sound attenuation experiments in superfluid turbulence.
A Taylor-Proudman theorem also emerges for the superfluid vortices that shows the
steady vortex line velocity becomes columnar under rapid rotation.

1 HVBK Equations
Recent experiments establish the Hall-Vinen-Bekarevich-Khalatnikov (HVBK)
equations as a leading model for describing superfluid Helium turbulence. See
Nemirovskii and Fiszdon [1995] and Donnelly [1999] for authoritative reviews.
See Henderson and Barenghi [2000] for a recent fluid mechanics study of steady
cylindrical Couette flow using computer simulations of the incompressible HVBK
equations.
In the Galilean frame of the normal fluid with velocity vn , the HVBK equa-
tions may be expressed as follows, upon ignoring thermal diffusivity and viscosity,
∂t ρ = − divJ , ∂t Ji = − ∂k Tik , ∂t S = − div(Svn ) + R/T ,
 1 
ρs ∂t vs + ρs (vs · ∇)vs = − ρs ∇ μ − |vs − vn |2 + ρs f , (1)
2
and summing over pair of upper and lower repeated indices. One may consult,
e.g., Bekarevich and Khalatnikov (BK) [1961] and Donnelly [1999] to compare
these equations with the form they take in the reference frame of the superfluid.1

1
In making this comparison it is useful to recall the Galilean transformation of the
chemical potential, μ  = μ − 12 |vs − vn |2 , where μ  is evaluated in the superfluid

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 114–130, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Introduction to HVBK Dynamics 115

Notation. Here ρ and ρs denote the total and superfluid mass densities, respec-
tively. The entropy density of the normal fluid is S, its temperature is denoted
T , and ρn = ρ − ρs denotes its mass density. The superfluid velocity is denoted
vs and
J = ρs vs + ρn vn
is the total momentum density. In the entropy equation R is the rate that heat is
produced by the phenomenological friction and reactive forces in f , which must
be specified to close the theory. We also denote

Stress tensor: Tik = ρs vs i vsk + ρn vn i vnk + (P + λ·ω) δik − λi ω k ,

Euler’s pressure law: P = − ε0 + T S + ρμ ,
Superfluid First Law: dε0 = μ dρ + T dS + (J − ρvn )·d(vs − vn ) + λ·dω . (2)

In the superfluid First Law, ω = curl vs is the superfluid vorticity with magnitude
|ω| = ω̂ ·ω, where ω̂ = ω/|ω| is its unit vector. BK [1961] takes the energy density
ε0 to depend on the magnitude of the superfluid vorticity, |ω|, as

ρs κ |ω| R
ε0 = ln .
4π a
This is the energy per unit length of a superfluid vortex line, ρs (κ2 /4π)ln(R/a),
with quantum of circulation κ = h/m 10−3 (cm2 /sec) and ratio R/a of mean
distance between vortices R to effective vortex radius a, times the vortex length
per unit volume, |ω|/κ. Hence, we find

∂ε0 ∂|ω| κ R
λ= = |λ| ω̂ with |λ|/ρs ≈ ln = λ0 ,
∂|ω| ∂ω 4π a

where one ignores the derivative of R κ/|ω| inside the logarithm. The ap-
pearance of λ in the stress tensor Tik shifts the pressure P , and div T introduces
an additional force − ω · ∇λ ≡ −ρs T into the motion equation. The quantity T
is called the “vortex line tension.”
BK [1961] assigned the following form to the phenomenological coupling force
f appearing in the superfluid velocity equation in (1),

f = (vL − vs ) × ω , vL = vn − ρs (α s0 + β ω̂ × s0 ) , (5)
where s0 = v0 − vn , v0 = vs + ρ−1
s curl λ , (6)

with “vortex velocity” vL and “slip velocity” s0 introduced as auxiliary quanti-

ties. The HVBK equations written as (1) in the normal-fluid frame conserve the
energy,  
1 
E= ρ|vn |2 + (J − ρvn )·vn + ε0 d 3 x . (7)
2
frame and μ in the normal-fluid frame. See Putterman [1974] for a clear discussion
of the role of Galilean transformations in superfluid hydrodynamics.
116 Darryl D. Holm

The BK [1961] form of the phenomenological force f also implies the dissipative
heating rate,
R = (J − ρvn + curl λ) · ω × (vL − vn ) ,
which is Galilean invariant and positive. Substituting this form of f into the
superfluid motion equation in (1) and taking its curl provides the following
equation for the superfluid vorticity, ω = curl vs ,

∂t ω = curl (vL × ω) .

This vorticity equation implies the HVBK superfluid Kelvin circulation

theorem,  
d d
ω · dS = vs · dx = 0 .
dt S dt ∂S(vL )
The Kelvin formula (10) expresses conservation of the flux of superfluid vorticity
through any surface S whose boundary ∂S moves with the velocity vL , so vL
may be regarded as the local velocity of a vortex line. Equivalently, the Kelvin
formula expresses conservation of superfluid velocity circulation around any loop
that moves with the vortex line velocity vL .
In a key phenomenological step that closed the theory, BK [1961] assigned
the undetermined functions α and β in the force f and its auxiliary vortex line
velocity vL as
B  ρn Bρn
1 + αρs = , βρs = . (11)
2ρ 2ρ
The dimensionless coefficients B and B  were introduced earlier in Hall and
Vinen (HV) [1956] to parameterize the Gorter-Mellink [1949] “mutual friction”
force (B) and its reactive component (B  ). Hence the name, HVBK equations
for this closure.
The assignments in BK [1961] of the undetermined functions α, β in (11), as
well as the vortex slip velocity s0 in (6) are designed to reproduce the phenomena
observed in HV [1956] and yet still conserve mass, momentum and energy. This
phenomenological approach used in BK [1961] is indeterminate, however, in the
sense that some freedom still remains in making these assignments. The HVBK
equations (1) that result from this approach do possess the desired conservation
laws for mass, momentum and energy. And they also possess a Kelvin theorem
for the circulation of superfluid velocity. However, because of the indeterminacy
inherent in the phenomenological approach, the HVBK equations (1) are not
unique in possessing these properties. An alternative assignment of the vortex
slip velocity is

s = v − vn with v = v̄ + ρ−1 curl λ , with v̄ = ρ−1 (ρs vs + ρn vn ) . (12)

The mean velocity v̄ also figures prominantly in Hills and Roberts [1977] discus-
sion of the HVBK equations. As we shall see, the alternative expression (12) for
the auxiliary vortex slip velocity in terms of the mean velocity v̄ arises naturally
in the Hamiltonian derivation of a slightly modified set of HVBK equations.
 Introduction to HVBK Dynamics 117

These equations possess the same formal conservation and circulation properties
as HVBK, modulo redefining the vortex slip velocity as s rather than s0 . The
vortex slip velocity s in (12) is defined relative to the Galilean frame of the nor-
mal fluid, which is present only at finite temperature. The HVBK s0 in (6) is the
limit of the vortex slip velocity s for zero temperature, at which no normal fluid
remains. The vortex slip velocity s in (12) is a slight modification of s0 in (6)
necessary to incorporate finite temperature effects, without changing the form of
the HVBK theory, obtained from a Hamiltonian derivation of these equations in
the normal fluid frame. In the Hamiltonian framework, the energy-momentum
conservation laws and Kelvin circulation theorem are all natual consequences.
Moreover, the velocities v and vn are identified as being dual to the momenta
given by ρvs and ρn (vn − vs ), respectively.
Outline. We shall use a Hamiltonian approach with Lie-Poisson brackets to de-
rive the expression (12) for the vortex line velocity v at finite temperature from
first principles by using the energy E in (7) as the Hamiltonian. The momenta
conjugate to the velocities v and vn shall be our basic dynamical variables. The
finite temperature vortex line velocity v and slip velocity s determined this way
turn out to be
ρ 
v = ρ−1 (J + curl λ) , and s = v − vn
s
(v0 − vn ) . (13)
ρ

At zero temperature, ρ → ρs and these reduce to the BK [1961] phenomenolog-

ical expressions with v0 given by (6). Thus, the finite temperature corrections
found by using the Hamiltonian approach renormalize the HVBK slip velocity
in the mutual friction force f by the factor ρs /ρ (the superfluid mass fraction).
Aside from this renormalization, the vortex line tension is left unaffected by
this renormalization, the superfluid vortex equation keeps its form and no new
parameters are added.
Technical details of deriving this renormalized theory from its Hamiltonian
and Lie-Poisson brackets are given in the Appendix.
Main results. We shall use the superfluid vortex dynamics for the renormalized
HVBK equations obtained via this Hamiltonian approach to write a dynamical
equation for the space-integrated total vortex tangle length, which is the
quantity measured in the Oregon experiments on superfluid turbulence reported
in Skrbek, Niemela and Donnelly [1999].
We shall also study the restriction of the renormalized HVBK equations for
the incompressible case, in which ρn and ρs are constants and one takes
∇ · vn = 0 and ∇ · vs = 0. Finally, we shall demonstrate the invariance of the
forms of these equations upon transforming into a rotating frame. The Coriolis
force in such a rotating frame couples to the vortex line velocity vL , which of
course differs from both the superfluid velocity and the normal velocity. We shall
derive a Taylor-Proudman theorem for steady superfluid vortices under rapid
rotation. According to this superfluid Taylor-Proudman theorem, the vortex line
velocity becomes columnar under sufficiently rapid rotation. That is, the lateral
vortex line velocity is nondivergent and independent of the axial coordinate, and
118 Darryl D. Holm

the axial velocity decouples from the lateral motion. Therefore, under sufficiently
rapid rotation, the superfluid vortex filaments will straighten and become parallel
to the axis of rotation as they approach a steady state.
Numerical implications. This renormalization of the vortex line element slip
velocity in the HVBK equations from s0 → s s0 ρs /ρ is sensitive to temper-
ature, but it does not affect the vortex line tension. Therefore, a temperature
sensitive trade-off arises between mutual friction and vortex line tension that
may be worth testing in numerical simulations such as those reported in Hen-
derson and Barenghi [2000]. The HVBK equations are thought to break down in
the presence of strong counterflow. However, as general conservation laws there
is no mechanism in the equations that would signal this breakdown. A rotating
Rayleigh-Besnard experiment might be useful in testing the range of validity of
the HVBK equations (Barenghi, private communication). Such an experiment
might also indicate how these equations should be modified in the presence of
strong counterflow.
Experimental implications. Temperature sensitivity of the coupling between
the superfluid vortices and the normal fluid component is an area of intense
current investigation in superfluid Helium turbulence, see Donnelly [1999]. One
would like to know whether the ρs /ρ renormalization of the mutual friction forces
relative to the vortex line tension would matter significantly in comparisons of
the predictions of the HVBK equations with modern experiments in Helium
turbulence at low, but finite temperatures.
Superfluid vortex dynamics. To begin addressing this issue, we may use the
superfluid vorticity equation for the renormalized HVBK equations obtained in
the Appendix via the Hamiltonian approach to write an explicit equation for
the dynamics of Vinen’s vortex length density L = |ω|/κ. In the superfluid
turbulence decay experiments reported by Skrbek, Niemela and Donnelly [1999]
the spatial integral of this quantity is measured as a function of time to decrease
over six decades as t−3/2 . The integrated vortex length measured in these ex-
periments is predicted by the renormalized HVBK equations to be governed by
the superfluid vorticity dynamics alone.
Upon including mutual friction, the superfluid vortex dynamics for the renor-
malized HVBK equations is expressed as, cf. equation (9),
∂t ω = curl(vL × ω) ,
in which the renormalized total vortex line velocity given by, cf. equation (5),
B  ρn Bρn
vL = v − s− ω̂ × s , where s = v − vn , (15)
2ρ 2ρ
and its Hamiltonian limit is found to be
v = v̄ + ρ−1 curl λ , with v̄ = ρ−1 J and λ = λ ω̂ . (16)
Thus, relative to the Hamiltonian approach, the terms in B and B  are additional
velocities introduced by phenomenology, while v is the vortex line velocity in
the absence of mutual friction.
 Introduction to HVBK Dynamics 119

The HVBK superfluid vorticity equation implies the following dynamics for
the integrated vortex length measured in the turbulence decay experiments,
 
d 3
Ld x = ω̂ · ∂t ω/κ d 3 x
dt
!" # !" #
length vorticity dynamics

= L ṽ · (ω̂ × curl ω̂) d 3 x
!" #
transport · curvature

Bρn
− L (ω̂ × curl λ ω̂) · (ω̂ × curl ω̂) d 3 x
2ρ2
!" #
damping by curvature

   Bρn 
+ L n̂ × ω̂ · (ω̂ × v ) + (v − vs ) dS . (17)
2ρ
!" #
creation and destruction at the boundary

Here ω̂ is the unit vector tangent to a superfluid vortex filament, so κ = (ω̂ ×

curl ω̂) is its local curvature. The transport and damping of the vortex tangle
length is proportional to this local curvature. The effective velocity ṽ in the
transport term is given by

B  ρn Bρn
ṽ = v − (v − vn ) − ω̂ × (v − vn ) . (18)
2ρ 2ρ

According to the last term in (17), vortex length is created or destroyed at the
boundary, unless the vortex filaments approach it in the normal direction, so
that n̂ × ω̂ = 0.
Formula (17) for the evolution of the total superfluid vortex length presents a
trade-off between the mass-weighted velocity v and the local induction velocity
(or filament curvature) ω̂ × curl ω̂, in the interior of the domain. This trade-
off in the interior competes with the process of creation and destruction at the
boundary. For example, in counterflow turbulence, the superfluid moves toward
the heater at the boundary, so the term in v would tend to be nonzero. In
contrast, for grid turbulence, v is small, so this term would tend to contribute
less.
 This formula governs the dynamics of the experimentally measured quantity
L d 3 x. However, it does not yet show how to obtain the t−3/2 decrease seen in
this quantity by Skrbek, Niemela and Donnelly [1999] in their experiments on
decay of turbulence.
Suppose the main source of decay were the term labeled “damping by cur-
vature” in formula (17) and the flow were isothermal and incompressible. This
would imply

1 d L/R2  3
L = − c0 (T ) λ0 =− , (19)
L dt L 2 (t + t0 )
120 Darryl D. Holm

where λ0 = λ/ρs = (κ/4π)ln(b/a) is the quantum vortex constant, t0 is a time

shift in the experimental analysis, c0 (T ) ≡ Bρn ρs /(2ρ2 ) and angle brackets ·
denote spatial integral over the measurement domain. In particular,
 
L ≡ L d x , L/R  ≡ L |ω̂ × curl ω̂|2 d 3 x .
3 2
(20)

The measured t−3/2 decrease in L implies via formula (19) that the length-
weighted mean curvature of the vortex tangle L/R2 /L decays due to mutual
friction as t−1 . Thus, on the average as the vortex length decays, the vor-
tices tend to straighten, under the effects of mutual friction damping.
Preservation of helicity versus preservation of vortex length. The he-
licity, or linkage number for the superfluid vorticity is defined as

Λ = (vs · ω) d 3 x .

The helicity satisfies an evolution equation obtained from the superfluid vortex
dynamics,
   
dΛ 1
= − (n̂ · ω) μ − vn2 − vs · (vL − vn ) dS − (n̂ · vL )(vs · ω) dS .
dt 2
Therefore, even with mutual friction, helicity is created and destroyed only on
the boundary. Moreover, helicity will be preserved, provided both ω and vL are
tangential at the boundary. The former condition, however, is the opposite of
that required for the creation and destruction of vortex length at the boundary
to cease. Therefore, no equilibrium should be expected that preserves both the
helicity and the vortex length in a superfluid.
Superfluid vortex equilibria are not ABC flows. The steady equilibrium
solutions of the superfluid vorticity dynamics satisfy
curl (vL × ω) = 0 . (23)
For example, a steady equilibrium exists when ω and vL are parallel. Note that
these “super-Beltrami flows” are not eigenfunctions of the curl. Therefore, they
are not Arnold-Beltrami-Childress (ABC) flows, as occur for the Euler equations.

2 Incompressible Renormalized HVBK Flows

To express the renormalized HVBK equations in the incompressible limit, we
begin by recollecting the compressible equations and abbreviating |vs − vn |2 =
vs2 n ,
∂t S = − div(Svn ) + R/T ,
∂t ρ = − div(ρs vs + ρn vn ) ,
   1 
ρs ∂t vs + (vs · ∇)vs = − ρs ∇ μ − vs2 n + ρs (vL − vs ) × ω ,
   2 
∂t ρs vs i + ρn vn i = − ∂k ρn vn i vnk + ρs vs i vsk − ∂i P − ∂k τik ,
 
τik = klm ∂ l vs i λm − λi ω k + δik λ · ω .
 Introduction to HVBK Dynamics 121

As we have seen, finite temperature effects renormalize the total vortex line
velocity as
 
ρn B B
vL = v − ω̂ × s + s , where s = v − vn , (24)
ρ 2 2

and the Hamiltonian part of the line velocity (with corrections for finite temper-
ature) is defined as

v = v̄ + ρ−1 curl λ , with v̄ = ρ−1 (ρs vs + ρn vn ) and λ = λ ω̂ . (25)

To the extent that ρ, ρs , ρn and S all may be taken as constants for a given
temperature and the heating rate R is negligible, then the velocities vn and vs
are incompressible, i.e.,

∇ · vn = 0 and ∇ · vs = 0 .

In this situation, the pressure P may be obtained from the Poisson equation,
   
−∇2 P + λ · ω = div ρs (vs · ∇) vs + ρn (vn · ∇) vn − ω · ∇λ , (27)

found from the divergence of the total momentum equation. Combining the su-
perfluid motion equation with total momentum conservation results in an equa-
tion for the normal fluid velocity in the incompressible case
 ρs 
ρn ∂t vn + ρn (vn · ∇)vn = − ∇ P  − ρs μ + vs2 n − ρs (vL − vs ) × ω + ω · ∇λ ,
2
where vL is given in equation (24). We set P  ≡ P + λ · ω and take it as the
total pressure. (One also could have absorbed λ · ω into P earlier, by including
it in Euler’s pressure law.) Since λ = |λ|ω̂ and ω̂ is a unit vector, we find for
constant ρs the standard relation for the vortex line tension denoted as T.
Namely,
ω · ∇λ = − λ0 ρs ω × curl ω̂ ≡ ρs T ,
where λ0 = λ/ρs = (κ/4π)ln(b/a) is a constant.
Remark. We note that the quantity T known as the vortex line tension first
appears in the normal fluid equation, as a reaction to the presence of the super-
fluid. The standard convention for introducing the mutual friction force has the
effect of shifting T into the superfluid equation. By action and reaction, though,
T could appear in either equation.
These equations of motion must be completed by providing an equation of
state relation for the quantity μ − 12 vs2 n . BK [1961] assumes a law of partial
pressures,
ρn  ρs  1
Pn = P = P  − Ps and Ps = P = ρs μ − ρs vs2 n .
ρ ρ 2
122 Darryl D. Holm

In this case, the renormalized HVBK motion equations for incompressible flow
reduce to
1 ρn
∂t vs + (vs · ∇)vs = − ∇P  − Fn s + T , (31)
ρ ρ
1 ρs
∂t vn + (vn · ∇)vn = − ∇P  + Fn s . (32)
ρ ρ
In these superfluid motion equations, the renormalized mutual friction force
Fn s is defined as the sum (with ω = curl vs )
ρ  B B  0
s
Fn s = (s 0 × ω) + Fn0 s , where Fn0 s = ω̂ × s 0 + s × ω . (33)
ρ 2 2
Here Fn0 s is the HVBK mutual friction force without any finite temperature cor-
rections. To acquire these formulas, we used the relations for the incompressible
case,
ρs   ρs  0  ρs 0
s = v − vn = vs + λ0 curl ω̂ − vn = v  − vn = s , (34)
ρ ρ ρ
with s 0 = vs + λ0 curl ω̂ − vn , and we eliminated vL by using the relation
 
  ρ s ρn B B
− ρs vL − vs − λ0 curl ω̂ = ρn s + ω̂ × s + s . (35)
ρ 2 2

In equation (33) for Fn s , the quantity (s 0 × ω) is the Hamiltonian reactive

force (which could be naturally absorbed into Vinen’s B  parameter) and Fn0 s
is the phenomenological mutual friction force defined according to the standard
convention as in BK [1961] and Donnelly [1999]. The finite-temperature correc-
tions contribute an overall factor of ρs /ρ to the standard zero-temperature
expression Fn0 s for the phenomenological mutual friction force. No new parame-
ters are added, but a temperature dependent trade-off is identified between
the renormalized mutual friction coupling and the vortex line tension, since the
vortex line tension remains unaffected by the finite-temperature corrections.
In the isothermal case, the motion equations are closed by the Poisson equa-
tion for P , since the other coefficients (B, B  , ρn /ρ, etc.) are specified functions
of temperature and they may be taken as constants, for an isothermal incom-
pressible superfluid flow.
Note that equations (31-32) may be rewritten with ωn = curl vn as
ρs 0 ρn ρs 1
∂t vs + ∇μs = vn × ω + s × ω − 2 Fn0 s , μs = P /ρ + vs2 , (36)
ρ ρ 2
ρs 0 ρ2s 0 1
∂t vn + ∇μn = vn × ωn + s × ω + 2 Fn s , μn = P /ρ + vn2 . (37)
ρ ρ 2
These equations imply an equation for the velocity difference,
1 ρs
∂t (vs −vn )+ ∇(vs2 −vn2 ) = vn ×(ω −ωn )− Fn0 s with ωn = curl vn . (38)
2 ρ
 Introduction to HVBK Dynamics 123

and there is no tendency for mutual friction to cause any alignment in the
vorticities of the superfluid and its normal component. Instead, the curl ω̂ part
of Fn0 s = 0 would break any such alignment, if it were to form spontaneously.
Indeed, alignments sufficient for steady solutions are

s 0 × (∇μs × ∇μn ) = 0 , s0 × ω = 0 and s 0 × vn = 0 , (39)

with s ≡ vs + λ0 curl ω̂ − vn ,
0

provided μs and μn are functionally unrelated. Thus, the steady equilibrium

alignments imposed by mutual friction involve vn , vs and curl ω̂, as well as the
independent gradients of μs and μn . For example, one class of equilibria has s 0 ,
vn , ω all aligned tangent to intersections of level surfaces of μn and μs .

3 Rotating Frame Renormalized HVBK Equations

We transform to a rotating frame with relative velocities denoted with an as-

terisk as vs∗ = vs − R(x), etc., and curl R = 2Ω. After a calculation involving
Legendre transformations, we obtain the Hamiltonian for the relative motion,
cf. the Hamiltonian in (61) of the Appendix,
 
∗ ∗ 1
h(M , ρ, S, u, A , n) = ρ |vn∗ + R(x)|2 + (M∗ − ρA∗ − ρvn∗ ) · vn∗ (40)
2
 
+ ε0 (ρ, S, vs∗ − vn∗ , ω ∗ + 2Ω) − R · ρ(vn∗ + R) + (ρ − n)(A∗ − R) d 3 x .

Here M∗ − ρA∗ = J∗ = J − ρR and vs = u − (A∗ − R). The equations resulting

from the Lie-Poisson bracket (55) of the Appendix in these relative variables
keep their forms and the condition n = ρ is still preserved. We conclude with
the following three remarks.
Superfluid Coriolis force couples to the vortex line velocity. The Hamil-
tonian evolution equation for the superfluid velocity in the rotating frame is
expressed as
 1 1 
∂t vs∗ + (vs∗ · ∇)vs∗ = − ∇ μ − |vs∗ − vn∗ |2 − |R|2 + (v∗ − vs∗ ) × ω ∗ + v∗ × 2Ω .
2 2
The last term is the Coriolis force and it involves the relative vortex line velocity.
The curl of this equation yields
 
∂t (ω ∗ + 2Ω) = curl v∗ × (ω ∗ + 2Ω) .

The form of the vortex dynamics equation is invariant under passing to a

steadily rotating frame, and the superfluid Coriolis force contains the renormal-
ized vortex line velocity, rather than the superfluid velocity. Therefore, this is
not merely a kinematic force. The vortex line velocity appearing in the superfluid
Coriolis force includes the interaction between the vortex lines and the superfluid
124 Darryl D. Holm

component. It also includes the interaction with the normal component, since
v depends on the relative momentum density and contains finite temperature
effects. The superfluid Coriolis force is essential in the spin up problem in He-II,
see, e.g., Reissenegger [1993].
Superfluid Taylor-Proudman theorem. For steady, or slow motions and
rapid rotation we have  
0 = curl v∗ × 2Ω .
If the rotation is uniform (∇Ω = 0) and oriented vertically (Ω = |Ω|ẑ) this
becomes
   T
0 = 2|Ω| ∂z v∗ − ẑ div v∗ = 2|Ω| ∂z v∗x , ∂z v∗y , −∂x v∗x − ∂y v∗y ,

where ( )T denotes transpose of a row vector into a column vector. Thus, for
steady, or slow motions and rapid uniform rotation, we find that vortex line
motion is columnar. That is, the lateral vortex line velocity is nondivergent
and independent of the axial coordinate, and the axial velocity decouples from
the lateral motion. Therefore, under sufficiently rapid rotation, the superfluid
vortex filaments will straighten and become parallel to the axis of rotation as
they approach a steady state. However, they may still undergo nondivergent
motion in the lateral plane. This superfluid Taylor-Proudman theorem explains
why steady superfluid vortices tend to be aligned with the rotation axis under
rapid uniform rotation. The same conclusion applies, if the velocity v∗ in the
Hamiltonian formulation is replaced by the phenomenological relative velocity
∗
vL = vL − R. Similar considerations are discussed in Sonin [1987] from a more
microscopic viewpoint.
Relative total momentum is not conserved for rotating compressible
flows. Since the Hamiltonian depends explicitly on spatial position, instead of
conserving relative total momentum, we have the balance
∂ h  ρ
∂t Ji∗ + ∂j Ti∗ j = − i  = ∂i |R|2 ,
∂x explicit 2
where h is the Hamiltonian density in equation (40). This relative momentum
balance is the effect of centrifugal force. Here we have dropped terms proportional
to ρ − n, since ρ = n is still preserved in a rotating frame. Consequently, the
stress tensor in the relative momentum equation also keeps its form in passing to
a rotating frame, although the total relative momentum is no longer conserved
if the flow is compressible.

Acknowledgments
I am grateful to H. R. Brand, A. Brandenburg, P. Constantin, R. Donnelly, V.
V. Lebedev, F. Lund, J. E. Marsden, J. Niemela, A. Reisenegger, L. Skrbek, K.
Sreenivasan and W. F. Vinen for stimulating discussions and encouragement. I
am also grateful for hospitality at the UC Santa Barbara Institute for Theoretical
Physics where this work was initiated during their Hydrodynamic Turbulence
 Introduction to HVBK Dynamics 125

program in spring 2000. This research was supported by the U.S. Department of
Energy under contracts W-7405-ENG-36 and the Applied Mathematical Sciences
Program KC-07-01-01.

Appendix: Lie-Poisson Hamiltonian Formulation

Conservation of the number of quantum vortices moving through superfluid 4 He
(and across the streamlines of the normal fluid component) is expressed by

d
ω · n̂ dS = 0 , (46)
dt S

where the superfluid vorticity ω is the areal density of vortices and n̂ is the unit
vector normal to the surface S whose boundary ∂S moves with the vortex line
velocity v . When ω = curl vs this is equivalent to a vortex Kelvin theorem

d
vs · dx = 0 , (47)
dt ∂S(v )

which in turn implies the fundamental relation

∂t vs − v × ω = ∇μ . (48)

The superfluid velocity naturally splits into vs = u − A, where u = ∇φ

and (minus) the curl of A yields the superfluid vorticity ω. The phase φ is
then a regular function without singularities. This splitting will reveal that the
Hamiltonian dynamics of superfluid 4 He with vortices may be expressed as an
invariant subsystem of a larger Hamiltonian system in which u and A have
independent evolution equations.
We begin by defining a phase frequency in the normal velocity frame as

∂t φ + vn · ∇φ = ν . (49)

The mass density ρ and the phase φ are canonically conjugate in the Hamiltonian
formulation. Therefore, one may set ν = − δh/δρ for a given Hamiltonian h and
the phase gradient u = ∇φ satisfies
δh
∂t u + vn · ∇u + (∇vn )T · u = − ∇ , (50)
δρ

where ( )T denotes transpose, so that (∇vn )T · u = uj ∇vnj . The mass density ρ

satisfies the dual equation
δh
∂t ρ + ∇ · (ρvn ) = − ∇ · . (51)
δu
Perhaps not surprisingly, the rotational and potential components of the super-
fluid velocity vs = u−A satisfy similar equations, but the rotational component
126 Darryl D. Holm

is advected by the vortex line velocity v , instead of the normal velocity vn . Ab-
sorbing all gradients into u yields

∂t A + v × ω = 0 . (52)

Taking the difference of the equations for u and A then recovers equation (48)
as  δh 
∂t vs − v × ω = − ∇ vn · u + with vs = u − A , (53)
δρ
in which one uses regularity of the phase φ to set curl u = 0. It remains to
determine v from the Hamiltonian formulation. Including the additional degree
of freedom A associated with the vortex lines allows them to move relative to
both the normal and super components of the fluid, and thereby introduces
an additional reactive force without introducing any additional inertia. This
Hamiltonian approach thus yields renormalized HVBK equations.
Proposition: Upon splitting the superfluid velocity into vs = u − A (with u =
∇φ so that ω = − curl A) the (renormalized) HVBK equations in the Galilean
frame of the normal fluid form an invariant subsystem of a Lie-Poisson
Hamiltonian system,
∂f
= {f, h} with f, h ∈ (M, ρ, S, u, A, n) ,
∂t
and Lie-Poisson bracket given by

{f, h} =
   
δf δh δh δh δh
− (Mk ∂j + ∂k Mj ) + ρ ∂j + S∂j + (∂k uj − uk , j )
δMj δMk δρ δS δuk
   
δf δf δf δh δf δh δf δh
+ ∂k ρ + ∂k S + (uk ∂j + uj , k ) + ∂k + ∂j
δρ δS δuj δMk δρ δuk δuj δρ
 
δf δh Aj , k − Ak , j δh δf δh
− ∂j + − ∂k d 3x . (55)
δAj δn n δAk δn δAk

Remarks: Here M is the total momentum density, the total mass density is ρ
and the entropy density is S. We shall interpret the density n later, after we
develop the Hamiltonian equations of motion. It shall emerge that n = ρ is an
invariant condition and, hence,

M − nA = J = p + ρvn ,

for n = ρ, where p = J − ρvn = ρs (vs − vn ) is the relative momentum density of

the superfluid in the Galilean frame of the normal fluid. The momentum density
associated with the vortex fluid will be N = −nA. The Hamiltonian will be the
energy E in (7).
The Lie-Poisson bracket in the Proposition appeared first in Holm and Ku-
pershmidt [1987] in a study of various approximate equations for the dynamics of
 Introduction to HVBK Dynamics 127

multicomponent superfluids with charged condensates. The mathematical struc-

ture of this Lie-Poisson bracket and its association with the dual of a certain Lie
algebra is discussed in Holm and Kupershmidt [1987]. Our re-interpretation of
this Poisson bracket introduced and studied earlier shall now yield a extension of
the HVBK equations that enables the vortex line velocity v and hence the vor-
tex reactive force and mutual friction force to be expressed at finite temperature.
Identifying this Poisson bracket as being dual to a Lie algebra establishes that
it satisfies the Jacobi identity, ijk {fi , {fj , fk }} = 0. The term in the Poisson
bracket responsible for the reactive vortex force will turn out to be {Ai , Aj } = 0.
The Poisson bracket {vs i , vs j } would vanish (as does {ui , uj } = 0) and thus
the reactive vortex force would be absent, in any Hamiltonian formulation for
which A and n were not independent degrees of freedom from M, ρ, S. Volovik
and Dotsenko [1979, 1980] obtain a different result and provide no Lie-algebraic
justification.
A Lagrangian formulation of these equations is also available. However, it
involves an equation for δl/δν about which nothing is known physically.
Corollary #1: The Lie-Poisson bracket is equivalent to the following separate
Hamiltonian matrix forms for the dynamical equations
⎡ ⎤ ⎡ ⎤⎡ ⎤
Mi Mj ∂i + ∂j Mi S∂i ρ∂i ∂j ui − uj , i δh/δMj
∂ ⎢ ⎥
⎢ S ⎥ = −⎢
⎢ ∂j S 0 0 0 ⎥ ⎢ δh/δS ⎥
⎥⎢ ⎥
⎣
∂t ρ ⎦ ⎣ ∂j ρ 0 0 ∂j ⎦ ⎣ δh/δρ ⎦ , (57)
ui uj ∂i + ui , j 0 ∂i 0 δh/δu j
and, upon defining N = −nA,
    
∂ Ni Nj ∂i + ∂j Ni n∂i δh/δNj
=− . (58)
∂t n ∂j n 0 δh/δn
These are individually expressed as
∂t S = {S, h} = − div(S δh/δM) ,
∂t n = {n, h} = − div(n δh/δN) ,
∂t ρ = {ρ, h} = − div(ρ δh/δM + δh/δu) ,
 
∂t u = {u, h} = − ∇ δh/δρ + (δh/δM) · u + (δh/δM) × curl u ,
∂t (N/n) = {(N/n), h}
   
= − ∇ δh/δn + (δh/δN) · (N/n) + δh/δN × curl (N/n) ,
 
∂t Mj + Nj = {Mj + Nj , h} = − ∂k Tjk .

Corollary #2: Consider a translation invariant Hamiltonian density with de-

pendence
h(M, ρ, S, n, vs , ω, A),
where vs = u−A, A = −N/n and ω = curl vs . The stress tensor Tjk is expressed
in terms of this Hamiltonian as
  
∂h ∂h  ∂h  ∂h ∂h 
k
T j = Mj +vs j + curl −vs l, j mlk + δj P −Aj
k
.
∂Mk ∂vs k ∂ω k ∂ωm ∂Ak vs
 128 Darryl D. Holm

where
∂h ∂h ∂h ∂h
P = Ml +ρ +S +n − h,
∂Ml ∂ρ ∂S ∂n
as in the Euler relation for pressure.
Remark. One notes many parallels and correspondences among these equations.
Note especially the expected similarities in the equations for u and N/n. Recall
that A = − N/n, so that the superfluid velocity is given by vs = u − A =
u + N/n. The evolution of the superfluid velocity is consistently composed as
the sum of these two separate dynamical pieces.
Proof of the Proposition: The following Hamiltonian h (and conserved en-
ergy) will yield the HVBK equations in the frame of the normal fluid upon using
this Lie-Poisson bracket
  
1 2
h = d x − ρ vn + (M − ρA) · vn + ε0 (ρ, S, vs − vn , ω) .
3
(61)
2
The variational derivatives of the Hamiltonian h are computed in this reference
frame by using the thermodynamic first law (2). Namely,
  1 
δh = d 3 x μ − vn2 − A · vn δρ + T δS + vn · δM + (p + curl λ) · δu
2
    
− p + curl λ + ρvn · δA + M − p − ρvn − ρA · δvn .

Here we have used the velocity split δvs = δu − δA and assumed the boundary
condition n̂ · ω × λ = 0 when integrating by parts. This boundary condition
is satisfied identically, since λ = λ ω̂ in the HVBK theory. Upon substituting
these variational derivatives into the Lie-Poisson bracket, Corollary #1 yields
the following equations expressed in the normal fluid reference frame,

∂t S = {S, h} = − div(Svn ) ,
∂t n = {n, h} = − div(ρvn + p + curl λ) ,
∂t ρ = {ρ, h} = − div(ρvn + p + curl λ) ,
( Hence, the condition n = ρ is preserved.)
1
∂t u = {u, h} = − ∇(μ − vn2 + vn · vs ) + vn × curl u ,
2
( Hence, curl u = 0 is preserved.)
 
∂t A = {A, h} = n−1 ρvn + p + curl λ × curl A ,
( Hence, v = vn + ρ−1 (p + curl λ) when n = ρ is used.)
 
∂t Mj − nAj = {Mj − nAj , h} = − ∂k Tjk

Remarks:
(1.) Preservation of the condition n = ρ by these equations allows the introduc-
tion of the momentum-carrying field A as an independent degree of freedom
 Introduction to HVBK Dynamics 129

without introducing additional material inertia, provided the dynamically pre-

served condition n = ρ holds initially. This is reminiscent of the preservation
of Gauss’s Law by the continuity equation for mass conservation in a fluid
plasma.
(2.) The curl of the dynamical equation for the field A implies the vortex line
velocity
1 δh
v = − = v̄ + ρ−1 curl λ , where v̄ = vn + ρ−1 p = ρ−1 J .
n δA
The velocity v̄ is the mass averaged velocity. The vortex slip velocity s corre-
sponding to the vortex line velocity v is the basis for the phenomenological
reactive and mutual friction forces f and Rayleigh dissipation function R in
the HVBK system. Namely,
s = v − vn = ρ−1 (p + curlλ) , with λ = λ ω̂ .
As expected, this expression agrees with BK [1961] at zero temperature. Note
that the renormalized HVBK equations introduce no new parame-
ters.
(3.) The corresponding equation for vs = u − A is then obtained as
1 2
∂t vs = − ∇(μ − v + vn · vs ) + v × ω , where v = v̄ + ρ−1 curl λ ω̂ .
2 n
This may be expressed equivalently in manifestly Galilean invariant form as
 1 
∂t vs + (vs · ∇)vs = − ∇ μ − |vs − vn |2 + f  , where f  = (v − vs ) × ω .
2
The term f  is the Hamiltonian contribution to the reactive vortex
force. This contribution would vanish if the vortex lines moved with the
superfluid velocity.
(4.) The stress tensor Tjk = πjk + τjk for total momentum conservation is given by
summing
 
πjk = ρs vs j vsk + ρn vn j vnk + P δjk ,
 
τjk = ∂l k l m vs j λm − λj ω k + ω · λ δjk .

The divergence of τjk defines the vortex line tension T as

∂k τjk = − ω · ∇λ + ∇(ω · λ) = − ρs T + ∇(ω · λ) ,

In the stress tensor πjk the pressure P is defined by the Euler relation,
P = − ε0 + μρ + T S ,
so that in the normal-fluid frame the pressure satisfies
dP = ρdμ + SdT − p · d(vs − vn ) − λ · d ω .
The stress tensor Tjk = πjk + τjk may be derived by using Corollary #2 for
the Hamiltonian formulation.
130 Darryl D. Holm

Implications of the HVBK vortex dynamics.

The new Hamiltonian formulation of the renormalized HVBK equations pre-
sented in the Proposition provides a formula for the slip velocity of a vortex
line element in a turbulent superfluid at finite temperature. Namely, for the
Hamiltonian h in equation (61), one finds
 
s = v − vn = ρ−1 p + curl λ . (69)
This formula for v recovers the HVBK expression in BK [1961] at zero tem-
perature. Otherwise, it provides an extension to finite temperature of the
HVBK vortex force
f = (vL − vs ) × ω , with vL = vn − ρs (α s + β ω̂ × s) , (70)
where the renormalized vortex slip velocity is given by
ρs 0
s = v  − vn = (v − vn ) , for constant ρs . (71)
ρ 
The corresponding heating rate R is given by
R = (J − ρvn + curl λ) · ω × (vL − vn ) = ρρs β ω |s × ω̂|2 ,
which is positive, as it must be.

References
1. Bekarevich, I. L. and I. M. Khalatnikov [1961] Phenomenological derivation of the
equations of vortex motion in He II, Sov. Phys. JETP 13 643-646.
2. Donnelly, R. J. [1999] Cryrogenic fluid dynamics, J. Phys.: Condens. Matter 11
7783-7834.
3. Gorter, C. J. and J. H. Mellink [1949] Physica 15 285.
4. Henderson, K. L. and C. F. Barenghi [2000] The anomalous motion of superfluid
helium in a rotating cavity, J. Fluid Mech. 406 199-219.
5. Hills, R. N. and P. H. Roberts [1977] Superfluid mechanics for a high density of
vortex lines, Arch. Rat. Mech. Anal. 66 43-71.
6. Holm, D. D. and B. Kupershmidt [1987] Superfluid plasmas: multivelocity non-
linear hydrodynamics of superfluid solutions with charged condensates coupled
electromagnetically, Phys. Rev. A 36 3947-3956.
7. Nemirovskii, S. K. and W. Fiszdon [1995] Chaotic quantized vortices and hydro-
dynamic processes in superfluid helium, Rev. Mod. Phys. 67 37-84.
8. Putterman, S. J. [1974] Superfluid Hydrodynamics, North Holland, Amsterdam.
9. Reissenegger, A. [1993] The spin up problem in Helium-II J. Low Temp. Phys. 92
77-106.
10. Skrbek, L., J. J. Niemela and R. J. Donnelly [1999] Turbulent flows at cryogenic
temperatures: a new frontier, J. Phys.: Condens. Matter 11 7761-7783.
11. Sonin, E. B. [1987] Vortex oscillations and hydrodynamics of rotating superfluids,
Rev. Mod. Phys. 59 87-155. 1987
12. Volovik, G. E. and V. S. Dotsenko [1979] Poisson brackets and continuous dynam-
ics of the vortex lattice in rotating He-II, JETP Lett. 29 576-579.
13. Volovik, G. E. and V. S. Dotsenko [1980] Hydrodynamics of defects in condensed
media in the concrete cases of vortices in rotating Helium-II and of disclinations
in planar magnetic substances, Sov. Phys. JETP 58 65-80.
Magnus Force, Aharonov–Bohm Effect,
and Berry Phase in Superfluids

Edouard Sonin

Racah Institute of Physics, Hebrew University of Jerusalem

1 Introduction
If the vortex moves with respect to a liquid, classical [1] or quantum, there
is a force on the vortex normal to the relative vortex velocity with respect to
the liquid. This is the Magnus force, which plays an important role in mod-
ern condensed-matter physics. In particular, it determines the mutual fiction in
superfluids [2,3,4] and the Hall effect in superconductors [5].
An obvious generalization of the classical Magnus force in the superfluid
seemed to be a force proportional to the superfluid density ρs [2]:

ρs [(v L − v s ) × κ] = F , (1)

where v L is the vortex velocity, v s is the superfluid velocity, and κ is a vector

along the vortex line with κ = h/m being the circulation of the superfluid
velocity around the vortex. In absence of the external force F on a superfluid,
the vortex moves with the superfluid velocity. However, quasiparticles (rotons [6]
and phonons [7]) produce an additional force transverse to the vortex velocity
and
F = −D(v L − v n ) − D [ẑ × (v L − v n )] . (2)
Here v n is the normal velocity. It was shown [8] that the additional transverse
force ∝ D = −κρn (the Iordanskii force) is connected with the analogue of the
Aharonov–Bohm effect [9] in superfluids. A similar force on the magnetic-flux
tube exists in the original Aharonov–Bohm effect for electrons [10].
Ao and Thouless [11] have pointed out that the Magnus force is connected
with the Berry phase [12], which is the phase variation of the quantum-mechanical
wave function of a quantum liquid generated by the adiabatic transport of the
vortex round a close loop. From the Berry-phase analysis Ao and Thouless [11,13]
concluded that the only force transverse to v L is the force proportional to ρs ,
i.e., D = 0 in Eq. (2). This conclusion disagreed with the previous calculations
and therefore was vividly discussed [14,15].
The present paper is an attempt to bring together two points of view in
order to find a source of disagreement [16]. I restrict myself with the problem of
the Galilean invariant quantum Bose-liquid described by the Gross–Pitaevskii
theory [17]. At large scales the theory yields equations of the hydrodynamics of
an ideal inviscous liquid. In presence of an ensemble of sound waves (phonons)
with the Planck distribution, which is characterized by a locally defined normal
velocity, one obtains the two-fluid hydrodynamics. The momentum balance in the

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 131–137, 2001.

c Springer-Verlag Berlin Heidelberg 2001
132 Edouard Sonin

area around a moving vortex demonstrates the existence of the Iordanskii force. I
also discuss the Berry phase. According to Refs. [11,13], the Berry phase and the
Magnus forces are proportional to the total current circulation at large distances.
But the total current circulation contains a normal-fluid contribution, which is
proportional to the Iordanski force. Taking this contribution into account, the
Berry-phase analysis agrees with the momentum-balance approach.

2 Gross–Pitaevskii Theory
and Two-Fluid Hydrodynamics
In the Gross–Pitaevskii theory [17] the ground state and weakly excited states
of a Bose-gas are described by the nonlinear Schrödinger equation
∂ψ 2 2
i =− ∇ ψ + V |ψ|2 ψ (3)
∂t 2m
for the condensate wave function ψ = a exp(iφ). The nonlinear Schrödinger
equation is the Euler–Lagrange equation for the Lagrangian
 
i ∂ψ ∂ψ ∗ 2 V
L= ψ∗ −ψ − |∇ψ|2 − |ψ|4 . (4)
2 ∂t ∂t 2m 2

The Noether theorem yields the momentum conservation law ∂ji /∂t+∇j Πij = 0
where j = Im{ψ ∗ ∇ψ} is the mass current, and the momentum-flux tensor is
 
2 ∗ ∗ V 2 2 2
Πij = (∇i ψ∇j ψ + ∇i ψ ∇j ψ) + δij |ψ| −
4
∇ |ψ| . (5)
2m 2 4m

Using the Madelung transformation [3], one obtains from complex Eq. (3)
two real equations for the liquid density ρ = ma2 and the liquid velocity v =
(/m)∇φ = (κ/2π)∇φ. Far from the vortex line these equations are hydrody-
namic equations for an ideal inviscous liquid:
∂ρ
+ ∇(ρv) = 0 , (6)
∂t

∂v
+ (v · ∇)v = −∇μ . (7)
∂t
Here μ = V a2 /m is the chemical potential. Equation (5) becomes the hydrody-
namic momentum-flux tensor Πij = P δij + ρvi (r)vj (r).
A plane sound wave propagating in the liquid generates the phase variation
φ(r, t) = φ0 exp(ik · r − iωt). Then ρ(r, t) = ρ0 + ρ(1) (r, t) and v(r, t) = v 0 +
v (1) (r, t), where ρ0 and v 0 are the average density and velocity in the liquid,
whereas ρ(1) (r, t) and v (1) (r, t) = (κ/2π)∇φ are periodical variations of the
density and the velocity due to the sound wave (ρ(1)  = 0, v (1)  = 0). Equations
(6) and (7) linearized with respect to ρ(1) and v (1) yield the sound equation for

φ with the sound velocity cs = V a2 /m and the spectrum ω = cs k + k · v 0 .
 Magnus Force, Aharonov–Bohm Effect, and Berry Phase 133

The total mass current expanded up to the terms of the second order with
respect the wave amplitude and averaged over time is

κ2 k
j = ρ0 v 0 + ρ(1) v (1)  = ρ0 v 0 + ρ0 φ20 k. (8)
8π 2 cs
If there is an ensemble of phonons with the Planck distribution
 −1  −1
E(p) − p · v n cs p + p · (v 0 − v n )
n0 (E, v n ) = exp −1 = exp −1 ,
T T
(9)
the total mass current linearized with respect to v 0 − v n is

1
j = ρ0 v 0 + 3 n0 (p)p d3 p = ρ0 v 0 + ρn (v n − v 0 ) . (10)
h
Here p = k is the phonon momentum, E = cs p+p·v 0 is the phonon energy, and
v n is the drift velocity of phonons. This expression is equivalent to the two-fluid
expression assuming that ρ = ρ0 = ρs + ρn , v 0 = v s , and  the normal density is
given by the usual two-fluid expression ρn = −(1/3h3 ) [∂n0 (ε, 0)/∂E]p2 d3 p.
Expanding the momentum-flux tensor up to terms of the second-order with
respect to the sound wave amplitude one obtains:
 
c2s ρ(1)  v(1) 
2 2
Πij = P0 δij + ρ0 v0i v0j + − ρ0 δij
ρ0 2 2
+ρ(1) (v(1) )i v0j + ρ(1) (v(1) )j v0i + ρ0 (v(1) )i (v(1) )j  . (11)

For the Planck distribution this yields the two-fluid momentum flux tensor

Πij = P δij + ρs vsi vsj + ρn vni vnj . (12)

Taking into account phonon-phonon interaction, which establishes the local

Planck distribution of phonons, we can derive all equations of the two-fluid
hydrodynamics, as shown in Ref. [18]. Thus the two-fluid hydrodynamics can
be derived from the hydrodynamics of an ideal inviscous fluid, and the latter
follows from the Gross–Pitaevskii theory.

3 Interaction of Phonons with a Vortex in Hydrodynamics

In presence of a vortex the sound equation is (see Refs. [8,14] for more details)

∂2φ ∂φ
2
− c2s ∇2 φ = −2v v (r) · ∇ , (13)
∂t ∂t
where
κ×r
v v (r) = (14)
2πr2
134 Edouard Sonin

is the circular velocity field induced by a vortex line. Here r is the position vector
in the plane xy. The sound wave produces the density variation
 
ρ0 κ ∂φ
ρ(1) = − 2 + v v · ∇φ(r) . (15)
cs 2π ∂t
One can calculate the differential cross-section in the Born approximation,
but since it is quadratic in κ this does not yield a transverse force [7,8,14]. Instead
we consider a quasiclassical solution of the sound equation:
   
iδS iκk
φ = φ0 exp −iωt + ik · r + = φ0 exp(−iωt + ik · r) 1 + θ , (16)
 2πcs
r
where δS = −(k/cs ) v v · dl = θκk/2πcs is the variation of the action due
to interaction with the circular velocity around the vortex. The angle θ is an
azimuth angle for the position vector r measured from the direction opposite
to the wave vector k. This choice provides that the quasiclassical correction
vanishes for the incident wave far from the vortex. One can check directly that
Eq. (16) satisfies the sound equation (13) in the first order of the parameter
κk/cs . The velocity generated by the sound wave around the vortex is
 
κ κ ik
v (1) = ∇φ = φ0 exp(−iωt + ik · r) ik − v v . (17)
2π 2π cs

According to Eq. (16) the phase φ is multivalued, and one must choose a cut for
an angle θ at the direction k, where θ = ±π . The jump of the phase on the cut
line behind the vortex is a manifestation of the Aharonov–Bohm effect [9]: the
sound wave after its interaction with the vortex has different phases on the left
and on the right of the vortex line. This results in an interference [8,14]. The
width of the interference region is dint ∼ r/k. 
Now we consider the momentum balance using the condition that dSj Π⊥j =
0 for a cylindrical surface around the vortex line. The subscript ⊥ points a com-
ponent normal to the wave vector k of the incident wave. The total momentum-
flux tensor can be obtained from Eq. (11) assuming v 0 (r) = v v (r) + v s :

Πij = −ρ0 (v s − v L ) · v v δij + ρ0 v0i v0j

+ρ(1) (v(1) )i vvj + ρ(1) (v(1) )j vvi + ρ0 (v(1) )i (v(1) )j  . (18)

The first two terms in this expression yield the momentum flux without phonons,
which produces the Magnus force for a liquid with the density ρ0 and the ve-
locity v s . The rest terms cancel except for the contribution from the term
ρ0 (v(1) )i (v(1) )j  in the interference region where v(1)⊥ = (κ/2πr)∂φ/∂θ. The
contribution depends on the phase jump and for a single sound wave is:

1 κ2 k 1 κ3 k 2
ρ0 (v(1) )⊥ (v(1) )r rdθ = ρ0 φ20 [δS(−) − δS(+)] = ρ0 φ20 ,
8π 2  8π 2 cs
(19)
where δS(±) = ∓κk/2cs are the action variations at θ → ∓π.
 Magnus Force, Aharonov–Bohm Effect, and Berry Phase 135

For the Planck phonon distribution the condition dSj Π⊥j = 0 yields:

ρ0 [(v L −v s )×κ]−[ρ(1) v (1) ×κ] = ρs [(v L −v s )×κ]+ρn [(v L −v n )×κ] = 0 . (20)

The term ∝ (v L − v n ) is the Iordanskii force, which corresponds to D = −κρn

in Eq. (2). Uniting in Eq. (20) terms linear in v L , we see that the total transverse
force (effective Magnus force) is proportional to the total density ρ0 = ρs + ρn .

4 Momentum Balance in the Two-Fluid Hydrodynamics

Up to now we analyzed spatial scales much less than the mean-free-path lph of
phonons (ballistic region). Now we shall see what is going on at scales much
larger than lph where the two-fluid hydrodynamics is valid.
Interaction between phonons and the vortex in the ballistic region produces
a force concentrated along the vortex line in the hydrodynamic region. The re-
sponse of the normal fluid to this force is described by the Navier-Stokes equation
with the dynamic viscosity ηn :
∂v n ∇P ρs S
+ (v n · ∇)v n = νn Δv n − − ∇T , (21)
∂t ρ ρn ρ
where νn = ηn /ρn is the kinematic viscosity, S is the entropy per unit volume,
and T is the temperature. At r < rm ∼ νn /|v n − v L | one may neglect * the
nonlinear inertial (convection) term (v n · ∇)v n . The line force −Fi = τij dSj
on the normal fluid produces a divergent velocity field (the Stokes paradox [1]):
F r
v n (r) = v n + ln , (22)
4πηn lph
where v n is the normal velocity at distances r ∼ lph , which separate the ballistic
and the hydrodynamic regions. Here τij = −ηn (∇i vnj + ∇j vni ) is the viscous
stress tensor. Due to viscosity the normal velocities v n∞ and v n at large (r ∼ rm )
and small (r ∼ lph ) distances from the vortex line are different (the viscous drag
[2]):
4πηn
F = (v n∞ − v n ) . (23)
ln(rm /lph )
The momentum
* balance
* in the two-fluid region must include the viscous
stress tensor: Πij dSj + τij dSj = 0, where Πij is given by Eq. (12). Since
* the
normal velocity field does not contain the circular velocity v v , the flux Πij dSj
yields the superfluid Magnus force, i.e., the force F satisfies Eq. (1).
However, at very large distances r  rm the nonlinear convection term is
more important than the viscous term. Thus the scale rm separates the viscous
and convection subregions. In the convection subregion the viscosity becomes
ineffective and the momentum flux related to the linear force on the normal
fluid should be connected with the normal part of the momentum flux tensor
Πij given by Eq. (12). The normal momentum transmission requires a circula-
tion of the normal velocity [19] with magnitude determined by the transverse
136 Edouard Sonin

force on the normal fluid. This is confirmed by the solution of the Navier-Stokes
equation obtained by Thouless et al. [20]. However, separation on longitudinal
and transverse components of a force should be done with respect to the normal
velocity v n∞ − v L , but not v n − v L . Using Eq. (23), Eq. (2) for the force from
the normal fluid can be rewritten (neglecting the longitudinal force ∝ D) as
 
1 D ln(rm /lph )
F =−   2 (v L − v n∞ ) + D [ẑ × (v L − v n∞ )] .
D ln(rm /lph ) 4πηn
1+ 4πηn
(24)
The transverse component of this force determines the normal circulation at very
large distances:

D κ
κn = dl · v n = −    2 =  2 , (25)
D ln(rm /lph ) κρn ln(rm /lph )
ρn 1 + 4πηn
1 + 4πηn

where we used the value D = −κρn for the Iordanskii force. In the limit of
a strong viscous drag κρn ln(rm /l)/4πηn  1 the transverse force and related
normal circulation are suppressed [20]. But the effect of the superfluid Magnus
force and the longitudinal force ∝ D is also suppressed in this limit.

5 Magnus Force and the Berry Phase

Let us consider now the Berry phase in the hydrodynamic description using the
Lagrangian obtained from Eq. (4) with the Madelung transformation:
κρ ∂φ κ2 ρ V
L= − ∇φ2 − ρ2 . (26)
2π ∂t 8π 2
The first term with the first time derivative of the phase φ (Wess-Zumino term)
is responsible for the Berry phase Θ = ΔSB /, which is the variation of the
phase of the quantum-mechanical wave function for an adiabatic motion of the
vortex around a closed loop [12]. Here
 
κρ ∂φ κρ
ΔSB = dr dt = − dr dt (v L · ∇L )φ . (27)
2π ∂t 2π
is the classical action variation around the loop and ∇L φ is the gradient of the
phase φ[r − r L (t)] with respect to the vortex position vector r L (t). However,
*∇L φ = −∇φ, where ∇φ is the gradient with respect to r. Then the loop integral
dl yields the circulation of the total current j = (κ/2π) < ρ∇φ > for points
inside the loop, but vanishes for points outside. As a result [13],

ΔSB κ
Θ= =V (dl · j) . (28)
 2π
where V is the volume inside the loop (a product of the loop area and the liquid
height along a vortex). Contrary to Eq. (27), the integral in Eq. (28) is related
with the variation of r, the vortex position vector r L being fixed.
 Magnus Force, Aharonov–Bohm Effect, and Berry Phase 137

If the *circulation of the normal velocity vanished, the current circulation

would be (dl · j) = ρs κ, and the Berry phase and the transverse force would
be proportional to ρs (see Geller et al. [13]). However, according to* Sec. 4, at
very large distances the normal circulation κn does not vanish and (dl · j) =
ρs κ + ρn κn . Using a proper value of κn given by Eq. (25), the Berry-phase
analysis yields the same transverse force as the momentum balance. But one
cannot find κn from the Berry-phase analysis, since the latter deals only with
very large distances, whereas κn is determined by the transverse force, which
arises at small distances (in the ballistic region) and appears in the small-distance
boundary condition for the Navier-Stokes equation in the two-fluid region. This
conclusion agrees with the recent analysis by Thouless et al. [20].
Discussions with Lev Pitaevskii and David Thouless during the present work-
shop had a great impact on my view presented in this article. I appreciate also
interesting discussions with Andrei Shelankov. The work was supported by the
grant of the Israel Academy of Sciences and Humanities.

References
1. H. Lamb, Hydrodynamics (Cambridge University Press, New York, 1975).
2. H.E. Hall and W.F. Vinen, Proc. Roy. Soc. A238, 204 (1956).
3. R.J. Donnelly, Quantized vortices in helium II (Cambridge University Press, Cam-
bridge, 1991), Sec. 2.8.3.
4. E.B. Sonin, Rev. Mod. Phys. 59, 87 (1987).
5. P. Nozières and W.F. Vinen, Phil. Mag. 14, 667 (1966).
6. E.M. Lifshitz and L.P. Pitaevskii, Zh. Eksp. Teor. Fiz. 33, 535 (1957) [Sov. Phys.-
JETP 6, 418 (1958)].
7. S.V. Iordanskii, Zh. Eksp. Teor. Fiz. 49, 225 (1965) [Sov. Phys.-JETP 22, 160
(1966)].
8. E.B. Sonin, Zh. Eksp. Teor. Fiz. 69, 921 (1975) [Sov. Phys.-JETP 42, 469 (1976)].
9. Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959).
10. A.L. Shelankov, Europhys. Lett., 43, 623 (1998).
11. P. Ao and D.J. Thouless, Phys. Rev. Lett. 70, 2158 (1993).
12. M.V. Berry, Proc. R. Soc. London A 392, 45 (1984).
13. M.R. Geller, C. Wexler, and D.J. Thouless, Phys. Rev. B 57, R8119 (1998).
14. E.B. Sonin, Phys. Rev. B 55, 485 (1997).
15. H.E. Hall and J.R. Hook, Phys. Rev. Lett., 80, 4356 (1998); E.B. Sonin, ibid. 81,
4276 (1998); C. Wexler et al., ibid. 80, 4357 (1998); 81, 4277 (1998).
16. A more detailed report on the present analysis is to be published in Proceedings
of the the workshop “Microscopic structure and dynamics of vortices in uncon-
ventional superconductors and superfluids” Dresden, Germany, March 2000, cond-
mat/0104221.
17. E.P. Gross, Nuovo Cimento 20, 454 (1961); L.P. Pitaevskii, Zh. Eksp. Teor. Fiz.
40, 646 (1961) [Sov. Phys.-JETP 13, 451 (1961)].
18. S.J. Putterman and P.H. Roberts, Physica 117 A, 369 (1983).
19. Existence of the normal circulation at large distances in the presence of the trans-
verse force on the vortex was pointed out by Pitaevskii (unpublished).
20. D.J. Thouless, M.R. Geller, W.F. Vinen, J-Y. Fortin, and S.W. Rhee, cond-
mat/0101297.
Using the HVBK Model
to Investigate the Couette Flow of Helium II

Karen L. Henderson

Faculty of Computer Studies & Mathematics, University of the West of England,

Bristol, BS16 1QY, U.K.

Abstract. We review the application of the two-fluid HVBK equations to helium II in

Couette geometry, that is flow between concentric, rotating cylinders. This application
is particularly interesting as a large number of experiments have been carried out in this
geometry and also because Couette flow is an exact solution of the HVBK equations
for both the normal fluid and superfluid.

1 Introduction
When the temperature of liquid helium drops below the transition temperature
of Tλ = 2.172 k a phase transition occurs and it becomes a quantum liquid
called helium II. Helium II can be described macroscopically by Landau’s two-
fluid model in which it is considered to be a mixture of a viscous normal fluid
and an inviscid superfluid. In addition, vortex lines appear in the superfluid
component when helium II rotates or when it moves along a tube faster than
a small critical velocity. Feynman [1] showed that the circulation about each
individual vortex line is quantised, taking the value of Γ = 9.97 × 10−4 cm2 /sec.
The most generally accepted equations for modelling the macroscopic flow of
helium II are the Hall-Vinen-Bekharevich-Khalatnikov (HVBK) equations which
were derived by a number of people over the years [2,3,4,5,6]. These equations
extend Landau’s two-fluid model to take into account the presence of quantized
vortex lines in the flow. The derivation of the equations is based on a continuum
approximation, assuming a high density of vortex lines, all aligned roughly in
the same direction.
The incompressible isothermal HVBK equations of motion of the two fluids
are

∂vn ρs
+ (vn ·∇)vn = −∇pn + νn ∇2 vn + F , (1a)
∂t ρ
∂vs ρ n
+ (vs ·∇)vs = −∇ps − νs T − F, (1b)
∂t ρ
∇· vn = 0, ∇· vs = 0. (1c)

where vn and vs are the velocity profiles of the normal fluid and superfluid
respectively, ρn and ρs the normal fluid and superfluid densities, ρ = ρn + ρs
helium’s total density, pn and ps effective pressures and νn the normal fluid

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 138–145, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Using the HVBK Model to Investigate the Couette Flow of Helium II 139

kinematic viscosity. The relative amount of normal fluid and superfluid present
in the flow depends on the temperature T of the liquid: if T → Tλ then ρs /ρ → 0
and ρn /ρ → 1; if T → 0 then ρs /ρ → 1 and ρn /ρ → 0.
The mutual friction force, F , describes the interaction between the normal
fluid and the vortex lines and is given by

F = 12 B ω
 × (ωs × (vn − vs − νs ∇× ω
 )) + 12 B  ωs × (vn − vs − νs ∇× ω
s s s
 ), (2)

where ωs represents the superfluid vorticity and is a measure of the number

and direction of vortex lines contained in a given small region of fluid. The term
ω s = ωs /|ωs | represents the unit vector in the direction of superfluid vorticity.
B and B  are temperature dependent coefficients which describe the interaction
between the normal fluid and the vortices [7,8].
The vortex tension force, −νs T , describes the energy in the vortex lines and
is such that
T = ωs × (∇× ω  s ), (3)
where ν s = (Γ/4π) log(b0 /a0 ) is the vortex tension parameter, a0 is the vortex
core radius and b0 = (|ωs |/Γ )−1/2 is the intervortex spacing.
In this paper we review the application of the HVBK equations to Taylor-
Couette flow, that is flow between two concentric rotating cylinders. Taylor-
Couette flow has been used as a bench-mark for fluid mechanics since Taylor’s [9]
pioneering work to investigate the transition from Couette flow to Taylor vor-
tices, which established a firm ground for using the Navier-Stokes equations and
the no-slip boundary conditions. Progress in helium II has been slower than for
classical fluids due in part to problems of flow visualization at such low temper-
atures. In considering a classical fluid, introduction of flakes or other small par-
ticles into the working fluid (usually oil or water), results in the Taylor vortices
being clearly evident. In contrast there are only limited visualisation techniques
available to the experimentalist at temperatures close to absolute zero. Recent
attempts have been made to reveal the flow pattern of helium II by adding small
particles [10]. However this was only successful at high rotation rates (40 times
the critical angular velocity at which linear stability analysis predicts Couette
flow becomes unstable).
Experiments on helium II between concentric cylinders were first performed
by Kapitza [11] in 1941 and Donnelly & LaMar [12] have written a review of
experiments involving helium II in Couette apparatus. We shall expand on two
types of experiments performed, which have been used to compare theoretical
predictions with. Early Taylor-Couette experiments were concerned with deter-
mining the viscosity of helium II by measuring the torque exerted by the flow
on the stationary cylinder. A break in the linear dependence of the torque with
the angular velocity of the rotating cylinder is taken to denote a transition from
one solution to another. The second experimental technique, that of measuring
the extra attenuation of a second sound wave, can be used to probe the super-
fluid vorticity. Second sound waves occur when there is a periodic counterflow
between the normal fluid and superfluid, which corresponds to a wave of heat.
Angular velocity is plotted against the attenuation factor and breaks in the curve
140 K.L. Henderson

are interpreted as transitions in the flow. By measuring the extra attenuation of

second sound waves in the axial, azimuthal and radial directions it is theoreti-
cally possible to get an idea of the number and direction of the quantized vortex
lines. In practice the information obtained is less complete than this.
In this paper we restrict our attention to rotation of the inner cylinder only,
keeping the outer cylinder fixed. The usual simplifying assumption of infinite
cylinders is adopted in Sects. 2 and 3.1 where we consider the stability of he-
lium II Couette flow and the nonlinear flow of helium II beyond this transition
respectively. In Sect. 3.2 we consider the basic flow of helium II between finite
cylinders with stationary endcaps.

2 Linear Theory

In the case of flow between infinite cylinders with inner radius R1 rotating with
angular velocity Ω and stationary outer radius R2 , Couette flow, whose velocity
profile is given by:
v c = (A + C/r) er (4)
is an exact solution of the HVBK equations for both the normal fluid and su-
perfluid. This is provided that Ω is greater than a small critical value at which
vortex lines first appear in the gap. A and C are constants depending on R1 ,
R2 and Ω determined by the no-slip condition imposed on the normal fluid at
the cylinder walls. The superfluid, being inviscid is not required to satisfy such
boundary conditions. The only restriction is that there is no penetration through
the boundary. From (4) it can be seen that the superfluid vorticity is purely ax-
ial and has magnitude 2|A|, thus the vortex lines are aligned in the direction of
rotation.
Early attempts to theoretically examine the stability of Couette flow theo-
retically were made by Chandrasekhar & Donnelly [13], however the issue has
only recently been resolved fully by Barenghi & Jones [14] and Barenghi [15].
They performed a linear stability analysis on the HVBK equations. The Couette
state was linearly perturbed and they numerically calculated the critical angular
velocity, Ωc , and corresponding critical axial wavenumber at which Couette flow
becomes linearly unstable. We summarize their findings below:

• the axisymmetric mode onsets first,

• the stability of helium II Couette flow is more stable than classical Couette
flow in the high temperature region but less stable in the low temperature
one,
• the critical axial wavenumber decreases as the temperature decreases, becom-
ing zero at relatively high temperatures.

For a classical fluid the critical axial wavelength at which Couette flow becomes
unstable and Taylor vortices form occurs at kc ≈ 3.1, in other words the result-
ing Taylor vortices are approximately square. This is not the case for helium II.
As the temperature decreases below the lambda temperature of Tλ = 2.172k the
 Using the HVBK Model to Investigate the Couette Flow of Helium II 141

critical axial wavenumber decreases resulting in an elongation of the Taylor cells.

This effect is strongly temperature dependent and the critical axial wavenum-
ber becomes zero at a relatively high temperature for certain parameter ranges
(e.g. T = 2 k).
Compared to the classical case T = 2.172k, the stability of helium II is
initially enhanced as the temperature decreases below the transition tempera-
ture, due to the tension in the vortex lines. However the stability is dramatically
reduced as the temperature drops further. These results prompted further exper-
iments which were performed by Swanson & Donnelly [16]. They carried out a
series of second sound experiments at temperatures close to Tλ . Comparison be-
tween theory [15] and experiments [16] gave excellent agreement, particularly for
temperatures close to the lambda temperature. Although the qualitative picture
was correct at lower temperatures it is not realistic to expect such good quanti-
tative agreement here, due to the breakdown of the infinite cylinder assumption.
For T < 2.07 k linear theory predicts that the critical axial wave number is zero.
In such a region, the Taylor cells would be so elongated that only a few would be
present in the apparatus and end effects would undoubtedly become important.
It was only at this stage that the validity of the HVBK equations was con-
firmed. This led the way for further research in this area, namely numerically
solving the nonlinear two-fluid HVBK equations of motion of helium II.

3 Nonlinear Solutions
3.1 Infinite Cylinder Assumption
Using the infinite cylinder assumption, Henderson, Barenghi & Jones [17] nu-
merically solved the HVBK equations for the first time to obtain the nonlinear
flow of helium II between infinite cylinders. The aim of the work was to ob-
tain solutions for helium II corresponding to what would correspond to Taylor
vortices in a classical flow and to investigate what happens to the vortex lines.
Axisymmetric solutions were considered since linear theory predicts that the ax-
isymmetric mode onsets first. To solve equations (1a-1c), boundary conditions
are required. The nonlinear problem is 6th order in both the normal fluid and
superfluid, however the linear problem is only 2nd order in the superfluid. Thus
two further boundary conditions for the superfluid are needed in addition to the
no penetration of the boundary used successfully in the linear stability analysis.
The extra boundary conditions employed were

ωφs = 0 at r = R1 and R2 (5a)

vφs = Ωr at r = R1 and R2 . (5b)

Equations (5a,5b) force the superfluid vorticity to be purely axial at the cylinder
walls. This is consistent with Couette flow, in which the vortex lines are purely
axial throughout the flow and results in the mutual friction being small at the
boundaries which is an advantage numerically. The normal fluid satisfies the
standard no slip boundary conditions as for the linear model.
142 K.L. Henderson

The HVBK equations were solved numerically using a pseudospectral method,

based on expansions in Chebychev polynomials in the radial direction and trigono-
metric functions in the axial direction [18]. Results were obtained for angular
velocities of up to 15% above the critical angular velocity at which linear theory
predicts that Couette flow becomes unstable. Apart from the axial stretching of
the Taylor cells, the normal fluid displays a velocity profile similar to that of a
classical fluid, However the superfluid velocity profile is markedly different to the
classical case; instead of a meridional flow consisting of single pair of cells in each
period, we find a more complex pattern of eddies and counter-eddies. Perhaps
what is of most interest is the orientation of the vortex lines. The numerical
results show that the superfluid vorticity is still predominantly axial and the
deflection in the azimuthal direction is smaller than that in the radial direction.
Considering the (r, z) plane the vortex lines are most densely situated near the
inner cylinder at positions of maximum inflow and are deflected towards the
outer cylinder at the centre of the cell, where there is outflow.
There are two ways of comparing the numerical results with experiments.
Firstly by comparing the additional attenuation of a second sound wave due to
the vortex lines and secondly by measuring the torque exerted on the outer cylin-
der. We compared the relative change in the azimuthal attenuation coefficient
at an angular velocity of 5.4% above the onset of Taylor vortices and found an
order of magnitude correspondence between the experimental value of Swanson
& Donnelly [16]. Our predicted value was lower than the observed value which
could be due to the following points:
• The observed attenuation will depend on the spatial structure of the mode
used to probe the flow.
• The observed attenuation does not discriminate the sense in which the vor-
tex lines point, however the HVBK equations are derived using an averaged
approach to the vorticity of the superfluid.
Although many experiments have been carried out to measure the torque in
helium II, few experiments have been carried out in a parameter range such that
the stability curve has a minimum at non-zero axial wavenumber. We compare
our torque measurements with those of the experiment of Donnelly [19] in which
values of the torque above the transition are reported. At small angular velocity,
in the Couette flow regime, the torque is proportional to the viscosity μ of the
fluid. At higher angular velocities there is a break in the curve, corresponding to
the onset of Taylor vortices, the torque exerted on the outer cylinder increases
as the axisymmetric taylor vortices are more efficient in transferring angular mo-
mentum than the azimuthal Couette motion. Excellent agreement was found [20]
between the calculation and the experimental data in the nonlinear regime for
Ω > Ωc .
These two results validated the HVBK equations for the first time in the
nonlinear regime. Although comparisons with available experiments in the non-
linear regime are encouraging, it is clear that end effects become important, even
at relatively high temperatures.
 Using the HVBK Model to Investigate the Couette Flow of Helium II 143

3.2 Unit Aspect Ratio

In order to be able to compare with further experimental data, end effects need
to be included in the model. Henderson & Barenghi [21] considered helium II
contained within a cylindrical annulus of inner radius R1 , outer radius R2 , height
H where the gap between the cylinders has been chosen such that H = R2 −
R1 . Thus the Couette annulus has unit aspect ratio, in that the gap between
the cylinders is equal to the height between the endcaps. The inner cylinder
rotates with constant angular velocity Ω, whilst the outer cylinder and two end
plates are stationary. This simple flow configuration enabled us to study how
the vortex lines respond to a shear in the presence of boundaries which are both
parallel and perpendicular to the natural axial direction of the vortex lines. The
axisymmetric form of the HVBK equations (1a-1c) were solved using a finite
difference approach taking a regular grid in both the r and z direction. The
boundary conditions on the curved cylinder walls were taken to be the same as
for the infinite cylinder case. However extra boundary conditions are also needed
on the two endcaps, z = 0, H. For the normal fluid standard no slip boundary
conditions were imposed. Whilst for the superfluid the following were used

vzs (r, 0) = vzs (r, H) = 0, (6a)

ωrs (r, 0) = ωrs (r, H) = 0, (6b)
ωφs (r, 0) = ωφs (r, H) = 0. (6c)

The first condition (6a) ensures that there is no penetration of the superfluid
through the boundary, whilst the last two conditions (6b,6c) correspond to per-
fect sliding of the vortex lines as discussed by Khalatnikov [5].
The main result of this investigation is the anomalous motion of helium II
when compared to the motion of a classical fluid. The velocity profile obtained
is a superposition of an azimuthal motion vφ around the inner cylinder and a
toroidal motion vr and vz in the vertical plane. The latter motion is in the form
of a pair of cells similar to a Taylor vortex pair, but being caused by boundaries
rather than a centrifugal instability, it is hereafter referred to an Ekman cell
pair. The first interesting finding is that vφs is almost z-independent, that is
the superfluid moves around the cylinders in a column-like fashion, which is
due to the tension in the vortex lines. This effect becomes more pronounced at
lower temperatures when the superfluid component is higher as is illustrated in
Fig. 1a,b. Each figure extends over the whole computational domain with the
inner cylinder and outer cylinder on the left and right respectively. In contrast,
vφn exhibits strong z-dependence due to the no-slip boundary conditions imposed
on the normal fluid at the ends and walls of the cylinders and has a similar profile
to that of a classical fluid, see Fig. 1c.
144 K.L. Henderson

Fig. 1. Azimuthal motion vφ of (a) the superfluid at T =1.8 K, (b) the superfluid at
T =2.17 K, (c) the normal fluid at T =2.17 K. Lighter regions correspond to larger
magnitude.

Fig. 2. Motion of helium II compared to a classical fluid. Classical: (a) radial veloc-
ity vr ; (b) axial velocity vz . Helium II at T =2.11 K: (c) vrn ; (d) vzn ; (e)vrs ; (f) vzs .
Lighter/darker regions correspond to positive/negative velocities.

The second interesting finding comes from looking at the Ekman cells in both
the normal fluid and superfluid. In a classical fluid the two Ekman cells form with
outflow at the centre and inflow at the ends of the cylinder, as in Fig. 2a,b. The
results for helium II are quite different. We find that the superfluid Ekman cells
 Using the HVBK Model to Investigate the Couette Flow of Helium II 145

always rotate in a counter-classical way due to the mutual friction force, that is
outflow occurs at the ends of the cylinder with inflow at the centre, see Fig. 2c,d.
It is also seen that the normal fluid Ekman cells rotate in a counter-classical way
at lower temperatures, see Fig. 2e,f, but revert to a classical direction close to the
transition temperature Tλ = 2.172k as one would expect. We also investigated
the magnitude and direction of the superfluid vorticity in order to gain a picture
of how the vortex lines are situated in the flow. As for the infinite cylinder case,
the superfluid vorticity is primarily axial with small deflections in both the r
and φ direction. However the vorticity is concentrated near the inner rotating
cylinder, which is a measurable result.

4 Discussion
The excellent agreement between the linear stability analysis [15] and exper-
imental data [16] was a rigorous test of the validity of the HVBK equations
at least in the linear regime. The HVBK model has been validated further by
the good agreement between the nonlinear calculation [18] and experimental
data [19]. Obtaining nonlinear solutions for the flow of helium II between infi-
nite and finite cylinders has allowed us to gain more insight into the flow which,
because of the low temperature environment, cannot be observed directly like
a classical fluid. It has also enabled us to explore the boundary conditions for
the superfluid. A possible future direction of the work would be to extend the
aspect ratio in order to investigate the transition to Taylor cells.

References
1. R.P. Feynman: ‘Application of quantum mechanics to liquid helium.’ In P rogress
in Low Temperature Physics 1. (C.J. Gorter, North Holland 1955)
2. H.E. Hall, W.F. Vinen: Proc. Roy. Soc. London A 238, 215 (1956)
3. H.E. Hall: Phil. Mag. Suppl. 9, 89 (1960)
4. I.L. Bekharevich and I.M. Khalatnikov: Sov. Phys. JETP 13 , 643, (1961)
5. I.M. Khalatnikov: An Introduction to the Theory Superfluidity. (Benjamin 1965)
6. R.N. Hills, P.H. Roberts: Arch. Rat. Mech. Anal. 66, 43 (1977)
7. C.F. Barenghi, R.J. Donnelly, W.F. Vinen: J. Low Temp. Phys. 52, 189 (1983)
8. D.C. Samuels & R.J. Donnelly: Phys. Rev. Lett. 65, 187 (1990)
9. G.I. Taylor: Phil. Trans. Roy. Soc. Lond. A 223, 289 (1923)
10. F. Bielert, G. Stamm: Cryogenics 33, 938 (1993)
11. P.L. Kapitza: J. Phys. USSR 4, 181 (1941)
12. R.J. Donnelly, M.M. Lamar: J. Fluid Mech. 186, 163 (1988)
13. S. Chandrasekhar, R.J. Donnelly: Proc. Roy. Soc. London A 241, 9 (1957)
14. C.F. Barenghi, C.A. Jones: J. Fluid Mech. 197, 551 (1988)
15. C.F. Barenghi: Phys. Rev. B 45, 2290 (1992)
16. C.J. Swanson, R.J. Donnelly: Phys. Rev. Lett. 67, 1578 (1991)
17. K.L Henderson, C.F. Barenghi, C.A. Jones: J. Fluid Mech. 283, 329 (1995)
18. K.L. Henderson, C.F. Barenghi, J. Low Temp. Phys. 98, 351 (1995)
19. R.J. Donnelly: Phys. Rev. Lett. 3, 507 (1959)
20. K.L. Henderson, C.F. Barenghi: Phys. Lett. A 191, 438 (1994)
21. K.L. Henderson and C.F. Barenghi: J. Fluid Mech. 406, 199 (2000)
An Introduction
to the Theory of Superfluid Turbulence

W.F. Vinen

School of Physics and Astronomy, University of Birmingham, Birmingham B15 2TT,

UK, and
Cryogenic Helium Turbulence Laboratory, Department of Physics, University of
Oregon, Eugene, Oregon 97403, USA

1 Introduction
In one sense superfluid turbulence is an old subject: it was mentioned as a theo-
retical possibility by Feynman in 1955[1]; and it has been known experimentally
since the early 1950s that flow of the superfluid component of helium II can
become turbulent when there is a steady counterflow of the two fluids, such as
occurs in a steady heat current[2]. The original experimental discovery was ac-
companied by the beginnings of a theory[3], and this theory has been developed
steadily, especially by Schwarz[4,5], so that many aspects of this type of turbu-
lence are now well understood. However, counterflow turbulence has no classical
analogue, and it has attracted little interest from those who study classical fluid
mechanics. Types of flow for which classical analogues do exist were observed
by low temperature physicists for many years, but the presence of the two fluids
were thought to make them very complicated, and they were not therefore stud-
ied in detail. More recently experiments have been reported on the analogue of a
rather simple case of classical turbulence, namely that produced by steady flow
through a grid[6,7]. In the classical analogue the turbulence is approximately ho-
mogeneous and isotropic, and its study has been important in the development
an understanding of classical turbulence[8]. The superfluid analogue promises to
be equally important.
In this paper I shall first describe some aspects of the theory of counterflow
turbulence. But I shall then devote most of the paper to grid turbulence, where
the theory is less well developed, although I shall make use of an important
experimental result obtained with a more complicated type of flow generated by
two counter-rotating discs[9]. My aim is to stimulate interest in the theory of
superfluid turbulence, particularly, at this stage, in the simple case of grid tur-
bulence, among both low temperature physicists and those with a background in
classical fluid mechanics. I shall focus on open questions and unsolved problems,
questions and problems that are clearly seen in grid turbulence, but which are
more widely relevant. My own background is in experimental quantum fluids,
and certainly not in theoretical fluid mechanics. I shall tend to speculate about
what I see as the physics of superfluid turbulence, and others will tell me where
my physical intuition is unreliable or, hopefully, where it can be developed along
more rigorous lines.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 149–161, 2001.

c Springer-Verlag Berlin Heidelberg 2001
150 W.F. Vinen

Superfluid grid turbulence illustrates in perhaps its simplest form what is

sometimes known as “vortex coupled superfluidity”; a turbulent state in which
the superfluid and normal components seem to be coupled together and behave
like a single classical fluid[10]. It has gradually become apparent that this type
of superfluid turbulence is quite common in cases where the two fluids are not
forced to move with different velocities. The ideas that we develop here in the
context of grid turbulence are likely therefore to be widely applicable.
A turbulent superfluid velocity field must take the form of a tangled array of
quantized vortex lines (circulation κ = h/m4 ). Motion of the lines is governed
by the classical Magnus effect. A force of mutual friction, f per unit length of
line, acting effectively on the core of the line normal to its length, arises from
the scattering of thermal excitations[11]. We shall assume that f = γ (vn − vL ),
where vn is the component of the velocity of the normal fluid perpendicular to
the length of the vortex and vL is the velocity of the line; for simplicity we ignore
any transverse component of f . Numerical values of γ are given in reference[12].
A number of different approaches have been used to develop our understand-
ing of superfluid turbulence. One relies on simple physical arguments backed
by dimensional analysis and ideas of dynamical similarity; another relies on
computer simulations. Perhaps there will in future be one that is based on rig-
orously established general principles; the fact that the turbulent velocity field
in the superfluid component is due entirely to discrete quantized vortex lines
may facilitate such a development (see the contribution of Gary Williams to
this workshop). The simulations are of two types: those pioneered by Schwarz[4]
and based on the idea that, except on very short length scales, quantized vor-
tices can be regarded as vortex filaments moving according to classical fluid
mechanics, with inclusion of the force of mutual friction f , which modifies the
motion through the Magnus effect; and those based on the Gross-Pitaevskii or
non-linear Schrodinger equation (NLSE), which includes a quantum description
of the vortex core, of the thermal excitations, and of their interaction, albeit one
that cannot be expected to be quantitatively applicable to liquid helium. For
many purposes the classical vortex filament approach is probably sufficient, but
it cannot give a fully satisfactory description of reconnections, which play an
important role in superfluid turbulence.
We shall confine our discussion to turbulence in superfluid 4 He; the case for
experiments on turbulence in superfluid 3 He is mentioned briefly in reference[13].
Experiments in which superfluid turbulence has probably been observed in 3 He
are reported by Fisher at this Workshop.

2 Counterflow Turbulence
In counterflow turbulence the vortex tangle is believed to be at least approxi-
mately homogeneous, provided that the average velocities of the two fluids, Vn
and Vs , are spatially uniform. The turbulence is maintained by the mutual fric-
tion. If the vortex lines move on average with the superfluid, the total average
force of mutual friction per unit volume, Fns , is equal to γL (Vn − Vs ), where L
 The Theory of Superfluid Turbulence 151

is the length of line per unit volume, and where we have ignored factors of order
unity arising from the random orientation of the lines. We can illustrate one
approach to superfluid turbulence by deriving the dependence of L on (Vn − Vs )
from a principle of dynamical similarity[3,14,15].
Let us assume that the vortex tangle is characterized by a single length,
 = L−1/2 , characteristic of both the vortex line spacing and the vortex radius
of curvature. Taking into account the Magnus effect and the force f , we can
easily show that the velocity with which any element of line moves is given by

vL = Vs + vs + ακ̂ × (Vn − Vs − vs ) , (1)

where vs is the superfluid velocity at the element due to the rest of tangle,
κ̂ is the unit vector parallel to the element, and α ≈ γ/ρs κ. There are two
contributions to the magnitude of vs : vs1 ≈ κ/2π, which is due to neighbouring
lines at the distance  from the element; and vs2 ≈ (κ/4π) ln (/ξ0 ), which is
due to the local curvature of the line. (We note in passing that computations
based on the“localized induction approximation”take into account only the latter
contribution.)
Suppose that we change the length scale in the vortex tangle by a factor g, so
that  → g, and let us ignore the logarithmic factor in vs2 . Then vs changes by
the factor g −1 . If we change Vs and Vn by the same factor, we see from equation
(1) that vL is also changed by the same factor (formally, we change the length
scale by g and the time scale by g 2 ). Therefore the whole tangle evolves in the
same way as it would before scaling, except for the scaling factor g, suggesting
that there is a principle of dynamical scaling. Application of this principle shows
−1
easily that  is proportional to (Vs − Vn ) , so that the mutual friction per
3
unit volume is proportional to (Vs − Vn ) , as is observed to be approximately
true. Taking into account the logarithmic term in vs2 introduces logarithmic
corrections, which are indeed probably observed.
As we have mentioned, the superfluid turbulence is maintained by the mutual
friction acting on individual elements of line. That the length of line can in
principle increase as a result of the mutual friction is clear from the fact[3]
that a vortex ring can grow if the self induced velocity of the ring is in the
direction of, but less in magnitude than, (Vn − Vs ); otherwise it will shrink.
Appropriately oriented parts of the tangle with low curvature can behave in a
similar way, although it is not obvious that a steady turbulent state (a finite
L) can be maintained. A detailed understanding of counterflow turbulence came
only from the simulations of Schwarz[4,5], which showed that a steady turbulent
state can be achieved through the effect of reconnections, which generate vortex
configurations that favour the growth of line.
This theory of counterflow turbulence is based on the assumption that the
flow of the normal fluid is laminar. The turbulent flow of the superfluid must
then occur on length scales not significantly larger than ; flow on larger length
scales would be damped out by mutual friction. Recently it has been suggested by
Melotte and Barenghi[16] that the laminar flow of the normal fluid in counterflow
may not always be stable, so that both fluids may become turbulent, probably
152 W.F. Vinen

on length scales significantly larger than . The theory for such a regime presents
us with a major challenge, which we mention again in Sect. 3.5.

3 Grid Turbulence in Superfluid Helium

Experiments on grid turbulence in superfluid helium at temperatures above 1K
have been described by Niemela at this workshop; the original measurements
were reported in reference[7]. I shall first remind you of the experimental results
and of the idea that they are consistent with a quasi-classical model, as discussed
by Skrbek at this workshop. Then I shall discuss why this model may work.
Finally, I shall extend the discussion to lower temperatures, where there are so
far no really satisfactory experimental results, but where new and potentially
interesting problems arise.

3.1 Measurements of the Decay of Vortex Lines,

and the Quasi-classical Model

In the experiments a grid is towed through the helium, and a measurement is

made of the time dependence of the excess attenuation of second sound in a small
fixed region in the helium behind the grid. The excess attenuation is caused by
mutual friction associated with vortex lines in the turbulent superfluid behind
the grid, and the measurements lead to data showing the decay of the line density,
L, with time. The grid moves through the helium at a velocity exceeding both
that required to create vortex lines (the mechanism need not concern us), and
that required to create turbulence in the normal fluid. Turbulence is therefore
produced in both fluids. The average velocity of each fluid must vanish (there
is nothing to maintain a counterflow, and any transient counterflow would be
damped rapidly by mutual friction), so the situation is quite different from that
discussed in Sect. 2.
As explained by Skrbek, the observed decay in L is consistent with a quasi-
classical model, based on the following two assumptions. First, on length scales
larger than the vortex line spacing, , which turns out to be close to the length
scale at which viscous dissipation is expected to occur in the turbulent normal
fluid, the two fluids are coupled together in the sense that the two velocity fields
are the same; the coupled fluids behave like a single classical fluid, the turbu-
lence exhibiting an inertial range of wavenumbers in which there is negligible
dissipation and in which the energy spectrum has the classical Kolmogorov form

E(k) = C2/3 k −5/3 , (2)

where C is a constant of order unity, and  is the rate of energy dissipation
per unit mass of helium at a high wavenumber, presumably of order or greater
than −1 . (Roughly speaking, energy is injected from the grid into eddies with
size of order the mesh size (wavenumbers of order the reciprocal of the mesh
size); non-linear terms in the equation of motion cause the energy to flow to
 The Theory of Superfluid Turbulence 153

smaller eddies (higher wavenumbers) in a cascade, until it can be dissipated by

viscosity.) Secondly, this rate of dissipation is given by
 = ν  κ2 L2 , (3)
where ν  has the dimensions of kinematic viscosity. If we interpret κ2 L2 as an ef-
fective mean square vorticity in the superfluid, which is of doubtful validity[17,13],
equation (3) is reminiscent of the rate of viscous dissipation in a turbulent clas-
sical fluid. Of course, as in a classical fluid, the Kolmogorov spectrum may rep-
resent an oversimplification, which fails to take account of, for example, inter-
mittency (see, for example, reference[18]); but it seems reasonable to ignore this
point at the present stage in the development of our understanding of superfluid
turbulence.
It is important to appreciate that we can say only that the experimental data
on grid turbulence are consistent with this quasi-classical model. Other models
might also be consistent: for example, one in which both the dissipation and
the energy spectrum take different forms (the energy spectrum could involve the
quantum of circulation). But the quasi-classical model is perhaps the simplest
model that will account for the experiments, and, as we shall argue, it can be
given some theoretical backing and is consistent with other experiments.
There is clearly a need to find direct experimental evidence for the valid-
ity of the Kolmogorov spectrum (2) in superfluid grid turbulence; i.e. a direct
measurement of the turbulent energy spectrum. We know of no such evidence
for grid turbulence, but evidence does exist for a more complicated type of flow
generated by two counter-rotating discs. This is from the important work of
Maurer and Tabeling[9], who measured pressure fluctuations in this type of flow
(on a rather course length scale), both above and below the superfluid phase
transition. They find that over a certain range of frequencies these fluctuations
do have a Kolmogorov spectrum, the spectrum being the same above and be-
low the transition. It seems likely therefore that the value of the Kolmogorov
constant, C, is the same above and below the phase transition and equal to the
value expected for a classical fluid. We assume that a similar result will hold for
grid turbulence. In that case the quasi-classical model of grid turbulence would
account unambiguously for the experimental results.

3.2 Superfluid Turbulence on Length Scales

Larger than the Vortex Line Spacing
Consider flow of the superfluid component on length scales large compared with
, and suppose that this flow can be achieved by a relatively small re-arrangement
of the vortex tangle, as turns out to be the case in grid turbulence[17]. Such a flow
can be characterized by a velocity field that is similar to that found in a classical
fluid, and we suggest that the dynamical behaviour of the superfluid is then sim-
ilar to that of a classical fluid with the same density at high Reynolds number.
An example, not involving turbulent flow, where the truth of this suggestion can
be verified is provided by wave motion in an otherwise uniformly rotating super-
fluid, which contains a uniform array of lines. Waves with wavevectors directed
154 W.F. Vinen

parallel to the axis of rotation have a character that depends on the magnitude
(q) of the wavevector relative to the line spacing . If q  1 the waves are Kelvin
waves on the individual vortices; if q  1 the waves become indistinguishable
from the classical “inertia waves”found in a classical rotating liquid (see, for
example, references[19,20]. In the case of turbulent flow we must remember that
the non-linear term in the Navier-Stokes equation couples motion on different
length scales, so that the validity of our suggestion depends on the hypothesis
of the “independence of Fourier components for distant wavevectors”[8]. But we
know of no formal proof of this validity.
We emphasize that this similarity between superfluid and classical flow can-
not extend to wavenumbers of order or greater than −1 , where the discrete
nature of the vorticity cannot be ignored.

3.3 The Turbulent Energy Spectra in Superfluid Grid Turbulence

We can now start to see some theoretical justification for the quasi-classical
model. We see that when the superfluid component flows with sufficient speed
through a grid it could lead to turbulence in that component, characterized by
the Kolmogorov spectrum for k  −1 , as in a classical fluid. When both fluids
flow through the grid two such turbulent flows could be generated. However, the
presence of vortex lines in the superfluid component gives rise to mutual friction,
which must tend to couple the two fluids. It can be shown[17] that, provided the
normal fluid flow is not significantly affected by viscosity, and provided that
k  −1 , this friction is sufficient to ensure that the two fluids have associated
with them the same velocity fields. (The demonstration involves a proof that the
time required for mutual friction to eliminate relative motion in the two fluids on
a length scale k −1 is considerably less than the (“turnover”) time for a turbulent
eddy of size k −1 to lose its energy by inertial transfer to other eddies.) Given, as
we have seen, that viscous dissipation is expected to occur in the normal fluid
only for k ≥ −1 , we can begin to understand the success of the quasi-classical
model.
The type of coupled motion of the two fluids that we are discussing here
relates of course to the vortex coupled superfluidity that we mentioned in Sect. 1.
The basic idea of such coupling is presumably more widely applicable than to the
simple case of grid turbulence; examples are provided by the flow investigated by
Maurer and Tabeling[9], to which we have already referred, and to that round a
sphere investigated recently by Smith et al[21]. The theory advanced here relies
on very general arguments; support from the theory of specific types of flow,
albeit very idealised, has been provided by Barenghi and his colleagues[33,23].
We emphasize that on length scales comparable with or less than , where
the flow of the superfluid component is strongly constrained by the fact that
vorticity can be associated only with discrete quantized vortex lines, the two
velocity fields cannot be the same, even in the absence of viscous dissipation in
the normal fluid.
The quasi-classical model requires not only the coupled motion of the two
fluids for k  1 but also the validity of equation (3) for the total rate of
 The Theory of Superfluid Turbulence 155

dissipation of turbulent energy, both fluids contributing to this energy. It is far

from obvious that equation (3) is correct.
There is also the question of the value of ν  . It turns out[7] that the existing
experiments on grid turbulence yield only the ratio C 3 /ν  . There has been much
private discussion about the value of the Kolmogorov constant C that ought to
be used. In the case of turbulence in a classical fluid C seems to be universal
(within fairly large experimental error and at reasonably high Reynolds number)
and equal to about 1.6[24]. The suggestion has been made that in the case of
superfluid grid turbulence we are dealing with a quantum liquid, so that C might
take a different value, which is perhaps temperature dependent. However, if we
accept evidence from the experiments of Maurer and Tabeling[9], to which we
have already referred, then C has its classical, temperature-independent, value.
We conclude therefore, at least for the present, that it is sensible to use the
value C = 1.6 to deduce the value of ν  from experiment, as is done by Niemela
and Skrbek at this Workshop. We remark that this value has the same order
of magnitude as ηn /ρ, where ηn is the viscosity of the normal fluid and ρ is
the total density of the helium, but that its temperature dependence is quite
different. Further development of our theoretical discussion requires therefore
both a demonstration that equation (3) is at least reasonable and some discussion
of the observed value and temperature dependence of ν  . It turns out that this
discussion is best postponed until we have discussed superfluid turbulence at
very low temperatures.
It may be relevant to add that the classical Kolmogorov spectrum does not
depend for its validity on the Navier-Stokes equation; it depends only on the prin-
ciple that the statistical properties of the turbulence are determined uniquely
and universally by the length scale involved and the rate of energy dissipation at
high wavenumbers. But it does depend on the absence of quantum effects, which
would introduce Planck’s constant, probably in the form of the quantum of cir-
culation, κ, as a relevant parameter. If quantum effects were important, the spec-
trum of the turbulent energy might take the form E(k) = 2/3 k −5/3 H(κ−3 k −4 ),
allowed by dimensional analysis, where H is some unknown function. Although
this may not be ruled out by the experiments on superfluid grid turbulence, it
does seem to be inconsistent with the experiments of Maurer and Tabeling[9].

3.4 Superfluid Turbulence at Very Low Temperatures

The experiments on grid turbulence that we have discussed so far were carried
out at temperatures above 1K, where there is a significant fraction of normal
fluid. We focus next on lower temperatures, including those so low that the frac-
tion of normal fluid can be neglected. No experiments have yet been reported on
superfluid turbulence produced by steady flow through a grid at low tempera-
tures. Preliminary study of turbulence produced by an oscillating grid has been
reported by Davis et al[25], and the latest results will be reported at this Work-
shop by McClintock; but this type of turbulence is not homogeneous and may
be difficult to interpret at this stage. Two simulations relevant to low tempera-
tures have been reported. Tsubota et al[26] have studied the decay of turbulence
156 W.F. Vinen

by the Schwarz technique. The turbulence is produced initially by simulated

counterflow at a high temperature: the counterflow is then turned off; the tem-
perature is changed to the value required; and the decay is then simulated. Nore
et al[27] perform a simulation relevant to zero temperature based on a solution
of the NLSE, with a Taylor-Green vortex as an initial flow. Our own approach
to low-temperature grid turbulence[17], which is quite speculative, has been to
try to identify the essential physics, with the help of an interesting paper by
Svistunov[28], and with confirmation where possible from the simulations. See
also the paper by Lipniacki at this Workshop.
Consider first the case of zero temperature, when there is no normal fluid. We
guess that flow through the grid leads to a Kolmogorov spectrum for k  1, as
explained in Sect. 3.3. The simulations of Nore et al tend to support this guess.
Energy flows towards smaller length scales (larger wavenumbers) in a classical
Kolmogorov cascade until it reaches length scales of order  (wavenumbers of
order −1 ), at which, as we emphasized in Sects. 3.2 and 3.3, this classical cascade
can no longer exist. However, there is no mechanism for energy dissipation at this
length scale: there is no mutual friction; and oscillatory motion of the lines is at
such a relatively low frequency that the radiation of sound (phonon generation)
is quite ineffective. Energy must therefore flow into structures smaller than ,
although flow at such large wavevectors cannot be described by any extension
of the Kolmogorov spectrum. Such structures can form only if vortex lines come
close together in some parts of the turbulent field.
This close approach of vortex lines was tentatively foreseen in reference[3],
and it became very evident, even at higher temperatures, in the simulations of
Schwarz[4], where it can lead, according to Schwarz, to the reconnections that
are necessary to maintain counterflow turbulence (see Sect. 2). Whether or not
reconnections occur, the close approach will lead to the formation of kinks on the
lines (one form of small-scale structure). For reasons that we mention later, these
kinks are more prominent at low temperatures, and they are seen very clearly
in the simulations of Tsubota el al[26]. The evolution of the kinks must involve
strongly non-linear processes, but crudely we can decompose the kinks into their
Fourier components, which are harmonic Kelvin waves propagating along the
lines with different wavevectors. The non-linear effects will lead to the transfer
of energy to Kelvin waves of larger and larger wavenumber and frequency, until
the frequency is high enough for the efficient production of phonons. Perhaps
therefore the turbulence is characterized by two cascades: energy fed in at the
grid first flows down a conventional Kolmogorov cascade (k < 1) and then down
a Kelvin wave cascade (k > 1), until it is dissipated by phonon production.
Recent simulations by Araki et al[29] and by Kivotides (this Workshop) give
support for this view. However, the two types of cascade may be different in
that energy may be fed into the Kelvin wave cascade over a wide range of
wavenumbers, and not merely at the smallest relevant wavenumber.
Reconnections might lead to other processes by which turbulent energy is
lost. When Schwarz demonstrated the importance of reconnections in counter-
flow turbulence, he did not have a microscopic theory of a reconnection; his simu-
 The Theory of Superfluid Turbulence 157

lations were based on a model of vortex lines as classical filaments, and he simply
assumed that a reconnection would occur when two lines came sufficiently close
together[4,26]. Simulations based on the NLSE can provide a microscopic theory,
albeit subject to the limited extent to which the NLSE provides a good descrip-
tion of helium. Such simulations were performed by Koplik and Levine[30], and
new simulations are reported by Adams at this Workshop. The work of Kop-
lik and Levine suggested that reconnections can occur without energy loss; i.e.
without the production of thermal excitations in the form of phonons. However,
the more sensitive simulations by Adams show that phonon production proba-
bly does occur, and we guess that in real helium both phonons and rotons are
produced. (Phonons are produced in the simulations of Nore et al[27], but it is
not clear whether they are produced by vortex oscillations or reconnections. See
also[31].) But we note that an energy loss per reconnection of order ρκ2 ξ0 , a
not unreasonable guess and one that is consistent with the Adams simulations,
can be shown to have a negligible effect on the overall vortex decay rate. Recon-
nections can also lead to the production of small vortex rings (another form of
small-scale structure), as is clear from the simulations of Tsubota et al[26]. This
does not lead in itself to dissipation into thermal excitations, and in an infinite
system the small rings may well be re-absorbed in subsequent reconnections. But
in the case where the turbulence is confined to a finite region the rings might
escape with a consequent effective loss of energy.
In this section we have so far considered only the case of strictly zero tem-
perature. At a finite temperature there is some normal fluid, which leads to the
force of mutual friction, f , introduced in Section 1. (At low temperatures the
viscosity of the normal fluid is high and its density low, so turbulence in the
normal fluid is unlikely to occur.) The mutual friction will lead to additional
damping of the Kelvin waves and to shrinkage of any small rings. The additional
damping of the Kelvin waves increases as the temperature rises, and it proba-
bly exceeds that due to phonon emission at temperatures exceeding about 0.4K.
(This result is uncertain because it is based on what is really a guess at the rate of
phonon emission from a Kelvin wave[17,35]. Here is another problem that needs
rigorous solution.) At still higher temperatures, approaching or greater than 1K,
the damping of the Kelvin waves becomes so large that they can hardly exist.
Vortex motion on the length scale  is then itself strongly damped by mutual
friction, and energy flow into the Kelvin wave cascade is inhibited; the vortex
lines become less kinked. The effect of the mutual friction in causing small rings
to collapse needs further study. Indeed the role of vortex rings in general, which
has been emphasized by Tsubota et al[26], requires further study.
These ideas can be used to estimate the rate of dissipation of turbulent
energy, , at low temperatures[17,28]. It seems that the rate of energy loss is
likely to be dominated by the rate at which energy flows from motion at length
scales of order  to smaller length scales. Probably therefore it is dominated by
the time scale associated with vortex motion on the length scale , which is of
order 2 /κ. This idea leads via a scaling analysis to a rate of loss of vortex line
158 W.F. Vinen

given approximately by
dL κ
= −χ2 L2 , (4)
dt 2π
where χ2 is a constant of order unity. More accurately, χ2 is a weak (logarithmic)
function of L and of other parameters such as the Kelvin wave cut-off due to
phonon emission or mutual friction (this more accurate form arises from the need
to take account of the relatively small length of line associated with the Kelvin
waves and small rings that exist on length scales less than ). The corresponding
rate of energy loss is given approximately by

 = ν  κ2 L2 , (5)

where the parameter ν  is of order κ but weakly dependent on L and temperature[17].

We see that equation (5) has the same form, approximately, as equation (3), and,
as a result of a numerical accident, the magnitude of ν  is similar to the magni-
tude of ν  deduced from the towed grid experiments at high temperatures. The
dissipation in the turbulent helium may therefore in a sense be characterized
by an effective kinematic viscosity that is not very dependent on temperature,
even at low temperatures, in spite of the fact that the mechanism of energy
dissipation is different at different temperatures.
A rate of loss of vortex line given by equation (4) has been observed in the
simulations of Tsubota et al[26]. These simulations lack the spatial resolution
required to include processes occurring on length scales much less than , and
they use the assumption that small rings are removed from the turbulence by
some unknown mechanism. The corresponding values of χ2 are not therefore
very reliable, although an expected increase in χ2 with increasing temperature
due to an increasing mutual friction is confirmed. One lesson that emerges from
a comparison of the calculations underlying equation (4) with the simulations is
that  is probably rather insensitive to the details of the dissipation occurring
at wavenumbers greater than −1 .
We emphasize that the ideas about superfluid turbulence outlined in this sec-
tion are very speculative and certainly not based on rigorous analysis. There are
serious gaps in the argument, relating, for example, to the fate of small vortex
rings. There are no experimental results with which to compare our predictions,
except in so far as the experiments of Davis et al[25] do probably confirm that
energy loss occurs at low temperatures at a rate not very different from that at
high temperatures. There is an urgent need for more experiments at low temper-
atures, but, as discussed elsewhere at this Workshop, the technical difficulties
are formidable.

3.5 Dissipation at Higher Temperatures

We must now return to dissipation in grid turbulence at higher temperatures,

above 1K, where our aim must be to understand the form of equation (3) and the
observed magnitude of ν  . In contrast to the situation at the lower temperatures,
 The Theory of Superfluid Turbulence 159

both fluids are now turbulent, and the total energy dissipation must include con-
tributions from both fluids. (It has recently been emphasized by Idowu el al [32]
that viscous dissipation can occur in the normal fluid even if the normal fluid is
not turbulent, because as a vortex moves through the normal fluid it drags part
of that fluid with it[11]; however, the resulting viscous dissipation is normally
taken into account in the drag coefficient, γ, appearing in the mutual friction
force (Sect. 1). But the local drag, extending possibly over a distance of order ,
might affect any small-scale normal-fluid turbulence. ) Viscous dissipation will
occur in the turbulent normal fluid, and there will be dissipation due to mutual
friction, the loss of energy in the superfluid being caused, as at somewhat lower
temperatures, by the mutual friction. Estimates of the total rate of dissipation
are difficult to make because the dissipation is occurring on a length scale of or-
der , where the discrete vortex structure is important and the two velocity fields
cannot be the same. Furthermore, the normal fluid is turbulent. Any simulation,
for example, would require the self-consistent determination of the vortex mo-
tion and the motion of the normal fluid, taking into account the coupling due to
mutual friction. The tools required for the solution of this difficult type of prob-
lem are being developed[33,34], but they have not yet been applied to realistic
problems.
The situation may be more simple at temperatures below about 1.6K, but
above those at which the considerations of Sect. 3.4 apply. It can be argued[35],
on the basis of a consideration of the various characteristic times involved, that
viscous dissipation in the normal fluid then occurs at wavenumbers that are
slightly, but significantly, smaller than −1 , and that mutual friction has a rather
small effect at these smaller wavenumbers, in spite of the fact that the motion
in the two fluids has become decoupled (the arguments are similar to those used
in reference[17] in connection with the coupling between the two fluids). Energy
dissipation in the superfluid then takes place by mutual friction at wavenumbers
close to −1 in a normal fluid that is at rest at these wavenumbers. The resulting
dissipation from the superfluid could then be very similar to that computed in
the simulations of Tsubota et al[26]. The total dissipation can then be calculated
without difficulty. The result is consistent with the form of (3), and the calculated
value of ν  as a function of temperature is quite close to that obtained from the
experiments. But more work is required before (3) can be understood fully.

4 Summary and Conclusions

The arguments presented in this paper have been speculative, but they offer a
picture of superfluid turbulence, particularly that formed behind a moving grid,
that is attractive and probably consistent with the presently available experi-
mental results. But there remain unsolved problems, and the whole speculative
approach requires rigorous appraisal. The situation for temperatures below 1K
is especially unsatisfactory, because as yet there exist virtually no relevant ex-
perimental results. From a practical point of view it is interesting to note that
in turbulent flow superfluid helium seems to behave in many ways like a clas-
160 W.F. Vinen

sical fluid with a kinematic viscosity similar to that of helium I, even at very
low temperatures; therefore any hope that the “vanishing viscosity”of superfluid
helium might allow the study of flow at phenomenally high Reynolds numbers
seems misplaced.

Acknowledgements
I am very grateful to Steve Stalp, Ladislav Skrbek, Russell Donnelly, Joe Niemela,
Makoto Tsubota, Carlo Barenghi, David Samuels, Peter McClintock and Edouard
Sonin for allowing me to see their results (experimental or theoretical) before
publication and for invaluable discussions. My work has been supported in part
by NSF Grant DMR-9529609.

References
1. R. P. Feynman, in Progress in Low Temperature Physics, 1, ch.2 ed. C. J. Gorter
(North Holland Publishing Co., Amsterdam, 1955)
2. W. F. Vinen, Proc. Roy. Soc. A240, 114, 128 (1957)
3. W. F. Vinen, Proc. Roy. Soc. A242, 493 (1957)
4. K. W. Schwarz, Phys. Rev. B31, 5782 (1985)
5. K. W. Schwarz, Phys. Rev. B38, 2398 (1988)
6. M. R. Smith, R. J. Donnelly, N. Goldenfeld, and W. F. Vinen, Phys. Rev. Letters
71, 2583 (1993)
7. S. R. Stalp, L. Skrbek, and J. J. Donnelly, Phys. Rev. Letters 82, 4831 (1999)
8. G. K. Batchelor, The Theory of Homogeneous Turbulence (Cambridge University
Press, Cambridge, 1953)
9. J. Maurer and P. Tabeling, Europhys. Letters, 43, 29 (1998)
10. R. J. Donnelly, in High Reynolds Number Flows Using Liquid and Gaseous He-
lium(Springer-Verlag, New York, 1991)
11. H. E. Hall and W. F. Vinen, Proc. Roy. Soc. A238, 215 (1956)
12. C. F. Barenghi, R. J. Donnelly, and W. F. Vinen, J. Low Temp. Physics, 52, 189
(1983)
13. W. F. Vinen, J. Low Temp. Physics, 121, 367 (2000).
14. K. W. Schwarz, Phys. Rev. Letters, 49, 283 (1982)
15. C. E. Swanson and R. J. Donnelly, J. Low Temp. Physics, 61, 363 (1985)
16. D. J. Melotte and C. F. Barenghi, Phys. Rev. Letters 80, 4181 (1998)
17. W. F. Vinen, Phys. Rev. B61, 1410 (2000)
18. U. Frisch, Turbulence (Cambridge University Press, Cambridge, 1995)
19. H. E. Hall, Proc. Roy. Soc. A245, 546 (1958)
20. R. J. Donnelly, Quantized Vortices in Helium II (Cambridge University Press, Cam-
bridge, 1991)
21. M. R. Smith, D. K. Hilton, and S. W. Van Sciver, Phys. Fluids 11, 751 (1999)
22. C. F. Barenghi, D.C. Samuels, G. H. Bauer, and R. J. Donnelly, Phys. Fluids 9,
2631 (1997)
23. C. F. Barenghi and D. C. Samuels, Phys. Rev. B60, 1252 (1999)
24. K. R. Sreenivasan, Phys. Fluids, 7, 2778 (1995)
25. S. I. Davis, P. C. Hendry, and P. V. E. McClintock, Proceedings of the 22nd In-
ternational Conference on Low Temperature Physics, Espoo and Helsinki to be
published in Physica B
 The Theory of Superfluid Turbulence 161

26. M. Tsubota, T. Araki, and S. K. Nemirovskii, Phys. Rev. B62, 11751 (2000).
27. C. Nore, M. Abid, and M. E. Brachet, Phys. Rev. Letters, 78, 3896 (1993), Phys.
Fluids, 9, 2644 (1997)
28. B. V. Svistunov, Phys. Rev B52, 3647 (1995)
29. T. Araki and M. Tsubota, J. Low Temp. Physics, 121, 405 (2000)
30. J. Koplik and H. Levine, Phys. Rev. Letters, 71, 1375 (1997)
31. M. Tsubota, S. Ogawa, and Y. Hattori, J. Low Temp. Physics, 121, 435 (2000)
32. O. C. Idowu, A. Willis, C. F. Barenghi, and D, C, Samuels, Phys. Rev. B
33. C. F. Barenghi and D. C. Samuels, Phys. Rev. B60, 1252 (1999)
34. C. F. Barenghi, D. Kivotides, O. C. Idowu and D. C. Samuels, J. Low Temp.
Physics, 121, 377 (2000).
35. W. F. Vinen, to be published
Numerical Methods for Coupled Normal-Fluid
and Superfluid Flows in Helium II

Olusola C. Idowu, Demosthenes Kivotides,

Carlo F. Barenghi, and David C. Samuels

Dept. of Mathematics, Univ. of Newcastle, Newcastle upon Tyne, NE1 7RU, UK

Abstract. Helium II at temperatures below Tλ can be described as a superposition

of two interacting fluids. A normal-fluid which has non-zero viscosity, and an inviscid
superfluid in which vorticity is confined to quantised vortex filaments. These two fluids
are coupled together by the mutual friction force. To model the normal-fluid flow we
use the forced Navier–Stokes equation while the superfluid flow is modelled as motion
of quantised vortex lines. In this article we describe the numerical methods used for the
coupled motion of the two fluids. We also briefly discuss the detailed flow structures
observed from the two-dimensional (2-D) and the three-dimensional (3-D) simulations
of this problem.

1 Introduction
Until recently, simulations of helium II flows have mainly involved calculating the
motion of superfluid vortex filaments under the mutual friction forcing from an
imposed normal-fluid flow [1]. A few studies have considered the reverse problem,
the calculation of properties of the normal-fluid flow due to a given mutual
friction forcing from the superfluid [2]. The problem of calculating the fully
coupled motion of both the normal-fluid and the superfluid components of helium
II is now being investigated, and we present here our first results from these
simulations [3]. We present the two-dimensional calculation in some detail, since
the localised nature of the mutual friction force on the normal-fluid is unusual in
fluid dynamics. The results of the closely related three-dimensional calculations
are also described but investigations are still preliminary in three-dimensions due
to the high computational costs. We end with a short and speculative discussion
of the role of these coupled flows in the generation of complex normal-fluid flows
through the triggering of flow instabilities.

2 The Self-Consistent Equation of Motion

The normal-fluid is a classical fluid and it evolves with the forced Navier–Stokes
equation.
∂v n 1 1
+ v n · ∇v n = − ∇p + νn ∇2 v n + F , (1)
∂t ρ ρn
where p is the pressure, νn = μ/ρn is the normal-fluid kinematic viscosity, μ is the
viscosity and F is the mutual friction force. Also, Eq. (1) must be supplemented

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 162–176, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Coupled HeII Flows 163

with the incompressibility condition. The mutual friction force is related to the
drag force on a superfluid vortex line which is a force per unit length of the
vortex line. To obtain the mutual friction force, we divide the drag force by the
normal-fluid area containing the superfluid vortex line. The length scale over
which the mutual friction force acts is quite small and can be considered to be
approximately a delta function force, non-zero only along the one-dimensional
superfluid vortex lines embedded in the 3-D normal-fluid flow.
In the fully coupled calculation of the local normal-fluid flow and the su-
perfluid vortex motion the velocity of each superfluid vortex line is given as [6]

v l = h1 (V s + v s ) + h2 s × (v n − V s − v s ) + h3 v n , (2)
where
ρs κD0
h1 = , (3a)
D02 + D2
ρs κD
h2 = 2 , (3b)
D0 + D2
D2 − D0 Dt
h3 = , (3c)
D02 + D2
D0 = ρs κ − Dt , (3d)

are friction coefficients and s is a unit vector in the direction of the vortex
line. V s is the averaged superfluid velocity, v s is the self-induced velocity of the
vortex line and κ is the quantum of circulation. The coefficients D and Dt are
the microscopic mutual friction drag coefficients defined as
ρs κ
[α − (α2 + α )] ,
2
D= (4)
q

ρs κ 2
[α + α − α ] ,
2
Dt = (5)
q
where
q = 1 − 2(α + α ) + (2 + 1)(α2 + α ) ,
2
(6)
and  = ρs κ/E. E is a temperature dependent constant. The coefficients α and
α in Eq. (4) and Eq. (5) are defined from the Hall and Vinen coefficients B and
B  [5,6,7,8]. α = ρn B/2ρ, and α = ρn B  /2ρ. Note that the normal-fluid velocity
here is not the average field normal-fluid velocity V n but the spatially rapidly
varying field v n which is a solution of the modified Navier–Stokes equation,
Eq. (1).

3 Numerical Methods for 2-D Flows

The direct numerical simulation method described in this section calculates the
coupled motion of a 2-D normal-fluid flow with one or more superfluid vortex
164 O.C. Idowu et al.

points (i.e. cross sections through the superfluid vortex lines). This is a fully-
coupled, self-consistent two-fluid calculation, solving for both the normal-fluid
velocity v n (x, t) and the motion of the superfluid vortex filaments v l (x, t). Re-
stricting our flow to the two-dimensional case still captures the most essential
physics of the problem and allows us to use a reasonably high grid resolution.
We used a grid resolution of 1282 for the results presented in this section.
In this calculation, we assume that V s , the externally applied superfluid
velocity in Eq. (2), is a uniform and constant flow. For a single vortex filament,
the self-induced velocity v s = 0 (in 2-D). The unit vector s is the direction vector
of each superfluid vortex (either +ẑ or −ẑ in the two-dimensional case). Since we
include the local motion of the normal-fluid in response to the superfluid vortex,
we must use a mutual friction force based on this local normal-fluid velocity, not
the mutual friction force based on the averaged normal-fluid velocity [6].
We used the stream function formulation to model the normal-fluid flow in
2-D. The stream function ψn for the normal-fluid is defined as [11,12]
 
∂ψn ∂ψn
vn = ,− ,0 . (7)
∂y ∂x

This definition ensures that the incompressibility condition is satisfied. The

stream function ψn is determined by the Poisson equation in 2-D

∇2 ψn + ωn,z = 0 (8)

where ωn,z is the z component of normal-fluid vorticity.

The equation for the normal-fluid vorticity vector ω n is given as
 
∂ω n
ρn + (v n · ∇)ω n − (ω n · ∇)v n = μ∇2 ω n + ∇ × F (9)
∂t

In 2-D flow only the ẑ component of this equation is non-zero and the vortex
stretching term (ω n · ∇)v n disappears. The mutual friction force per unit area,
F, exerted on the normal-fluid by the superfluid vortex points can be written as
[3,6,13]
D+ Dt +
F= ŝ × [ŝ × (vn − v l )] + ŝ × (vn − v l ) (10)
A p A p

where A is the area of the normal-fluid over which the mutual friction force is
distributed and p is the number of vortex lines. The force F is a function of
position and is non-zero only in areas of size A (taken to be the computational
grid spacing) which contain superfluid vortex points.
The motion of the superfluid vortex line given by Eq. (2), is an ordinary
differential equation which can easily be solved numerically in 2-D by different
methods. For this problem we used the Euler time-stepping method.
 Coupled HeII Flows 165

3.1 The Normal-Fluid Flow in 2-D

The response of the normal-fluid on the motion of the superfluid vortex line
was calculated by solving equations (8) and (9). We solved the Poisson equation
(Eq. 8) using a standard numerical method discussed by Press et al (1983) [12].
To solve the Navier–Stokes equation (Eq. (9)) we used the finite difference
method with periodic boundary conditions. The 2-D Navier–Stokes equation in
vorticity form can be written in the stream function formulation as
    2   
∂ω n ∂ψn ∂ω n ∂ψn ∂ω n ∂ ωn ∂ 2 ωn ∂Fy ∂Fx
ρn + − =μ + + − .
∂t ∂y ∂x ∂x ∂y ∂x2 ∂y 2 ∂x ∂y
(11)
The above equation is for the z component since this is the only non-zero com-
ponent in the 2-D formulation. We can discretised the normal-fluid vorticity
equation using Eq. (11) to have,
 
ω t+1 (i, j) = ω t (i, j) + δt Δx ψ t (i, j)Δy ω t (i, j) − Δy ψ t (i, j)Δx ω t (i, j)
 
1 
+ δt νn (Δxx + Δyy ) ω t (i, j) + Δx Fyt (i, j) − Δy Fx(t
i, j) ,(12)
ρn

where the operators are defined as

1 , -
Δx f (i, j) = f (i + 1, j) − f (i − 1, j) (13)
2δx

1 , -
Δy f (i, j) = f (i, j + 1) − f (i, j − 1) (14)
2δy
1 , -
Δxx f (i, j) = 2 f (i + 1, j) − 2f (i, j) + f (i − 1, j) (15)
δx
1 , -
Δyy f (i, j) = 2 f (i, j + 1) − 2f (i, j) + f (i, j − 1) (16)
δy
The discretisation scheme here is first-order accurate in time and second-order
accurate in space. The normal-fluid vorticity obtained from Eq. (12) at time step
t is used to calculate the stream function using Eq. (8) at time step t + 1.
We applied the periodic boundary condition on both the normal-fluid vortic-
ity and the stream function defined as

ψ(i, j) = ψ(i + N, j) = ψ(i, j + M ) (17)

ω(i, j) = ω(i + N, j) = ω(i, j + M ) (18)

To match the periodicity in the normal-fluid we need to define a similar boundary
condition for the superfluid. To achieve this we create images vortices for each
superfluid vortex line in the computational box. For the results discussed in this
166 O.C. Idowu et al.

paper we used eight images for each vortex. The flow field produced by these
image vortices have very negligible effect on the formation of the normal-fluid
peak velocity.
Other boundary conditions such as the slip and the no-slip boundary condi-
tions on the velocity can be imposed on the coupled helium II flow. In this case
the method of flow computation will be different from the one described above.
For instance the normal-fluid velocity and the vorticity will have to be computed
differently at the boundary.

3.2 Delta Function Forcing on a Grid

The friction force exerted on the normal-fluid by the superfluid vortex line is a
delta function force at the position of the superfluid vortex line. As the superfluid
vortex line moves through the grid space the mutual friction force on the normal-
fluid also changes. To make the transition of the mutual friction force between
the grid spaces smooth, we weighted the mutual friction force over the four
neighbouring normal-fluid grid points (Fig. 1). To do this we chose a simple
normalised linear weighting

wi,j = (1 − δxp )(1 − δyp ), (19)

where δxp and δyp represents the distance of the superfluid vortex line at point
(xp , yp ) from the four corner grid points (Fig. 1). The sum of these four weight-
ings is one. With this method the mutual friction forces on the normal-fluid grid
always change continuously as the superfluid vortex line moves across the grid.
This method described is similar to the Cloud-in-Cell numerical method [11]
used in modelling the motion of point vortices in classical hydrodynamics. We
tested this model by changing the grid spacing and we saw no significant effect
of the grid spacing on the normal-fluid velocity generated by the mutual friction
forcing. Using this approach, the mutual friction force is then computed on the
normal-fluid grid using Eq. (10). The value obtained is then substituted into
Eq. (11) to obtain the normal-fluid vorticity.

3.3 Extrapolation of the Normal-Fluid Flow in the Neighbourhood

of the Superfluid Vortex Line

The normal-fluid flow field near a vortex line cannot be estimated from the
normal-fluid grid through grid interpolation because we expect v n to have a
sharp peak at the position (xp , yp ) of the superfluid vortex point due to the
mutual friction force. To estimated the normal-fluid flow field we extrapolate
the flow to the point (xp , yp ) using the twelve neighbouring normal-fluid grid
points shown in Fig. 1. A 2-D vector field f at point (xp , yp ) (Fig. 1) can be
extrapolated on four neighbouring grid points as
⏐
⏐
f (xp , yp )⏐ = f (xi , yj ) + (δxp )fx (xi , yj ) + (δyp )fy (xi , yj ) , (20)
(i,j)
 Coupled HeII Flows 167
i, j + 2 i + 1, j + 2

i − 1, j + 1 i, j + 1 i + 1, j + 1 i + 2, j + 1
wi,j+1 wi+1,j+1

xp , yp

δyp wi,j wi+1,j

i − 1, j i, j δxp i + 1, j i + 2, j

i, j − 1 i + 1, j − 1

Fig. 1. Finite difference grid for the normal-fluid showing the position of the super-
fluid vortex line xp , yp and the weighting on the neighbouring grid points used for
interpolation.

⏐
⏐
f (xp , yp )⏐ = f (xi , yj+1 ) + (δxp )fx (xi , yj+1 ) + (1 − δyp )fy (xi , yj+1 ) , (21)
(i,j+1)
⏐
⏐
f (xp , yp )⏐ = f (xi+1 , yj ) + (1 − δxp )fx (xi+1 , yj ) + (δyp )fy (xi+1 , yj ) , (22)
(i+1,j)
⏐
⏐
f (xp , yp )⏐ = f (xi+1 , yj+1 )+(1−δxp )fx (xi+1 , yj+1 )+(1−δyp )fy (xi+1 , yj+1 ) ,
(i+1,j+1)
(23)
where fx = ∂f /∂x and fy = ∂f /∂y. The derivatives at the normal-fluid grid
points are computed using points outside the grid area containing the superfluid
vortex line (Fig 1). The derivatives are defined as
   
fx (xi , yj ) [f (i, j) − f (i − 1, j)] /δx
= (24)
fy (xi , yj ) [f (i, j) − f (i, j − 1)] /δy
   
fx (xi , yj+1 ) [f (i, j + 1) − f (i − 1, j + 1)] /δx
= (25)
fy (xi , yj+1 ) [f (i, j + 2) − f (i, j + 1)] /δy
   
fx (xi+1 , yj ) [f (i + 2, j) − f (i + 1, j)] /δx
= (26)
fy (xi+1 , yj ) [f (i + 1, j) − f (i + 1, j − 1)] /δy
   
fx (xi+1 , yj+1 ) [f (i + 2, j + 1) − f (i + 1, j + 1)] /δx
= (27)
fy (xi+1 , yj+1 ) [f (i + 1, j + 2) − f (i + 1, j + 1)] /δy
The estimated normal-fluid flow at (xp , yp ) using Eqs. (20)–(23) and the weight-
ing function defined in Eq. (19) now becomes
⏐ ⏐
⏐ ⏐
f ∗ (xp , yp ) = wi,j ∗ f (xp , yp )⏐ + wi+1,j ∗ f (xp , yp )⏐ +
(i,j) (i+1,j)
⏐ ⏐
⏐ ⏐
wi,j+1 ∗ f (xp , yp )⏐ + wi+1,j+1 ∗ f (xp , yp )⏐ (28)
(i,j+1) (i+1,j+1)
168 O.C. Idowu et al.

The extrapolated function f ∗ (xp , yp ) could be the normal-fluid velocity, vorticity

or pressure (in a scalar form of the equation) at point (xp , yp ).

3.4 Numerical Stability and Time Stepping

With two simultaneous calculations for the normal-fluid and the superfluid, we
must choose a time step to satisfy both calculations. One limit on the time step
is that we must keep it below the limit set by the normal-fluid viscosity,

ρn δx2
Δtvisc ≤ . (29)
μ

This is a necessary stability criterion for finite difference schemes [11]. Also we
want the superfluid vortex points to move smoothly through the normal-fluid
computational grid, so that we sets another limit
δx
Δtsf ≤ , (30)
VL
where VL is the velocity magnitude of the superfluid vortex line. We set the time
step of the calculation to be 1/10 of the minimum of these two values.

4 Results in 2-D Flows

The normal-fluid flow vn was dragged by the superfluid via the mutual friction
force F. As a result of the drag, the normal-fluid forms a localised jet-like struc-
ture shown in Fig. 2a and c. A detailed description of this flow structure is given
in [3].
At low driving superfluid velocity V s recirculation flow can be seen around
the edges of the computational box (Fig. 2a). The position of the vortex lines
corresponds to the center of the normal-fluid jet. The spatial structure of the jet
shown in Fig. 2a shows that the jet velocity is sharply peaked toward the center
and the shape of the jet is slightly elongated in the direction of the jet velocity.
This shape is maintained at all temperatures. The measured length (≈ 0.3mm)
and width (≈ 0.1mm) of the jet shows very little temperature dependence. We
do not currently have any predictive theory for these length scales.
The peak velocity of the normal-fluid jet however shows a very strong tem-
perature dependence. It has a maximum at temperatures of ≈ 1.9 K and falls
rapidly to zero at 1.3 K and at the lambda transition temperature. The drag on
the normal-fluid by the superfluid is approximately 60% at 1.9 K
The normal-fluid energy En in the coupled flow is approximately proportional
to V 2s . This energy comes from the superfluid flow through mutual friction. The
normal-fluid flow structure described here is very dissipative and maintains a
steady flow only because it is constantly renewed by the energy gained from
the superfluid through mutual friction. The simulation further reinforced this
by showing that the normal-fluid jet moves along with the superfluid vortex line
 Coupled HeII Flows 169

(A ) (B )

|v n |
|v l |

(C ) (D )

y y

x x
Fig. 2. Velocity field in the normal-fluid due to the mutual friction forcing of superfluid
vortex lines. The calculation was made on a 1282 grid. The size of the computational
box is 1 mm. The temperature dependent parameters were taken at T = 1.9 K. (A)
Magnitude of jet formed by a single vortex line with driving velocity of VS = 10−2
cm/s. (B) Velocity arrow plot of jet formed by a single vortex line. (C) Normal-fluid
velocity magnitude for motion of four vortex line. (D) Normal-fluid velocity arrow plot
showing the rotation of four vortex line.

even when the velocity of the jet differs in magnitude and direction from that of
the superfluid vortex line velocity.
Four superfluid vortex lines orientated in the same direction will rotate
around each other. Starting with this initial condition, we observed the develop-
ment of a net rotation in the normal-fluid (shown in Fig. 2b and d). For these
multiple vortex line simulations, the vortex line motion is due to the advection
170 O.C. Idowu et al.

by the other vortices and no external superfluid driving velocity is needed to

induce the normal-fluid flow.
The localised normal-fluid flow structures formed by the mutual friction force
have an unusual spatial form characterised by the length and the width of the
jet. This type of forcing (usually confined to large length scales) may affect the
characteristics of the normal-fluid turbulence in coupled helium II flow.

5 Numerical Methods for 3-D Flows

Our model of interacting superfluid combines the two fluid idea of Tisza-Landau
with the Onsager-Feynman idea of the quantized superfluid vortices. In other
words, this combines the Eulerian with the Lagrangian formulations of fluid
dynamics.
Correspondingly, the numerical problem of calculating three-dimensional in-
teracting superfluid flows can be split into three parts. The first part concerns
the solution of the equations that describe the evolution of the free superfluid
line vortices. The second part has to do with the solution of the system of non-
linear partial differential equations that describe the free normal-fluid flow, the
well known (incompressible) Navier-Stokes equations.
The third sub-problem is related to the numerical modelling of the mutual
friction interaction, as well as, to the coupling of the two already mentioned
calculations. This is the novel part in every method treating interacting super-
fluid. In the past there have been works that calculated interacting flows, but in
these works the normal-fluid was modelled as having an infinite inertia, and so
one could use a kinematic prescription for it (See the article by Samuels in this
volume). In this way therefore, one needs neither to employ advanced methods
of numerical fluid dynamics or to model the mutual friction force acting on the
classical fluid. The methods discussed in this section overcome this simplification
and allowing the normal-fluid to react makes possible the investigation of novel
physics which were beyond the capabilities of the previous methods.
For the calculation of the free normal-fluid we use Eulerian grid finite volume
methods. For the free superfluid we use Lagrangian vortex methods. We have
employed a vortex point method for the two-dimensional case and a vortex
filament method for the three-dimensional one. Vortex point methods are capable
of calculating three-dimensional flows too but the vortex filament method is the
natural one for superfluid since in the latter vorticity has support only on very
thin line vortices. In addition, this approach is already been used by Schwarz in
a series of papers and it is proven to be a useful approach. Next, we describe our
methods in greater detail.

5.1 The Free Normal-Fluid

The normal-fluid is described by the incompressible Navier-Stokes equation (1),
with the last term at the right hand side put to zero. In this way one can use
the many methods developed in the field of classical numerical fluid dynamics.
 Coupled HeII Flows 171

We use a finite volume method which was first discussed in Harlow and Welch
[14], later improved by Kim and Moin [15], and finally by Wray [16]. The details
of the method can be found in these papers. A useful text in which many basic
concepts are discussed is Strikwerda [17]. The method is well known in the
classical turbulence community, and although it has not a very high order of
accuracy, it has the advantage of being simple. This is particularly a virtue,
because one wants to combine this method together with a very different one (a
vortex filament method) for the superfluid. In this respect, one can first have the
whole of the numerical procedure constructed, and later if needed can introduce
higher order discretization schemes.
We note here in brief some important points:
(1) The convective and body-force terms are advanced in time with a Runge-
Kutta scheme of an O(Δt3 ) error, and space differenced with central schemes √ of
an O(Δx2 ) error. The stability analysis of the method requires CF L ≤ 3.
(2) The diffusion terms are discretized with a Crank-Nicholson scheme of
error O(Δx2 ) in space and O(Δt2 ) in time. Since this scheme is an implicit
method, it requires the inversion of matrices. In order to have a method that
is not too computationally expensive an operator splitting is employed which
reduces the matrix inversion problem to the inversion of tridiagonal matrices
(one inversion for each direction). This method of fractional steps is of error
O(Δt3 ) in time (and thus is not inconsistent with the accuracy of the Runge-
Kutta scheme), and by not being iterative it is fast.
(3) The pressure is advanced in time with an Euler method of error O(Δt).
The space discretization is done with central differencing and introduces errors
of O(Δx2 ).
(4) The incompressibility condition is imposed by solving a Poisson equation
for pressure.
We have done at present many calculations with CF L = 1.5. All these cal-
culations were done with periodic boundary conditions. The calculations were
robust, with no instability problems whatsoever, and with no problems in satis-
fying the condition for the incompressibility of the velocity field. We have used
standard tridiagonal matrix inversion methods, as found in Press et al [12], and
numerical fast fourier transforms for the solution of the Poisson equation for
pressure (suitable for problems with periodic boundary conditions). Although,
the implicit method of Crank-Nicholson allows one to integrate the diffusion op-
erator with a time step much greater than the time step of the Runge-Kutta,
we have done calculations with the much smaller viscous time step in order to
resolve efficiently the viscous dominated phenomena in the normal-fluid. This is
in most times a necessity in interacting superfluid as the following scaling argu-
ment suggests. If the superfluid is put to interact with a stationary normal-fluid
one can (sufficiently away from the superfluid vortex core) assume that the mag-
nitude of the mutual friction force is of the order of the quantum of circulation κ.
The mutual friction forcing, in order to generate convective motion of a high Re
number, must overcome the friction forces which scale with ν. Now, because of a
numerical accident these constants have very close values for liquid helium. This
172 O.C. Idowu et al.

implies that the structures seen in normal-fluid when stirred by the superfluid
should be highly dissipative and thus need to be resolved all the way down to
viscous time scales. One should not expect a wide range of normal-fluid scales
to be excited by the mutual friction force. However, one should be careful when
applying such scaling arguments close to the vortex core.
Another important point is that of space resolution. In order to be consistent
with a continuum description of the normal-fluid we must resolve all possible
scales for which a continuum approach should be valid (we refer to them as
continuum scales). It is conceivable of course that the normal-fluid has important
dynamics even at scales where the continuum theory is not valid; in such a
case our model is also useless and one should resort to kinetic theory kind of
calculations. One hopes that what is left out has not an important effect on what
is resolved. This is a major and ever occurring problem in physics.
We do calculations with Knudsen numbers ranging from 100 to 1000 (a very
big value of the Knudsen number corresponds to the continuum regime). We
note here that when calculating superfluid is not enough to just resolve the
Kolmogorov scale for the normal-fluid; interesting physics (in fact perhaps the
only interesting physics) should appear below this scale. Finally, we might want
to resolve all possible time scales at which the continuum theory holds. In such
case we should choose a time step of the order 100 or 1000 mean free path
times. Such a time step is expected to be smaller than the turn over time of the
Kolmogorov scale size eddies.

5.2 The Superfluid

The free superfluid is modelled numerically through a vortex filament method
(see the article by Samuels in this volume). Our vortex method has fewer prob-
lems than the analogous methods used in classical fluid dynamics. The main dif-
ference is in the initialisation of the vorticity. In normal-fluid one has a smooth
vorticity field as an initial condition, which one has to approximate with singular
concentrations of vorticity: vortex points or vortex filaments. In the case of su-
perfluid such a problem does not exist, because in superfluid vortex filaments are
not an idealisation as in classical fluids but they are what one observes macro-
scopically in the laboratory. From the mathematics point of view the vorticity
is not defined on a superfluid vortex line. Tracking the motion of vortex lines
though it is useful, since the circulation of quantum lines is not only well de-
fined, but also always constant. This implies that the free superfluid turbulence
should be simpler than the classical one. Indeed, in the classical case not only
is not correct to say that the flow can be reduced to an assembly of a finite
number of classical vortex filaments, but even in case such a thing could be true
we would have to deal with an infinite number of classical vortex configurations
corresponding to a possible infinite range of possible circulation values. These
considerations, although are motivated from numerical analysis they have a value
from the physics point of view too, suggesting that the relative simplicity of free
quantum turbulence could help in gaining some insight into the complexities of
classical turbulence.
 Coupled HeII Flows 173

The equation for the Lagrangian vortex line dynamics is equation (2). This
equation models the physics of the balance between the Magnus, the mutual
friction, and the Iordanskii forces. The last two forces are been set to zero when
studying free quantum turbulence by putting zero all the terms except those
corresponding to the self-induction one. Although we can do calculations with
both the exact Biot-Savart law, as well as, with the Local Induction Approxima-
tion (LIA), we use the later only when the full Biot-Savart computation becomes
impractical because of computational complexity.
In implementing our method we discretize each vortex line using a Lagrangian
grid along each vortex line. In this way we reduce the problem from an integrod-
ifferential one to a system of ordinary differential equations. For the later system
we use a standard Runge-Kutta-Fehlberg method of an O(Δt4 ) accuracy. The
time stepping is done with a predictor-corrector procedure. We note that the
discretization along each vortex is not an arbitrary procedure. If it results in
a very dense grid one sees that in order to retain stability one has to employ
very small time stepping. The adequate (for stability) time step for a particular
Lagrangian grid choice should come from a linear stability analysis of the full
integrodifferential equation. At present we are not aware of such an analysis. So
we decide first on a particular space discretization and then by trial and error we
decide upon an appropriate time stepping. The Lagrangian grid size is chosen to
be the same as the Eulerian one for the normal-fluid. The time stepping on the
Lagrangian calculation should be appropriate for resolving the fastest possible
Kelvin waves present in the system. Its calculation is as follows: the group ve-
locity of a Kelvin wave of wavelength λ on a quantum vortex line of circulation
κ and core size a ∼ 10−8 cm, is given by:
 
κ λ
Vwave (λ) = ln (31)
2λ 2πa
Next, we find that should the Lagrangian grid spacing be Δl, it takes time
Δt = Δl/Vwave for the fastest Kelvin wave (which corresponds to λ = 2Δl) to
travel exactly the smallest resolvable distance. Then one needs a minimum time
step equal to Δt/2 in order to resolve all Kelvin waves in the system. All our
calculations meet at least this requirement. A note is in order here: although
the superfluid vortex ring has the tendency to develop structure at smaller and
smaller scales, this structure formation has an ultraviolet cut-off. This is de-
termined by the strongest of the following three mechanisms: acoustic damping,
mutual friction [18], and (mostly active at finite temperatures) a mechanism due
to Glabertson-Ostermeier instability, as discussed in Samuels and Kivotides [19].
Of course in general one might not be able to resolve all relevant scales due to
space and time computational complexity. In such a case one should at least be
careful to resolve scaling regimes as information about them comes available in
the literature. Finally, one needs to ensure that a vortex line does not transverse
more than an Eulerian grid cell at a particular time step. This condition we refer
to as superfluid CFL (SFCFL) condition.
Resolution considerations are very important in calculating interacting super-
fluid. Although it is easy to choose the smallest of the superfluid and normal-fluid
174 O.C. Idowu et al.

time steps to do the calculation, the choice is not so easy in space resolution.
Even if we resolve all normal-fluid continuum scales (obviously fully resolving
the Kolmogorov scale) it could still be meaningful for the superfluid vortex lines
(only) to develop structure at even smallest scales (should the temperature is
such that there is no effective damping at continuum scales). In such a case
one notes that the benefit of allowing the superfluid grid to be denser than the
normal-fluid grid is not balanced by the resulting increase in the logical depth of
the algorithms or the extra computational complexity . In such cases, we have
chosen the same grid for both fluids.

5.3 The Interaction Modelling

The modelling of the mutual force is based on the observation that the term in
the Navier-Stokes which corresponds to this force should have units of accelera-
tion. The relation (10) gives the mutual friction force per unit length of vortex
line. We use this relation as follows:
(1) We do a search first over the Eulerian domain to track which fluid volumes
are threaded by a vortex line.
(2) We next (for each fluid element with the above property) calculate the
integral of formula (10) over the length of line enclosed inside a particular fluid
volume.
(3) Finally in order to have the correct units we divide the total force that we
have obtained at step 2, with the density of the normal-fluid times the volume
of the particular fluid volume. The result of this operation is appropriate for
inclusion into the equation of motion (1).
Modelling the mutual friction force is not the only new aspect of any interact-
ing superfluid calculation. One additionally considers the problem of increasing
computational complexity if, for example, in a turbulence calculation one has
a grid 2563 which is quite common in well resolved turbulent flows, and in ad-
dition one must combine such a calculation with a vortex filament method. As
noted before, one needs (as far as the superfluid is concerned) to resolve Kelvin
wave times which are expected to be much smaller than the viscous times for
the normal-fluid. It is not practical in such calculations to advance the normal-
fluid with the same time step as the superfluid. We have developed a sub-cycling
method which allows the superfluid to evolve inside one normal-fluid outer time
step during which evolution the normal-fluid velocity and pressure fields remain
frozen. We allow up to 10 sub-cycling steps but one should adjust this number
according to the particular physics he is looking after. This method gave us no
stability problems, although from the mathematics point of view it is possible
to lead to instability.

5.4 Preliminary Results in 3-D Flows

Our program for investigating interacting superfluid flows has just started [20].
At the moment we have used our methods to investigate the motion of superfluid
rings in a stationary normal-fluid. The results, are closely related to those of the
 Coupled HeII Flows 175

2-D flow. Fig. 3 shows a section of the normal-fluid velocity field generated by
the mutual friction forcing from superfluid vortex ring. The velocity field shown
is taken in the plane perpendicular to the plane of the superfluid vortex ring, so
the superfluid vortex filaments pass through this plane at right angles, much as
in the 2-D simulations. The normal-fluid near the superfluid vortex filament is
dragged by the superfluid vortex, producing localised jets of normal-fluid.
In 3-D, this dragging of the normal-fluid by the superfluid vortex forms a
pair of normal-fluid vortex rings associated with the superfluid vortex ring. The
outer normal-fluid vortex ring is oriented in the same direction as the superfluid
vortex ring, while the inner normal-fluid vortex ring has the opposite orientation.
Without the mutual friction force from the superfluid vortex, this normal-fluid
flow structure would rapidly dissipate. The Reynolds number of the normal-
fluid flow, calculated using the peak normal-fluid velocity and the width of the
normal-fluid jet, is of order unity, confirming the dissipative nature of the flow
structure.

Fig. 3. The normal-fluid velocity field in the 3-D simulation. The velocity shown is
measured on the plane perpendicular to the plane of the superfluid vortex ring.

6 Discussion and Conclusion

Does the mutual friction force of the superfluid vortex filaments generate any
complicated, possibly turbulent flow in the normal-fluid? In the simulations re-
ported here, it clearly does not. The normal-fluid flow structures generated are all
of low Reynolds number and are thus rapidly dissipated. No significant normal-
fluid flow is seen far from the source of the forcing, the superfluid vortex. In this
sense, the mutual friction force does not significantly ‘stir’ the normal-fluid.
But even though mutual friction does not directly generate any complicated
flows in the normal-fluid, there are other possibilities that must be considered.
It is possible that the mutual friction force could trigger instabilities in normal-
fluid shear flows. Stability calculations of a normal-fluid flow under a specified
mutual friction force distribution show that such instabilities can be triggered,
176 O.C. Idowu et al.

at sufficiently large mutual friction forcing amplitudes [2]. This behaviour has
not yet been tested in the fully coupled two fluid calculations described in this
paper, since we have only considered normal-fluid flows generated in an initially
stationary normal-fluid. The direct effect of the mutual friction force on the
normal-fluid appears to be limited to the formation of the dissipative jet. The
indirect effect of this forcing on the normal-fluid flow stability now needs to be
considered (See Godfrey’s article in this volume).

Acknowledgements

This research is supported by University of Newcastle upon Tyne and the Lev-
erhulme Trust.

References
1. K. W. Schwarz: Phys. Rev. B 31 5782 (1985).
2. D. J. Melotte, C. F. Barenghi: Phys. Rev. Lett. 80 4181 (1998).
3. O. C. Idowu, A. Willis, D. C. Samuels, C. F. Barenghi: Phys. Rev. B 62 3409
(2000).
4. C. F. Barenghi: J. Phys., Condens. Matter 11 7751 (1999).
5. H. E. Hall and W. F. Vinen: Proc. Roy. Soc. A 238 215 (1956).
6. C. O. Idowu, D. Kivotides, C. F. Barenghi, D. C. Samuels: J. Low Temp. Phys.
120 269 (2000)
7. C. F. Barenghi, R. J. Donnelly, W. F. Vinen: J. Low Temp. Phys. 52 189 (1983).
8. W. F. Vinen: Proc. Roy. Soc. A 242 493 (1957).
9. R. J. Donnelly: Quantized Vortices in Helium II , Cambridge University Press
(1991).
10. R. J. Donnelly, C. F. Barenghi: J. Phys. Chem. Ref. Data, Vol.27, No.6, (1998).
11. R. Peyret, T. D. Taylor: Computational Methods for Fluid Flow , Springer-Verlag
(1983).
12. W. H. Press, S. A. Teukolsky, W. T. Vetterling, B. P. Flannery: Numerical Recipes
in Fortran 77 second edition , Cambridge University Press (1992).
13. R.G.K.M. Aarts, A.T.A.M. deWaele: Phys. Rev. B 50 10069 (1994).
14. M. W. Harlow, J. E. Welch: Phys. Fluids 8 2182 (1965).
15. J. Kim, P. Moin: J. Comp. Physics 59 308 (1985).
16. A. A. Wray: Very Low Storage Time-Advancement Schemes, Internal Report,
NASA Ames Research Center, Moffet Field, California (1987).
17. J. Strikwerda: Finite Difference Schemes and Partial Differential Equations,
Wadsworth and Brooks/Cole (1989).
18. W. F. Vinen: Phys. Rev. B 61 1410 (2000).
19. D. C. Samuels, D. Kivotides: Phys. Rev. Lett. 83 5306 (1999).
20. D. Kivotides, C. F. Barenghi, D. C. Samuels: Science 290 777 (2000)
From Vortex Reconnections
to Quantum Turbulence

Tomasz Lipniacki

Institute of Fundamental Technological Research,

Świȩtokrzyska St. 21, 00-049 Warsaw, Poland

Abstract. An alternative approach to quantum turbulence is proposed in order to

derive the evolution equation for vortex line-length density. Special attention is paid
to reconnections of vortex lines. The summed line-length change ΔS of two vortex
lines resulting from the reconnection (in the presence of counterflow Vns ) can be ap-
proximated in the form: ΔS = −at1/2 + bVns 2 3/2
t , with a > 0, b ≥ 0, at least until
ΔS ≤ 0. For steady-state turbulence, the average line-length change ΔS between
reconnections has to be zero. If, for a given value of the counterflow, the line density
is smaller than the equilibrium one, the reconnections occur less frequently and ΔS
becomes positive and the line density grows until the equilibrium is restored. When
the line-density is too large, the reconnections are more frequent, the lines shorten be-
tween reconnections and the line density gets smaller. The time derivative of the total
line density is proportional to the reconnection frequency multiplied by the average
line-length change due to a single reconnection. The evolution equation obtained in
the proposed approach resembles the alternative Vinen equation.

1 Introduction

The variety of the dynamic phenomena exhibited by superfluid 4 He involves

the appearance and motion of quantized vortices. Due to the existence of these
singularities the superfluid component is coupled dissipatively with the normal
component. At low velocities He II (superfluid 4 He) flows in the frictionless,
presumably laminar manner consistent with the ideal fluid description. When the
counterflow (the relative velocity of the two helium components) vns = vn − vs
becomes sufficiently large, however, the superfluid laminar flow develops into
superfluid turbulent flow in which the quantum vortices form a chaotic tangle.
The aim this work is to derive the evolution equation for vortex line-length
density L. In the proposed approach special attention is paid to reconnections
of vortex lines. The vortex line evolution is analyzed in the localized induction
approximation, supplemented by the assumption that when two vortex lines
cross, they undergo a reconnection.
Let us recall that if the curve traced out by a vortex filament is specified in the
parametric form s(ξ, t), then in the superfluid reference frame the instantaneous
velocity of a given point of the filament is given by the equation (in scaled units
in which the elementary circulation around vortex line equals to unity)

ṡ = s × s + αs + αs × vns , (1)

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 177–183, 2001.

c Springer-Verlag Berlin Heidelberg 2001
178 T. Lipniacki

where dot and prime denote instantaneous derivatives with respect to the scaled
time τ and arc length ξ, respectively, α is the non dimensional friction coefficient
and vns the counterflow. 
The line-length of the vortex filament l = dξ, which motion is given by (1),
satisfies the equation
  
∂l
= αvns · (s × s ) − α|s |2 dξ . (2)
∂τ

2 Vortex Motion Following Reconnection

We consider an idealized reconnection of two infinite straight vortex filaments.

In the moment of reconnection these two filaments transform into two ”new”
sharply bent vortex filaments. We analyze the evolution of a vortex line forming
at the initial state (τ = 0) angle 2φ.

2.1 The Case vns = 0

With vn = vs = 0 Eq.(1) simplifies to

ṡ = s × s + αs . (3)

It is not difficult to show that in this case the vortex line for all
√ times will have
a similar shape (Fig.1), whose spatial scale D is growing as τ . This leads to
the result
Δl = −A τ 1/2 , (4)
where Δl is the line change and A = A(α, φ) > 0.
Let us consider now the dynamical equation (3) without self the induction
term s × s . In such an equation the coefficient α can be absorbed into time τ

Fig. 1. Vortex line evolution after reconnection for reconnection angle φ = π/4, and
friction coefficient α = 0.1. The line positions are shown at τ = i2 τo .
 From Vortex Reconnections to Quantum Turbulence 179

and one gets

∂s
= s . (5)
∂(ατ )
It is clear that the line-length reduction Δlo in the evolution governed by the
above equation satisfies
Δlo = Bo (φ) (ατ )1/2 . (6)
The simulations revealed that Δl ≈ Δlo . This result can be justified as follows
(Lipniacki [1]) For α = 0, the vortex evolution equation (1) reduces to the non
dissipative one,
∂s
= s × s . (7)
∂τ
During the evolution governed by Eq.(7), the following quantities are constant
in time 
l = const , |s |2 dξ = const. (8)

Now, one can evaluate the vortex line step by step using the method of fractional
steps. In the first half of time step we solve Eq.(5), in the second one we solve
Eq.(7) using the previous result as an initial condition. The evolution governed
by non dissipative Eq.(7) does not change the line length and does not change
the squared curvature - which determines the rate of line-length changes (see
Eq.2) during the evolution under Eq.(5). Roughly speaking, although the first
term in Eq.(3) significantly changes the evolution of the vortex (line shape), it
does not significantly influence the rate of the line-length reduction. As the result
we have:
A ≈ B(φ) α1/2 , i.e. Δl ≈ −B(φ) (ατ )1/2 . (9)

2.2 The Case vns = const = 0

In this case, for a given α and φ, we consider the summary line-length change
ΔS of two vortex lines resulting from the reconnection:
ΔS = Δl1 + Δl2 . (10)
Let us note, that just after the reconnection, close to the reconnection point,
the two vortices have the binormals (s × s ) opposite to each other. The first
term in Eq.(2) when integrated over two lines resulting from reconnection gives
no contribution to total line length change. Because the second term of Eq.(2)
is always negative the total length of the resulting vortices decreases in the
beginning. However, during the further evolution, the characteristic curvatures
(and so the absolute value of second term) get smaller, and the two vortices turn
so that the average value of vns ·(s ×s ) becomes positive. Then the total length
of vortices starts growing.
It can be shown (Lipniacki [1]) that after the idealized reconnection of two
straight vortex lines the summary line-length change ΔS of two vortices resulting
from the reconnection can be expanded as function of vns and τ
ΔS = a1 τ 1/2 + a2 τ 3/2 vns
2
+ a3 τ 5/2 vns
4
+ ... . (11)
180 T. Lipniacki

2
Fig. 2. The vortex evolution in configuration 2 for α = 0.3 , τ vns = 32, 64, 96, 128.
The separated vortex ring is seen the final frame.

The coefficients ai depend on α, φ and the direction of vns with respect to re-
connecting lines.

Moreover, the numerical simulations revealed (Lipniacki [1]) that at least

until ΔS ≤ 0 one can restrict to first two terms.

ΔS(τ ) = −a τ 1/2 + b τ 3/2 vns

2
, (12)

where a > 0, b ≥ 0.
To apply the above result to the analysis of the dynamics of the whole vortex
tangle one should average the function ΔS(τ ) over a representative sample of
reconnection configurations. We found that when estimating the average line-
length change ΔSo (τ ) = ΔS(τ ) one can use the simplified dynamical equation

ṡ = αs + αs × vns , (13)

(i.e. without self induction term) and get roughly the same result as for Eq.(1)
(Lipniacki [1]). We should note, however, that Eq.(1) can be replaced by Eq.(13)
only when calculating the line length change averaged over a representative
sample of configurations. In Eq.(13) α may be absorbed into time scale; in the
result we have
ΔSo = −ao (ατ )1/2 + bo (ατ )3/2 vns
2
. (14)
To estimate ao , bo we assume that every line in the moment of reconnection
has to be parallel or antiparallel to one of 3 ”main” directions: x̂, ŷ, ẑ and that
the direction of vns is (1, 1, 0) in Cartesian coordinates. In principle, this means
that there are 12 possible reconnection configurations, but it is not difficult to
 From Vortex Reconnections to Quantum Turbulence 181

check that only 3 of them are essentially different. The average line-length change
ΔSo (τ ) is then estimated by averaging over these configurations. In the result
we get ao = 2.34, bo = 0.514 .
Let us notice that the coefficients ao , bo determine the characteristic non
dimensional time τ ∗ = τ vns2
= ao /bo for which the average line-length change
after reconnection is zero.

3 The Model

Equation (14) gives one a hint for the explanation of how the vortex tangle is
sustained: for the steady-state turbulence, the average line-length change be-
tween reconnections has to be zero. If for a given value of the counterflow vns ,
the line density is smaller then the equilibrium value, the reconnections occur
less frequently, and so, the characteristic time between reconnections is longer;
hence according to Eq.(14) the line-length change between reconnections be-
comes positive. As the result the line density of vortex the tangle grows until the
equilibrium is restored. Inversely, when the line density is too large, the recon-
nections are much more frequent, so the decaying term in Eq.(14) prevails and
the line density gets smaller.
In the proposed approach (see Lipniacki [2] for more details) the dynamics
of a vortex tangle will be considered as a sequence of reconnection followed by
a ”free” evolution of vortex lines resulting from each reconnection.
Let the instantenous line-length density be L. This gives one the characteristic
spacing between lines lo and the characteristic (average) line curvature |s |

lo = L−1/2 , |s | = 1/lo = L1/2 . (15)

Let us divide the vortex lines into segments with length lo . Because the length
lo is equal to the spacing between lines we may expect that each segment moves
(more or less) as a unity, but the motions of the neighboring segments are not
strongly correlated. Moreover, because the length of segments is equal to the
characteristic radius of curvature, when considering collisions, the segments can
be roughly treated as straight ones. It is assumed that any collision of vortex
segments leads to a reconnection; such assumption is justified by numerical sim-
ulations of Schwarz [5]. For the sake of simplicity we assume that all segments
move with the same speed vo which can be estimated basing on (1):
/
2
vo = |ṡ| = L(1 + α2 ) + (αvns )2 . (16)
3

Having vo and lo (i.e. the number of segments per unit volume n = L/lo = L3/2 )
one can estimate the reconnection frequency (per unit volume) fr ,

vo L2
fr = , (17)
2
182 T. Lipniacki

and the average time between reconnections of line segments

1
τc = . (18)
L1/2 vo
The estimation of fr needs some algebra since the cross section for collision of
two given segments depends on their orientation with respect to their relative
velocity. The time derivative of line-length density is then
dL
= fr ΔS(τc ) , (19)
dτ
where ΔS(τc ) is the average line-length change due to a single reconnection.
The resulting evolution equation is
dL 1 1
= − ao α1/2 L7/4 vo1/2 + bo α3/2 L5/4 vo−1/2 vns
2
, (20)
dτ 2 2
with vo given by Eq.(16).

4 Results
The obtained evolution equation (20) differs from the classical Vinen equation
(Vinen [8])
∂L
= −βv L2 + αv |vns |L3/2 , (21)
∂t
where αv and βv are temperature dependent coefficients. When α  1 and
2
αvns  L the line velocity vo can be approximated by L1/2 and the evolution
equation simplifies to
dL 1 1
= − ao α1/2 L2 + β −1 α3/2 L vns
2
. (22)
dt 2 2
which is know as the alternative Vinen equation. The generation term is pro-
2
portional to Lvns what is closer to the phenomenological theory of classical
turbulence. Indeed, (Niemirowski, Fiszdon [4]) by assuming that turbulence can
be characterized by a parameter, say, L, and that its time derivative dL/dt is an
analytic function of L, the alternative form of Vinen equation can be interpreted
as the first two terms in series expansion. Furthermore, as the generation term
is the scalar function of vector argument vns , it is reasonable that the series
expansion starts with this argument squared. The last comment concerns also
the full form of the evolution equation obtained in the model, i.e. Eq.(20) where
the counterflow velocity vns is also everywhere squared. The presence of absolute
value of vns in the classical Vinen equation is a little bit strange.
The experimental data are not precise enough to chose between Vinen, al-
ternative Vinen equation, or the Eq.(20). It makes sense only to compare with
experimental data the steady-state value of L following from (20) i.e. L∞
 α4  bo α 2 α 2 
2
L∞ = vns +8 − (23)
9 πao 3
 From Vortex Reconnections to Quantum Turbulence 183

In a broad range of α (α ∈ [0.01, 1]) the value of L∞ predicted by (23) generally

agrees with experimental data and Schwarz’s [6] numerical simulations of steady-
state turbulence (see [2]).
Equation (20) may be compared with data from Schwarz and Rozen[7]. The
authors analyzed in numerical simulation of non-equilibrium turbulence large
transients, in which the line-length density grows from small to large values,
and found that the coefficients in the Vinen equation are not constant. One
can calculate how those coefficients vary in time by comparing the growth term
of our evolution equation (20) with the Vinen one. Then one may check [2]
that Schwarz-Rozen data are considerably better fitted by Eq.(20) than by the
classical Vinen equation.
Our main result is the construction of the simple model in which the micro-
scopic analysis a of quantum tangle leads to a macroscopic evolution equation
for line-length density. The numerical simulations needed to estimate the coef-
ficients ao , bo in Eq.(20) are relatively simple, when compared with simulations
of Schwarz. The main advantage of the presented approach is the possibility to
generalize it to anisotropic flows with significant macroscopic superfluid vortic-
ity. Such flows are expected in such phenomena as spin-up or boundary layer
forming. The viscous forces in a cylinder, which starts spinning from rest, act-
ing on the normal component, may give rise to counterflow, large enough to
cause quantum turbulence which may significantly influence the dynamics of
both components. In this way the angular momentum can be transferred from
the cylinder via the normal component to the superfluid component. One may
expect the following scenario: due to viscous forces the normal component starts
spinning, and this implies counterflow which generates quantum turbulence. The
mutual friction forces couple two components (vns → 0) and in the end quantum
vortices polarize to form a pattern of straight parallel lines and both fluids spin
together. However, because of the large normal fluid velocity gradients, even
in the first stages of spin-up process the arising quantum turbulence is highly
anisotropic; the tangle of quantized vortices is polarized to carry considerable
macroscopic superfluid vorticity. This is probably why the description of spin-up
process in terms of Vinen model was found to be inconsistent (Lipniacki [3]).
The line-length density calculated from Vinen equation was smaller then the line
density calculated from superfluid velocity profiles. The origin of these vortices
cannot be explained within the Vinen model.

References
1. T. Lipniacki: European Journal of Mechanics B/Fluids 19, (3), 361 (2000)
2. T. Lipniacki: Arch. Mech. 53, (1), 23 (2001)
3. T. Lipniacki: Arch. Mech. 49, (4), 615 (1997)
4. S. K. Nemirowski, W. Fiszdon: Rev. Mod. Phys. 67 (1), 37 (1995)
5. K. W. Schwarz: Phys. Rev. B 31, 5782 (1985)
6. K. W. Schwarz: Phys. Rev. B 38, 2398 (1988)
7. K. W. Schwarz, J. R. Rozen: Phys. Rev. B 44, 7563 (1991)
8. W. F. Vinen: Proc. R. Soc. London Ser. A 242, 493 (1957)
Vortices and Stability
in Superfluid Boundary Layers

Simon P. Godfrey, David C. Samuels, and Carlo F. Barenghi

Dept. of Mathematics, Univ. of Newcastle, Newcastle upon Tyne, NE1 7RU, UK

Abstract. Boundary layer flows are of critical importance to the motion and physics
of a fluid flow. In superfluid helium II, the problem of boundary layer flow highlights a
significant difference between the two fluid components that make up the current model
of helium II: the difference in the boundary conditions. This two fluid model also has a
fundamental effect on the stability of the flow, as one must consider the stability of both
of the two fluid components. We describe superfluid vortex structures which remain
stable in a laminar boundary layer, and discuss their effect on the boundary layer. We
also calculate the stability characteristics of the plane Poiseuille flow of helium II, and
have discovered a new unstable mode caused by mutual friction, that can dominate
the stability characteristics of the flow.

1 Introduction
1.1 The Two-Fluid Model
Superfluid helium II (i.e. below a temperature of 2.17o K) is best described as a
superposition of two fluid components: a normal fluid and the superfluid. Both
of these two components have their own velocity fields and densities (denoted
by the subscripts n and s respectively). The normal fluid component is basically
a classical Navier-Stokes (i.e. viscous) fluid whereas the superfluid component
is an inviscid Euler fluid with some quantum effects. These two fluids interact
via the mechanism of mutual friction, which exchanges energy and momentum
between the two fluids. Mutual friction is caused by the interaction of the rotons
and phonons (constituents of the normal fluid) with quantized superfluid vortex
lines[1].

2 Boundary Layer Vortices

We shall first consider the laminar helium II boundary layer. The normal fluid
obeys no-slip boundary conditions, whereas the superfluid is allowed to slip freely
at the boundary. This difference in boundary conditions suggests there may be
a region in the flow, near the boundary, where there are large differences in the
two velocity fields.

The vorticity in the superfluid component is confined to quantized vortex

filaments, which have velocity Ṡ approximated by
Ṡ = Vs + αS × (Vn − Vs ) (1)

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 184–190, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Vortices and Stability in Superfluid Boundary Layers 185

where S represents a position along the vortex filament, the dot represents a
time derivative, prime the derivative with respect to arclength ξ, and α is the
mutual friction parameter. This is an approximate equation of motion since we
are neglecting the α term which is typically quite small. The term proportional
to α gives the primary response of the vortex filament to the mutual friction.
This force will be largest where the velocity difference is largest. Thus we would
expect a strong mutual friction force on superfluid vortex lines near a boundary.
We model the helium II boundary layer flow by assuming that the normal
fluid has a Blasius profile (with a velocity Un far from the boundary), and the
superfluid has a uniform velocity profile with a value Us . Both velocity fields are
taken to be in the x̂ direction, along the boundary, and the parameters Us and
Un are independent of each other.
We shall also assume that we are far enough from the leading edge of the
boundary layer so that the boundary layer thickness δ is approximately constant
over the region which we are considering. We seek to find non-trivial superfluid
vortex filament structures that are stable, even at non-zero temperatures where
energy dissipation occurs.
In an inviscid Euler flow, there exist many vortex structures which can move
without changing their size or shape (vortex rings, helical vortex waves, vortex
solitons) [2]. However, in superfluid helium II at non-zero temperatures (when
the normal fluid component is present) energy and momentum transfer between
the two fluid components causes all those examples to become unstable. Trivial
vortex structures (e.g. vortex lines parallel to the direction of both the velocity
fields) with the term αS × (Vn − Vs ) in (1) equal to zero at every point along
the vortex structure will remain stable at non-zero temperatures. We seek to
find non-trivial superfluid vortex structures that retain both their shape and
size, but that have non-zero mutual friction force at almost all points along their
length [3]. Such structures would constantly exchange energy and interact with
the normal fluid, and as such would be important to the dynamics of the helium
II boundary layer flow.
We seek to define a vortex configuration S(ξ, t) which will keep the same
shape while moving in the streamwise direction x̂ with a uniform and steady
speed UD relative to the mean superfluid flow Us . Note that we are not cal-
culating vortex structures which are pinned to the boundary. Using the Local
Induction Approximation (and a lot of algebra) we derive an equation for such
a structure. Further details of this calculation are given in [4]). This nondimen-
sionalized equation is

S = A {S × UD + α2 [S (Vn − Us )]

+ α(Vn − Us ) − UD − S [S · (Vn − Us − UD )]} (2)

where A is a constant of order 1. The boundary conditions are that the vortex
filament must be perpendicular to the boundary on the boundary, so that the
condition of zero flow through the boundary is not violated. There are some
solutions of (2) that do not return to the boundary, but we are not concerned
with these. This equation was solved numerically using a third order Adams-
186 Simon P. Godfrey, David C. Samuels, and Carlo F. Barenghi

Bashforth-Moulton predictor-corrector method. Equation (2) can be used to find

steady state superfluid vortex configurations in any geometry. The geometry of
the flow is determined by the boundary conditions on the end of the superfluid
vortex filament, and by the steady normal fluid velocity V n (x).

2.1 Properties of the Vortex Line Solutions

We hold two of the system parameters (α and the ratio Us /Un ) constant, and
vary the other two (UD and Un ), and thus map out the solutions to (2) of this
plane of the parameter space. A typical plot of this is shown in Fig. 1.

2 2

0 0
D

-2 D -2 α = 0 .2
V o r t e x D r if t V e lo c it y , U

V o r te x D r ift V e lo c ity , U
-4 α = 0 .1 -4 U s
/U n
= 0 .7 5
-6 U s
/U n
= 0 .7 5 -6

-8 -8

-1 0 -1 0

-1 2 -1 2

-1 4
(a ) -1 4 (b )
-1 6 -1 6
0 5 1 0 1 5 2 0 2 5 3 0 3 5 4 0 0 5 1 0 1 5 2 0
N o r m a l F lu id V e lo c it y a t y = in f in it y , U n
N o r m a l F lu id V e lo c ity a t y = in f in ity , U n

Fig. 1. Values of UD and Un which give give solutions to (2) and satisfy the boundary
conditions. The middle branch of the plot is stable, and the upper and lower branches
were found to be unstable. Parameter values are given in the figure.

For other parameter ranges (α and Us /Un ) the plot is qualitatively the same,
with three branches. Using another vortex simulation (based on the Biot-Savart
law, which provides an independent test of these vortices) we have determined
that the middle branch is the stable branch, with the lower branch solutions
growing to become their middle branch counterparts, and the upper branch
solutions shrinking rapidly to zero length.
Plots of these middle branch solutions are given in Fig. 2. These vortices move
with a constant velocity through the boundary layer flow without changing their
shape or form. They extend to a typical height of 2δ to 4δ (the boundary layer
extends to a height of approximately 5δ). These structures are physically similar
to the classical horseshoe vortex (which are associated with turbulent boundary
layers, laminar flow around obstacles, interactions between boundary layers and
jets and have been linked with turbulent bursting phenomena[5]), both in size
and velocity (although our superfluid vortices move with greater speed), yet the
underlying physics are completely different in all of these cases.
These superfluid vortex structures are not static objects, but have a dy-
namical equilibrium between vortex stretching and decay. The lower part of the
structure is growing, whereas the upper section is shrinking. For further details
 Vortices and Stability in Superfluid Boundary Layers 187

2 0
N o rm a l (a )

1 0

A B C D

0
0 1 0 2 0 3 0 4 0 5 0 6 0
1 0
(b )
5
S p a n w is e

0
-5
-1 0
0 1 0 2 0 3 0 4 0 5 0 6 0
S tr e a m w is e
Fig. 2. Examples of the stable vortex configurations (middle branch). Parameter values
are α = 0.2, Us /Un = 0.75 and (A) Un =5, (B) 10, (C) 15, (D) 17. Note the increase in
length as the velocity Un increases (b), but they never extend up out of the boundary
layer (a).

of this mechanism, see Ref. [4]. This implies a constant energy transfer between
the normal fluid and the superfluid. Near the boundary, on the legs of the vortex
the normal fluid loses energy to the superfluid. Further away from the boundary,
in the centre section of the vortex, this energy is returned from the superfluid to
the normal fluid. Thus, this provides a mechanism for moving energy upwards
in the flow. It should be noted that this mechanism takes energy away from an
already low energy region of the flow.

2.2 Discussion
We have seen that these vortices are attractors, and thus we would expect rem-
nant vortex filaments within a boundary layer flow to develop into these struc-
tures. The existence of such structures should be considered in boundary layer
experiments, such as flow past a sphere [6], oscillating u-tube, oscillating discs
and spheres [7], as these geometries all have non-dimensional velocities within
the range where we would expect our structures to be present. These superfluid
vortex structures also act as conduits of energy in the normal fluid, transporting
kinetic energy away from the boundary. This transport of kinetic energy away
from the boundary is likely to have a destabilising effect on the normal fluid.
188 Simon P. Godfrey, David C. Samuels, and Carlo F. Barenghi

3 Stability Analysis

The question of the stability of the normal fluid flow due to vortices present in the
boundary layer is now addressed. We shall consider the 2D laminar flow between
two infinite boundaries. In the absence of a mutual friction coupling between the
two fluid components, the normal fluid will have a plane Poiseuille profile, and
the superfluid a constant profile. However, with mutual friction, the normal fluid
profile will be changed. We assume a vorticity distribution (Gaussian, centred at
the points where the velocity profiles are equal), and model the mutual friction
force Fmf by
Fmf = F(z)(Vn − Vs ) (3)
where F(z) represents the vorticity distribution. With this term added to the
equations of motion for the two fluid components we calculated the forced normal
fluid velocity profile. Further details can be found in [8] The next step is to
investigate the linear stability of this velocity profile.

3.1 Linear Stability

We follow the method of linear stability outlined in [9]. Starting with the mutual
friction forced Navier-Stokes equation for the normal fluid,
∂ 1 2
Vn + (Vn · ∇)Vn = −∇P + ∇ Vn + F[Vn − Vs ] (4)
∂t Re
we assume that the normal fluid velocity profile Vn is made up of a mean
flow U (z)x̂ (previously calculated) and a perturbation velocity u = (u , 0, w ).
Substituting this, and introducing the streamfunction Ψ of the form

Ψ (x, z, t) = Φ(z)eiβ(x−ct) (5)

we arrive at a modified Orr-Sommerfeld equation. Where β is the wavenumber of

the disturbance velocity, and c = cr +ici is the complex wavespeed, Note that the
growth rate of the perturbation is given by αci . The modified Orr-Sommerfeld
equation is

(U − c)(D2 − β 2 )Φ − U  Φ = (iβRe)−1 (D2 − β 2 )2 Φ

+ (iβ)−1 (F  D + FD2 − β 2 F)Φ (6)

with boundary conditions Φ(−1) = Φ(1) = 0 and Φ (−1) = Φ (1) = 0. Both D

and prime denote derivative with respect to z. This is an eigenvalue problem
which was solved numerically. The neutral stability curves are given by ci =0,
although the reader should note that the growth rate is given by βci , not just ci
alone. Since we are only considering the case of positive β, then if ci is greater
than zero, the perturbation grows exponentially, whereas is shrinks exponentially
if ci is less than zero.
 Vortices and Stability in Superfluid Boundary Layers 189

3.2 Stability Results

A typical plot of the neutral stability curves is shown in Fig. 3. The effect of
the mutual friction is to slightly alter the stability profile of the main branch,
but it also produces an entirely new unstable mode, which occurs for very low
wavenumber β. This new instability can have a critical Reynolds number which

1 .2
S ta b le

1 .0

0 .8
β L

0 .6
F o r c e d P r o file U n s ta b le
0 .0 4 P la n e P o is e u ille
0 .0 2
0 .0 0
1 0 0 0 1 0 0 0 0
R e
Fig. 3. Neutral stability curve for the forced normal fluid velocity profile. Forcing
parameters are peak forcing fmax =0.005, and width of forcing σ = 0.15.

is lower than the main unstable branch. We always find that this lower branch
has its most unstable eigenmode occurring at zero wavenumber. This stability
analysis predicts that the channel flow of helium II goes unstable in a fundamen-
tally different way than the classical Navier-Stokes flow. The critical Reynolds
number of the lower branch may also be calculated analytically, and gives the
result as
π2
Recrit = (7)
fmax
where fmax is the peak value of the vorticity distribution, related to the vortex
line density. This analytic result agrees very well with our numerical data, as do
the predicted and calculated eigenmodes. We have found that the shape of the
Gaussian vorticity distribution is not critical to the stability of the flow, but the
maximum value of it is important. A contour plot of the spatial structure of the
streamfunction [the real part of Φ(z) exp (iβx)] is plotted in Fig. 4. The lower un-
stable mode has a much simpler spatial structure than its classical main branch
counterpart, as the dependence on the downstream coordinate x̂ is removed as
β → 0.

3.3 Discussion
From the nondimensionalization of the system, we have calculated the vortex line
density required to produce our forcing magnitude fmax . For a flow with a length
scale of 1cm, a peak velocity of 1cm/s, at 1.90o K the required superfluid vortex
line density is approximately 35cm−2 . This is a relatively modest line density.
190 Simon P. Godfrey, David C. Samuels, and Carlo F. Barenghi

Fig. 4. Contour plot of the streamfunction Φ of the forced modes. (a) the upper mode,
β = 2.00, (b) the lower mode β = 0. The Reynolds number is 20,000 with the forcing
parameters as in Fig. 3.

Above a critical forcing magnitude, a new unstable mode appears, which for
sufficient forcing can dominate the classical instability. This new instability has
a much simpler geometry, and is not very sensitive to the width of the superfluid
vorticity distribution. We would speculate that the important (in terms of the
new instability) part of the mutual friction (3) is not the exact form of the
vorticity distribution, but that the mutual friction is proportional to Vn − Vs .
For these velocity profiles, this means that we have a strong shear effect in the
mutual friction term with the force changing sign at the position where this new
instability has the greatest effect.

References
1. D. C. Samuels, R. J. Donnelly: Phys. Rev. Lett. 65, 187 (1990)
2. R. J. Donnelly: Quantized Vortices in Helium II (Cambridge University Press,
1991)
3. D. C. Samuels: Phys. Rev. B, 46, (1992)
4. S. P. Godfrey, D. C. Samuels: Phys. Rev. B, 61, 6 (2000)
5. B. J. Cantwell: Ann. Rev. Fluid Mech., 13, (1981) C. J. Baker: J. Fluid. Mech., 95
(1979) H. Makita et al: AIAA J. 27 (1989) R. M. Kelso, A. J. Smits: Phys. Fluids,
7 (1995) G. R. Offen, S. J. Kline: J. Fluid Mech.,70 (1975)
6. M. R. Smith, D. K. Hilton, W. W. Van Sciver: Phys. Fluids, 11 (1999)
7. R. J. Donnelly, A. C. Hollis-Hallett: Can. J. Phys., 33 (1955)
8. S. P. Godfrey, D. C. Samuels, C. F. Barenghi: Phys. Fluids, 13 983 (2001)
9. P. G. Drazin, W. H. Reid: Hydrodynamic Stability (Cambridge University Press,
1984)
Grid Generated He II Turbulence
in a Finite Channel – Theoretical Interpretation

L. Skrbek and J.J. Niemela

Cryogenic Helium Turbulence Laboratory, Department of Physics,

University of Oregon, Eugene, OR 97403, USA

Abstract. Up to six orders of magnitude of He II vorticity decaying over three orders

of magnitude in time in the temperature range 1.2 K ≤ T ≤ 2 K can be described by
a purely classical spectral model for homogeneous and isotropic turbulence. The He II
vorticity is defined as ω = κL, where κ is the circulation quantum and L represents the
total length of the vortex line per unit volume, an experimental observable. The model
accounts for the quantum effects by introducing a temperature dependent effective
kinematic viscosity. In agreement with experimental observation, the spectral decay
model predicts four different regimes of the decay of vorticity in a finite channel - they
switch as the energy containing and dissipative Kolmogorov length scales grow during
the decay, finally both being saturated by the size of the channel.

1 Introduction
Studies of grid generated nearly homogeneous and isotropic turbulence (HIT)
and its decay belong to the most important and extensively explored problems
in “conventional” turbulence. Most of the experimental work is related to wind
tunnels, where the turbulence is studied as it decays downstream (e.g., [1]).
Turbulence without a mean flow, generated by using an oscillating grid and by
towing a grid through a stationary sample of fluid has also been studied (see [2]
and reference therein). Despite the long history of the subject, a general theory
describing the decay of turbulence based on first principles has not yet been
developed. However, experimental data containing information about the turbu-
lent energy spectra provide a solid foundation for a phenomenological approach
first outlined in [1,4], as the generally accepted forms of the three-dimensional
turbulent energy spectra uniquely determine the temporal decay of turbulence.
We have generalized this approach taking into account the finite size of the tur-
bulence box, intermittency and viscosity effects and discuss the final period of
decay [2,3]. In particular, we show that distinctly different decay regimes exist
due to the physical restriction that eddies larger than the size of the turbulent
box cannot exist[2,7].
It is interesting to compare the decay of grid generated classical and “su-
perfluid” turbulence, which consists of a dense tangle of quantized vortex lines
and normal fluid eddies. Here by “superfluid” we mean the turbulence generated
by towing a grid with a velocity vg through a stationary sample of He II in a
channel, as described in the companion article in this book. We stress that our
analysis is not applicable to a counterflow turbulence created by applying a heat
pulse to He II.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 191–197, 2001.

c Springer-Verlag Berlin Heidelberg 2001
192 L. Skrbek and J.J. Niemela
4
1 0
2 0 0 0 0 0
1 5 0 0 0 0
3
1 0 1 0 0 0 0 0 1 .3 K
5 0 0 0 0
2
1 0 2 5 0 0 0

1
1 0 1 0 0 0 0
(1 /s )

5 0 0 0
0
1 0
ω

5 0
2 0 0 0
)

-1
1 /1 0

1 0
(s
1 1 /1 0

4 0
-2
1 0
ω τ

0 .1 τ (s ) 1
-3
1 0
0 .0 1 0 .1 1 1 0
τ (s )

Fig. 1. The decaying He II vorticity measured at T = 1.3 K for indicated ReM . Each
curve represents an average of three individual decays. As the decay curves tend to
collapse on the universal curve, for clarity we shifted them subsequently by factor of
two downwards, the uppermost remaining unchanged. The virtual origin time tvo1
corresponds to a virtual origin position 3.5M downstream in a classical wind tunnel.
The inset shows the early part of the normalized vorticity decay data, displaying power
law with exponent −11/10. After saturation, typically several orders of magnitude of
decaying vorticity follow the power law with exponent −3/2, represented by the thick
solid line.

In the framework of the phenomenological two-fluid model He II is described

as consisting of the inviscid superfluid and the viscous normal fluid. In turbulent
flow the presence of quantized vortices couples the two fluids together, via mutual
friction[5]. Based on recent theoretical and experimental investigations[3,5,7], in
the temperature range covered in the present study, 1.2 K≤ T ≤ 2 K, turbulent
He II flow resembles classical flow possessing an effective kinematic viscosity νef f
of order μ/ρ, where μ is the dynamic viscosity of the normal component and ρ the
total density. In particular, the usual HIT relationship between turbulent energy
dissipation per unit volume and the mean square vorticity applies (limitations
of applicability of this equation are discussed by Vinen[5]):

−dE/dt = ε = νef f κ2 L2 = νef f ω 2 (1)

Here the vorticity is defined, in analogy with a rotating bucket of He II, as

ω = κL, where κ is the quantum of circulation and L is the length of quantized
vortex line per unit volume, obtained from the experiment. At present it is not
entirely clear how closely this definition of ω corresponds to the rms vorticity in
classical turbulence. Discussing this problem involves careful considerations of
the nature of He II turbulence arising from the quantized circulation in superfluid
and details of the coupling via mutual friction and would bring us beyond the
scope of this paper. Here we emphasize the similarity between classical and
 Grid Generated He II Turbulence in a Finite Channel 193

1 0 0 0
T = 1 .3 K
5
R e M
= 1 .5 x 1 0
4
R e M
= 2 .5 x 1 0
3
R e M
= 5 x 1 0

)
-1 /6
(s
5 /6
ωx (t+ tv o 1 ) 1 0 0

1 0
0 .1 1 1 0
t+ tv o 1 (s )

Fig. 2. Parts of the decaying He II vorticity multiplied by (t + tvo1)5/6 measured at

T = 1.3 K for the indicated ReM . The plateau indicates the −5/6 power law, the
second regime of the decay.

superfluid turbulence and show that considering them in parallel leads to a

better understanding of the phenomenon of a fluid turbulence in general.

2 The Spectral Decay Model

Let us introduce a purely classical phenomenological spectral model of decaying
HIT which predicts and quantitatively describes the data of the temporal decay
of the quantized vortex line density obtained in the experiment where He II
turbulence is created by towing a grid through a stationary sample of He II
(for experimental details, see the companion article in this book). The model
treats He II as a classical liquid possessing the temperature dependent effective
kinematic viscosity.
The model assumes that at early times the decaying grid turbulence displays
the generally accepted form of the 3D spectrum for HIT

E(k) = 0; k ≤ kd = 2π/d (2)

E(k) = Ak m ; m = 2; 2π/d ≤ k ≤ k1 (t) (3)

E(k) = Cε2/3 k −5/3 ; k2 (t) ≤ k ≤ γ(ε/νef

3
f)
1/4
= 2πηef f (4)

E(k) = 0; k ≥ 2πηef f (5)

which reflects a physical restriction that eddies larger than the width of the
channel, d, cannot exist and neglects intermittency. Also, the high wave number
exponential tail of the spectrum is approximated by a sharp cutoff at the effective
194 L. Skrbek and J.J. Niemela
4
1 0
5
R e = 2 x 1 0
T = 1 .7 5 K M
3 5
1 0 R e M
= 1 0
4
R e M
= 6 x 1 0
2
1 0 4
R e M
= 4 x 1 0
4
1 R e = 2 x 1 0
(1 /s )

M
1 0 2 0 0 4
R e = 1 0
)
1 /2

M
ω

(s

0
1 0
3 /2

1 0 0
x τ

-1
ω

1 0
0
1 1 0 1 0 0
τ (s )
-2
1 0
0 .1 1 1 0 1 0 0
tim e ( s )
Fig. 3. Vorticity decay curves measured for several indicated ReM collapse on the
universal decay curve, represented by the thick solid line - a plot of formula (8). The
inset shows the normalized data, the plateau indicates the −3/2 power law.

Kolmogorov length scale, ηef f , by introducing the dimensionless factor γ of

order unity. We assume the 3D Kolmogorov constant C = 1.62 ± 0.17, based
on a number of classical experiments[6]. In the vicinity of the energy containing
length scale e = 2π/ke (t), where k1 (t) ≤ ke (t) ≤ k2 (t), the spectral energy
density displays a smooth broad maximum whose analytical form is not specified.
Evaluating the total turbulent energy by integrating the 3D energy spectrum
over all k leads to a differential equation for decaying turbulent energy. Applying
ε = νef f ω 2 leads to a differential equation for decaying vorticity (for details of
calculation including intermittency corrections, see[2]). At the early decay the
spectrum stays self-similar. For d >> e >> ηef f , the energy decay is predicted
to follow[2,7]
m+1 2 3m+5
E(t + tvo1 ) = E(τ ) ∝ τ −2 m+3 ; e ∝ τ m+3 ; ω ∝ τ 2m+6 (6)

Comparison with both the wind tunnel[1,2] and He II data[2] suggests the
virtual origin position within few mesh units downstream the grid. Assuming
validity of the Saffman invariant[4] (m=2) we obtain ω ∝ τ −11/10 , i.e., the first
regime of the decaying vorticity, as illustrated in Fig. 1. Experimentally, the
power law is very sensitive to the exact value of virtual origin time. Due to the
geometry of our apparatus, the parameter space for observing this first regime
of the decay is rather limited, but clearly observable[2], see the inset in Fig.1.
As the turbulence decays further and e grows, the lowest physically signifi-
cant wave number becomes closer to the broad maximum around 2π/e . The low
wavenumber part of the spectrum can no longer be approximated as Ak m with
 Grid Generated He II Turbulence in a Finite Channel 195

1 0 1 0 0 0

1 0 0

(1 /s )
(s )
1

s a t
1 0
a t
ts

ω
T = 1 .3 K
T = 1 .6 5 K 1
T = 1 .9 5 K
0 .1
3 4 5
1 0 1 0 1 0
R e
Fig. 4. The saturation time (open symbols, left axis) and corresponding vorticity val-
ues (filled symbpols, right axis) versus ReM at different temperatures. The solid line
represents 3.3 × 104 /ReM .

m = 2. Instead, it can be characterized by an effective power that decreases as

the turbulence decays, such that 0 ≤ m < 2. Formula (6) then shows that the
decay rate slows down. As e approaches d, m becomes effectively zero and we
arrive at the second regime of the decay characterized by ω ∝ τ −5/6 - see Fig.2.
Note that these simplified power law arguments do not hold for the decay of the
turbulent energy as discussed in[2].
At the saturation time, tsat , the vorticity reaches its saturation value, ωsat ,
and the growth of e is completed. Still neglecting the cutoff of the spectral
energy at ηef f , the further decay - the third regime - is predicted universal:
√
27d C 3 −3/2
ω(τ ) = τ (7)
2π νef f

with the virtual origin time tvo2 . Therefore, no matter what value ReM is (pro-
viding it is high enough to neglect viscosity corrections[2]), in the finite size box
the decaying turbulence ought to reach this universal third regime of the de-
cay. Note that the virtual origin time tvo2 generally differs from tvo1 introduced
above[2], but our analysis shows that the difference is small and in our further
discussion of the decay we therefore assume tvo2 ∼ = tvo1 ∼
= 3.5M/vg .
We used individual decay curves to define tsat and ωsat as an intersection
point of the power laws ω ∝ τ −5/6 and ω ∝ τ −3/2 superimposed on the decay
data. The result of this fitting procedure is summarized in Fig.4. We found tsat ∝
1/ReM and values of tsat and ωsat hardly dependent on temperature, although
the normal fluid density changes over an order of magnitude. It strongly suggests
that the role of quantum effects in He II turbulence in this temperature range
can be at least approximately accounted for by introducing νef f and justifies
applicability of a purely classical model for the decay.
196 L. Skrbek and J.J. Niemela

T = 1 .6 5 K 5
R e M
= 2 x 1 0
1 0
5
R e M
= 1 0
4
(1 /s ) R e M
= 2 .5 x 1 0
4
1 R e M
= 1 0
ω

0 .1

0 2 0 4 0 6 0 8 0 1 0 0
tim e ( s )

Fig. 5. The late part of the decay data measured at indicated ReM . The thick solid line
represents a plot of formula (8) with parameters C = 1.62, νef f = 2.28 × 10−4 cm2 /s,
γ = 0.418. The late decay can be characterized as exponential, of a form exp(t/t0 ) with
t0 = 29 s, represented by a dashed line.

So far in our discussion we neglected a role of the high wave number cutoff of
the energy spectrum at ηef f . As the turbulent energy (or vorticity) decays and
the Kolmogorov length scale grows, the relative importance of this cutoff grows
and a simple power law cannot any longer describe the decay of vorticity. It is
possible to show[2] that after saturation the universal decay of vorticity can be
more accurately described by
 2  3/2
33 νef f 2π tB
ω(τ ) = cos3 θ (8)
23 γ d τ
4/3  
where cos2 (3θ) = τ /τB and tB ∼= 16Cγ
9νef f
d 2
2π . Naturally, for γ → ∞ expression
(8) reduces to a simple power law (7). Formally, as the vorticity decays, ηef f
becomes the size of the channel and the turbulent energy vanishes. Close to
this stage the applicability of the spectral model is no longer justified. From
Fig.4 we can estimate tB of order 100 s and γ ∼ = 0.4. This matches the value
of γ needed to describe the viscous corrections in classical wind tunnel data[1].
Expression (8) is used for comparison with experimental data in Figs. 4 and 5.
It describes the universal decay - up to five orders of magnitude of decaying He
II vorticity - measured for all ReM at any temperature. The departure from a
single power law with exponent -3/2 illustrates the increasing influence of the
growing Kolmogorov length scale in the decay.
Experimentally we have found a slight increase of the experimental data from
the theoretical line predicted by the model towards higher values of vorticity (see
Figs. 4 and 5). There might be several reasons for this behavior. First, it might
occur due to extra production of vorticity by counterflow in the channel. The
turbulence created by the towed grid decays and, as a result, the temperature
inside the channel grows, stimulating counterflow inside the channel due to the
 Grid Generated He II Turbulence in a Finite Channel 197

fountain effect. The faster the grid is pulled, the more extra vorticity is produced
by counterflow. The extra heat input could also result from friction between the
grid and the channel walls.
It is possible, however, that there is a deeper physical reason for this behavior
originating from the quantum nature of the He II turbulence. The quantum
effects were taken into account simply by introducing νef f , roughly a factor of
2-3 higher than the kinematic viscosity based on total fluid density, and νef f
may depend on the Reynolds number.
As the Kolmogorov scale grows, it gradually approaches d and thus e which
is already saturated by it. Expression (8) describes the experimental data down
to surprisingly low level of vorticity, of order ω ∼ = 0.1 s−1 , corresponding to
∼
a vortex line density L = ω/κ = 100 cm −2
and to a mean distance between
quantized vortex lines about 1 mm. Still, the essentially classical description of
the decaying vorticity holds.
With no inertial scale left there is no energy transfer towards higher wavenum-
bers and the only possibility for further decay is the exponential viscous decay.
This is the fourth and last regime of decaying vorticity in a finite channel. Note
that it differs from the final period of decay observed in classical wind tunnels, as
these can be essentially regarded as infinitely large[2]. The last regime is evident
from Fig.5, where the late decay curves originating from various ReM display
an exponential decay of the form ω(t) ∝ exp(−t/t0 ), practically indistinguish-
able from the spectral model prediction. Performing systematic measurements
for various ReM at T = 1.3 K, 1.65 K and 1.9 K, we obtained t0 = (27 ± 6) s.
This last decay regime can be considered in analogy with the decay of the
oscillatory motion in viscous fluids, characterized by exponential decay of the en-
ergy E = E0 exp(−βt), where the decay coefficient β = 2νk 2 . For k ∼ = 2π/d and
ν∼ = 10−4 cm2 s−1 it suggests a characteristic decay time close to the observed
one.

Acknowledgements

We acknowledge stimulating discussions with R.J. Donnelly, G.L. Eyink, D.

Holm, K. R. Sreenivasan S.R. Stalp and W.F. Vinen. This research was sup-
ported by NSF under grant DMR-9529609.

References
1. G. Comte-Bellot, S. Corrsin: J. Fluid Mech. 25, 657 (1966); 48, 273 (1971)
2. L. Skrbek, S.R. Stalp: Phys. Fluids 12, 1997 (2000)
3. S.R. Stalp, L. Skrbek, R.J. Donnelly: Phys. Rev. Lett. 82, 4831 (1999)
4. P.G. Saffman: J. Fluid Mech. 27, 581 (1967); Phys. Fluids 10, 1349 (1967)
5. W.F. Vinen: Phys. Rev. B 61, 1410 (2000)
6. K.R. Sreenivasan: Phys. Fluids 7, 2778 (1995)
7. L. Skrbek, J.J. Niemela, R.J. Donnelly: Phys. Rev. Lett. 85, 2973, (2000)
Vortex Tangle Dynamics
Without Mutual Friction in Superfluid 4 He

Makoto Tsubota1 , Tsunehiko Araki1 , and Sergey K. Nemirovskii2

1
Department of Physics, Osaka City University, Osaka 558-8585, Japan
2
Institute of Thermophysics, Academy of Science, Novosibirsk 630090, Russia

1 Introduction
Recently Davis et al. observed the free decay of the vortices at mK temperatures
where the normal fluid density became vanishingly small and the mutual friction
did not work effectively [1]. It is unclear how the vortices decay. Motivated by
this experimental work, we studied numerically the vortex dynamics without
the mutual friction, thus finding some cascade process which was obscured by
the normal fluid at higher temperatures. This paper reviews our recent works on
this problem. The numerical procedure based on the vortex filament formulation
is described in detail in Ref. [2]. Section 2 describes the dynamics of waves
excited along the reconnected vortex lines under the full Biot-Savart law and the
energy spectrum characteristic of the cascade process [3]. Section 3 studies the
dynamics of a dense vortex tangle(VT) by the calculation under the localized
induction approximation(LIA) [2]. The absence of the mutual friction makes
the vortices kinked, which promotes vortex reconnections. Consequently small
vortices are cut off from a large one through the reconnections. The resulting
vortices also follow the self-similar process to break up to smaller ones. Although
our formulation cannot describe the final destiny of the minimum vortex, the
decay of the VT is found to be connected with this cascade process, which is
just the cascade process at zero temperature Feynman proposed [4]. It should
be noted that this cascade process in a VT includes not only the breaking up of
vortices but also the vortex wave process described in Sect. 2.

2 Vortex Wave Cascade Process

Recently Vinen discussed the acoustic emission from an oscillating vortex at very
low temperatures [5]. The scenario is the following. Reconnection of two vortices
leave sharp kinks on them. The kinks, propagating along the vortex lines, are
evolved to the vortex waves whose wavenumbers are much larger than the inverse
of the average vortex spacing . Eventually the vortex waves with wavenumber
larger than a critical value are strongly damped by the acoustic emission. The
scenario except for the final stage can be confirmed by our formulation.
As a typical example, we calculated the collision of a straight vortex line
and a moving ring by the full Biot-Savart law. Figure 1(a) shows the initial
configuration of vortex lines. Toward the reconnection, the ring and the line

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 198–204, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Vortex Tangle Dynamics Without Mutual Friction 199

(a) (b) (c)

(d) (e) (f)

Fig. 1. Collision of a straight vortex and a ring at t=0(a), t=20(b), t=21(c), t=22(d),
t=24(e) and t=26(f). At t=20.4, two vortices reconnect.

twist themselves so that they become locally antiparallel at the closest place
(Fig.1(b)). At t 20.4, two vortices reconnect. After the reconnection (Fig.1(c),
(d) and (e)), the resulting local cusps broaden while exciting vortex waves with
various wavenumbers (Fig.1(f)).
We will introduce the energy spectrum of the velocity field v made by the
vortices [6]. The kinetic energy can be defined as the integral of the square of a
field: 
1 √
Ekin = 3
d3 x( ρv)2 , (1)
2(2π)
where ρ is the density of fluid. The energy spectrum Ekin (k) is defined as
∞
Ekin = 0 dkEkin (k). Using the Parseval’s theorem, one gets the following en-
ergy spectrum:
   2
1  1 3 ir · k √

Ekin (k) = 
dΩk  d re ρv  , (2)
2 (2π)3

where dΩk denotes the volume element k 2 sin θdθdφ in the spherical coordinates.
The energy spectrum Ekin (k) represents the contribution from the velocity field
with wave number k to the kinetic energy.
Figure 2 shows the energy spectrum in the process of Fig.1. Before the
reconnection(t ≤ 20.4), the energy spectrum is almost constant in time. After the
reconnection(t ≥ 20.4), the energy spectrum begins to fluctuate suddenly. The
waves generated by the cusps evolve chaotically to waves with other wavenum-
ber by the nonlinear interaction; the reconnection process results in the ergodic
energy distribution of vortex waves. This is consistent with the study of the side-
band instability by Samuels and Donnelly [7]. Figure 2 shows that some energy
peaks move to small k region. Compared with Fig.1, this behavior may reflect
the broadening of the local cusps and the ring’s leaving the line. The total kinetic
200 M. Tsubota, T. Araki, and S.K. Nemirovskii

Fig. 2. Time evolution of the energy spectrum E(k) before and after the reconnection.

energy is conserved despite the mode mixing. If some dissipative mechanism is

introduced in the k region above a critical wave number, as in the Vinen’s the-
ory [5], the kinetic energy is transferred from small k region to large one, being
dissipated at the critical value, so that the total energy decays. This is consistent
with Vinen’s theory.

3 Cascade Process in the Vortex Tangle

3.1 Decay of the Vortex Tangle
This section studies the free decay of the dense VT without mutual friction
under the LIA. The VT in a 1cm3 cube is calculated with the space resolution
Δξ = 1.83 × 10−2 cm and the time resolution Δt = 1.0 × 10−3 sec. The initial
VT for free decay is prepared by the development of six vortex rings subject
to thermal counterflow and mutual friction [8]. After turning off the thermal
counterflow, we follow the dynamics of the VT. Figure 3 shows transient VTs
with and without the mutual friction. The difference is marked.
The VT subject to the mutual friction consists of relatively smooth vortex
lines, while the absence of the mutual friction makes the vortices kinked, which
promotes vortex reconnections [2]. The small vortices are separated from a large
one through the reconnections. The resulting vortices also follow the self-similar
cascade process to break up to smaller ones. Our numerical calculation cannot
follow the dynamics of vortices smaller than its space resolution; such vortices
are eliminated numerically and its justification will be discussed later.
202 M. Tsubota, T. Araki, and S.K. Nemirovskii

decay rate is almost independent of the space resolution when it is in the range
between 4Δξ and Δξ/4.
It is possible to classify the kinds of reconnection in the VT dynamics. The
vortex reconnection is divided topologically into three classes. The first refers to
the process whereby two vortices reconnect to two vortices, which is most usual.
The second is the process which divides one vortex into two vortices (the split
type); the cascade process is driven by this kind of reconnection. Third is the
process whereby two vortices are combined to one vortex against the cascade
process (the combination type). Investigating the number of reconnection events
in the VT dynamics of Fig. 4, we find that most reconnections belong to the
first class but the second split type occupies about 17% of the total reconnec-
tions, being superior to that of the third combination type of about 10%. The
reconnection of the split type actually promotes the cascade process, against the
reverse process due to that of the combination type.
The cascade process is revealed further by investigating the size distribution
of vortices. Figure 5 shows the change of the size distribution in the VT dynam-
ics of Fig. 4. Each figure shows the number n(x) of vortices as a function of their
length x. The system size a(=1cm) and the space resolution Δξ(= 1.83×10−2 cm)
are the characteristic scales in this system. The vortices longer than a are orig-
inally few, and most vortices are concentrated in the scale range [Δξ, a]. As
the cascade process progresses, every vortex generally divides into smaller ones
through the split type reconnections, although some combination type recon-
nections may occur. As a result, the vortices between Δξ and a are decreased
in number because they become smaller than Δξ and eliminated. Although the
vortices larger than a become few too, some of them survive. Such vortices ex-
tend over the whole system and have small self-induced velocity, being almost
straight. Of course they may happen to reconnect with other vortices, then they
can be divided to smaller vortices and follow the cascade process. However their
small self-induced velocity makes their reconnections uncommon, thus stabilizing
them. Hence the decay of the vortex line density (VLD) is attributable mainly
to the cascade process of the vortices smaller than a.
The final destiny of small vortices through the cascade process may be inter-
preted several ways. First, the vortices can vanish at a small scale by radiating
phonons, which is discussed recently by Vinen [5]. Secondly, the vortices whose
size is eventually reduced to the order of the interatomic distance no longer sus-
tain the vortex state, probably changing into such short-wavelength excitation
as roton. Since both mechanisms work only at a small scale, some process that
transfers energy from a large scale to smaller scales is necessary for the decay
of the VT; this is just the cascade process. Thirdly, in a real system, the small
vortices may collide with the vessel walls [2]. Since only the vortices in the bulk
are observed experimentally, the reconnection with the walls may reduce the
observed VLD effectively.
 Vortex Tangle Dynamics Without Mutual Friction 203

Fig. 5. Bar chart showing the number of vortices n(x) as a function of the length x.
The time is t=0(a), 50(b) and 100(c).

3.2 Comparison with the Vinen’s Equation

Considering that cascade process at zero temperature proposed by Feynman [4],

Vinen obtained an evolution equation for the VLD L(t), what we call the Vinen’s
equation [9]
dL κ
= −χ2 L2 , (3)
dt 2π
where χ2 is a parameter and κ the quantized circulation. Its solution is given by
1 1 κ
= + χ2 t, (4)
L L0 2π
204 M. Tsubota, T. Araki, and S.K. Nemirovskii

Fig. 6. Comparison of the decay of L(t) and the solution of the Vinen’s equation.

where L0 is the VLD at t = 0.

Figure 6 shows the comparison of our numerical results and the solution of
the Vinen’s equation. The solid line refers to our result for the VT decay of Fig.4,
while three other lines denote Eq.(4) with the parameters χ2 = 0.5, 0.3, 0.2.
Our result agrees excellently with the solution of χ2 = 0.3. Since the Vinen’s
equation is based closely on the cascade process, this agreement supports that
the cascade process occurs really in the numerical simulation. The values of χ2
obtained at finite temperatures [5] are extrapolated towards zero temperature,
then seeming to reach reasonably to χ2 0.3; the value obtained numerically
may be consistent quantitatively.

Acknowledgements

We acknowledge W.F. Vinen and P.V.E. McClintock for useful discussions.

References
1. [Link], [Link], [Link]: Physica B 280, 43(2000)
2. [Link], [Link], [Link]: Phy. Rev. B Vol.62, Num.17(2000)
3. [Link], [Link]: J. Low Temp. Phys.(to be published)
4. [Link]: ‘Application of Quantum Mechanics to Liquid Helium’. In: Progress
in Low Temperature Physics Vol. 1. ed. by [Link](North-Holland, Ameterdam,
1955) pp. 17–53
5. [Link]: Phy. Rev. B 61, 1410(2000)
6. [Link], [Link], [Link]: Phys. Fluids 9, 2644(1997)
7. [Link], [Link]: Phys. Rev. Lett. 64, 1385(1990)
8. [Link]: Phys. Rev. B 38, 2398(1988)
9. [Link], Proc. R. Soc. London A 242, 493(1957)
Applications of the Gaussian Model of the
Vortex Tangle in the Superfluid Turbulent He II

Sergey K. Nemirovskii and Mikhail V. Nedoboiko

Institute of Thermophysics, 630090 Novosibirsk, RUSSIA

Abstract. In spite of an appearance of some impressive recent results in understanding

of the superfluid turbulence in HeII they fail to evaluate many characteristics of vortex
tangle needed for both applications and fundamental study. Early we reported the
Gaussian model of the vortex tangle in superfluid turbulent HeII. That model is just
trial distribution functional in space of vortex loop configurations constructed on the
basis of well established properties of vortex tangle. It is designed to calculate various
averages taken over stochastic vortex loop configurations. In this paper we use this
model to calculate some important characteristics of the vortex tangle. In particular
we evaluate the average superfluid mass current J induced by vortices and the average
energy E associated with the chaotic vortex filament.

1 Introduction
The presence of vortex tangle appearing in the superfluid turbulent HeII essen-
tially changes hydrodynamic properties of the latter (see e.g.[1] ,[2],[3]). Accord-
ing phenomena are studied in frame of so called Phenomenological Theory (PT)
pioneered by Vinen [4] and greatly modified by Schwarz [5]. The PT describes
superfluid turbulence (ST) in terms of the total length of vortex lines (per unit of
volume) or the vortex line density (VLD) L(t) and of the structure parameters of
the VT. Knowledge of these quantities allows to calculate some of hydrodynamic
characteristics of superfluid turbulent HeII such as a mutual friction, sound at-
tenuation etc. Meanwhile there exist many other physical quantities connected
to distribution of the filaments and their interaction related with other physical
phenomena which can not be expressed in terms of the PT. The relevant phe-
nomena should be covered by appropriate stochastic theory of chaotic vortex
filaments. Of course, the most honest way to develop such theory is to study
stochastic dynamics of vortex filaments on the base of equations of motion with
some source of chaos. However due to extremely involved dynamics of vortex
lines this way seems to be almost hopeless. Thus, a necessity of a developing
an advanced phenomenological approach appeared. We offer one variant of such
approach. The main idea and the main strategy are the following. Although the
phenomenological theory of the superfluid turbulence deals with macroscopical
characteristics of the vortex tangle, it conveys the rich information concerning
the instantaneous structure of the vortex tangle. Namely we know that the VT
consists of the closed loops labelled by sj (ξ), uniformly distributed in space and
having the total length L(t) per unit of volume. From acoustical experiments
it follows that filaments are distributed in anisotropic manner and quantitative

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 205–211, 2001.

c Springer-Verlag Berlin Heidelberg 2001
206 S.K. Nemirovskii and M.V. Nedoboiko

characteristics of this anisotropy can be expressed by some structure parameters

(see [1], [3], [5]). Beside this usual anisotropy there is more subtle anisotropy
connected with averaged polarization of the vortex loops. Furthermore there are
some proofs that the averaged curvature of the vortex lines is proportional to the
inverse interline space and coefficient of this proportionality (which is of order
of unit) was obtained in numerical simulations made by Schwarz [4].
The master idea of our proposal is to construct a trial distribution function
(TDF) in the space of the vortex loops of the most general form which satisfies
to all of the properties of the VT introduced above. We assume that this trial
distribution function will enable us to calculate any physical quantities due to
the VT. In the paper we describe typical shape of vortex loop obtained from
evaluating of the correlation functions. We also calculate the average hydrody-
namic impulse (or Lamb impulse) JV in the counterflowing superfluid turbulent
HeII and the average kinetic energy E associated with the chaotic vortex loop.

2 Constructing the Trial Distribution Function

According to general prescriptions the average of any quantity B({sj (ξj )})
depending on vortex loop configurations is given by
+
B({sj (ξj )}) = B({sj (ξj )})P({sj (ξj )}). (1)
{sj (ξj )}

Here P({sj (ξj )}) is a probability of the vortex tangle to have a particular configu-
ration {sj (ξj )}. Index j distinguishes0different loops. The meaning of summation
over all vortex loop configurations {sj (ξj )} in formula (1) will be clear from
further presentation . We put the usual in the statistical physics supposition that
all configuration corresponding to the same macroscopic state have equal prob-
abilities. Thus the probability P({sj (ξj )}) for vortex tangle to have a particular
configuration {sj (ξj )} should be proportional to 1/Nallowed , where Nallowed
is the number of allowed configurations, of course infinite
1
P({sj (ξj )}) ∝ . (2)
Nallowed

Under term ”allowed configurations” Nallowed we mean only the configurations

that will lead to the correct values for all average quantities known from experi-
ment and numerical simulations. Formally it can be expressed as a path integral
in space of three-dimensional (closed) curves supplemented with some constrains
connected to properties of the VT.
1
Nallowed ∝ D{sj (ξ)} × constraints {sj (ξ)}. (3)
j

The constraints entering this relation are expressed by delta functions expressing
fixed properties of the VT. For instance constrain δ((s j (ξ))2 − 1) expresses that
 Applications of Gaussian Model of the Vortex Tangle 207

parameter ξ is the arc length. However this condition will lead to not tractable
theory. We will use a trick known from the theory of polymer chains (see e.g. [6]) ,
namely we will relax rigorous condition and change delta function by continuous
(Gaussian) distribution of the link length with the same value of integral. This
trick leads to the following expression for number of way:
1 L  2
Nallowed ∝ D{sj (ξ)} × e−λ1 0 |s | dξ . (4)
j

In the same manner we are able to introduce and treat other constrains con-
nected to the known properties of the VT structure. The detailed calculations
are exposed in paper of one of the author [7] , now we write down final expression
for probability of configurations

Nallowed ∝ D{s(κ)} exp (−L{s(κ)}) . (5)

Here s(κ) is one-dimensional Fourier transform of variable s(ξ) and Lagrangian

L{s(κ)} is a quadratic form of the components of the vector variable s(κ)
+ αβ
L{s(κ)} = sα
x (κ)Λ (κ)sβx (κ). (6)
κ=0

In practice to calculate various averages it is convenient to work with the char-

acteristic (generating) functional (CF) which is defined as a following average:
⎛ ⎞
++ μ
W ({Pj (κ)}) = exp ⎝ − Pj (κ)sj (−κ)⎠.
μ

j κ=0

Due to that our Lagrangian 6is a quadratic form (in s(κ)) and, consequently,
the trial distribution function is a Gaussian one, calculation of the CF can be
made by accomplishing the full square procedure to give a result
⎛ ⎞
++ μ
W ({Pj (κ)}) = exp ⎝− Pj (κ)Njμν (κ)Pνj (−κ)⎠ . (7)
j κ=0

Elements of matrix Njμν (κ) are specified from calculation of total length, anisotropy
coefficient, curvature and polarization. The explicit form of them is written down
in [7].
Thus we reached the put goal and have written the expression for trial CF
which, we repeat, enables us to calculate any averaged of the vortex filament
configuration. For instance calculating some of the correlation functions we are
able to describe a typical shape of the averaged curve. It is sketched out in Fig.1
208 S.K. Nemirovskii and M.V. Nedoboiko

Fig. 1. A snapshot of the averaged vortex loop obtained from analysis of the statistical
properties. Position of the vortex line element is described as sj (ξj ), where ξj is arc
length, sj (ξj ) = dsj (ξj )/dξj is a tangent vector, unit vector along the vortex line;
sj (ξj ) = d2 sj (ξj )/dξj2 is the local curvature vector; vector production sj (ξj ) × sj (ξj ) is
binormal which is responsible for mutual orientation of the tangent vector and vector
of curvature. Close (Δξ  R , where R is the mean curvature ) parts of the line are
separated in 3D space √ by distance Δξ. The distant part (R  Δξ) are separated in 3D
space by distance 2πRΔξ (with correction due to the closeness). The latter property
reflects a random walk structure of the vortex loops. As a whole the  loop is not isotropic
having a ”pancake” form. In addition it has a total polarization sj (ξj ) × sj (ξj )dξj
forcing the loop to drift along vector Vn and to produce nonzero superfluid mass
current in z-direction

3 Hydrodynamic Impulse of the Vortex Tangle

As an one more illustration to the developed theory we discuss hydrodynamic
impulse of the vortex tangle JV which is defined as

6+
ρs κ
JV =  sj (ξj ) × sj (ξj ) dξj  (8)
2 j

The quantity JV is closely related to momentum of fluid (see [8]). The averaged
< sj (ξj )×sj (ξj ) > is immediately evaluated by use of CF (7) to give the following
result:  
z ρ6
κIl αv
JV = − ρs V s (9)
ρn c22 βv
Note that the coefficient includes no fitting parameters but only characteristics
known from the Phenomenological Theory (see [5]). Relation (9) shows that the
vortex tangle induces the superfluid current directed against the external super-
fluid current. It should be expected since there is some preferable polarization
 Applications of Gaussian Model of the Vortex Tangle 209

of the vortex loops. In the experiments this additional superfluid current should
display itself as suppression of the superfluid density. This effect is 3D analog to
the famous Kosterlitz-Thoulless effect except of that distribution of the vortex
lines is not calculated but is obtained appealing to the experimental data.
Since superfluid density enters an expression for second sound velocity, it
seems attractive to detect it using transverse second sound testing. To do it we
have firstly to evaluate transverse change of the ρs and, secondly, to develop
the theory to match it to nonstationary case. The general theory asserts that
while applying a harmonic external second sound field suppression of superfluid
density becomes the function of frequency ω of the following form:
 x 
x δJV 1
Δρs (ω) = x
. (10)
δVs transv 1 + iωτJ

Here transverse (δJxV /δVsx )transv is half of the one given by rel. (9). The quantity
τJ is the time of relaxation of the superfluid current JV . which is to be found
from dynamical consideration. First, we have to derive dJxV /dt with help of
the equation of motion of the vortex line elements and, second, to evaluate
various averaged appearing in right-hand side. function we obtain the following
final result for change of the second sound velocity. Performing all of described
procedures one obtains that the relative change Δu2 /u2 of the second sound
velocity is given by
Δu2 V4
= −f (T ) ns . (11)
u2 ω2
Here the function f (T ) is composed of the structure parameters of the vortex
tangle
κIl2 α2 (1 − Ixx )2
4ρ6
f (T ) = . (12)
ρn c42 β 3
Decreasing of the second sound velocity in the counterflowing HeII has been re-
ally observed about two decades ago by Vidal with coauthors [9]. Let us compare
our result (11) with the Vidal’s experiment. Using the data on the structure pa-
rameters one obtains that e.g. for the temperatures 1.44K the value of function
f (T ) is about 620s2 /cm4 . Taking the frequency ω = 4.3 rad/s, used in [9], and
Vns = 2 cm/s one obtains that Δu2 /u2 ≈ 4× 10−4 , which is very close to the
observed value.

4 Energy of the Vortex Tangle

In this section we calculate the averaged energy of the stochastic vortex loop
distributed according trial distribution function (5) . The general expression for
the energy associated with linear vortices can be written as (see e.g. [8])

7  8 9 Li Lj :
1 2 3 ρs κ2 + si (ξi )sj (ξj )
E= ρs vs d r = dξi dξj . (13)
2 8π j,i |si (ξi ) − sj (ξj )|
0 0
210 S.K. Nemirovskii and M.V. Nedoboiko

In 3D Fourier space the average energy E (13) can be rewritten as

9  Li Lj :
ρs κ2 + d3 k   ik(si (ξi ) − sj (ξj ))
E= sj (ξi )sj (ξj )dξi dξj e . (14)
2 i,j (2π)3 k2
k 0 0

Comparing (14) and (7) it is possible to express the energy E in terms of the
characteristic Functional
 Li Lj
ρs κ2 + d3 k δ2 W
E = dξi dξj eik(si (0) − sj (0)) ×
3
2 i,j k (2π) k 2 iδPi (ξi ) iδPα
α
j (ξj )
0 0
(15)
Here set of Pn (ξn ) in CF W ({Pn (ξn )}) is again determined with help of the
θ-functions
Pi (ξi ) = kθ(ξi )θ(ξi − ξi ), Pj (ξj ) = kθ(ξj )θ(ξj − ξj ),
(16)
Pn (ξn ) = 0, n = i, j

The relation (16) implies that we have to choose in integrand in exponent of CF

only points lying in interval from 0 to ξi on i -curve and from 0 to ξj on j-curve.
While evaluation of self-energy of the same loop , i = j , one has to distinguish
points ξi , and to put them to be e.g. ξi and ξi . Further results concern the case
of the only loop of length L. Omitting tremendous calculations we write down
the final answer in the following form:
 
ρκ2 L R ρκ2 L 2 R
E= ln + 1 − √ (f2 − f1 ) ln (17)
4π a0 4π π a0

ρκ2 L 1 2f3 f2
+ √ 1/2 5/2 2
· Il +
2
√ 1/2
,
4π ( π − 1) π c2 3 π 3/2 ( π − 1)

where the quantities f (of order of unit)

 are expressed via the structure param-
eters of the VT as follows (belowβ = Ix − Iz /Ix )

f1 (β) = 2 (3 − β 2 )(arcsin (β)/β), (18)

    
f2 (β) = 1 − β 2 + 2 − β 2 arcsin (β)/β / 2 (3 − β 2 ), (19)

  3/2  
f3 (β) = 2 3 − β 2 1 − β 2 − (arcsin β)/β /β (20)

Let us comment expression (17). The first term in the right-hand side of (17)
is just the energy of unit of length of a straight vortex filament (see e.g. [1])
multiplied by its length. In this form it is frequently used in theory of superfluid
turbulence (see e.g. [3]) and in other applications. But there are additional terms.
The third and forth terms appeared from long-range interaction, they are smaller
 Applications of Gaussian Model of the Vortex Tangle 211

then logarithmic ones (about ten percents). The third term is of especial interest.
It appeared due to polarization of the vortex loop and its presence implies that
there is some elasticity of the vortex tangle in Vns direction. Results of the
previous section showed that the VT induces some additional superfluid flow.
Therefore one can expect that combination of longitudinal elasticity combining
with inertia of additional will lead to appearing of elastic waves, 3D analog of
the Tkachenko waves.
The second one is also logarithmically large. Logarithmic behaviour points
out that this contribution came from denominator |s(ξ) − s(ξ  )|. But it was the
first (local) term which collected contributions from neighbor points along the
line. Therefore the third term appeared from accidental self-crossing of remote
(along the line) parts of the vortex filament. The fact that this term is propor-
tional to L and is of of the first (local) contribution is due to that the line is
fractal object with Haussdorf Hd dimension equal Hd = 2. According to general
theory of fractal lines it has an infinite number of self-crossing with cardinal
number 2Hd − 3 i.e. it is equivalent to line.

5 Conclusion

We briefly exposed an essence of Gaussian model of the vortex tangle and give
several examples how it can be used for evaluation of important physical char-
acteristics such as induced momentum and energy of interaction. These charac-
teristics has been discussed early (see e.g. [1]), however their evaluation has not
been performed because of lack of a proper theory. We think that these illustra-
tions convince that Gaussian model can serve as effective tool to study chaotic
vortex filaments.
The work was partly funded by Russian Foundation for Basic Research, Grant
N 99-02-16942

References
1. R.J. Donnelly:Quantized Vortices in Helium II, (Cambridge University Press,
Cambridge 1991)
2. J.T. Tough: ’Superfluid Turbulence’. In: Progress in Low Temperature Physics,
VIII [Link] D.F. Brewer (North Holland, Amsterdam 1982), pp. 133-219
3. [Link], [Link]: Rev. Mod. Phys., 67 , 37 (1995)
4. W.F. Vinen:Proc. R. Soc. London A 243, 400(1958)
5. [Link]:, Phys. Rev. B38, 2398 (1988)
6. M. Doi and S.F. Edwards: The theory of polymer dynamics (Clarendon Press,
Oxford 1986)
7. S.K. Nemirovskii:Phys. Rev B57, 5792 (1997)
8. G.K. Batchelor, An introduction to fluid mechanics, (Cambridge University Press
1967)
9. F.C. Vidal:C.R. Acad Sci., B 275, , 609 (1972)
Stochastic Dynamics of a Vortex Loop.
Thermal Equilibrium

Sergey K. Nemirovskii1 , L.P. Kondaurova1 , and M. Tsubota2

1
Institute for Thermophysics, Lavrentyeva,1, 630090 Novosibirsk, Russia
2
Department of Physics, Osaka City University, Osaka, Japan

Abstract. We study stochastic behavior of a single vortex loop appeared in imperfect

Bose gas. Dynamics of Bose-condensate is supposed to obey Gross-Pitaevskii equa-
tion with additional noise satisfying fluctuation-dissipation relation. The correspond-
ing Fokker-Planck equation for probability functional has a solution P({ψ(r)}) =
N exp(−H {ψ(r)} /T ), where H {ψ(r)} is a Ginzburg-Landau free energy. Consider-
ing a vortex filaments as a topological defects of the field ψ(r) we derive a Langevin-
type equation of motion of the line with correspondingly transformed stirring force.
The respective Fokker-Planck equation for probability functional P({s(ξ)}) in vor-
tex loop configuration space is shown to have a solution of the form P({s(ξ)}) =
N exp(−H {s} /T ), where N is a normalizing factor and H {s} is energy of vortex line
configurations. In other words a thermal equilibrium of Bose-condensate results in a
thermal equilibrium of vortex loops appeared in Bose-condensate. Some consequences
of that fact and possible violations are discussed.

1 Introduction and Scientific Background

Quantized vortices appeared in quantum fluids have been an object of inten-

sive study for many years (see for review and bibliography the famous book
by Donnelly[1]). The greatest success in investigations of dynamics of quantized
vortices has been achieved in relatively simple cases such as vortices in rotating
helium (where they form a vortex array orientated along an axis of rotation) or
vortex rings. However these simple cases are rather exception than a rule. Due
to extremely involved dynamics initially straight lines or rings evolve to form
highly entangled chaotic structure. Thus a necessity of use of statistic methods
to describe chaotic vortex loop configurations arises. A most tempting way is
to treat ”gas” vortices as a kind of excitation and to use thermodynamic meth-
ods. One of first examples of that way was an use of the Landau criterium for
critical velocity where vortex energy and momentum were applied to relation
having pure thermodynamic sense. More extended examples would be the fa-
mous Kosterlitz-Thouless description of 2D vortices or its 3D variant intensively
elaborated currently (for review and bibliography see e.g.[2]).
In the examples above and in many other it is assumed that chaotic vortex
loop configurations are in a thermal equilibrium and their statistics obeys the
Gibbs distribution . That supposition is based on fundamental physical principles
and can be justified in a standard way considering vortex lops as a subsystem
submerged into a thermostat and exchanging energy with the latter. The role of

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 212–218, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Stochastic Dynamics of a Vortex Loop. Thermal Equilibrium 213

thermostat in the case of vortices is played by the other excitations (phonons and
rotons) of an underlying physical field. In the case of Bose-Einstein condensate
(BEC), which we consider in this paper, that field is an order parameter ψ(r,t).
Vortex lines are just the crossings of surfaces where both real and imaginary part
of ψ(r,t) vanish. Excitations of order parameter (phonons and rotons) interact
with vortices driving the latter to statistic distribution which in accordance with
general principles should be the Gibbs distribution.
It is well known however that the Gibbs distribution can be alternatively
obtained in the frame of some reduced model like kinetic equations or Fokker–
Planck equation. That way of course is not of such great generality as a principle
of maximum entropy, but instead it allows us to clarify the mechanisms of how
the Gibbs distribution is established and to discuss possible deviations and vi-
olations of equilibrium state. We choose that way to examine how a thermal
equilibrium of chaotic vortex loops follows from a thermal equilibrium of BEC.

2 Langevin Equation
We perform our consideration on the basis of the Gross–Pitaevskii model[3],[4].
To describe stochastic behavior of BEC let us add to the right–hand side of the
Gross–Pitaevskii equation the random stirring force ζ(x, t).
 ∂ψ δH(ψ)
= − (Λ + i) + ζ(x, t). (1)
m ∂t δψ ∗
Here H {ψ}-is the Ginzburg–Landau free energy functional
  2 
 2 μ 2 V0 4
H {ψ} = d3 x |∇ψ| − |ψ| + |ψ| . (2)
2m2 m 2m
The thermal noise obeys the following fluctuation–dissipation theorem
2kB T Λ
ζ(x1 , t1 )ζ ∗ (x2 , t2 ) = δ(x1 − x2 ) δ(t1 − t2 ) . (3)
(m/)
The stochastic problem introduced by equations (1)–(3) has a solution de-
scribing thermal equilibrium, where probability of some configuration of ψ–filed
is proportional to exp(−H {ψ(r)} /T ). This readily follows from the correspon-
dent Fokker–Planck equation[5].
Let’s go on to the problem of obtaining of the vortex dynamics appear-
ing in BEC, whose own dynamics obeys relations (1)–(3). That problem, as
well as more general problems of that kind has been considered many times by
many methods[6] starting from pioneering papers by Pitaevskii [7] and Fetter
[8]. We develop one more method, neither too rigorous nor principally new but
convenient for purpose of this paper. Details of that method will be published
elsewhere, here we briefly describe it to an extent to comprehend the put goal.
As any motion of topological defects the vortex line dynamics is determined
by the one of the underlying field theory. On the other hand if one ignores a pres-
ence of other excitations except of the vortices one can say that all motion of
214 S.K. Nemirovskii, L.P. Kondaurova, and M. Tsubota

BEC is determined by the one of quantum vortices. Thus there is mutual corre-
spondence and the order parameter ψ(x, t) can be considered as some functional
of a whole vortex loop configuration ψ(x | {s(ξ, t)}). Temporal dependance of
field ψ is connected with motion of lines and its rate of change (at some point
x) is expressed by the following chain rule:

∂ψ(x, t) δψ(x | {s(ξ, t)}) ∂s(ξ  , t) 
= dξ . (4)
∂t δs(ξ  , t) ∂t
Γ

Vortices in quantum fluids are very slender tubes (except of vicinity of phase
transition) and their dynamics is analogous to the one of the strings. That implies
that we have to aim our efforts at integrating out radial degrees of freedom. It
can be reached by the following procedure. Let us further multiply equation(1)
1
by −Λ+i δψ ∗ /δs(ξ0 , t), where ξ0 some chosen point on the curve. Combining the
result obtained with the complex conjugate and integrating over whole space we
have
  
 3 Λ − i ∂ψ δψ ∗ Λ + i ∂ψ ∗ δψ
d x + =
m Λ2 + 1 ∂t δs(ξ0 , t) Λ2 + 1 ∂t δs(ξ0 , t)
  
δH(ψ, ψ ∗ ) δψ ∗ δH(ψ, ψ ∗ ) δψ
− d3 x + + (5)
δψ ∗ δs(ξ0 , t) δψ ∗ δs(ξ0 , t)
  
Λ−i δψ ∗ Λ+i ∗ δψ
d3 x 2
ζ(x, t) + 2 ζ (x, t) .
Λ +1 δs(ξ0 , t) Λ + 1 δs(ξ0 , t)
The first integral in the right–hand side of (5) expresses a chain rule for
functional derivative δH(s)/δs(ξ0 , t) where H(s) is the energy of moving BEC
expressed via vortex line position. Consequently considering them to be very
slender tubes (which is justified when the radius of the curvature R is much
larger of the core size r0 ) and neglecting an energy associated with the core, the
quantity H(s) is just kinetic energy of the superfluid flow created by vortices
(see e.g. [9],[10])
˜2  
ρs κ s (ξ)s (ξ)
H(s) = dξdξ  . (6)
8π |s(ξ) − s(ξ  )|
Γ Γ 


Here s (ξ) is tangent vector, double integration is performed along a whole
˜
line, κ is a quantum of circulation equal 2π/m. Calculation of functional deriva-
tive
δH(s)/δs(ξ0 , t)
is straightforward and leads to result
δH(s) ˜
= ρs κs (ξ0 ) × B(ξ0 ). (7)
δs(ξ0 , t)
·
Quantity B(ξ0 ) is the velocity of the line element s ( ξ 0 ) expressed by well
known Biot–Savart law. The terms in the left–hand side of equation (5) (first
 Stochastic Dynamics of a Vortex Loop. Thermal Equilibrium 215

line) can be evaluated in general form by observing that the major contribution
into integrals appears from vicinity of the vortex filament (see e.g. [11]). Thus
to evaluate integral we replace ψ(x | s(ξ, t)) by ψv (x⊥ ) = ψv (s(ξcl , t) − x) where
ψv is well studied 2D vortex and integration over d3 x by d2 x⊥ dξcl . Functional
derivative δψ ∗ /δs(ξ  , t) should be evaluated by a following rule: δψ ∗ /δs(ξ  , t) =
∇⊥ ψv (x⊥ )δ(ξ  − ξcl ). Here ξcl is the label of point of the line closest to point x.
Using the said above and calculating integrals of squared gradients of ψv (x⊥ )
we conclude that left–hand side of equation (5) transforms into
 2πρs ·  2πρs σΛ ·
s ( ξ 0 ) × s (ξ0 ) + s ( ξ 0 ). (8)
m Λ2 + 1 m Λ2 + 1
Let us now discuss the rest terms of equation (5) including random force
ζ(x, t) (the third line). Consequently considering that the all motion of BEC is
connected to motion of line, we have to consider Langevin force ζ(x, t) as some
secondary quantity stemming from random displacements of filaments. Con-
nection between displacements (random) of filaments δs and deviations (ran-
dom) of δψ(x, t) may be written in form similar to (4) with formal substitution
·
∂ψ(x, t)/∂t → δψ(x, t) and s→ δs. Taking into account that δψ(x, t) = ζ(x, t)δt,
and δs = ζ(ξ, t)δt we conclude that quantities ζ(x, t) and ζ(ξ, t) are connected
to each other by a chain rule

δψ(x | {s(ξ, t)})
ζ(x, t) = ζ(ξ  , t)dξ  . (9)
δs(ξ  , t)
Γ

That implies that to take into consideration random displacements of line

we have to change the last term in equation (5) by the one similar to (8) with
·
substitution s (xi0 ) → ζ(ξ0 , t). Gathering all terms we obtain a vector equation,
·
which can be resolved up to tangential velocity s ( ξ 0 ) along the curve. The
latter does not have any physical meaning and can be removed by suitable pa-
rameterization of the label variable ξ. Solving that vector equation we arrive at

· 1 + Λ2 (1 + Λ2 )Λσ 
s (xi0 ) = B(ξ0 ) + s (ξ0 ) × B(ξ0 ) + (m/)ζ(ξ0 , t). (10)
1 + Λ2 σ 2 1 + Λ2 σ 2
Equation (10) describes motion of vortex line in terms of the line itself. It is
remarkable fact (not obvious in advance that noise ζ(ξ0 , t) acting on line is also
additive (does not depend on line variables).
The last effort we have to do is to ascertain both the statistic properties
of noise ζ(ξ0 , t) and its intensity. Shortly, it can be done by comparison of
·
equation(8) with substitution s (xi0 ) → ζ(ξ0 , t) with the last term of equa-
tion (5). Clearly the former appeared as result of transformation of the latter.
Equating them and taking the scalar productions of both parts of the resulting
relation we arrive at
 2 
kB T Λ + 1 Λσ
ζη1 (ξ1 , t1 )ζη2 (ξ2 , t2 ) = δ(ξ1 −ξ2 ) δ(t1 −t2 )δη1 ,η2 . (11)
ρs π(/m) 1 + Λ2 σ 2
216 S.K. Nemirovskii, L.P. Kondaurova, and M. Tsubota

Here ζη1 and ζη2 are components of random velocities in η1 , η2 directions lying
in the plain normal to the line.
Thus starting from dynamics of BEC (equation (1)) with the fluctuation–
dissipation theorem (3) we derive equation (10 ) describing motion of vortex line
in terms of line itself with the additive noise obeying the fluctuation–dissipation
theorem (11). These relations complete a stochastic problem of quantized vortex
dynamics under thermal noise stemming from the one stirring the underlying
field. In the next section we demonstrate that this problem has an equilibrium
solution given by Gibbs distribution exp(−H {s} /kB T ), where H {s} is the func-
tional of energy due to vortex loop (equation (6)) and T is the temperature of
Bose–condensate.

3 Fokker–Planck Equation

To show it we, first, derive the Fokker–Planck equation corresponding to Langevin

type dynamics obeyed (10) and (11). Let us introduce probability distribution
functional (PDF)
P({s(ξ)}, t) = δ (s(ξ) − s(ξ, t)) . (12)
Here δ is delta functional in space of vortex loop configurations. Averaging is
fulfilled over ensemble of random force. The Fokker–Planck equation can be
derived in standard way (see e.g. [12])
  
∂P δ 1 + Λ2 (1 + Λ2 )Λσ 
+ dξ B(ξ) + s (ξ) × B(ξ) P + (13)
∂t δs(ξ) 1 + Λ2 σ 2 1 + Λ2 σ 2
   2 
kB T Λ + 1 Λσ δ δ
dξdξ  δ(ξ−ξ  ) δ(t1 − t2 )δη1 ,η2 P=0
2
2πρs (/m) 1 + Λ σ 2 2 δs(ξ) δs(ξ  )

Equation (13 ) possesses the equilibrium solution in form of the Gibbs distribu-
tion P({s(ξ)}) = N exp(−H {s} /T ), where N is a normalizing factor. Let us
show that the first integral term vanishes identically for that solution. To do it
we exploit relation (7) and parametrization of label variable ξ in which velocity
·
s (ξ0 ) is normal to the line. Using a tensor notation we rewrite the first term in
integrand in form (we omit the coefficient and factor exp(−H {s} /kB T ))

δsβ (ξ) δH(s) δ 2 H(s) δH(s) δH(s)
αβγ
 + sβ (ξ) + sβ (ξ)
δsα (ξ) δsγ (ξ, t) δsγ (ξ, t)δsα (ξ) δsγ (ξ, t) δsα (ξ, t)

The functional derivative δsβ (ξ)/δsα (ξ) ∝ δβα therefore all terms vanish due to
symmetry. Thus the reversible term gives no contribution to flux of probability
(in the configuration space) equation (13), one says it is divergence free. Further-
more exploiting again relation 7 ) one convinces himself that second (dissipative)
term in (13) and third (due to stirring force) term exactly compensate each other
(locally) as it should be in the thermal equilibrium.
 Stochastic Dynamics of a Vortex Loop. Thermal Equilibrium 217

4 Possible Violation of Thermal Equilibrium

Thus we have proved that the thermal equilibrium of BEC results in the thermal
equilibrium of vortex loop. We are now in position to discuss how it can be
destroyed. Analyzing the proof one can see that the following steps were crucial.
1. Additive white noise ζ(x, t) acting on field ψ(x, t) is transformed into additive
white noise f (ξ, t) acting on vortex line position s(ξ,t). 2. Intensity of noise f (ξ, t)
expressed by (11) is that it locally compensates dissipative flux of probability
distribution functional in the Fokker–Planck equation (13).
That observation points out how the thermal equilibrium in space of vortex
loops can be destroyed for real vortex tangles appeared e.g. in counterflowing
HeII or formed in quenched superfluids. Being a macroscopical objects vortex
loops inevitably undergo large scale perturbations generated e.g. by nonuniform
flow or by action of other vortex loops, randomly placed with respect to the
studied loop. One more essential source of large scale perturbations might be
long wave instabilities of vortex filament motion. That type of random action
drastically differs from δ–correlated in ξ space thermal noise considered above.
It obviously cannot compensate dissipative flux of probability, which is pro-
portional curvature and acts accordingly in small scales. Instead the following
scenario seems to be realized [13]. Due to nonlinear character of the equation of
motion the large–scale perturbations on an initially smooth filament interact cre-
ating higher harmonics. They in turn generate harmonics with larger κ , where
κ is one–dimensional wave vector arisen in 1D Fourier transform (with respect
to label variable ξ) of quantity s(ξ, t). One can say that an additional curvature
created by large scale stirring force propagates in region of small scales. In real
space that corresponds to entangling of vortex loop and creation of vortex line
segments with large local curvature. Then dissipative processes come into play,
their role now is that they remove from the system harmonics with very large
wave vectors κ. So the Kolmogorov cascade–like solution with a flux of a cur-
vature in space of κ̇ is established and, as a result, the stochastic distribution is
far from equilibrium. That scenario is quite similar to the one which is realized
in classical turbulence or in the so called weak (wave) turbulence. A difference
is that if in case of the wave and classical turbulence) an exchange an energy
between
 4 harmonics
 is realized, whereas in our case there is exchange a curvature
κ s(κ)s(−κ) .
The scenario described above is especially relevant in the so called low tem-
perature superfluid turbulence case, when the normal component is very small
and the usual (considered here) dissipation monotonically increasing with the
curvature is absent. Some recent experiments and numerical simulations [14]
show that vortex tangle in HeII decays at extremely low temperature (about 1
mK), where dissipation due to normal component is negligibly small. Obviously
some other strong mechanisms of dissipation must take place. It can be e.g.
emission of phonons and rotons from speedy moving parts of line or just col-
lapse of ”hairpin” segments of the filament. That mechanisms are concentrated
on very small scales (of order of the vortex core size) or equivalently in region
of very large wave numbers κ. Therefore regions of the pumping and the sink
218 S.K. Nemirovskii, L.P. Kondaurova, and M. Tsubota

of an additional curvature are greatly remote in κ space and a system must be

essentially nonequilibrium.
One more reason of a violation of thermal equilibrium might be a reconnec-
tion of lines. In real vortex tangle consisting of many loops the vortex filaments
undergo frequent collisions and reconnections. Just after reconnection there ap-
pear kinks on the curves disappearing later on. From mathematical point of
view a kink on the curve can be described as a discontinuity of tangent vector
s (ξ, t) which has Fourier transform of type s (κ) ∝ κ−1 . Thus the reconnec-
tion processes supply the selected curve with discontinuities having a spectrum
s(κ) ∝ κ−2 . Taking into account a random nature of vortex line collision the
reconnection processes can be modelled, in some measure as a random stirring
of filament with spectrum of type ζ(κ)ζ(−κ) ∝ κ−4 . Remember now that the
establishing of a thermal equilibrium requires that the random force correlation
function is proportional ∝ δ(ξ1 −ξ2 ) and, consequently the spectrum does not
depend on wave number κ, ζ(κ)ζ(−κ) = const . Therefore the colored noise
ζ(κ)ζ(−κ) ∝ κ−4 coming from reconnection processes can also lead to the
Kolmogorov type nonequilibrium state. To clarify which of mechanisms forming
nonequilibrium state prevails in real vortex tangle one has to investigate much
more involved problem.
This work was partly supported under grant N 99-02-16942 from Russian
Foundation of Basic research.

References
1. R.J. Donnelly, Quantized Vortices in Helium II (Cambridge University Press, Cam-
bridge 1991).
2. [Link], J. Low Temp. Phys., 101, 421, (1993)
3. L.P. Pitaevskii: Zh. Exp. Teor. Fiz. (U.S.S.R.) 35, 408 (1958) [Sov. Phys. JETP
8, 282 (1959)].
4. E.P. Gross: Nuovo Cimento 20, 454 (1961)
5. P.C. Hohenberg and B.I. Halperin: [Link] 49, 435 (1972)
6. More or less full bibliography can be found in book by Pismen (see later). Some
more recent papers would be E. Schroder and [Link], [Link]. 40,
1908 (1997).
7. L.P. Pitaevskii: Sov. Phys. JETP 13, 451 (1961)
8. A.L. Fetter: [Link]. 151, 100, (1966)
9. G.K. Batchelor, An introduction to fluid mechanics (Cambridge University Press,
Cambridge 1967)
10. P.G. Saffman: Vortex Dynamics (Cambridge University Press, Cambridge 1992).
11. L.M. Pismen: Vortices in Nonlinear Fields, (Claberson Press, Oxford, 1999)
12. Jean Zinn-Justin: Quantum Field Theory and Critical Phenomena (Claberson
Press, Oxford,1992)
13. S.K. Nemirovskii, J. Pakleza, W. Poppe:Stochasti behaviour of a vortex filament.
Notes et Documents LIMSI N91-14, Orsay (1991)
14. M. Tsubota, Tsunehiko Araki, S.K. Nemirovskii: Phys Rev.B, 2000, (to be pub-
lished).
Stochastic Dynamics of a Vortex Loop.
Large-Scale Stirring Force

S.K. Nemirovskii and [Link]. Baltsevich

Institute of Thermophysics, 630090, Novosibirsk, Lavrent’eva, 1.

Abstract. Stochastic dynamics of a vortex filament obeying local induced approx-

imation equation plus random agitation is investigated by analytical and numerical
methods. The character of a stirring force is supposed to be a white noise with spatial
correlator concentrated at largedistances comparable with size of the loop. Dependence
of the spectral function sα β
κ sκ of the vortex line on both the one-dimensional wave
vector κ and intensity of the external force correlator ζκα ζκα was studied. Here sα κ is
the Fourier transform of the line element position sα (ξ, t). It is shown that under the
influence of an external random force a vortex ring becomes a small tangle whose mean
size depends on external force intensity. The theoretical predictions and the numerical
results are in reasonable agreement.

1 Introduction
In the previous paper [1] we discuss how the large-scale perturbations can de-
stroy the thermal equilibrium state in the space of vortex loop configurations.
In this paper we elaborate that idea and present results of the both analytical
and numerical investigations on stochastic dynamics of a vortex filament in HeII
undergoing an action of the large scale random displacements. Moreover we con-
sider the case when the smooth dissipation connected with normal component
is small (that correspond to the case of very small temperature ) and the only
strong dissipative mechanisms appear at very small scales comparable with the
core radius of vortex. Thus detailed (at which scale) balance between the pump-
ing and dissipation required for themal equilibrium is violated and, as it has
been discussed in [1], essentially nonequilibrium picture state must develop.
From a formal point of view that problem is significantly more involved,
therefore we restrict ourselves to a consideration of the local induction approx-
imation [2],[3] in the equation of motion and omit processes of reconnection.
This statement of problem is, of course, far from real superfluid turbulence in
He II. A value of that work is that it enables us to understand mechanisms of
entanglement of vortex filament and of appearing the strongly nonequilibrium
state. We would remind that idea of the vortex tangle had been launched by
Feynman more than 40 years ago [2] and only about 10-15 years ago Schwarz
demonstrated and confirmed that idea in his famous numerical simulations [3].
To our knowledge a similar success in analytic study is absent.
A structure of the paper is following. In the first part of this paper we de-
velop nonequilibrium. diagram technique analogous to the one elaborated by

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 219–225, 2001.

c Springer-Verlag Berlin Heidelberg 2001
220 S.K. Nemirovskii and [Link]. Baltsevich

Wyld [4] for classical turbulence. Using further method of direct interaction ap-
proximation we derive a set of Dyson equation for the pair correlators and for
the Green functions. Assuming that region of stirring force and dissipation are
widely separated in space of scales we seek for a scale invariant solution in the
so called inertial interval. We also present results of direct numerical simulation
of the vortex tangle dynamics. Numerical results confirm the ones obtained in
the analytical investigations, however some discrepancies remained.

2 Analytical Investigation
In the local induction approximation (LIA) the equation of motion of quantized
vortex filament in HeII reads
d s(ξ, t)
= β̃s  × s 
+ δ + ζ(ξ, t). (1)
dt
Here s(ξ, t) is a point of the filament labeled by variable ξ, 0 ≤ ξ ≤ 2π, which
coincides here with the arclength. The quantity δ stands for dissipation, which
is small for usual scales and large for marginally small scales comparable with
the core size r0 . External Langevin force ζ(ξ, t) is supposed to be Gaussian with
correlator
 α 
ζ (ξ1 , t1 )ζ β (ξ2 , t2 ) = F α (ξ1 − ξ2 ) δ(t1 − t2 ) δ αβ , α, β = 1, 2, 3 , (2)

where F α (ξ1 −ξ2 ) is changing on the large scale of order of the line length (∼ 2π).
˜ ˜
κ
The quantity β̃ is β̃ = 4π log rR0 , with circulation κ and cutting parameters
R (external size, e.g. averaged radius of curvature) and r0 . For our numerical
˜
calculations we have chosen κ = 10−3 cm2 /s. This value corresponds to the
case of superfluid helium.
In Fourier space equation (2) has the form

−iωsq =
α αβγ
Γκκ sβ sγ δ(q − q1 − q2 )dq1 dq2 + δq + ζqα .
1 κ2 q1 q2
(3)

Here sα α
q is the spacial and temporal Fourier component of s (ξ, t), defined as
follows:  
sα
q = sα (ξ, t) ei(ωt − κξ) dtdξ . (4)
αβγ
The vertex Γκκ1 κ2
responsible for nonlinear interaction has the form

iβ̃
αβγ
Γκκ1 κ2
= √ αβγ κ1 κ2 (κ2 − κ1 ), (5)
2 2π
where αβγ it the antisymmetric unit tensor. One can show the vertex Γκκ αβγ
1 κ2
to
satisfy the so called Jacoby identities
, n αβγ -
κ Γκκ1 κ2 + κn2 Γκγαβ2 κκ1
+ κn1 Γκβγα
1 κ2 κ
δ(κ + κ1 + κ2 ) = 0, n = 2, 4. (6)
 Stochastic Dynamics of Vortex Loop. Large-Scale Stirring Force 221

This relations express are tightly connected with the laws of conservation of total
length L and curvature K
 2π  2π
 
L = s s dξ = const, K = s  s  dξ = const . (7)
0 0

Conservation of these quantities is readely derived from either of relations (1),(3).

It is understood that conservation law is held in absence of the both dissipation
and stirring force.
One of the regular approaches to describe random fields is based on the Wyld
diagram technique [4], originally developed to study hydrodynamic turbulence.
Following this technique we introduce for the description of random processes the
following averages: the spectral density tensor (or correlator, or simply spectrum)
Sqαβ and the Green tensor Gαβ q (or simply Green function) which are defined by
 
Sqαβ δ(q + q1 ) = sα β
q sq1 , (8)
9 :
δsα
q
Gαβ
q δ(q + q1 ) = . (9)
δζqβ1
Analysis of diagrams shows that due to the antisymmetry of tensor αβγ con-
αβγ
tained in the expression for the vertex Γκκ1 κ2
, both Sqαβ and Gαβ
q are propor-
tional to δαβ , i.e. Sq ≡ Sq and Gq ≡ Gq . Details of that technique are
αβ α αβ α

described in [5].
The renormalized quantities Sqα and Gα q (taking into account interactions)
satisfy a Dyson set of diagram equations:
◦ ◦
Gα
q = Gα
q + Gα α α
q Σq Gq , (10)

 α  α
Sqα = Gα α
q Fq + Φq Gq . (11)
Here ◦ Gα
q is the “bare” Green function which is equal to (ω − δκ )
−1
. The mass
α α
operators Φq and Σq can be written in form of diagram series: These series
frequently used in nonequilibrium processes have a standard form, explicit form
of them is given in [4],[5].

3 Conservation Laws and Pair Correlators

Dyson equations have shapes indicating a cumbersome handling, therefore they
can be studied for some special cases. One of them is considered in the present
paper. It is connected with conservation laws for the total length and the curva-
ture expressed by (7). Let us consider conservation of total curvature (for total
length there is the same consideration). In Fourier space the conservation laws
for total curvature κ can be expressed in the following form
∂Kκ ∂PκK
+ κ
= I+ (κ) − I−
κ
(κ) (12)
∂t ∂κ
222 S.K. Nemirovskii and [Link]. Baltsevich
 2π
where Kκ = √12π 0 s  s  e−iκξ dξ is the curvature density and PκK is the flux
of this quantity in Fourier space (or, equally, in space of scales). The right-hand
K
side of equation (12) describes creation of additional curvature (with rate I+ (κ))
due to external force and annihilation of it due to dissipative mechanism (with
rate −I− K
(κ)). In the equilibrium case the flux PκK is absent and source and sink
K κ
terms must compensate each other locally for each κ, i.e. I+ (κ) = I− (κ). In
the case under consideration when source and sink terms are widely separated
in κ-space that condition is obviously violated. Therefore a flux of curvature PκK
in Fourier space appears. In region of wave numbers κ remote from both region
of the pumping κ+ and of the sink κ− , κ+  κ  κ+ , the so called inertial
interval, derivative ∂PκK /∂κ = 0 , so PκK is constant equal to P κ . Resuming
we conclude that the problem reduces to study the set of Dyson equation (10)-
(11) in inertial interval under condition of constant flux of the curvature. In this
case Sqα and Gα q are expected to be independent on the concrete type of both
the source and the sink but to be dependent on value of P K . Furthermore, the
αβγ
vertices Γκκ 1 κ2
are homogeneous functions of its arguments. This property, as
well as the condition κ+  κ− by virtue of which one can put κ+ = 0 and
κ− = ∞, leads to the assumption that the problem is the scale invariant, i.e. it
has no characteristic scale for κ. This suggests a power-law form of Sqα and Gα q

1  ω  1  ω 
Sqα = r+p f r , Gαq = r
g r . (13)
κ κ κ κ
Here both f and g are dimensionless functions of their arguments. We aim now
to find the scaling indices r and p.
The first relation between indices r and p can be found from an analysis of
diagram series, claiming all terms to have the same powers of argument κ. This
leads to the first scaling condition
2r + p = 7 . (14)
Another relation between r and p can be obtained from the Dyson equations
(10), (11) which can be rewritten in the form (see e.g. [6])

; <
dω Im Sqα Σqα − Φα α
q Gq = 0. (15)

This relation plays the role of kinetic equations for systems with a weak inter-
action . It has been obtained with help of the expression for the Green function
α −1
Gαq = (ω − Σq ) ; the external force correlator Fκα disappears in the iner-
tial interval. To find a relation of interest between r and p we rewrite relation
(15), disclosing expressions for mass operators Φα α
q , Σq and restricting ourselves
to first order terms in diagram series. That procedure called direct interaction
approximation is frequently used in classical turbulence (see e.g. [6]). After some
calculation we arrive at the following relation (see also [8]):

Im dωdω1 dω2 dκ1 dκ2 δ(q + q1 + q2 ) × Γκκ
αβγ
1 κ2

; αβγ α β γ <
× Γκκ1 κ2 Gq Sq1 Sq2 + Γκ2 κκ1 Gq2 Sq Sq1 + Γκβγα
γαβ γ α β
1 κ2 κ
Gβq1 Sqγ2 Sqα = 0 . (16)
 Stochastic Dynamics of Vortex Loop. Large-Scale Stirring Force 223

To move further we perform conformal transformations in the second and third

term within the braces in integrand, known as Zakharov transformations (see
e.g. Zakharov [7] and Kuznetsov and L’vov [8]). For example for the second
term these transformations have the form

κ = κ  (κ/κ  ) , κ1 = κ  (κ/κ  ) , κ2 = κ(κ/κ  ) , (17)

ω = ω  (κ/κ  )r , ω1 = ω  (κ/κ  )r , ω2 = ω(κ/κ  )r . (18)

The third term is transformed in similar manner. As a result the integrand in
(16) becomes
  x  x
αβγ α β γ αβγ κ γαβ κ
Γκκ1 κ2 Gq Sq1 Sq2 Γκκ1 κ2 + Γκ2 κκ1 + Γκβγα
1 κ2 κ
(19)
κ2 κ1

where
x = 7 − r − 2p . (20)
Due to Jacoby identities (6) the integrand vanishes when x = − 2 for con-
servation of total length and x = − 4 for conservation of total curvature.
Substiting these values into (20) and solving equations (14), (20) we obtaine a
set of couples of indices r, p corresponding to nonequilibrium. states with fluxes
of the length (r = 5/3 , p = 11/3) and of the curvature (r = 1, p = 5). One
time correlators can be found then integrating over frequences ω
 =ω>  − 11
1 κ 3 for length
α
Sκ = dω r+p f ∝ (21)
κ κ s κ−5 for curvature

So we have got solutions for the correlators Sκα which correspond to different
conservation laws in (6). Since there are no sources and sinks acting in the inter-
mediate range these solutions guarantee that the according fluxes are constant.
Depending on the way of agitation of the system one can get the real spectrum

Fig. 1. See text for explanations

224 S.K. Nemirovskii and [Link]. Baltsevich

as some mixture of the obtained solutions in which the fluxes of length and cur-
vature are present simultaneously. A similar situation for wave systems has been
discussed earlier (see e.g. [6] and bibliography therein) and it is called multi-flux
solution.
Having in mind to compare our result with the both numerical and experi-
mental investigation we have to take into account a presence of δ-correlated (in
ξ-space) random force considered in our previous paper [1]. In the local induc-
tion approximation the energy H{s} of line is proportional of its length and in
parametrization when ξ is arclength. can be expressed as

˜2 
ρs κ R
H{s} = ln s (ξ)s (ξ)dξ (22)
4π r0
Γ

It is easy to see that the equilibrium distribution described in [1] leads in that
case to correlator Sκα ∝ 1/κ2 . The final solution is a mix of equilibrium solution
and of the ones expressed by relation (21). Because of nonlinearity it, in general,
is not a simple superposition except of the cases when one of stirring action
prevails and the other can be considered as small deviations. For instance if a
large-scale random stirring is small in comparison with δ-correlated (in ξ-space)
action we have
A B C
Sκα = 2 + 11/3 + 5 . (23)
κ κ κ
The second and third terms in the right-hand side of 23 are small. The constants
A, B, C entering are connected with both intensity of random stirring and its
structure. The further specification requires some additional analysis.

Fig. 2. See text for explanations

 Stochastic Dynamics of Vortex Loop. Large-Scale Stirring Force 225

4 Some Numerical Results

In this section we present some preliminary results on a direct numerical sim-

ulations of a vortex ring evolution under action of a random stirring displace-
ments. The large scale character of noise was guaranteed by calculating it from
a Fourier series taking into account only the first few harmonics. Besides some
(uncontrolled) white noise due to numerical procedure has been excited. Figure 1
shows the projection of the line in the x, y - plane (where the ring was placed
initially) for several times. As predicted, an consequent arising of higher har-
monics takes place leading eventually to an entanglement of the initially smooth
vortex loop.
Another numerical results is shown in Fig. 2 where logarithm of quantity
Sκxx averaged over several realizations is depicted as a functions of log κ. The
average slopes the graphs depend on intensity large-scale stirring force. In several
realizations the slope lies between −2.5 and −3.5, which agrees with theoretical
prediction 23.
This work was partly supported under grant N 99-02-16942 from Russian
Foundation of Basic research.

References
1. S.K. Nemirovskii, L.P. Kondaurova, M. Tsubota: ’ Stochastic Dynamics of Vortex
Loop. Thermal Equilibrium’. In: The same issue????
2. R.J. Donnelly:Quantized Vortices in Helium II (Cambridge University Press, Cam-
bridge 1991)
3. K. W. Schwarz: Phys. Rev. B 18, 245 (1978), K. W. Schwarz, Phys. Rev. B 38
2398 (1988)
4. H. W. Wyld, Ann. Phys. 14, 134 (1961)
5. S.K. Nemirovskii, J. Pakleza, W. Poppe:Stochasti behaviour of a vortex filament.
Notes et Documents LIMSI N91-14, Orsay (1991)
6. V. E. Zakharov, V. S. L’vov, G. Falkovich: Kolmogorov Spectra of Turbulence I,
(Springer-Verlag, 1992)
7. V. E. Zakharov: Zh. Experim. Theor. Phys. 51, 688 (1966)
8. E. A. Kuznetsov, V. S. L’vov: Phys. Lett. 64A, 157 (1977)
Nonequilibrium Vortex Dynamics in Superfluid
Phase Transitions and Superfluid Turbulence

Han-Ching Chu and Gary A. Williams

Department of Physics and Astronomy,

University of California,
Los Angeles, CA 90095 USA

Abstract. The nonequilibrium dynamics of superfluid phase transitions and super-

fluid turbulence are calculated using vortex renormalization methods. In two dimen-
sions the Kosterlitz–Thouless recursion relations are solved in conjunction with the
Fokker–Planck equation for the vortex–pair dynamics, yielding complete solutions for
quenched 2D superfluid transitions that are in agreement with scaling predictions and
numerical simulations. For the case of 2D superfluid turbulence the steady–state solu-
tion is found to be a forward cascade of enstrophy, and it is argued that the energy
spectrum varies as k−3 . The extension of the theory to three dimensions using vortex–
loop renormalization is discussed.

1 Introduction
The topics of superfluid turbulence and superfluid phase transitions are not usu-
ally discussed together, although both are known to involve complex arrange-
ments of tangled vorticity (vortex pairs [1] in two dimensions (2D) and vortex
loops [2] in three dimensions (3D)). Here we illustrate that there is a fundamental
connection between the two problems. The same vortex renormalization tech-
niques that are used to solve the phase transition problem are applied directly
to find solutions for the case of superfluid turbulence.
We show this initially in two dimensions, considering a thin superfluid helium
film on a flat substrate, which is coupled to a thermal reservoir at temperature
T . The vortex pairs in the film are characterized by the distribution function
Γ (r, t), which is the density of pairs of separation between r and r+dr. It is
determined from the 2D Fokker–Planck equation [3],
  
∂Γ 2D ∂ ∂Γ ∂ U − p · (v n − v s )
= 2 · +Γ , (1)
∂t ao ∂r  ∂r  ∂r  kB T
where r = r/ao with ao the vortex core size, U and p are the pair interaction
energy and impulse, v n and v s are any externally applied superfluid and normal
fluid flow fields, and D is the diffusion coefficient characterizing the mutual
friction drag force on the vortex cores of a pair. It can be seen that the time in
this equation is scaled by τo = a2o /2D, the diffusion time of the smallest pairs of
separation ao . In equilibrium where ∂∂ Γt = 0 the solution of (1) is just the usual
Boltzmann distribution,
 
1 −(U + 2Ec ) + p · (v n − v s )
Γ = 4 exp (2)
ao kB T
C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 226–232, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Nonequilibrium Vortex Dynamics 227

where Ec is the vortex core energy. The vortex interaction energy U in (1) is
determined using the Kosterlitz–Thouless (KT) vortex renormalization methods
[1]. The Kosterlitz recursion relations can be written in the form

1 ∂U 2π K
= (3)
kB T ∂ r r
and
∂K
= − 4 π3 r3 K 2 Γ , (4)
∂r
where K = 2 σs /m2 kB T is the dimensionless areal superfluid density. In thermal
equilibrium (2), (3), and (4) lead to the well-known result that the superfluid
density has a universal jump to zero at the transition temperature TKT .

2 Quenched Superfluid Transition

To study the quenched 2D superfluid transition [4] we first equilibrate the film
with the heat bath at temperature T , generating an equilibrium distribution
of vortex pairs. The top curve in Fig. 1 at t = 0 is the distribution at TKT
found by iterating (2), (3), and (4); it varies asymptotically as (r/ao )−2πK . The
temperature of the heat bath is then reduced abruptly to a low temperature,
0.1 TKT , where in equilibrium the vortex density is over 20 orders of magnitude
smaller. However, the pairs cannot suddenly disappear, since the only way they
can be extracted by the heat bath is if the plus–minus pairs annihilate at r = ao ,
where the remaining core energy is converted to phonons. This can take quite a
long time to occur, however, since the pairs only slowly lose kinetic energy to the
diffusive frictional force of the heat bath as they move in towards annihilation.

Fig. 1. Time dependence (in units of the diffusion time τo ) of the pair distribution
function Γ (in units a−4
o ) for an instantaneous quench from TKT to 0.1 TKT .
228 H.-C. Chu and G.A. Williams

Fig. 2. a) Decay of the pair density (in units a−2

o ) and b) recovery of the superfluid
fraction in time after a quench to 0.1 TKT from different starting temperatures Ti .

To solve for the time dependence of the distribution function after the quench
we combine (1) and (3), which with no applied flows becomes
 
∂Γ 2D ∂ ∂Γ
= r + 2πK Γ . (5)
∂t r ∂r ∂r

We solve this in conjunction with (4) using the method of lines with a third–
order Runge–Kutta technique on a finite domain, to a maximum pair separation
R/ao = e10 = 2.2×104 . The boundary conditions used are that the flux of pairs
J = −2D(r ∂Γ∂r + 2πK Γ ) be continuous across the boundary r = R, and that Γ
drop abruptly to zero at r = ao , which is equivalent to putting a delta-function
sink term at ao on the right-hand side of (5) to generate the annihilation at that
point.
Figure 1 shows the time dependence of the distribution function for an in-
stantaneous quench from TKT to 0.1 TKT . The smallest pairs decay away first,
 Nonequilibrium Vortex Dynamics 229

Fig. 3. Dynamic exponent z calculated from the slope of the curves in Fig. 2a, compared
with various theories and simulations.

in the classic pattern of a phase–ordering transition where at long times the

largest scales become more dominant. By integrating the distribution function
over d2 r the pair density is obtained, shown as the top curve in Fig. 2a. At long
times the density decreases nearly as t−1 , in agreement with scaling theories
and simulations [5]. We find, however, that the exponent only approaches −1
logarithmically slowly, being −1.12 at t = 100 and −1.09 at t = 2×104 . This
is a consequence of the slow recovery of the superfluid density after the quench
(bottom curve of Fig. 2b), which has a logarithmic approach to its equilibrium
value.
Quenches from starting temperatures below TKT also display power-law vari-
ation of the vortex density, since the KT transition is a line of critical points. The
plots in Fig. 2 show this for a series of initial temperatures between 0.8 TKT and
TKT . At the lower starting temperatures the superfluid density recovers more
quickly, and the vortex decay then varies accurately with the form t−z/2 where
z is a temperature–dependent constant. The increasing slopes in Fig. 2a give the
increasing values of z plotted in Fig. 3. We find complete agreement (except near
TKT due to the logarithmic slowness) with the scaling prediction of Minnhagen
and co-workers [6] for the dynamic exponent

σs (T ) TKT
zscale = 4 −2 , (6)
σs (TKT ) T

and disagreement with other predictions [3,5]. Minnhagen has recently shown
[6] that simulations [5,7] of the vortex decay can be misleading, since different
results for z are obtained depending on the boundary conditions used: peri-
odic boundary conditions (PBC) yield only the spin-wave result z = 2, while
fluctuating–twist boundary conditions (FTBC) give zscale . Our results in agree-
ment with zscale confirm that this is in fact the correct exponent characterizing
the vortex dynamics.
230 H.-C. Chu and G.A. Williams

3 Superfluid Turbulence

We can apply the same KT renormalization techniques to 2D superfluid turbu-

lence. For this the film is held at a low temperature 0.1 TKT , and vortex pairs of
large separation R are injected externally into the film at random positions and
random orientations, at a rate Q̇ pairs per unit area per second. These pairs will
begin to cascade down to smaller separations under the action of the frictional
force, finally reaching the annihilation scale ao where they are absorbed by the
thermal bath. Eventually a steady–state will be achieved where the distribution
function becomes constant in time, with the annihilation current at ao being
equal to the injected current at R. This is a forward enstrophy cascade, which
is a conserved quantity in 2D [8].
The steady-state distribution can be found by solving the Fokker–Planck
equation with a source term added,
 
∂Γ 2D ∂ ∂Γ
=0= r + 2πK Γ + Q̇δ(R − r) , (7)
∂t r ∂r ∂r

in conjunction with (4), and again assuming as above that Γ drops to zero at
ao . Integrating this over d2 r gives
∂Γ α
r + 2πK Γ = H(R − r) , (8)
∂r 4π 2 a4o

where we have defined a dimensionless injection rate α = a2o Q̇τo , and H is the
Heaviside unit step function. For low injection rates the solutions of (8) and (4)
are a constant distribution function Γ = Γo H(R − r) and a superfluid fraction
unchanged from unity, where Γo = α/4πKo a4o and Ko is the value of K at the
scale ao (where the areal superfluid density is just the areal density of the fluid at
low T ). At higher injection rates the density of the injected vortices can start to
approach densities found at TKT , and this has the effect of driving the superfluid
density to zero at some length scale ro . Setting K = 0 in (8) then gives the
variation of the distribution at larger scales, Γ (r) − Γ (ro ) = (α/4πa4o ) ln(r/ro ).
Full numerical solutions of (8) and (4) are shown in Fig. 4 for a wide range
of injection rates α. At the lowest rates the injected distribution adds to the
thermal vortices at the bath temperature of 0.1 TKT , while at intermediate rates
the distribution is the constant value Γo . As the injected distribution starts to
approach values comparable to the thermal distribution at TKT , the superfluid
density is reduced by the pairs, finally dropping to zero, where Γ then takes on
the logarithmic increase noted above.
The work done in forming a given distribution of the pairs is the negative of
the free energy, since the system is coupled to the bath. The work per unit mass
can be calculated from the Kosterlitz free energy [1], which in our variables is

F 2πkB T R
2 Γo 2
W =− o
= Γ (r) r dr = π (R − a2o ) , (9)
σs σso ao m2 K o
 Nonequilibrium Vortex Dynamics 231

Fig. 4. a) Pair distribution resulting from injection of pairs at R/ao = 2.2×104 , for a
wide range of injection rates α as indicated along the right side, b) superfluid fraction
versus length scale for several of the higher injection rates.

where the last equality holds at the intermediate injection rates

 where Γ is a
constant. Defining a k-space energy spectrum W (k) by W = W (k) dk, it can
be seen that over the relevant range 1/R < k < 1/ao the result of (9) is achieved
for W (k) ∝ k −3 . This is the same dependence found for the forward cascade
in classical 2D Navier-Stokes turbulence [8]. The superfluid cascade differs from
the classical case, however, since the spectrum varies linearly with the injected
enstrophy flux η = 2(h/m)2 Q̇a−2 o and not as η
2/3
. The spectrum takes the form
−3
W (k) = C η τo k with C a dimensionless constant, where it can be seen that
the k −3 variation is the only dimensionally consistent possibility, apart from
possible ln(kR) corrections. The occurrence of τo (which is “universal” for the
cascade since these identical vortices all have the same drag force on them) is
easily understood physically: the longer it takes for a pair to cascade down the
phase space the more the vortices pile up and increase the total energy. We note
that for the superfluid system the enstrophy cascade is the only cascade that
232 H.-C. Chu and G.A. Williams

forms, since energy is not conserved due to the coupling to the thermal bath. The
reverse energy cascade that is found in the classical turbulence [8] is not possible
because the quantized circulation prevents the merger of like–sign vortices which
is necessary for such a cascade. It will be of great interest if computer simulations
[8] can check these predictions for the cascades in 2D superfluid turbulence.
In further work it should be possible to study the time decay of the turbulence
by switching off the injection term in (7), and then solving (5) in the same manner
as for the quenched transitions. In fact (5) is then readily identified as the 2D
equivalent of the phenomenological Vinen equation [9] governing the dynamics
of superfluid turbulence.

4 Three Dimensions
The analytic solutions described above for both 2D quenched superfluid tran-
sitions and 2D superfluid turbulence are the result of the simplicity of the KT
renormalization methods. It should be quite feasible to extend these techniques
to three dimensions, where the relevant excitations are vortex loops. A simple
theory [2] using the KT ideas in 3D has been able to accurately model many
features of the superfluid λ–transition. The renormalization allows the complex
tangle of interacting noncircular loops to be reduced to a simpler problem, by
averaging out the effects of shorter length scales, until only a dilute gas of loops
of size equal to the coherence length are left. For the turbulence problem rings
of large size will be injected, followed by a cascade to smaller loops due to the
mutual friction force, until the smallest loops are removed at the scale ao . It
should be noted that in this approach the very complicated motion and recon-
nections of the individual loops do not need to be considered (just as they were
not for the 2D pairs above); all that needs to be taken into account is a statistical
“snapshot” of the distribution function at each timestep. It seems quite possible
that the 3D Fokker–Plank equation [10] will again prove to be the source of the
Vinen equation for the turbulent dynamics.
Useful discussions with G. Eyink are acknowledged. This work is supported
by the US National Science Foundation, grant DMR 97-31523.

References
1. J.M. Kosterlitz, D. Thouless: J. Phys. C 6, 1181 (1973) J.M. Kosterlitz: J. Phys.
C 7, 1046 (1974)
2. G. A. Williams: Phys. Rev. Lett. 83, 2347 (1999)
3. V. Ambegaokar, B. Halperin, D. Nelson, E. Siggia: Phys. Rev. B 21, 1806 (1980)
4. H. Chu, G. A. Williams: Physica B 284-288, 115 (2000)
5. A. Bray, A. Briant, D. Jervis: Phys. Rev. Lett. 84, 1503 (2000)
6. B. Kim, P. Minnhagen, P. Ollson: Phys. Rev. B 59, 11506 (1999)
7. H. Luo, B. Zheng: Mod. Phys. Lett. 11, 615 (1997)
8. R. Kraichnan: Phys. Fluids 10, 1417 (1967) E. Lindborg, K. Alvelius: Phys. Fluids
12, 945 (2000)
9. W. F. Vinen: Proc. Roy. Soc. London A 242, 493 (1957)
10. G. A. Williams: J. Low Temp. Phys. 93, 1079 (1993)
The Nonlinear Schrödinger Equation
as a Model of Superfluidity

Paul H. Roberts and Natalia G. Berloff

Department of Mathematics
University of California, Los Angeles, CA, 90095-1555
roberts@[Link], nberloff@[Link]

Abstract. The results of theoretical and numerical studies of the Gross-Pitaevskii

(GP) model are reviewed. This model is used to elucidate different aspects of super-
fluid behaviour: the motion, interactions, annihilations, nucleation and reconnections
of vortex lines, vortex rings, and vortex loops; the motion of impurities; flow through
apertures; superfluid turbulence and the capture of impurities by vortex lines. The
review also considers some generalizations of the model.

1 Introduction
One of the most useful ways of describing superfluid helium at zero temperature
begins with Schrödinger’s equation for the one-particle wave function Ψ . Since
liquid helium is a strongly correlated system dominated by collective effects, the
form of the Hamiltonian in Schrödinger’s equation cannot be derived starting
from first principles. At zero temperature 4 He has large interatomic spacing and
low density, which suggests a description in terms of a weakly interacting Bose
gas for which such a derivation can be made rigorous. Continuum mechanical
equations for 4 He are usually built on the assumption that a Bose condensate
gives the exact description at zero temperature.
The imperfect Bose condensate in the Hartree approximation is governed by
equations that were derived by Ginsburg and Pitaevskii [22] and by Gross [26],
and generally called “GP equations”. In terms of the single-particle wavefunction
Ψ (x, t) for N bosons of mass M , the time-dependent self-consistent field equation
is 
2 2
iΨt = − ∇ Ψ + Ψ |Ψ (x , t)|2 V (|x − x |) dx , (1)
2M
where V (|x − x |) is the potential of the two-body interactions between bosons.
The normalization condition is

|Ψ |2 dx = N. (2)

For a weakly interacting Bose system, (1) is simplified by replacing V (|x − x |)

with a δ - function repulsive potential of strength V0 . This does not alter the
nature of the results since the characteristic length of the weakly interacting Bose

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 235–257, 2001.

c Springer-Verlag Berlin Heidelberg 2001
236 P.H. Roberts and N.G. Berloff

gas is larger than the range of the force [26]. Equation (1) for such a potential
is
2 2
iΨt = − ∇ Ψ + V0 |Ψ |2 Ψ. (3)
2M
Equations (1)-(3) define Hamiltonian systems, the following integrals being con-
served: mass 
M = M |Ψ |2 dx, (4)

momentum density 

p= [Ψ ∗ ∇Ψ − Ψ ∇Ψ ∗ ] dx, (5)
2i
and energy; in the case of (3) this is expressed by
 
2 V0
E= |∇Ψ |2 dx + |Ψ |4 dx. (6)
2M 2
Our starting point is a condensate everywhere at rest in what we shall call
the “laboratory frame”, so that Ψ = exp(iEv /) in that frame, where Ev is the
chemical potential of a boson (i.e., the increase in ground state energy when one
boson is added to the system). We then consider deviations from that state by
studying the evolution of ψ where

ψ = Ψ exp(iEv t/), (7)

so that (3) becomes

2 2
iψt = − ∇ ψ + V0 |ψ|2 ψ − Ev ψ. (8)
2M
It is usually convenient to model phenomena in an infinite domain in which,
prior to the onset of a disturbance, ψ = ψ∞ everywhere, where, by (8)

ψ∞ = (Ev /V0 )1/2 . (9)

We then modify (4)-(6) to forms measuring departures from this uniform state
(see, e.g. [44]) 
(|ψ|2 − ψ∞
2
) dx = 0, (10)


p= [(ψ ∗ − ψ∞ )∇ψ − (ψ − ψ∞ )∇ψ ∗ ] dx, (11)
2i
 
2 V0
E= |∇ψ| dx +2
(|ψ|2 − ψ∞
2 2
) dx. (12)
2M 2
The mass density and flux are

ρ = M ψψ ∗ , j= ((ψ ∗ − ψ∞ )∇ψ − (ψ − ψ∞ )∇ψ ∗ ). (13)
2i
 Nonlinear Schrödinger Equation 237

2 The Fluid Equations

One of the principal aims of this review is to interpret the consequences of the
condensate model, not from the perspective of solid state physics, but through
the eyes of a fluid dynamicist. This is achieved by recasting the nonlinear Schrödinger
equation by the Madelung transformation. We write

ψ = ReiS , (14)

so that
ρ = M R2 , j = ρu = ρ∇φ, φ = (/M )S. (15)
The real and imaginary parts of (8) then yield a continuity equation

∂ρ
+ ∇ · (ρu) = 0, (16)
∂t
and an integrated form of the momentum equation
 
∂φ 1 2 ρ ∇2 ρ1/2
+ u +c 2
− 1 − c2 a2 1/2 = 0, (17)
∂t 2 ρ∞ ρ

where ρ∞ is the density at infinity and c is the speed of sound:

2
ρ∞ = M ψ ∞ , c2 = Ev /M = (V0 /M 2 )ρ∞ . (18)

The last term in (17) is sometimes called the “quantum pressure” although
it is dimensionally a chemical potential. It is significant where “healing” of the
wavefunction ψ is important, as for instance in a vortex core, or within a “healing
layer” adjacent to a boundary, of approximate thickness

a= , (19)
(2M Ev )1/2

which is called the “healing length”. Elsewhere, in the “bulk” of the fluid, the
last term in (17) in insignificant, and (17) assumes the form appropriate for a
classical inviscid fluid, in which the pressure, P , is proportional to ρ2 .
The unintegrated form of the momentum equation is obtained by taking the
gradient of (17). This gives
 
∂ui ∂ui ∂P ∂Σij
ρ + uj =− + , (20)
∂t ∂xj ∂xi ∂xj

where P is pressure and the Σij are the quantum stresses

 2
V0 2  ∂ 2 ln ρ
P = ρ , Σij = ρ ; (21)
2M 2 2M ∂xi ∂xj

see [24].
238 P.H. Roberts and N.G. Berloff

Curves on which the wavefunction ψ vanishes correspond to vortex lines in

the superfluid with the correct unit of quantization, κ = h/M . For simple zeros,
of the kind with which we shall be concerned, the circulation

C= u · ds (22)
Γ

round a closed contour Γ is equal to the number of vortex lines that Γ contains.
Since u = ρ∇φ, the circulation C is preserved until a vortex line crosses Γ ,
whereupon C increases or decreases by κ. At the instant at which the vortex
line meets Γ , the density ρ on Γ is zero at the point of intersection; Γ is no longer
a closed curve, and Kelvin’s theorem on the constancy of C is inapplicable.
This marks an extremely basic difference between the condensate equations
and the Euler equations of classical fluid dynamics. Both describe conservative
(Hamiltonian) systems, but the Euler fluid does not allow circulation to change.
In describing topological changes in the vorticity of a classical fluid, it is neces-
sary to invoke a mechanism that breaks the constraint of Kelvin’s theorem. For
example, one method of modeling the severing or coalescence of vortex filaments
is to restore viscosity to the fluid, so allowing the vorticity to diffuse from one
filament to the other. Another method is described in Sect. 5 below. Such devices
are unnecessary when the condensate model is employed.
The GP model not only enjoys the advantage of comparative simplicity but
also describes qualitatively correct superfluid behaviors at low temperatures T .
In this paper we review how the GP model is successfully used to elucidate the
motion, interactions, annihilations, nucleation and reconnections of vortex lines,
vortex rings, and vortex loops; the motion of impurities; the flow of superfluid
through apertures; superfluid turbulence; and the capture of impurities by vortex
lines.

3 Shortcomings of the GP Model

Despite its success, several aspects of the local model (8) are qualitatively or
quantitatively unrealistic for superfluid helium. The dispersion relation between
the frequency, ω, and wave number, k, of sound waves according to (8) is
 2

ω 2 = c2 k 2 + k4 . (23)
2M
1
The velocity, c, of long wavelength sound waves is therefore proportional to ρ 2 see
(21). (We have here denoted the bulk density, ρ∞ , by ρ.) That this is unrealistic
is seen from the experiments on Grüneisen constant UG = (ρ∂c/c∂ρ)T , which
shows that, in the bulk (i.e., on length scales long compared with the healing
length, κ/c), the fluid behaves as a barotropic fluid (p ∝ ργ ) with γ = 2.8 (see
[10] and references therein).
By writing p = k and (p) = ω, we convert (23) into the dispersion curve
of GP theory:
, -1/2
(p) = p c2 + (p/2M )2 . (24)
 Nonlinear Schrödinger Equation 239

At best, this describes the phonon branch of the excitation spectrum. The roton
branch is lost. At first sight this does not seem important, because it is only the
phonon branch that is thermally populated at low temperatures. Nevertheless,
the lack of a roton branch is seen to be serious when, as in Sects. 6 and 7 below,
we use the GP model to simulate superflow past boundaries. We shall describe
in Sect. 8 a generalization of (8) that restores the roton branch.
To expose a related deficiency of the GP model, we invoke the “ghost” of a
normal fluid by supposing that T is infinitesimal but nonzero. In local thermo-
dynamic equilibrium, the density of quasi-particles in the fluid is
    −1
1 −w·p
Nq = exp −1 dp, (25)
3 kT

where w = un − us , is the velocity of the normal fluid1 relative to the superfluid;

e.g. see [34]. It is clear from (24) that Nq is negligibly small when w < c, because
then −w·p > 0 for all p. When w > c, however, the integrand in (25) is negative
and even infinite for p for which  − w · p ≤ 0. This unphysicality shows that the
assumption of local thermodynamic equilibrium is untenable wherever w > c.
Then quasi-particle production destroys superfluidity.
In real superfluid helium, the limiting value of w for the destruction of super-
flow is known as the ‘Landau critical velocity’; we shall denote this by uL . The
argument just given correctly indicates that uL is determined by a condition
of tangency: uL = (p)/p = d(p)/dp and, because of the roton branch, this is
much less than c for superfluid helium. Compressibility therefore plays a greater
role in the condensate than in real helium.
GP theory has no preferred reference frame. This is hardly surprising in a
model which applies at T = 0, where not even the “ghost” of normal fluid exists.
Nevertheless, it is a non-uniformity for T → 0 that is an undesirable attribute of
GP theory. Carlson [13] proposed a way of removing this by augmenting (8) with
a term motivated by Khalatnikov’s (1952) theory of mutual friction between the
superfluid and normal fluid at finite T , which involved three independent kinetic
coefficients, of which only the third, ζ3 (> 0), concerns us here. Khalatnikov
modified the integrated superfluid momentum equation to be
∂φ 1 2
+ us + μ = −ζ3 ∇ · (ρs w), (26)
∂t 2
where μ is a chemical potential; Using the mass conservation equation
∂ρ
+ ∇ · (ρu) = 0, (27)
∂t
1
We are attempting here to make a point of principle, and for this purpose we may
ignore the fact that the mean free path of phonons at low temperatures may be so
large that they are better represented by a Boltzmann equation than by a normal
fluid continuum. We temporarily write the density of the superfluid and normal
fluid by ρs and ρn (where ρs + ρn = ρ) and their velocities as us and un (where
ρu = ρs us + ρn un is the total momentum density).
240 P.H. Roberts and N.G. Berloff

we see that (26) can also be written as

 
∂φ 1 2 ∂ρ
+ us + μ = −ζ3 ∇ · + ∇ · (ρun ) . (28)
∂t 2 ∂t

Since ρ ≈ ρs for small T , this suggests that a useful generalization of (8) is

2 2 , -
iψt = − ∇ ψ + V0 |ψ|2 ψ − Ev ψ + ζM 2 ψ ∂t |ψ|2 + ∇ · (|ψ|2 un ) , (29)
2M
where ζ > 0. This implies, in place of (17),
   
2 2∇ ρ
2 1/2
∂φ 1 2 ρ ∂ρ
+ u +c 2
−1 −c a = −ζ + ∇ · (ρvn ) , (30)
∂t 2 ρ∞ ρ1/2 ∂t

and (16) is unchanged. Unlike (1) or (8) equations (29) and (30) single out a
preferred reference frame, the one in which vn = 0. It is reasonable to suppose
that ζ → 0 as T → 0 since otherwise the normal fluid, which is subjected to the
frictional force that is the counterpart of the ζ term in (30), would be infinitely
accelerated in the limit.
Let us again suppose infinitesimal but nonzero T , and determine how the
right-hand side of (30) modifies the dispersion relation (23). We add, to a uniform
counterflow w, perturbations proportional to exp(ik · x − iωt), and quickly find
from (16) and (30) that (23) is replaced by

ω 2 = c2 k 2 (1 + 12 a2 k 2 ) + iζρ∞ k 2 (ω − w · k). (31)

For small ζ, the roots of (31) are approximately real; one of them is ω ≈ ωr =
ck(1 + 12 a2 k 2 )1/2 > 0, and its imaginary part is defined by the imaginary part
of (31):  
ωi ≈ −(ζρ∞ k 3 /2ωr ) w − c(1 + 12 a2 k 2 )1/2 . (32)
If w < c, this is positive for all k, but if w > c it is negative for all sufficiently
small k, so establishing the breakdown of superfluidity once the Landau critical
velocity (here c) is exceeded.
Although, as shown by Berloff (1999), the ζ term can be a useful practical
tool in computing vortex structure, our interest in ζ in this review is that it
represents a potentially significant source of non-thermal quasi-particles. Con-
sider superflow past a body such as a positive ion. It is convenient, as in §6
below, to treat the body as an infinite potential barrier to the condensate, and
if this is done the GP equation allows the superfluid to have any velocity along
the boundary. In reality, however, ψ penetrates a very short distance into the
body and collisions within the solid can generate quasi-particles that, on aver-
age, move with it. In the parlance of fluid mechanics, the normal fluid velocity
un obeys the no-slip condition on the boundary un = 0. The ζ−term leads to
quasi-particle emission whenever the superflow along the boundary exceeds the
Landau critical velocity; for the local GP model, the quasi-particles are seen as
sound waves (phonons).
 Nonlinear Schrödinger Equation 241

It may be worth pointing out that the condition w > uL for the destruction of
superfluidity is invalid where healing occurs and the ∇2 ρ1/2 term in (17) or (30)
is significant. The superfluid vortex provides a prime example, in which us → ∞
in the vortex core, although no instabilities are seen in numerical simulations
(provided the vortex as a whole does not move with super-Landau speeds; see
Berloff and Roberts (1999). For the remainder of this review, us will again be
denoted by u.
Finally, we should not ignore what is perhaps the most fundamental objection
to GP theory. Although by including  it provides the verisimilitude of being
a quantum description of superfluidity, it is in reality semi-classical. A proper
operator description is, however, so complex that it cannot be usefully employed
to study situations as complicated as those described in this review. We should
not, however, overlook the progress that has been made towards a more realistic
description of a vortex core (see [46]).

4 Vortices
A vortex line is defined by a zero of the wave function ψ = 0. The straight-line
vortex creates a velocity around this line of u = /M s, where s is the distance
from the vortex line. The equation for the amplitude of the steady straight-line
vortex is found by substituting ψ = R(s) exp(iχ) into (8) written in cylindrical
coordinates (s, χ, z):

d2 R 1 dR 1
2
+ − 2 R + R − R3 = 0, (33)
ds s ds s
where distance is scaled by a and R by R∞ . For small distances from the axis,
R is proportional to s; for s → ∞, R ∼ 1 − 1/2s2 . The energy per unit length of
the vortex line is
∞  2 ∞ ∞ 
κ 2 ρ∞ dR R2 1
E = s ds + ds + (1 − R ) s ds .
2 2
(34)
4π ds s 2
0 0 0

The first term can be regarded as a “quantum energy”, the second term is
the classical vortex kinetic energy that diverges logarithmically unless a cut-off
distance L is introduced; the third term in (34) represents the potential energy.
The energy per unit length of the vortex is usually expressed in the form
 
ρκ2 L
E = ln + L0 , (35)
4π a

where the constant L0 is called the “vortex core parameter” and was determined
numerically by Pitaevskii [40] as L0 = 0.3809.
A dense array of straight-line vortices are encountered when a bucket of
superfluid is rotated rapidly. This greatly affects its dynamics. This gives rise to
HVBK theory; see, for example Hills and Roberts [29] and Holm (this meeting).
242 P.H. Roberts and N.G. Berloff

Straight-line vortices can transmit energy along their length by Kelvin waves.
These have been comprehensively analyzed for incompressible Euler fluids; see
for example Chapter 11 of Saffman [47]. They have been less well studied for
compressible fluids such as the GP condensate, for which the Kelvin wave may be
accompanied by acoustic emission that ultimately damps out the waves, unless
they are sustained by a source. This emission is of interest in studies of super-
fluid turbulence (§5). According to the analysis of Pitaevskii [40], the Kelvin
waves that bend the rectilinear vortex do not emit sound if their wavelength
is sufficiently long; more general types of wave have been studied by Rowlands
[45].
Large vortex rings were investigated by Roberts and Grant [44] and they
obtained the following expressions for the energy per unit length, momentum,
and velocity, v:
 
8R
E = 12 ρκ2 R ln + L0 − 2 , p = ρκπR2 , (36)
a
 
8R
v = (κ/4πR) + L0 − 1 . (37)
a
As expected, E ≈ 2πRE , but also rings obey Hamilton’s equation

v = ∂E/∂p. (38)

Jones and Roberts [33] determined the entire sequence of vortex rings nu-
merically for the GP model (8). They calculated the energy E and momentum
p and showed how the location of the sequence in the Ep-plane relates to the
superfluid helium dispersion curve. They found two branches meeting at a cusp
where p and E assume their minimum values, pm and Em . As p → ∞ on each
branch, E → ∞. On the lower branch the solutions are asymptotic to the large
vortex rings (36)-(37). Since the GP model has a healing length (based on the
sound velocity) different from the vortex core parameter there are two possible
ways to introduce dimensional units and to plot the solitary wave sequence next
to the Landau dispersion curve on the pE− plane. If the dimensional units based
on the vortex core parameter are chosen, the cusp lies just above the Landau
dispersion curve; if instead the healing length (sound speed) is selected the cusp
meets the dispersion curve of the GP model, which (we recall) does not have a
roton branch.
As E and p decrease from infinity along the lower branch, the solutions begin
to lose their similarity to large vortex rings, and (36) - (37) determine E, p,
and v less and less accurately, although (38) still holds. Eventually, for a mo-
mentum p0 slightly greater than pm , the rings lose their vorticity (ψ loses its
zero), and thereafter the solitary solutions may better be described as ‘rarefac-
tion waves’. The upper branch consists entirely of these and, as p → ∞ on this
branch, the solutions asymptotically approach the rational soliton solution of
the Kadomtsev-Petviashvili (KP) equation and are unstable.
 Nonlinear Schrödinger Equation 243

An interesting, and still incompletely answered question is whether the GP

sequence of solitary waves is stable or not. The question is not frivolous since
the classical circular rings in an incompressible fluid are known to be unstable
(Widnall and Sullivan [54]). Jones et al. [32] concluded that the entire upper
branch of rarefaction waves [33] is unstable, but that the lower branch solutions
may be stable because their axisymmetric expansion or contraction is forbidden
due to energy and momentum conservation. Grant [23] examined asymmetric
perturbations of a large ring in the form of infinitesimal Kelvin waves, but he
did not locate any that grew and, moreover, his analysis is inconclusive since
it omitted the acoustic emission that accompanies the Kelvin waves. This was
however included by Pismen and Nepomnyashchy [39], who found that it is
stabilizing.

5 Superfluid Turbulence; Vortex Line Reconnection

Superfluid turbulence has been the focus of many experimental studies (e.g., [52],
[18]), especially in “the high temperature regime,” by which we mean 0.6◦ K ≤
T < Tλ ≈ 2.172◦ K, where Tλ is the λ-point, which marks the transition between
the normal and superfluid phases of helium. In this regime, the density, ρs ,
of superfluid is smaller than the normal fluid density, ρn , and in consequence
turbulence in the superfluid is, to a large degree, determined by the turbulence
taking place in the normal fluid. This is reflected by the success of theories
of superfluid turbulence, such as that of Barenghi et al. [1], [2], in which the
superfluid vorticity is largely tied to that of the normal fluid. In contrast, in the
low temperature range (T ≤ 0.6◦ K), where ρn is smaller than ρs , we may expect
turbulence in the superfluid largely to determine turbulence in the normal fluid,
rather than the reverse.
For the study of superfluid turbulence in the low temperature range it is
necessary to follow the evolution of only two fields, the real and imaginary parts
of ψ, rather than ρ and v, or ρ and a non-single-valued velocity potential φ. There
is however no high wavenumber sink of energy to terminate the Kolmogorov
cascade. This obstacle may restrict the use of the simplest forms (1) or (8) of
the GP model for turbulence studies, but the Carlson generalization, that uses
(30 instead of (17), may be the remedy, since this recognizes that, in real helium
even in the low temperature range, normal fluid is present that is coupled to the
superfluid and which, through its viscosity, provides a high wavenumber sink.
Another difficulty is that strong turbulence contains a complicated mixture of
condensate, phonons, rotons, quantized vortices, vortex waves and shocks. Such
a variety of processes may make it difficult to determine universal scaling laws
for the turbulent correlation functions.
Recently several papers have appeared that discuss decaying Kolmogorov
turbulence using GP theory. Nore et al. [38] studied superfluid turbulence us-
ing numerical simulations of the GP model. They decomposed the total energy
(which is conserved) into incompressible kinetic, internal, and “quantum” com-
ponents (that corresponds to acoustic excitations), and they computed the cor-
244 P.H. Roberts and N.G. Berloff

responding energy spectra. They found that the rate of transfer of kinetic energy
into other energy components is comparable with the rate of energy dissipation
through viscosity in classical turbulence. At the moment of maximum energy
dissipation, the energy spectrum resembles the Kolmogorov inertial range.
The term “superfluid turbulence” is often used synonymously for the “evolu-
tion of a superfluid vortex tangle”. The dynamics of the turbulent state depend
crucially on the interactions of the vortex filaments. In a set of pioneering pa-
pers, Schwarz [48] developed a numerical technique to simulate the dynamics
of the vortex tangle. This was based on the classical theory of vortex filaments
in an incompressible Euler fluid. The reconnection of vortex lines was therefore
forbidden by Kelvin’s theorem, and he was therefore compelled, when studying
changes in vortex line topology, to introduce ad hoc reconnection rules, e.g., that
vortex filaments reconnect if, and only if, they approach within a distance of Δ
of one another, where Δ is an ad hoc constant of the order of the core radius.
As we saw in §2 above, Kelvin’s theorem does not apply to GP models when a
zero of ψ lies on the circuit Γ in (22). The coalescence of vortex lines merely
corresponds to the merging of two zeros of ψ, and the creation of vortices corre-
sponds to the appearance of a new zero line of ψ. These processes were first seen
to happen in the GP calculations of Jones and Roberts [33] and later, in a more
graphic form, by Koplik and Levine ([35] and [36]), who simulated the interac-
tion between straight-line vortices and their reconnection. They also witnessed
the annihilation of vortex rings of similar radii. Vortices can also be nucleated
by moving bodies such as ions, but again this process cannot occur in an Euler
fluid, but is allowed by the GP models, as is seen in the next Section. Another
grave disadvantage of using the incompressible Euler fluid to model the super-
fluid is that it eliminates sound in transferring, and perhaps cascading, energy
down the turbulence spectrum. Vinen (this meeting) has stressed the potential
importance of this process. Although the simplest form (8) of GP theory gives
c incorrectly (see §1), the generalized theory to be described in §8 below, may
provide a viable route to progress. Svistunov [51] has stressed the role of Kelvin
waves in the turbulent cascade.
In Fig. 1, the results of unpublished calculations by Berloff of the evolution of
two-dimensional solutions of (8) are shown, in terms of the density ρ which, being
depleted near a vortex core, is a clear marker for the line vortices. The walls of the
computational box were taken to be reflective. It may be seen that the number
of vortex lines is not conserved during the simulation; vortex nucleation and
annihilation occurs. At the start there were 4 vortices of positive circulation and
4 of negative circulation. Over the period of integration the energy is increasingly
transferred to sound waves and the number of vortices gradually diminishes.

6 Intrinsic Vortex Nucleation

In this section we consider the nucleation of quantized vortices from the stand-
point of GP theory. Understanding vortex nucleation and the critical veloci-
ties associated with it is one of the most significant questions of superfluidity.
 Nonlinear Schrödinger Equation 245

Fig. 1. The evolution of eight straight-line vortices in 2D within a box, four of each
direction of circulation. As a result of the interactions, the number of vortices decreases
and their energies are transmitted to sound waves

Breakdown of ideal superflow in channels and in rotating containers is usually

attributed to the formation of vortices. Flow through apertures and around im-
purities is known to generate vortex rings and vortex loops. According to Don-
nelly [18] we should distinguish between ‘extrinsic nucleation’, which concerns
the growth of pre-existing vorticity from ‘intrinsic nucleation’, which is creation
of vorticity from “nothing”. It is virtually impossible to distinguish between
these two experimentally. The energy of the trapped vortex may be close to the
energy of an elementary excitation such as a roton. It could have been excited
thermally and remained trapped at a pinning center during the cooling down
of the superfluid to low temperatures, and then be enlarged and “released” by
superflow past that center.
The intrinsic nucleation is hard to understand as there is no truly microscopic
theory of the superfluid. We do not know how to describe the processes on the
scale of the coherence length, where the motion is governed by the full quantum
many-body structure of the superfluid. We confine attention here therefore to two
possible approaches to vortex nucleation. The first is based on the GP model; the
second relies on semiclassical, hydrodynamic (large-scale) theory with the vortex
assumed ab initio with tunneling introduced in an ad hoc way. As was emphasized
by Fischer [20], and foreshadowed long ago by, for example, Schwarz and Jang
[50], there are difficulties faced by the quasi-classical theory of nucleation that
render it incapable of describing intrinsic nucleation. A semiclassical description
is valid only if the quantum core structure of the tunneling object does not come
246 P.H. Roberts and N.G. Berloff

into contact with the boundary. The principal advantage of using the GP model
is that semiclassical and tunneling processes are joined seamlessly.
Vortex nucleation by an impurity such as the positive ion 4 He+ 2 moving in
superfluid helium at low temperature with velocity v has been studied experi-
mentally and theoretically (see, e.g. Donnelly, 1991), and has uncovered some
interesting physics. The flow round an ion that is moving with a sufficiently
small velocity, v, is well represented by one of the classical solutions of fluid
mechanics, namely the flow of an inviscid incompressible fluid around a sphere.
In this solution, the maximum flow velocity, u, relative to the sphere is 3v/2,
and occurs on the equator of the sphere (defined with respect to the direction
of motion of the sphere as polar axis). Above some critical velocity, vc , the ideal
superflow around the ion breaks down, leading to the creation of a vortex ring
(Rayfield and Reif, 1964). Muirhead et al. [37] created a theory of vortex nu-
cleation that allowed them to estimate three useful quantities (i) vc , (ii) the
form of the potential barrier that must be overcome for the creation of vortices
both as encircling rings and vortex loops, and (iii) the nucleation rate. These
calculations were carried out for a smooth rigid sphere moving through an ideal
incompressible fluid using the semiclassical approach.
An important scale defined by the condensate model is the ‘healing length’,
a, defined in (19). This determines the radius of a vortex core and the thickness
of the ‘healing layer’ that forms at a potential barrier (such as the ion surface in
GP model). The radius, b, of the ion is large compared with a, and asymptotic
solutions for  ≡ a/b → 0 become relevant. Such a solution has two parts, an
interior or ‘boundary layer’ structure that matches smoothly to an exterior or
‘mainstream’ flow. Berloff and Roberts [6] showed that the dimensionless flow
velocity U = u/c is greatest on the equator of the sphere where it is
Uθ = 3V /2 + 0.313V 3 + 0.3924V 5 + 0.648V 7 + 1.24V 9 + 2.63V 11 + · · · (39)
 
+ 2.12V + 1.58V 3 + 2.89V 5 + · · · ,

where V = v/c. This gives vc ≈ 0.53c for  → 0. (We recall here that the criterion
u = c for criticality strictly applies only for  = 0; see §3.)
There is some similarity between the flow of the condensate past the ion
and the motion of a viscous fluid past a sphere at large Reynolds numbers,
the healing layer being the counterpart of the viscous boundary layer. There
are, however, important differences. At subcritical velocities, the flow of the
condensate is symmetric fore and aft of the direction of motion, and the sphere
experiences no drag. In contrast, the viscous boundary layer separates from the
sphere, so evading D’Alembert’s paradox, destroying the fore and aft symmetry,
and therefore bringing about a drag on the sphere. Moreover, when v > vc ,
shocks form at or near the sphere, but shocks are disallowed in the condensate
since they represent a violation of the Landau criterion and a breakdown of
superfluidity. When v > vc , the condensate evades shocks through a different
mode of boundary layer separation. The sphere sheds circular vortex rings that
move more slowly than the sphere and form a vortex street that trails behind it,
maintained by vortices that the sphere sheds. As the velocity of the ion increases
 Nonlinear Schrödinger Equation 247

such a shedding becomes more and more irregular. Each ring is born at one
particular latitude within the healing layer on the sphere. As it breaks away
into the mainstream, it at first contributes a flow that depresses the mainstream
velocity on the sphere below critical. As it moves further downstream however,
its influence on the surface flow diminishes. The surface flow increases until it
again reaches criticality, when a new ring is nucleated and the whole sequence
is repeated. The vortex street trailing behind the ion creates a drag on the ion
that decreases as the nearest vortex moves downstream, but which is refreshed
when a new vortex is born.
Berloff [4] considered the vortex nucleation when the symmetry of the system
is broken in the presence of the random noise or by introducing a plane wall at
some distance from the ion. Figure 2 shows the formation and evolution of a
vortex loop on the positive ion when it moves subcritically. As the velocity
reaches criticality on the side of the sphere closer to the boundary, a vortex loop
appears, spreads laterally, interacts with the boundary, the feet of the vortex
line on the surface of the ion come closer to one another, and they detach to
form a loop on the boundary.
Frisch et al. [21] and Winiecki et al. [55] have solved numerically the conden-
sate equation for flow past a circular cylinder and Berloff and Roberts [6] for the
flow around a positive ion and have confirmed the main features of the scenario
just described. Below the critical velocity steady solutions of the GP model exist.
Huepe and Brachet [31] numerically computed stationary steady and unsteady

Fig. 2. The results of numerical integration of (3) for the positive ion moving with
the velocity 0.51c for the isosurface  = 0.2∞ at (a) t = 132a/c, (b) t = 168a/c, (c)
t = 216a/c, and (d) t = 240a/c.(After [4])
248 P.H. Roberts and N.G. Berloff

solutions for flow around a cylinder. They confirmed the critical velocity of nu-
cleation vc ≈ 0.42c found numerically by Frisch et al. [21] and Winiecki et al.
[55] and through the asymptotic expansion by Berloff and Roberts [6]. Winiecki
et al. [56] carried out integrations similar to those of Huepe and Brachet [31] but
for three dimensional flow around a penetrable sphere. They also found three
branches of steady solutions that exist up to the nucleation point. In order of
decreasing energy these are (i) irrotational flow round the sphere, (ii) a vortex
loop attached to the sphere, and (iii) a vortex ring surrounding the sphere sym-
metrically. These three solutions are all stable. Perhaps a transition from one
branch to another can be achieved by quantum tunneling in the way envisaged
by Muirhead et al. [37].
In addition to the positive ion, other impurities have proved to be useful
experimental probes, including neutral atoms such as 3 He (radius ∼ 4Å) and
the negative ion, which is an electron in a bubble cut out of the liquid owing to
the repulsive interaction between the electron and surrounding helium atoms.
We now consider the negative ion.
In the Hartree approximation, the equations governing the one particle wave-
function of the condensate, ψ, and the wave function of the electron, φ, are a
pair of coupled equations suggested by Gross [27] and by Clark [14], [15]:

∂ψ 2 2
i =− ∇ ψ + (U0 |φ|2 + V0 |ψ|2 − E)ψ, (40)
∂t 2M

∂φ 2
i = − ∇2 φ + (U0 |ψ|2 − Ee )φ, (41)
∂t 2μ
where M and E are the mass and single particle energy for the bosons; μ and Ee
are the mass and energy of the electron. The interaction potentials between boson
and electron and between bosons are here assumed to be of δ−function form
U0 δ(x − x ) and V0 δ(x − x ), respectively. To lowest order, perturbation theory
predicts such interaction potentials to be U0 = 2πl2 /μ and V0 = 4πd2 /M ,
where l is the boson-impurity scattering length, and d is the boson diameter.
The normalization conditions on the wave functions are
 
|ψ|2 dV = N, |φ|2 dV = 1. (42)

Using the system (40)-(41), Grant and Roberts [25] studied the motion of a
negative ion moving with speed v using an asymptotic expansion in v/c, where
c is the speed of sound, so that their leading order flow is incompressible. Treat-
ing  ≡ (aμ/lM )1/5 as a small parameter they calculated the effective (hy-
drodynamic) radius and effective mass of the electron bubble. By employing a
convergent series expansion suitable for u = O(c), Berloff and Roberts (2000)
determined vc in the limit  → 0. They have shown that vc for the negative ion is
about 20% less than vc for the positive ion, in agreement with the experimental
findings of Zoll (1976); see also Table 8.2 of Donnelly (1991). This reduction
may be attributed to the flattening of the electron bubble by its motion through
 Nonlinear Schrödinger Equation 249

the condensate. The “equatorial bulge” is created by the difference in pressure

between the poles and equator associated with the greater condensate velocity
at the latter than at the former. The existence of the bulge also enhances these
differences in velocity (and pressure), as compared with a spherical impurity,
with the result that, if v is gradually increased from zero, the flow ue on the
equator of the electron bubble attains the velocity of sound before ue does for
the positive ion.
Finally, we mention the nucleation of the vorticity by superfluid flow through
apertures. Burkhart et al. [11] considered the properties of a vortex close to a
wall as derived from the GP model and obtained a critical velocity and vortex
radius at nucleation that agree with experiments.

7 Capture of Impurities by Vortex Lines

Rayfield and Reif [43] used an ion time-of-flight spectrometer to determine the
dynamics of ion-quantized vortex ring complexes. They observed that above
some critical velocity, vc , ideal superflow around an ion breaks down. The moving
ion produces vortex rings and the ion becomes trapped in one of these. The
capture of negative ions by quantized vortex lines in a rotating bucket was first
demonstrated by Careri et al. [12]. Schwarz and Donnelly [49] observed that at
low temperatures (< 0.5 K) straight-line vortices can trap positive ions.
Berloff and Roberts [8] solved equations (40)-(41) numerically to observe and
elucidate the process of capture; see Fig. 3. They have shown that this process
can be better characterized as the reconnection of the vortex line with its pseudo-
image inside the impurity. Initially the vortex line bends towards the impurity
at the point where the distance between them is least. As the result of this
interaction the impurity moves around the vortex axis. The process of capture
continues as the vortex line terminates on the surface of impurity and its feet
move to opposite poles of the impurity surface. At the same time helical waves
start propagating from the impurity along the two segments of the vortex line.
Such helical waves have been observed during the relaxation of the vortex angle
when two vortex lines reconnect [28], and are just Kelvin waves. These waves,
and associated sound waves, radiate energy to infinity.
During the time in which the vortex merges with the healing layer round the
ion (now the layer of thickness ∼ a in which neither φ in (40) nor ψ in (41) can
be neglected), the character of the solution alters rapidly, corresponding to the
topological change that defines the capture of the ion by the vortex. Once the
vortex has divided into two, with separate feet attached to the healing layer, the
flow round the ion has acquired circulation that it previously could not possess.
It is clear from the discussion above that the trapped impurity is in the
lower energy state than the free impurity. The difference V is the ‘substitution
energy’, also called the ‘binding energy’. Donnelly and Roberts [17], using the
healing model of the vortex core, estimated that
 
V = 2πρs (/M )2 b (1 + a2 /b2 )1/2 sinh−1 (b/a) − 1 . (43)
250 P.H. Roberts and N.G. Berloff

Fig. 3. The results of numerical integration of (40)-(41) for the negative ion initially
placed a distance 6a apart from the rectilinear vortex line. The pictures show the
isosurface  = 0.2∞ . (After [8])

The more sophisticated calculations of Berloff and Roberts [8] gave 21.7 K for a
positive ion of radius 10a, where a = 0.47 Å for the nondimensionalization based
on the speed of sound. If instead a = 1 Å is used, they obtained V /kB ≈ 46 K,
which may be compared with 33.5 K according to (43). For the electron bubble
 Nonlinear Schrödinger Equation 251

our result is V /kB = 55 K, and (43) gives V /kB = 66 K if R = 16 Å is used.

The healing layers around the positive and negative ions differ dramatically and
the healing layer close to the vortex feet is less depleted, which was not taken
into account by (43). This explains why (43) overestimates V .
The capture of a penetrable sphere moving with supercritical velocity by
a vortex ring that it itself created was simulated by Winiecki and Adams [57]
through the full 3D numerical integration of the GP equation, supplemented
with the calculation of the drag on the surface of the sphere. They have shown
that after the positive ion emits the vortex ring, the drag on the ion slows it
down and the ion is captured by the vortex ring; see Fig. 4.

Fig. 4. Capture of the penetrable sphere by the vortex it itself created for different
values of the applied force. (After [57])

Berloff and Roberts [7] performed integrations of (40)-(41) for the following
configuration: a moving negative ion catches up with a vortex ring moving in
the same direction. The axis of the ring does not coincide with the axis of
the impurity. Such a condition is necessary to destroy the axisymmetry of the
system. Figure 5 shows the process of capture of the electron by the vortex ring.
Initially, the faster moving ion passes the vortex ring. The Bernoulli effect of the
flow propels the ion and vortex towards one another with a force approximately
proportional to s−3 , where s is the closest distance between them, similarly to
the process of capture of the ion by a straight line vortex.
252 P.H. Roberts and N.G. Berloff

Fig. 5. Capture of the moving ion by the vortex ring: the results of numerical integra-
tion of (40)-(41) for the isosurface  = 0.2∞ . (After [7])

8 Nonlocal Models
In §3 we considered different shortcomings of the GP model. The natural question
arises whether it is possible to remedy (3) to make it more quantitatively realistic
for superfluid helium. For some time there has been a belief that, as soon as
the nonlocal model (1) with a realistic two-particle potential, V , that leads to
phonon-roton-like spectra is solved, the properties of superfluid helium will be
well represented. The minimum requirements on such a potential would be (i)
the correct position of the roton minimum and (ii) the correct speed of sound.
Actually such a fit can be obtained with a variety of potentials. Pomeau and
Rica [41] pioneered the use of nonlocal models for study superfluidity, but their
model did not have the correct sound velocity (slope of the dispersion curve
at the origin). Berloff [3] investigated the applicability of (1) with a potential
that adequately represents the dispersion curve. It was shown that for liquid
helium having the correct Landau dispersion curve, solutions of equation (1)
develop non-physical mass concentrations. In particular, the “Eulerian part” of
the momentum equation (without the quantum stress tensor) may become no
longer hyperbolic in some parts of the integration volume. A virial theorem,
similar to the one used to establish the catastrophic blow-up in the focusing
nonlinear Schrödinger equation, can be used to establish similar catastrophes
in bounded volume for (1). This indicates that the assumptions underlying the
derivation of the equation break down and that higher order nonlinearities must
be introduced.
A more accurate approach in modeling liquid helium is through density-
functional theory done by Dalfovo et al. [16], which attempts to give an ade-
quate microscopic description of interactions. In this approach the total energy
is still written as a functional of the one-body density, but it includes short-range
correlations [19]. This approach has provided a quantitatively and qualitatively
 Nonlinear Schrödinger Equation 253

reliable representation of the superfluid properties of free surfaces, helium films,

and droplets (see [16] and references therein). At the same time this approach is
phenomenological and results in rather complicated forms of the energy function-
als with many parameters that are chosen to reproduce liquid helium properties.
Berloff and Roberts [5] attempted to modify the nonlocal model (1) in the
spirit of a density - functional approach, by introducing only one additional
nonlinear term in the expression for the correlation energy. This allowed them
to remedy the nonphysical features of model (1), while retaining not only an
adequate representation of the Landau dispersion relation, but also simplicity in
the analytical and numerical studies.
The correlation energy of the Skyrme interactions in nuclei [53] is given by
  
1 W0 2 W1 2+γ 2
Wc (ρ) = 2 ρ + ρ + W2 (∇ρ) , (44)
M 2 2+γ
where W0 , W1 , W2 and γ are phenomenological constants. The first two terms
give a local density approximation, and the gradient term corresponds to finite
range interactions. The necessary nonlocality of interactions was added directly
into the first term of (44) by introducing a two-body interaction potential, V (|x−
x |), so that (44) becomes
   
1 1 W1 2+γ
Wc (ρ) = 2 ρ(x)V (|x − x |)ρ(x ) dx + ρ dx. (45)
M 2 2+γ

This incorporates and generalizes the W2 interaction term in (44), which has
therefore been abandoned; V (|x − x |) is chosen so that the implied dispersion
relation is a good fit to the Landau dispersion curve. Following private commu-
nications with C. Jones, Berloff and Roberts [5] considered a potential of the
form
V (|x − x |) = V (r) = (α + βA2 r2 + δA4 r4 ) exp(−A2 r2 ), (46)
and also the slightly modified potential

V (|x − x |) = V (r) = (α + βA2 r2 + δA4 r4 ) exp(−A2 r2 ) + η exp(−B 2 r2 ), (47)

where A, B, α, β, δ and η are parameters that can be chosen to give excellent

agreement with the experimentally determined dispersion curve.
On adopting (45), one can see that (1) is replaced by

2 2
iΨt = − ∇ Ψ + Ψ |Ψ (x , t)|2 V (|x − x |) dx + W1 Ψ |Ψ |2(1+γ) . (48)
2M
with (7) this equation becomes
 
2 2   
iψt = − ∇ ψ+ψ |ψ(x , t)| V (|x − x |) dx + W1 |ψ|
2 2(1+γ)
− Ev . (49)
2M
This model not only produces the structure and energy per unit length of the
straight-line vortex that are very close to the ones obtained from the Monte
254 P.H. Roberts and N.G. Berloff

Carlo simulations by Sadd et al. [46], but it also made it possible to bring
the vortex core parameter (35) and the healing length into agreement. Figure
6 compares the experimentally determined dispersion curve with that employed
by (49). The insets give the density in the core of the straight line vortex and in
the healing layer at a solid boundary, both for (49) and for the GP model. The
vortex rings of large radii satisfy (36)-(38). Berloff and Roberts [5] integrated
(48) numerically to elucidate the behavior of the small vortex ring. The Berloff-
Roberts calculations indicate that when the velocity of the vortex ring reaches
the Landau critical velocity the ring becomes unstable and evanesces into sound
waves. For any ring traveling with speed greater than the Landau critical velocity,
the amplitude of the far-field solution will not decay exponentially at infinity,
which makes the existence of such a ring impossible. One of the goals of these
calculations was to clarify Onsager’s concept of the roton as “the ghost of a
vanished vortex ring.” One can hope that the transition from the vortex ring
to the sound pulse and the concomitant loss of vorticity would occur close to
the roton minimum in energy-momentum space, or (more probable) close to the
point where the group velocity and the phase velocity are equal (the Landau
critical velocity uL ). Their calculations show that indeed there is a point on
the pE− plane where the ring ceases to exist and where uL = ∂E/∂p, but this
point lies far from the roton minimum. It remains to be seen whether the idea
of the roton as a ghostly vortex ring will ever be vindicated. As one has a great
variety of potentials that lead to the Landau dispersion curve one can tune the
parameters so that the line E = uL p, meets the pE− curve for the family of the
vortex rings, to allow this sequence of vortex rings to be terminated at a lower
energy and momentum level. Whether this process will lead to coalescence with
the roton minimum is not yet clear.
Berloff and Roberts [9] used (49) to elucidate the differences between the
processes of vortex nucleation and roton emission. They argued that vortices
are nucleated when the velocity around the positive ion exceeds the velocity of
sound. The moving ion generates rotons when it moves with the velocity greater
than the Landau critical velocity.

9 Conclusions
Despite the fact that the interatomic spacing in helium is short compared with
the coherence length, the GP equation has proved itself useful in modeling, in at
least a qualitatively faithful way, many of the phenomena that have been stud-
ied experimentally. This is partially a result of the great strides taken by com-
puter technology during the past decade that have made it possible to undertake
three-dimensional simulations of processes such as vortex-vortex and vortex-ion
interactions, including the nucleation of vortices by ions and the capture of ions
by vortices. It has also become possible to explore more general equations of
Gross-Pitaevskii type from which a more realistic dispersion curve emerges, one
that possesses a roton minimum at approximately the correct momentum and
energy.
 Nonlinear Schrödinger Equation 255

Fig. 6. The dispersion relation p − E. The solid line corresponds to the nonlocal poten-
tial V (|x − x |) with A = 0.9, χ = 0.2, and γ = 1. The dots are based on experiment.
The insets show (a) the amplitude |ψ|/ψ∞ of the straight line vortex for the nonlocal
model (49) (solid line) and the GP model (dashed line); (b) the amplitude |ψ|/ψ∞ of
the healing layer at a solid boundary (an infinite potential barrier) placed at r = 0
for the nonlocal model (solid line) and the GP model (dashed line); (c) the potential
V (|x − x |) plotted as a function of r = |x − x |, in the nondimensional units defined
in [9]. (After [9])

One topic that has not been addressed in this review concerns generaliza-
tion to finite temperature. As is well known, the Landau two-fluid description of
helium dynamics was devised before it was demonstrated that vorticity in the su-
perfluid is quantized, and Landau theory encounters difficulties when quantized
vortices are present . These difficulties are analogous to those encountered when
the classical Euler equations are used to model the superfluid. One attempt to
modify Landau theory in order to evade these difficulties was made by Hills and
Roberts [30]. This theory was also, in a sense, a consequence of the one-fluid
theory of Putterman and Roberts [42].
Despite the successes of Gross-Pitaevskii theory and its generalizations, much
remains to be done, possibly by generalizing the theory itself, before it can
realistically be used for superfluid helium. It is fortunate however that it can be
legitimately applied to low density condensates, where the interatomic spacing
is greater than the coherence length.
This review was written by two people whose backgrounds are in fluid me-
chanics. Not surprisingly therefore the perspective throughout has been fluid
mechanical. Undoubtedly, we have neglected to cite work that we should have
256 P.H. Roberts and N.G. Berloff

referenced. We apologize for any such omissions, which arise through ignorance
and not malice.

Acknowledgments

We are grateful to “Sandy” Fetter for his criticisms of the previous version of
this review, and “Lev” Pitaevskii and “Joe” Vinen for their helpful comments.
This work is supported by the grant DMS-9803480 from the National Science
Foundation.

References
1. [Link], D.C. Samuels, G.H. Bauer, R.J. Donnelly: Physics of Fluids 9,
2631 (1997)
2. [Link], C.J. Swanson, R.J. Donnelly: J. Low. Temp. Phys.100, 385 (1995)
3. N.G. Berloff: J. Low. Temp. Phys. 116, 359 (1999)
4. N.G. Berloff: Phys. Rev. Lett., submitted (2000)
5. N.G. Berloff, P. H. Roberts: J. Phys.: Math. Gen. 32, 5611 (1999)
6. N.G. Berloff, P. H. Roberts: J. Phys.: Math. Gen. 33, 4025 (2000)
7. N.G. Berloff, P. H. Roberts: J. Phys.: Math. Gen. accepted (2000)
8. N.G. Berloff, P. H. Roberts: Phys. Rev. B., accepted (2000)
9. N.G. Berloff, P. H. Roberts: Phys. Lett. A, in press (2000)
10. J. B. Brooks, R. J. Donnelly: J. Phys. Chem. Ref. Data 6, 51 (1977)
11. S. Burkhart, M. Bernard, O. Avenel, E. Varoquaux: Phys. Rev. Lett. 72, 380
(1994)
12. G. Careri, W. D. McCormick, F. Scaramuzzi: Phys. Lett. 1, 61 (1962)
13. N.N. Carlson: Physica D 98,183 (1996)
14. R.C. Clark: Phys. Lett. 16, 42 (1965)
15. R.C. Clark: in Superfluid Helium ed. J. F. Allen (Amsterdam, North Holland,
1966)
16. F. Dalfovo, A. Lastri, L. Pricaupenko, S Stringari, J Treiner: Phys. Rev. B 52,
1193 (1995)
17. R. J. Donnelly, P.H. Roberts: Proc. Roy. Soc.A312, 519 (1969)
18. R.J. Donnelly: Quantized Vortices in Helium II, (Cambridge University Press,
Cambridge, 1991)
19. J. Dupont-Roc, M. Himbert, N. Pavlov, and J. Treiner: J. Low Temp. Phys. 81,
31 (1990)
20. U.R. Fischer: Physica B 255, 41 (1998)
21. T. Frisch, Y. Pomeau, S. Rica: Phys. Rev. Letts. 69, 1644 (1992)
22. V.L. Ginzburg, L.P. Pitaevskii: Zh. [Link]. Fiz. 34, 1240 (1958)
23. J. Grant: J. Phys. A: Gen. Phys. 4, 695 (1971)
24. J. Grant: J. Phys. A: Math., Nucl. Gen. 6, L151 (1973)
25. J. Grant, P. H. Roberts: J. Phys. A: Math., Nucl. Gen. 7, 260 (1974)
26. E. P. Gross: J. Math. Phys. 4, 195 (1963)
27. E. P. Gross: in Quantum Fluids. ed. D. B. Brewer (Amsterdam, North Holland,
1966)
28. F.R. Hama: Phys. Fluids 5, 1156 (1962)
29. R.N. Hills, [Link]: Archiv. Rat. Mech. Anal. 66, 43 (1977)
 Nonlinear Schrödinger Equation 257

30. R.N. Hills, [Link]: J. Low Temp. Phys. 40, 117 (1980)
31. C. Huepe, M. Brachet: C. R. Acad. Sci. Paris 325, 195 (1997)
32. C.A. Jones, S. J. Putterman, P.H. Roberts: J. Phys. A: Gen. Phys. 19, 2991
(1986)
33. C.A. Jones, P.H. Roberts: J. Phys. A: Gen. Phys. 15, 2599 (1982)
34. I.M. Khalatnikov: Introduction to the Theory of Superfluidity, (W.A. Benjamin,
Inc., New York Amsterdam, 1965)
35. J. Koplik, H. Levine: Phys. Rev. Letts. 71, 1375 (1993)
36. J. Koplik, H. Levine: Phys. Rev. Letts. 76, 4745 (1996)
37. C.M. Muirhead, W.F. Vinen, R.J. Donnelly: Phyl. Thans. R. Soc. Lond. A 3,
433 (1984)
38. C. Nore, M. Abid, M. E. Brachet: Phys. Fluids 9, 2644 (1997)
39. L.M. Pismen, A.A. Nepomnyashchy: Physica D 69, 163 (1993)
40. [Link]: Sov. Phys. JETP 13, 451 (1961)
41. Y. Pomeau, S. Rica: Phys. Rev. Lett. 71, 247 (1993)
42. S. J. Putterman, P. H. Roberts: Physica 117 A, 369 (1983)
43. G. W. Rayfield, F. Reif: Phys. Rev. 136, A1194 (1964)
44. P.H. Roberts, J. Grant: J. Phys. A: Gen. Phys. 4, 55 (1971)
45. G. Rowlands: J. Phys. A: Nucl. Gen. 6, 322 (1973)
46. M. Sadd, G.V. Chester, L. Reatto: Phys. Rev. Lett. 79, 2490 (1997)
47. P.G. Saffman: Vortex Dynamics, (Cambridge University Press, Cambridge, 1992)
48. K.W. Schwarz: Phys. Rev. B 38, 2398 (1988)
49. K.W. Schwarz, R. J. Donnelly: Phys. Rev. Lett. 17, 1088 (1966)
50. K.W. Schwarz, F. Jang: Phys. Rev. A 8, 3199 (1973)
51. B. V. Svistunov: Phys. Rev. B 52, 3647 (1995)
52. J. T. Tough: “Superfluid turbulence”. In Progress in Low Temperature Physics,
VIII, ed. D.F. Brewer. Amsterdam: North-Holland, 133 (1982).
53. D. Vautherin, D. M. Brink: Phys. Rev. C 5, 626 (1972)
54. S.E. Widnall, J. Sullivan: Proc. R. Soc. Lond. A 287, 273 (1973)
55. T. Winiecki, J. F. McCann, C. S. Adams: Phys. Rev. Letts 82, 5186 (1999)
56. T. Winiecki, J. F. McCann, C. S. Adams: Europhys. Letts 48, 475 (1999)
57. T. Winiecki, C. S. Adams: Europhys. Letts submitted (2000)
Vortex Nucleation and Limit Speed
for a Flow Passing Nonlinearly Around a Disk
in the Nonlinear Schrödinger Equation

Sergio Rica

Laboratoire de Physique Statistique,

Ecole Normale Supérieure,
24, rue Lhomond, 75231 Paris Cedex 05, France.

Abstract. I review a study over almost nine years in collaboration with T. Frisch,
C. Josserand and Y. Pomeau [1–5] on the problem of critical velocities and vortex
nucleation for a superflow passing around a disk in the nonlinear Schrödinger equation.

1 Introduction and Formulation of the Problem

The Gross–Pitaevskii[6] or nonlinear Schrödinger equation (NLSE for short here-
after) reads, in a dimensionless form1 :
1
i∂t ψ = − ∇2 ψ + |ψ|2 ψ. (1)
2
This is a partial differential equation for a complex wave function ψ = ψ(x, y, t).
We shall consider the boundary conditions

ψ(x, y, t) = 0 if x2 + y 2 < R , (2a)
∂y ψ = 0 at y = ±Ly , (2b)
∂x ψ = iv∞ ψ at x = ±Lx . (2c)

The boundary term (2c) imposes a flow in the x direction and for the analysis
we shall take the limits (Lx , Ly ) → ∞. This problem deals with two distinct
independent parameters the velocity v∞ and the disk radii R.
A direct numerical study of (1) with (2a,2b,2c), as well as simple analytical
arguments that I will review in this article, show that the flow around a disk
releases vortices from the perimeter of the disk creating a net drag force beyond
a critical velocity, that leads to a vortical and dissipative flow on large scales.
Vortices, when they appear in our simulations are produced by the system itself,
without being introduced at the beginning. At low speed, after some transient the
solution accomodates the boundary conditions everywhere to yield a stationary
1
In their
 original work the nonlinear term comes as a mean field interaction term:
ψ(r) V (|r − r |)|ψ(r )|2 dr . We have studied that model, in particular whenever
interparticle interaction V allows possible a roton minimum in the dispersion relation
[7].

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 258–267, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Vortex Nucleation and Limit Speed in the NLSE 259

Fig. 1. Vortex shedding by a flow (from left to right) around a disk in NLSE. Two
pairs of vortices are seen downstream. It appears that the vortices are nucleated when
the flow becomes locally (at the edge of the disk) supersonic.

velocity field without dissipation, because of d’Alembert’s paradox for perfect

fluids. Beyond a critical velocity, that we shall characterize later, vortices begin
to be released more or less periodically from the disk, yielding an average drag
on it, this drag tends linearly to zero at threshold.
This has been a long study that has spaned several years in collaboration with
T. Frisch, C. Josserand and Y. Pomeau [1,2,3,4,5]. The most interesting aspect
is that we have determined that the criteria for vortex nucleation in the NLSE is
closely related to the transonic transition [8]. Indeed, the nonlinear Schrödinger
equation could be transformed into a set of “hydrodynamical” equations for a
compressible fluid with a certain state equation in a long wavelength approxi-
mation. In ordinary fluid mechanics, this would lead to the formation of a shock
wave inside the supersonic bubble. Nothing similar is possible in NLSE, because
of the lack of built-in irreversibility, something that is necessary to balance non-
linearities inside the shock wave. In NLSE, the formation of shock waves is
replaced by the nucleation of vortices with a quantized circulation. In fact, these
vortices are topological defect solutions of NLSE [6]. In two space dimensions
they are points where the complex field ψ(x, y, t) vanishes, making a ±2π phase
jump when one turns around the defect. The aim of this article is to present a
review that summarizes the results of Refs. [1,2,3,4,5].
Writing ψ = ρ1/2 eiφ , we obtain two “real” hydrodynamical fields, ρ and φ,
representing respectively the particle fluid density and the velocity potential:

∂t ρ = −∇ · (ρ∇φ); (3a)
1 1
∂t φ = − (∇φ)2 + 1/2 ∇2 ρ1/2 − ρ. (3b)
2 2ρ
The first one is the density mass conservation equation, identifying ∇φ by the
local velocity v. Long wavelength and low amplitude perturbations around the
260 S. Rica
√ √
homogeneous state: ψ0 = ρ0 e−iρ0 t propagate with the sound speed cs = ρ0 .
In the second equation, the term 2ρ11/2 ∇2 ρ1/2 , often called quantum pressure, is
negligible for large scale flows, that is for flows with a space scale much larger
than the only intrinsic microscopic length, ξ0 = √1ρ0 (ξ0 is, also, the vortex core
size).
Therefore, the problem of a flow around a disk possesses two independent
dimensionless parameters, the Mach number M = vc∞ s
and the ratio ξR0 .
After our original work of 1991/92, several papers appeared in the literature
producing a great confusion in the basics of the problem. However, in a recent
article, Huepe and Brachet [9] re-examined the problem of (superfluid-)vortex
nucleation near an obstacle in the context of the nonlinear Schrödinger equa-
tion. This article contains good evidence for the scenario of vortex nucleation
described by our previous work in the subject .
The discovery of Bose–Einstein condensation in atomic vapors [10] opens
the way to test in a rather detailed fashion some predictions of the equilib-
rium and nonequilibrium quantum statistical mechanics. As those vapours are
weakly interacting systems there is hope to compare the predictions of the Gross–
Pitaevskii theory with experimental results. In particular the sound propagation
was tested successfully [11]. And more recently it was observed some evidence of
the existence of a superflow around an obstacle with a well defined critical speed
(around one fourth the speed of sound) [12]. Ketterle’s results agree quantita-
tively with results coming from NLSE in a rugby balloon-shape geometry in 3D,
a detailed discussion on that was given by C.S. Adams in this conference and in
Ref. [13].
After this introduction to the problem, in Sect. 2, I will derive the nonlinear
PDE (5a,5b,5c) for the stationary flow in the limit ξR0 → ∞, and discuss the
limit speed, due to change of type of (5a). In Sect. 5, I will show how to solve
this nonlinear PDE by series expansions in Mach number, leading to a critical
Mach number in the limit ξR0 → ∞:

Mc ≈ 0.36969(7) (4)

Section 4 deals with the existence of vortical unstables stationary flows of (1)
for M < Mc . Finally, in Sect. 5 I review, with the help of the Euler–Tricomi
equation, the transonic transition, and how that matches with vortex shedding.

2 Critical Velocities
Far from the disk boundary, in terms of the microscopic distance, the usual
hydrodynamic assumption holds: the quantum pressure term can be neglected.
Then one obtains an unique nonlinear equation for the velocity potential, which
 Vortex Nucleation and Limit Speed in the NLSE 261

is an equation for a compressible fluid2 :

  
1 1
0 = ∇· + (1 − (∇φ) 2
) ∇φ ; (5a)
M2 2
∂r φ = 0 at r = 1 (5b)
φ = r cos θ as r → ∞, (5c)

Here φ is the velocity potential whose gradient is normalized in absolute value

to one, as required by (5c). In this limit ( ξR0 → ∞) one may set R = 1, with
M the Mach number at infinity. (r, θ) are polar coordinates with origin at the
center of the unitary disk.
Let us point out that the boundary conditions (2b,2c) match directly with
(5c), but (2a) requires a more sophisticated argument to match (5b) in the limit
ξ0
R → 0. In fact, the inner
value of the wavefunction near the disk boundary van-

ishes as ψ ≈ λ(x, y, t)×( x2 + y 2 −R)Θ( x2 + y 2 −R)eiφ(x,y,t) , being λ(x, y, t)
a smooth non vanishing function and Θ(·) the Heaviside function. Replacing this
in (1) one gets at leading order that λ(x, y, t)r̂ · ∇φ must vanish, that is (5b).
At low Mach number the second order partial differential equation (5a) is
elliptic everywhere outside the disk. However, as M increases equation (5a) be-
comes
 hyperbolic beyond a critical Mach number Mc . This happens whenever
∂v Mv 2 + v2 (1 − v 2 ) vanishes at least at one point in the space. This criteria
gives the value for the critical Mach number, as the real root of the equation:
1 1 3
+ − (∇φ)2max = 0. (6)
M2 2 2
Here (∇φ)2max is the maximum local value of (∇φ)2 . One can show, via an
hodograph transformation (see [8]), that this maximum is reached always at the
boundary of the disk. Since the explicit value |∇φ|max depends on the geometry
and the Mach number, there is no close exact solution to this problem, i.e. it is
impossible to know the exact value of the critical Mach number. One may only
estimate Mc perturbatively as in Sect. 5.
Recently, G. Richardson [14] solved numerically the problem (5a,5b,5c). After
the numerics, he found a critical Mach number around Mc ≈ 0.373, i.e. close to
the one given by perturbation series Mc ≈ 0.36969(7).
Critical velocities depend strongly on the geometry because (∇φ)2max in (6)
does depend on it. It is instructive to repeat the argument in [1] which leads to a
zero-th order approximation for the critical speed for the case of a disk. The idea
originates by considering as a zero-th order approximation the solution for the
incompressible flow (M → 0 in (5a)) satisfying the boundary conditions (5b,5c),
2
For normal gas of polytropic index γ one has that (5a) reads
  
1 1 2 γ−2  
∇· + (1 − (∇φ) ) ∇φ = (∇φ)2 − 1 ∇2 φ.
M2 2 2

NLSE corresponds to γ = 2.
262 S. Rica

that is  
1
φ0 = r+ cos θ. (7)
r
At zero order one has that the maximum speed arises at θ = ±π/2 and |∇φ|max =
2, putting this value into (6), one gets for the critical Mach number at zero order
/
(0) 2
Mc = = 0.426401, (8)
11
our first estimate in [1].
(0)
 The same argument could be repeated for different geometries, it gives Mc =
≈ 0.589768 for a sphere in three spatial dimensions, whenever |∇φ|max = 32 .
8
23
This dependence becomes very dramatic for very sharp objects, as the one of
the figure.

i h

Fig. 2. Example of a flow on a sharp obstacle.

One may compute the incompressible flow using complex variables (here
z = x + iy):
 1/2
z2
φ0 = 1 + 2 .
h
The maximum speed diverges at the point z = ih; however there is a natural
cut-off, ξ0 . The critical Mach number appears balancing only the first and third
 1/2
(0)
terms in (6): Mc = √23 ξh0 . This dependence has been verified numerically
[3].

3 Flow Around a Disk via a Janzen–Rayleigh Expansion3

As mentioned the problem (5a,5b,5c) deals with the compressible fluid passing
around an obstacle. For the case of a disk, the effect of compressibility has been
studied as a perturbation of the incompressible solution by Rayleigh [16] (the
Janzen–Rayleigh expansion). He presented only the first order correction due the
technical difficulties. Then, a similar expansion was extended up to order M 16
3
Since my remark [5], it has appeared a paper by Berloff and Roberts [15] with similar
expansions for a disk and a sphere in 2D and 3D.
 Vortex Nucleation and Limit Speed in the NLSE 263

for the case of compressible air. However the question about the convergence of
the method remains (see the book of van Dyke [17]).
In this section, I realize a Janzen–Rayleigh expansion for low Mach number
to compute order by order the full superfluid flow around a disk and then the
critical speed is obtained.
The main difference with original method of Rayleigh is the use of the com-
plex variable z = x + iy and z̄ = x − iy, thus equation (5a) and the boundary
condition (5b) become
M 2   
∂zz̄ φ = ∂z 4|∂z φ|2 − 1 ∂z̄ φ + cc., (9a)
4
0 = z∂z φ + z̄∂z̄ φ = 0 at z = eiθ & z̄ = e−iθ . (9b)
Equation (9a) is transformed into an integro-differential equation in the com-
plex variables z and z̄:

M2  
φ(z, z̄) = φh (z) + dz̄ 4|∂z φ|2 − 1 ∂z̄ φ + cc. (10)
4
Here φh (z) is a boundary term (analytic on z), fixed by the boundary conditions
(9b) on the disk.
Now we expand the velocity potential as
φ = φ0 + M 2 φ1 + M 4 φ2 + . . .
The first order φ0 is the solution of the incompressible flow satisfying the bound-
ary conditions, that is
   
1 1 1 1
φ0 = z+ + z̄ + , (11)
2 z 2 z̄
i.e. the same as (7) after taking z = reiθ . This zero order solution ensures the
boundary condition at infinity (5c) for all the expansion, that is all higher order
must be of lower degree than z, i.e. lim|z|→∞ φn (z, z̄)/|z| → 0. At zero order one
has that the maximum speed arises at z = ±i and |∇φ|max = 2. Replacing the
later value into (6), one obtains the critical Mach number at zero order (8).
Now using φ = φ0 + M 2 φ1 in (10) one obtains the second order correction
to the critical Mach number
√
233 − 11
(1)
Mc = √ ≈ 0.390253,
2 7
and so on. After some iterations (which I have done with the help of Mathemat-
ica) one obtains the full velocity field potential φ as a series in M 2 . In particular,
for the maximum of the speed, the result is
7 44 1511639 6
|∇φ|max ≈ 2 + M 2 + M 4 + M
6 15 151200
5084105183 8 311688814107079 10
+ M + M
127008000 1760330880000
132895513753510095163 12
+ M + ... (12)
158588208979200000
264 S. Rica

This expansion, together with equation (6), leads (after terms up to M 50 and
an improvement of the convergence, see [5] for details), the final value for the
critical Mach number, Mc ≈ 0.36969(7).
Finally, the perturbation series (as proposed) cannot be convergent for any
M . As Rayleigh suggested, it is reasonable to believe that the convergence breaks
down at the transonic transition, because one cannot expand the solution for the
velocity potential into an hyperbolic domain. On the other hand, the full solution
possesses a multiple pole at z = 0. Even if this pole is inside the disk, it could
be a source of problems as M increases.
A simpler task (and probably sufficient) is the study of the local convergence
at the most singular point, that is the point at the maximum speed given by the
series (12). The study realized in Ref. [5] suggests that the radius of convergence
seems to be Mc , as expected, but the singularity seems to be non rational:
1/(M∗2 −M 2 )1.35 . However, 25 terms in (12) are not sufficient and only a powerful
method or faster computers could dimiss.

4 Unstable Solutions
As shown in [2,4] vortex nucleation happens via a saddle-node bifurcation4 ,
that is two different stationary (one stable and the other unstable) solutions
collides. The stable solution was computed via the Janzen–Rayleigh series in the
preceeding section. As argued in [2], unstable stationary solutions must contain a
vortex because it cannot exist two potential (non-vortical) solutions for equations
(5a,5b,5c). A proposed solution is one that haves two vortices in the up and down
poles of the disk (x = 0, y = ±a), as it is seen in Fig. 3.
One has up to zero order
 
1 1
φ0 = z+
2 z
    
Γ i i
+i ln (z−ia) − ln z− − ln (z+ia) + ln z+ + cc. (13)
2v∞ R a a
Here Γ is the quantum of circulation (Γ = 1 in this units and /m in real units),
the distance a (a parameter) is fixed by imposing no motion of the vortex, i.e.
one imposes dzd
φ(z = ±ia) = 0 after one removes the self-interaction. Up to zero
order one has that the distance a diverges to infinity as v∞ Γ
R
goes to zero, and
is a monotonic decreasing function of a, that reaches asymptotically a = 1 as
v∞ R
Γ → ∞. This dependence is slightly modified due to finite values of the Mach
number and the intrinsic microscopic length ξ0 . The dimensionless parameter
Γ
v∞ R can be written as the ratio of the two dimensionless parameters, in fact
ξ0
v∞ R ≡ M × R .
Γ 1

4
The one dimension case has been studied by V. Hakim [18] who has shown (ana-
lytically) the existence of a stable and an unstable solution below a critical speed
and a saddle-node bifurcation at threshold. The release of vortices is replaced by a
periodic nucleation of one dimensional dark solitons.
 Vortex Nucleation and Limit Speed in the NLSE 265

− Γ

i a

Γ
i/a
+ O
-i/a
− Γ

-i a
Γ

Fig. 3. Schematic configuration of an unstable stationary solution. The vortex circu-

lations are ±Γ , a is the distance from the upper vortex to the origin O and the images
are located at ±i/a.

Γ ξ0
One sees, that v∞ R goes to zero with R , therefore the zero-th order (13)
is the same than the non-vortical solution (11). No difference is expected at all
orders. Therefore one expects that as ξR0 → 0 the stable and unstable branches
collapse into a single one. This effect is clearly shown in figure 2 of Ref. [9], where
the energy of both stable and unstable solutions are plotted as a function of the
Mach number. Finally, near threshold one knows, after the cathastrophe theory,
that the energy barrier is universal and behaves as ΔE ∼ (Mc − M )3/2 (and not
the one suggested by Packard in his lecture).

5 The Euler–Tricomi Equation near the Transonic Region

Near threshold for the transonic transition the series expansion in Mach number
does not seem to be appropiate. We shall approximate (5a) assuming that at
the dominant order the surface of the disk is flat. In this limit, the hydrody-
namic boundary conditions are satisfied with an uniform tangent velocity field.
Corrections have to be added to this velocity field in order to take into account
the curvature of the disk boundary. The first correction is trivial, and only the
second one is crucial for the analysis. This second order correction satisfies the
Euler–Tricomi equation [8]:
− ( − y) ∂xx ϕ + ∂yy ϕ = 0 (14)
with the boundary condition
∂y ϕ ∼ −x3 ; at y = 0. (15)
I refer to Ref. [4] for the derivation of this equation and the boundary condition.
Here, the point (x = 0, y = 1) is a new origin of coordinates, the x-axis being
tangential to the disk, and the y-axis perpendicular,  ∼ (M − Mc ), and ||  1.
266 S. Rica

Because ϕ is a small correction we neglect nonlinear terms like ∂x ϕ∂xx ϕ besides

y∂xx ϕ, and the boundary condition for ϕ does not require the curved boundary.
Euler–Tricomi is, in some sense, the normal form of the transonic transition,
being ϕ its amplitude. This Euler–Tricomi equation may be interpreted as fol-
lows: −( − y) represents a generic tangential local velocity profile of an ideal
compressible flow near a body, since the local main speed diminishes as y in-
creases, that is as one moves far-away from the obstacle. The transonic transition
arises at y = .
The characteristics of (14) are sesqui-parabolas of the type
2
x = ± ( − y)3/2 + x0
3
that are inside the obstacle (y < 0) for  < 0 and move out to the real space as
 > 0.
Let us point out that the Euler–Tricomi possess multivalued solutions [8]
that might give some idea of what arises when crossing the critical velocity. In
fact, the roots z(x, y) of the cubic polynomial (as well as any linear combination
of these three roots):
z 3 + 3 (y − ) z + 3x (16)
are exact solutions of equation (14). Note that for y > , the cubic polynomial
has only one real root for all values of x, whereas, for y <  there are three real
roots inside a sesqui-parabola: |x| ≤ 23 ( − y)3/2 . This multivaluedness of the real
roots of the cubic equation (16) means that it is not possible, generally, to follow
continuously a root of (16) along a closed path around the origin. More precisely,
such solution will admit a discontinuity in the region inside the semi-cubic. Unless
one can regularize the discontinuity which arises along the multivalued region,
there is no hope of having a stationary solution of our problem5 . Finally, one
notes that the discontinuity should be regularized via the general time dependent
nonlinear problem.
In Ref. [4], we have considered this nonlinear time dependent Euler–Tricomi
equation including the quantum pressure term. The result is
−( − y)∂xx ϕ + ∂yy ϕ = ∂x4 ϕ + ∂tx ϕ + ∂x ϕ∂xx ϕ + . . . (17)
which is a kind of Kadomtsev-Petviashvili equation. The Kadomtsev-Petviashvili
equation possess exact solutions in 2D [19]; however, because of the inhomoge-
neous boundary condition one cannot use them here. We have followed a different
approach using a perturbation expansion. A solvability condition leads to an am-
plitude equation describing a saddle-node bifurcation of the type iŻ = −μ + Z 2 ,
being Z the amplitude of ϕ and μ is related to . As a result, the critical velocity
behaves as /
te ξ0
Mc + C
R
5
There is a singular exception if the discontinuity gap is 2π such a case is not generic
because the phase jump depends explicitly on .
 Vortex Nucleation and Limit Speed in the NLSE 267

(here Mc is the ξ0 = 0 limit, (4)). This behavior appears since quantum pressure
stabilizes the flow above the transonic transition.
This saddle-node bifurcation gives rise to a time-dependent dynamics cor-
responding to vortex emission; nevertheless, for that it remains to match this
outer velocity potential, solution of the Euler–Tricomi equation with an inner
solution, close to the disk boundary with a vortex. To do that, one expects to
relate the amplitude of ϕ, i.e. Z, to an order parameter which parametrizes
a continuous family of solutions of the full nonlinear Schödinger equation (1).
Jones and Roberts [20] found the kind of solution that we are interested in, con-
sisting in axisymmetric solitary structures, this solutions match to solution of
the Kadomtsev-Petviashvili equation as M → Mc .
Concluding, the author acknowledges Yves Pomeau who outlined with deep
originality our collaboration over the years. He also thanks T. Frisch and C.
Josserand for their participation and for several original contributions, and to
M.E. Brachet and V. Hakim for constant interest and enlightened discussions.

References
1. T. Frisch, Y. Pomeau and S. Rica, Phys. Rev. Lett., 69, 1644 (1992)
2. Y. Pomeau and S. Rica, Comptes Rendus Acad. Sc. (Paris), t. 316 Série II, 1523
(1993)
3. S. Rica and Y. Pomeau, “Critical Velocities and Nucleation of Vortices in a Model
of Superflow”, in Instabilities and Nonequilibrium Structures IV, eds. E. Tirapegui
& W. Zeller, Kluwer (1993)
4. C. Josserand, Y. Pomeau and S. Rica, Physica D 134, 111 (1999)
5. S. Rica, Physica D 148, 221 (2001)
6. V.L. Ginzburg and L.P. Pitaevskii, Sov. Phys. JETP 7, 858 (1958); L.P. Pitaevskii,
Sov. Phys. JETP, 13, 451 (1961); E.P. Gross, J. Math. Phys. 4, 195 (1963)
7. Y. Pomeau and S. Rica, Phys. Rev. Lett. 71, 247 (1993), Ibid. 72, 2426 (1994)
8. L.D. Landau and E.M. Lifshitz, Fluid Mechanics, Pergamon Press (Oxford 1987)
9. C. Huepe et M.E. Brachet, Comptes Rendus Acad. Sc. (Paris), t. 325 Série II, 195
(1997); Physica D 140, 126 (2000)
10. M.H. Anderson, et al. Science 269, 198 (1995); C.C. Bradley, et al. Phys. Rev.
Lett. 75, 1687 (1995); K.B. Davis, et al. Phys. Rev. Lett. 75, 3969 (1995)
11. M.R. Andrews, et al. Phys. Rev. Lett. 79, 553 (1997); Phys. Rev. Lett. 80, 2697
(1997)
12. C. Raman, et al., Phys. Rev. Lett. 83, 2502 (1999)
13. B. Jackson, J.F. McCann and C.S. Adams, Phys. Rev. A61, 051603(R) (2000)
14. G. Richardson, “Vortex motion in type-II superconductor”, D. Phil. Thesis Oxford
University, Oxford (1995)
15. N.G. Berloff and P.H. Roberts, J. Phys. A: Math. Gen. 33, 4025 (2000)
16. L. Rayleigh, Phil. Mag., 32, 1 (1916)
17. M. van Dyke, “Perturbation Methods in Fluid Mechanics”, the Parabolic Press
(Stanford 1975), pages 4, 15 and 215
18. V. Hakim, Phys. Rev. E 55, 2835, (1997)
19. S.V. Manakov et al., Phys. Lett. 63A, 205 (1977)
20. C.A. Jones and P.H. Roberts, J. Phys. A: Math. Gen. 15, 2599 (1982)
Vortices in Nonlocal Condensate Models
of Superfluid Helium

Natalia G. Berloff and Paul H. Roberts

Department of Mathematics
University of California, Los Angeles, CA, 90095-1555
nberloff@[Link], roberts@[Link],

Abstract. Nonlocal nonlinear Schrödinger equations are considered as models of su-

perfluid helium. The models contain a nonlocal interaction potential that leads to
a phonon-roton-like dispersion relation. It is shown that for any such potential the
generalized Gross-Pitaevskii (GP) model has non-physical features, specifically the de-
velopment of catastrophic singularities and unphysical mass concentrations. The GP
equation is remedied by introducing a higher order term in the local density approx-
imation for the correlation energy. The resulting theory is applied in two ways. The
family of superfluid vortex rings is derived. The nucleation of vortex rings by a moving
ion is considered.

1 Introduction
Superfluid helium at 0◦ K has a large interatomic spacing and is often described
in terms of a weakly interacting Bose gas. The imperfect Bose condensate in the
Hartree approximation is governed by equations that were derived by Gross and
by Ginsburg and Pitaevskii. In terms of the single-particle wavefunction ψ(x, t)
for N bosons of mass M, the time-dependent self-consistent field equation is

2 2
iψt = − ∇ ψ + ψ |ψ(x , t)|2 V (|x − x |) dx , (1)
2M

where V (|x − x |) is the potential

 of the two-body interactions between bosons.
The normalization condition is |ψ|2 dx = N.
The internal energy per unit volume, E, at point x and time t is given by

2 1
E(ρ) = 2
(∇ρ) + ρ(x)V (|x − x |)ρ(x ) dx , (2)
8M 2 ρ 2M 2
and the total energy, W , is
 
2
W = E(ρ) dx = (∇ρ)2 dx + Wc (ρ). (3)
8M 2 ρ

The first term on the right-hand side of (3) describes the quantum kinetic energy
of a Bose gas of nonuniform density; Wc (ρ) is a potential or correlation energy
that incorporates the effect of interactions.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 268–275, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Vortices in Nonlocal Condensate Models of Superfluid Helium 269

For a weakly interacting Bose system (1) is simplified by replacing V (|x−x |)
with a δ - function repulsive potential of strength V0 = V dx , which leads to
the local GP model. Several aspects of the local GP model are qualitatively or
quantitatively unrealistic for superfluid helium. The dispersion relation between
the frequency, ω, and wave number, k, of sound waves according to the local GP
model is  2
2 2 2 
ω =c k + k4 , (4)
2M
1
where c = (V0 ρ∞ ) 2 /M , ρ∞ = Ev /V0 . This shows that the velocity, c, of long
1
wavelength sound waves is proportional to ρ 2 (here we have replaced the bulk
density, ρ∞ , by ρ). That this is unrealistic is seen from the experiments on
Grüneisen constant UG = (ρ∂c/c∂ρ)T ≈ 2.8 at the vapor pressure [1]. Also, the
dispersion curve (4) has no roton minimum. At best, (4) describes the phonon
branch of the excitation spectrum.
There is significant interest attached to the question of whether the introduc-
tion of a realistic two-particle interaction potential, V , that leads to a phonon-
roton-like spectra in the GP model, gives a better description of the properties
of superfluid helium than the local model [2]. The minimum requirements on
such a potential would be the correct position of the roton minimum and the
correct bulk normalization (see below).

2 Applicability of the Generalized

Gross–Pitaevskii Model
We transform (1) by introducing the average energy level Ev , so that
ψ = Ψ exp(−iEv t/), and rescale it by

 
x→ x, t→ t. (5)
(2M Ev )1/2 2Ev

The dimensionless form of (1) becomes

  
∂Ψ
−2i = ∇ Ψ + Ψ 1 − |Ψ (x , t)|2 V (|x − x |) dx ,
2
(6)
∂t

with the bulk normalization condition V (|x |) dx = 1. We get the dispersion
curve by linearizing about the uniform state. We write Ψ = 1 + Ψ  and consider
plane waves of the form Re(Ψ  ) = exp i(ωt − kx). Then the dispersion relation
can be written as
 ∞
2 1 4
ω = k + 2kπ sin(kr)V (r)r dr. (7)
4 0

We require ω  (krot ) = 0, ω(krot ) = ωrot , where (krot , ωrot ) is the position of the
roton minimum on the kω− dispersion curve, which in dimensional units is found
from experiments [3] to be krot = 1.926Å−1 , and ω = 8.62K ◦ kB /. By taking
270 N.G. Berloff and P.H. Roberts

the limit of (7) for k → √ 0 and using the normalization condition, the sound
speed is found to be 1/ 2 as in the local model. By relating this to the known
value of the sound speed at low k in He II, 238 m/s, we find that the healing
length of the model (6) is fixed as [L] = 0.47Å, and therefore krot = 0.907 and
ωrot = 0.158.
After V has been selected to give a good account of the roton minimum it
is typically found that, after the normalization condition has been enforced, the
convolution 
δP 1
= ρ(x )V (|x − x |) dx (8)
δρ 2M 2
is not necessary positive. This convolution is the variational derivative of the
mechanical pressure, P , of the hydrodynamic formulation of (1) in the semi-
classsical limit ( → 0). When the integral on the right-hand side of (8) is
negative the pressure P decreases when the mass density increases, which is
unphysical.
It can be shown [4] that a virial theorem, similar to the one used in es-
tablishing the catastrophic singularities of the focusing nonlinear Schrödinger
equation, indicates that the solutions of the nonlocal model (1) blow up in a
bounded domain.
Finally, to illustrate the development of mass concentrations we solved (6)
numerically for the flow around a positive ion moving with the dimensionless
velocity U = 12 UL , where UL is the Landau critical velocity. The ion is modeled
as the infinite potential barrier so that Ψ = 0 on r = b, where b is the radius
of the positive ion. The interaction potential was used in the form suggested by
Jones [5]
V (r) = (α + βA2 r2 + γA4 r4 ) exp(−A2 r2 ), (9)
where the parameters α, β, γ, and A are chosen to give agreement with the ex-
perimentally determined dispersion curve as discussed above. Notice that the
local flow velocity is below UL everywhere. Nevertheless, persistent mass con-
centrations develop along the axis of symmetry; see Fig. 1.
Such an unphysical behavior indicates that assumptions made in the deriva-
tion of the equation must be unjustified. This difficulty could be overcome by

Fig. 1. The density plot in a cross-section of the solution of (6) for the flow around a
sphere of radius 10a moving to the right with velocity 0.5vL . After [4].
 Vortices in Nonlocal Condensate Models of Superfluid Helium 271

introducing into the weakly nonlinear theory dissipation or higher order nonlin-
ear terms. But we would like to preserve the Hamiltonian character of the GP
model. Instead, we will take the density - functional approach [6], which tries to
introduce an accurate microscopic picture of liquid helium.

3 Nonlocal Nonlinear Schrödinger Equation

The correlation energy of the Skyrme interactions in nuclei [7] is given by
  
1 W0 2 W1 2+γ
Wc (ρ) = 2 ρ + ρ + W2 (∇ρ)2 dx, (10)
M 2 2+γ

where W0 , W1 , W2 and γ are phenomenological constants. The first two terms

give a local density approximation, and the gradient term corresponds to fi-
nite range interactions. In a somewhat similar way to [6], we add the necessary
nonlocality of interactions directly into the first term of (10) by introducing a
two-body interaction potential, V (|x − x |), so that (10) becomes
   
1 1    W1 2+γ
Wc (ρ) = 2 ρ(x)V (|x − x |)ρ(x ) dx + ρ dx. (11)
M 2 2+γ

This incorporates and generalizes the W2 interaction term in (10), which has
therefore been abandoned.
On adopting (11), we find that the nonlinear Schrödinger equation replacing
(1) is

2 2
iψt = − ∇ ψ + ψ |ψ(x , t)|2 V (|x − x |) dx + W1 ψ|ψ|2(1+γ) , (12)
2M

and equation (6) is replaced by

  
∂Ψ   
−2i = ∇ Ψ + Ψ 1 − |Ψ (x , t)| V (|x − x |) dx − χ|Ψ |
2 2 2(1+γ)
. (13)
∂t

The bulk normalization condition becomes V (|x |) dx = 1 − χ. The dispersion
relation of (13) is modified in comparison with (4) by adding the term 12 (1+γ)χk 2
to its right-hand side. The bulk normalization condition gives the slope at the
origin (the dimensionless speed√of sound) as (1 + γχ)/2 and the unit of length
(healing length) as [L] = 0.47 1 + γχ Å. A fit to the Landau dispersion curve
can be obtained as in §2.
There are two logical choices of the parameter γ. First, we can view the term
W1 ρ2+γ in (10) as the second term in the nonlinear expansion of the correlation
energy in powers of ρ, and that yields γ = 1. The second possible choice is to take
γ = 2.8, which gives c ∝ ρ2.8 in agreement with the experimentally determined
Grüneisen constant UG ≈ 2.8.
272 N.G. Berloff and P.H. Roberts

Next, we shall use (13) to study the family of the vortex rings [8]. For a
vortex ring of large radius R the results for the straight-line vortex can be used
to give [9] the energy and momentum of such ring as
  
1 8R 
E = κ2 ρ∞ R ln −2+c ,
2 L

and
p = κρ∞ πR2 .
After differentiating E and p with respect to R and substituting into the Hamil-
ton’s equation v = ∂E/∂p we get the expression for the velocity of the large
vortex ring as   
κ 8R 
v= ln −1+c .
4πR L
Glaberson and Donnelly [10] used the experimental results of Rayfield and
Reif [11] on the relation between the energy and velocity of large vortex rings
to estimate the vortex core parameter L. These estimates were based on the
hollow core vortex model with c = 0 and produced L ≈ 0.81Å. Jones [5] did
similar calculations for the nonlocal model (6) and found c = −0.13, so that
L ≈ 0.71Å for the optimal choice of the parameter A. For the local GP model
with c = 0.381 the vortex core parameter is L ≈ 1.19Å. These values of L are
much larger that the healing length found from the sound speed, which is 0.47
Å for any of the above models. Jones [5] posed the question of whether a self-
consistent theory is possible, i.e., one where the vortex core parameter and the
healing length are brought into harmony. The answer is “Yes.” Our model (13)
with V (r) = V (r) = (α + βA2 r2 + γA4 r4 ) exp(−A2 r2 ) + δ exp(−B 2 r2 ) is able
to bring about agreement. For γ = 1, χ = 3.5, A = 1.6, B = 1, and δ = 1 we
numerically integrated (13) to find c = 0.1825, so that L ≈ 1Å, which is the
healing length of our model. This gives the energy of a vortex ring traveling at
27 cm/sec as 10 ev, which agrees with the experiments of Rayfield and Reif [11].
This choice of parameters is not very practical and, in the calculations below, we
shall use instead the interaction potential (9) with χ = 0.2, γ = 1, and A = 0.9,
which also represent the roton minimum satisfactorily.
A sequence of vortex rings of small radius has been derived numerically [8].
When the velocity of the vortex ring reaches the Landau critical velocity the ring
becomes unstable and evanesces into sound waves. For any ring traveling with
speed greater than the Landau critical velocity, the amplitude of the far-field
solution will not decay exponentially at infinity, which makes the existence of
such a ring impossible. Figure 2 plots the sequence of the vortex rings on the
pE plane together with the dispersion curve of (13). Note that, as parameters
χ, γ, and A are varied, the position of a corresponding family of vortex rings on
pE-plane changes dramatically relative to dispersion curve. Actually it is even
possible to bring the termination point of this family close to the roton minimum.
 Vortices in Nonlocal Condensate Models of Superfluid Helium 273

8 0

6 0

4 0

2 0

0
0 2 4 6 8 1 0
p/
Fig. 2. The dispersion relation p − E and a family of the vortex rings as solutions of
(13). Each dot represents the position of a vortex ring on pE-plane.

4 Vortex Nucleation and Roton Emission

In this section we shall use the model (13) to elucidate vortex nucleation from,
and roton emission by, moving ions [12]. The picture of nucleation that emerges
as a result of experiments performed over the years by the McClintock group in
Lancaster University (see, for instance [13]) shows that vortex nucleation and
roton emission are independent processes, and that the latter is linked to vL
but the former is not. We numerically integrated (13) for the axisymmetric flow
around the positive ion of radius b = 10 moving uniformly with velocity U .
Our numerical calculations (for the details of the numerics see [4]) indicate that,
provided U does not exceed the dimensionless Landau critical velocity UL , the
ion experiences no drag and the flow is steady in the frame of reference moving
with the ion. Notice that the velocity on the equator of the ion may exceed
UL (for incompressible flow the velocity on the equator would be 3U/2 but is
even larger when compressibility is allowed for), but this leads neither to vortex
nucleation nor roton emission.
When U > UL a modulated wave envelope is formed involving wave numbers
from the neighborhood of the Landau point, where ω  (kL ) = ω(kL )/kL , ω(k) =
ω̄(k) − U · k, and ω̄(k) refers to stagnant helium. These waves radiate energy
to infinity, resulting in drag on the ion. We have not so far been able to observe
vortex nucleation for U > UL , but we obtained insight into vortex nucleation
with the help of an artificial example which tended to confirm the hypothesis
[13] that roton emission and vortex nucleation are different processes.
Our artificial example is motivated by the fact that the critical velocity vc
for vortex nucleation by an electron bubble is (according to the local GP model)
reduced by its shape which, when moving, is oblate [14]. The presence of 3 He
274 N.G. Berloff and P.H. Roberts

would enhance this effect through the concomitant reduction in surface tension.
We can make vc even smaller by artificially increasing the flattening, to such an
extent that vc becomes less than vL , so that nucleation can be studied with the
model (13) without the complications of roton emission. We therefore consider
an ion with an oblate spheroidal surface moving in the direction of its short
(symmetry) axis with a velocity less than vL . The ratio of lengths of axes is 5.
Nucleation of vortices occurs when v = vc ≈ 0.148 ± 0.007c (when the speed
of sound is reached on the equator); see Fig. 3. To compare this with the cor-
responding result for the Bose condensate, we performed similar calculations
using the local GP model. The critical velocity of nucleation in this case was
found to be 0.205 ± 0.007c. Such a significant drop (∼ 30%) in the critical veloc-
ities between local and nonlocal model can be partially explained by the greater
compressibility of the fluid, according to the nonlocal model.

Fig. 3. The density plot in a cross-section of the solution of (13) for the flow around
an oblate spheroid (see text) moving to the right with velocity 0.156c at t = 100 (left)
and t = 300 (right). The white circles show the core of a vortex ring nucleated from
the spheroid and gradually falling astern of it. After [12].

5 Conclusions
In summary, we considered a nonlocal nonlinear Schrödinger equation (13) as a
model of superfluidity. The model has a finite range interaction potential that
leads to a dispersion curve with a roton minimum and can accommodate a more
realistic relationship between the speed of sound, the density and the pressure.
The parameters of the model can be chosen to bring the healing length into
agreement with the vortex core parameter. According to our model, there is no
drag on a positive ion moving with v < vL . As the velocity of the ion exceeds the
Landau critical velocity vL , it starts to experience drag and it creates modulated
waves with wave numbers corresponding to the roton minimum. Our model
 Vortices in Nonlocal Condensate Models of Superfluid Helium 275

failed to describe vortex nucleation in such circumstances. Nevertheless it could,

through an artificial example, provide strong indications that roton emission
and vortex nucleation are different processes, the former being connected to the
Landau critical velocity, and the latter to the speed of sound.
This research was supported by the NSF grant DMS-9803480.

References
1. J. B. Brooks, R. J. Donnelly: J. Phys. Chem. Ref. Data, 6, 51 (1977)
2. N.N. Bogolyubov: J. Phys. USSR 11 23 (1947); E. P. Gross: J. Math. Phys., 4 195
(1963); R. N. Hills and P.H. Roberts: Q. J. Mech. Appl. Math. 40 279 (1987); S.
Putterman and P. H. Roberts: in Macroscopic Theories of Superfluids (Cambridge
University Press) 4 (1991); Y. Pomeau and S. Rica: Phys. Rev. Lett. 71, 247 (1993)
3. R. J. Donnelly, J. A. Donnelly, R. N. Hills: J. Low Temp. Phys. 44, 471 (1981)
4. N.G. Berloff: J. Low Temp. Phys. 116 359 (1999)
5. C. A. Jones: private communications (1993)
6. F. Dalfovo, A. Lastri, L. Pricaupenko, S. Stringari, and J. Treiner: Physical Review
B 52, 1193 (1995) ; J. Dupont-Roc, M. Himbert, N. Pavlov, and J. Treiner: J. Low
Temp. Phys. 81, 31 (1990)
7. D. Vautherin, D. M. Brink: Phys. Rev. C 5, 626 (1972)
8. N. G. Berloff, P. H. Roberts: J. Phys. A: Math. Gen. 32 5611 (1999)
9. P.H. Roberts, J. Grant: J. Phys. A: Gen. Phys. 4, 55 (1971); C.A. Jones, P.H.
Roberts: J. Phys. A: Gen. Phys. 15, 2599 (1982)
10. W. I. Glaberson, R. J. Donnelly: in Progress in Low Temperature Physics IX ed
D.F. Brewer, North Holland, Amsterdam (1986)
11. G. W. Rayfield and F. Reif: Phys. Rev. A 136 1194 (1964)
12. N. G. Berloff, P. H. Roberts: Phys. Letts. A, 274, 69 (2000)
13. P.C. Hendry, N. S. Lawson, P.V.E. McClintock, C.D.H. Williams, and R.M. Bow-
ley: Phys. Rev. Lett. 60 604 (1988).
14. N. G. Berloff, P. H. Roberts: J. Phys. A: Math. Gen. (2000) submitted
Ginzburg–Landau Description of Vortex
Nucleation in a Rotating Superfluid

Igor Aranson1 and Victor Steinberg2

1
Argonne National Laboratory, 9700 South Cass Avenue, Argonne, Illinois 60439
2
Department of Physics of Complex Systems, Weizmann Institute of Science,
Rehovot, 76100, Israel

Abstract. Nucleation of vortices in rotating superfluid by spin-up and rapid thermal

quench is discussed in the framework of the time-dependent Ginzburg–Landau equa-
tion (TDGLE). An analysis of the instability in inhomogeneous rotationally-invariant
system results in the expression for the critical rotational velocity. A stability analysis
of multicharged vortices is presented. It is shown that they are very long-living objects
with lifetime inversely proportional to the dissipation rate. It was found by numerical
and analytical solution of the TDGLE that vortex nucleation by rapid thermal quench
in the presence of superflow is dominated by a transverse instability of the moving
normal-superfluid interface.

1 Introduction
There are two major ways to generate vorticity in rotating superfluid: to spin-up
a bucket with superfluid helium from initially steady state without any vortices
to a state with a rotation velocity Ω ≥ Ωc , where Ωc is the critical rotation speed
for a vortex nucleation, and to quench it thermally from temperature, T , above
the superfluid transition temperature, Tλ , to temperature below it. These two
different scenaria can be described by the same model, namely time-dependent
Ginzburg–Landau equation (TDGLE) for the superfluid order parameter cor-
rected for presence of normal component. This equation for superfluid helium
was first suggested by L. Pitaevskii[1,2].
Further we review our results based on the stability analysis of the TDGLE
for the rotating superfluid. The vortex nucleation in a rotating superfluid was
a rather hot topic during 60th and 70th. Remnant vorticity did not permit to
define correctly the critical rotational velocity for vortex nucleation at that time.
We believe that next generation of experiments will overcome this problem, e.g.,
in the way suggested about 25 years ago[3]. Together with new experimental
methods of vortex detection suggested recently[4,5], it can provide experimental
test of the theory.
Formation of vortices under a rapid quench is recognized as a fundamental
problem of contemporary physics [6]. Superfluid 3 He offers a unique “testing
ground” for rapid phase transitions [7]. Recent experiments where a rotating su-
perfluid 3 He was locally heated well above the critical temperature by absorption
of neutrons [8] revealed vortex formation under a rapid second–order phase tran-
sition. We will discuss the dynamics of vortex nucleation under a rapid quench
in the framework of the TDGLE.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 276–282, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Ginzburg-Landau Description of Vortex Nucleation 277

2 Spin-Up and Nucleation of Vortices

in Superfluid Helium
Let us consider a cell containing a superfluid helium rather close to the superfluid
transition temperature Tλ . When the cell is rotated with an angular velocity Ω,
the normal component is involved into a solid body rotation with Vn = Ω × r.
The superfluid component cannot participate in the uniform rotation up to the
point where the potential flow condition is satisfied, i.e. ∇ × Vs = 0.
A conventional approach to the spin-up problem is to describe it by a two-
fluid hydrodynamic model corrected by an equation for the vortex line dynamics
[10,11]. The vortex lines interact with the normal component that leads to mu-
tual friction[11]. It is evident that this hydrodynamic description does not catch
the key point of the spin-up problem, namely, the vortex nucleation which ac-
tually causes the spin-up of the superfluid component, the primary superfluid
relaxation mechanism toward a steady rotation.
Another approach to the spin-up problem, which can describe both the dy-
namics and the nucleation of quantized vortices in a superfluid helium, is to
use TDGLE together with a two-fluid hydrodynamic model[2]. Outside the vor-
tex core which is normal, one gets the superfluid velocity circulation around a
single quantum vortex κ = 2π/m [11]. Corresponding set of equations, which
describe the dynamics of the complex order parameter of the superfluid conden-
sate, Ψ = |Ψ | exp(iχ), in a rigid steady rotation of the normal component, looks
in the scaled variables as follows [2]( see for details [12]):

i  Λ 
∂t Ψ = − ΔΨ + Ψ − |Ψ |2 Ψ + (∇ − iVn )2 Ψ + Ψ − |Ψ |2 Ψ (1)
2 2
where Vn = Ωr, Vs = ∇χ, and Λ is a temperature-dependent parameter.
Equation (1) is reminiscent of that of the Ginzburg–Landau equation for
superconductors in the London limit [13]. The role of an external magnetic field
is played by the angular velocity, and of the corresponding vector-potential by the
velocity of the normal component. Then by analogy one expects that at Ω ≤ Ωc1 ,
there exists a motionless superfluid component with no vortices. At Ω > Ωc1
vortices will be nucleated and penetrate into the fluid producing a vortex lattice
in the interior of a helium container. As follows from the experiments on the
superfluid 4 He, Ωc1 is too low to be detected [11]. On the other hand, Ωc2
(which is analogous to Hc2 in superconductors and at which superfluidity will
be completely destroyed in the sample) is too high to be reached experimentally.
Equation (1) describes the spin-up of the superfluid part in a rigidly rotating
flow of the normal component. This equation is asymptotically correct in the
vicinity of the λ point. Moreover, one can speculate that the equation quali-
tatively describes some aspects of spin-up for 4 He near zero temperature with
Λ → 0. Therefore, we will consider Eq. (1) for all temperatures with Λ → 0 for
T → 0 and Λ ∼ (Tλ − T )−1/3 for T → Tλ .
The critical rotational velocity, Ωc , for the onset of the vortex nucleation for
T → Tλ can be found from linear stability analysis of a stationary solution. Then
278 I. Aranson and V. Steinberg

for this solution χ = 0, and F = |Ψ | is defined by the following equation:

∂r F
∂r2 F + + F − F 3 − Ω 2 r2 F = 0. (2)
r
Equation (2) has to be complimented by the conditions at the r = 0 and the
condition at the outer wall r = R, where R is the radius of the container. As a
boundary condition at the wall we take a condition of finite suppression of the
superfluid density by the wall, i.e., ∂r Ψ +γΨ = 0 for r = R where γ characterizes
the suppression of the order parameter. For γ → 0 we have no-flux boundary
condition (∂r Ψ = 0). Solution of Eq. (2) for arbitrary Ω and R is accessible only
numerically. Selected results are presented in Fig.1 of Ref[12].
The stationary solution is stable for Ω < Ωc and looses its stability above the
critical angular velocity Ωc . Instability of the stationary solution leads to nucle-
ation of vortices and the corresponding spin-up of superfluid. By substitution of
a solution of the form Ψ = F (r) + W (r, θ, t), where W is a small generic per-
turbation, one obtains a linear equation for W . Ωc is found from the existence
condition of the first nontrivial eigenmode satisfying the boundary conditions
[12]. In the limit of R  1 the solution is found by matching of the bulk solution
with the solution near the wall. The analysis shows that the most unstable eigen-
modes are localized√in the narrow layer of the width rb near the container wall.
We obtained rb ∼ R  R for large R. The value of most unstable azimuthal
number n and the critical frequency Ωc for the container radius R is given by
the expressions: /
3/4 1 1 Δ(γ)
n = Q(γ)R , Ωc = + 1/2 . (3)
R 3 R
The dimensionless parameters Q(γ), Δ(γ) are the functions of the suppression
rate γ, are obtained by the matching of outer and inner expansions, and are
shown in Fig. 2 of Ref [12]. One can make estimates of Ωc based √ on Eq.(3).
Indeed, in dimensional variables one gets Ωc = (/mRd )[2/3 /( 3ξ0 )] = 3.3 ×
103 2/3 /Rd 1/sec, where  = (Tλ − T )/Tλ , ξ0 = 2.74 × 10−8 cm is the correlation
length far from Tλ , and the power 2/3 is introduced to assure correct scaling of
the superfluid density near the λ point [14]. The conventional Feynman equation
[11] has different scaling with R and no temperature dependence. On the other
hand, as we discussed in Ref.[12], the temperature dependence of Ωc found is the
same as in the theory of thermal nucleation of quantized rings [11]. However, the
nucleation rate in our case was calculated up to prefactor (see Ref.[12]), while
in the former theory it was obtained from heuristic arguments. Moreover, we
consider non-uniform distribution of ρs due to Vn that is impossible task for the
equilibrium theory. The latter can become significant at small values of  and
R. Thus, at  = 10−6 and Rd = 0.1cm scaled R is of the order 103 , and the
correction to the frequency from Eq.(3) can be of the order of several percent.
We performed numerical simulations of Eq.(1), which details are presented
in Ref[12]. We observed nucleation and consequent tearing off of the vortices
for Ω ≥ Ωc irrespective of Λ. However, the character of the nucleation and
asymptotic states depends on Λ. For Λ → ∞ and slightly above Ωc we observed
 Ginzburg-Landau Description of Vortex Nucleation 279

nucleation of several vortices. Nucleation occurs at nonlinear stage of the insta-

bility when a set of single zeros (four zeros for R = 65) is torn off at the radius
R. These zeros are the seeds for the vortex cores. The vortices propagate into
the interior of the container and finally form a perfect vortex lattice, reminiscent
of that of the Abrikosov lattice [13] (see Fig. 1). Further increase of Ω results in
formation of additional vortices.

a b c

Fig. 1. Grey-coded images of |Ψ | demonstrating nucleation of vortices and creation

of the vortex lattice. The dark shade corresponds to zero of |Ψ |; the white one to its
maximum value. Vortices are seen as black dots. The parameters are: Ω = 0.01, R =
65, Λ 1. The initial condition is Ψ = 1 plus small amplitude broad-band noise. a)
t = 100Λ, b) t = 200Λ, c) t = 1100Λ.

3 Stability of Multicharged Vortices

As we pointed out above, Eq.(1) can describe qualitatively the spin-up of su-
perfluid helium also at T → 0 with Λ → 0. Then the TDGLE becomes the
nonlinear Schödinger equation (NLSE). Numerical simulations show the charac-
ter of the vortex nucleation is drastically different for Λ → 0. Typically whole
clusters of vortices are torn off. These clusters can be considered as a perturbed
multicharged vortex. The multicharged vortex is unstable and breaks down into
single charged vortices. However, the lifetime happens to be proportional to Λ
and diverges for Λ → 0. The multicharged vortices with the topological charge
±n are known to have higher energy and decay into n single-charged, or elemen-
tary vortices. Since in the NLSE with small dissipation the decay time of the
multicharged vortices can be arbitrarily large, one can expect to detect them
in experiments at very low T . Some indirect indications of the multicharged
vortices can be found in several experiments on vortex nucleation [11].
We consider the perturbative solution Ψ = [F (r) + η(x, y, t)] exp[inθ] of the
NLSE with dissipation[16]
 
∂t Ψ = (ε + i) ΔΨ + Ψ − |Ψ |2 Ψ , (4)
Here η is the complex function, and ε  1 is the phenomenological parameter
which describes the bulk dissipation of superflow towards the condensate. We
280 I. Aranson and V. Steinberg

assume that the only channel for the bulk dissipation at T → 0 is the absorption
of acoustic excitations of superflow by the normal component which presents,
e.g., due to normal 3 He atoms. This assumption is based on consideration that
the excess energy of n-charged vortex can decay at T → 0 only by acoustic
radiation of the bounding energy of n single charges. In the presence of the energy
conservation and other integrals of motion in the NLSE this transformation of
the bounding energy into acoustic field is a very slow process, i.e. multiple vortex
can be long-lived.

Fig. 2. Images of |Ψ (x, y)| for double-charged vortex(a)-(c); and triple-charged (d)-(f)
at the moments of time : (a) t=1700; (b) t=2000; (c) t=2500;(d) t=1700; (e) t=2000;
and (f) t=2500. The domain size 100 × 100 units, number of FFT harmonics 128 × 128;
ε = 0.001, boundary conditions no-flux. Single vortices are presented by black spots,
the acoustic field is seen in gray shade.

Numerical analysis of the stability problem reveals that the lifetime of the
multicharged vortex diverges as τ ∼ (ελ1 )−1 and is formally infinite for NLSE.
However, one has slower (non-exponential) instability mode at  = 0. In particu-
larly, generic perturbations grow linearly in time. In this sense the multicharged
vortex is metastable and may exist for a very long time. The instability has a
nonlinear nature and originates from the interaction between localized eigen-
modes and the continuous spectrum of the vortex radiation. The crucial point
here is that the simultaneous existence of the localized eigenmodes with n ≥ 2
and extended excitations does not contradict to the energy conservation of the
Hamiltonian NLSE, since they contribute to the energy with opposite signs. This
process is similar to the growth of waves with negative energy [17]. In Fig.2 one
clearly sees that a rotating double-charged and triple-charged vortices radiate
away the acoustic waves. Thus, the decaying multicharged vortices are effective
source of the acoustic radiation.
 Ginzburg-Landau Description of Vortex Nucleation 281

4 Nucleation of Vortices by Rapid Thermal Quench

Nucleation of vortices by neutron irradiation in 3 He-B in the presence of rotation
was studied experimentally in Ref. [8]. Ignoring non-relevant complexity of the
3
He-B specific multicomponent order parameter, we will use the TDGLE for a
scalar order parameter ψ [9]:

∂t ψ = Δψ + (1 − f (r, t))ψ − |ψ|2 ψ + ζ(r, t). (5)

Close to Tλ the local temperature is controlled by normal-state heat diffusion

and evolves as f (r, t) = E0 exp(−r2 /σt)t−3/2 , where σ is the normalized diffusion
coefficient. E0  1 determines the initial temperature of the hot bubble T ∗ due
to the nuclear reaction between the neutron and 3 He atom and is proportional
to the deposited energy E0 . The Langevin force ζ with the correlator ζζ   =
2Tf δ(r − r )δ(t − t ) describes thermal fluctuations with a strength Tf .

a c

js

b d

Fig. 3. 3D isosurface of |ψ| = 0.4 for σ = 400, E0 = 30 and k = 0.5. (a-b) Tf = 0.

Images are taken at times t = 36, 80. (c-d), Tf = 0.002, t = 24, 80.

Numerical simulations of Eq. (5) in 2D and 3D shows that without fluc-

tuations (Tf = 0) the vortex rings nucleate upon the passage of the thermal
front, Fig. 3a,b. Not all of the rings survive: the small ones collapse and only
the big ones grow. Although the vortex lines are centered around the point of
the quench, they exhibit a certain degree of entanglement. After a long transient
period, most of the vortex rings reconnect and form the almost axisymmetric
configuration. We find that the fluctuations have a strong effect at early stages:
the vortices nucleate not only at the normal-superfluid interface, but also in the
282 I. Aranson and V. Steinberg

bulk of the supercooled region (Fig. 3c), according to the Kibble-Zurek “cos-
mological” mechanism [6]. However, later on, small vortex rings in the interior
collapse, and only larger rings (primary vortices) survive and expand (Fig. 3d).
We conclude that the primary source of vortices in our numerical simula-
tions is the instability of normal-superfluid interface in the presence of flow. The
analysis results in the following expression for the number of survived vortices
√ 1/3 
(vs /vc ) − β 2 log(Tf−1 )/E0
2 2/3
N ∼ σE0 (6)

where β = const, while vs and vc are the imposed and critical GL superflow
velocity, respectively. This estimate is in agreement with simulations, see Ref.
[9]. Eq. (6) exhibits a slow logarithmic dependence of the number of vortices
on the level of fluctuations. For the experimental values of the parameters our
analysis results in about 10 surviving vortices per heating event. It is consistent
with Ref. [8] where as many as 6-20 vortices per neutron were detected.
This research is supported by US DOE, grant W-31-109-ENG-38 I.A.), and
by the Minerva Center for Nonlinear Physics of Complex Systems (V.S.).

References
1. V. L. Ginzburg and L. Pitaevskii. Sov. Phys. JETP 34, 858 (1958).
2. L. Pitaevskii, Sov. Phys. JETP 35, 282 (1959).
3. J. Hulin et al., Phys. Rev. A9, 885 (1974).
4. F. Lund and V. Steinberg, Phys. Rev. Lett., 75, 1102 (1995).
5. H. Davidovitz and V. Steinberg, Europhys. Lett. 38, 297 (1997).
6. G.E. Volovik, Physica B 280, 122 (2000); T.W.B. Kibble, J. Phys. A: Math Gen
9, 1387 (1976); W. H. Zurek: Nature 317, 505 (1985)
7. V.B. Eltsov, M. Krusius, and G.E. Volovik: cond-mat/9809125, to be published.
8. V.M.H. Ruutu et al: Nature 382, 334 (1996); V.M.H. Ruutu et al: Phys. Rev. Lett.
80, 1465 (1998).
9. I.S. Aranson, N.B. Kopnin and V.M. Vinokur: Phys. Rev. Lett. 83, 2600 (1999)
10. A. Reisenegger, J. Low Temp. Phys., 92, 77 (1993) and references therein.
171 (1985).
B/Fluids 9, 259 (1990).
11. R. J. Donnelly, Quantized Vortices in Helium II, Cambridge University Press, Cam-
bridge, 1991.
12. I. Aranson and V. Steinberg, Phys. Rev. B54, 13072 (1996-II).
13. P. G. de Gennes, Superconductivity of Metals and Alloys, (Addison-Wesley, Red-
wood City, 1989).
14. V.L. Ginzburg and A.A. Sobaynin, Sov. Phys. Usp., 19, 773 (1976); J. Low. Temp.
Phys., 49, 507 (1982).
15. Kramer, L., and W. Zimmermann, Physica D 16, 221, 1985
16. I. Aranson and V. Steinberg, Phys. Rev. B53, 75 (1996-I).
17. L. A. Ostrovsky et al., Usp. Fiz. Nauk 150, 417 (1986) [ Sov. Phys. Usp. 11, 1040
(1986)].
Weak Turbulence Theory
for the Gross–Pitaevskii Equation

Sergey Nazarenko1 , Yuri Lvov2 , and Robert West1

1
Mathematics Institute, University of Warwick, Coventry, CV4 7AL, UK.
2
Rensselaer Department of Mathematical Sciences, New York, 12180-3590, USA.

1 Motivation and Background

Recent developments in the theory of Bose-Einstein condensates have resulted
in a renewed interest in the Gross Pitaevskii equation (GPE),

i∂t ψ + ψ − |ψ|2 ψ − U ψ = 0, (1)

where the potential U is a given function of coordinate, see for example Fig.
1. In the limit U = 0, the GPE is called the defocusing Nonlinear Schrödinger
equation (NLSE), an equation which plays a central role in the understanding
of non-linear optics.
Weak Turbulence Theory (WTT) is a well developed statistical theory which
has been used to great effect in understanding the behaviour of large ensembles
of weakly nonlinear NLSE waves, [1,2,3,4]. A remarkable feature of the WTT
is that it provides a rigorous closure and allows one to obtain Kolmogorov-
type spectra (corresponding to constant spectral cascades of motion integrals)
as exact analytical solutions; this being impossible in the case of Navier-Stokes
turbulence. In particular, there are two spectra
 corresponding to the down-scale
cascade of the integral of energy,
 2 E = [(∇ψ)2
+ 12 ψ 4 ] dx, and the up-scale
cascade of “particles”, N = ψ dx, respectively. The up-scale cascade of par-
ticles generates a large-scale condensate which, in turn, modifies the turbulence
dynamics, changing the dominant process from being 4-wave to a 3-wave one.
The goal of this paper is to use the ideas developed for the NLSE to derive
a WTT for a large set of random waves described by the GPE. An interesting
picture emerges even from a naive application of the results already obtained for
the NLSE case. Imagine an arbitrary initial excitation; a superposition of modes

Fig. 1. Turbulent cascades of energy E and particle number N .

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 283–289, 2001.

c Springer-Verlag Berlin Heidelberg 2001
284 S. Nazarenko, Y. Lvov, and R. West

with energies somewhere in the middle of the potential well U , see Fig. 1. Be-
cause of the nonlinear interaction (“collisions”) there will be a redistribution of
energy E and particles N among the modes. Having in mind the NLSE results,
one can interpret the propagation of excitations towards the low levels (conden-
sation) as an up-scale cascade of particles. The spreading towards higher energy
states and spilling over the barrier (cooling) can be interpreted as a down-scale
energy cascade. Both of these processes will be described by Kolmogorov-type
energy distributions over the levels (scales) which are dramatically different from
any thermodynamic equilibrium distributions. Thus, the condensation and the
cooling rates will also be significantly different from the ones obtained based on
the assumption of a thermodynamic equilibrium and a Boltzmann distribution.
Interesting WTT solutions describing the finite-time condensation were obtained
in [4] and are described in another paper by Svistunov found elsewhere in this
volume. Further, formation of a strong condensate will eventually modify the
dynamics and change the energy distribution in a manner similar to the NLSE
case.
There is of course a problem with such a naive application of the NLSE
results to the GPE. Indeed, WTT for NLSE deals with spatially homogeneous
wave turbulence, whereas for GPE, the turbulence is trapped by an external
potential and is, therefore, intrinsicly inhomogeneous (e.g. a turbulent spot).
Additional inhomogeneity of the turbulence arises because of the condensate,
which in the GPE case is itself coordinate dependent.
The effects of the coordinate dependent potential and condensate can be
observed already at the level of the linear dynamics. This can be most eas-
ily understood using a wavepacket (WKB) formalism that is applicable if the
wavepackets’ wavelengths l are much shorter than the characteristic width of the
potential well L. This naturally leads it a small parameter ε such that
l
ε=  1.
L
The coordinate dependent potential and the condensate distort the wavepackets
so that their wavenumbers change. This has a dramatic effect on nonlinear res-
onant wave interactions because now waves can only be in resonance for a finite
time.

2 Weak Turbulence Theory for NLSE

Before considering WTT for GPE, let us briefly outline the results previously
obtained for NLSE. The main object of WTT is the spectrum of homogeneous
turbulence, n(k), which in case of NLSE is defined as
? @
ψ̂k ψ̂k = nk δ(k − k ), (2)

where the angle brackets mean an ensemble average. Starting with NLSE in
Fourier space and assuming that the amplitudes ψ̂k are small enough (for the
 Weak Turbulence Theory for the Gross–Pitaevskii Equation 285

linear wave period to be much less than the characteristic nonlinear time) one
can derive the following kinetic equation for waves (KEW) [1],
  
1 1 1 1 1
∂t nk = nk n1 n2 n3 + − − (3)
π nk n1 n2 n3
δ(k + k1 − k2 − k3 )δ(ωk + ω1 − ω2 − ω3 ) dk1 dk2 dk3 ,

where ωk = k 2 is the linear wave frequency (we use the notation k = |k|), nj =
n(kj ) and ωj = ω(kj ), j = 1, 2, 3. This equation describes spectral redistribution
due to a 4-wave resonant process, which has a characteristic time

τ4w = 1/n2 k 6 .

The right hand side of this equation is called the collision term. Like the original
NLSE, KEW conserves energy E and particles N , and there exist exact analytical
solutions to KEW corresponding the constant spectral fluxes of these invariants
[2].
An important consequence of the up-scale particle cascade is the generation
of a condensate at large scales, which can be approximated by a coordinate
√
independent solution ψ =  exp(−it), where  is a constant. In this case, one
can derive WTT by considering weak perturbations about the condensate as it
was done in [2]. The resulting expressions are quite lengthy and for our purposes
here it will suffice to note that there will be a 3-wave resonant interaction with
characteristic time
τ3w = 2/nk 3
(if k 2 greater or equal  which will be the only interesting case for our WKB
theory). Thus, the 3-wave process will dominate the 4-wave one if the condensate
is strong enough,  > nk 3 . Now, let us consider the case of inhomogeneous
turbulence trapped by an external potential. The inhomogeneity affects not only
nonlinear interactions but also the linear propagation and we, therefore, will start
by considering the linear dynamics.

3 Linear Dynamics of the GPE

We will now develop a WKB theory for small-scale wave-packets, described by
a linearised GPE, with and without a background condensate. As is traditional
with any WKB-type method we assume the existence of a scale separation ε  1,
as explained in Sect. 1. In this analysis we will take l ∼ 1 so that any spatial
derivatives of a given large-scale quantity (e.g. the potential U or the condensate)
are of order ε. The transition to WKB phase-space is achieved through the
implementation of Gabor transforms, which are defined as,

ĝ(x, k, t) = f (ε∗ |x − x0 |)eik·(x−x0 ) g(x0 , t)dx0 , (4)

where f is a rapidly decreasing function of x, for instance a Gaussian. The

parameter ε∗ is small and such that ε  ε∗  1. Hence, our kernel f varies at
286 S. Nazarenko, Y. Lvov, and R. West

the intermediate-scale. A Gabor transform can therefore be thought of a localised

Fourier transform. Physically, one can view a Gabor transform as a wavepacket
distribution function over positions x and wavevectors k.

3.1 Without a Condensate

Linearising the GPE, we describe wavepackets ψ without the presence of a con-

densate via
i∂t ψ + ψ − U ψ = 0, (5)
where U is a slowly varying potential. Note that (5) is just the linear Schrödinger
equation commonly considered in quantum mechanics. Gabor transforming this
equation and combining the result with its complex conjugate we find the fol-
lowing WKB transport equation,

Dt |ψ̂|2 = 0, (6)

where Dt ≡ ∂t + ẋ · ∇ + k̇ · ∂k represents the total time derivative along the

wavepacket trajectories in phase-space. The ray equations are used to describe
wavepacket trajectories in (k, x) phase-space,

ẋ = ∂k ω, k̇ = −∇ω. (7)

The frequency ω, in this case, is given by ω = k 2 + U , (again we use the notation

k = |k|). Equations (6) and (7) are nothing more than the famous Ehrenfest
theorem from quantum mechanics. According to (7), the wavepackets will get
2
reflected by the potential at points rR where U (rR ) = kmax . We will now move
on to consider linear wavepackets in the presence of a background condensate.

3.2 With a Condensate

One common misconception in the BEC theory is that the presence of a con-
densate acts on the higher levels by just modifing the confining potential U , see
for example [5]. If this was the case the linear dynamics would still be described
by the Ehrenfest theorem (with some effective potential), but as we will now
see this is not the case. We will assume that the condensate ψ0 is a (nonlinear)
ground state solution of equation (1). We start by considering a small perturba-
tion φ  1, such that ψ = ψ0 (1 + φ). Substituting this into (1) and linearising
with respect to φ, we find
∇ψ0
i∂t φ + φ + 2 · ∇φ − (φ + φ∗ ) = 0. (8)
ψ0

where  = (x) = |ψ0 |2 is a slowly varying condensate density. In a similar

manner to the previous section, the rest of this derivation consists of Gabor
transforming (8), combining the result with its complex conjugate and finding
a suitable wave-action variable such that the transport equation represents a
 Weak Turbulence Theory for the Gross–Pitaevskii Equation 287

conservation equation along the rays. This derivation is quite lengthy and can
be found in [6], here we will only present the results. As one might expect from a
WKB based theory, at the zeroth order of ε, we derive a dispersion relationship

ω = k k 2 + 2, (9)

which is identical to the one obtained for linear waves in the presence of a
homogeneous condensate (ε = 0) in [2]. At the first order we obtain a transport
equation given by
Dt n = 0, (10)
which represents the conservation of wave-action n along wavepacket trajectories
described by the ray equations (7). The wave-action is given by
 2
1 ωρ  A ik 2 A 
n= Reφ − Imφ  . (11)
2 k2  ω

Obviously, the dynamics in this case cannot be reduced to the Ehrenfest theorem
with a renormalised potential U . It is interesting that such a wave-action agrees
with that found in [2] for the homogeneous case with non-zero nonlinearity (ε =
0, σ = 0). This is the opposite limit to the one considered above (ε = 0, σ = 0);
see Sect. 2.

4 Applicability of WKB Descriptions

In this section we will investigate the applicability of the above theory. Firstly
let us consider the case of a steady weak condensate so that the effect of the
nonlinear term is small in comparison to the linear ones, |ψ0 |  |ψ0 |. This
corresponds to the eigenvalue problem ∂t ψ0 = −iΩψ0 . The GPE will therefore
become
Ωψ0 + ψ0 − ψ0 − U ψ0 = 0.

Fig. 2. Regions of applicability of WKB descriptions.

288 S. Nazarenko, Y. Lvov, and R. West

Since Ω is a constant we observe that the Laplacian term acts to balance the
external potential term (like in the linear Schrödinger equation) and the nonlin-
ear term can be at most as big as the linear ones Ω ∼ r12 ∼ U (r0 )  , where
0
r0 is the characteristic size of the condensate (it is defined as the condensate
“reflection” point from the condition Ω = U (r0 ), see below).
Now for a WKB description to be valid we require kr0  1, i.e. we require
the characteristic length-scale of our wavepackets to be a lot smaller than that
of the large scales. Using this fact we find k 2  r12 ∼ U (r0 )  . Therefore, the
0
condensate correction to the frequency, given by (9), is small. In other words
2
the wavepacket does not “feel” the condensate. Indeed, from kmax = U (rR ) we
have U (rR )  U (r0 ) and this implies that rR  r0 (where rR is the wavepacket
reflection point, see Fig. 2). Thus, the condensate in this case occupies a tiny
space at the bottom of the potential well and hence does not affect a wavepacket’s
motion. Therefore, a wavepacket moves as a “classical” particle described by the
Ehrenfest equations (6) and (7). In fact, in this case it would be incorrect to
try to describe the small condensate corrections via our WKB approach because
these corrections are of order  ∼ ε2 (the ε2 terms being ignored in a WKB
description).
Now consider a strong condensate such that
ψ0
Ω∼
=U + , (12)
|ψ0 |

i.e. the r dependence of the potential U is now balanced by the nonlinearity.

This case is usually refered to as the Thomas-Fermi limit, see for example the
paper by Fetter in this volume.
Furthermore, now wavepackets can “feel” the presence of a strong condensate
if  ∼ k 2 . We see that the WKB approach is applicable because k 2 ∼   r12 ∼
0
| ψ0 |
|ψ0 | .
According to the ray equations ω is a constant along a wavepackets trajec-

tory, so we can find the packet’s wavenumber from k 2 = 2 + ω 2 − . One can
see that k 2 remains positive for any value of  which means that the presence of
the condensate does not lead to any new wavepacket reflection points (at which
k would turn into zero). Thus, turbulence is allowed to penetrate into the centre
of the potential well. However, the group velocity increases when the condensate
√
becomes stronger, ∂k ω ∼ ρ. This means that the density of wavepackets de-
creases towards the centre of well. Therefore, the condensate tends to push the
turbulence away from the centre, towards the edges of the potential trap.
To summarize, in the presence of a strong condensate we have two regions of
applicability for our WKB descriptions, see Fig. 2a. Wavepackets at a position
r < r0 , in the central region of the potential well will evolve according to the
WKB-condensate description (10). The Laplacian term only becomes important
for r > r0 where  is exponentially small. In this case the Ehrenfest description
is appropriate.
 Weak Turbulence Theory for the Gross–Pitaevskii Equation 289

5 Weakly Nonlinear GPE Waves

Derivation of WTT for nonuniform turbulence governed by GPE consists of a

combination of the WKB derivation for the linear part of dynamics and a stan-
dard WTT derivation (see e.g. [2]) where Gabor transforms are used instead of
the Fourier transforms. The most straightforward example of this procedure is
when the nonlinear time (described in Sect. 2) is greater than the characteris-
tic linear time k/L (the time between consequent wavepacket reflections). This
regime corresponds to an opaque gas of wavepackets. In this case, one can ig-
nore the inhomogeneity in the derivation of the collision term found in the KEW
and simply replace any Fourier transforms by Gabor transforms multiplied by
exp(−ik · x) (this spatially slowly varying variable is directly analogous to the
Fourier transform). Note that U will not enter into the collision term because
the U -dependence of ω drops out from the 4-wave resonance condition, whereas,
in the case of turbulence about the condensate, ω is completely independent
of U . Therefore, KEW in this case can be readily rewritten by replacing the
partial time derivative on the left hand side of the standard 4-wave or 3-wave
equation, with Dt . The choice of the 4-wave or 3-wave equation depends on the
relative intensity of the turbulence n and the condensate , see the condition
given in Sect. 2. In particular, in the regions where the linear dynamics is de-
scribed by the condensate WKB the 3-wave interaction is always dominant if
max ∼ kmax2
∼ 1 (because n  1); whereas, for the Ehrenfest type WKB,
the dominant process can be either 4-wave or 3-wave. The case when the time
between consequent wavepacket reflections is less than the nonlinear time is less
obvious and requires further study.

References
1. A.C. Newell: Rev. Geophys. 6, 1 (1968)
2. S. Dyachenko, A.C. Newell, A. Pushkarev, V.E. Zakharov: Physica D 57, 96 (1992)
3. V.E. Zakharov, S.L. Musher, A.M. Rubenchik: Phys. Rep. 129, 285 (1985)
4. B.V. Svistunov: J. Moscow Phys. Soc, 1, 373, (1991)
5. C.W. Gardiner, M.D. Lee, R.J. Ballagh, M.J. Davis, P. Zoller: PRL 81, 24 (1998)
6. S. Nazarenko, Y. Lvov, R.J. West: to be published (2000)
Dissipative Vortex Dynamics and Magnus Force

L.M. Pismen

Department of Chemical Engineering and Minerva Center for Nonlinear Physics of

Complex Systems, Technion – Israel Institute of Technology, 32000 Haifa, Israel

1 Basic Equations

A single vortex in the infinite plane, emerging as a solution of a conservative PDE

with global U (1) and Galilean symmetry, is a very special object. Its singular
properties, as well as special symmetries of the governing equation, which are
broken in more realistic models, have been a source of paradoxes and erroneous
conclusions during the long history of theory of vortex motion. A detailed review
containing further references is found in the recent monograph by this author
[1]. This communication sketches main ideas and pitfalls of singular perturbation
techniques that allow to describe slow vortex motion including effects of weak
dissipation.
Our starting point is the Gross–Pitaevskii or defocusing nonlinear Schrödinger
(NLS) equation for complex field u(x) in R2 :

−iut = ∇2 u + (1 − |u|2 )u. (1)

A symmetric vortex solution with topological charge N = ±1 is expressed in

polar coordinates r, φ as u0 = ρ0 (r)e±iφ , where the real amplitude ρ0 (r) verifies

ρ0 + r−1 ρ0 + (1 − r−2 − ρ20 )ρ0 = 0. (2)

The relevant special properties of this solution are

• logarithmic divergence of the total energy of the vortex within a circle with
the radius L:
 L
, - √
E =π |∇ρ0 |2 + (ρ0 /r)2 + 12 (1 − ρ20 )2 dr = π ln(L e/a0 ); (3)
0

• invariance to planar translations and global phase shifts (the corresponding

Goldstone modes are ∇u0 and iu0 , and the conserved quantities are energy
and momentum).
• Galilean invariance:
,  -
x → x−v s t, ∇ → ∇, ∂t → ∂t −v s ·∇; u → u exp i 12 v s · x − 14 vs2 t . (4)

As a consequence of Galilean invariance, the vortex remains symmetric when

placed in a constant phase gradient v s = i(u∇u − u∇u)/|u|2 and viewed in the

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 290–296, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Dissipative Vortex Dynamics 291

respective Galilean frame, or, in other words, when advected with a constant
velocity v s .
The vortex motion is viewed at large distances, long periods of time and
slow velocities, scaled, respectively, as , 2 , and , where  is the ratio of the
microscopic scale of Eq. (1) – the “healing length” – to a relevant macroscopic
scale L. In view of Eq. (4), the static solution remains invariant in the first order
at v s = O(). This implies a remarkably simple law of vortex motion in a slowly
varying field: the vortex is embedded in the ambient flow and moves with the
local velocity. In this approximation, the vortex has no inertia.
The large-scale flow that may include many interacting vortices is conve-
niently described in the fluid-mechanical representation using rescaled coordi-
nates and time and the field variable presented in the Madelung form
√
u=  eiϑ/2 , (5)

where  = |u|2 is density and ϑ is the flow potential, defining the flow velocity
v = ∇ϑ. Then the imaginary part of Eq. (1) takes the form of the continuity
equation
t + ∇ · (v) = 0, (6)
and the real part is written as the Bernoulli equation

ϑt + 12 |v|2 + p = 0, (7)

in which p is pressure introduced via the equation of state

 
p = 2 ( − 1) − 2 −1/2 ∇2 1/2 . (8)

The far field approximation is obtained by neglecting the last term in Eq. (8),
called in the superfluid context quantum pressure. Further approximation is
incompressibility. Since  = 1 − O(2 ) outside the vortex core, the continuity
equation reduces in the leading order to ∇ · v = 0. The problem is mapped
then on classical ideal fluid dynamics: the gradient of Eq. (7) yields the Euler
equation
v t + (v · ∇)v + ∇p = 0, (9)
and the far field velocity potential ϑ verifies the Laplace equation ∇2 ϑ = 0.
This approximation is applicable as long as vortices are removed at distances
far exceeding the core size and move with a speed far less than the speed of
sound, which is of O(1) on the scale of Eq. (1) or of O(−1 ) on the scale of
Eqs. (6), (7). It is perturbed, however, in a singular way by small corrections to
the basic model that violate Galilean invariance and energy conservation. These
corrections can be properly taken into account only through matched asymp-
totic expansion combining different approximations at widely separated scales.
Moreover, weak compressibility effects become important at long times, when
acoustic dispersion drives the system to lower energy states, notwithstanding
the energy conservation.
292 L.M. Pismen

2 Magnus Force
Perturbations of different nature may cause a drift of the vortex relative to
the ambient flow. A driving force of such a drift is commonly called a Magnus
force. This force is in fact a universal notion, as it is required for closing the
overall momentum balance whenever a vortex moves relative to the ambient flow,
whatever the cause of this motion might be. This is a purely classical effect, which
can be fully described in the framework of classical fluid mechanics. Consider a
vortex motion with a constant velocity v l under the action of a point force F
applied at the vortex location. The flow field should be viewed in the comoving
frame, where it is described by the stationary Euler equation
(v − v l ) · ∇v + ∇p = F δ 2 (x). (10)
Taking the momentum balance over a large circle yields

F = {np + [n · (v − v l )](v − v l )}ds
 2π
= lim r {n(v · v l ) − (n · v)v l }dφ = 4πN J v l , (11)
r→∞ 0

where J is the operator of rotation by π/2. The contour integral is transformed

here by replacing the pressure with the help of the Bernoulli law, and observing
that the velocity field induced by the vortex is purely angular and does not
project on the normal vector n. The two terms, one due to pressure, and the
other, due to the kinetic energy, contribute equally to the final result. The motion
is across the flow in such a way that the vortex is displaced towards the region
where the induced velocity is added to the extrinsic flow, and the increased
average velocity results, by the Bernoulli law, in a reduced pressure.
The result, based on the momentum balance only and independent of the
nature of the applied force, means in fact that a force normal to the direction
of drift relative to the ambient flow must always be applied to close the over-
all momentum balance. It depends, however, on a presumption that the Euler
equation is applicable everywhere. We know that the hydrodynamic description
fails in the vortex core; thus, Eq. (11) is applicable only when the force causing
the vortex drift is accumulated outside the core region. This is true, for example,
when the drift is caused by scattering of long-wave phonons [2], but the action
of inhomogeneities or interactions with normal fluid (incorporating also short-
wave excitations) provide adverse examples, as we shall see below. Therefore
the “universality” of Eq. (11), stipulating a linear mobility relation between the
force and the drift velocity, is illusory. A force accumulating in the core can be
recognized by a divergence in the core expansion that would typically generate
a logarithmic correction to the mobility relation.
As an example, consider a model including interaction between superfluid and
normal fluid components. The field equation replacing Eq. (1) can be written in
the form
, -
(ut + iΛ v n · ∇u) = (i + Λ) ∇2 u + (1 − 14 vn2 − |u|2 )u . (12)
 Dissipative Vortex Dynamics 293

where v n is velocity of the normal fluid and Λ is a parameter characterizing the

relaxation rate. This equation retains the property of Galilean invariance (4)
that has to be complemented now by the appropriate shift of v n .
The problem of vortex motion is treated perturbatively at |v n | = O(), i.e
when the normal fluid velocity is of the same order of magnitude as the superfluid
velocity v s at typical inter-vortex separations far exceeding the size of the vortex
core. We rescale the vortex velocity as v l = V l (1 + Λ2 ), and the velocity of the
normal fluid as v n = V n (1+Λ2 ), and carry out the Galilean transformation (4)
to the coordinate frame propagating with the vortex velocity with an additional
phase shift:
, -
∂t → ∂t − vl ·∇, u → u exp − 12 i (V l + Λ2 V n ) · x . (13)

Assuming that the motion is stationary in the comoving frame, Eq. (12) is rewrit-
ten to O() as

∇2 u + (1 − |u|2 )u + Λ(V l − V n ) · ∇u = 0. (14)

The drift velocity V l can be obtained from the solvability condition after the
solution of this equation is expanded in . The perturbation is, however, singular,
as the first-order term diverges in the far field. The solvability condition should
be computed therefore in a circle of radius r0 large compared to the core size
but small on the far field scale, i.e. 1  r0  −1 , and involves both area and
contour integrals:
 r0  2π  2π
Re r dr ϕΨ · ∇u0 dφ + r0 (ϕ∂r u1 − u1 ∂r ϕ)r=r0 dφ = 0. (15)
0 0 0

where Ψ = Λ(V l − V n ) · ∇u0 is the inhomogeneity in the first-order expansion

of Eq. (14) and ϕ = ∇u0 is the translational Goldstone mode; overline denotes
the complex conjugate.
The missing value of the first-order correction u1 on the bounding circle
should be obtained by matching with the far field solution. The outer equation
should have a fitting form to insure successful matching. In the far field, Eq. (12)
is rewritten in extended coordinates X = x, T = 2 t. In the leading order, the
density ρ = 1 − O(2 ) is almost constant, while the phase equation reduces to
the convective diffusion equation that degenerates into the Laplace equation at
Λ → 0. Assuming that the vortex speed is stationary, the phase equation in the
comoving frame is
Λ(V l − V n ) · ∇θ + ∇2 θ = 0. (16)
This equation can be solved, and the function u1 ≈ eiθ(r,φ) obtained as the limit
of the solution at r → 0. Computing the integrals in Eq. (15) and reversing the
transformation (13) yields the mobility relation
 
1 v0 (1 + Λ2 )
vs − vn = (v l − v n ) + J (v l − v n )N Λ ln . (17)
1 + Λ2 Λ|v l − v n |
294 L.M. Pismen

At Λ → 0 this reduces to v l = v s , i.e. to a simple advection of the vortex by

the ambient superfluid. At Λ = 0, there is a Magnus force driving the vortex
normally to the direction of superflow. This part of the mobility relation contains
a logarithmic correction dependent on the core structure; in the standard model,
v0 ≈ 3.29. The mobility relation (17) is weakly nonlinear. The nonlinearity does
not appear in the logarithmic factor when divergences in the core integrals are
removed by means of an artificial long-scale cut-off rather than by matching with
the far region [3,4,5]. We can also see that inferring the absence of a transverse
force due to interaction with the normal fluid [6] on the basis of Eq. (11) is
erroneous.
The direction of vortex motion is generally oblique. It is directed in such a way
that like vortices repel and unlike attract each other, but two interacting vortices
never move directly along the connecting straight line. Due to the logarithmic
correction, the angle with the connecting line decreases as the vortices accelerate
at closer approach. Since the superfluid velocity in the far field obeys at Λ =
0 the heat equation rather than the Laplace equation, the vortex motion is
generally history-dependent, and is not uniquely determined by instantaneous
vortex positions. Clearly, the lifetime of a vortex pair is finite, and decreases
with growing Λ.

3 Three-Dimensional Effects
The three factors determining the motion of a line vortex in 3D are the local
curvature, the advection by superflow (or, in other words, the nonlocal action of
the phase field), and interaction with the normal fluid. The velocity due to local
curvature is computed with the help of a double matching procedure [7]. First,
the equation in the core region is rewritten in the aligned coordinate frame,
introducing a curvature correction, and the solvability condition is evaluated on
a circle large compared with the core size as above. The flow field on this circle
is obtained by solving the far field equation in two steps, matching the velocities
induced by adjacent and removed parts of the line vortex. The mobility relation
obtained in the non-dissipative case is
λ0
v 0l = v s + N κb ln , (18)

where b is the binormal to the line vortex, κ is curvature and λ0 is a numerical
parameter. One cannot, however, simply combine the curvature-induced velocity
and the dissipative terms in Eq. (17). An impediment is that Eq. (17) has been
obtained by matching with the solution of the far field equation (16) obtained
under the assumption of stationary motion. The nonlinear logarithmic correction
in Eq. (17) is the result of this matching. Generally, the argument of the log-
arithm obtained by matching with a non-stationary far field solution would be
history-dependent, in the same way as the argument of the logarithm in Eq. (18)
depends on the global shape of the line vortex. The formal analogy between the
terms containing the velocity and curvature ends at this point, since, on the one
hand, motion with a constant curvature is not a reasonable assumption, and, on
the other hand, making the local mobility relation dependent on the history of
 Dissipative Vortex Dynamics 295

motion is not practical. The remaining alternative is to treat the argument of

the logarithm as an adjustable parameter, keeping in mind that it must be an
O(−1 ) quantity. An additional coup de force is to allow for different logarithmic
factors in the velocity and curvature terms. Then the combined mobility relation
is written as
 
vl − vn Λ(v l − v n ) λ1 λ0
vs − vn = + Nl × ln − κn ln . (19)
1 + Λ2 1 + Λ2  

This relation can be further resolved with respect to the dissipative part of the
vortex drift velocity, v l − v 0l , and represented in a transparent form

v l = v 0l + αJχ (v n − v 0l ), (20)

where Jχ is the operator of rotation through angle χ about the local tangent
direction l, and

Λ2 (ν 2 + Λ2 ) ν(1 + 2Λ2 )
α2 = , tan χ = , (21)
2 2
1 + (2 + ν )Λ + Λ 4 Λ(ν 2 − 1 − Λ2 )

where ν = ln(λ1 /). This can be interpreted as a non-isotropic friction rule,

where α is the modulus of the friction coefficient, and χ is the deflection angle.
The equivalent form postulated by Schwarz [8] is
 
v l = v 0l + α1 l × (v n − v 0l ) − α2 l × l × (v n − v 0l ) . (22)

The above derivation links this expression (albeit not in an impeccable way)
to the two-fluid model, so that the phenomenological friction coefficients are
expressed as

Λ(ν 2 − 1 − Λ2 ) ν(1 + 2Λ2 )

α1 = α cos χ = , α2 = α sin χ = .
1 + (2 + ν 2 )Λ2 + Λ4 1 + (2 + ν 2 )Λ2 + Λ4
(23)
Applying this mobility relationship to a ring vortex, we can see that, due to
a large logarithmic factor, the rotation angle χ is positive, and therefore the
vector v n − v 0l , directed against the binormal, is rotated anticlockwise towards
the direction of the normal. As the ring velocity acquires a normal component,
the ring shrinks as it propagates, and its lifetime becomes finite, decreasing with
growing χ or α2 .

4 Failure of Mechanistic Reduction

We can observe that logarithmic factors in mobility relations arise whenever
divergent integrals appear in the perturbation scheme. Besides the above ex-
amples, logarithmic corrections arise when vortex motion is influenced by back-
ground inhomogeneities and inertia. The matching technique allows to overcome
divergence problems, but a remaining troublesome point is that constants under
296 L.M. Pismen

the logarithms are not universal. One needs to know the far field configuration
to compute these values, and all known analytical results are applicable, strictly
speaking, only to the configurations used in these studies, which usually imply
stationary motion.
The situation is still less certain when higher order acoustic effects are con-
sidered. A system of moving vortices may relax to a state of lower energy (even-
tually, to the quiescent state) even in a non-dissipative model Eq. (1) through
emission of second sound which is either radiated out to infinity or absorbed
on bounding walls. This effect is of second order in , as it is caused by local
density depletion and accelerations. Acoustic radiation spreading out to infinity
or adsorbed at distant walls effectively cancels the energy and momentum con-
servation and blurs the distinction between conservative and dissipative media.
Scattering of phonons on a vortex has been originally studied [9] as a model
of interaction with the normal fluid; it gives, indeed, both drag and lift com-
ponents, although the respective mobility coefficients differ substantially from
those produced by the two-fluid model (Sect. 2).
The common way to estimate the effects of acoustic emission and scattering
is (1) to compute the flow field induced by non-dissipative vortex motion; (2)
to consider the variable flow field as the source of an acoustic wave and (3) to
compute the loss of energy and momentum through radiation by estimating the
respective fluxes far from the source. This works only when vortices are confined
to a region much smaller than the overall extent of the nonlinear medium, and
allows to obtain closed equations of motion only in simplest configurations, like
a single vortex advected by an oncoming wave, a pair of rotating point vortices,
or an oscillating vortex ring [1]. In more complex configurations, there is no way
to incorporate acoustic emission into mobility relations.
The cautionary message is that any factor that causes deviation of a vortex
trajectory from the local superflow streamlines disrupts straightforward reduc-
tion of field equations to “mechanical” equations of vortex motion. The reduc-
tion to a “particle–field” description survives only in a weaker sense. It is still
possible to replace the full field equation (1) or (12) by the phase equation,
provided vortices are removed at distances far exceeding the core size, but nu-
merical matching of the inner limit of the outer solution with the perturbed core
structure is necessary to make the approximation consistent.

References
1. L.M. Pismen, Vortices in Nonlinear Fields, Oxford University Press, 1999.
2. E.B. Sonin, Phys. Rev. B 55, 485 (1997).
3. H.E. Hall and W.F. Vinen, Proc. Roy. Soc. A238, 215 (1956).
4. A. Onuki, J. Phys. C15 L1089 (1982).
5. E.B. Sonin, this issue.
6. C. Wexler, Phys. Rev. Lett. 79 1321 (1997).
7. L.M. Pismen and J. Rubinstein, Physica (Amsterdam) D47 353 (1991).
8. K.W. Schwarz, Phys. Rev. B 31 5782 (1985).
9. L.P. Pitaevskii, Sov. Phys. JETP 8 888, 1959.
Transition to Dissipation
in Two- and Three-Dimensional Superflows

Cristián Huepe1 , Caroline Nore2 , and Marc-Etienne Brachet3

1
James Franck Institute, University of Chicago, 5640 S. Ellis Avenue, Chicago, IL
60637, USA
2
Université de Paris-Sud, LIMSI, Bâtiment 508, F-91403 Orsay Cedex, France
3
Laboratoire de Physique Statistique de l’Ecole Normale Supérieure associé au
CNRS et aux Universités Paris 6 et 7, 24 rue Lhomond, 75005 Paris, France

Abstract. Vortex nucleation in two-dimensional (2D) and three-dimensional (3D) su-

perflows past a cylinder is studied. The superflow is described by a Nonlinear Schrödinger
like equation. In the 2D case, a continuation method is used to characterize the bifur-
cation of stationary states leading to vortex formation. A saddle-node followed by a
secondary pitchfork bifurcation that leads to the branch of nucleation solutions (one
vortex in an asymmetric field) is found. The dependence of the bifurcation diagram on
the ratio of the coherence length to the disc diameter is studied. Using the 2D station-
ary vortex nucleation solutions to construct the initial condition, the 3D dynamics of
a vortex pinned to the surface of the cylinder is numerically studied. Quasistationary
half-ring vortices, pinned at the sides of the cylinder, are generated after a short time.
On a longer time scale, a vortex stretching may occur, inducing dissipation and drag.
The corresponding 3D critical velocity is found to be well below the 2D one.

1 Introduction
Above a certain critical velocity, superfluid flows are known to enter a dissipative
regime. The Nonlinear Schrödinger equation (NLSE) describes the dynamics of
superfluid 4 He, at temperatures low enough for the normal fluid to be negligible.
In the homogeneous two-dimensional NLSE flow past a disc, the existence of a
transition to dissipation due to the periodic emission of pairs of counterrotat-
ing vortices was found in [1]. Although this model system is not quantitatively
equivalent to a 4 He flow, it has been studied in detail for its universal properties
and mathematical interest [2,3,4]. A closer connection to experiments was given
by the recent experimental success in producing and manipulating dilute Bose
Einstein condensates. The NLSE has been used to accurately describe the dy-
namics of such systems, allowing direct quantitative comparison between theory
and experiment [5]. In particular, a recent experiment which studies the dissipa-
tion produced in a Bose Einstein condensed gas by moving a blue detuned laser
beam through it [6] has renewed the interest in dynamics of a NLSE superflow
past an obstacle.
The present paper reviews some recent work in the NLSE description of a
superflow past a cylinder in 2D and 3D [2,3,4,7,8]. The paper is organized as
follows. In Sect. 2 we present the hydrodynamic form of the NLSE. In Sect. 3,
the numerical tools used in this work are described. The bifurcation diagram

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 297–304, 2001.

c Springer-Verlag Berlin Heidelberg 2001
298 C. Huepe, C. Nore, and M.-E. Brachet

that governs the 2D system is explained in Sect. 4. Finally, Sect. 5 explores

the possibility that a vortex stretching mechanism (as the one found in super-
fluid turbulence [9,10,11]) is responsible for low flow velocity dissipation in 3D
systems.

2 Definition of the System

In this section, we briefly present the hydrodynamic form of the NLSE that
models the effect of a moving cylinder of diameter D in a superfluid at rest.
Consider the following action functional
    
√ i ∂ψ ∂ ψ̄
A = dt 2cξ d3 x ψ̄ −ψ −F , (1)
2 ∂ t̃ ∂ t̃

where ψ is a complex field, ψ̄ its conjugate and F is the energy of the system.
The coherence length ξ and the speed of sound c (for a mean fluid density ρ0 = 1)
are the physical parameters of the superfluid. The energy F reads F = E −P ·U ,
with
  
1 1
E = c2 d3 x [−1 + V (x)]|ψ|2 + |ψ|4 + ξ 2 |∇ψ|2 − (2)
2 2
√ 
i 
P = 2cξ d3 x ψ∇ψ̄ − ψ̄∇ψ . (3)
2
Here, E is the fluid internal energy, P the fluid impulsion and U the velocity
of the moving cylinder. The potential V (r) = (V0 /2)(tanh[4(r − d/2)/Δ] − 1) is
used to represent a cylindrical obstacle of diameter D. The calculations presented
below were realized with V0 = 10 and Δ = ξ. With these values, the fluid density
inside the disc is negligible and the density boundary layer is well resolved with a
mesh adapted to the coherence length. The NLSE is the Euler Lagrange equation
corresponding to (1)
δψ i δF c  
= −√ = i√ [1 − V (x)]ψ − |ψ|2 ψ + ξ 2 ∇2 ψ + U · ∇ψ. (4)
δt 2cξ δ ψ̄ 2ξ
It can be mapped into two hydrodynamical
  equations by applying Madelung’s
√
transformation [12,13] ψ = ρ exp √iφ2cξ
. The real and imaginary parts of the
NLSE produce, for a fluid of density ρ and velocity v = ∇φ − U , the following
equations of motion
∂ρ
+ ∇ (ρv) = 0 (5)
∂t
  2√
∂φ 1 2 2∇ ρ
− U · ∇φ + (∇φ) + c [ρ − Ω(x)] − c ξ √
2 2
= 0. (6)
∂t 2 ρ

These correspond to the continuity and Bernoulli equations [14]. The last term in
(6) is a dispersive supplementary quantum pressure term that is relevant only at
 Transition to Dissipation in Two- and Three-Dimensional Superflows 299

length scales smaller than ξ. In the limit where ξ/D → 0, the quantum pressure
term vanishes and we recover the system of equations describing an Eulerian
flow. Note that the NLSE admits vortical solutions of characteristic core size
∼ ξ. These are topological defects (zeros) of ψ and thus appear as points in 2D
and lines in 3D.

3 Numerical Methods
Time integration was done by using a fractional step (Operator-Splitting) method
[15]. The bifurcation diagrams of stationary solutions to the NLSE describing a
2D superflow around a disc were obtained using Fourier pseudospectral methods
[16] and continuation techniques [17].
When the extremum of F is a local minimum, the stationary solution ψS
of (4) can be reached by integrating to relaxation the√associated real Ginzburg-
Landau equation (RGLE) [3] given by ∂ψ/∂t = −(1/ 2cξ)δF/δ ψ̄.
The unstable stationary solutions of the RGLE were found by using Newton’s
method [17]. In order to work with a well-conditioned system [18], we search for
the roots of the following associated relation
 
c  
f (ψ) = Θ−1 (1 − i σ U · ∇) + σ √ [1 − V (x)] − |ψ(t)|2 ψ(t) − ψ(t), (7)
2ξ
, √ -
with Θ = 1 − σ (cξ/ 2)∇2 . Calling ψ(j) the value of the field ψ over the j-th
collocation point, finding the roots of f (ψ) is equivalent iterating the Newton
step ψ(j) = ψ(j) + δψ(j), up to convergence.
- Every Newton step requires the
0
solution for δψ(k) of k df(j) /dψ(k) δψ(k) = −f(j) (ψ). This solution is obtained
by an iterative bi-conjugate gradient method (BCGM) [19]. Here, σ is a free
parameter that can be used to adjust the pre-conditioning of the system in
order to optimise the convergence of the BCGM [18].

4 Bifurcation Diagram and Scaling in 2D

Using the numerical techniques described above, the bifurcation diagrams were
computed for various values of ξ/D.
The functional E and energy F of the stationary solutions are shown in
Fig. 1 for ξ/D = 1/10 as a function of the Mach number (M = |U |/c). The
stable branch (a) disappears with the unstable solution (c) at a saddle-node
bifurcation when M = M c ≈ 0.4286. The energy F has a cusp at the bifurcation
point, which is the generic behaviour for a saddle-node. There are no stationary
solutions beyond this point. When M ≡ M pf ≈ 0.4282, the unstable symmetric
branch (c) bifurcates at a pitchfork to a pair of nonsymmetric branches (b).
The nucleation energy barrier of these branches is given by (Fb − Fa ) which is
roughly half of the barrier for the symmetric branch (Fc − Fa ). The arrow on
the bifurcation diagram in Fig. 1 shows the Mach number value at which a vortex
appears in unstable stationary solutions. When M n ≤ M ≤ M c , solutions are
 Transition to Dissipation in Two- and Three-Dimensional Superflows 301

0 .5

0 .4 5

M *
1
M

0 .4
M *
2

0 .3 5
0 0 .0 5 0 .1 0 .1 5 0 .2
ξ /d

Fig. 2. Saddle-node bifurcation Mach number M c (×) and pitchfork bifurcation Mach
number M pf (+), as a function of ξ/D. The dotted curve corresponds to a fit to the
polynomial law M c = K1 (ξ/D)K2 + MEuler
c
. The dashed lines M1∗ and M2∗ correspond
respectively to first and second order compressible corrections to M c (see text)

c

can be used to improve the M = 2/11 result. The next order [23] to the
√ √
critical Mach number is M2∗ = 233 − 11/(2 7) ≈ 0.3903.

5 Subcriticality and Vortex-Stretching in 3D

We used the 2D laminar stationary solution ψ0V (x, y) (corresponding to branch
(a) on Fig. 1) and the one-vortex unstable stationary solution ψ1V (x, y) (branch
(b)) to construct the 3D initial condition

ψ3D (x, y, z) = fI (z)ψ1V (x, y) + [1 − fI (z)]ψ0V (x, y) (8)

(Fig. 3a). The function fI (z) = (tanh[(z − z1 )/Δz ] − tanh[(z − z2 )/Δz ])/2, takes
the value 1 for z1 ≤ z ≤ z2 and 0 elsewhere (Δz is an adaptation length). The
resulting dynamical evolution can be schematically described in terms of short-
time and long-time dynamics.
During the short-time dynamics, the initial pinned vortex line rapidly con-
tracts, evolving through a decreasing number of half-ring-like loops, down to
a single quasi-stationary half-ring (see Figs. 3b-3d). The evolution takes place
near the plane perpendicular to the flow, provided that the initial vortex is long
enough to contract to a quasi-stationary half-ring as shown on Fig. 3d. Oth-
erwise, the vortex line collapses against the cylinder while moving upstream.
On a longer time scale, the quasistationary half-ring can either start moving
upstream or downstream. When driven downstream, the vortex loop is continu-
ously stretched while the pinning points move towards the back of the cylinder.
When the half ring moves upstream, it eventually collapses against the cylinder,
generating a laminar superflow.
302 C. Huepe, C. Nore, and M.-E. Brachet

(Aa ) (Ee )
U

( Bb )
U

( Cc )
U

( Dd )
U U U

Fig. 3. Left: Short-time dynamics

√ of a vortex pinned to the cylinder with ξ/D = 1/40,
√
|U |/c = 0.26√and Δz = 2 2ξ in the
√ [Lx × Ly × Lz ] periodicity box (Lx /D = 2.4 2π,
Ly /D = 1.2 2π and Lz /D = 0.4 2π). The surface |ψ3D | = 0.5 is shown at times (A)
t = 0, (B) t = 5ξ/c, (C) t = 10ξ/c and (D) t = 15ξ/c. Right (E): Vortex stretching at
t = 150ξ/c with |U |/c = 0.35 and ξ/D = 1/20

0 .5

0 .4

0 .3
M
0 .2

0 .1
1 /4 0 1 /3 0 1 /2 5 1 /2 0 1 /1 5 1 /1 0

ξ /D
Fig. 4. Parametric study of vortex stretching in a Mach number M versus ξ/D log-
log diagram. Circles: stretching, crosses: no stretching. The asterisk represents the
experiment reported in [6]. The solid line is the 2D saddle-node bifurcation Mach
c
number M2D . The dashed line corresponds to Rs = 5.5 (see text)

In order to distinguish between the two situations we have carried out the 3D
runs that we display schematically on Fig. 4. The runs with vortex stretching are
labeled by circles and those without by ×. Note that all runs were performed at
 Transition to Dissipation in Two- and Three-Dimensional Superflows 303

c
Mach numbers below the 2D saddle-node bifurcation Mach number M2D (ξ/D).
The critical Mach number for the onset of dissipation in the experiment where
a blue detuned laser beam is moved through a Bose Einstein condensed gas [6]
is marked by an asterisk.
Note that the numerical study presented above does not focus on the critical
c
velocity (M3D ) at which 3D stationary solutions disappear. This problem has
been recently addressed in [24,20]. Our approach is rather to propose a mecha-
c
nism and criteria for dissipation through vortex stretching at M < M3D .
A frontier between the dissipative and nondissipative cases can be drawn
approximately as the dashed line in Fig. 4, which corresponds to the expression
RS = 5.5 with RS ≡ |U |D/cξ = M D/ξ. This superfluid “Reynolds” number
is defined in the same way as the standard (viscous) Reynolds number Re ≡
|U |D/ν (with ν the kinematic viscosity). In the superfluid turbulent regime
(RS  1), RS was shown to be equivalent to the standard (viscous) Reynolds
number Re [9,10,11].
We acknowledge useful scientific discussions with V. Hakim, L. Tuckerman
and E. Varoquaux. This work was supported in part by the ECOS-CONICYT
program no. C96E01 and by the MRSEC Program of the National Science Foun-
dation under Award Number DMR 9808595. Computations were performed at
the Institut du Développement et des Ressources en Informatique Scientifique.

References
1. T. Frisch, Y. Pomeau, and S. Rica: [Link]., 69, 1644 (1992)
2. C. Huepe and M.-E. Brachet: C. R. Acad. Sci. Paris, 325(II), 195–202 (1997)
3. C. Huepe and M. E. Brachet: Physica D, 140, 126–140 (2000)
4. C. Huepe: Bifurcations et instabilités dans les condensats de Bose Einstein et les
écoulements superfluides. PhD thesis, Ecole Normale Supérieure, Paris (1999)
5. Franco Dalfovo, Stefano Giorgini, Lev P. Pitaevskii, and Sandro Stringari: Reviews
of Modern Physics, 71(3), (1999)
6. C. Raman, M. Köhl, R. Onofrio, D.S. Durfee a nd C.E. Kuklewicz, Z. Hadzibabic,
and W. Ketterle: Phys. Rev. Lett., 83(13), 2502 (1999)
7. C. Nore, C. Huepe, and M. E. Brachet: Phys. Rev. Lett., 84(10), 2191 (2000)
8. C. Huepe, S. Métens, G. Dewel, P. Borckmans, and M.-E. Brachet: Phys. Rev.
Lett., 82(2), 1616 (1999)
9. C. Nore, M. Abid, and M. Brachet: Phys. Rev. Lett., 78, 3896 (1997)
10. C. Nore, M. Abid, and M. Brachet: Phys. Fluids, 9(9), 2644 (1997)
11. M. Abid, M. E. Brachet, J. Maurer, C. Nore, and P. Tabeling: Eur. J. Mech.
B/Fluids, 17(4), 665 (1998)
12. R. J. Donnelly: Quantized Vortices in Helium II. (Cambridge Univ. Press, Cam-
bridge 1991)
13. E. A. Spiegel: Physica D, 1, 236 (1980)
14. L. Landau and E. Lifschitz: Mécanique des fluides, vol. 6. (Editions Mir, 1989)
15. R. Klein and A. J. Majda: Physica D, 53, 267 (1991)
16. D. Gottlieb and S. A. Orszag: Numerical Analysis of Spectral Methods. (SIAM,
Philadelphia 1977)
17. R. Seydel: From Equilibrium to Chaos: Practical Bifurcation and Stability Analysis.
(Elsevier, New York 1988)
304 C. Huepe, C. Nore, and M.-E. Brachet

18. C. Mamun and L. Tuckerman: Phys. Fluids, 7(1), 80 (1995)

19. W. Press, S. Teukolsky, W. Vetterling, and [Link] Numerical Recipes. (Cam-
bridge Univ. Press, Cambridge, 1994)
20. T. Winiecki,J.F. McCann, C.S. Adams: Europhys. Lett., 48(5), 475–481 (1999)
21. C. Josserand: Dynamique des Superfluides: Nucléation de vortex et Transition de
phase du permier ordre. PhD thesis, Ecole Normale Supérieure, Paris (1997)
22. C. Josserand, Y. Pomeau, and S. Rica: Physica D, 134(1), 111–125 (1999)
23. V. Hakim. private communication. 1999.
24. N.G. Berloff, and P.H. Roberts: J. Phys. A, 33, 4025–4038, 2000.
Motion of Objects
Through Dilute Bose–Einstein Condensates

C.S. Adams1 , B. Jackson1 , M. Leadbeater1 , J.F. McCann2 , and T. Winiecki1

1
Dept. of Physics, University of Durham, Rochester Building, South Road, Durham,
DH1 3LE, England. UK
2
Dept. of Applied Mathematics and Theoretical Physics, Queen’s University,
Belfast, BT7 1NN, Northern Ireland. UK

Abstract. This paper discusses the motion of objects through quantum fluids de-
scribed by the Gross Pitaevskii (GP) equation. The object moves without dissipation
at velocities below a threshold which corresponds to the critical velocity for vortex
nucleation. Above the critical velocity, vortex shedding is the dominant mechanism of
energy transfer between the object and the fluid. We compare the predictions of the GP
model with experiments on an oscillating laser beam in an alkali vapour Bose Einstein
condensate and ions in superfluid helium-4.

1 Introduction
A wide variety of quantum fluids are now accessible experimentally, ranging
from the dilute Bose-Einstein condensates produced in atomic vapours [1] to
liquid helium and superconductors [2]. The dilute Bose gas is a special case in
that the interactions can be accurately represented by a mean-field leading to
a relatively simple theoretical description based on the Gross Pitaevskii (GP)
equation, a form of the non-linear Schrödinger equation [3]. Consequently one
enticing product of the recent discovery of Bose Einstein condensation in dilute
alkali vapours is the potential for refining our understanding of quantum fluids.
The dilute Bose Einstein condensate is identical to a classical Euler fluid
except for the quantisation of circulation and the influence of the kinetic energy
term appearing in the GP equation. The latter produces fluid healing and shear
stresses which enable vortex formation without viscosity [4]. The conservation of
circulation (Kelvin’s theorem) requires that vortex lines are created in pairs or
as rings which emerge from a point [5], or at boundaries [6]. Recent experiments
on dilute atomic vapours have confirmed many of the predictions of the GP
model. For example, the expected vortex solutions have been observed in rotating
condensates [7,8], and experiments with a moving object [9,10] are consistent
with GP predictions of the critical velocity for vortex formation [11].
In this paper, we consider the motion of objects through homogeneous and
inhomogeneous condensates. In Sect. 2 we briefly review the fluid properties
described by the GP equation. In Sect. 3 we consider the process of vortex for-
mation in terms of the stationary uniform flow solutions of the GP equation. We
discuss how vortices can exist in laminar flow solutions below the critical velocity,
and that there is a ‘smooth’ transition between flows without and with vortices.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 307–319, 2001.

c Springer-Verlag Berlin Heidelberg 2001
308 C.S. Adams et al.

In Sect. 4 we use the uniform flow solutions to determine the critical velocity,
and in Section 5 we return to the time-dependent solutions to study the link
between vortex shedding and drag. In Sect. 6, we consider the motion through
an inhomogeneous condensate, and show that if the object passes through lower
density regions, then the critical velocity is significantly lowered, as observed
experimentally [9,10]. Finally in Sect. 7 we compare the results of experiments
on ions in superfluid helium and the predictions of the GP model.

2 Fluid Equations
The evolution of a dilute Bose-Einstein condensate in the limit of low tem-
perature is given by the solution of the time-dependent Gross-Pitaevskii (GP)
equation,
 
2 2
i∂t ψ(r, t) = − ∇ + V (r, t) + g|ψ(r, t)| ψ(r, t) ,
2
(1)
2m

where ψ is normalised to the number of condensate particles, the coefficient of

the non-linear term, g = 4π2 a/m, a is the s-wave scattering length, V (r, t)
represents external potentials arising from the trap and any moving obstacle.
The link between equation (1) and the equivalent equations of fluid mechanics is
made by defining a mass density, ρ = mψ ∗ ψ, and a momentum current density,
Jk = (/2i)(ψ ∗ ∂k ψ − ψ∂k ψ ∗ ), then the fluid velocity is given by, vk = Jk /ρ,
or equivalently vk = (/m)∂k ϕ, where ϕ is the phase of the field ψ. By defini-
tion, the velocity field corresponds to a potential flow, however, it can support
circulation at singularities (vortices).
The continuity equation (conservation of mass or probability) follows from
the definition of ρ and equation (1),

∂t ρ + ∂k Jk = 0 . (2)

The conservation of momentum equation may be found by considering the rate

of change of the momentum current density,

∂t Jk + ∂j Tjk + ρ∂k (V /m) = 0 , (3)

where the momentum flux density tensor takes the form [12,13],

2 g
Tjk = (∂j ψ ∗ ∂k ψ − ψ ∗ ∂j ∂k ψ + c.c.) + δjk |ψ|4 . (4)
4m 2
This can be rewritten as,
Tjk = ρvj vk − σjk , (5)
where the stress tensor σjk is given by,

σjk = − 12 δjk g(ρ/m)2 + (/2m)2 ρ∂j ∂k ln ρ . (6)

 Motion of Objects Through Dilute Bose–Einstein Condensates 309

The form of equations (2), (3), and (5) is identical to those for classical fluid
flow [15], except for the stress, equation (6). The normal stress, −σjk , j = k, or
pressure is given by,
p = 12 g(ρ/m)2 − (/2m)2 ρ∇2 ln ρ . (7)
The first term, the interaction pressure, supports the propagation of sound. In
the limit of long wavelength excitations (i.e, the linear part of the Bogoliubov
spectrum), the second term, known as the quantum pressure, is neglibible, and
the speed of sound is given by [14,15],
 1
c = ∂p/∂ρ = (gρ/m2 ) 2 . (8)
For a dilute Bose-Einstein condensate shear stresses, σjk , j = k, arise from
density gradients, due the second term in (6). This property gives rise to the
possibility of vortex formation and drag without viscosity.
Combining equations (2) and (3), along with the identity, ρ−1 ∂j [ρ∂j ∂k ln ρ] =
1 1
2∂k [ρ− 2 ∂j ∂j ρ 2 ] the momentum equation may be rewritten as,
1 1
∂t vk + vj ∂j vk + ∂k [gρ/m2 − (2 /2m)ρ− 2 ∂j ∂j ρ 2 + V /m] = 0 , (9)
which corresponds to the GP form of the Euler equation. By integrating, or
more directly from the real part of equation (1), one obtains the conservation of
energy or Bernoulli equation,
1 1
∂t φ + 12 mv 2 + gρ/m − (2 /2m)ρ− 2 ∇2 ρ 2 + V = 0 . (10)
The conservation of angular momentum (Kelvin’s theorem), follows from Euler’s
equation (9) and states that the circulation around a closed ‘fluid’ contour does
not change in time. This means that within the fluid, vorticity must be created
in the form of rings or pairs of lines which emerge from a point [5]. The exception
is at boundaries, where the wavefunction is clamped to zero and no closed fluid
loop can be drawn, e.g., at the surface of an impenetrable object [15] or from
the edge of a trapped condensate [6].

3 Time-Independent Solutions in the Object Frame

For homogeneous fluids, where the external potential is due to the obstacle only,
it is convenient to rescale length and velocity  in terms of the healing length,
√
ξ = / mn0 g, and the sound speed, c = n0 g/m, respectively, where n0 is
the number density far from the object. In this case, for an object moving with
velocity U , equation (1) becomes
 
i∂t ψ̃(r, t) = − 12 ∇2 + V (r − U t) + |ψ̃(r, t)|2 ψ̃(r, t) , (11)
√
where ψ̃ = ψ/ n0 . Unless otherwise stated we use these units throughout. If
ψ  (r  , t) = ψ̃(r, t) is the field in the fluid frame written in terms of the object
frame coordinates, r  = r − U t, then the GP equation for ψ  (r  , t) is
i∂t ψ  = − 12 ∇2 ψ  + V (r  )ψ  + |ψ  |2 ψ  + iU · ∇ ψ  . (12)
310 C.S. Adams et al.

Solutions of (12) of the form, ψ̃(r  , t) = φ(r  )eiμt , where μ is the chemical poten-
tial, are found only for U ≤ Uc , where Uc is the critical object velocity for vortex
formation [16,17,18]. Consequently, such time-independent solutions provides a
useful technique to determine the critical velocity as illustrated in Sect. 4.
The time-independent solutions of (12) also characterise the process of vor-
tex formation. In Fig. 1(left) we plot the energy dependence of the solutions
corresponding to a penetrable sphere with radius R = 3.3. We define the energy
relative to a laminar flow state with V = 0 and having the same number of
particles, i.e.,
  2 
1 2 2 1 2
E = dr |∇φ| + V |φ| + |φ| − 1 . (13)
2 2

Away from the object, any deviation of the local particle density |φ|2 from 1 con-
stitutes an excitation. The three branches correspond to (in order of increasing
energy) laminar flow with no vortices, a vortex ring pinned to the sphere, and
a vortex ring encircling the sphere. These states are illustrated by surfaces of
constant density shown in Fig. 2. A more detailed discussion is given in Ref. [17].

2 5 0 4

3
2 0 0
2

1 5 0 1
e n e rg y

0
z

1 0 0
− 1

5 0 − 2

− 3
0
− 4
0 0 .2 0 .4 0 .6 − 4 − 2 0 2 4
v e lo c ity ( s o u n d s p e e d ) x

Fig. 1. Left: The energy of the time-independent solutions of the GP equation for flow
past a penetrable sphere with radius R = 3.3 (V (r) = 1.0, r ≤ R, V (r) = 0.0, r > R)
as a function of the flow velocity. No time-independent solutions exist for flow above
a critical velocity, Uc = 0.68. The three branches correspond to (in order of increasing
energy) laminar flow, a pinned vortex ring, and an encircling vortex ring. The vortex
ring appears in the laminar flow solution at U = 0.62 (as indicated), well below the
critical velocity. Right: Cross sections showing the phase pattern of the laminar flow
solutions at U = 0.615 (left side) and U = 0.625 (right side). The object is indicated
by the white circle. At U = 0.625 the phase changes from −π (black) to π (white) near
the centre of the object indicating the appearance of the vortex ring. As the energy
increases the vortex ring grows, eventually crossing the object boundary at U = 0.61
on the encircling ring branch.
312 C.S. Adams et al.

(a ) (b ) (c )

0 .6 0 .6 0 .6

(1 )
0 .4 0 .4 0 .4
v e lo c ity

(2 )

(3 )
0 .2 0 .2 0 .2

0 0 0
0 2 5 0 5 0 0 7 5 0 0 2 5 0 5 0 0 7 5 0 0 2 5 0 5 0 0 7 5 0
m o m e n tu m m o m e n tu m m o m e n tu m

Fig. 3. The evolution of the object velocity, U , with initial value U = 0.5, due to
a constant external force, F . The time-dependent solution (bold line) is plotted on
top of the corresponding time-independent solutions of (12) as a function of the total
momentum P = F t for (a) F = 0.05, (b) F = 2 and (c) F = 4. In (a), the initial
laminar flow state (1) evolves into an encircling vortex ring (2), followed by an abrupt
decrease in velocity when the object jumps into the vortex core. This transition to
the pinned ring state (3) excites vibrations of the ring leading to oscillations in the
object velocity. As the vortex ring radius increases the object velocity tends towards
the velocity of a free vortex ring, as indicated by the dotted line. In (c), the force is
sufficient to detach the object from the ring and the cycle repeats. The dots in (b) and
(c) correspond to the times of the isosurface plots shown in Roberts and Berloff, Fig. 4.

due to the quantum pressure term. By solving equation (12), one finds that the
critical object velocity, Uc , is reached when the flow velocity, v, exceeds the local
sound speed, c = |ψ|, in a region of the fluid where the quantum pressure is zero.
This is illustrated in Fig. 4 which shows the flow velocity, the sound speed, and
the quantum pressure term near the surface of a large (R = 50) impenetrable
sphere.
In the limit of large radius, R → ∞ (where one can neglect the boundary
layer and hence the quantum pressure term), one can predict the critical velocity
analytically using an asymptotic expansion for the speed at the equator, v =
3 3 5
2 U + 0.313U + 0.392U + . . . [20], and correcting for the dependence of the
sound speed on density. In steady state, neglecting the quantum pressure, the
Bernoulli equation (10) states that the sum of the interaction and kinetic energy
terms near the object and in the far-field must be equal, i.e.,
1 1
1 + U 2 = n + v2 .
2 2
The solution becomes
√ unstable when the flow velocity becomes equal to the
sound speed, v = n, [21,22]. Combining these results (including terms up to
U 11 ) gives Uc = 0.53004 for R → ∞, in good agreement with our numerical
value Uc = 0.53285 for R = 50. The critical velocity decreases with increasing R
tending asymptotically towards the R = ∞ value [17,18].
 Motion of Objects Through Dilute Bose–Einstein Condensates 313

1 1
(a )

0 .5 0 .5
(b )

0 0
(c ) − 3
x 1 0
1
− 0 .5 − 0 .5
0
U = 0 .2 − 1
U = 0 .5 3 2 8 5
− 1 − 1 5 1 0

0 5 1 0 1 5 2 0 0 5 1 0 1 5 2 0
d is ta n c e fr o m s p h e re d is ta n c e fr o m s p h e re

Fig. 4. Laminar flow solution for the motion of a sphere with radius R = 50 and velocity
U = 0.2 (left) and U = 0.53285 (right). The three curves show, (a) the ‘bulk’ sound
√
speed, |ψ| = n, (b) the velocity, v, and (c) the quantum pressure term, ∇2 |ψ|/2|ψ|, as
√
a function of position. Note that, at the critical velocity Uc = 0.53285 (right), v = n,
where the quantum pressure is zero.

For trapped condensates the speed of sound and hence the critical velocity
depend upon position. This can lead to a significant lowering of the critical
velocity if the object passes through lower density regions of the condensate as
discussed in Section 6.

5 Vortex Shedding and Drag

The main significance of the critical velocity is that it marks the threshold for
the breakdown of superfluidity and the on-set of dissipation. For macroscopic
objects, where vortex pinning can be neglected, motion faster than the critical
velocity, U > Uc , results in periodic vortex shedding. The energy and momentum
transfer to the fluid leads to a drag force on the object. The drag force F may
be calculated either from the energy transfer
dE
= F .v , (14)
dt
or the rate of momentum transfer: By integrating equation (3), one finds that
the k-th component of the force,
  
Fk = ∂t dΩJk = − dS nj Tjk − dΩ ρ∂k (V /m) , (15)
Ω S Ω

where S is the surface of the object or control surface within the fluid [13], Ω is
the volume enclosed by S, nj is the j-component of the normal vector to S, and
dS is a surface element. The second term on the right-hand side can be likened to
the buoyancy of the fluid. In the case of homogeneous flow past an impenetrable
object, Sect. 4 and 5, ψ = 0 on the object surface and the potential is uniform
elsewhere, therefore, only the first term contributes. Conversely, for a penetrable
314 C.S. Adams et al.

object in a trapped condensate, Sect. 6, Ω may be chosen to encompass the

entire fluid, and the first term is zero.
A plot of the drag force as a function of velocity for a cylindrical object is
shown in Fig. 5. The drag is calculated by applying equation (15) to the solutions
of equation (1). The vortices in the wake of a moving object interact creating
fluctuations in the flow pattern and the drag, therefore the data shown in Fig. 5
correspond to an average over many
 vortex emission cycles. The force per unit
length is measured in units of  n30 g/m. The threshold velocity for the on-set
of drag is 0.4 as expected for a cylindrical object [22].
The contribution of vortex shedding to the total drag force can be estimated
by considering the momentum transfer due to vortex emission, i.e.,

F v = fv pv , (16)

where fv is the vortex shedding frequency and pv is the momentum of a vortex

pair as it is created in the equatorial plane. The comparison between equa-
tions (15) and (16) shown in Fig. 5 suggests that for U < c vortex shedding
is the dominant dissipation mechanism, whereas for U > c, an increasingly sig-
nificant contribution arises from sound waves. For U > c, the reflected sound
waves create a standing wave pattern in front of the object as shown in the time-
averaged density images of Fig. 6. In the far-field the orientation of the standing
waves corresponds to the Mach angle as in classical acoustics.
Figure 6 also displays some interesting features in the wake. The dark streaks
emerging from the shadow region behind the object correspond to sound waves
emitted by interacting vortices. As a pair of vortex lines loses energy they move
closer together and eventually annihilate. Far downstream, a significant fraction
of the vortex energy is converted into sound. This conversion of vortex energy
into sound energy due to vortex motion and reconnections [23] is an important
contributary process in the the decay of superfluid turbulence (see Chapters by
Vinen and McClintock).

6 The Critical Velocity in Inhomogeneous Condensates

Experiments on moving laser beams in trapped atomic condensates suggest a
critical velocity of Uc ∼ 0.1 [9,10], much lower than the homogeneous value
for an cylindrical object Uc ∼ 0.4 [22]. Vortex stretching [24] and enhanced
phonon excitation at low velocity due to the spatial inhomogeneity [25] have
been proposed as possible mechanisms for this reduction. Numerical simulations
of trapped atomic condensates in three dimensions indicate that Uc ∼ 0.1 − 0.2
[11] in rough agreement with experiment. Fig. 7(left) shows the energy transfer to
the condensate as a function of the object velocity. In the experiments this energy
is measured as a heating effect [9,10]. At finite temperature, the conversion of
energy into heat may be accerlated by the interaction with the thermal cloud
[26], however, even in the limit T → 0 one would expect a decay of the vortex
structures as discussed in connection with Fig. 6. Apart from the lower critical
velocity the behaviour is similar to the idealised homogenous case discussed in
 Motion of Objects Through Dilute Bose–Einstein Condensates 315

4 0

3 5

3 0

2 5
d ra g fo rc e

2 0

1 5

1 0

0
0 0 .5 1 1 .5 2 2 .5
v e lo c ity
Fig. 5. The time-averaged drag force (black circles) as a function of velocity for an
impenetrable cylinder with radius R = 3. The force per unit length is measured in
units of  n30 g/m. The open circles indicates the contribution due to vortex shedding.
The error bars indicate the residual fluctuations in the time-averaged drag.

Fig. 6. Time-averaged fluid density in the frame of a cylindrical object (radius R = 3)

moving from left to right with speeds of 0.9 (left) and 1.2 (right). Dark and light
indicate low and high density respectively. For supersonic motion a standing wave
pattern appears in front of the object.
316 C.S. Adams et al.

Sect. 5. We suggest that the lower critical velocity is mainly due to formation
of vortices where the object intersects lower density regions of the condensate.
Fig. 7(right) shows how at low object speeds (U = 0.24c) vortices enter the
condensates as half rings at the surface. These half rings penetrate towards
the centre along the low density regions close to the object axis, eventually
connecting to form a pair of lines. In this case, the reconnection is being driven
by an external force and the vortex line length increases. This is the inverse of
the usual reconnection process where the vortex line length decreases [23]. The
subsequent motion of the vortex lines is determined by density gradients and
the interaction with other vortices [27].

0 .6
E n e r g y tr a n s fe r [a r b . u n its ]

0 .4

0 .2

0
0 1 2 3
v

Fig. 7. Three dimensional GP simulations of a moving laser beam in a trapped Bose-

Einstein condensate. In this example, distance and time are measured in terms of
the harmonic oscillator length, /2mω, and the inverse trap frequency, 1/ω. Left:
The rate of energy transfer to the condensate as a function of the laser velocity. The
critical velocity is Uc ∼ 0.1c, where c = 2.54 is the sound speed at the centre of the
condensate (indicated by the dashed line). Right: Isosurface plots (v 2 = 10−3 ), showing
the evolution of the vortices at times, (A) 6.8, (B) 7.6, and (C) 8.4, for an object velocity,
U = 0.24c. Also shown in (B) is a dashed circle in the x−y plane, indicating the spatial
extent of the condensate given by the Thomas-Fermi radius, RTF = 3.59. The laser
beam is moving into the paper and downwards along the normal to the dashed circle.
Vortices are seen to nucleate at the surface as half-rings, before penetrating into the
bulk to reconnect as a vortex pair. Other parameters match those given in Fig. 5 of
[11].

The notion that vortices are responsible for dissipation is further supported
by the recent experimental observations of the density dependence of the critical
velocity and good agreement with predictions of the heating rate [10]. More evi-
dence could be gained by comparing with theoretical predictions of the variation
of Uc with laser beam intensity [11] and radius [21].

7 Comparison to Ions in Helium

In superfluid He-4 interactions lead to a depletion of the condensate and a differ-
ent dispersion curve to the dilute Bose gas [2,28]. A modified GP model which
 Motion of Objects Through Dilute Bose–Einstein Condensates 317

includes the non-local nature of the interactions has been used to reproduce
the correct dispersion curve [29,30,31]. The GP equation can also be applied to
describe a finite temperature system [32,33], or similarly the depletion of the
condensate due to interactions [34]. However, some insight can be gained by ap-
plying the standard GP model to the motion of objects in helium. For example,
existing theories of vortex nucleation cannot treat the intermediate states where
the vortex is less than a healing length from the object [35]. An advantage of
the GP equation is that fluid healing and the quantisation of circulation are
included explicitly, making it possible to identify a complete path of vortex nu-
cleation. Using the GP model, both the time-independent solutions (Sect. 3) and
time-dependent simulations [19] suggest that an object can be accelerated up to
a maximum velocity, where a vortex ring gradually emerges encircling the ob-
ject, and subsequently becomes pinned to it. An important question is whether
the spherical symmetry favours the encircling ring state. In Fig. 8 we show the
time-independent states for a sphere with a bump. The bump lowers the critical
velocity from 0.68 to 0.65, and now there is a continuous transition between
laminar flow and an vortex ring encircling the bump. However, the small vortex
loops of the type discussed by Muirhead et al. [35] are not observed. Smaller
rings have a lower energy alternative, to exist as a ‘virtual’ ring inside the ob-
ject, as in Fig. 1(right). The vortex appears only when it has a radius larger
than that of the object.

Fig. 8. Uniform flow solutions for a sphere (R = 3.3) with a bump (R = 1.5). Left:
The velocity-momentum dependence of the laminar and four vortex solutions. Right:
Cross sections showing the fluid density (dark corresponds to low density around the
object) for the four allowed vortex configurations: (1) Vortex ring encircling the object
and bump; (2) pinned to the object; (3) encircling the bump; and (4) pinned to the
bump. Note that laminar flow evolves continuously into a vortex ring encircling the
bump, solution (3). Subsequently, the system evolves to the lowest energy state (4).
318 C.S. Adams et al.

To convert the critical velocity to √ a value pertaining to ions in HeII we

use the measured healing length, ξ/ 2 = 0.128 nm [36], and the GP sound
speed c = /mξ = 88 ms−1 . For an object with radius R = 3.3, equivalent to
0.6 nm which is close to the size of the ‘snowball’ that surrounds a positive ion
[28], one obtains vc ∼ 60 ms−1 , in rough agreement with experiments [37]. The
surface roughness effect discussed above tends to reduce the critical velocity.
In addition, there is experimental support for the GP picture of a continuous
transition between a bare ion and an ion plus vortex [37].
For ions in HeII, there is also strong experimental evidence that vortex nu-
cleation involves an energy barrier [38]. In the GP model the energy barrier
corresponds to the energy required to transfer from the laminar flow to the
pinned ring solution in Fig. 1(left). This energy decreases with increasing flow
velocity but remains finite at the critical velocity because the pinned ring branch
joins the laminar-encircling ring curve at an energy above the critical point. The
size of the energy barrier depends on the object radius, mass, and penetrability.
For an impenetrable sphere with radius R = 3.3 (see Fig. 5 in Ref. [17]), we find
an energy barrier of approximately 8n0 cξ 2 . For HeII, this would correspond to
∼ 3.8 K, similar to the value derived by Muirhead et al. using arguments based
on classical fluid mechanics [35]. For HeII in the limit of low temperature, the
transition across the energy barrier can still be driven by local fluctuations of
the condensate or non-condensate density.

8 Conclusion
The main aim of this paper was to illustrate the use of the GP equation as a
fluid dynamical model of dilute Bose-Einstein condensates, and to consider to
what extent experimental and theoretical studies of the dilute quantum fluids
can provide new insight into more complex superfluids. We have shown that the
process of vortex formation is characterised by a ‘smooth’ transition between
flows without and with vortices, and that vortex rings initially emerge encircling
the object. For inhomogeneous condensates, the critical velocity is lowered by the
formation of vortices where an object intersects lower density regions. Finally,
we suggest that an extension of the GP model may provide a good description
of the mechanism of vortex formation by ions in liquid helium.

Acknowledgements

We thank N. G. Berloff, S. Rica, and W. F. Vinen for stimulating discussions.

References
1. See e.g. Bose-Einstein condensation in atomic gases, Proc. Int. School of Physics
Enrico Fermi, eds. M. Inguscio, S. Stringari and C. Wieman (IOS Press, Amster-
dam, 1999).
 Motion of Objects Through Dilute Bose–Einstein Condensates 319

2. D. R. Tilley and J. Tilley, Superfluidity and superconductivity, 3rd Ed., (IoP, Bristol,
1990).
3. V. L. Ginzberg and L. P. Pitaevskii, Zh. Eksp. Teor. Fiz. 34, 1240 (1958) [Sov.
Phys. JETP 7, 858 (1958)]; E. P. Gross, Nuovo Cimento 20, 454 (1961) and J.
Math. Phys. 4, 195 (1963).
4. T. Winiecki, B. Jackson, J. F. McCann, and C. S. Adams, J. Phys. B. 33, 4069
(2000).
5. B. Jackson, J. F. McCann, and C. S. Adams, Phys. Rev. Lett. 80, 3903 (1998).
6. B. M. Caradoc-Davies, R. J. Ballagh, and K. Burnett, Phys. Rev. Lett. 83, 895
(1999).
7. M. R. Matthews, B. P. Anderson, P. C. Haljan, D. S. Hall, C. E. Wieman, and E.
A. Cornell, Phys. Rev. Lett. 83, 2498 (1999).
8. K. W. Madison, F. Chevy, W. Wohlleben, and J. Dalibard, Phys. Rev. Lett. 84,
806 (2000).
9. C. Raman, M. Köhl, R. Onofrio, D. S. Durfee, C. E. Kuklewicz, Z. Hadzibabic, and
W. Ketterle, Phys. Rev. Lett. 83, 2502 (1999).
10. R. Onofrio, C. Raman, J. M. Vogels, J. Abo-Shaeer, A. P. Chikkatur, and W.
Ketterle, preprint, Phys. Rev. Lett. 85, 2228 (2000).
11. B. Jackson, J. F. McCann, and C. S. Adams, Phys. Rev. A 61, 051603 (R) (2000).
12. T. Frisch, Y. Pomeau, and S. Rica, Phys. Rev. Lett. 69, 1644 (1992).
13. T. Winiecki, J. F. McCann, and C. S. Adams, Phys. Rev. Lett. 82, 5186 (1999).
14. P. Nozières and D. Pines, Theory of Quantum Liquids Vol II (Addison-Wesley,
Redwood City, 1990).
15. Fluid Mechanics (2nd ed.), L. D. Landau and E. M. Lifshitz (Pergamon, Oxford,
1987).
16. C. Huepe and M.-É. Brachet, C. R. Acad. Sci. Paris, 325, Série 2 b, 195 (1997).
17. T. Winiecki, J. F. McCann, and C. S. Adams, Europhys. Lett. 48, 475 (1999).
18. C. Huepe, M.-É. Brachet, Physica D, 140, 126 (2000).
19. T. Winiecki and C. S. Adams, Europhys. Lett. 52, 257 (2000).
20. N. G. Berloff and P. H. Roberts, J. Phys. A 33, 4025 (2000).
21. C. Josserand, Y. Pomeau, and S. Rica, Physica D, 134 111 (1999);
22. S. Rica, Physcia D 148, 221 (2001).
23. M. Leadbeater, T. Winiecki, D. C. Samuels, C. F. Barenghi, and C. S. Adams,
Phys. Rev. Lett. 86 no. 7 to appear (2001)
24. C. Nore, C. Huepe, and M. E. Brachet, Phys. Rev. Lett. 84, 2191 (2000).
25. P. O. Fedichev and G. V. Shlyapnikov, cond-mat/0004430.
26. P. O. Fedichev and G. V. Shlyapnikov, Phys. Rev. A 60 R1779 (1999).
27. B. Jackson, J. F. McCann, and C. S. Adams, Phys. Rev. A 61, 013604 (2000).
28. R. J. Donnelly Quantized vortices in Helium II, (CUP, Cambridge, 1991).
29. N. G. Berloff, J. Low Temp. Phys. 116, 359 (1999).
30. N. G. Berloff and P. H. Roberts, J. Phys. A 32, 5611 (1999).
31. N. G. Berloff and P. H. Roberts, Phys. Lett. A 274, 569 (2000).
32. R. J. Marshall, G. H. C. New, K. Burnett, and S. Choi, Phys. Rev. A 59, 2085
(1999).
33. M. J. Davis, S. A. Morgan and K. Burnett, cond-mat/0011431
34. M. Leadbeater and C. S. Adams, unpublished
35. C. M. Muirhead, W. F. Vinen, and R. J. Donnelly, Phil. Trans. R. Soc. Lond. A
311, 433 (1984).
36. G. W. Rayfield and F. Reif, Phys. Rev. 136, 1194 (1964).
37. G. W. Rayfield, Phys. Rev. Lett. 19, 1371 (1967).
38. P. C. Hendry, N. S. Lawson, P. V. E. McClintock, and C. D. H. Williams, Phys.
Rev. Lett. 60, 604 (1988).
Stability of a Vortex in a Rotating Trapped
Bose–Einstein Condensate

Alexander L. Fetter and Anatoly A. Svidzinsky

Geballe Laboratory for Advanced Materials, Stanford University, Stanford, CA

94305-4045, USA

1 Time-Dependent Gross–Pitaevskii Equation

The remarkable achievement of Bose-Einstein condensation in dilute trapped
alkali-metal atomic gases [1,2,3] has stimulated the (now successful) search for
quantized vortices that are usually associated with external rotation [4,5,6,7,8].
An essential feature of these condensates is the order parameter, characterized
by a complex macroscopic wave function Ψ (r, t). Theoretical descriptions of vor-
tices in trapped low-temperature condensates have relied on the time-dependent
Gross-Pitaevskii (GP) equation [9,10,11], which omits dissipation. For a trap
rotating at angular velocity Ω about ẑ, it is a nonlinear Schrödinger equation
∂Ψ  
i = T + Vtr − Ω Lz + g|Ψ |2 Ψ, (1)
∂t
0
where T = −2 ∇2 /2M is the kinetic-energy operator, Vtr = 12 M j ωj2 x2j is
the harmonic trap potential, Lz = xpy − ypx is the z component of angular
momentum, and g ≡ 4π2 a/M characterizes the strength of the short-range
interparticle potential (here, a is the positive s-wave scattering length for repul-
sive two-body interactions; typically a is a few nm). Current experiments involve
“dilute” systems,
 so that the zero-temperature condensate contains nearly all N
particles, with dV |Ψ |2 ≈ N . For a steady solution, Ψ (r, t) = Ψ (r)e−iμt/ ,
where μ is the chemical potential.

1.1 Equivalent Hydrodynamics of Compressible Isentropic Fluid

If the condensate wave function is written as Ψ = eiS |Ψ |, the real and imaginary
parts of Eq. (1) precisely reproduce the time-dependent irrotational hydrody-
namics of a compressible isentropic fluid, written in terms of the number density
n = |Ψ |2 and the velocity potential Φ = S/M , where v = ∇Φ [9,10,12]. In
principle, the dynamics of a curved vortex line in a rotating trap follows directly
from general hydrodynamics, but the trap potential and resulting nonuniform
density complicate the problem considerably. Thus it is preferable to start with
simple situations.

This work was supported in part by NSF Grant No. 99-71518. The manuscript
benefited from our participation in workshops organized by H. Stoof at the Lorentz
Center, Leiden, The Netherlands and by S. Stringari at ECT∗ , Trento, Italy; we are
grateful for their hospitality.

Supported in part by Stanford University.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 320–326, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Stability of a Vortex in a Rotating Trapped Bose–Einstein Condensate 321

1.2 Thomas–Fermi Limit for Large Condensates

For a noninteracting condensate, the macroscopic wave function is the ground

state of the harmonic oscillator, with spatial extent dj = (/M ωj )1/2 and
j = x, y, z; typically dj is a few μm. The repulsive interactions act to expand
the condensate, and the relevant dimensionless interaction parameter is N a/d0 ,
1/3
where d0 = (dx dy dz ) is a suitable geometric mean. Recent experiments fo-
cus on the regime N a/d0  1, when the quantum-mechanical energy associated
with the density gradients is negligible. In this Thomas-Fermi (TF) limit [13],
the nonrotating GP equation has the solution g|ΨT F |2 +Vtr ≈ μ, with a parabolic
density profile  
x2 y2 z2
nT F (r) ≈ nT F (0) 1 − 2 − 2 − 2 . (2)
Rx Ry Rz
Here Rj = (2μ/M ωj2 )1/2  dj fixes the condensate’s dimensions and nT F (0) =
μ/g is the central density. The (large) dimensionless ratios R05 /d50 = 15N a/d0
and μ/ω0 = 12 R02 /d20 characterize the effect of the repulsive interactions, where
ω0 and R0 are appropriate geometric means.

2 Energy of a Vortex in a Large Rotating Trap

The time-dependent GP equation (1) follows from a variational energy functional


 
E(Ω) = dV Ψ ∗ T + Vtr + 12 g|Ψ |2 − Ω Lz Ψ . (3)

If Ψ represents a straight singly quantized vortex in a rotating disk-shape trap

(R⊥  Rz ), the TF density yields the increase in energy ΔE(r 0 , Ω) associated
with the vortex at the transverse position r 0 [14]. For a nonrotating trap with
Ω = 0, this energy ΔE(r 0 , 0) decreases monotonically with lateral displacement.
Energy conservation requires that the allowed motion is a precession at fixed trap
potential (the trajectory is elliptical for an anisotropic trap). In the presence of
weak dissipation, the vortex slowly spirals outward, lowering its energy.
For arbitrary Ω and small lateral displacements, ΔE(r 0 , Ω) has the form
   
4π Rz n(0)2 R⊥ 2M ΩR⊥ 2
ΔE(r 0 , Ω) ≈ ln −
3 M ξ 5
 2 2
   2

1 x0 y0 R⊥ 2M ΩR⊥
− + 2 3 ln − , (4)
2 Rx2 Ry ξ 

where ξ = /(2M μ)1/2 is the vortex-core radius and 2/R⊥ 2

= 1/Rx2 + 1/Ry2
defines the mean transverse radius of the condensate. In the TF limit, ξ is small,
with ξR0 = d20 , ensuring a clear separation of length scales ξ  d0  R0 .

• For small positive Ω, ΔE(r 0 , Ω) decreases with increasing lateral displace-

ments so that a vortex will spiral outward in the presence of weak dissipation.
322 A.L. Fetter and A.A. Svidzinsky

• The curvature at r0 = 0 vanishes at Ωm = 32 (/M R⊥ 2

) ln(R⊥ /ξ), signaling the
onset of metastability. Since ΔE(0, Ωm ) > 0, this state is not truly stable.
• When Ω > Ωm , the trap center becomes a local minimum. For weak dissipa-
tion, a vortex that is slightly displaced from the center will spiral inward.
• When Ω reaches Ωc = 52 (/M R⊥ 2
) ln(R⊥ /ξ) = 53 Ωm , a central vortex becomes
stable because ΔE(0, Ωc ) = 0. Hence Ωc is the thermodynamic critical angular
velocity for vortex creation.
• For Ω > Ωc , the form of ΔE(r 0 , Ω) shows that a vortex at the outer edge
remains metastable, but the barrier for entry into the condensate decreases.
Spontaneous nucleation of a vortex may eventually occur through a surface
instability [15,16,17]. The observed critical angular velocity for vortex cre-
ation [5,7] is ≈ 70% higher than the predicted TF value Ωc , in qualitative
agreement with such a surface mechanism (but see Ref. [18] and Sec. 4.2 for
an alternative explanation).

3 Small-Amplitude Excitation of a Vortex

in a Rotating Trap
A macroscopic Bose condensate acts like an external particle source, in the sense
that the same physical excited state (here labeled by j) can be achieved ei-
ther by adding or by subtracting one particle. The actual eigenstates involve
“quasiparticle” operators αj† and αj that are linear combinations of the two
quantum-mechanical states. For a given normal mode, the resulting pair of cou-
pled complex amplitudes uj (r) and vj (r) obey the Bogoliubov equations [10,19]
that determine the corresponding eigenfrequency ωj . Imposing Bose-Einstein
commutation relations [αj , αk† ] = δjk requires that these amplitudes obey the
particular normalization dV (|uj |2 − |vj |2 ) = 1, and the resulting quasiparti-
0
cle Hamiltonian reduces to a set of uncoupled harmonic oscillators j ωj αj† αj ,
summed over all modes with positive normalization. In the simplest case of a
uniform condensate, the eigenfrequencies are all positive, which ensures that
the system is stable because the energy then has a lower bound. For a uniform
condensate moving with velocity v, however, some of the eigenfrequencies can
become negative as soon as |v| exceeds the speed of sound, which corresponds
to the instability associated with the Landau critical velocity [20]. Physically,
the system can spontaneously generate quasiparticles because the Hamiltonian
is no longer bounded from below.
These general ideas have direct relevance to the stability of a singly quan-
tized vortex in a rotating trapped condensate. For simplicity, it is convenient to
consider an axisymmetric condensate in equilibrium with a trap rotating at an
angular velocity Ω. In this case, states can be characterized by their azimuthal
angular quantum numbers mj , and the transformation to rotating coordinates
H → H − ΩLz ensures that the eigenfrequencies ωj (Ω) in the rotating frame
are simply ωj (Ω) = ωj − mj Ω, where ωj is the corresponding eigenfrequency in
the nonrotating frame.
 Stability of a Vortex in a Rotating Trapped Bose–Einstein Condensate 323

3.1 Stability of a Vortex

The first numerical study of the Bogoliubov equations for a singly quantized vor-
tex in a nonrotating trap found a single negative-frequency mode (the “anoma-
lous mode”) [21,22], implying that the vortex in the condensate is unstable.
Physically, this instability arises because the condensate with a vortex has a
higher energy than the vortex-free condensate.
This situation is especially clear for a noninteracting Bose gas in an axisym-
metric harmonic trap, when the first excited (vortex) state has an excitation en-
ergy ω⊥ and an angular momentum . The transition back to the true ground
state involves the (negative) frequency −ω⊥ and the (negative) change in angu-
lar quantum number −1. More generally, the numerical analysis for small and
medium interaction strength N a/d⊥ [21] found that the anomalous frequency ωa
remained negative (with ma = −1 unchanged). In a frame rotating with angular
velocity Ω, the anomalous frequency becomes ωa (Ω) = ωa + Ω. Since ωa < 0,
the eigenfrequency ωa (Ω) in the rotating frame rises toward 0 with increasing Ω
and vanishes at a critical rotation speed Ω ∗ = −ωa = |ωa |. The Bogoliubov de-
scription of small oscillations implies that a condensate with a singly quantized
vortex is unstable for Ω < Ω ∗ but becomes stable for Ω > Ω ∗ .
In contrast to the numerical study [21] for small and medium values of the di-
mensionless coupling parameter N a/d0 , a direct perturbation analysis is feasible
for a large condensate (N a/d0  1) containing an axisymmetric singly quan-
tized vortex (the TF limit). Detailed study of the Bogoliubov equations [14]
yields the explicit
 expression
 for the anomalous frequency in the rotating frame
ωa (Ω) = − 32 /M R⊥ 2
ln (R⊥ /ξ) + Ω, which has the expected form. The result-
ing critical
 angular  velocity
Ω ∗ = 32 /M R⊥ 2
ln (R⊥ /ξ) for the onset of local stability agrees with Ωm in-
ferred from Eq. (4) for the onset of metastability.

3.2 Splitting of Normal-Mode Frequencies Caused by a Vortex

The ground-state condensate can sustain dynamical oscillations driven by the
mean-field repulsive interaction (analogous to plasma oscillations in a charged
medium). These normal modes become particularly simple in the TF limit of a
large condensate [23], and experiments have confirmed the predictions for the
lowest few modes in considerable detail [11]. For an axisymmetric condensate,
the normal modes can be classified by their azimuthal quantum number m, and
modes with ±m are degenerate.
When the condensate contains a vortex, the asymmetric circulating flow af-
fects the preceding normal modes. In particular, the originally degenerate modes
are split by the Doppler shift of the local frequency (analogous to the splitting
of magnetic sublevels in the Zeeman effect). In the TF limit, this small frac-
tional splitting of the degenerate modes is proportional to |m|d20 /R02 [24,25]; it
has served to detect the presence of a vortex and to infer the circulation and
angular momentum [7,26].
324 A.L. Fetter and A.A. Svidzinsky

4 Vortex Dynamics

At zero temperature, the time-dependent GP equation (1) determines the dy-

namics of the condensate in a rotating trap. A vortex line will move in response
to the nonuniform trap potential, the external rotation, and its own local cur-
vature. This problem is especially tractable in the TF limit, because the small
vortex core radius ξ permits a clear separation of length scales; the method of
matched asymptotic expansions yields an explicit expression for the local velocity
of a vortex line [27,28].

4.1 Dynamics of Straight Vortex

It is simplest to consider a straight vortex [14], which applies to a disk-shape

condensate with R⊥  Rz . Assume that the vortex is located near the trap
center at a transverse position r 0 (t). In this region, the trap potential does not
change significantly on a scale of order ξ. The solution proceeds in two steps.
(a) First, consider the region near the vortex core that is assumed to move
with a transverse velocity V ⊥ ẑ. Transform to a co-moving frame centered on
the vortex core, where the trap potential exerts a force proportional to ∇⊥ Vtr
evaluated at r 0 (t). The resulting solution includes both the detailed core struc-
ture and the “asymptotic” region |r ⊥ − r ⊥0 |  ξ
(b) Second, consider the region far from the vortex, where the core can be
treated as a distant singularity. The short-distance behavior of this solution
includes the region ξ  |r ⊥ −r ⊥0 |. The two solutions must agree in the common
region, which determines the translational velocity V of the vortex line.
The details become intricate, but the final answer is elegant and physical:
   2

3 R⊥ 2M ΩR⊥
V = ln − ẑ × ∇⊥ Vtr , (5)
4M μ ξ 3

where R⊥ for an asymmetric trap is defined below Eq. (4). This expression has
several important aspects

• The motion follows an equipotential line along the direction ẑ × ∇⊥ Vtr , con-
serving energy, as appropriate for the GP equation at zero temperature. For
an asymmetric harmonic trap, the trajectory is elliptical.
• For a nonrotating trap, the motion is counterclockwise, in the positive sense.
• With increasing external rotation, the translational velocity V decreases and
vanishes at the special value Ωm = 32 (/M R⊥ 2
) ln(R⊥ /ξ) discussed below
Eq. (4).
• For Ω > Ωm , the motion as seen in the rotating frame is clockwise.
• A detailed analysis of the normalization of the Bogoliubov amplitudes shows
that the positive-norm state has the frequency
2ωx ωy
ω= (Ω − Ωm ) . (6)
ωx2 + ωy2
 Stability of a Vortex in a Rotating Trapped Bose–Einstein Condensate 325

This normal-mode frequency is negative (and hence locally unstable) for Ω <
Ωm , but it becomes locally stable for Ω > Ωm , in agreement with the discus-
sion below Eq. (4).

4.2 Inclusion of Curvature

Equation (5) for the local velocity of a straight vortex oriented along ẑ can be
generalized to include the possibility of a different orientation of the vector t̂
locally tangent to the vortex core. In addition, local curvature k of the vortex
line defines a plane that includes both t̂ and the local normal n̂, inducing an
additional translational velocity along the binormal vector b̂ ≡ t̂ × n̂. A detailed
analysis [29] yields the translational velocity of the element located at r 0
   
 t̂ × ∇Vtr (r 0 ) 1 k2 2∇Vtr (r 0 ) × Ω
V (r 0 ) = − + k b̂ ln ξ + + ,
2M g|ΨT F |2 R⊥2 8 ∇2⊥ Vtr (r 0 )
(7)
where ∇2⊥ is the Laplacian in the plane perpendicular to Ω. In the first term,
|ΨT F |2 vanishes near the condensate boundary; hence t̂ × ∇Vtr (r 0 ) must also
vanish there, implying that the vortex is locally perpendicular to the surface.
Equation (7) allows a study of the dynamics of small-amplitude displacements
of the vortex from the z axis, when x(z, t) and y(z, t) obey coupled equations.
In the limit ωz = 0, there is no confinement in the z direction, and the density
is independent of z. The resulting two-dimensional dynamics exhibits helical
solutions that are linear combinations of two plane standing waves.
More generally, for ωz = 0, the density near the z axis has the TF parabolic
form, and the solutions become more complicated. It is convenient to define the
asymmetry parameters α = Rx2 /Rz2 and β = Ry2 /Rz2 , where α > 1, β > 1 indicate
a disk shape and α < 1, β < 1 indicate a cigar shape.
• For a nonrotating trap with the special asymmetry values α = 2/[n(n + 1)]
(here, n a positive integer), the effects of the nonuniform trap potential and
the curvature just balance, and the condensate has stationary solutions with
the vortex at rest in the xz plane. A disk-shape trap has no such states,
and the first one occurs for the spherical trap with α = 1. The next such
state occurs for α = 13 , when the condensate is significantly elongated. Similar
considerations for β apply to stationary states in the yz plane.
• For other values of α and β, solutions necessarily involve motion of the vortex
line relative to the stationary condensate.
• Analytical solutions can be found for an extremely flat disk with α  1 and
β  1, reproducing the frequency Ωm found in Eq. (4).
• For small deformations of a vortex line in an axisymmetric trap with α = β,
a disk-shape or spherical condensate (α ≥ 1) has only a single (unstable)
precessing normal mode with a negative frequency ωa < 0. In this case, an
external rotation Ω ≥ Ωm = |ωa | stabilizes the vortex. For these geometries,
Ωm is less than the thermodynamic critical value Ωc . In a spherical conden-
sate, the one anomalous mode |ωa | agrees with the observed vortex precession
frequency seen in the JILA experiments [8,18].
326 A.L. Fetter and A.A. Svidzinsky

• In contrast, an axisymmetric cigar-shape condensate has additional negative-

frequency precessing modes, and Ωm can exceed Ωc for sufficiently elongated
condensates; such behavior seems relevant for the ENS experiments [5,7,18],
where Ωm ≈ 1.7Ωc is close to the observed rotation speed for creating the first
vortex.
• For small axisymmetric deviations from a spherical trap, a straight vortex
line can execute large-amplitude periodic trajectories. In this case, the vortex
line becomes invisible when it tips away from the line of sight, and it then
periodically returns to full visibility. Such revivals agree with preliminary ob-
servations at JILA [26].

References
1. M. H. Anderson et al., Science 269, 198 (1995).
2. C. C. Bradley et al., Phys. Rev. Lett. 75, 1687 (1995).
3. K. B. Davis et al., Phys. Rev. Lett. 75, 3969 (1995).
4. M. R. Matthews et al., Phys. Rev. Lett. 83, 2498 (1999).
5. K. W. Madison et al., Phys. Rev. Lett. 84, 806 (2000).
6. K. W. Madison et al., e-print cond-mat/0004037.
7. F. Chevy, K. W. Madison, and J. Dalibard, e-print cond-mat/0005221.
8. B. P. Anderson et al., e-print cond-mat/0005368.
9. E. P. Gross, Nuovo Cimento 20, 454 (1961).
10. L. P. Pitaevskii, Zh. Eksp. Teor. Fiz. 40, 646 (1961) [Sov. Phys. JETP 13, 451
(1961)].
11. F. Dalfovo et al., Rev. Mod. Phys. 71, 463 (1999).
12. A. L. Fetter, Phys. Rev. A 53, 4245 (1996).
13. G. Baym and C. J. Pethick, Phys. Rev. Lett. 76, 6 (1996).
14. A. A. Svidzinsky and A. L. Fetter, Phys. Rev. Lett. 84, 5919 (2000); e-print cond-
mat/9811348.
15. F. Dalfovo et al., Phys. Rev. A 56, 3840 (1997).
16. D. L. Feder, C. W. Clark, and B. I. Schneider, Phys. Rev. A 61, 011601(R) (1999).
17. T. Isoshima and K. Machida, Phys. Rev. A 60, 3313 (1999).
18. D. L. Feder et al., e-print cond-mat/0009086.
19. A. L. Fetter, Ann. Phys. (N.Y.) 70, 67 (1972).
20. L. D. Landau, J. Phys. (USSR) 5, 71 (1941).
21. R. J. Dodd et al., Phys. Rev. A 56, 587 (1997).
22. D. S. Rokhsar, Phys. Rev. Lett. 79, 2164 (1997).
23. S. Stringari, Phys. Rev. Lett. 77, 2360 (1996).
24. A. A. Svidzinsky and A. L. Fetter, Phys. Rev. A 58, 3168 (1998).
25. F. Zambelli and S. Stringari, Phys. Rev. Lett. 81, 1754 (1998).
26. E. Cornell, private communication.
27. L. M. Pismen and J. Rubinstein, Physica D 47, 353 (1991).
28. B. Y. Rubinstein and L. M. Pismen, Physics D 78, 1 (1994).
29. A. A. Svidzinsky and A. L. Fetter, e-print cond-mat/0007139.
Kinetics of Strongly Non-equilibrium
Bose–Einstein Condensation

Boris Svistunov

Russian Research Center “Kurchatov Institute”, 123182 Moscow, Russia

Abstract. We consider the ordering kinetics in a strongly non-equilibrium state of a

(weakly) interacting Bose gas, characterized, on one hand, by large occupation num-
bers, and, on the other hand, by the absence of long-range order. Up to higher-order
corrections in inverse occupation numbers, the evolution is described by non-linear
Schrödinger equation with a turbulent initial state. The ordering process is rather rich
and involves a number of qualitatively different regimes that take place in different
regions of energy space. Specially addressed is the case of evolution in an external
potential.

1 Introduction
Kinetics of Bose–Einstein condensation (BEC) in a weakly interacting Bose gas is
one of the most fundamental problems of non-equilibrium statistical mechanics.
The exciting progress in the experiment with BEC in ultracold gases initiated by
the pioneer works [1] opens up an opportunity of laboratory study of non-trivial
regimes of BEC kinetics.
From the very beginning it should be realized that the statement of the
problem of BEC kinetics involves a number of aspects that are of crucial impor-
tance to the very character of the evolution process. The nature of the process
strongly depends on how the BEC is being achieved: Say, by slow cooling, or
by self-evolution of an essentially non-equilibrium initial state. More generally,
it is important to take into account whether the considerable deviation from
equilibrium occurs only at some sufficiently large length scales and only in the
fluctuation region (so that the kinetics is of universal character and does not
reflect specifics of weakly interacting system), or the non-equilibrium situation
arises far enough from the critical region and the resulting ordering kinetics is
characteristic only to weakly interacting gas. Obviously, the picture of evolution
can be “trivialized” if some portion of condensate is already present in the initial
state. Finally, and especially importantly for the realistic case of a trapped gas,
finite size of the system, or just only that of the condensate can partially or
completely change the relaxation scenario, if this size turns out to be less than
some correlation length relevant to the ordering process in the infinite system.
In this paper we concentrate on a statement of BEC kinetics problem that
we believe to be the most characteristic of the case of weakly interacting gas.
Namely, we consider the self-evolution of weakly interacting gas with a strongly
non-equilibrium initial state. To maximally simplify the consideration without
qualitatively changing the nature of the process, we assume that in the initial

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 327–333, 2001.

c Springer-Verlag Berlin Heidelberg 2001
328 B. Svistunov

state all occupation numbers are either much larger than unity, or equal to
zero, and that there are no correlations between different single-particle modes.
The advantage of choosing such an initial condition is that from the very begin-
ning one can employ classical-field description in terms of non-linear Schrödinger
equation (NLSE), which in the theory of Bose gases is known as Gross-Pitaevskii
equation [2], with a certain turbulent initial condition. It should be stressed that,
in contrast to a wide-spread prejudice, the very description in terms of NLSE
does not imply the presence of condensate, or any sort of dynamical phase tran-
sition (see, e.g., discussion in [3]). The question of the presence of condensate, or,
more generally, the question of (long-range) order is the question of the (long-
range) structure of corresponding classical field.
Hence, the dynamical model for our problem reads ( = 1)
∂ψ Δ
i = − ψ + V (r)ψ + U | ψ |2 ψ , (1)
∂t 2m
where | ψ |2 is interpreted as particle density (not the condensate density!), m
is the particle mass, V (r) is the external potential; U = 4πa/m is the vertex
of the effective pair interaction, a is the scattering length. To introduce the
initial condition to (1) one has to consider the expansion of ψ(r, t) in terms of
eigen
0 modes ϕε (r) [(−Δ/2m + V )ϕε = εϕε ] of linear part of NLSE: ψ(r, t) =
ε a ε (t)ϕ ε (r). Then, at the initial moment t = 0, the phases of the complex
amplitudes aε can be considered as random, while | aε |2 is identified with the
occupation number nε of the mode ε (see, e.g., [3] for more details). To the
best of our knowledge, the first formulation of BEC kinetics problem in terms
of NLSE was given in [4].
A full-scale numeric simulation of NLSE with the turbulent initial condition
could, in principle, cross almost all the t’s in the strongly non-equilibrium BEC
kinetics problem. Such a simulation has not been done yet [5]. Nevertheless, we
will see that from general considerations it is possible to propose the evolution
scenario and to obtain all relevant estimates.
In very general terms, the direction in which the field ψ will evolve is clear
from the following considerations. First, it is natural to expect that the system
must relax to a certain equilibrium state. Secondly, this equilibrium state should
correspond to zero temperature, since the classical field described by the equation
(1) forms a heatbath at absolute zero with respect to itself. Hence, if there is a
stable groundstate (that is if U > 0) for a given particle density, then the system
should approach it in this or that way, the excess energy being carried away
(to higher and higher harmonics) by ever decreasing portion of high-frequency
fraction of the field. At the final stage of evolution all the particles are condensed
except for an infinitesimally small high-frequency portion.
Though the general tendency of evolution is clear, the particular relaxation
scenario is not at all self-evident. A detailed analysis [6,7,8] leads to a rather
sophisticated scenario that involves a number of qualitatively different stages.
The evolution starts with an explosion-like wave in energy space, propagating
from higher energies towards the lower ones, that leads to a formation of a
specific power-law distribution of particles. Immediately after its formation, this
 Kinetics of Strongly Non-equilibrium Bose–Einstein Condensation 329

distribution starts to relax. Simultaneously, in the low-energy region the so-

called coherent regime sets in that leads to the formation of quasi-condensate
correlation properties. Basically, the quasi-condensate state corresponds to what
is known in the theory of superfluidity as the state of superfluid turbulence. It
can be viewed as a condensate containing a tangle of vortex lines (plus a specific
sharply non-equilibrium distribution of long-wave phonons). The formation of
the quasicondensate occurs very rapidly (characteristic time is much smaller than
the time of the wave formation). In contrast to it, the final stage of long-range
ordering, associated with relaxing superfluid turbulence and long-wave phonons,
takes a macroscopically large time.
In the present paper we render the homogeneous BEC scenario [6,7,8] (Sects.
2 and 3) and project it onto the case of a trapped gas. We find out that in an
external potential the evolution picture can be even more rich.

2 Kinetic Regime
During some initial period of evolution the correlations between different ampli-
tudes aε are vanishingly small. Such a regime (known in the theory of non-linear
classical-field dynamics as weak turbulence [9]) admits a description in terms of
kinetic equation. This stage is thus referred to as kinetic stage.
Kinetic equation corresponding to the weak-turbulence regime of non-linear
Schrödinger equation belongs to a generic class of scale-invariant models with
four-wave particle- and energy-conserving interaction, that allows an analysis
of evolution kinetics in general terms (see, e.g., [6]). For the BEC kinetics the
analysis suggests that there are two alternative ways of the initial evolution: (i)
shrinking of the particle distribution as a whole towards ε = 0 (during infinite
time), or (ii) a specific wave in the energetic space leading to a singularization of
distribution at the point ε = 0 at some finite time moment t = t∗ . The answer
to the question of which scenario takes place for a given model depends only on
the scaling properties of the collision integral and the density of states; and the
case of NLSE corresponds to the scenario (ii).
The evolution at the beginning of the kinetic stage results in the formation
of self-similar wave in the energy space propagating in an explosion-like fashion
from the high-energy region (where the particles are initially distributed) towards
lower energy scales. Corresponding self-similar solution of the kinetic equation
has the form [6]

nε (t) = Aε−α
0 (t)f (ε/ε0 (t)) , t ≤ t∗ , (2a)
ε0 (t) = B | t∗ − t |
1/2(α−1)
. (2b)

Here A and B are dimensional constants depending on the initial condition

and related to each other by the formula B = const(m3 U 2 A2 )1/2(α−1) . The
dimensionless function f (numeric data for f see in [6]) is defined up to an
obvious scaling freedom. The explosion character of the evolution guarantees
that the wave reaches the point ε = 0 at some finite time moment t = t∗ [t = 0
330 B. Svistunov

corresponds to the beginning of evolution], the value of t∗ being on the order of

the typical time of (stimulated) collisions in the gas at t = 0. Physically, this
explosion-like evolution is supported by the stimulation of the collision rate at
the head of the wave, ε ∼ ε0 , by ever growing occupation numbers.
Generally speaking, the index α in (2a)-(2b) cannot be established from the
scaling properties of the collision term of the kinetic equation, being related
thus to the particular form of the latter. It is possible, however, to specify lower
and upper limits for α following from the consistency of (2a) and (2b) with
the requirement that these formulae describe an explosion-like singularization of
distribution (rather than infinite-time shrinking). To this end we note that from
the scale invariance it follows that f (x) behaves like some power of x at x  1.
At t = t∗ the occupation numbers have to be finite at ε > 0, hence
f (x) → x−α at x → ∞ . (3)
The requirement that the particle distribution does not shrink as a whole implies
that the number-of-particles integral for the distribution (2a), (3) is divergent at
ε → ∞. This immediately yields α < 3/2. The condition α > 1 is necessary for
ε0 (2b) to approach zero at t = t∗ . So we have 1 < α < 3/2). The most accurate
up-to-date numeric analysis of α was performed in [10] with the result α ≈ 1.24.
At t > t∗ kinetic description is still valid for not so small energies, but
to obtain an adequate sewing with the solution (2a)-(2b) one has to explicitly
introduce the (quasi)condensate, employing the conservation of the total number
of particles (see the discussion in [6]). The structure of the self-similar solution
nε (t) = Aε−α ˜
0 (t)f (ε/ε0 (t)) , ε > 0 , t ≥ t∗ (4)
[f˜(x) → f (x) at x → ∞] corresponds to a back wave in the energy space,
destroying the singular distribution created by the wave (2a)-(2b). The particles
being released during this destruction go directly to quasicondensate. For the
quasicondensate density n0 we thus have
 ∞
A √
n0 (t) = 3/2 3/2−α
(2m) ε0 (t) dx x[x−α − f˜(x)] ∝ (t − t∗ )(3−2α)/4(α−1) .
4π 2 0
(5)
As follows from general considerations and is supported by direct numeric anal-
ysis [6], f˜(x) ∝ 1/x at x  1, which means that the back wave creates a quasi-
equilibrium distribution at ε  ε0 (t) [with infinite at t = t∗ and ever decreasing
afterwards temperature ∝ ε−α 0 (t)].
To estimate the parameter A for a given initial conditions one extrapolates
the solution (2a)-(2b) to a region of energies ∼ εinit , where the particles were
initially concentrated with typical occupation numbers nεinit . This immediately
yields A ∼ nεinit εαinit .

3 Coherent Regime
Strictly speaking, evolution in a kinetic regime does not lead to the ordering.
It is seen from the fact that the description in terms of kinetic equation is
 Kinetics of Strongly Non-equilibrium Bose–Einstein Condensation 331

associated with the random phase approximation (RPA) and thus valid only
when the phases of aε ’s are practically uncorrelated. In such a state even local
order (quasicondensate) is absent. Quasicondensation implies a strong change
of the correlation properties as compared to the RPA state [7]. It occurs in the
regime of strong turbulence (so-called coherent regime), when typical time of
evolution is comparable to the time of oscillation of the phases of relevant aε ’s.
The essence of the process of the quasicondensate formation is the transformation
of the strong turbulence into the state known as superfluid turbulence.
The degrees of freedom associated with the quasicondensate are the same
as in a genuine condensate. These are phonons and topological defects (vortex
lines). That is quasicondensate can be viewed as a condensate with (i) a tangle of
vortex lines and (ii) strongly non-equilibrium distribution of long-wave phonons
implying strong fluctuations of the phase of the quasicondensate part ψ0 of the
field ψ at large distances.
Given the solution (2a)-(2b) of the kinetic equation, one readily estimates
where and when the coherent regime sets in, and what is the typical density of
the quasicondensate upon its formation. The characteristic time of evolution at
−1
the energy scale ε is the collision time τcoll (ε) ∼ m3 (U εnε )2 . For nε ∼ Aε−α the
RPA criterion τcoll (ε)ε  1 becomes invalid at ε ∼ εcoh = (m3 U 2 A2 )1/(2α−1) .
The distribution nε ∼ Aε−α at the scale εcoh is formed at the time tcoh obeing an
obvious relation ε0 (t = t∗ − tcoh ) ∼ εcoh . Hence, tcoh estimates the time moment
when the coherent regime sets in. The time interval | t∗ −tcoh | ε−1 coh  t∗ is a typ-
ical time of the process of quasicondensate formation (in the strong turbulent
regime all characteristic times and distances scale with ε−1 coh and (mεcoh )
−1/2
,
init
correspondingly). The initial quasicondensate density n0 is the density corre-
sponding to the harmonics ε ∼ εcoh . It obeys an obvious relation ninit 0 U ∼ εcoh ,
in accordance with the fact that in strong turbulent regime kinetic and potential
energies are of the same order. The initial spacing between the vortex lines in
the quasicondensate scales as (mεcoh )−1/2 .
The coherent regime and the back wave are practically independent processes,
with a reservation that the coherent part of the field is being pumped with
particles from the high-energy region. This pumping, however, does not affect
the character of the coherent evolution.
Relaxation of the quasicondensate towards genuine condensate goes in two
directions: (i) relaxation of the vortex tangle [that is relaxation of the superfluid
turbulence] and (ii) relaxation of the long-wave phonons. Both processes require
a macroscopically large time (an analysis of this stage of evolution see in [8]).

4 External Potential
In the experiments with trapped ultracold gases normally there takes place the
Knudsen regime, when free path length of a particle with the energy ε, lfree (ε),
is much larger than the typical radius of the particle’s trajectory, Rε . For defi-
niteness we consider a parabolic trap with all the three frequencies of the same
332 B. Svistunov

order ω0 . So that Rε ∼ v(ε)/ω0 [v(ε) is the typical velocity corresponding to the

energy ε], and the condition lfree (ε)  Rε is equivalent to τcoll (ε)ω0  1.
Knudsen regime is a very convenient starting point for analyzing kinetics
in a potential. Almost in all qualitative aspects it corresponds to an isotropic
homogeneous case, since the distribution of particles depends only on the two
variables, ε and t (ergodic approximation). The main quantitative difference
comes from the difference in the density of states, the scaling of the collision
time remaining the same.
Initial picture of evolution is described by the self-similar wave (2a)-(2b)
with α ≈ 1.6 [11,12]. The question then is: What happens when t → t∗ ? The
effect of the potential can be associated with two rather different reasons. The
first reason is the essential discreteness of the low-lying energy levels, which
−1
drastically changes the kinetics when τcoll (ε) < Δ(ε), where Δ(ε) is the typical
interlevel spacing. The second reason is the violation of the Knudsen regime
at the head of the wave at some stage of evolution [because of the decreasing
τcoll (ε0 (t)) with t].
Remarkably, the above-mentioned two circumstances arise always separately,
and (apart a certain cross-over region) there is nothing in between. More specif-
ically, as it immediately follows from the non-equality Δ(ε) ≤ ω0 and the fact
that the relevant collision time τcoll (ε0 (t)) permanently decreases, if there occurs
a break-down of Knudsen regime (in corresponding region of coordinate space
with a typical size RKn around the center of the potential), the discreteness of
levels will never become relevant.
The case, when the discreteness of levels starts to act within the Knudsen
regime, is rather transparent physically and is studied to a large extent both
experimentally [13] and theoretically [11,12]. The evolution scenario in this case
is as follows. When the wave reaches the scale where the level discreteness be-
comes relevant, its further propagation is suppressed (essentially discrete har-
monics practically do not interact with each other) and the back wave is formed.
At ε < ε0 (t), the back wave generates quasi-equilibrium distribution with time-
dependent permanently decreasing temperature T (t) and permanently increasing
number of particles. Condensation thus occurs in a quasi-equilibrium way, with-
out the coherent stage (interaction between low-lying harmonics is negligible),
and the whole process can be described within the kinetic approach [11,12].
We are mostly interested in the case, when at some t = tKn the Knudsen
regime breaks down for energies εKn ∼ ε0 (t = tKn ). From (2a)-(2b) we estimate
εKn ∼ [m3 U 2 A2 /ω0 ]1/2(α−1) . The size of corresponding spatial region is defined
by mω02 RKn 2
∼ εKn . We argue that within the region r < RKn at t > tKn
the external potential becomes irrelevant at least until the quasicondensate is
formed, so that the most important evolution stage basically does not differ from
the homogeneous case. Indeed, it is quite natural that further evolution within
the region r < RKn will result in the formation of anti-Knudsen regime lfree (ε) 
RKn for ε  ε0 (tKn ), because of increasing collision rate with increasing the
occupations numbers. In the anti-Knudsen regime the evolution during the time
period on the order of collision time is insensitive to the external potential (the
 Kinetics of Strongly Non-equilibrium Bose–Einstein Condensation 333

criteria for the Knudsen regime and for the sensitivity to the potential within the
collision time coincide). But this time is enough to form the wave (2a)-(2b) in the
energy space (with the exponent α corresponding to the homogeneous case) and
then to form quasicondensate. During the wave evolution in the energy space, the
free-path length of the particles with ε ∼ ε0 (t) is getting progressively smaller,
which renders the proposed scenario self-consistent. A minor deviation from the
pure homogeneous picture is that now the moment t∗ depends on the distance
from the center of the potential, so that the coherent regime first should start
at r = 0 (the point of maximal initial density) and then gradually occupy all
the anti-Knudsen region up to r ∼ RKn . By this moment the quasicondensate
is formed at r ≤ RKn . In terms of the total number of particles, N , and the
typical single-particle energy of the initial distribution in the potential, εin , the
2(α−3) 1/4(α−1)
estimate for RKn is: RKn ∼ [U 2 N 2 m5−2α ω09−4α εin ] .
On the basis of the above discussion one can introduce the parameter

p = N 2 ω0 m3 U 2 (ω0 /εin )6−2α , (6)

that determines which of the two regimes takes place under given initial con-
ditions: the discrete-harmonic regime (p  1), or the superfluid-turbulence one
(p  1).
For more details on the ordering kinetics in a trapped gas see [14].

References
1. M.H. Anderson et al.: Science 269, 198 (1995); C.C. Bradley et al.: Phys. Rev.
Lett. 75, 1687 (1995); K.B. Davis et al.: Phys. Rev. Lett. 75, 3969 (1995)
2. V.L. Ginzburg and L.P. Pitaevskii: Zh. Eksp. Teor. Fiz. 34, 1240 (1958) [Sov. Phys.
JETP 7, 858 (1958); E.P. Gross: J. Math. Phys. 4, 195 (1963)
3. Yu. Kagan and B.V. Svistunov: Phys. Rev. Lett. 79, 3331 (1997)
4. E. Levich and V. Yakhot: J. Phys. A: Math. Gen. 11, 2237 (1978)
5. Note, however, an interesting preliminary study performed in [K. Damle, S.N. Ma-
jumdar, and S. Sachdev: Phys. Rev. A 54, 5037 (1996)]; as well as the all-quantum
simulation [P.D. Drummond and J.F. Corney, Phys. Rev. A 60, R2661 (1999)],
where the evidence of spontaneous vortex formation in a trapped gas was reported.
6. B.V. Svistunov: J. Moscow Phys. Soc. 1, 373 (1991)
7. Yu. Kagan, B.V. Svistunov, and G.V. Shlyapnikov: Zh. Eksp. Teor. Fiz. 101, 528
(1992) [Sov. Phys. JETP 75, 387 (1992)]
8. Yu. Kagan and B.V. Svistunov: Zh. Eksp. Theor. Fiz. 105, 353 (1994) [Sov. Phys.
JETP 78, 187 (1994)]
9. See the contribution by S. Nazarenko to this volume for details.
10. D.V. Semikoz and I.I. Tkachev: Phys. Rev. D 55, 489 (1997)
11. C.W. Gardiner et al.: Phys. Rev. Lett. 81, 5266 (1998); M.J. Davis, C.W. Gardiner,
and R.J. Ballagh: Phys. Rev. A 62, 63608 (2000)
12. M.J. Bijlsma, E. Zaremba, and H.T.C. Stoof: Phys. Rev. A 62, 63609 (2000)
13. H.-J. Miesner et al.: Science 279, 1005 (1998)
14. B.V. Svistunov: cond-mat/0009295 (to be published in Phys. Lett. A)
Quantum Nucleation of Phase Slips
in Bose–Einstein Condensates

H.P. Büchler1 , V.B. Geshkenbein1,2 , and G. Blatter1

1
Theoretische Physik, ETH-Hönggerberg, CH-8093 Zürich, Switzerland
2
Landau Institute for Theoretical Physics, 117940 Moscow, Russia

Abstract. We present a theoretical study of quantum fluctuations in a Bose-Einstein

condensate confined within a thin cylindrical trap and perturbed by a moving impurity.
We derive an effective action which maps the problem to that of a massive particle with
damping in a periodic potential. Quantum fluctuations lead to a finite nucleation rate
of phase slips and we make use of known results in our determination of the transport
characteristic. Real Bose-Einstein condensate are finite systems and exhibit interesting
effects depending on topology: in superfluid rings we obtain a critical velocity below
which the nucleation rate is quenched. In a cigar shaped condensate the low-energy
action is equivalent to that of a capacitively shunted Josephson junction. The state
with a well defined phase difference across the impurity then is unstable towards a
decoupled state with a fixed number of particles on either side of the impurity.

1 Introduction
A spectacular phenomenon of quantum fluids is the friction free transport, often
studied in terms of the Josephson effect across a constriction or orifice separating
two reservoirs. Recently, new types of experiments with a moving laser beam per-
turbing a condensate of sodium atoms enable the observation of superfluid flow
in Bose-Einstein condensates [1] and the presence of a critical velocity beyond
which the nucleation of phase slips leads to dissipation. Reducing the transverse
dimension of the condensate increases the effect of quantum fluctuations and
leads to drastic changes in the transport characteristics. Here, we study the ef-
fect of quantum fluctuations on the superfluid flow through an impurity in a
quasi one-dimensional Bose-Einstein condensate.
Transport properties in charged superfluids are conveniently studied by cur-
rent biasing the circuit via external electrical devices. In particular, using such
experimental setups, the quantum fluctuations of the superconducting wave func-
tion in a small Josephson junction device [2] and in thin superconducting and
superfluid wires have been examined [3,4,5]. Recently, a new type of experiments
has been set up determining the superfluid flow of a weakly interacting Bose-
Einstein condensate around an impurity [1]. In this setup the flow is induced by
the motion of an impurity, e.g., a laser beam repelling the atoms from its focus.
In our analysis we chose this setup of a moving impurity in a Bose-Einstein
condensate for studying the quantum nucleation of phase slips.
The effect of phase slips is most conveniently studied in a superfluid ring
with length L where the impurity is moving with uniform velocity v around the

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 334–341, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Quantum Nucleation of Phase Slips 335

a ) v b ) v c ) v
Im p u rity S u p e rflu id

Fig. 1. Superfluid in a small thin cylindrical trap with a moving impurity: a) ring
structure with periodic boundary condition for the condensate wave function, b) infinite
wire, c) finite system of length L with closed ends.

ring (see Fig. 1a). The creation of a phase slip leads to an additional winding
of the superfluid phase describing an acceleration of the superfluid, which in
turn is related to a suppression of the order parameter, i.e., particles leave the
condensate. The finite length L together with the periodic boundary conditions
provide us with a quantization of the velocity of the superfluid in units 2vL =
2π/Lm, i.e., a phase slip leads to an increase of the mean velocity by 2vL . As a
consequence, the relative velocity vrel between the superfluid and the uniformly
moving impurity decreases by a rate which can be estimated as
∂t vrel = −2 Γ (vrel ) vL , (1)
where Γ denotes the quantum nucleation rate for phase slips. When the relative
velocity between the superfluid and the impurity is smaller than the critical
velocity necessary for the nucleation of phase slips we reach an equilibrium state.
Switching on the laser beam during a time Δt the amount of energy transferred
to the superfluid is given by

ΔE = dtΓ (v)EPS (v), (2)
Δt

with EPS (v) the energy transferred by a single phase slip event. Two methods
for measuring this energy transfer have been applied in experiments on bulk
superfluids: determination of the heating of the condensate by measuring the
thermal fraction [1,6] and by direct observation of the flow field in the superfluid
via in situ non-destructive imaging of the condensate [6].
Increasing the length of the superfluid ring, finite size effects become irrel-
evant and the behavior of the ring approaches that of an infinitely long wire.
This situation is illustrated in Fig. 1b. In contrast to the ring setup, the impurity
never returns to its initial position. As a consequence, only parts of the super-
fluid are accelerated by a phase slip and waves traveling with a velocity close to
the sound velocity carry away the energy transferred to the superfluid.
Closing the wire with impenetrable boundaries at a distance L (see Fig. 1c),
these sound waves are reflected and provide a homogeneous heating of the sample
due to the nucleation of phase slips. In contrast to the previous setups the
impurity oscillates instead of moving with uniform velocity, and the mean relative
velocity between the superfluid and the impurity does not decrease. The amount
of energy transferred to the superfluid is given by Eq. (2).
336 H.P. Büchler, V.B. Geshkenbein, and G. Blatter

2 Effective Action
We start with the Gross-Pitaevskii Lagrangian describing the dynamics of the
confined condensate wave function of the weakly interacting bosons [7]. With m
the mass of the bosons, U the strength of the repulsion, and ρ0 the condensate
density, the Lagrangian in one dimension reads
2 U, -2
LGP = i ψ∂t ψ − (∂x ψ)(∂x ψ) − ψψ − ρ0 . (3)
2m 2
√
The repulsive interaction produces the healing length ξ= / mρ 0 U and a
finite compressibility κ related to the sound velocity cs = κ/mρ0 = ρ0 U/m.
Comparing the length scale ξ with the scale given by the condensate density 1/ρ0
provides us with a dimensionless parameter K = πξρ0 , the number of particles
per healing volume measuring the relevance of fluctuations. The condition of
weakly interacting bosons where a mean field description is valid translates into
K  1. The energy scale of the interaction is given by the chemical potential
μ = ρ0 U . The impurity is described by a suppression of the chemical potential

Lint = −V (x − vt)ψψ (4)

with v the velocity of the impurity. We will consider a short range interaction
between the impurity and the superfluid, i.e., V (x) = gξδ(x), with impurity
strength g  μ. The finite size of the system is taken into account via boundary
conditions for the condensate wave function. In the following we will consider
three different setups (see Fig. 1): a superfluid ring with periodic boundary
condition ψ(x, t) = ψ(x+L, t), the infinite system, and the superfluid tube where
the flow vanishes at the ends of the tube, i.e., ∂x φ(L/2, t) = ∂x φ(−L/2, t) = 0,
√
where φ denotes the phase of the wave function ψ = ρ0 (1 + h) exp(iφ).
The integration over all degrees of freedom except the phase difference ϕ
across the impurity is carried out in the imaginary time description of the su-
perfluid and provides us with the effective action [8]
 β 
T +
2
S |ϕ(ωs )| EJ
= α(ωs ) + dτ [1 − cos ϕ(τ )] − ρ0 vϕ(τ ) (5)
  s 2 0 

where β denotes the inverse temperature T , while ωs = 2πT s/ are the Mat-
subara frequencies with s ∈ Z. The static Josephson like potential with the
Josephson energy EJ = Kμ2 /gπ derives from the interaction of the superfluid
with the impurity, while the driving term is induced by the motion of the im-
purity. The first term describes the dynamics of ϕ and follows from integrating
out the low energy excitations in the leads. The kernel α(ω) depends on the
boundary conditions of the condensate wave function.
We start with the simplest setup: the infinitely long tube shown in Fig. 1b.
Then the kernel α(ω) reads
K  2
α(ω) = αd (ω) + αm (ω) = |ω| + ω . (6)
2π EC
 Quantum Nucleation of Phase Slips 337

The first term describes a Caldeira-Leggett type [9] dissipation via radiation of
sound waves, while the second term introduces a kinetic term with ‘charging’
energy EC = 4π 2 EJ /K 2 . The characteristic time scale τc of the action (5) is

K 1 4π 1
τc = = (7)
π EJ K EC

and derives from a comparison of the different terms in the action K/4π ∼
/EC τc ∼ EJ τc /. This scale is much slower than the time scale τ0 = /μ of the
unperturbed superfluid, τc = (g/μ)τ0 , justifying the assumption that only low
energy excitations in the leads contribute to the effective action (5).
The action (5) with the kernel (6) is also known to describe a resistively
and capacitively shunted Josephson junction (RCSJ-model) [10]between two
charged superconductors. The quality factor defined by Q = 2π EJ /(K 2 EC )
separates two regimes: the junction is overdamped with the damping dominating
over the inertia if Q < 1, while for Q > 1 we enter the underdamped regime
with a hysteretic current-voltage characteristic [10]. In the uncharged situation
considered here, the relation EC = 4π 2 EJ /K 2 between the ‘charging’ energy
and the Josephson energy leads to Q = 1, and the action is at the border
between the overdamped and the underdamped regime. Then effects leading
to hysteretic behavior become only relevant for velocities close to the critical
velocity v ∼ vc = EJ /ρ0 .
The effect of quantum fluctuations in the RCSJ model has been extensively
studied [9,11,12]. The system shows a quantum phase transition at K = 1 [11].
For K < 1 quantum fluctuations dominate and the ground state is described by
a delocalized phase and the system exhibits a linear response Δμ = ∂t ϕ =
ρ0 v/K characteristic for a normal fluid, i.e., the phase coherence of the con-
densate across the impurity is suppressed and a fixed particle number N is
established [12]. On the other hand, for K > 1 describing a weakly interacting
Bose-Einstein condensate the damping is strong and reduces the effect of quan-
tum fluctuations. Then the ground state is characterized by a fixed (localized)
phase ϕ allowing for a superflow v = vc sin ϕ across the impurity with a vanishing
linear response Δμ = ∂t ϕ = 0 at zero temperature. However, the quantum
nucleation of phase slips provides us with algebraic corrections to the response
function ∂t ϕ = 2πΓ with the rate [13] (see Fig. 2)
 2K−1 2
y2 2πτc |Γ (K + iρ0 vβ)|
Γ = sinh (πρ0 βv) . (8)
τc β Γ (2K)

At zero temperature, the quantum nucleation is algebraic in v, Γ ∼ v 2K−1 , while

a finite temperature and small driving forces lead to a linear quantum nucleation
rate Γ ∼ vT 2K−2 with an algebraic temperature dependence. Finally, the energy
EPS transferred to the superfluid by a single phase slip takes the form
  
v
EPS = iv dxψ∂x ψ = 2Kμ . (9)
τ =0 cs
338 H.P. Büchler, V.B. Geshkenbein, and G. Blatter

ln ( Γ )
Γ ∼ v
2 K -1

T = 0
Γ ∼ v

T = 0

ln ( v T ) ln ( v )

Fig. 2. Quantum nucleation rate of phase slips: algebraic behavior Γ ∝ v 2K−1 at

zero temperature, and a linear behavior Γ ∝ vT 2K−2 for small velocities and finite
temperature with a crossover at vT = πT /mcs .

3 Finite Size Effects

Ring: The periodic boundary conditions in a superfluid ring of length L (see
Fig. 1a) provide us with a quantization of the velocity of the superfluid by
2vL = 2π/Lm. The relative velocity between the impurity and the superfluid
takes the form
vL
vrel = vc sin ϕ = v − ϕ, (10)
π
where the first term is the flow induced by the motion of the impurity, while the
second term describes the flow of the superfluid in the leads with ϕ ∈ [2πk, 2π(k+
1)); k ∈ Z is the winding number of the phase. The relative velocity vrel drives
the system and determines the quantum nucleation rate. For large rings with
L > ξc = cs τc and k = 0 we can drop the second term and the relative velocity
is determined by the velocity of the impurity alone. The kernel
⎧
cosh ωL/cs + 1 ⎨
2Kcs /L, ω < cs /L,
αd (ω) = Kω = (11)
sinh ωL/cs ⎩
K|ω|, ω > cs /L.
is modified at low frequencies: first, the finite size produces a quantization of the
available sound modes, leading to a suppression of the spectral density at low
frequencies. Second, the closed structure of the setup induces a self-interaction
resulting in an additional static potential ∝ (KL/cs )ϕ2 which quenches the low
frequency quantum fluctuations in the phase ϕ. For any value of K the ground
state then is characterized by a fixed phase ϕ allowing for a superfluid flow across
the impurity. For K > 1, the velocity vL is a critical velocity: for v < vL the
phase coherence is not reduced by quantum fluctuations and the superfluid flows
free of dissipation through the impurity; at the critical velocity vL , there is a
sharp onset of phase slips which goes over to the nucleation rate for the infinite
system (8) as the velocity increases.

Tube: Next, we study a superfluid in a tube of finite length L (see Fig. 1c). The
finite size again produces a quantization of the modes in the leads which sup-
presses the spectral density and modifies the damping kernel at low frequencies,
 Quantum Nucleation of Phase Slips 339

⎧
cosh ωL/cs − 1 ⎨ (KL/2cs ) ω 2 , ω < cs /L,
αd (ω) = Kω = (12)
sinh ωL/cs ⎩
K|ω|, ω > cs /L.
The low frequency action then reduces to that of a capacitively shunted Joseph-
son junction,
 2
S  (∂τ ϕ) EJ
= dτ + [1 − cos ϕ] + L> (13)
 EL 2 

with the ‘charging’ energy EL = cs /KL. The additional term L> accounts for
high frequency contributions ω > cs /L and includes a kinetic (αm , see (6)) and
a dissipative term. For long wires L > ξc the mass term αm (ω) = (/EC ) ω 2 is
small, while it does not appear altogether in the case of short wires L < ξc .
We start with systems where the characteristic frequency is smaller than cs /L
allowing us to drop L> in determining the ground state properties of (13). This
condition
√ is satisfied
 for short tubes L < ξc with the characteristic frequency
EJ EL / = L/πξc (cs /L) < cs /L. Then, the low frequency eigenstates of
the system (13) are characterized by Bloch states in the ‘quasinumber’ N , see
Ref. [12] (the ‘quasinumber’ N plays the role of the ‘quasimomentum’ k in the
periodic crystals). For weakly interacting bosons with EJ /EL = K 2 L/πξc  1
a tight binding analysis is applicable and the lowest energy band takes the form

E(N ) = W cos (2πN ) , (14)

where the band width W can be calculated in the instanton approach using a
single kink solution [14]
 L   
W ∼ exp −K L/ξc , (15)
τc ξc
with ξc = cs τc . Comparing to the classical Josephson effect, the fixed phase ϕ
driving the superfluid velocity v = vc sin ϕ in the Josephson junction is now
replaced by the fixed ‘quasinumber’ N driving the chemical potential difference
Δμ = 2πW sin (2πN ). Similarly, ∂t ϕ = μ/ → ∂t N = ρ0 v. Then for a slowly
moving impurity the system exhibits ‘Bloch oscillations’ in the chemical po-
tential Δμ(t) = 2πW sin(2πρ0 vt) in analogy to the ac Josephson effect. These
oscillations are due to the accumulation of particles in front of the impurity, the
latter allowing only discrete particles to tunnel. Each ‘Umklapp’ process then
describes a particle tunneling through the impurity. Two mechanisms lead to the
disappearance of the ‘Bloch oscillations’ at larger driving forces: first, transitions
to higher bands, known as √ ‘Zener tunneling’, become
 relevant for frequencies ρ0 v
larger than the band gap EJ EL (v > (vL /K) L/ξc ) and describe the accu-
mulation of particles in front of the impurity, i.e., absence of ‘Umklapp’ processes
quenches the tunneling. The flow across the impurity then does not match the
flow of particles towards the impurity, producing a steady increase in the chem-
ical potential difference Δμ. Second, for ρ0 v > cs /L, or equivalent v > vL /K,
340 H.P. Büchler, V.B. Geshkenbein, and G. Blatter

the damping terms in L> localize the phase and a crossover to the classical
Josephson effect appears: the dissipation projects the phase ϕ and renders it
sharp. In an actual experiment, we start with the sample in the superfluid state
characterized by a homogeneous phase. Switching on the laser decouples the
Bose-Einstein condensate into two subsystems. For a vanishing velocity of the
impurity, the initially well defined phase difference ϕ evolves according to the
action (13) and spreads over the various minima within the characteristic time
scale /W . The observed phenomenaare then determined by the Bloch states
and a weak driving force v < (vL /K) L/ξc provides us with ‘Bloch oscillations’
in the chemical potential Δμ. Increasing the driving force, v > vL /K the sys-
tems enters a regime where ‘Zener’ tunneling and dissipation compete with the
physics of ‘Bloch oscillations’. Note that this regime is terminated by the critical
velocity vc of the link, as vc < vL in these short tubes.
For long wires with√L > ξc the terms L> in (13) are relevant and the char-
acteristic frequency is EJ EC / ∼ cs /ξc > cs /L. The ground state properties
of the system and low drive response then involves a more difficult interplay
between dissipation and Bloch physics which requires more study. The situation
simplifies at larger drives v > vL : all processes are fast involving frequencies
ω > cs /L and the damping term in (12) dominates over the inertia. Then we
recover the physics of the infinitely long system: the phase ϕ becomes localized
allowing for a superfluid flow through the impurity and the quantum nucleation
rate calculated via the instanton approach reduces to the result (8).
In conclusion, quantum fluctuations in the condensate wave function of Bose-
Einstein condensates brings about fascinating new phenomena in superfluids and
their experimental observation is a challenging task for the future providing us
with a better understanding of macroscopic quantum effects.

References
1. C. Raman et al., Phys. Rev. Lett. 83, 2502 (1999).
2. J. M. Martinis, M. H. Devoret, and J. Clarke, Phys. Rev. B 35, 4682 (1987).
3. N. Giordano, Physica B 203, 460 (1994).
4. A. Bezryadin, C. N. Lau, and M. Tinkham, Nature 404, 971 (2000).
5. A. D. Zaikin, D. S. Golubev, A. van Otterlo, and G. T. Zimányi, Phys. Rev. Lett.
78, 1552 (1997); V. A. Kashurnikov, A. I. Podlivaev, N. V. Prokof’ev, and B. V.
Svistunov, Phys. Rev. B 53, 13091 (1996); Y. Kagan, V. N. Prokof’ev, and B. V.
Svistunov, Phys. Rev. A 61, 045601 (2000).
6. R. Onofrio et al., Phys. Rev. Lett. 85, 2228 (2000).
7. F. Dalfovo, S. Giorgini, L. P. Pitaevksii, and S. Stringari, Rev. Mod. Phys. 71,
463 (1999).
8. H. P. Büchler, V. D. Geshkenbein, and G. Blatter, to be published.
9. A. O. Caldeira and A. J. Leggett, Ann. Phys. (N.Y.) 149, 374 (1983).
10. M. Tinkham, Introduction to Superconductivity, McGraw-Hill, 1996.
11. A. Schmid, Phys. Rev. Lett. 51, 1506 (1983); S. A. Bulgadaev, Sov. Phys. JETP
Lett. 39, 317 (1984).
12. G. Schön and A. D. Zaikin, Phys. Rep. 198, 237 (1990).
 Quantum Nucleation of Phase Slips 341

13. U. Weiss, H. Grabert, P. Hänggi, and P. Riseborough, Phys. Rev. B 35, 9535
(1987).
14. S. Coleman, Phys. Rev. D 16, 2929 (1977); C. G. Callan and S. Coleman, Phys.
Rev. D 16, 1762 (1977); S. Coleman, Aspects of symmetry, Cambridge University
Press, 1988.
Vortex Reconnection in Normal and Superfluids

Joel Koplik

Benjamin Levich Institute and Department of Physics

City College of the City University of New York
New York, NY 10031 USA

1 Introduction
An example of vortex reconnection is shown in a time sequence in Fig. 1, wherein
two distinct vortex filaments in a fluid move together, merge, and then divide into
two or more filaments moving away, with part of one initial filament connected
to part of the other. The physics underlying this example [1] will be presented
later, but the key feature is the evident change in the topology of the vortices. In
this lecture, we will discuss vortex reconnection in both normal and superfluids,
emphasizing the relevance of the process to their respective turbulent flows, the
similarities between the two cases, and the computational issues. The lecture
is aimed at a fairly general audience: no detailed knowledge of fluid mechanics
is assumed beyond a nodding acquaintance with the Navier–Stokes equation,
and nothing about superfluidity beyond the idea of a two-fluid system with an
quantum-mechanically condensed component, and a willingness to accept the
Gross-Pitaevskii model for the latter. My emphasis will be on the superfluid
case, and the reconnection process in normal fluids is discussed in more detail in
the cited literature. My original work reviewed here was done in collaboration
with Herbert Levine. Related and more recent work along these lines is described
in the lectures by Adams and Roberts in this volume.

2 Some Vortex Generalities

The simplest example of a vortex flow field [2] is the infinite line vortex,
κ
u(r) = φ̂ (1)
2πr
corresponding to steady rotation about the z-axis (in cylindrical coordinates
r, φ, z). The vorticity is defined generally as as ω = ∇ × u, which in this case has
the value (κ/r)δ(r)ẑ, so that the orientation of ω is the rotation axis, and the
magnitude is related to the strength of the rotation. The constant prefactor has
been written in terms of the circulation Γ through any closed curve enclosing
the z-axis,  
Γ ≡ dr · u = dS · ω (2)
C S
where S is a surface whose boundary is C; for the line vortex (1) one finds Γ = κ.
This velocity field is an exact solution of the Euler equation, the Navier–Stokes

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 345–357, 2001.

c Springer-Verlag Berlin Heidelberg 2001
346 J. Koplik

Fig. 1. Reconnection of two superfluid vortex filaments [1].

equation for normal fluids

 
∂u
ρ + u·∇ u = −∇p + μ∇2 u (3)
∂t

with the viscosity μ set to zero, with pressure p = p0 − ρκ2 /8π 2 r2 , and constant
density ρ. The singularity along the axis can be avoided by smoothing the vor-
ticity over a finite core region [3], and various choices are available. For example,
the “Rankine vortex” has uφ = κr/2πa2 inside the core region r < a and the
previous form for r > a outside, so that inside the core the vorticity is uniform,
ω = ẑκ/πa2 , but still zero outside. In the presence of viscosity, these vortices
are no longer exact solutions, and in fact a vortex will decay due to friction in
the absence of external forcing, but for finite intervals one can get often by with
a suitable time-dependent core.
In a superfluid the situation is rather analogous. In this talk we adopt the
Gross-Pitaveskii (GP) model [4] for the wave function of the condensate

∂ψ 2 2 , -
i =− ∇ ψ + V0 |ψ|2 − n0 ψ (4)
∂t 2m
 Vortex Reconnection in Normal and Superfluids 347

As in any quantum mechanical system, we can define a density ρ = |ψ|2 and a

velocity u = j/ρ, where the usual probability current is j = Re [(/im)ψ ∗ ∇ψ].
After some algebra [5], the real and imaginary parts of the GP equation become
the equation of continuity of a compressible fluid, and the Euler equation with
an unusual and singular equation of state,
√
2 ∇2 ρ V 0
p= √ + (n0 − ρ2 ) (5)
2m ρ m

The GP model (4) has exact, steady, stable solutions of the form ψ1 = f (r)eiθ ,
again in cylindrical coordinates, where f (r) is a smooth function with the asymp-
√
totic behaviors f → n0 as r → ∞ and f ∼ r as r → 0. The crossover between
the two limiting forms of f occurs at a “core radius” a0 = 2 /2mn0 V0 , which
for parameters appropriate for 4 He is O(1Å). The corresponding velocity field is

h φ̂
u= (6)
m 2πr
for r  a0 , which is nothing but a line vortex with circulation h/m. Solutions
whose circulation is any integer multiple of this value exist, but are unstable:
the circulation is thus quantized. At smaller r the velocity is messier, but the
vanishing of f (r) at the origin leads to a current which is finite everywhere.
Furthermore, the density ρ = f 2 is a constant outside of the core, and tends
smoothly to zero as r → 0, while the vorticity is non-zero only within the core
region where ρ varies. Thus, outside of a microscopic core region, a superfluid has
vortex excitations whose dynamics is given by the incompressible Euler equation,
and thus isolated superfluid vortices have the same hydrodynamical behavior as
those of an inviscid normal fluid.
An important feature of inviscid fluids which bears on vortex stability and
possible reconnection is the conservation of circulation, or Kelvin’s theorem. The
theorem states that the circulation around a contour which moves with the fluid
is constant in time, and its relevance here is that single vortex filaments cannot
just disappear, whereas two adjacent filaments of equal and opposite vorticity
could annihilate locally. More generally, it appears superficially in Fig. 1 that
some contours may enclose vortices initially but not later, so it is useful to
present the derivation of Kelvin’s theorem, simply to see where it could fail. The
argument is to consider the total or Lagrangian time derivative of the circulation:
  
d du(t) d
u(t) · dr = · dr + u(t) · (dr)
dt C(t) C(t) dt C(t) dt
Note that because the curve C(t) moves with the fluid, it is not differentiated. In
the first integral, du/dt can be replaced with the right-hand side of the Navier-
Stokes equation, ρ−1 [−∇p + μ∇2 u], and in the second integral we write u ·
(d/dt)dr = u · d(dr/dt) = (1/2)d(u2 ). The two total derivatives integrate to
zero, leaving 
dΓ (t) μ
= ∇2 u
dt ρ C(t)
348 J. Koplik

As noted above, circulation is only conserved in the absence of viscosity. A

further limitation on the theorem is that the derivation tacitly assumes that the
functions involved are sufficiently regular to be differentiated inside the integral
and manipulated as above. At a superfluid vortex core, the density vanishes, so
that in view of (5), the pressure is singular and there is cause for concern.
The motion of a vortex filament is a combination of the background fluid
motion plus any self-induced velocity it produces. In general, the velocity field
due to any vorticity distribution may be recovered by a Biot-Savart integral.
In simple cases, the nature of the motion may be deduced by symmetry and
application of the right-hand rule, as in electrodynamics. For example, an iso-
lated infinite straight vortex filament does not move, since there is no preferred
direction. Two such parallel filaments rotate about their center of mass, while
anti-parallel filaments translate in tandem, as can be easily seen by appropriate
gestures with one’s right hand. More generally, for vanishing filament thickness
and the inviscid limit the Biot-Savart law can be reduced [3] to the “local in-
duction approximation”

∂R κ log L/a ∂R ∂ 2 R
≈ × (9)
∂t 4π ∂s ∂s2
Here, points R on the the filament are parametrized by arclength s and time
t, and (a, L) are short and long-distance cutoffs, respectively. The self-induced
velocity is then along the binormal to the filament axis, so for example an isolated
non-straight filament does move, and a circular vortex ring moves along its
centerline, perpendicular to the plane of the ring.

3 The Importance of Reconnection

Vorticity and vortex filaments are an important ingredient in high velocity flows
of both normal and superfluids. Although the turbulent states have important
similarities that motivate the study of reconnection, which we will discuss here,
the corresponding mechanisms by which they arise are of course rather different.
Consider first the behavior of a bucket of a normal viscous fluid as it is spun
up from rest about its axis. The fluid will begin to rotate along with its container,
and at low velocities develop a laminar state with constant angular velocity and
a parabolic free surface at the top. At large enough angular velocity, this uniform
state is unstable and time-fluctuating structures such as spiral vortices appear.
With further increases in rotation rate, these localized fluctuations occur more
frequently, and grow in size and lifetime and completely overwhelm the laminar
flow. At high rotation rates, one sees velocity fields dominated by fluctuations,
whose length scales extend up to the size of the container, and whose time scales
range from the microscopic on up. One crucial feature of fully-developed turbu-
lent flows in normal fluids is the presence of regions of concentrated vorticity,
as in the simulation example shown in Fig. 2a [6], where values of |ω| greater
than a selected threshold are shown. Experimental studies [7], using bubbles
for visualization, likewise indicate the presence of regions of both line-like and
 Vortex Reconnection in Normal and Superfluids 349

sheet-like excitations, which fluctuate in time and space. The precise connection
between vortical structures and turbulence in normal fluid flows is a subject of
current work [8], but there is at least a school of thought that they underlie
the “intermittency” phenomena observed in the turbulent energy spectrum and
structure functions. In terms of the subject at hand, although the evolution of
these high vorticity structures in a fully-developed turbulent flow does not cor-
respond precisely to the mutual interaction of isolated filaments, consideration
of such processes is an obvious ingredient. Aside from its relevance to turbu-
lence, vortex reconnection can occur in many other contexts in normal fluid
dynamics where vortex filaments or other structures are produced, for example
in the wakes of airplane wings or ships, and would have an impact on issues such
as drag reduction. A further application is to the question of whether singular
solutions of the Navier-Stokes equations develop from smooth initial data [9].
Typical configurations which have been used for numerical investigations have
concentrated vorticity, and reconnection may regularize them.

Fig. 2. (Left) Regions of high vorticity in a simulated turbulent flow [6]; (Right) Sim-
ulation of a superfluid vortex tangle [10].

If a bucket of superfluid is rotated about its axis, one sees a rather differ-
ent sequence of states as the rotation rate is increased. At low angular velocity
nothing happens to the fluid at all, until at a critical value a vortex filament is
nucleated, parallel to the rotation axis. The configuration persists until a second
critical velocity, whereupon a second filament appears, and so on. The filaments
are parallel lines, distributed in the orthogonal plane at regular lattice points,
and aside from stability considerations higher rotation rate just produces more
filaments. Of course this configuration is unstable, and any fluctuation of one
filament creates a perturbing velocity field which acts on the others. Unlike the
normal fluid case, however, the individual superfluid vortices are “topologically”
stable due to quantization of circulation, and they simply begin to move in
response to the local velocity field. The result is a disordered and fluctuating
collection of vortex filaments, resembling a pot of boiling spaghetti, the start-
350 J. Koplik

ing point for the calculations of Schwarz [10], and subsequently others, of the
statistical properties of the turbulent superfluid state. There are no experimen-
tal results visualizing this vortex tangle, but a simulation snapshot is given in
Fig. 2b, which evidently resembles its normal fluid counterpart Fig. 2a. Such
calculations are based on the local induction approximation, which is quite rea-
sonable for the widely separated vortices in a superfluid, but the calculation
shows that as these filaments move about they are are prone to intersect each
other. At this point a more microscopic calculation is needed to determine the
outcome of a vortex collision.

4 Reconnection in Normal Fluids

Given an evolution equation for the appropriate fields, either Navier-Stokes or
Gross-Pitaevskii, it is ostensibly a straightforward matter to integrate numer-
ically to find the time-evolution of an initial two-vortex configuration. In the
former case, although there are many well-developed numerical techniques, vor-
tex interactions involve both long and short length scales, and the calculation is
expensive to carry out in detail. Most work I’m aware of dates from ca. 1990, and
is reviewed by Kida and Takaoka [11] with a different slant than in the present
talk. In Fig. 3a,b, we show two representative examples, filament reconnection as
studied by Melander and Hussain [12], and vortex ring merger following Kida,
Takaoka and Hussain [13]. In both cases a spectral (Fourier series) numerical
method was used, the figures are isosurfaces of a selected vorticity magnitude,
the fluid is incompressible, and the initial condition is symmetric in that the two
starting vortices have the same strength. The general result is that adjacent,
equal and oppositely oriented vortex structures tend to merge and then move
apart, perhaps with some transient secondary regions of high vorticity left be-
hind. For the vortex ring case at least, there are visualization experiments [14]
which show the same merger into one ring and subsequent breakup into two new
rings moving in an orthogonal plane.
In terms of understanding the mechanism, the first stage is the fact that
oppositely oriented segments of concentrated vorticity move towards each other.
This behavior follows intuitively from the local induction approximation (9),
as explained by Siggia [15]. In the typical configuration for interaction, vortex
filaments which approach are each bent into a hairpin shape with opposite ori-
entations. When the cores begin to overlap, there is a cancelation of opposite
signed vorticity, leading to the bridging seen in the figures. The words which
usually accompany figures of the intermediate state are that the closest seg-
ments of adjacent opposite vortex lines annihilate in the merger region while
the remainder of these lines simply connect up outside, and the more distant
parts of the opposed vortex lines persist to form the thin bridge connecting the
main filaments as they pull apart. The initial motion of the outer parts of the
filaments then carries the reconnected configuration away.
It would of course be desirable to have an analytic model of the reconnec-
tion process, but those available are not entirely consistent with numerical tests
 Vortex Reconnection in Normal and Superfluids 351

Fig. 3. (Left) Reconnection of anti-parallel vortex filaments in a normal fluid [11];

(Right) Vortex ring merger and breakup in a normal fluid [13].

[16,17]. These authors present very detailed numerical analyses of reconnection

simulations, which differ somewhat in detail, but with no clear and simple in-
sight emerging. Among the additional features which can be identified in the
simulations are vortex dipole formation, flattening of the core regions, localized
vorticity amplification, and jet formation. Much of the difficulty is already evi-
dent in the complicated three-dimensional spatial variation seen in the figures.
What about Kelvin’s theorem? The core of a vortex will entail some viscous
effects (in the line vortex case, u ∼ 1/r would make the viscosity term dominant
at small r, and in the Rankine case velocity gradients are discontinuous at edge
of the core), so the theorem is only valid if applied to contours outside the core,
and therefore outside the reconnection region. If two vortex filaments were to
approach with parallel vorticity, a contour encircling both would have a net
circulation that is the sum of the individual ones, and it would not be possible
for them to reconnect and annihilate this vorticity. Fortunately the relevant case
is anti-parallel, when the locally induction flow outside the core leads to natural
attraction, and there a zero total circulation initially is quite compatible with
reconnection.
If the initial condition is asymmetric, with vortices of different circulation
interacting, it is not possible to cleanly reconnect in this way, because Kelvin’s
theorem applied to a contour encircling the reconnection region implies that
the final circulation is non-zero. Numerical simulations show that an assortment
of final states with exotic final states can occur [18], such as a weak vortex
352 J. Koplik

wrapping around a stronger one. Other simulations have studied the evolution
of intertwined vortex rings, which produce their own bestiary of curious shapes
[19]. Likewise, other effects such as compressibility of the fluid can complicate
the dynamics and alter the final topology [20].

5 Reconnection in Superfluids
The superfluid case turns out to be much simpler and cleaner. One one hand,
there is a very simple and elegant numerical technique available for the nonlin-
ear Schroedinger equation, and on the other, the fact that vorticity is quantized
means that only symmetric cases are relevant, and furthermore it is not possible
to have persistent residual bridges. Two cautionary remarks are needed. First,
the GP model is precisely that, and is not a completely faithful description of
Helium-II. However, it does exhibit vortex configurations which have reasonably
correct behavior as far as is known, and we may regard the details of its wave
function as representing a particular approximation for regularizing what would
otherwise be a singular core region. Most of the motion is just that of an Euler
fluid, and the behavior we find resembles that of normal fluids, so it is plausible
to suppose that we just miss some details of the excitations in the final state. A
second limitation is that we consider pure condensate and no explicit coupling
to normal fluid at all, so strictly speaking we have a zero-temperature system.
Again, to the extent that reconnection phenomena involve only the general as-
pects of the short distance properties of the superfluid, we may hope that the
results are general.
The numerical method used here is a split-time-step algorithm [21] first in-
troduced for the GP equation with the opposite sign of the nonlinearity (an
equation relevant to laser focusing). We [1] first rescale length and time to get
the dimensionless equation
∂ψ
= i∇2 ψ − iψ(|ψ|2 − 1)
∂t
As a variant of the simple Euler method ψ(t + dt) = ψ(t) + dt · ∂ψ/∂t, break the
right-hand side into its two terms and treat them successively. In Fourier space,
assuming periodic boundary conditions, the Laplacian term just integrates into
a phase
2
ψ(k, t) → ψ̃(k, t + dt) = e−ik dt ψ(k, t) (11)
while in real space, since without the Laplacian term one has ∂/∂t |ψ|2 = 0, the
nonlinear term also integrates to a phase,
2
ψ̃(x, t + dt) → ψ(x, t + dt) = e−i(|ψ̃| −1)dt
ψ̃(x, t + dt)

of course, the wave function must be Fourier transformed twice at every time
step, but with periodic boundary conditions this is efficiently done by FFT
algorithms. Thus, the schematic numerical code is
 Vortex Reconnection in Normal and Superfluids 353

do i=1, nsteps
FFT ψ
2
ψ → e−ik dt ψ
inverse FFT ψ
2
ψ → e−i(|ψ| −1)dt ψ
enddo
and the actual code is not much more complicated than this. Aside from sim-
plicity, because at each sub-time-step the wave function is just multiplied by a
phase, the norm  
d3 x |ψ(x, t)|2 ∼ d3 k |ψ(k, t)|2

is conserved, so total probability is constant in time. Similarly, the energy

  
1 1 1 1
E = d3 x |∇ψ|2 + |ψ|4 − |ψ|2 +
2 4 2 4
is conserved; since the energy is proportional to the total length of vortex line,
there is no possibility of vortex stretching here.
The remaining ingredient is the initial wave function corresponding to the
particular vortex configuration of interest. Beginning with a background value
ψ = 1, a trivial solution of (10), a single line vortex along the z-axis is obtained
by multiplying it by the wave function ψ1 = f (r)eiθ discussed above. For a two-
vortex configuration, the latter is multiplied by a second factor ψ1 , where now

Fig. 4. Reconnection of anti-parallel vortex filaments in superfluid [1].

354 J. Koplik

r and θ are defined with respect to the new axis of the second filament [22].
This provides a reasonable approximation, provided the filaments are initially
separated by a distance much larger than the core size. Any error is taken care
of by the integration of the GP equation itself, provided the vortices have time
to relax to the correct form before they begin to interact with each other. For a
vortex ring [23], a similar procedure may be used, provided the radius of the ring
is much larger than the core size. At any point x we consider the plane passing
through the point and containing the axis of the ring. The ring then intersects
this plane at two points x± , once in the positive and once in the negative sense,
and we take ψ(x, t = 0) = ψ1 (x − x+ )ψ1∗ (x − x− ). The test of this procedure is
to observe the translation of a single ring along its center axis, and verify that
the observed velocity agrees with analytic calculations.
It is now straightforward to study reconnection. In Fig. 1, the initial con-
figuration has two filaments with axes at 90◦ , and an initial separation of 4 in
units of the core size. The filaments bend towards each other, merge, and then
split with an obvious topological rearrangement. Figure 4 shows the case most
relevant to reconnection in a vortex tangle, where the initial vortices have op-
posite orientation, and have a mild hairpin-like bend towards each other. In a
periodic box, only certain relative orientations are consistent with a two-vortex
state; we find that initial relative angles of 180◦ , 135◦ and 90◦ reconnect while
0◦ and 45◦ do not. Since it is only in the former cases that the vortices tend
to approach each other, unless there is a strong external flow in addition to the
mutual induction velocity, we conclude that intersecting vortices almost always
reconnect.
Note the similarity to the normal fluid reconnection simulation in Fig. 3a;
the main difference is the absence of the bridge connecting the withdrawing final
vortices. In a normal fluid it is possible to leave behind high-vorticity regions
after an interaction in which viscosity enters and invalidates Kelvin’s theorem,
but in a superfluid this would require a new segment of quantized filament, and
this is not possible. As for the reconnection event itself, the superfluid density
vanishes at the vortex core, so Kelvin’s theorem applies only to contours which
avoid the merger region and there is no contradiction. In contrast to the Navier-
Stokes simulations, nothing very interesting happens in the cores during merger
processes: the density contours just smoothly merge into each other, without
rapid variation in space or time. Although this calculation conserves energy
globally, the reconnection process can redistribute it, and in fact there is an
emission of radiation from the reconnection region [24]
Vortex ring interactions resemble the normal fluid case, but the simplicity of
the numerical methods allowed us to examine a number of cases with various
initial orientations for long times [23]. If we begin with two rings, moving along
their axes so as to intersect at 90◦ , the result is similar to that in Fig. 3b, again
without the bridge joining the two final rings. The key geometrical feature ob-
served in the normal fluid vortex ring merger is present – the fact that there
are two final rings which move off in a plane perpendicular to the plane of inci-
dence. In the GP calculations, the rings continue to move periodically through
 Vortex Reconnection in Normal and Superfluids 355

the simulation box, occasionally repeating the merger and dissociation process,
whereas the Navier-Stokes calculations are dissipative and awkward to follow
for long times, so the ultimate fate of the bridge is not known. If the initial
condition is varied, however, we can obtain other final states. In Fig. 5, the ini-
tial intersection angle is 120◦ , and the final state has four rings; the difference
evidently is related to the detailed shape of the intermediate, distorted single-
ring state, and its tendency to close on itself. Other less-symmetric initial states
where the initial rings intersect off-axis can give three final rings, or presumably
any number.

Fig. 5. Vortex ring merger and breakup in a superfluid [23].

356 J. Koplik

6 Conclusions
We have discussed vortex reconnection in a comparative way for normal and
superfluids. The dynamics at the scale of the vortex cores is rather different,
and apparently much smoother for superfluids, whereas the large scale behavior
of vortex filaments and rings is very much the same. Since both systems are
essentially Euler fluids outside a vortex core, the latter behavior is unsurprising,
and provides further evidence for the similarities between normal and superfluid
turbulence.
The superfluid reconnection calculations discussed here incorporated the min-
imal physics needed to address the phenomena intelligently – the simplest reason-
able model of the condensed state, zero temperature, and neglect of the normal
fluid and mutual friction. It is our belief that in a model more realistic than
GP one would see perhaps more structure in the core region, corresponding to
a richer excitation spectrum, but no significant change in general behavior or
vortex topology. The role of the GP model in this sense is to provide a phys-
ically reasonable smoothing of the core singularity present in an idealized line
vortex, and that more refinements would just affect the details. The similarities
between the superfluid and Navier-Stokes results discussed above already justify
this assertion.
Adding the normal fluid and finite temperature is a very interesting avenue
for further work at the atomic scale considered here, since issues such as thermal
nucleation and vortex coupling to normal fluid are likely to be qualitatively sig-
nificant. Similarly, the generation of vorticity due to interaction of the superfluid
with solid walls would be better understood if further microscopic information
were available. It would be a pleasure to hear these topics reviewed at the next
meeting on superfluid vortices.

References
1. J. Koplik and H. Levine: Phys. Rev. Lett. 71, 1375 (1993)
2. G. K. Batchelor: An Introduction to Fluid Dynamics (Cambridge University Press,
Cambridge 1967)
3. P. G. Saffman: Vortex Dynamics (Cambridge University Press, Cambridge 1992)
4. E. P. Gross: J. Math. Phys. 4, 195 (1963); L. P. Pitaevskii: Sov. Phys. JETP 13,
451 (1961)
5. R. J. Donnelly: Quantized Vortices in Helium II (Cambridge University Press,
Cambridge 1991)
6. A. Vincent and M. Meneguzzi: J. Fluid Mech. 225, 1 (1991)
7. S. Douady, Y. Couder and M. E. Brachet: Phys. Rev. Lett. 67, 983 (1991)
8. U. Frisch: Turbulence (Cambridge University Press, Cambridge 1995)
9. H. K. Moffatt: J. Fluid Mech. 409, 51 (2000), and earlier references therein.
10. K. W. Schwarz: Phys. Rev. B 38, 2398 (1988), and earlier references therein. For
later work, see the lecture by Tsubota in this volume, and its references.
11. S. Kida and M. Takaoka: Annu. Rev. Fluid Mech. 26, 169 (1994)
12. M. V. Melander and F. Hussain: Phys Fluids A 1, 633 (1989)
13. S. Kida, M. Takaoka and F. Hussain, J. Fluid Mech. 230, 583 (1991)
 Vortex Reconnection in Normal and Superfluids 357

14. T. Fohl and J. S. Turner: Phys Fluids 18, 433 (1975); Y. Oshima and N. Izutsu:
ibid. 31, 2401 (1988)
15. E. D. Siggia: Phys Fluids 28, 794 (1985)
16. O. N. Boratov, R. B. Pelz and N. J. Zabusky: Phys Fluids A 4, 581 (1992)
17. M. J. Shelley, D. I. Meiron and S. A. Orszag: J. Fluid Mech. 246, 613 (1993)
18. N. J. Zabusky and M. V. Melander: Physica D 37, 555 (1989)
19. H. Aref and I. Zawadski: Nature 354, 50 (1991)
20. D. Virk, F. Hussein and R. M. Kerr: J. Fluid Mech. 304, 47 (1995)
21. M. Taha and M. Ablowitz: J. Comput. Phys. 55, 203 (1984)
22. A. L. Fetter: Phys. Rev. 138, A429 (1965)
23. J. Koplik and H. Levine: Phys. Rev. Lett. 76, 4745 (1996)
24. M. Leadbeater, T. Winiecki, D. C. Samules, C. F. Barenghi and C. S. Adams,
cond-mat/0009060; see the paper by Adams in this volume.
Helicity in Hydro and MHD Reconnection

Axel Brandenburg1,2 and Robert M. Kerr3,4

1
Department of Mathematics, University of Newcastle upon Tyne, NE1 7RU, UK
2
NORDITA, Blegdamsvej 17, DK-2100 Copenhagen Ø, Denmark
3
NCAR, Boulder, CO 80307-3000, USA
4
Department of Atmospheric Sciences, University of Arizona, Tucson, AZ
85721-0081, USA

Abstract. Helicity, a measure of the linkage of flux lines, has subtle and largely un-
known effects upon dynamics. Both magnetic and hydrodynamic helicity are conserved
for ideal systems and could suppress nonlinear dynamics. What actually happens is not
clear because in a fully three-dimensional system there are additional channels whereby
intense, small-scale dynamics can occur. This contribution shows one magnetic and one
hydrodynamic case where for each the presence of helicity does not suppress small-scale
intense dynamics of the type that might lead to reconnection.

1 Introduction
The term reconnection is used in both the MHD and fluids communities to
describe topological changes in magnetic or vorticity fields due to resistivity or
viscosity and could not occur in ideal cases where these dissipative terms are zero.
In the strictly ideal limit the connectivity of the field lines would not change, but
this is a singular limit and even the smallest amount of resistivity or viscosity
allows the connectivity to change, albeit on small length scales. In the presence
of finite dissipative terms, large amounts of energy can be converted into heat
(or other forms of energy if one goes beyond the hydrodynamic approximations).
MHD reconnection plays a role in understanding why the solar corona (i.e. the
tenuous layers above the solar surface) are heated to ∼ 106 K, even though at
the surface of the sun the temperature is only ∼ 6000 K. The other aspect is
that the dissipative terms allow the field line connectivity to change. There are
strong indications from observations of the solar corona in X-rays that field loops
originally tied to the solar surface all of a sudden break loose and transport
large amounts of flux into outer space. This raises the issue of how fast can
reconnection occur. This is perhaps the single most challenging aspect of the
problem.
Early work on MHD reconnection was concerned with steady state config-
urations, allowing a constant flux of material to pass through an X-point type
configuration in two-dimensional field line configurations. However, because the
reconnection site becomes very thin as the magnetic resistivity decreases, the
amount of flux processed through the reconnection site decreases like the square
root of the resistivity and by this mechanism finite reconnection in a dynamical
timescale is not feasible for typical astrophysical values of resistivity. Other more

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 358–365, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Helicity in Hydro and MHD Reconnection 359

complicated initial conditions can lead to slow shocks that increase the recon-
nection rate, as discussed in a recent textbook [1]. Is not clear, however, whether
the various boundary conditions studied so far represent anything physical in
the corona and whether the results could explain the nanosecond timescales
over which hard X-ray output associated with reconnection is seen to rise.

2 Dissipation of Energy and Helicity

Magnetic reconnection has two distinct aspects. One is the speed at which mag-
netic energy can be converted into heat and the other is the speed at which
the magnetic topology can change. The two need not be the same. The perhaps
worst possible type of topology to change is one that invokes mutual linkage of
flux tubes, which can be described by the magnetic helicity H defined as

H = A · B dV, (1)

where B is the magnetic field and A is the vector potential such that B = ∇×A.
Obviously, A is not uniquely defined, because adding an arbitrary gradient field
to A would not change B. However, the value of H is unaffected by this if the
integral is taken over a domain where the normal component of the field vanishes
on the boundaries. In that case
  
(A + grad ϕ) · B dV = A · B dV + ϕ∇ · B dV = H, (2)

because the magnetic field is always solenoidal, ∇ · B = 0. Another conserved

quantity is the cross helicity, Hc = u · B dV , which describes the linkage be-
tween flux tubes and vortex
 tubes. In the absence of magnetic fields the hydro-
dynamic helicity, Hh = u · ω dV , is conserved by the inviscid Euler equations,
and it describes the linkage of vortex tubes with themselves.
The standard example that highlights the connection between magnetic he-
licity and topology is an interlocked pair of flux rings (see, e.g., the first panel of
Fig. 1), for which the magnetic helicity is given by twice the product of the two
magnetic fluxes of each of the two flux rings. However, helicity is also associated
with two orthogonal flux tubes as shown in Fig. 5.
The dramatic difference between the dissipation of magnetic energy and mag-
netic helicity can best be seen by contrasting the equations of the conservation
of magnetic energy and magnetic helicity,
d
1
2 B · B = −u · (J × B) − ηJ · J , (3)
dt

d
1
2 A · B = −u · (B × B) − ηJ · B, (4)
dt
where angular brackets denote volume averages, and surface terms are assumed
to vanish.
360 A. Brandenburg and R.M. Kerr

Fig. 1. Resistive evolution of an initially interlocked pair of flux rings. Isosurfaces of

the magnetic field are shown at different times.

The important point to note here is that the magnetic energy can maintain
a steady state where Joule dissipation, ηJ 2 , can be finite and large if work is
done against the Lorentz force, i.e. if −u · (J × B) > 0. At the same time,
however, there is no such term in the magnetic helicity equation, so in that case
a steady state is only possible if the current helicity, J · B, vanishes.
In the absence of any forcing, f , the hydrodynamic helicity is conserved
 in a
similar manner, but there are two important
 differences. First, because u·ω dV
contains one more derivative than u2 dV , it dissipates faster than the energy
if there is dissipation. Second, if there is forcing, the hydrodynamic helicity is no
longer conserved. This difference to the hydromagnetic case can best be seen by
contrasting eqs. (3) and (4) with the corresponding equations in hydrodynamics,
d
1
2 u · u = u · f  − νω · ω, (5)
dt
d
1
2 u · ω = ω · f  − νθ · ω, (6)
dt
where θ = ∇×ω is the curl of the vorticity, and surface terms are again assumed
to vanish. Thus, unlike the magnetic counterpart, kinetic helicity conservation is
only possible in the special case where the forcing is perpendicular to the vorticity
and the flow is inviscid. With dissipation and the absence of forcing both kinetic
energy and kinetic helicity are decaying, but kinetic helicity contains an extra
derivative more than the kinetic energy and so decays faster than energy and
does not pose a hard constraint.

3 Interlocked Flux Rings

In Fig. 1 we show an example of an initial flux tube configuration where, in our
case, each tube has the flux Φ = B · dS = 0.7B0 d2 , where B0 is the maximum
field strength in the core of each tube and d is its radius. The magnetic helicity
is measured to be H = A · B dV = 0.98B02 d4 , in perfect agreement with the
formula H = 2Φ2 .
The subsequent evolution of this flux tube configuration is governed by the
curvature force acting separately in each flux tube trying to make them contract.
Eventually the two tubes come into contact and produce an intense current sheet
where they touch. Figure 2 shows the peak current, J ∞ , plotted in two ways,
 Helicity in Hydro and MHD Reconnection 361

Fig. 2. Semi-logarithmic plot of J ∞ for a compressible 2403 calculation in a domain

of size 4 (dotted line: filtered, and solid line: unfiltered initial conditions) together with
fits to exponential growth and blow-up behavior, respectively. The blow-up scaling fits
better at later times.

one showing a period of singular growth and the other showing a period of
exponential growth. This initial period is represented in Fig. 1 by the first two
frames showing the approach and initial deformation of the linked flux tubes. At
the last time visualized, the surfaces in the outer region appear to merge into a
single continuous flux tube, while the inner region appears to be annihilated in
a complicated reconnected structure with writhe. Figure 3 takes another look at
this time using flux lines instead of surfaces. The flux lines in the outer region
that appeared to be continuous can now be seen to change direction abruptly
where they plunge into the inner region. And the inner region is now seen to
be continuously connected to the outer flux lines and instead of being a single
flux tubes with writhe, it now appears to be the original flux lines just twisted
around each other with almost no reconnection.
In the ideal case, the magnetic helicity is conserved for all time. Even in
the resistive case the magnetic helicity is very nearly constant. Furthermore, the
peak current increases to large values, which appear to be limited only by the
numerical resolution. This is related to the newly posed millennium question of
whether regularity of the Navier-Stokes equations can be shown [2]. A singularity
probably does not develop for the full viscous and resistive equations due to the
development of reconnection. However, singularities do seem possible for Euler
and ideal MHD. Numerical calculations have been used to provide insight into the
interaction of anti-parallel vortex tubes using the incompressible Euler equations
[3]. The key to providing useful results was the direct comparison with hard
analytic bounds for the maximum growth rate of the vorticity [4]. This initial
condition was very contrived with special symmetries and no helicity, unlike real
flows, and its generality remains uncertain.
A similar analytic bound has been shown to exist for the ideal MHD equa-
tions [5]. Therefore two intertwined questions have arisen. First, is there an ini-
tial condition for ideal MHD which might show similar singular growth? Second,
what role might the various types of helicity play in suppressing or enhancing
362 A. Brandenburg and R.M. Kerr

Fig. 3. Magnetic field lines (white) together with some field vectors (in grey) indicating
the field orientation for a resistive flux ring calculation at t = 4, i.e. shortly after the
time of the suspected singularity in the ideal calculation.

the reconnection rate? It has been found [6] that in the ideal case the two inter-
locked magnetic flux tubes go through a phase where behavior consistent with
a singularity seems plausible. This was surprising because as noted above this is
a nearly maximally helical initial condition and was expected to suppress non-
linearity. For hydrodynamics, it has also been claimed [7] that the helical initial
condition of two orthogonal vortex tubes showed signs of a singularity, although
the analytic test [4] was never applied. These two cases raise two possibilities.
In hydrodynamics there might exist a mechanism whereby helicity is shed per-
mitting stronger nonlinear growth, while in MHD the nonlinearity really comes
from J × B and J · B is not conserved, so there might in fact be no constraint
upon locally strong nonlinearity.
Let us consider an argument for why helicity might be required to allow, at
least for a period, nearly singular growth for ideal MHD. This is based upon old
arguments for why there could not be a singularity of Euler. It has been argued
that a singularity would not occur for Euler because what drives the growth
in the vorticity is the axial strain stretching the vorticity, and this strain must
grow at the same rate as the vorticity to sustain this growth. This could only
be achieved by an enormous growth in the curvature of vortex lines [8], which
in turn would require a delicate balance in the growth of the pressure Hessian.
This was originally thought not to be feasible, but newer analysis of the anti-
parallel Euler calculations [9,10] has shown that all of the analytic requirements
needed to achieve this delicate state are in fact obeyed. This is possible because
the vorticity and the strain are in fact just different manifestations of the same
vector field and can be strongly aligned.
While it has not been shown analytically, one would expect that for ideal
MHD to show similar singular growth, a similar delicate balance would have
to exist between the vorticity, the current, and the magnetic and velocity strain
fields. Current and magnetic strains are just different manifestations of the same
 Helicity in Hydro and MHD Reconnection 363

vector magnetic field, just as vorticity and strain are manifestations of the same
vector velocity field. However, one is still left with the current being completely
distinct from vorticity. Only if there is some property of the vector fields that
strongly couples these two fields could they act in concert to give a singularity.
Perhaps because helicity is conserved, for strongly helical structures there exists
such a constraint. Therefore only for strongly helical magnetic structures could
singular, or nearly singular, nonlinear growth occur.
New analysis has shown that the location of the peak in the current is at the
juncture between the outer flux lines and the inner flux lines where the maximum
in the curvature is located. This would be consistent with new mathematical
analysis by J. D. Gibbon (unpublished) that strong magnetic field line curvature
should be associated with any singular growth. The vorticity is also the strongest
in this region, suggesting the type of symbiotic growth of current and vorticity
that we believe is needed if there is to be a singularity of ideal MHD.
Figure 3 is resistive, but is very similar to visualizations of new ideal calcu-
lations that were run at higher resolution, up to the equivalent of 12963 mesh
points if a uniform mesh had been used. However, these new calculations, while
they do extend the period of seemingly singular growth, now appear to show
that the singular growth eventually is suppressed in the incompressible case.
The evidence relies on consistent behavior between the two highest resolution
calculations.
What might be the cause of this suppression? While it will take time to
fully understand these massive data sets, the initial indications are that it is
occurring as the peak in the current moves outside the inner region with its
strongly aligned vorticity, magnetic, and current fields. Our suspicion is that the
importance of the inner region is that, through twist, the location of most of
the initial helicity associated with the linked flux tubes is in the inner region.
If this can be shown, then it might tell us that the secret to maintaining nearly
singular growth is to maintain as high a level of local helicity for as long as
possible. This would be consistent with arguments [11] that fast reconnection is
associated with the entanglement of flux lines due to footpoint motion, which is
known to produce the required heating rates [12]. Helicity has also been shown
to play a role in coronal simulations of an arcade and a twisted flux loop [13],
with nearly singular growth in current similar to what we have observed.
There are other possible mechanisms that could suppress singular growth.
In a simulation [14] of nearly the same initial condition as the one used in our
original paper [6], there is only exponential growth that is associated with the
appearance of current sheets. More recent detailed analysis of our calculation
shows that the exponential growth is actually associated with the appearance
of two nearly overlapping orthogonal current sheets and the pressure barrier
between them that suppresses stretching terms and growth. This is an important
result because in some sense the more physical initial situation might be two flux
tubes that do not overlap at all. Our simulations that show stronger growth in
the current all have some overlapping between the initial flux tubes, something
364 A. Brandenburg and R.M. Kerr

Fig. 4. 1/ω∞ and 1/ ωi eij ωj in Euler for orthogonal vortices.

Fig. 5. Isosurfaces of vorticity as a fraction of the peak vorticity. The three frames are
t = 0, 6 and 10. Arms are pulled out of the original vortices, become anti-parallel, then
vorticity within the arms develops singular behavior.

that should not happen in an astrophysical situation where the flux tubes are
initially separated by large distances.

4 Orthogonal Vortex Tubes

We now turn to the case of straight tubes that are orthogonal to each other.
We note that also in this case there is finite helicity. The magnetic case has
been studied previously [15], but here we focus on the hydrodynamic case with
vortex tubes. Figure 4 shows the inverse of the peak vorticity and the inverse
of the enstrophy production rate for the orthogonal vortex tubes whose inviscid
evolution is shown in Fig. 5. Plotting these inverses was previously shown to
be the most effective way to highlight the 1/(tc − t) singular behavior. Figure 4
shows that the initial growth is weak, unlike the anti-parallel case. Then the
growth of peak vorticity, ωp , and enstrophy production, Ωpr , becomes stronger
with their inverses going roughly linearly to zero at the same singular time.
What is the configuration around the peak vorticity once singular growth
starts? And what role does helicity play? Analysis of the three-dimensional fields
shows that the peak vorticity is located in the arms that are being pulled off
 Helicity in Hydro and MHD Reconnection 365

of the two original orthogonal vortices. The last time shows that these isosur-
faces are parallel, and analysis shows that the vorticity within these surfaces
is anti-parallel. That is, to develop singular growth exactly the same alignment
of vorticity that was previously described as a contrived situation is actually
what the dynamics generate by themselves. This is consistent with vortex fil-
ament work [16]. In terms of helicity, locally around the anti-parallel vortices
there is no kinetic helicity density. Therefore in order for orthogonal vortices to
develop singular growth, the flow must realign itself to be non-helical, shedding
any helicity to achieve this.
In conclusion, these calculations have demonstrated that the role of helicity
can be rather complex. In the hydrodynamic case it appears that the absence
of helicity is required for there to be singular growth and in the MHD case
helicity seems to be required. The role of helicity upon reconnection should now
be investigated for these and similar configurations [15].

References
1. E. R. Priest and T. G. Forbes: Magnetic Reconnection. CUP (2000).
2. [Link] problems/[Link]
3. R. M. Kerr: Phys. Fluids 5, 1725 (1993).
4. J. T. Beale, T. Kato, A. Majda: Comm. Math. Phys. 94, 61 (1984).
5. R. E. Caflisch, I. Klapper, G. Steele: Comm. Math. Phys. 184, 443 (1997).
6. R.M. Kerr, A. Brandenburg: Phys. Rev Lett. 83, 1155 (1999).
7. O.N Boratav, R.B. Pelz, N.J. Zabusky: Phys. Fluids A 4, 581 (1992).
8. J.D. Gibbon, M. Heritage: Phys. Fluids 9, 901 (1997).
9. R.M. Kerr: ‘The role of singularities in turbulence’. In 19th ICTAM Kyoto ’96, ed.
by T. Tatsumi, E. Watanabe, T. Kambe (Elsevier Science 1997) pp. 57–70.
10. R.M. Kerr: ‘The outer regions in singular Euler’. In Fundamental problematic
issues in turbulence, ed. by A. Tsinober, A. Gyr (Birkhauser 1998) pp. 57–70.
11. E.N. Parker: ’Nanoflares and the solar X-ray corona’. Astrophys. J. 330, 474
(1988).
12. K. Galsgaard, Å. Nordlund: ‘Heating and activity of the solar corona: I. boundary
shearing of an initially homogeneous magnetic-field’. J. Geophys. Res. 101, 13445
(1996).
13. T. Amari, J.F. Luciani: ‘Helicity redistribution during relaxation of astrophysical
plasmas’. Phys. Rev. Lett. 84, 1196 (2000).
14. R. Grauer, C. Marliani: ‘Current-sheet formation in 3D ideal incompressible mag-
netohydrodynamics’. Phys. Rev. Lett. 84, 4850-4853 (2000).
15. R.B. Dahlburg, D. Norton: ‘Parallel computation of magnetic flux-tube reconnec-
tion’. In Small-scale structures in three-dimensional hydro and magnetohydrody-
namic turbulence, ed. by M. Meneguzzi, A. Pouquet, P.L. Sulem (Lecture notes in
physics 462 Springer-Verlag 1995) pp. 331–339.
16. A. Pumir, E. D. Siggia: ‘Vortex dynamics and the existence of solutions of the
Navier-Stokes equations’. Phys. Fluids. 30, 1606 (1987).
Tropicity and Complexity Measures
for Vortex Tangles

Renzo L. Ricca
1
Mathematics Department, University College London
Gower Street, London WC1E 6BT, UK
2
Isaac Newton Institute for Mathematical Sciences
20 Clarkson Road, Cambridge CB3 0EH, UK
ricca@[Link]

Abstract. In this paper we introduce and discuss new concepts useful to analyse and
characterize patterns of vortex lines in fluid flows. We define measures of tropicity
to identify ‘tubeness’, ‘sheetness’ and ‘bulkiness’ of vortex lines and to measure the
spreading of field lines about preferred directions. Algebraic, geometric and topological
measures based on crossing number information are discussed and are put in relation
to the kinetic helicity and the energy of the fluid system.

1 Vortex Structures and Tangles

in Classical and Quantized Vortex Flows
Coherent structures represent an essential feature of classical turbulent flows [2],
[5]. Experimental and numerical results have indeed shown that vorticity has
a tendency to coalesce into highly localized regions. As modern visualizations
of classical and superfluid flows show [16], [14], [15], [17], strong anisotropies
emerge as vortical flows re-organize themselves to form tubes, sheets or quantized
complex tangles of vortex lines. As vortex structures evolve, different types of
non-linear effects and instabilities take place, until continuous break-up and re-
structuring are overcome by total dissipation.
The tremendous progress in visualization techniques and real-time diagnos-
tics of complex flow patterns [18] makes it now possible accurate recognition of
vortex pattern formation and interaction; detailed mechanisms of braiding, link-
ing and re-structuring of vortex lines can be analysed by real-time simulations
to a high degree of accuracy. In this context geometric and topological informa-
tion is available from data-sets of numerical simulations; properly analysed, it
provides valuable help to understand fundamental properties of dynamics and
energetics of turbulent flows [13], [11]. New concepts and tools based on geometry
and topology are therefore being developed [12] to quantify physical information
associated with structural complexity of such flows. In the following sections we
shall introduce measures based on geometric and topological concepts that will
provide useful tools to quantify structural complexity of vortical flows and help
to develop new measures to estimate physical properties.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 366–372, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Tropicity and Complexity Measures for Vortex Tangles 367

2 Measures of Tropicity for Vortex Tangles:

Tubeness, Sheetness and Bulkiness
A first step in the application of measures of structural complexity is to develop
tools for pattern recognition of vortex structures. These must be based on es-
timates of anisotropy and spatial extension of vortex lines in the fluid. Let us
E a fluid region D in IR and a generic tangle T of n vortex lines L, i.e.
3
consider
T = n Ln in D. A vortex line is given by a line of vorticity, whose support is
identified with a smooth, simple space curve, not necessarily closed in D (D can
be a sub-region of the entire fluid domain), with vorticity everywhere tangent to
the curve. It is useful to introduce the concept of ‘tropicity’ (a measure of the
space configuration of the vortex region) to characterize the degree of tubeness,
sheetness and bulkiness of T in D. Consider a single vortex line L as frozen
in space and time. A measure of the spatial extent of L in D is given by the
maximal distance D1 between two points Pi and Pj on L, i.e.

D1 = max d(Pi , Pj ) ≡ P0 P1 = X 2 + Y 2 + Z 2 , (1)
i,j

where P0 = (x0 , y0 , z0 ) ∈ L, P1 = (x1 , y1 , z1 ) ∈ L and X = x1 − x0 , Y = y1 − y0 ,

Z = z1 −z0 . The unit vector T̂1 = (P1 −P0 )/D1 (first directional tropicity vector )
given by the director cosines X/D1 , Y /D1 , Z/D1 , is the principal directional
axis of L. Let us consider then the transversal spatial extension of L. Take the
maximal distance of a third point Pi ∈ L (not aligned with P0 and P1 ) from the
line l(P0 , P1 ): a measure of maximal width spanned by L in D is given by

D2 = max d(Pi , l(P0 , P1 )) ≡ P2 O , (2)

i

where P2 ∈ L and O ∈ l(P0 , P1 ), where O, being not necessarily a point on

L, is the footpoint of the orthogonal projection of P2 onto P0 P1 . The second
directional tropicity vector is the unit vector T̂2 = (P2 −O)/D2 , which prescribes
direction and orientation of the second principal directional axis of L in D.
We can now define the plane Π = Π(P0 , P1 , P2 ) and, by taking the maximal
distance of a fourth point Pi ∈ L from Π, we have a measure of the maximal
three-dimensional extension of L in D, i.e.

D3 = max d(Pi , π(P0 , P1 , P2 )) ≡ P3 Q , (3)

i

(P3 ∈ L) taken along the direction of the third directional tropicity vector T̂3 =
T̂1 × T̂2 .
Let us consider the whole tangle of vortex lines T . The spatial extension of
T in D is measured by the three maximal distances Di , this time taken with
respect to the principal axes λi (along T̂i , i = 1, 2, 3) determined by sampling
Pi over the whole tangle T . The tropicity volume is given by V (D) = D1 D2 D3 ,
and measures of tropicity can be defined by taking the relative ratio of these
quantities. We have
368 R.L. Ricca

def D1
• if D3 = O(D2 ) and D2  D1 : tubeness = Λ1 ≡ ;
D2
def D2
• if D3  D2 and D2 = O(D1 ): sheetness = Λ2 ≡ ;
D3
def D1 D2 D3
• if D3 = O(D2 ) and D2 = O(D1 ): bulkiness = Λ3 ≡ .
D33
These quantities provide a first crude information of the space configuration
of the pattern. For classical turbulent flows these quantities are particularly use-
ful to detect regions of high tubeness and sheetness, whereas for quantized flows
bulkiness may provide more useful information related to vortex line density.
Moreover, as we shall see below, information on directional tropicity finds useful
applications for geometric and topological estimates of structural complexity.

3 Measures of Geometric Complexity:

Directional Alignment and Writhing
In classical fluids strong anisotropy is characterized by the presence of elongated,
tubular regions. If ω = ∇ × u denotes vorticity (u = u(x) being the velocity
field function of the position vector x), and σ vortex stretching, defined by
 
1 ∂ui ∂uj
σi = Si,j ωj = + ωj , (4)
2 ∂xj ∂xi
then a measure of how vorticity and vortex stretching lines are spread about the
principal directional axis T̂1 is given by the angles ηω and ησ , defined by

|ω̄ × T̂1 | |σ̄ × T̂1 |

tan ηω = , tan ησ = , (5)
ω̄ · T̂1 σ̄ · T̂1
where ω̄ = ωD and σ̄ = σD denote space averages of the field lines over
the tropicity domain D. Similarly, we can define the spread of vorticity and
stretching (through the angles χω and χσ ) about a sheet-like distribution by
taking
ω̄ · T̂3 σ̄ · T̂3
cot χω = , cot χσ = . (6)
|ω̄ × T̂3 | |σ̄ × T̂3 |
For quantized vortex tangles directional writhing provides a measure of aver-
age coiling of vortex lines in space. Let Πi = Π(T̂i ) denote a plane of projection
with normal T̂i and consider the projection of the tangle T onto the plane Πi .
A ‘good’ diagram Ti = T (T̂i ) (which of course depends on the line of projection
T̂i ) is given by a curve graph, whose self-intersections are given by countably
many points, where the segments intersect transversally. Good projections can
be found by an appropriate choice of the projection plane. By keeping track of
the orientation of the curve (induced by the vorticity vector), we obtain an ori-
ented diagram (see Fig. 1) and by assigning the value r = ±1 to each projected
 Tropicity and Complexity Measures for Vortex Tangles 369

Fig. 1. Example of an oriented space curve projected onto the plane Πi .

crossing r (according to standard convention on signs [6]), we can quantify the

directional writhing in terms of algebraic sum of positive and negative crossings
on Πi (see [10])
 +
1 dX × dX ∗ · (X − X ∗ )
W r(Ti ) = ∗
= r . (7)
4π Ti |X − X | 3
r∈Ti

The writhing number W r(T ) is given by averaging the directional writhe over
the whole solid angle. It may be computationally convenient to approximate this
quantity by taking the algebraic mean over the three principal orthogonal planes
as reference projection planes; hence, the estimated writhing number given by
9 :
+
W r⊥ = r ≈ W r(T ) , (8)
r∈Ti
⊥

will provide a reasonable, simple measure of the average coiling of T .

4 Algebraic Measure of Structural Complexity:

Average Crossing Number
Structural complexity can be measured by counting the total number of cross-
ings, that are present in a tangle at a given time. This quantity, which is as-
sociated with the un-oriented tangle diagram, is given by the average crossing
number C̄ by counting the total number of un-signed crossings in Ti and aver-
aging over the whole domain D [3]. We have
9 :
+
C̄(T ) = |r | . (9)
r∈Ti
D
370 R.L. Ricca

Once again, it is computationally convenient to approximate this measure by

the algebraic mean taken over the three principal orthogonal planes, hence
9 :
+
C̄⊥ = |r | ≈ C̄(T ) . (10)
r∈Ti
⊥

Current work done by Barenghi et al., [1] shows that C̄⊥ provides indeed a
good approximation to C̄ and it seems to be very effective to detect structural
complexity.

5 Measures of Topological Entanglement:

Kinetic Helicity and Directional Linking
Topological entanglement can be calculated on the basis of information about
directional tropicity of T . In this context the concept of linking number is fun-
damental, since it is related to the kinetic helicity H of the flow (see [7])

H(T ) = u · ω dV . (11)
D
For superfluid vortices helicity can be estimated through linking number mea-
sures made directly on the vortex tangle, hence avoiding the difficulties associ-
ated with the integration over the vorticity field. These are based on the concept
of (Gauss) linking number Lkij = Lk(Li , Lj ) of two loops Li and Lj , given by
 
1 dX i × dX j · (X i − X j )
Lkij = , (12)
4π Li Lj |X i − X j |3
and its limit form Lkii , given by the the Călugăreanu-White formula
Lkii = W ri + T wi , (13)
where W ri = W r(Li ) is the writhing number of Li and T wi = Ti + Ni is the
total twist, sum of the total torsion (Ti ) and intrinsic twist (Ni ) of Li divided
by 2π (see [10] for precise definitions and physical meaning of these quantities).
For a tangle of n vortex lines, each of circulation κi (i ∈ [1, . . . , n]), the kinetic
helicity is given by (see [11])
+ +
H(T ) = 2 Lkij κi κj + Lkii κ2i . (14)
i=j i

Note that in the case of superfluid vortices, since quantized vortex lines have no
internal structure we can assume T wi = Ti .
In the case of strong anisotropy of vortex flows directional tropicity provides
natural reference directions to measure localized winding of vortex lines. A rela-
tive measure of winding is given by the directional linking number kλi of vortex
lines with respect to one of the principal tropicity axes λi , i.e.
kλi = k(Li , λi )D , (15)
keeping λi fixed in the average process.
 Tropicity and Complexity Measures for Vortex Tangles 371

6 Relationships Between Complexity Measures

and Energy Levels
It is of fundamental importance to relate measures of algebraic, geometric and
topological complexity to physical properties of the system, such as kinetic he-
licity and energy. For this purpose it is convenient to re-write eq. (14) in a more
compact form. Consider the linking numbers Lij (i, j ∈ [1, . . . , n]) as elements of
a square matrix; since Lij = Lji , we can reduce the linking matrix to diagonal
form, i.e.
⎛ ⎞ ⎛ ⎞
L11 L12 . . . L1n M11 0 . . . 0
⎜ L21 L22 . . . L2n ⎟ ⎜ 0 M22 . . . 0 ⎟
⎜ ⎟ ⎜ ⎟
⎝ ... ... ... ... ⎠ → ⎝ ... ... ... ... ⎠ , (16)
Ln1 Ln2 . . . Lnn 0 0 . . . Mnn

where each element Mii takes into account self- and mutual linking of the vortex
lines. We can therefore re-cast eq. (14) in the form
+
H(T ) = Mii f (κi ) , (17)
i=1,...,n

where f (·) is a linear function of quadratic terms in κi .

If the tangle is made on average of vortex filaments of same length L (obtained
by an average measure over the tropicity domain D), we can show [9] that on
dimensional grounds the enstrophy Ω of the system is given by a relationship of
the form 
1 +
Ω(T ) = |ω|2 dV = Mii f (κi ) , (18)
D L i=1,...,n

that provides an interesting connection with helicity. Moreover, since the mag-
netic energy of a perfectly conducting magnetized fluid is bounded from below
by the magnetic helicity Hm [8], according to the inequality

Emin ≥ q0 |Hm | , (19)

where q0 is a positive constant, then we can expect that in steady state conditions
similar bounds hold for minimum enstrophy levels or for other types of ground
state energy in relation to the complexity of the physical system.

Acknowledgements
Financial support from UK PPARC (Grant GR/L63143) and EPSRC (Grant
GR/K99015) is kindly acknowledged.
372 R.L. Ricca

References
1. Barenghi, C.F., Ricca, R.L. & Samuels, D.C. (2000) How tangled is a tangle?
Submitted.
2. Betchov, R. (1956) An inequality concerning the production of vorticity in
isotropic turbulence. J. Fluid Mech. 1, 497–504.
3. Freedman, M.H. & He, Z.-X. (1991) Divergence-free fields: energy and asymptotic
crossing number. Ann. Math. 134, 189–229.
4. Galanti, B., Gibbon, J.D. & Heritage, M. (1997) Vorticity alignment results for
the three-dimensional Euler and Navier-Stokes equations. Nonlinearity 10, 1675–
1694.
5. Hussain, A.K.M.F. (1986) Coherent structures and turbulence. J. Fluid Mech.
173, 303–356.
6. Kauffman, L.H. (1987) On Knots. Annals Study 115, Princeton University Press.
7. Moffatt, H.K. (1969) The degree of knottedness of tangled vortex lines. J. Fluid
Mech. 35, 117–129.
8. Moffatt, H.K. (1991) Relaxation under topological constraints. In Topological As-
pects of the Dynamics of Fluids and Plasmas (ed. H.K. Moffatt et al.), pp. 3–28.
NATO ASI Series E: Applied Sciences 218. Kluwer, Dordrecht.
9. Moffatt, H.K. & Ricca, R.L. (1991) Interpretation of invariants of the Betchov-Da
Rios equations and of the Euler equations. In The Global Geometry of Turbulence
(ed. J. Jiménez), pp. 257–264. Plenum Press, New York.
10. Moffatt, H.K. & Ricca, R.L. (1992) Helicity and the Călugăreanu invariant. Proc.
R. Soc. Lond. A 439, 411–429.
11. Ricca, R.L. (1998) Applications of knot theory in fluid mechanics. In Knot Theory
(ed. V.F.R. Jones et al.), pp. 321–346. Banach Center Publications, Institute of
Mathematics 42, Polish Academy of Sciences, Warsaw.
12. Ricca, R.L. (2000) Towards a complexity measure theory for vortex tangles. In
Knots in Hellas ‘98 (Ed. McA. Gordon et al.), pp. 361–379. Series on Knots and
Everything 24, World Scientific, Singapore.
13. Ricca, R.L. & Berger, M.A. (1996) Topological ideas and fluid mechanics. Phys.
Today 49 (12), 24–30.
14. Schwarz, K.W. (1988) Three-dimensional vortex dynamics in superfluid 4 He. Phys.
Rev. B 38, 2398–2417.
15. She, Z.-S., Jackson, E. & Orszag, S.A. (1990) Intermittent vortex structures in
homogeneous isotropic turbulence. Nature 344, 226–228.
16. Swanson, C.E. & Donnely, R.J. (1985) Vortex dynamics and scaling in turbulent
counterflowing helium II. J. Low Temp. Phys. 61, 363–399.
17. Vincent, A. & Meneguzzi, M. (1991) The spatial structure and statistical proper-
ties of homogeneous turbulence. J. Fluid Mech. 225, 1–20.
18. Zabusky, N.J., Silver, D. & Pelz, R. (1993) Visiometrics, juxtaposition and mod-
eling. Phys. Today 46, 24–31.
The Geometry
of Magnetic and Vortex Reconnection

Gunnar Hornig

Ruhr-Universität Bochum, 44780 Bochum, Germany

Abstract. Reconnection is an important process of structure formation in fluid dy-

namics, occurring in the form of vortex reconnection in hydrodynamics as well as in
the form of magnetic reconnection in plasmas. There is a close analogy between the
quantities involved in both phenomena but, surprisingly, the process of magnetic re-
connection, although complicated by the presence of a magnetic field, is geometrically
simpler than vortex reconnection; it may thus serve as good starting point to under-
stand the geometry of vortex reconnection.
A general covariant definition of reconnection is given and, starting from a simple
analytic model of magnetic reconnection, the basic process of reconnection is analyzed.
The model is then modified to meet the additional constraints of vortex reconnection.
It is shown that, although the evolution of the vorticity near the reconnection site is
stationary and two-dimensional the flow velocity is inevitably three-dimensional, and
time dependent. Explicit expressions for the reconnected flux and the reconnection
time are given.

1 Introduction
The notion of reconnection is found in many fields of physics: in hydrodynamics
[1] for the reconnection of vortex tubes, in plasma physics for magnetic reconnec-
tion [2], in the theory of superfluids [3] for the reconnection of quantized vortex
elements, as well as in cosmology for the interaction of cosmic strings [4]1 . In
this contribution we will consider only reconnection as a process in the evolu-
tion of a divergence-free vector field, therefore involving the first two examples.
The divergence-free field is the common ingredient to both vortex reconnection
in hydrodynamics (HD) and magnetic reconnection in magnetohydrodynamics
(MHD). In HD the evolution of the vorticity field is determined by the curl of
the Navier-Stokes equation, while the evolution of the magnetic field in MHD is
determined by the curl of Ohm’s law. Both equations have the same structure:

∂t G −∇ × (v × G) = ∇ × N with: (1)
v : the velocity of the fluid flow
G : the transported field; G = ∇ × v in HD, G = B in MHD;
N : The non-ideal term
 
N = ∂k ν(∂k v i + ∂i v k ) ei (Viscous, incompressible HD);
N = η∇ × B (Resistive MHD) ,
1
The references given here are only examples that provide an entry to the literature

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 373–380, 2001.

c Springer-Verlag Berlin Heidelberg 2001
374 G. Hornig

where ∂i denotes the derivative with respect to the coordinate xi , ei is the

corresponding unit vector and summation of repeated indices is assumed.
Because only the evolution of the divergence-free field is relevant for the
definition of reconnection the other equations of hydrodynamics (HD) and mag-
netohydrodynamics (MHD), i.e. the equations of continuity, energy and the mo-
mentum balance in MHD are not considered here.
The main difference between HD and MHD is that in HD the transporting
field, v, and the field which is transported, w, are coupled by w = ∇ × v. A
corresponding equation in MHD (B = ∇ × v) does not exist. This extra freedom
allows for more simple reconnection solutions in MHD compared to HD. It is
therefore natural to start with an example of magnetic reconnection

2 Magnetic Reconnection
Note first that for vanishing resistivity equation (1) is ideal, i.e. the magnetic
field is frozen into the fluid flow and the magnetic flux integrated over a comoving
surface is conserved. In this case the topology of the field (B) is conserved as
well and no reconnection is possible.

∂t B − ∇ × (v × B) = 0 , (2)

⇒ B · n da = const.

Only if the non-idealness N is non-vanishing is reconnection possible. Con-

sider for instance a configuration with an initially anti-parallel two-dimensional
magnetic field as shown in Fig. 1. A localized non-idealness N perpendicular to
the plane in which the magnetic field lies can produce a curl which reduces the
initial anti-parallel field components according to (1) and adds a new perpen-
dicular component, such that the field lines are reconnected. The non-idealness
N or more exact ∇ × N plays an important role in the process. ∇ × N has
to be present to allow for reconnection but it also has to be localized, that is
there has to be an outside region where ∇ × N is negligible, at least on the
time scales under consideration. Otherwise we would have a situation of global
diffusion of the magnetic field and the term ‘reconnection’ would be meaningless

⇒ ⇒
Fig. 1. Basic reconnection process with a localized non-idealness N perpendicular to
the plane shown.
 The Geometry of Magnetic and Vortex Reconnection 375

because there is no way of identifying field lines or flux tubes in time. Note that
the localization of ∇ × N is not as restrictive as it may look at first sight. It is
not necessary that N is localized, only the curl of N has to be localized. The
localization is usually not a serious problem in MHD because in the most of the
applications of magnetic reconnection the resistivity is very low and enhanced
only in thin current sheets. In contrast in HD the viscosity is usually assumed
to be constant and the localization of ∇ × N in HD is, therefore, a result of the
structure of vortex sheets in most cases.
Note that ‘localization’ also means that there is no closed flux within the
region where ∇ × N is not negligble, otherwise the same argument as above
applies to this closed flux. But if all flux of the non-ideal region is connected to
the ‘ideal’ (∇ × N = 0) surrounding, then the conservation of flux in the outside
region implies a conservation of flux in the non-ideal region as well. Therefore,
the effect of a localized N is only to dissipate energy and rearrange the flux but
not to disspate the flux (at least on the time scale on which the reconnection
acts). Hence it should be possible to describe the process as being ‘ideal’, i.e.
satisfying (2) but with a new transport velocity u which may have a singularity.
For example

B = [y, kx, 0], u = [−x k Ez /(k 2 x2 + y 2 ), y Ez /(k 2 x2 + y 2 ), 0] (3)

is a stationary solution (Fig. 2) of (2), which has a singularity at the origin

L
E

a) B b)
Fig. 2. (a) The flow structure and (b) the evolution of cross-sections of the magnetic
flux under the singular flow u for a stationary reconnection process

of the coordinate system such that it transports the magnetic flux in a finite
time along the x-z plane onto the z-axis where the flux is split and reconnected
(Fig. 2b). The flux is then transported outwards along both directions of the z-y
plane. The example can be derived for a magnetic field B satisfying

E + v × B = η ∇ × B with
v = [−x, y, 0], E = [0, 0, Ez ], η = (Ez − kx2 − y 2 )/(k − 1) (4)
376 G. Hornig

The solution holds only in the region where η > 0. Since both E and J are
perpendicular to B the transport velocity of the magnetic flux is simply
u = (E − ηJ ) × B/B 2 .
Note that u is not the fluid velocity, but where η vanishes u and v coincide.
Now (2) is only a special case of an equation which leads to reconnective
solutions. Since the transport velocity has to become infinite at the reconnection
line (the line where u is singular), an adequate description is a covariant one
and the most general form, which leads to an equation like (2) (see [5],[6]) is
αβγδ ∂α U ν Fνβ = 0 (5)

∂0 (U 0 E + U × B) + ∇(E ·U ) = 0
⇔
U 0 ∂0 B − ∇ × (U × B) − ∇U 0 × E = 0
⇔ LU ωF2 = 0 (6)
where Fαβ denotes the electromagnetic field tensor. The four-velocity

U (4) = (U 0 , U 1 , U 2 , U 3 ) = (U 0 , U )
is related to the usual velocity in space by
U dX/ds
u := 0
= , (7)
U dX 0 /ds
and LU ωF2 = 0 is the Lie-derivative of the 2-form of the electromagnetic field
with respect to the flow U (4) . The Lie-derivative theorem then implies that the
electromagnetic flux integrated over a two-dimensional comoving surface (C)
is constant with respect to the parameter s which describes the U (4) -flow in
space-time.  
(6) ⇒ F dA = Fμν dxμ dxν = const.
C C
The system (5) can be derived from

E ·U = −∂0 Φ
U ν Fνβ = ∂β Φ ⇔ (8)
U 0 E + U × B = ∇Φ .
These equations can be considered as the most general form of an Ohm’s law
which lead to a conservation of electromagnetic flux in the form of (5). The po-
tential Φ is not relevant for the two-dimensional reconnection example considered
above because, for this case, E · B vanishes and U (4) can be chosen such that
Φ = 0 holds everywhere. It is, however, important for the case of reconnection
in a non-vanishing magnetic field [7], which requires E · B = 0 and hence Φ = 0.
An additional advantage of the covariant formulation is that now the singu-
larity in u can be represented as a null of U (4) and is thus open to an analysis
of its structure. For instance, for the example given above the corresponding
4-velocity is given by
U (4) = [(k 2 x2 + y 2 ), −kEz x, Ez y, 0] . (9)
 The Geometry of Magnetic and Vortex Reconnection 377

This solution is typical for reconnection because U 0 > 0 requires that U 0 vanishes
quadratically at the null of U (4) , while the space components are antisymmetric
and thus in lowest order are linear in x and y near the null. The existence of
a null of U (4) with an X-point structure in the space components can be used
for a definition of magnetic reconnection [6]. In general the reconnection process
requires a line in space along which U (4) vanishes, the reconnection line, denoted
by L in Fig. 2 (b). For the example given this line is the z-axis, but in general
the line can be curved and moving, and coincides with a field line of B as long
as its velocity is small. In this case the total reconnected flux is given by the
integral along the part of reconnection line where U (4) vanishes and over the
time interval when this occurs, one has

Φrec = E · dl dt .

Correspondingly E · dl is the reconnection rate for a stationary process.
It is worth mentioning that a uni-directional magnetic field (B = bz (x, y)ez )
can also show an evolution, which requires a non-continuous u. An example is
where two initially isolated regions of positive bz in an environment of negative
bz merge. We do not consider this type of ‘reconnection’ here since it does not
require a ‘cut and paste’ of magnetic flux and, correspondingly, the reconnected
flux as defined above vanishes in these cases.

3 Vortex Reconnection

As mentioned in the introduction there is no one-to-one correspondence between

solutions of magnetic reconnection and vortex reconnection since the latter have
to satisfy w = ∇ × v. Moreover, apart from solutions with constant viscosity,
the non-ideal term N is more complicated in HD.
We start by constructing a stationary two-dimensional solution for vortex
reconnection, that is a solution where all quantities depend only on two space
coordinates (x, y) and the vorticity vectors lie in the x-y-plane. Using the analogy
with magnetic reconnection we assume that the vorticity has the form w =
[y, kx, 0] as in example (3). However, the flow velocity v = [−x, y, 0], which we
used in this example, does not satisfy w = ∇ × v. It has to be modified with an
additional z-component vz (x, y), such that

w = ∇vz (x, y) × ez ,

which is inevitable if v is not to depend on z.

Moreover, from magnetic reconnection we know that an electric field along
the reconnection line (the z-axis in our example) has to be present for a non-
vanishing reconnection rate. The analogue of the electromagnetic field tensor in
hydrodynamics is the vorticity tensor

Wμν = ∂μ Vν − ∂ν Vμ μ, ν ∈ {0, 1, 2, 3}
378 G. Hornig

The ‘electric’ field component of Wμν is therefore −∂t V − ∇V 0 . Because, V (4)

or v, respectively, is a real fluid velocity, which we assume to be non-relativistic,
we have V 0 = 1 and V = v. Thus the ‘electric’ field is given by −∂t v. For a
stationary reconnection process this field has to be constant, hence vz ∼ t in our
model.
Taking these conditions into account a full solution is given by

v = [−x, y, y 2 /2 − kx2 /2 − tEz ] (10a)

ν = Ez /(k − 1) − k/(3 k − 1) x2 − 1/(k − 3) y 2 (10b)
p = p0 + (k + 1)/(6 k − 2) x2 − (k + 1)/(2 k − 6) y 2 . (10c)

This solution satisfies (1) with a non-ideal term for viscous, incompressible HD,
as given in (1), for a region near the reconnection axis where p > 0, ν > 0 and
k > 3.
Before discussing the properties of this solution we remark that concerning
the covariant transport (5) of the vorticity tensor, the magnetic reconnection
in example (3) and the above example of a vorticity reconnection are identical.
They both satisfy (8) for Φ = 0 and U (4) given by (9).

− U 0 ∂t v + U × w = 0 (11)
U 0E + U × B = 0 (12)

The solution (10a) is only a solution for the neighborhood of the reconnection
line due to the simple polynomial assumption for the fields. Therefore, it cannot
reproduce all the complex features of vortex reconnection such as ‘bridging’, sec-
ondary reconnection, etc. This simple system, however, already contains several
basic properties of vortex reconnection. First, although we started with the most
elementary reconnection process, with respect to the fluid velocity v the process
is not stationary any more, since the vz component grows linearly with time.
Thus there is no true stationary vortex reconnection. Moreover, the existence
of a non-vanishing vz component breaks the two-dimensionality present in the
vorticity field. This complicates the flow structure of the process enormously.
But despite this complexity in the flow velocity, the reconnection process of the
vorticity is comparatively simple.
Because we know the transport velocity of the vorticity we can calculate the
time necessary for two flux tubes to reconnect in our example. This merging time
is given by the time which is required for a point (x0 ) on the x-axis to reach the
z-axis under the flow u. Hence
 0  0  0
dt = U /U dx = −
0 x
kx/Ez dx = x20 k/(2Ez ) .
x0 x0 x0

This result is in accordance with numerical results [8], which show a quadratic
dependence of the reconnection time on the distance between the flux tubes.
Note that this is not an accidental feature of our model, but is due to the
generic structure of reconnecting flows in space-time as explained above.
 The Geometry of Magnetic and Vortex Reconnection 379

In analogy to magnetic reconnection, the reconnected flux is given by the

integral along the reconnection line. The reconnected flux and the reconnection
rate are therefore,
 
dΦrec
Φrec = − ∂t v · dl dt = − ∂t v · dl . (13)
dt

4 Conclusions
It has been shown that the process of reconnection of a divergence-free field is
basically a flux conserving process and thus can be represented by an ideal ad-
vection of the flux under a velocity field u which has a singularity. A transition
to the covariant equations in space-time allows one to resolve this singularity,
which was shown to be a null point of the corresponding 4-velocity U (4) . Within
this concept we strictly distinguish between the transport velocity of the flux, u
or U (4) in space-time, respectively, and the fluid flow (v). Starting from an ele-
mentary two-dimensional stationary model for magnetic reconnection, a similar
solution for vortex reconnection can be constructed. This solution is, however,
stationary and two-dimensional only with respect to the velocity U (4) , whereas
the construction of the corresponding flow field v requires a three-dimensional
time-dependent solution. Thus, with respect to v, vortex reconnection is inher-
ently three-dimensional and time-dependent, even in its most elementary solu-
tion. It also shows that care has to be taken if passive tracers are used to follow
vortex tubes because, within the region where the non-idealness is relevant, the
vortex velocity u and the fluid velocity v differ significantly.
The covariant formulation also shows that the analogous quantity to the
electric field in MHD is −∂t v in HD and thus the reconnection rate, which is
known from magnetic reconnection to be  the integral of the electric field parallel
to the reconnection line, is given by − ∂t v · dl.

Acknowledgment
This work was supported by the Volkswagen Foundation.

References
1. S. Kida, M. Takaoka, Vortex Reconnection, Annu. Rev. Fluid Mech.26, 169 (1994)
2. D. Biskamp, Magnetic Reconnection, Physics Reports 237, 179 (1994)
3. J. Koplik, H. Levine, Vortex reconnection in superfluid helium, Phys. Rev. Letters
71, 1375, (1993)
4. E.P.S. Shellard, Cosmic String interactions, Nucl. Phys. B 282, 624 (1987)
5. G. Hornig, The covariant transport of electromagnetic fields and its relation to
magnetohydrodynamics, Phys. Plasmas, 4, 646, 1997.
6. G. Hornig, ‘The Evolution of Magnetic Helicity Under Reconnection’. In: Magnetic
Helicity, [Link] A.A. Pevtsov, (American Geophysical Union, Washington 1999), pp.
157-165
380 G. Hornig

7. G. Hornig, L. Rastätter, The magnetic structure of B = 0 reconnection, Physica

Scripta, T 74, 34, (1997)
8. A.T.A.M. de Waele, R.G.K.M. Aarts, Route to vortex reconnection, Phys. Rev.
Lett., 72, 482 (1994)
Current-Sheet Formation
near a Hyperbolic Magnetic Neutral Line

Bhimsen K. Shivamoggi

Institute of Theoretical Physics, University of California,

Santa Barbara, CA 93106-4030, USA, and
University of Central Florida, Orlando, FL 32816, USA

Abstract. Two-dimensional flow of an incompressible plasma in a hyperbolic mag-

netic field is discussed. The effects of sweeping [7] as well as shearing [11] of the magnetic
field lines by the plasma flow are considered. Exact solutions describing current-sheet
evolution in this setting are given which compare very well with laboratory experiments
([3], [8]-[10]).

1 Introduction
When a plasma collapses near the neutral line of the applied magnetic field, a thin
neutral current sheet is formed there. The magnetic flux continually accumulates
in the region of the neutral sheet and causes the total current and the sheet width
to increase. Thus, as Syrovatskii [1] pointed out, a stationary state is not really
possible for a current sheet. Nonetheless, laboratory experiments (Bratenahl et
al. [2], Frank [3], Gekelman et al. [4]) have shown that a quasi-stationary state is
still possible for a few Alfvén times as the current sheet collapse is impeded by
a pressure build-up. However, a non-stationary state ensues eventually! From a
theoretical point of view, the development of a current sheet in the vicinity of a
magnetic neutral line is an essentially nonlinear phenomenon. An exact solution
for the MHD1 equations for a time-dependent, two-dimensional flow of a plasma
in a hyperbolic magnetic field was given by Uberoi [5] and Chapman and Kendall
[6]. This solution describes the plasma carrying oppositely-directed magnetic
field lines from two sides toward the neutral point, where they slip through
the plasma, and are cut and reconnected, and carried outwards (see Figure 1).
However, this solution had an initially current-free magnetic field, so it was not
appropriate for the reconnection problem. Shivamoggi [7] modified this solution
so as to remedy this defect. This modified solution predicted a sequence of events
associated with the evolution of a current sheet in a hyperbolic magnetic field,
1
The MHD model offers the easiest approach to describe the macroscopic interaction
between plasmas and magnetic fields. Besides, many space- and laboratory-plasma
phenomena can be adequately described by the MHD model. The MHD approx-
imation is essentially valid if there is sufficient localization of the interaction of
particles in physical space. This localization can be achieved either by collisions be-
tween the particles or by gyrations of the particles in a strong magnetic field. In
the MHD model, the ion-dynamics play an important role and electrons with their
fast-response ability provide almost instant shielding to any charge imbalances.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 381–388, 2001.

c Springer-Verlag Berlin Heidelberg 2001
382 B.K. Shivamoggi

in agreement with laboratory experiments (Frank [3], Kirii et al. [8], [9], and
Bogdanov et al. [10]) on the collapse of a plasma near the hyperbolic magnetic
neutral line.
Recently, this solution was generalized (Shivamoggi [11]) to incorporate a
uniform shear-strain rate in the plasma flow so that the magnetic field lines now
undergo not only sweeping but also shearing by the plasma flow. The effect of
the shear strain in the plasma flow was found to impede the current-sheet forma-
tion. The integrability aspects of the system of nonlinearly-coupled differential
equations governing these dynamics have been investigated (Rollins and Shiva-
mog [12]) which seem to show that the effect of the shear strain in the plasma
flow is to produce chaotic evolutions in the dynamical system in question.

Fig. 1. Evolution of the streamlines and field lines.

2 Current-Sheet Formation at a Hyperbolic Magnetic

Neutral Line in a Stagnation-Point Plasma Flow
Consider a two-dimensional problem with the flow velocity and the magnetic
field confined to the xy-plane and independent of the z-coordinate.2
2
A generalized procedure to construct solutions relaxing the latter constraint was
sketched by Shivamoggi and Uberoi [13].
 Current-Sheet Formation near a Hyperbolic Magnetic Neutral Line 383

The MHD equations for an incompressible plasma are

 
∂
 + v · ∇ v = −∇p − (∇2 A)∇A (1)
∂t
 
∂
+ v · ∇ A = η∇2 A (2)
∂t
where v is the plasma velocity, p is the plasma pressure,  is the plasma mass
density (which will be put equal to unity in the following) and the magnetic field
B is given by
B = ∇A × îz . (3)
Let us choose the following initial conditions (Shivamoggi [7]):
t=0: Vx = −ax, Vy = ay, A = K(kx2 − y 2 ), (4)
where a, K and k are externally-determined parameters, with a > 0 and k > 1.
(4) describes a stagnation-point plasma flow impinging transversely onto the
x = 0 plane of the type involved in the laboratory experiments ([3], [8]-[10]).
The initial value of the current density J for the arrangement (4) is nonzero
and is given by
t = 0 : J ≡ ∇ × B = 2K(1 − k)iz . (5)
The Lorentz force associated with the initial current density (5) is
t = 0 : J × B = 2K 2 k(1 − k)xîx − 2K 2 (1 − k)y îy (6)
which has components directed toward the origin along the x-axis and away from
it along the y-axis so as to maintain the prescribed initial stagnation-point flow
(4).
Let us now assume that the solution, for t > 0, of equations (1) and (2),
commensurate with the initial conditions (4), is of the form
Vx (x, y, t) = −γ̇(t)x, Vy (x, y, t) = γ̇(t)y
1
A(x, y, t) = K[α(t)kx2 − β(t)y 2 ], p(x, y, t) = − ν(t)(x2 + y 2 ) + p0 (7)
2
with the initial conditions:
t=0:α=β=1 and γ̇ = a. (8)
Note that, for the solution (7), ∇2 A is a function only of t, so that the effect
of resistivity, according to equation (2), is to add a function of t to A (which
leaves the magnetic field unaltered) and hence to introduce an electric field along
the z-axis. We therefore neglect the effect of resistivity in the following.
Substituting (7), we obtain from equations (1) and (2),
α̇ − 2γ̇α = 0 (9)
β̇ + 2γ̇β = 0 (10)
γ̈ = 2K (k α − β )
2 2 2 2
(11)
2
K
ν = γ̇ 2 + (kα − β)2 (12)
4π
384 B.K. Shivamoggi

from which, we obtain, on using (8),

α(t) = e2γ(t) (13)

−2γ(t)
β(t) = e (14)
−4γ
2 2
γ̇ = K (k e 2 4γ
+e )−C (15)

with
t = 0 : γ = 0. (16)
Here
C ≡ K 2 (k 2 + 1) − a2 .
From equations (13)-(16), we obtain for small t,

γ(t) ≈ at + K 2 (k 2 − 1)t2 (17)

and for large γ(t),  

1 1
γ(t) ≈ ln (18)
2 2Kk(t0 − t)
where t is a constant of integration. According to (17), γ(t) becomes infinite
in a finite time (another finite-time singularity!) .3 This is perhaps plausible
because of the steady infinite energy input into the system via the stagnation-
point plasma flow .
The solution (13)-(18) shows that the evolution of the streamline/field line
configuration for t > 0 is as follows:
* the streamlines have a fixed shape (only the magnitude of the velocity
changes with t),
* the magnetic separatices rotate toward each other as t increases,
(see Figure 1).
Therefore, as t increases, the magnetic field becomes tangential to the y-axis,
with intensity increasing indefinitely, and shows a discontinuity on the y-axis; so,
a very thin neutral current sheet is formed on the y-axis. The solution (13)-(18)
also shows that as the initial magnetic field gradient K increases, the magnetic
separatices will rotate towards each other faster.
This evolution scenario has been confirmed by the laboratory experiments
([3], [8]-[10]) on the collapse of a plasma near the hyperbolic neutral line, which
showed that:
* a thin neutral current sheet separating oppositely-directed magnetic fields
formed,
3
Bulanov and Sakai [14] recently demonstrated the existence of another solution of
two-dimensional incompressible MHD equations which also shows a finite-time singu-
larity. These results are compatible with a recent rigorous result of Klapper [15] that,
for smooth initial conditions, no current sheet (or any other type of singularity) can
form unless ∇v or ∇B blow up outside a finite region around the two-dimensional
magnetic null point.
 Current-Sheet Formation near a Hyperbolic Magnetic Neutral Line 385

* during the process of sheet formation, a two-dimensional stagnation-point

plasma flow in the plane normal to the current direction developed — plasma
flowed toward the sheet surface at right angles on both sides and flowed out of
the sheet along the sheet surface,
* an increase in the initial magnetic field gradient produced a rapid change
in the structure of the magnetic field near the neutral line and hastened the
current-sheet formation.

3 Effect of a Uniform Shear–Strain in the Plasma Flow

Let us now generalize the solution (13)-(18) to incorporate a uniform shear-strain
in the plasma flow, and hence, investigate the effect of this shear-strain on the
current-sheet formation.
Let us now choose the initial conditions to be (Shivamoggi [11]):
t = 0 : Vx = −ax + by, Vy = ay + bx, A = kx2 − y 2 + bxy. (19)
(19) incorporate in the plasma flow a uniform shear-strain rate characterized by
the parameter b (let us take b > 0). Note that the initial plasma-flow is, however,
irrotational because the initial value of the vorticity Ω, from (19), is
t = 0 : Ω ≡ ∇ × v = 0. (20)
The initial value of the current density J is, on the other hand, nonzero and is
given by
t = 0 : J ≡ ∇ × B = 2(1 − k)îz (21)
and leads to the Lorentz force required to maintain the initial plasma flow (19).
Let us now assume that the solution, for t > 0, of equations (1) and (2),
commensurate with the initial conditions (19), is of the form
Vx (x, y, t) = −P (t)x + S(t)y, Vy (x, y, t) = P (t)y + R(t)x
1
A(x, y, t) = kα(t)x2 − β(t)y 2 + σ(t)xy, P (x, y, t) = − ν(t)(x2 + y 2 ) + p0
2
(22)
with the initial conditions
t = 0 : α = β = 1, P = a, R = S = σ = b. (23)
Note that for the solution (22), ∇2 A is a function only of t again, so that
resistivity does not play an important role in the evolution of solution (22) and
is therefore neglected.
The vorticity
Γ = [R(t) − S(t)]îz
evolves according to the equation (obtained by taking the curl of equation (1)):
 
∂
+ v · ∇ Ω = (Ω · ∇) v + (B · ∇) J − (J · ∇) B. (24)
∂t
386 B.K. Shivamoggi

This gives
d
[R(t) − S(t)] = 0 (25)
dt
from which, on using (20), we obtain

R(t) = S(t), (26)

which implies that the plasma-flow remains irrotational for all t.

Substituting (22), and using (26), we have from equations (1) and (2),

k α̇ − 2kP α + R2 = 0 (27)
β̇ + 2P β − R = 0
2
(28)
Ṙ + 2R(kα − β) = 0 (29)
Ṗ − 2(k 2 α2 − β 2 ) = 0 (30)

Equations (27)-(30) admit an integral:

1
I = P 2 + R2 − (k 2 α2 + β 2 ) = Ĉ (31)
2
where
1
Ĉ ≡ a2 + b2 − (k + 1). (32)
2
The invariance of I is confirmed by the numerical solution of equations (27)-(30)
(see Figure 2).
(31) implies that P , R, α and β may grow with t even though the quantity
(P 2 + 12 R2 −k 2 α2 −β 2 ) is bounded. Physically, this of course implies the formation
of a very thin current sheet on the y-axis, as t ⇒ ∞.
(31) also shows the shear strain in the plasma flow (represented by R):
* acts as a sink for the magnetic energy,
* impedes the current-sheet formation.
This may be seen alternatively by considering the small-time behavior of the
solutions of equations (27)-(30).
Putting
P (t) = γ̇(t) (33)
we obtain from equations (27) and (28), for small t,
1
α(t) ≈ 1 + 2γ(t) − [R(t)]2 t (34)
k
1
β(t) ≈ 1 − 2γ(t) + [R(t)]2 t. (35)
k
Using (33), (34) and (35), equations (29) and (30) give
4 2
γ̈ = 2(k 2 − 1) + 8(k 2 + 1)γ − (k + 1)R2 t (36)
k
Ṙ + 2(k − 1)R = 0 (37)
 Current-Sheet Formation near a Hyperbolic Magnetic Neutral Line 387

Fig. 2. I vs. t from a numerical solution of equations (27)-(30) for the initial conditions
t = 0 : P = 0.5, R = 1.0, α = 1.0, β = 1.0, and the parameter k = 1.

from which, we have for small t,

2b2 2
γ(t) ≈ at + (k 2 − 1)t2 − (k + 1)t3 . (38)
3k
(38) confirms that the presence of a shear strain in the plasma-flow (i.e., b = 0)
tends to impede the current-sheet formation.

4 Discussion
Current-sheet formation near a hyperbolic magnetic neutral line has been inves-
tigated by including the effects of both sweeping and shearing of the magnetic
field lines by the plasma flow. The field-line shearing has been found to impede
the current-sheet formation. On the other hand, investigation of the integrabil-
ity aspects of the system of nonlinearly-coupled differential equations governing
these dynamics indicates [12] the possibility of shear-induced chaotic evolutions
in the dynamical system in question.

Acknowledgements
The author is thankful to the referee for his valuable remarks and suggestions.

References
1. S.I. Syrovatskii: Sov. Astron. J. 10, 270 (1966).
2. P.J. Baum, A. Bratenahl, M. Kao and R.S. White: Phys. Fluids 16, 1501 (1973).
3. A.G. Frank: Proc. Lebedev Phys. Inst. (Moscow) 74, 108 (1974).
388 B.K. Shivamoggi

4. W. Gekelman, R.L. Stenzel and N. Wild: Phys. Sciprta T2/2, 277 (1982).
5. M.S. Uberoi: Phys. Fluids 6, 1379 (1963).
6. S. Chapman and P.C. Kendall: Proc. Roy. Soc. (London) A 271, 435 (1963).
7. B.K. Shivamoggi: Phys. Fluids 29, 769 (1986).
8. N.P. Kirii, V.S. Markov, S.O. Syrovatski, A.G. Frank and A.Z. Khodzhaev: Proc.
Lebedev Phys. Inst. (Moscow) 110, 121 (1979).
9. N.P. Kirii, V.S. Markov, A.G. Frank and A.Z. Khodzhaev: Fiz. Plazmy 3, 538
(1977).
10. S. Yu Bogdanov, V.S. Markov, A.G. Frank and A.Z. Khodzhaev: J. Phys. Col. C7,
40, Suppl. 7, C7-221-2 (1979).
11. B.K. Shivamoggi: Phys. Lett. A 258, 131 (1999).
12. D.K. Rollins and B.K. Shivamoggi: Europhys. A, submitted for publication (2000).
13. B.K. Shivamoggi and M.S. Uberoi: Phys. Fluids 22, 2247 (1979).
14. S. Bulanov and J. Sakai, J. Phys. Soc. Japan 66, 3477 (1997).
15. I. Klapper, Phys. Plasmas 5, 910 (1998).
Nonlocality in Turbulence

Arkady Tsinober

Faculty of Engineering, Tel Aviv University, Tel Aviv 69978

1 Introduction and Simple Examples

Nonlocality is among the three main reasons1 why the problem of turbulence is
so difficult. The term ‘nonlocality’ is used here in several related meanings which
are clarified in the course of the discussion of the issues throughout this paper.
Taking the position that velocity fluctuations represent the large scales and
the velocity derivatives represent the small scales one can state that in homo-
geneous (not necessarily isotropic) the large and the small scales are uncor-
related. This is expressed quantitatively via correlations between velocity and
vorticity. For example, in a homogeneous turbulent flow the mean Lamb vec-
tor ω × u = 0, and also (u · ∇)u = 0. If the flow is statistically reflexionally
symmetric then the mean helicity ω · u vanishes too. However, vanishing corre-
lations do not necessarily mean absence of dynamically important relations. In-
deed, the quantities (u · ∇)u ≡ ω × u + ∇(u2 /2) and ω × u, are the main ‘guilty’
for all we call turbulence. Both contain the large scales (velocity) and small scales
(velocity derivatives, vorticity). So some kind of coupling between the two is un-
avoidable.
Let us begin from the kinematic relation between velocity and vorticity,
which is just a consequence of the relation ω =curl u. Therefore any alter-
ing of ω (and strain) results in its ‘reacting back’ on the velocity field. This
point is not that trivial as may seem. Indeed, take a Helmholz decomposition
of the most significant part of the nonlinear term in NSE the Lamb vector,
ω × u = ∇α + ∇ × β. Assuming that ω and  u are random
 Gaussian
 and un-
related , i.e. ω =rotu the result is, that (∇α)2 = (∇ × β)2 . However, if
ω =rotu and u is quasi-Gaussian,
  i.e. obeys the zero-forth-cumulant relation,
then (∇α)2 ∼ 2 (∇ × β)2 i. e. in this case the rms of the potential part
of the Lamb vector is twice larger than its solenoidal part2 . In other words ω
‘reacts back’ on velocity, and consequently on ω × u, even for purely kinematic
reasons.
More generally, vorticity is not just small scales. It is of special importance,
since together with boundary conditions the whole flow field is determined en-
tirely by the field of vorticity. This, of course, includes the velocity field itself,
and therefore the large scales are determined by the small scales and vice versa.
1
The three N’ s: nonlinearity, nonintegrability and nonlocality.
2
In real turbulent flows this difference is even larger.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 389–395, 2001.

c Springer-Verlag Berlin Heidelberg 2001
390 A. Tsinober

This is the simplest indication not only for direct interaction/coupling between
large and small scales but also that this interaction is bidirectional.
A natural question is what about the correlation(s) and coupling between
velocity, ui and the strain, sij . Again in a homogeneous turbulent flow they
are uncorrelated ui sij  = 0. However, velocity
 is correlated with small
 scales

of ‘higher order’. Namely, the correlation, ui ∇2 ui = −2 sij sij  = − ω 2 , is
essentially nonvanishing again for purely kinematic reasons. Nevertheless, it is
of special dynamical significance as being directly related to the dissipation of
turbulent energy,  = 2ν sij sij  , which for dynamical
 reasons
 remains finite
for whatever small ν. Therefore the correlation ui ∇2 ui becomes very large
at small ν. However, the corresponding correlation coefficient at large Reynolds
numbers is roughly of the order Re−1/4 , i.e. becomes very small. This, of course,
does not mean that the coupling between u and ∇2 u becomes unimportant at
large Reynolds numbers. Moreover, just like in case of vorticity, strain is not
just small scales, since together with boundary conditions the whole flow field is
determined entirely by the field of strain [1].
One can use the above example with the Lamb vector to illustrate the dy-
namical aspect of this coupling. For this we retreat from homogeneous flows and
consider a unidirectional in the mean fully developed turbulent flow such as the
flow in a plane channel in which all statistical properties depend on the coordi-
nate normal to the channel boundary, x2 , only. In such a flow a simple precise
kinematic relation is valid

du1 u2 /dx2 ≡ ω × u1 = ω2 u3 − ω3 u2  = 0, (1)

which is just a consequence of the vector identity (u · ∇)u ≡ ω × u + ∇(u2 /2)

in which incompressibility and d· · · /dx1,3 = 0 where used, and · · ·  means an
average in some sense (e.g. time or/and over the planes x2 = const, etc.). The
dynamical aspect is that in turbulent channel flows du1 u2 /dx2 = 0 is essen-
tially different from zero at any arbitrarily large Reynolds number. Therefore one
can see from (1) that at least some correlations between velocity and vorticity in
such flows are essentially different from zero. Since vorticity is essentially a small
scale quantity the relation (1) is a clear indication of a dynamically important
statistical dependence between the large scales (u) and small scales (ω). With-
out this dependence du1 u2 /dx2 ≡ 0, which means that the mean flow would
not ‘know’ about its turbulent part at all. It is noteworthy that corresponding
correlation coefficients (and many other statistical characteristics) are of order
10−2 even at rather small Reynolds numbers. Nevertheless, as we have seen, in
view of the dynamical importance of interaction between velocity and vorticity
in turbulent shear flows such ‘small’ correlation by no means does not imply
absence of dynamicaly important statistical dependence and direct interaction
between large and small scales. Indeed it is this interaction that results in drastic
changes of the whole mean flow.
 Nonlocality in Turbulence 391

2 Different Aspects of Nonlocality

From the formal point a process is called local if all the terms in the governing
equations are differential. If the governing equations contain integral terms, then
the process is nonlocal. The Navier-Stokes equations are integro-differential for
the velocity field in both physical and Fourier space (and any other). Therefore,
generally, the Navier-Stokes equations describe nonlocal processes. The prob-
lem is intimately related to the one of decompositions/representations, which is
briefly mentioned below.
The property of nonlocality of Navier-Stokes equations in physical space
is two-fold. On one hand, it is due to pressure (‘dynamic’ nonlocality), since
ρ−1 ∇2 p = ω 2 −2sij sij , so that pressure is nonlocal due to nonlocality of the oper-
ator ∇−2 . This nonlocality is strongly associated with essentially non-Lagrangian
nature of pressure. For example, replacing in the Euler equations the pressure
2
Hessian ∂x∂i ∂xp
j
≡ Πij , which is both nonlocal and non-Lagrangian, by a local
quantity (1/3)δij ∇2 p = (1/6)ρ{ω 2 − 2sij sij } turns the problem into a local and
integrable one and allows to integrate the equations for the invariants of the
tensor of velocity derivatives ∂ui /∂xj in terms of a Lagrangian system of coordi-
nates moving with a particle, see references in [1]. The reason for disappearance
of turbulence (and formation of singularity in finite time) in such models, called
restricted Euler models, is that the eigenframe of sij in these models is fixed in
space [2], whereas in a real turbulent flow it is oriented randomly. This means
that nonlocality due to presure is essential for (self-) sustaining turbulence: no
presure Hessian - no turbulence. A related aspect is that the Lagrangian accel-
eration Du/Dt - a kind of small scale quantity – is dominated by the pressure
gradient, ∇p, [3].
Taking curl of the NSE and getting rid of the pressure does not remove the
nonlocality. Indeed, the equations for vorticity and enstrophy are nonlocal in
vorticity, ω, since they contain the rate of strain tensor, sij , due to nonlocal
relation between vorticity, ω, and the rate of strain tensor, sij (‘kinematic’ non-
locality)3 . The two aspects of nonlocality are related, but are not the same. For
example, in compresible flows there is no such relatively simple relation between
pressure and velocity gradient tensor as above, but the vorticity-strain relation
remains the same.
Both aspects of nonlocality are reflected in the equations for the rate of
strain tensor and total strain/dissipation, s2 ≡ sij sij , and the equations for the
third order quantities ωi ωj sij and sij sjk ski . An important aspect is that the
latter equations contain invariant quantities ωi ωj Πij and sik skj Πij reflecting
the nonlocal dynamical effects due to pressure and can be interpreted as inter-
action between vorticity and pressure and between vorticity and and strain. For
3
Nonlocality of the same kind is encountered in problems dealing with the behaviour
of vortex filaments in an ideal fluid. Its importance is manifested in the breakdown
of the so called localized induction approximation (LIA) as compared with the full
Bio-Savart induction law, see [4] and references therein.
392 A. Tsinober

example, the equation for ωi ωj sij

D(ωi ωj sij )/Dt = ωi sij ωk ski − ωi ωj Πij + νωi ∇2 ωi , (2)

shows both aspects of nonlocality of vortex stretching process. The first term in
(2) (which is just the squared magnitude of the vortex stretcing vector) is strictly
positive ωi sij ωk ski ≡ W 2 > 0. This means that the nonlinear processes involving
vortex stretching (or direct interaction of vorticity and strain) always tend to
increase even the instantaneous enstrophy production. However, the inviscid
rate of change of enstrophy generation contains also a second term reflecting
the interaction between vorticity and the pressure Hessian Πij . This is the term
−ωi ωj Πij . Without this term the question on why ωi ωj sij  > 0 would be
immeadiately answered. It appears [1] that ωi ωj Πij  is positive and is about
W 2 /3, i.e. in the mean the nonlinearity in (2) is reduced by this nonlocal term,
since for a Gaussian velocity field ωi ωj Πij  ≡ 0. The nonvanishing correlation
ωi ωj Πij  (and also sik skj Πij ) is also one of the manifestations of nonlocality,
the direct coupling between the large and small scales.

2.1 Direct Coupling Between Large and Small Scales

Nonlocality in the sense as discussed above is an indication of direct coupling

between large and small scales. There exist massive evidence that this is re-
ally the case as there are many indications both for direct interaction/coupling
between large and small scales but also that this interaction is bidirectional4 .
We mention first the well known effective use of fine honeycombs and screens
in reducing large scale turbulence in various experimental facilities. The exper-
imentally observed phenomenon of strong drag reduction in turbulent flows of
dilute polymer solutions and other drag reducing additives is another example
of such a ‘reacting back’ effect of small scales on the large scales [5]. Third, one
can substantially increase the dissipation and the rate of mixing in a turbulent
flow by directly exciting the small scales [6].

Anisotropy. One of the the manifestations of direct interaction between large

and small scales is the anisotropy in the small scales. Though local isotropy
is believed to be one of the universal properties of high Re turbulent flows it
appears that it is not so universal: in many situations the small scales do not
forget the anisotropy of the large ones. There exist considerable evidence for
this which has a long history starting somewhere in the 50-ies, see references
in [1], [7], [8]. Along with other manifestations of direct interaction between
large and small scales the deviations from local isotropy seem to occur due to
various external constraints like boundaries, initial conditions, forcing (e.g. as
in DNS), mean shear/strain, centrifugal forces (rotation), buoyancy, magnetic
field, etc., which usually act as an organizing factor, favoring the formation of
4
Note that in case of passive objects there is no such a bidirectional relation – it is
only one way.
 Nonlocality in Turbulence 393

coherent structures of different kinds (quasi-two-dimensional, helical, hairpins,

etc.). These are as a rule large scale features which depend on the particularities
of a given flow and thus are not universal. These structures, especially their
edges seem to be responsible for the contamination of the small scales. This
‘contamination’ is unavoidable even in homogeneous and isotropic turbulence,
since there are many ways to produce such a flow, i.e. many ways to produce
the large scales. It is the difference in the mechanisms of large scales production
which ‘contaminates’ the small scales. Hence, nonuniversality.
Let us turn first again to the ‘simlpe’ example above and look at the prop-
erties in the proximity of the midplane, x2 ≈ 0, of the turbulent channel flow.
In this region dU/dx2 ≈ 0, but the flow is neither homogeneous nor isotropic,
since though u1 u2  ≈ 0 in this region too, d u1 u2  /dx2 is essentially nonzero
and is finite independently of Reynolds number as far as the data allow to make
such a claim. This is also a clear indication of nonlocality, since in the bulk of
the flow, i.e. far from the boundaries, dU/dx2 ∼ 0.
The first experimental evidence on anisotropy in small scales at large Reynolds
numbers in the atmospheric boundary layer experiments goes back to the fact
that the skewness of the derivative of temperature fluctuations is not small, as
should be in locally isotropic flow, and is of order 1, whereas for a locally isotropic
flow it should be close to zero, see [8] for further references. An important fea-
ture of these flows is the presence of a mean gradient of the passive scalar – the
rest is not so important: the phenomenon is observed for a Gaussian and two
dimensional velocity field [8].
Recently similar observations were made for the velocity increments and ve-
locity derivatives in the direction of the mean shear both numerical and lab-
oratory, see references in [7]. It was found that the stastistical properties of
velocity increments and velocity derivatives in the direction of the mean shear
do not conform with and do not confirm the hypothesis of local isotropy. More-
over, our results imply that the large scales are directly coupled to the small
scales. (The anisotropy disappears when the large scale shear is removed), [7].
More precisely these results imply that there is a direct influence of mean shear
on the small scales, which is possible due to the permanent bias of the mean
shear to which is exposed the field of fluctuations due to its very large residence
time in the mean shear. One of the explanations of the results from [7] is that
the large scales are directly coupled to the small scales. However, this does not
mean that there is no such coupling when the large scale shear is removed. An-
other exlanation is that in this experiment the value of the Corrsin criterion
∗
SC = (dU/dx2 )(ν/ )1/2 ≈ 2.4 · 10−2 . This is the ratio of the Komlogorov time
scale, τη = (ν/ )1/2 , to the time scale, (dU/dx2 )−1 , associated with the mean
shear, and it should be small enough in order to have isotropy in small scales.
The main problem is how small. There is no agreement on this issue, but there is
evidence that in order to have one decade of isotropic inertial range in boundary
∗
layer flows (both simple and complex) at Reλ ≈ 1500 it is necessary that SC
−2
< 10 , see [10] and references therein. This brings us to the next issue.
394 A. Tsinober

Other manifestations of direct coupling between large and small scales.

An important recent observation, [9] and references therein, is that conditional
statistics of small scale quantities (e.g. velocity increments, estrophy, total strain)
conditioned on large scale quantities (velocity) is not independent of the large
scale quantities as should be if if the large scales are not coupled directly to the
small scales. We stress that this observation was made in [9], in spite that in this
experiment the mean shear was rather small, less than 0.1s−1 . This corresponds
∗
to the value of the Corrsin criterion SC = (dU/dx2 )(ν/ )1/2 ≈ 2 · 10−3 , which
is an order of magnitude smaller than in [7] and five times is lower than the value
0.01 required for local isotropy in presence of mean shear mentioned above. Two
aspects deserve special comment regarding the experiment in [9]. First, there is
a clear tendency of increase of the conditional averages of the structure functions
with the energy of fluctuations. Second, such a tendency, that is the direct cou-
pling, is observed also for the smallest distance of the order of Kolmogorov scale
∼ η, which was used for estimates of the derivatives in the streamwise direction.
Similar behaviour is exhibited by the enstrophy ω 2 and the total strain sij sij .
The observations on the the coupling between and the ‘reaction back’ of the
small scales on the large by no means are not exausted by the references given
above. As an example from the atmospheric physics we bring a quotation of the
first conclusion rached at the Symposium on the nature fo so clear air turbulence
[11]: The energy dissipated at small-scale by clear air trubulence influences the
large-scale atmospheric motion.


Helicity. Helicity, ω · udx, and its density, ω · u, deserve here also special
mentioning. The formal reason is that if u · ω = 0, this is a clear indication
of direct coupling of large and small scales. So it is not surprising that in flows
with nonzero mean helicity the direct coupling between small and large scales
is stronger than otherwise. The stronger coupling between the large and small
scales in flows with nonzero mean helicity u · ω aids creation of large scale
structures out small scale turbuence, see [12] and references therein. As men-
tioned this does not mean that in case u · ω = 0, or even u · ω = 0 as in
two-dimensional flows, such a coupling does not exist.

On closures and constitutive relations. Memory effects. The nonlocality

due to the coupling between large and small scales is also manifested in problems
related to various decompositions of turbulent flows and in the so called closure
problem. For example, in the Reynolds decomposition of the flow field into the
mean and the fluctuations and in similar decompositions associated with large
eddy simulations (LES) the relation between the fluctuations and the mean flow
(or resolved and unresolved scales in LES, etc.) is a functional. That is the field
of fluctuations at each time/space point depends on the mean (resolved) field
in the whole time/space domain. Vice versa the mean (resolved) flow at each
time/space point depends on the field of fluctuations (unresolved scales) in the
whole time/space domain. This is because the equations for the fluctuations (in-
resolved scales) contain as coefficients the mean (resolved) field. This means that
 Nonlocality in Turbulence 395

in turbulent flows point-wise flow independent ’constitutive’ relations analogous

to real material constitutive relations for fluids (such as stress/strain relations)
can not exist, though the ‘eddy viscosity’ and ‘eddy diffusivity’ are used fre-
quently as a crude approximation for taking into account the reaction back of
fluctuations (unresolved scales) on the mean flows (resolved scales). The fact
that the ‘eddy viscosity’ and ‘eddy diffusivity’ are flow (and space/time) depen-
dent is just another expression of the strong coupling between the large and the
small scales.

3 Concluding Remarks
The general conclusion is that the small scales are not decoupled from the large
scales even at Reynolds numbers (based on the Taylor microscale) as large as
104 . In other words the Galilean invariance is broken in the restricted sense
that the properties of small scale turbulence are not independent of parameters
characterizing the large scales, such as, e.g. the energy of velocity fluctuations.
This direct and bidirectional interaction/coupling of large and small scales is a
generic property of all turbulent flows and one of the main reasons for small scale
intermittency, non-universality, and quite modest manifestations of scaling. In
view of such coupling it is not clear how meaningful is the notion of inertial
range and things like eddy viscosity representations of the subgrid scales.

References
1. A. Tsinober: ‘Turbulence - Beyond Phenomenology’, Lect. Notes Phys., 511,
(Springer, Berlin, Heidelberg 1998), pp. 85-143; A. Tsinober, An informal intro-
duction to turbulence (Kluwer, Dordrecht, Boston 2001).
2. E.A. Novikov: Fluid Dyn. Res., 6, 79 (1990).
3. P. Vedula and P.K. Yeung , Phys. Fluids, 11, 1208 (1999).
4. R.L. Ricca, D. Samuels, C.F. Barenghi: J. Fluid Mech., 391, 29 (1999).
5. A. Gyr and H.-W. Bewersdorff: Drag reduction of turbulent flows by additives,
(Kluwer, Dordrecht, Boston 1995).
6. J. M. Wiltse and A. Gledzer.: Phys. Fluids, 10, 2026 (1998).
7. X. Shen and Z. Warhaft: Phys. Fluids, 12, 2976 (2000).
8. Z. Warhaft: Ann. Rev. Fluid Mech., 32, 203 (2000).
9. M. Kholmyansky, and A. Tsinober: ‘On the origins of intermittency in real tur-
bulent flows’, In: Intermittency in Turbulent Flows, ed. by J.C. Vassilicos, (Cam-
bridge, Cambridge University Press 2000), pp. 183-192.
10. S.G. Saddoughi: J. Fluid Mech., 348, 201 (2000).
11. Y.-H. Pao, A. Goldburg: Clear air turbulence and its detection, (Plenum, New York
1969).
12. Droegemeier, K.K., Lazarus, S.M. and Davies-Jones, R.: Month. Weather Rev.,
121, 2006 (1993).
Quantized Vorticity in Superfluid 3 He-A:
Structure and Dynamics

R. Blaauwgeers1,2 , V.B. Eltsov1,3 , M. Krusius1 , J. Ruohio1 , and R. Schanen1,4

1
Low Temperature Lab, Helsinki University of Technology, FIN-02015 HUT, Finland
2
Kamerlingh Onnes Lab, Leiden University, 2300 RA Leiden, The Netherlands
3
Kapitza Institute, Kosygina 2, Moscow 117334, Russia
4
CRTBT-CNRS, BP 166, F-38042 Grenoble Cedex 09 FRA, France

Abstract. Superfluid 3 He-A displays the largest variety in vortex structure among the
presently known coherent quantum systems. The experimentally verified information
comes mostly from NMR measurements on the rotating fluid, from which the order-
parameter texture can often be worked out. The various vortex structures differ in
the topology of their order-parameter field, in energy, critical velocity, and in their
response to temporal variations in the externally applied flow. They require different
experimental conditions for their creation. When the flow is applied in the superfluid
state, the structure with the lowest critical velocity is formed. In 3 He-A this leads to the
various forms of continuous (or singularity-free) vorticity. Which particular structure is
created depends on the externally applied conditions and on the global order-parameter
texture.

1 Superfluid 3 He

The accepted textbook example of a superfluid has traditionally been 4 He-II.

However, in many respects the 3 He superfluids display more ideal behaviour,
both in their theoretical description and their macroscopic properties. One re-
markable difference is the absence of remanent vorticity in most experimental
setups for superfluid 3 He. In 4 He-II, the formation of vortex lines via mechanisms
intrinsic to the superfluid itself is observed only in rare cases, most notably in
the superflow through a sub-micron-size orifice where the vortex lines are blown
out of the immediate vicinity of the aperture and are trapped on surface sites far
away where the flow is small. In 3 He superfluids, where the vortex core radii are
at least 100 times larger, intrinsic critical velocities can be measured simply with
bulk superfluid flowing past a flat wall. For this reason rotating measurements
have proven very efficient in their study, quite unlike in 4 He-II.
The hydrodynamics of the 3 He superfluids abound with new features which
have only marginally been investigated. Some experimental work has been per-
formed on vortex tangles in the context of quench-cooled non-equilibrium tran-
sitions from the normal liquid to the quasi-isotropic 3 He-B [1] and also around
vibrating wire resonators in the zero temperature limit [2]. Little if any work has
been reported on turbulent flow or on vortex networks in the highly anisotropic
3
He-A phase [3]. Nevertheless, this is the phase with the much richer variety of
response to externally applied flow. Here we describe preliminary studies in this

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 399–420, 2001.

c Springer-Verlag Berlin Heidelberg 2001
400 R. Blaauwgeers et al.

vortex cluster
7 mm

counterflow
v = vn - vs

v vn = Ω R
Ω
κ
v= (N0 - N)
2 π R
vortex deficit
r
0 R κ N
vs =
2 π R

Fig. 1. Vortex lines in a rotating cylindrical container. Metastable states, which include
fewer vortex lines than the equilibrium state, consist of a central cluster with rectilinear
vortex lines and a surrounding annular region of vortex-free counterflow. The Magnus
force from the circulating counterflow confines the lines in the cluster to the density
nv = 2Ω/(νκ), when the cluster rotates like a solid body with the container. If Ω
is increased, the cluster contracts, the counterflow velocity increases and ultimately
reaches the critical velocity limit vc , where a new vortex is created. If Ω is reduced,
the counterflow velocity diminishes, the cluster expands and eventually reaches the
annihilation threshold where vortices in the outermost circle of lines are pushed to
the cylindrical wall for annihilation. The graph on the bottom illustrates how the
counterflow velocity changes as a function of the radial position in the container. Also
shown how its value v(R) at the cylinder wall depends on the vortex-line number
N , when N0 ≈ πR2 nv denotes their number in the equilibrium state at the rotation
velocity Ω.

direction, which explain the critical flow velocity [4] and the dynamic response
to rotation in the A phase [5]. Unusual new features are observed, which all
can be explained by the structural properties of the anisotropic order-parameter
field, but which also suggest that more surprises can be expected. The critical
velocity [6] and the response of vortex lines to a dynamic rotation drive [7] in
3
He-B have been measured earlier. These measurements attest to more tradi-
tional behaviour, as can be expected for vortex lines with singular cores in an
order-parameter field with many orders of magnitude weaker anisotropy.
Much of the existing information on new vortex structures in the 3 He su-
perfluids has been derived from NMR measurements on bulk liquid samples
contained in a rotating cylinder with smooth walls. The principle of this “rotat-
ing bucket” method is outlined in Fig. 1. Here all sensors can be placed outside
 Quantized Vorticity in Superfluid 3 He-A: Structure and Dynamics 401

the cylinder and one can investigate the undisturbed order-parameter field in
the bulk superfluid, in the presence of different types of quantized vorticity as
well as the conditions in which they are created.

2 Order-Parameter Texture and Superflow in 3 He-A

The beauty of 3 He-A lies in its anisotropy: Although the underlying material,
liquid 3 He, is isotropic, an all-pervading anisotropy arises from the condensation
into a coherent p-wave paired state. Here the two fermion quasiparticles forming
a Cooper pair have relative angular momentum L = 1 and total spin S = 1 [8].
The spin structure of the condensate is characterized by the formation of spin
up-up and down-down pairs so that the total spin S has zero projection on an
axis, which traditionally is denoted by the unit vector d̂ (⊥ S). The Cooper pairs
also have a preferred direction for their orbital momentum, denoted by the unit
ˆ so that the projection of the net orbital momentum L on the ˆ axis
vector ,
is positive. Finally, also the quasiparticle excitation spectrum is anisotropic: In
momentum space the energy gap vanishes at two opposite poles of the Fermi
surface, located on the ˆ axis.
A number of competing interactions act in unison to produce a smooth
variation of the ˆ and d̂ vector fields over the container volume. These vec-
tor fields, which in the absence of singularities are smoothly continuous, are
called the orbital and spin textures. The dipole (or spin-orbit) coupling con-
nects the orbital and spin textures via a free-energy density fD = −gD (d̂ · ) ˆ 2.
The magnetic anisotropy energy in an externally applied magnetic field H is
written as fH = gH (d̂ · H)2 . These two orientational forces on d̂ balance each
other at
 a characteristic value of the field H. This is called the dipole field,
HD = gD /gH ∼ 1 mT, below which the dipole coupling wins and the texture
becomes dipole locked, with d̂ and ˆ either parallel or anti-parallel. At higher
fields, H  HD , d̂ is forced to lie perpendicular to H. This is the case in NMR
measurements: d̂ is everywhere contained in the plane perpendicular to H and
varies smoothly in this plane. This texture is coupled to the orbital alignment
such that ˆ is also for the most part forced parallel to d̂.
Important exceptions, when ˆ becomes decoupled from the planar d̂ orienta-
tion (in the presence of a magnetic field H  HD ), are topological defects like
the dipole-unlocked central part of a vortex line, which is also called the “soft
vortex core” (Fig. 2). To minimize the loss in dipole energy, the soft core has
a radius on the  order of the dipolar healing length ξD = /(2m3 vD ) ∼ 10 μm,
where vD ≈ gD /ρs ∼ 1 mm/s is the order of magnitude of the so-called
dipolar velocity, ie. the flow velocity at which the orienting force on ˆ from an
externally applied superflow matches the bending or gradient energy of , ˆ which
maintains the spatial coherence of the order parameter.
The rigidity of the continuous ˆ texture, which arises from the anisotropy
forces, from superfluid coherence, and from the boundary conditions, explains
why fluid flow in the A phase is not always dissipative, in spite of the gap nodes
in the quasiparticle spectrum. For most ˆ textures the velocity of the superfluid
 Quantized Vorticity in Superfluid 3 He-A: Structure and Dynamics 403

becomes nonvanishing in those regions of the orbital texture where the ˆ ori-
entation is not contained within one single plane. On forming the circulation
of vs along a closed path, which encircles such a region with inhomogeneous ˆ
orientations, 
 ˆ ,
νκ = vs · dr = S() (3)
2m3
one finds that the number (ν) of circulation quanta (κ = h/(2m3 )) is given by
ˆ over which the ˆ orientations extend within the encircled
the solid angle S()
region. In other words, the circulation is related to the topological charge of the
ˆ field.
In the soft core of the dipole-unlocked singularity-free vortex (Fig. 2), the
orientational distribution of the ˆ field covers a solid angle of 4π and ν = 2.
This configuration is an example of a skyrmion, which consists of two halves, a
circular and a hyperbolic Mermin-Ho vortex, which also are known as merons.
The circular half covers the 2π orientations in the positive half sphere and the
hyperbolic those in the negative half.
The dipole coupling exerts an extra torque on spin precession and gives rise
to frequency shifts in NMR. Experimentally a most valuable consequence is the
fact that dipole-unlocked regions experience a frequency shift which is different
from that of the locked bulk liquid and moreover a characteristic of the ˆ texture
within the soft core of the defect. Different structures of topological defects give
rise to absorption in satellite peaks where both the frequency shift of the peak
and its intensity are a characteristic of the defect structure [11]. This property
provides a measuring tool which differentiates between defects and where the
absorption intensity can be calibrated to give the number of defects (Fig. 3).

3 Double-Quantum Vortex Line

The generic rotating experiment consists of an acceleration – deceleration cycle in
the rotation drive, as shown in Fig. 4. This experiment gives reproducible results
in 3 He superfluids where remanent trapped vortex filaments can be avoided in a
container with smooth walls. In Fig. 4 the NMR spectrometer has been tuned to
the frequency of the satellite from the double-quantum vortex line and its peak
height is recorded as a function of the rotation velocity Ω. The rate of change
| dΩ/dt |∼ 10−4 rad/s2 is kept as slow as possible so that dynamic effects do not
influence the result, but such that long-term drifts neither become important.
During increasing Ω it is possible to discern from the measuring noise a stair-
case-like pattern in the satellite peak height when Ω ≥ Ωc . The periodicity
in this signal as a function of Ω calibrates the circulation associated with one
vortex line: ΔΩ = νκ/(2πR2 ). This provides the proof for ν = 2, namely that
the vortex line created in the externally applied flow is doubly quantized [13].
The intercept of the acceleration record with the horizontal axis, when Ω ≥
Ωc , determines the critical velocity vc in Fig. 4. The fact that this section
is linear proves that vc remains constant during all of the acceleration. On
plotting the corresponding counterflow velocity at the cylinder wall, v(Ω) =
404 R. Blaauwgeers et al.

P = 34.2 bar
H = 11.8 mT
NMR absorption
0.27 T = 0.50 Tc

0.70
0.28

0.81
CF+
0.63
CUV
0.61 VS
0.23 CUV

0.59 Main SS
0.66 peak TS

0 5 10 15 20 25
Δ f (kHz)
Fig. 3. NMR spectroscopy of topologically stable defects of the order-parameter field
in 3 He-A. The measured NMR absorption is plotted as a function of the frequency
shift f − f0 from the Larmor value f0 = γH0 , where the resonance takes place in
the normal phase. The large truncated peak on the right represents the resonance
2
absorption of the dipole-locked bulk superfluid, with a relative frequency shift R⊥ = 1.
The different satellite peaks represent: CUV — double-quantum vortex lines (Fig. 2) in
the equilibrium rotating state at Ω = 0.58 rad/s, which in a cylinder of radius R = 2.5
mm corresponds to 150 vortex lines. At a temperature as low as 0.49 Tc , the satellite
spectrum includes both a large primary and a small secondary peak. VS — equilibrium
state of the vortex sheet in the same conditions. SS — soliton sheet with splay structure
and with the sheet oriented vertically parallel to the cylinder axis (Ω = 0). TS —
soliton sheet with twist structure and oriented transverse to the cylinder axis (Ω = 0.11
rad/s, which is below but close to Ωc ). CUV + CF — double-quantum vortes lines in
a cluster surrounded by vortex-free counterflow close to the critical velocity threshold
2
(Ω = 2.0 rad/s). The two satellites with the normalized frequency shifts R⊥ = 0.27
and 0.70 are the primary and secondary double-quantum vortex peaks while that with
2
R⊥ = 0.81 is caused by the counterflow. (In this measurement the temperature is
2
higher (T = 0.54 Tc ) which explains the larger R⊥ values of the vortex satellites than
in the CUV spectrum.) The NMR field H is oriented parallel to the rotation axis.
(From Ref. [11])

ΩR − νκN/(2πR), it is seen that the noise in vc can be explained to arise from

experimental sources. Thus vortex formation proceeds here in the form of a reg-
ular periodic process and displays no measurable stochastic behaviour, which
could be associated with nucleation across an energy barrier. Indeed, a simple
argument, which is presented below, shows that the nucleation energy barrier is
 Quantized Vorticity in Superfluid 3 He-A: Structure and Dynamics 405

2 → →
H = 9.56 mT ( H ⊥ Ω )
P = 33.5 bar
T = 0.73 Tc
Satellite amplitude (mV) 1.5 .
Ω = 3.10-4 rad/s2

0.5 29 V/vortex
Ω c

0 5.9 mrad/s

0 0.1 0.2 0.3 0.4 0.5 0.6

(rad/s)
Fig. 4. Response of double-quantum vortex lines to a closed acceleration – deceleration
cycle in the rotation drive Ω: The peak height of the primary NMR satellite plotted
as a function of the externally applied rotation. Initially on increasing Ω from zero,
vortex-free counterflow is created. This is known as the Landau state in superfluids
and as the Meissner state (with complete flux expulsion) in superconductors. At Ωc
the critical counterflow velocity, vc = Ωc R, is reached at the cylinder wall and the first
vortex line is created. This brings about a reduction in the flow velocity at the wall
by Δv = νκ/(2πR). When Ω is increased further, the process is repeated periodically
and the slanting section with the slope dN/dΩ = 2πR2 /(νκ) is recorded. This means
that here the critical velocity remains constant at vc = Ωc R = ΩR − νκN/(2πR). At
maximum amplification this section can be seen to mimic a periodic signal of staircase
pattern, as shown in the insert. The upper branch of the acceleration – deceleration
cycle is measured during decreasing Ω. Here the excess vortex-free counterflow is first
reduced, until the cluster reaches the annihilation threshold, upon which vortex lines
start to annihilate during further deceleration. At the annihilation threshold the cluster
is separated
√ from the cylinder wall by a counterflow annulus of minimum width: d ≈
βR/(2 Ω). This is comparable to the inter-vortex distance within the cluster: d ≈
[νκ/(2πΩ)]1/2 . In practice the number of vortex lines at the annihilation
√ threshold
[12] is equal to that in the equilibrium state: N0 = πR2 nv (1 − β/ Ω), where β ≈
0.09 for double-quantum vortex lines in the present experimental container. The true
equilibrium state is obtained, in principle, by cooling through Tc in rotation at constant
Ω. Annihilation in the upper branch can be seen to be a more random process, in
which a larger number of lines may be removed approximately simultaneously from
the outermost circle of vortex lines. (From Ref. [13])

so large compared to thermal energy that it cannot be overcome by any usual

nucleation mechanisms. Instead, the applied counterflow has to be increased to
the point where the barrier height goes to zero [6]. Thereby vortex formation
becomes essentially an instability.
The reason for the high nucleation barrier at low applied flow velocities is the
large length scale ξD on which the vortex has to be formed. The energy stored
per unit length in the superflow around the vortex core is of order ∼ ρs ν 2 κ2 . This
406 R. Blaauwgeers et al.

Ω (rad/s) Δ N / Δ Nmax
0 0.2 0.4 0.6 0.8 0

Probability (%)
0.25 0.5 0.75 1
3 20
Transverse field

NMR absorption
T = 0.80 Tc 15
2 10
Ω b
5
Ω c
50 Vortices 0
1
Axial field
T = 0.82 Tc
0 1.5
vc2

vc (mm/s)
1

0.5
vcUS

0
0.5 0.6 0.7 0.8 0.9 1.0
T / Tc

Fig. 5. Critical velocity for the formation of the double-quantum vortex. Vortex for-
mation can proceed as a regular periodic process, as in Fig. 4, or it may start as a
burst-like event which then goes over into the periodic process, as shown in the top-left
insert. In the burst-like event a large number of vortex lines is formed simultaneously
at a bulk-liquid texture instability. Here a first order transition occurs in the order-
parameter texture and it is suddenly transformed to a new configuration in which the
critical velocity is generally lower than in the original texture. In the transformed tex-
ture the periodic process turns on during a further increase of Ω and the corresponding
Ωc is obtained by extrapolating back to zero peak height of the vortex satellite. Vortex
formation at a texture instability is possible only at high temperatures (T  0.7 Tc ),
where the energy barriers separating different textures can still be overcome. The inset
on the top right shows as a histogramme the number of vortex lines ΔN which are
produced in the burst, normalized to the equilibrium number √ ΔNmax at the rotation
velocity Ωb after the burst: ΔNmax = [2πR2 Ωb /(νκ)] (1 − β/ Ωb ). The data for the
periodic process are marked as (◦) while burst-like events are denoted as (•). Measuring
conditions: H  Ω, H =9.9 – 15.8 mT, P =29.3 – 34.2 bar. (From Ref. [4])

has to be compared to the kinetic energy of the applied superflow at the cylinder
wall in a volume comparable to that where the instability occurs, ∼ ρs v 2 ξD
2
. Thus
the flow velocity for creating the vortex has to be of order vc ∼ νκ/(2πξD ). This
shows that the velocity for reaching the instability decreases with length scale
and is in 3 He-A comparable to the dipole velocity vD ∝ 1/ξD , ie. the velocity
required to break dipole locking. The barrier height, in contrast, increases ∝ ξD 2
.
In He-II the appropriate length scale is the superfluid coherence length ξ ∼ 0.1
4

nm, which is of atomic size, and gives a barrier height of order 1 K. In 3 He

superfluids the barrier is higher and the temperature lower, both by at least three
orders of magnitude. Therefore vortex-formation takes place at an instability.
 Quantized Vorticity in Superfluid 3 He-A: Structure and Dynamics 407

2.0
(a) H (b) H

vc2
1.5

vc (mm/s)
2.0
vc||

0.5
vcUS

0
0.5 0.6 0.7 0.8 0.9 1
T / Tc

Fig. 6. Critical velocity of vortex formation for selected order-parameter textures. The
curves represent the respective calculated bulk superfluid flow instability [14]. The
absolute instability limit at vc2 applies to the originally homogeneous (ˆ  v) orbital
texture in inset (a). Here ˆ is confined to the transverse plane and is dipole-locked
everywhere, when Ω < Ωc , except within the surface layer and in the very center (where
there is either a singular disclination line or a dipole-unlocked radial 2π Mermin-Ho
vortex). This limit can be compared to measurements (•, ) in which the sample is
cooled in rotation through Tc , to obtain the global equilibrium texture. The lowest
vc is measured (×) when the NMR spectrum shows the signature from the transverse
twist soliton (cf. Fig. 3). These data points can be compared to the calculated vc of
a dipole-unlocked soliton vcU S . In transverse magnetic field, H ⊥ Ω, the equilibrium
global texture in inset (b) is also obtained by cooling through Tc in rotation and gives
data points (◦, 2) which should be compared to the calculated instability at vc . This
texture includes two singular disclination lines, located diametrically opposite each
other at the cylinder wall (marked with a black dot on the horizontal diameter in
inset (b)). The transverse field measurements seem to be less sensitive to the annealing
requirement since other data () with varied prehistories are not very different. (From
Ref. [4])

As outlined above, one might think that the instability-determined critical

velocity in 3 He-A is a well-defined quantity. It should depend only on the ex-
ternally applied conditions, such as temperature (T ) and pressure (P ), which
determine the 3 He-A properties ρs and ξD . In a smooth-walled container it
should not depend on the wall properties, since in the A phase vortex formation
has to occur essentially within the bulk liquid: At the cylinder wall ˆ is oriented
perpendicular within a surface layer of width comparable to ξD , owing to a rigid
boundary condition. However, experimentally it is immediately concluded that
the formation process has a lot of variability: It can have the regular appearance
shown in Fig. 4, or it can take the burst-like form shown in Fig. 5 (inset at top
408 R. Blaauwgeers et al.

left). Also vc depends on the previous experimental history which the sample
has been subjected to in the superfluid state [4].
All this variation in the characteristics of the critical velocity can be measured
in one and the same smooth-walled sample container. This is quite different
from 3 He-B, where vortex formation in the same cylinder proceeds as a rule as
a reproducible and well-behaved regular process, similar to that in Fig. 4 [6].
In 3 He-B the critical velocity is not history dependent and typically at least an
order of magnitude larger, as can be expected when a singular-core vortex has
to be formed on the length scale of the superfluid coherence length ξ(T, P ) ∼ 10
– 100 nm.
The seemingly unruly behaviour of 3 He-A can be explained by the depen-
dence of the instability velocity on the global order parameter texture, which
is still a poorly controlled and understood feature of the experiments. In Fig. 5
various measurements on vc have been collected, regardless of the earlier history
of the samples. The data points, accumulated from three different sample cylin-
ders, seem to fall between a maximum and a minimum limit. Note that there
are no measurements with zero or very small vc , proving that 3 He-A is a true
superfluid. The global order parameter texture in the cylinder depends on the
history of sample preparation. In Fig. 6 the measurements are grouped accord-
ing to what type of global texture is expected on the basis of the procedure
which was used to prepare the sample. This figure now provides some credibility
to the notion that the global texture can be influenced by the sample prepara-
tion method, that the critical velocity indeed depends on the texture, and that
theoretically calculated estimates [14] of the bulk-liquid flow instability for the
different textures provide reasonable upper or lower bounds for the measured
data. The measurements on vc seem to provide the first experimental tool for
characterizing the global texture, however indirectly. Also the independence of
the observed features on the container exemplifies the fact that in the A phase
vortex formation is a truly intrinsic bulk-liquid process, well separated from the
container wall (unlike even 3 He-B).
The lowest critical velocities in Fig. 6 are recorded when a dipole-unlocked
soliton (cf. Fig. 3) is present in the container and is oriented perpendicular to
the rotation axis. The soliton is a planar wall of width ∼ ξD which separates
bulk liquid in two different, but degenerate minima of the dipole energy, d̂ ↑↑ ˆ
and d̂ ↑↓ , ˆ while within the wall the dipole energy is not minimized. This
explains the reduced critical velocity since in this case a dipole-unlocked region
exists at the cylinder wall where the counterflow velocity is maximized and which
can seed the formation of a dipole-unlocked vortex. The resulting structure, the
intersection of a double-quantum vortex line with a transverse soliton sheet, is an
example of a metastable ˆ field with complicated knot-like continuous topology.
This unstable configuration can be maintained in a long cylinder to moderately
high rotation (Ω  0.5 rad/s) [15].
However, the most surprising case is that when a vertical dipole-unlocked
soliton is present. Here the soliton sheet is oriented approximately parallel to
the rotation axis and connected to the cylinder wall along two dipole-unlocked
410 R. Blaauwgeers et al.

vn = Ω R
v

vs

b
r
0 R

Fig. 8. Large-scale structure of the vortex sheet in its equilibrium configuration in the
rotating container. The vortex sheet is formed when 3 He-A is accelerated to rotation
in an applied magnetic field (H > HD ) and a soliton wall parallel to the rotation
axis exists in the container. If the magnetic field is oriented along Ω, the equilibrium
configuration of the folding is a double spiral. The graph on the top right illustrates the
radial distributions of the normal and superfluid velocities. The latter is discontinuous
across the sheet and differs from that of individual vortex lines in Fig. 1.

the sheet is made up of a alternating linear chain of singly quantized vortices

with circular and hyperbolic ˆ winding such that a fully periodic structure results.
Like in the case of vortex lines, the vortex sheet is translationally invariant in
the direction parallel to the rotation axis Ω. The large-scale structure in the
equilibrium configuration is a continuously meandering foil which is attached at
both ends at two vertical connection lines to the cylindrical wall (Fig. 8). It is
these connection lines where circulation quanta are added to or removed from
the sheet.
The hydrodynamic stability of the vortex sheet was calculated by Landau
and Lifshitz [20]. By considering the equilibrium state with the kinetic energy
from the flow between the folds and the surface tension σ from the soliton
sheet, it is concluded that the distance between the parallel folds has to be
b = (3σ/ρs )1/3 Ω −2/3 . This is somewhat larger than the inter-vortex distance
in a cluster of vortex lines. The areal density of circulation quanta has approx-
imately the solid-body value nv = 2Ω/κ. This means that the length of the
vortex sheet per two circulation quanta is p = κ/(bΩ), which is the periodicity
of the order-parameter structure in Fig. 7. The NMR absorption in the vortex-
sheet satellite measures the total volume of the sheet which is proportional to
1/b ∝ Ω 2/3 . The nonlinear dependence of absorption on rotation is the exper-
imental signature of the vortex sheet, in addition to its low value of critical
velocity.
 Quantized Vorticity in Superfluid 3 He-A: Structure and Dynamics 411

6.9

6.8
NMR absorption [mV]

6.7 a f

6.6

6.5

6.4

6.3
1 1.1 1.2
[rad/s]
Fig. 9. Response of the equilibrium vortex sheet to a closed acceleration – deceleration
cycle in the rotation drive. This measurement is used to define the critical velocity
of the vortex sheet as a function of Ω. During increasing Ω the threshold, where
more circulation is added, is denoted as Ωf , while the threshold of annihilation, where
circulation is removed, is Ωa . The experimental result (thick line) is compared to a
simulation calculation [21] (thin line) on the vortex-sheet configuration which is shown
on the right at Ωa = 1.0 rad/s. In the vortex-sheet meander the circles denote the actual
center positions of each circulation quantum. The measurement has been performed in
a smooth-walled fused quartz cylinder (2R = 3.9 mm). The measuring conditions are
the same as in Figs. 13 and 14.

The vortex sheet is formed whenever a vertical dipole-unlocked soliton sheet

is present in the container and rotation is started. The reason for this is the low
critical velocity at the dipole-unlocked connection lines between the sheet and
the cylinder wall. This facilitates the creation of new circulation quanta, which is
then immediately incorpotated in the structure of the sheet. The critical velocity
of the vortex sheet is well below that of isolated vortex lines. Nevertheless,
when a new circulation quantum is added to the sheet, it experiences repulsion
from the circulation which already resides in the sheet close to the connection
line. Owing to this small Ω-dependent barrier, the critical velocity becomes Ω-
dependent. Experimentally the critical velocity can be defined from Fig. 9 as
vc∗ = (Ωf − Ωa )R, the separation in rotation velocities between the thresholds
where new circulation is formed at (Ωf ) and existing circulation is annihilated
at (Ωa ).
The measured critical velocity in Fig. 10 follows qualitatively the depen-
dence √
vc∗ (Ω) ∝ Ω. This approximate relation illustrates the characteristics of the
vortex sheet: The Magnus force FM = ρs κv from the vortex-free counterflow,
with the velocity v = (Ω − Ωa )R ≈ 2dΩ at the cylinder wall, attempts to pull
412 R. Blaauwgeers et al.

0 .4
Epoxy cell

0 .3
vc (mm/s)

0 .2
*
Q u a rtz c e ll

0 .1
P = 33 bar
T = 0.7 Tc
0
0 1 2 3

Ω (rad/s)
Fig. 10. Measured critical velocity vc∗
of vortex-sheet expansion, when new circulation
is added as a function of the applied rotation Ω. In contrast to the regular periodic
process in vortex-line formation, where vc is a constant as a function of Ω, here vc∗ (Ω) ∝
√
Ω. The measurements represent the situation when a vertical soliton wall is already
present in the cylinder and the vortex sheet has been adiabatically grown in an axially
oriented
√ magnetic field to its equilibrium configuration. The solid curves are fits with
Ω dependence. The rougher epoxy wall has a larger magnitude of critical velocity,
perhaps because of pinning of the connection lines which resists smooth readjustment
in the folding of the vortex sheet as a function of Ω. The different symbols of data
points (quartz cell) illustrate the reproducibility of the results from one adiabatically
grown vortex sheet to another.

additional circulation into the sheet. It is opposed by the repulsion from the
circulation, which already resides in the sheet closest to the connection line,
at a distance equal to the width d of the circulation-free annulus. The balance
2
between the repulsion Fr = (ρ√ s κ /d) ln [d/(αξD )] and FM gives a critical coun-
terflow velocity of order vc ∼ 2κΩ.
Actually, because of the presence of the two connection lines, a vortex-sheet
state does not have full axial symmetry in the distribution of its vortex-free
counterflow even in the double-spiral configuration of Fig. 9 (ie. at the connec-
tion line the distance of the first circulation quantum from the wall is different
from that in the outermost circular spiral part). This causes a small difference
in the measured vc∗ = (Ωf − Ωa )R from the real value of vc at the connection
lines. The two can be connected if the sheet configuration is known. A qualita-
tive difference between these two critical velocities is that the true value of vc
does not approach zero when Ω → 0, but remains finite. This is expected since
some counterflow velocity is required when new circulation is formed even at a
dipole unlocked connection line of a circulation-free soliton sheet, since also the
 Quantized Vorticity in Superfluid 3 He-A: Structure and Dynamics 413

attractive interaction of the emerging vortex line with its image forces within
the wall have to be overcome.

5 Dynamic Response
So far we have looked at the response of the 3 He-A order-parameter field to what
are essentially adiabatic changes in the rotation velocity. But what happens if the
rotation is changing rapidly with time? Because of its lower critical velocity, it is
found that the vortex sheet becomes the preferred structure in rapidly changing
rotation, rather than individual double-quantum vortex lines. Consequently in
3
He-A, the response to high-frequency perturbations in the rotation drive is
explained by the dynamic properties of the vortex sheet. The central feature
becomes the interplay between the large-scale configuration of the sheet and its
confined circulation. The first illustration of these considerations is provided by
the measurements in Fig. 11.
In the experiment of Fig. 11 the complete NMR absorption spectrum is slowly
recorded while the rotation is harmonically swinging back and forth around zero
with a short period of 50 s. At low amplitude of back-and-forth rotation the
circulation enters in the form of double-quantum vortex lines (upper spectrum).
The lines are periodically formed and then annihilated during both the positive
and negative half cycles of rotation. During the positive and negative half cycles
the vorticity has opposite signs of circulation: All vortex lines are annihilated
approximately by the time when rotation goes through zero. Therefore the en-
velope of the vortex-line satellite is modulated at twice the frequency of the
rotation drive. To record the spectrum the rate of the NMR field sweep has to
be much slower than the period of the rotation drive: The horizontal scale, which
here is plotted in terms of the linearly changing NMR field sweep, is actually
the common time axis for all variables.
A rarely seen feature of the upper spectrum is the modulation in the ab-
sorption on the high-field flank of the large bulk-liquid peak. This absorption
component is created by the slight dipole unlocking in the vicinity of the surface
layer on the cylinder wall when the counterflow velocity gets sufficiently large.
The modulation of this absorption is approximately in anti-phase with that in
the vortex-line satellite, ie. minimum absorption in the vortex-satellite is reached
approximately at the same time when absorption in the counterflow signal is at
maximum. Since the total area in the absorption spectrum must be constant as a
function of time, this means that absorption is transferred periodically between
the different parts of the spectrum.
The lower spectrum in Fig. 11 is recorded with a larger amplitude of back-
and-forth rotation. In this case the circulation enters in the form of a vortex
sheet, which now provides a more stable response than double-quantum vortex
lines. (During the first few half cycles the response may be in the form of lines,
but it soon goes spontaneously over into the vortex sheet.) The modulation of the
absorption in the vortex-sheet satellite is similar to that of the double-quantum
vortex peak. The major change is the absence of the counterflow signal, since
414 R. Blaauwgeers et al.

100

Periodic rotation drive

Δ Ω
80

− Δ Ω

50 s
NMR absorption (mV)

60

40

Counterflow Splay soliton

and and
vortex lines vortex sheet
(,9 = 0.40 rad/s) (,9 = 0.50 rad/s)

20

9 9.1 9.2 9.3 9.4 9.5

H (mT)
Fig. 11. NMR absorption spectra in back-and-forth rotation. The measurement is
performed by sweeping slowly the polarizing magnetic field H at constant rf excitation
frequency (f = 310 kHz) across the NMR region while the sample is in sinusoidal back-
and-forth rotation: Ω(t) = ΔΩ sin ωt. The field sweep is much slower than the period of
the rotation drive (2π/ω = 50 s). Thus the rotation appears as a modulation envelope
on the field-dependent NMR absorption. 1) When the amplitude ΔΩ of the sinusoidal
rotation is sufficiently small (vD /R  ΔΩ = 0.40 rad/s) the circulation enters in
the form of double-quantum vortex lines (upper spectrum). Here the absorption is
transferred periodically between the vortex and counterflow satellites (spectrum CUV
+ CF in Fig. 3). 2) When the rotation amplitude is increased (ΔΩ = 0.50 rad/s) the
circulation goes into the vortex sheet (lower spectrum). Now the absorption is shifted
between the vortex-sheet and soliton satellites (spectra labeled as VS and SS in Fig. 3).
(Measuring conditions: 2R = 3.87 mm, H  Ω, P = 34 bar, T = 0.7 Tc )
416 R. Blaauwgeers et al.

180

0.6 rad/s
150 1.2 rad/s
1.8 rad/s 0.2
2.4 rad/s

Amplitude (mV)
120
Phase ( )

90 P = 34 bar 0.1
T = 0.7 Tc
60 H = 9.56 mT
H ||
30 0
0 0.05 0.1 0 0.05 0.1 0.15
Modulation frequency [Hz]
Fig. 13. Dynamic response of the equilibrium vortex sheet to harmonic modulation
of the rotation drive: Ω(t) = Ω0 + ΔΩ sin ωt. The left panel shows the phase of the
modulated absorption component in the peak height of the vortex-sheet satellite, rel-
ative to the rotation drive (in degrees). The right panel gives the amplitude of the
modulated absorption (peak to peak), referred to in mV at the output of the cryogenic
preamplifier operated at LHe temperature [6]. The solid curves are guides to the eye.
The measuring conditions are the same as in Fig. 11 with ΔΩ = 0.050 rad/s. (From
Ref. [24])

sheet is not that of the equilibrium double spiral but one where multiple pieces
of sheet exist which all are attached separately along two contact lines to the
vertical cylinder wall [5].
A straightforward continuation of the measurements in Fig. 12 is to monitor
the satellite peak height as a function of the modulation frequency, to find the
dynamic response to changes in the rotation drive. In the case of double-quantum
vortex lines, the satellite intensity (or peak height at constant line width) mon-
itors in first approximation the number of lines. In this situation only the fast
radial motion of vortex lines can be measured. This is performed by stopping
rotation very abruptly and by measuring the time dependence of vortex-line an-
nihilation. The tail of this signal is given by N (t) = N (0)(1 + t/τF )−1 , where
N (0) is the number of vortex lines in the container at the moment t = 0, when
rotation has come to a stop (Ω(t) = 0). Such measurements give the charac-
teristic decay time τF = (νκnv (0) ρn B/ρ)−1 , where νκnv (0) is the density of
circulation quanta at t = 0 and ρn B/ρ is the dissipative mutual friction. This
mode of vortex-line motion has been monitored in the 3 He superfluids for double-
quantum vortices in A phase [22] and for singular-core vortices in B phase [7].
Both measurements are consistent with other determinations of mutual friction
[23]. Other resonance techniques can be used to record also the slow approach
of vortex lines to equilibrium at constant Ω, after an initial disturbance has
been switched off. This motion is predominantly azimuthal in character for an
 Quantized Vorticity in Superfluid 3 He-A: Structure and Dynamics 417

Time [s]
0 2 0 4 0 6 0 8 0 1 0 0 1 1 0

12.4 = 2.4 [rad/s] | | - vc / R = 0.03 rad/s

= 1.9 s = 2.6 s
12.2
NMR absorption [mV]

= 0.6 [rad/s] | | - vc / R = 0.11 rad/s

4.4
= 4.1 s = 5.3 s
4.2
6

4
4.0 [s]
2

3.8 0
Ω Ω 1 2
[rad/s]
Fig. 14. Response of the equilibrium vortex sheet to a step change in the rotation
drive. On the left Ω is reduced by ΔΩ = 0.15 rad/s and on the right it is increased
back to the original value. The solid curves are exponential fits. The corresponding
time constants are plotted in the inset, with () for downward and () for upward
step changes. (From Ref. [24])

isolated vortex cluster and in the 3 He superfluids it is over damped, ie. has pure
exponential time dependence [7].
In the vortex sheet the circulation is topologically confined within the sheet
where it can move in two ways: either along the sheet or by forcing the entire
sheet to contract or expand. Both types of motion proceed only via changes in
the length and folding of the sheet. A slow final readjustment in the configuration
of the sheet follows these faster initial responses, if the texture is left to anneal
at constant Ω. The central feature is thus the interplay between the large-scale
configuration of the sheet and its confined circulation. This aspect is illustrated
by the measurements in Figs. 13 and 14.
The satellite intensity of the vortex sheet is proportional to the total volume
of the sheet. The circulation is distributed as a continuous periodic chain along
the sheet where the total number of quanta N is given by the length L of the
meander in the transverse plane: N = L/(p/2) = 2bLΩ/κ. Thus the satellite
signal has to be proportional to
1  ρ 1/3
s
L = κN , (4)
2 3σΩ
which means that it displays a dependence on Ω even at constant N , as evident
in Fig. 9. Therefore, also the readjustment to small disturbances in Ω at constant
circulation can be measured for the vortex sheet with this technique [24].
418 R. Blaauwgeers et al.

The response of the peak height of the vortex-sheet satellite is shown in

Fig. 13 for harmonic modulation of the rotation drive and in Fig. 14 for a step
change. In both cases the change by ΔΩ about the average rotation Ω0 is kept as
small as possible, to maintain N constant. However, because of finite measuring
resolution the condition 2ΔΩ < Ωf − Ωa (cf. Fig. 9) is valid only at large Ω0 in
the measurement of the harmonic response. In the step-response measurement
it is not satisfied even at the highest Ω0 value.
Let us first look at the characteristics in the limit of large Ω0 . The left panel
in Fig. 13 shows that the phase shift between the response and the drive is
then approaching 180◦ . The right panel shows that this response occurs at small
amplitude. The out-of-phase behaviour at constant N is expected from Eq. (4):
To maintain solid-body-rotation, the vortex sheet contracts during increasing
rotation and expands during decreasing Ω.
This out-of-phase signal at constant N identifies the origin of the sharp out-
of-phase spike in the step response at large Ω0 (upper trace in Fig. 14). The
spike is then followed by a slow exponential in-phase response which has to be
associated with the change in N . The amplitude of this component is small at
large Ω0 (upper trace) but grows much larger at low Ω0 where the change in
N becomes the dominant effect. The time constant of the exponential signal is
plotted as a function of Ω0 in the inset at bottom right in Fig. 14. In the harmonic
response this means that with decreasing Ω0 the phase shift is reduced and the
amplitude grows.
The time sequence of different processes in the dynamic response of the equi-
librium vortex sheet with double-spiral configuration is thus the following: (i)
The fastest response with a time constant of order 1 s occurs in the density of
the circulation quanta. In the axial field with circularly spiraling folding this
can happen only perpendicular to the sheet, ie. the sheet either contracts or
expands at constant N . (ii) The adjustment of the number of circulation quanta
to the equilibrium value occurs slower on a time scale of order 10 s. This requires
the motion of the circulation along the sheet in unison and the corresponding
readjustment in the length of the sheet. (iii) The slowest component is the an-
nealing of the vortex sheet at constant Ω which takes place on a time scale of
minutes. In the present range of high and intermediate temperatures all motions
are exponentially over damped – a generic property of 3 He superfluids where the
kinematic viscosity of the normal component is so large that it can be considered
to be clamped to corotation with the container.
More recent measurements demonstrate that the response of 3 He-A to a ro-
tation drive, which includes components at high frequency and large amplitude,
is dominated by the vortex sheet and its properties. Again the basic features
are here derived from structural considerations: The large-scale configuration of
the vortex sheet is altered substantially from the equilibrium state such that
the dynamic response becomes greatly enhanced [5]. This leads to considerable
gains in the kinetic energy under rapidly changing flow conditions which readily
compensate the small increases in textural energies.
 Quantized Vorticity in Superfluid 3 He-A: Structure and Dynamics 419

6 Summary and Future Work

The hydrodynamics of 3 He-A will undoubtedly be a source for more surprises.
It is well known that with heat flow superfluid – normal fluid thermal counter
currents can be produced which drive dissipative time-dependent ˆ textures [25].
In rotational flow a large number of different structures of quantized vortic-
ity have been identified. These make it possible for the superfluid to mimic
optimally solid-body rotation, depending on temperature, magnetic field, the
oder-parameter texture, and the properties of the rotation drive.
NMR has been the most efficient method for distinguishing between differ-
ent vortex textures, by providing the possibility to probe the order-parameter
texture from outside the rotating container. The method is well suited for the
measurement of a vortex cluster which is translationally invariant in the direc-
tion parallel to the rotation axis. However, both the vortex texture and the NMR
spectrum change if the orientation of the polarization field is rotated. Therefore
the NMR method is less suited for precise measurements of vortex states which
are less regular in configuration than rotating vortex clusters. An exception are
random tangles of vorticity where the average density of dipole-unlocked vortic-
ity can be continuously monitored.
Uniform rotation is not generally suited for the study of turbulence. Suf-
ficiently long-lived turbulent states, which could efficiently be recorded with
present continuous-wave NMR techniques, have not been produced by rotating
3
He-A at temperatures above 0.5 Tc . At lower temperatures the normal fluid frac-
tion is rapidly depleted and the hydrodynamic response time of the superfluid
fraction increases, owing to the reduced coupling with the walls of the rotating
container. If the A phase can be maintained in metastable state in low magnetic
fields to these low temperatures [26], then it might become possible to gener-
ate transient, but long-lived turbulent states by rotation techniques. For such
studies present measuring techniques would be adequate. We might thus expect
to see new development in the hydrodynamics of 3 He-A which would illuminate
the properties of an anisotropic superfluid in the zero temperature limit.

Acknowledgements
This work was funded in part by the EU – Improving Human Potential – Access
to Research Infrastructures programme under contract EC HPRI-CT-1999-50.

References
1. V.B. Eltsov, M. Krusius, G.E. Volovik in: Proc. Low Temp. Phys., Vol.
XV, ed. W.P. Halperin (Elsevier Sciece B.V. Amsterdam, 2001); preprint
([Link]
2. S.N. Fisher, A.J. Hale, A.M. Guénault, G.R. Pickett: Phys. Rev. Lett. 86, 244
(2001)
3. H.M. Bozler in: Helium three, eds. W.P. Halperin, L.P. Pitaevskii (Elsevier Sciece
B.V. Amsterdam, 1990)
420 R. Blaauwgeers et al.

4. V.M. Ruutu, J. Kopu, M. Krusius, Ü. Parts, B. Plaçais, E.V. Thuneberg, W. Xu:
Phys. Rev. Lett. 79, 5058 (1997)
5. V.B. Eltsov, R. Blaauwgeers, N.B. Kopnin, M. Krusius, J.J. Ruohio, R. Schanen,
E.V. Thuneberg: to be published
6. Ü. Parts, V.M. Ruutu, J.H. Koivuniemi, Yu.M. Bunkov, V.V. Dmitriev, M. Fo-
gelström, M. Huebner, Y. Kondo, N.B. Kopnin, J.S. Korhonen, M. Krusius, O.V.
Lounasmaa, P.I. Soininen, G.E. Volovik: Europhys. Lett. 31, 449 (1995); V.M.
Ruutu, Ü. Parts, J.H. Koivuniemi, N.B. Kopnin, M. Krusius, J. Low Temp. Phys.
107, 93 (1997)
7. E.B. Sonin, Y. Kondo, J.S. Korhonen, M. Krusius: Europhys. Lett. 22, 125 (1993);
Phys. Rev. B 47, 15113 (1993-II)
8. D. Vollhardt, P. Wölfle: The superfluid phases of helium 3 (Taylor & Francis,
London, 1990)
9. V.B. Eltsov, M. Krusius in: Topological defects in 3 He superfluids and the non-
equilibrium dynamics of symmetry-breaking phase transitions, eds. Yu. Bunkov
and H. Godfrin (Kluwer Academic Publ., Dordrecht, 2000), p. 325
10. N.D. Mermin, T.L. Ho: Phys. Rev. Lett. 36, 594 (1976); N.D. Mermin: Phys.
Today 34, 46 (1981)
11. V.M. Ruutu, Ü. Parts, M. Krusius: J. Low Temp. Phys. 103, 331 (1996)
12. V.M. Ruutu, J.J. Ruohio, M. Krusius, B. Plaçais, E.B. Sonin, Wen Xu: Phys. Rev.
B 56, 14089 (1997-I); Physica B 255, 27 (1998)
13. R. Blaauwgeers, V.B. Eltsov, M. Krusius, J.J. Ruohio, R. Schanen, G.E. Volovik:
Nature 404, 471 (2000)
14. R. Kopu, R. Hänninen, E.V. Thuneberg: Phys. Rev. B 62, 1 Nov (2000)
15. V.M.H. Ruutu, Ü. Parts, J.H. Koivuniemi, M. Krusius, E.V. Thuneberg, G.E.
Volovik: Pis’ma v ZhETF 60, 659 (1994) [JETP. Lett. 60, (1994)
16. P.G. Saffman: Vortex dynamics (Cambridge Univ. Press, Cambridge UK, 1992)
17. For the early history on quantized vorticity in 4 He-II see R.J. Donnelly: Quantized
vortices in Helium II (Cambridge Univ. Press, Cambridge UK, 1991)
18. Ü. Parts, E.V. Thuneberg, G.E. Volovik, J.H. Koivuniemi, V.H. Ruutu, M. Heinilä,
J.M. Karimäki, M. Krusius: Phys. Rev. Lett 72, 3839 (1994); E.V. Thuneberg:
Physica B 210, 287 (1995); M.T. Heinilä, G.E. Volovik: Physica B 210, 300
(1995); Ü. Parts, V.M. Ruutu, J.H. Koivuniemi, M. Krusius, E.V. Thuneberg,
G.E. Volovik: Physica B 210, 311 (1995)
19. M. Sigrist: Physica B, 280, 154 (2000) and references therein; T. Kita: Phys. Rev.
Lett. 83, 1846 (1999)
20. L. Landau, E. Lifshitz: Dokl. Akad. Nauk. 100, 669 (1955)
21. V.B. Eltsov: J. Low Temp. Phys. 121, 387 (2000)
22. P.J. Hakonen, V.P. Mineev: J. Low Temp. Phys. 67, 313 (1987)
23. T.D.C. Bevan, A.J. Manninen, J.B. Cook, H. Alles, J.R. Hook, H.E. Hall: J. Low
Temp. Phys. 109, 423 (1997); Phys. Rev. Lett. 77, 5086 (1996); ibid. 74, 750
(1996)
24. V.B. Eltsov, R. Blaauwgeers, M. Krusius, J.J. Ruohio, R. Schanen: Physica B
284–288, 252 (2000)
25. D.N. Paulson, M. Krusius, J.C. Wheatley: Phys. Rev. Lett. 37, 599 (1976); H.E.
Hall, J.R. Hook: The hydrodynamics of superfluid 3 He, in: Prog. Low Temp. Phys.,
Vol. IX, ed. D.F. Brewer (Elsevier Sciece B.V. Amsterdam, 1985), p. 143
26. P. Schiffer, M.T. O’Keefe, M.D. Hildreth, H. Fukuyama, D.D. Osheroff: Phys. Rev.
Lett. 69, 120 (1992); ibid. 69, 3096 (1992); Prog. Low Temp. Phys., Vol. XIV, ed.
W.P. Halperin (Elsevier Sciece B.V. Amsterdam, 1995), p. 159
Vortices in Metastable 4 He Films

Ralf Blossey

Universität Essen, Fachbereich Physik, D-45117 Essen, Germany

1 Wetting Properties of 4 He
on Weak-Binding Alkali Metals
1.1 Wetting Transitions of Liquid Helium

Liquid helium wets almost all surfaces. This behaviour results from its small
dielectric permittivity, ε = 1.057, which leads to attractive forces between the
helium atoms weaker than those between helium and other atoms [1]. In 1991
Cheng et al., however, predicted that 4 He does not wet certain alkali metals. In-
stead, for some of these metallic substrates, a first-order wetting transition was
foreseen [2]. In the alkali metals, weakly bound outer electrons cause a short-
range repulsion for the helium atoms. The resulting temperature-dependent bal-
ance between repulsive short-range and attractive long-range forces is at the
origin of the wetting phenomenon in liquid helium.
Shortly after the theoretical prediction the non-wetting behaviour of 4 He
on Cs at ultralow temperatures was experimentally confirmed for 4 He/Cs ([3],
for a review [4]). Both the wetting transition at Tw ≈ 1.9 K, and first-order
prewetting transitions were observed, accompanied by strong hysteresis effects
upon undercooling [5] (termed ‘anomalous nucleation’, [6]). Subsequently, the
wetting behavior of 4 He on Rb has been investigated ([7], [8] and references
therein). The findings are still somewhat controversial; e.g., it is not entirely
clear whether there is a wetting phase transition of 4 He on (nominally pure) Rb
at coexistence; if it exists, the wetting temperature is very low (Tw ≤ 300 mK).
Further, the superfluid transition in the 4 He-film on Rb appears to deviate in
several respects from the standard Kosterlitz-Thouless transition scenario [8].
Figure 1 shows a schematic phase diagram of the system 4 He/Cs in temper-
ature T and chemical potential difference Δμ. The first-order wetting transition
occurs at coexistence of bulk liquid and gas phases, Δμ = 0, at T = Tw . At
T = Tλ occurs the superfluid transition. The off-coexistence continuation of the
first-order transition is the prewetting line, Δμp (T ), ending in the prewetting
critical point at Δμp (Tp ). The superfluid transition in a film of finite thickness
occurs at the Kosterlitz-Thouless line Δμλ (T ).

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 421–427, 2001.

c Springer-Verlag Berlin Heidelberg 2001
422 R. Blossey

Δμ
• •
Tw Tλ T
 •
Δμλ (T )

Δμls (T ) Δμp (T )

Δμp (Tp )
•

Fig. 1. Equilibrium wetting phase diagram of 4 He/Cs in the parameters temperature

T and chemical potential difference Δμ between the bulk liquid and gas phase. Further:
lower surface spinodal Δμls (dotted); quench line (arrow) (see Secs.2/3)

1.2 Interface Model Description of Wetting Transitions

A convenient description of both equilibrium and non-equilibrium aspects of
wetting transitions is based on the effective interface Hamiltonian [9]

√
H0 [f ] = d 2 x [γ g + V (f ) − Δμf ] (1)

where f (x) is the local height above the (planar) substrate x  R 2 . The first term
√
γ g = γ(1 + (∇f )2 ) 1/2 is the capillary contribution of the liquid-vapor inter-
face with surface tension γLV = γ. Typically, a squared-gradient approximation
√
to g ≈ 1 + (1/2)(∇f )2 + ... suffices. The second term in (1) is an effective
interface potential V (f ) resulting from effective, i.e. temperature-dependent, in-
teractions between the liquid-vapor interface and the solid. For the case of a
first-order wetting transition between a microscopic film of thickness f = f0
and a macroscopic film with f = ∞, V (f ) generally has a strongly asymmetric
double-well structure, as shown in Figure 2. For long-ranged molecular interac-
tions the asymptotic decay of V (f ) for f → ∞ is given by V (f ) = A/f 2 where
A is the Hamaker constant, determined by the dielectric properties of the three
media involved (solid, liquid, gas) [1]. Finally, the last term in eq.(1) is the chem-
ical potential contribution with Δμ ≡ μ − μc , where μc is the chemical potential
at coexistence of the bulk gas and liquid phases.
The average thickness f¯ of a wetting layer on a wall is a convenient order pa-
rameter for a discussion of equilibrium wetting states and the transitions between
 Vortices in Metastable 4 He Films 423

Fig. 2. Effective interface potential V (f ) for a first-order wetting transition

them. It fulfills the equation of state obtained from the variation δH/δf |f =f¯ = 0,
V  (f )|f =f¯ = Δμ , where V  ≡ dV /df . For homogeneous film states, the discus-
sion of the minima of V allows the construction of the wetting part of the phase
diagram of Figure 1.

2 Lifetime of an Undercooled Film

A metastable thick wetting layer of 4 He on Cs can be created by undercooling a
film from above the prewetting line into the nucleation region located between
μp (T ) and a lower spinodal line μls (T ), whereby the location of the lower spinodal
is determined by the conditions V  (f ) = Δμ and V  (f ) = 0, where f = f¯ is
the thickness of the undercooled thick film. The undercooled layer resides on
the substrate until, by thermal fluctuations, a hole of critical size (or excess
free energy) has formed. The shape of the critical nucleus is determined by the
variational equation to (1), δH0 [f ]/δf = 0, assuming cylindrical symmetry, i.e.
f = f (r),  
1 
γ f (r) + f (r) = V  (f ) − Δμ

(2)
r
where the prime denotes differentiation with respect to the argument. The term
on the left-hand side of the equation corresponds to the curvature-contribution
of the hole profile. A critical hole obeys the boundary conditions

f (∞) = f¯ ≡ F , f  (0) = 0 . (3)

424 R. Blossey

The solution to eqs.(2) and (3) determines the excess free energy of the critical
hole via
Ec (T, Δμ) ≡ H0 [f (r)] − H0 [F ] . (4)
Within a Fokker-Planck description of the nucleation process [10] the excess free
energy determines the probability for the formation of a critical hole in a wetting
film via the Arrhenius-type factor

Γ ∝ exp −[Ec /(kB T )] (5)

where kB is Boltzmann’s constant, and T is the temperature. The lifetime τ of

the film is then given by τ ∼ Γ −1 .
The experimentally observed anomalously long lifetime of an undercooled he-
lium film can be explained by the properties of critical holes at deep temperature
quenches close to coexistence. When Δμ → −0 for T  Tw , F and consequently
the critical depth Fc ≡ F − f0 diverges. In this regime the critical hole profile at
distances far from the wall, f  f0 , is determined by the asymptotic behaviour
of V (f ) for f → ∞. The critical hole then has a funnel-like shape with a finite
inner radius Rc [11]. With Δμ → −0 one has Rc = const., while the central
depth Fc of the funnel-shaped hole diverges according to
1
Fc ∼ |Δμ| − 3 , (6)

and the excess free energy scales as

Ec ∼ Fc2 ∼ F 2 . (7)

If the quench is performed at a finite chemical potential difference and ap-

proaches the lower spinodal, the excess free energy of the critical hole vanishes
according to [12]
Ec ∼ T − Tls . (8)

3 Application to 4 He/Cs
The exponential dependence of the nucleation rate Γ on Ec stands in the way of
an experimental verification of the scaling results: the observation times t > τ
are very long close to coexistence Δμ → −0, while upon the approach to the
spinodal, hole nucleation becomes an exponentially fast process. Recently, it
has been proposed that for a superfluid helium film a controlled deformation of
the liquid-vapor interface (‘surface dimple’) can be a suitable candidate for a
reduction of the nucleation barrier [13]. The presence of a surface deformation
affects the nucleation kinetics: the nucleation of a hole can now occur from
an already existing ‘hole’-like excitation of the interface. This can lead to a
reduction of the excess free energy barrier of the critical nucleus so that, to a
first approximation, the excess free energy is modified by replacing Ec by Ec − Ed ,
where Ed denotes the excess surface free energy of the surface dimple.
 Vortices in Metastable 4 He Films 425

For concreteness, we compare two temperature quenches from high to low

temperatures. Consider Δμ < 0 fixed. Quench 1 starts at a temperature slightly
above the KT-line, i.e. from a normal 4 He film. Quench 2 starts from a tempera-
ture slightly below the KT-line, hence from a superfluid film. Both quenches end
at the same temperature Tf  Tw . Since both initial states can be very close to
each other, the probability for homogeneous nucleation of a critical hole differs
only little.
When the hole is nucleated from a superfluid film, however, the excess free
energy required for the formation of a critical hole will be different from that
of a normal fluid due to the presence of vortices with the associated Bernoulli
pressure [14] 
−R0 /r2 for r > ξ
P(r) = (9)
−(R0 /ξ 2 )(2 − (r/ξ)2 ) for r < ξ
where R0 = s κ2c /8π 2 with the mass density s of the superfluid and κc as
the circulation quantum. The Bernoulli
 pressure modifies the effective interface
Hamiltonian H[f ] = H0 [f ] + 2π dr rP(r)f . By comparison of terms with H0
we see that the Bernoulli pressure effectively acts as a local change in chemical
potential.
We can now determine the shape of the vortex-caused surface deformation
in the same manner as that of the critical hole. We first focus on a ‘thick’
but finite film. The interface potential V (f ) − Δμf for the thick film can be
approximated by a parabola centered at the equilibrium thickness f = F . Using
F as a reference level by introducing ζ ≡ f − F , the surface deformation of the
interface is determined by the harmonic functional
 ∞
H[ζ] = 2π drr[(γ/2)(∇ζ)2 + φ(ζ) − P(r)ζ] (10)
0


with φ(ζ) ≡ (V (F )/2)ζ = (3A/F 4 )ζ 2 . The depth of the dimple is found to
2

behave according to [13,14]

ν
Fd ≡ |ζ(0)| ≈ [ln(2a/ξ)]2 , (11)
2
with the capillary length a = (γ/V  (F )) 1/2 . For the dimple radius one has

Rd ≡ |r(ζ = ζ(0)/2)| ≈ ξ exp[(Fd /ν)1/2 ] , (12)

while for the dimple free energy one has from eq.(11)

Ed ≡ −H[ζ] = γν 2 I0 ∼ Fd2 , (13)

where I0 is of the order O(ln(a/ξ)). H[ζ] is negative, expressing the fact that the
thick film lowers its free energy by forming the dimple. Table I lists the results
for the capillary length, the dimple depth, radius and free energy for 4 He/Cs
at two film thicknesses, F = 100 Å and F = 200 Åwith A = 10K/kB . The
magnitude of Fd for 4 He is fairly small, while the radius of the dimple is already
426 R. Blossey

Table 1. Dimple properties

F (Å) a (Å) Fd (Å) Rd (Å) Ed (K)

100 670 12.28 149.98 24.46

200 2680 17.67 405.49 41.59

considerable: for F = 200 Å, Rd ≈ 2F . These results indicate that the vortex
dimple is, at least in the 4 He/Cs system, indeed a weak perturbation for a thick
film F → ∞ for Δμ → −0. However, when the size of the dimple matches that of
the critical hole, which occurs upon the approach to the spinodal, the difference
in surface free energies ΔE ≡ Ec − Ed shrinks. It vanishes according to the power
law [15]
ΔE ∼ (Tf − T∗ ) 1/2 (14)
where T∗ (Δμ) is the temperature, defined as the point beyond which surface
deformations have become unstable. At a given value of Δμ, T∗ > T , so that T∗
corresponds to a shifted spinodal temperature [13].

4 Conclusions
The consideration of a nucleation process in the presence of localized surface de-
formations might open up the possibility of performing quantitative experiments
on the nucleation of holes in undercooled wetting layers. A comparison of the de-
cay of metastable normal liquid and superfluid layers could therefore allow a test
of the predictions of the scaling theory of critical holes. The scenario described
here is in addition affected by the number n of circulation quanta present in the
vortex, since Ed ∼ R02 ∼ (nκc )4 . Further modifications of the dimple-assisted
dewetting scenario are: 1) the thinning of the superfluid film due to the Kon-
torovich effect [16]. Superfluid flow of the film modifies the asymptotics of the
effective interface potential via V (f ) = A/f 2 − s j 2 /(2f ) where j = vc f is the
critical current. For typical critical velocities vc ≈ 0.6 m/s the Kontorovich effect
will be appreciable for film thicknesses F > 200 Å[15]. 2) P.G. de Gennes has
recently pointed out that on an extremely weak-binding surface such as Rb the
balance between Bernoulli pressure and the boundary tension of 4 He ‘pancakes’
may give rise to unusually large vortex cores [17]. 3) The disordered surfaces
that are common to Cs and Rb substrates might affect the presented scenario
by introducing disorder-dominated hysteresis effects [18]. They are however less
relevant for dewetting than for wetting processes, since in the former case their
difference is clearly discernable [18].

Acknowledgments
Support by the DFG under the Leibniz grant (Di387/2-1) and the Schwerpunkt-
sprogramm ’Wetting and Structure formation at Interfaces’ (Bl356/2-1) is grate-
fully acknowledged.
 Vortices in Metastable 4 He Films 427

References
1. Israelachvili J. (1992) Intermolecular and surface forces. 2nd ed., Academic Press,
London
2. Cheng E., Cole M.W. et al. (1991) Helium prewetting and nonwetting on weak-
binding substrates. Phys. Rev. Lett. 67, 1007-1010
3. Nacher P.-J., Dupont-Roc J. (1991) Exprimental evidence for nonwetting with
superfluid helium. Phys. Rev. Lett. 67, 2966-2969
4. Cheng E., Cole M.W. et al. (1993) Novel wetting behaviour in quantum films.
Rev. Mod. Phys. 65, 557-567
5. Rutledge J.E., Taborek P. (1992) Prewetting phase diagram of 4 He on cesium.
Phys. Rev. Lett. 69, 937-940
6. Schick M., Taborek P. (1992) Anomalous nucleation at first-order wetting transi-
tions. Phys. Rev. B 46, 7312-7314
7. Wyatt A.F.G., Klier J., Stefanyi P. (1995) Prewetting of 4 He on Rb: coexistence
of two superfluid films? Phys. Rev. Lett. 74, 1151-1154
8. Philips J.A., Ross D. et al. (1998) Superfluid onset and prewetting of 4 He on
rubidium. Phys. Rev. B 58, 3361-3370
9. Brezin E., Halperin B.I., Leibler S. (1983) Critical wetting: the domain of validity
of mean field theory. J. Phys. (Paris) 44, 775-783
10. Blossey R. (1995) Nucleation at first-order wetting transitions. Int. J. Mod. Phys.
9, 3489-3525
11. Bausch R., Blossey R. (1994) Lifetime of undercooled wetting layers. Phys. Rev.
E 50, R1759-R1761
12. Bonn D., Indekeu J.O. (1995) Nucleation and wetting near surface spinodals. Phys.
Rev. Lett. 74, 3844-3847
13. Blossey R. (1998) Dimple-assisted dewetting in rotating superfluid films. Phys.
Rev. B 57, R14048-R14051
14. Harvey K.C., Fetter A.L. (1973) Free surface of a rotating superfluid. J. Low.
Temp. Phys. 11, 473-481
15. Blossey R. (2000) Dimples and Dents. Heterogeneous Nucleation in Undercooled
Wetting Layers. Habilitation Thesis Heinrich-Heine-Universität Düsseldorf
16. Kontorovich V.M. (1956) Effect of the rate of flow of a He II film on its thickness.
Sov. Phys. JETP 3, 770-771
17. de Gennes P.G. (1999) Dry vortices in thin helium films. C.R. Acad. Sci. Paris
t.327 II, 1337-1343
18. Blossey R., Kinoshita T., Dupont-Roc J. (1998) Random-field Ising model for the
hysteresis of the prewetting transition on a disordered substrate. Physica A 248,
247-272
Quantum Hall Effect Breakdown Steps
and Possible Analogies with Classical
and Superfluid Hydrodynamics

L. Eaves

School of Physics & Astronomy, University of Nottingham,

Nottingham NG7 2RD, UK

Abstract. The breakdown of the integer quantum Hall effect at high currents some-
times occurs as a series of regular steps in the dissipative voltage drop measured along
the Hall bar. The steps were first seen clearly in two Hall bars used to maintain the
US Resistance Standard, but have also been reported in other devices. This paper
discusses the origin of the steps in terms of an instability in the dissipationless flow
induced by inter-Landau level tunnelling processes in local microscopic regions of the
Hall bar. It is proposed that electron-hole pairs are generated in the quantum Hall fluid
in these regions and that the electronic motion can be envisaged as a quantum analogue
of the von Karman vortex street which forms when a classical fluid flows past an ob-
stacle. A possible analogue with vortex formation in superfluids is also discussed briefly.

Keywords: Quantum Hall effect, Tunnelling, Hydrodynamics, Quantum Vortices

1 Introduction
The relation between the quantum Hall effect (QHE) [1,2] and superfluid hydro-
dynamics appears a tenuous one at first sight. Yet in 1985 Muirhead, Vinen and
Donnelly[3] pointed out an interesting analogy between the motion of a vortex
in a superfluid film and electron motion in two dimensions under the action of
crossed electric and magnetic fields. In the analogy, the vortex circulation cor-
responds to the electron charge, the Magnus force to the Lorentz force and the
superfluid density to the strength of the magnetic field. These authors then went
on to consider the quantum description of the vortex motion and how vortices
can be generated by tunnelling from the edge of the film.
Another possible analogy for the two systems is the small value of kinematic
viscosity of He II and the low level of resistive dissipation when current flows in
a semiconducting Hall bar sample under ideal quantum Hall effect conditions.
The purpose of this article is two-fold: first, to propose a model based on
inter-Landau level tunnelling to explain an unusual step-like breakdown of the
dissipationless state in the quantum Hall effect and, second, to draw attention
to possibly interesting parallels between this model and well-established ideas of
vortex pair formation in superfluids. When considering the analogies we should
be aware that the electrons which make up the quantum Hall fluid are moving
in an electrostatic potential which we can consider as comprising two parts:
the first part arises from the presence of the Hall electric field due to the edge

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 428–437, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Quantum Hall Effect Breakdown Steps 429

charges, and the second part arises from the randomly-located remote donor
ions. It is well-established that this random potential plays an important role in
the QHE [2]. The parallel situation for He II wou ld be flow in the presence of
small obstacles, e.g. charged ions or possibly 3 He atoms.
In the integer QHE [1], a two-dimensional electron fluid carries an almost dis-
sipationless current and the ratio of the current to the Hall voltage is quantised
in units of e2 /h. This quantisation is so precise that it is used to define the unit of
electrical resistance. Above a critical value of current, the dissipative voltage, Vx ,
measured along the direction of current flow, increases rapidly, leading to QHE
breakdown. The microscopic mechanism, or mechanisms, responsible for break-
down have been a topic of interest and controversy for almost twenty years [2].
In two samples used to maintain the US Resistance Standard [4,5], Cage and
co-workers observed a staircase of ten or more steps in the longitudinal voltage,
Vx , at breakdown as shown in Fig. 1.
The steps have a regular height, ΔVx ≈ 5mV and are accompanied by hys-
teresis and intermittent noise. The data were obtained by measuring Vx with a
large current I flowing along the length of a bar-shaped sample (see Fig. 1(a) -
inset) made from a GaAs/(AlGa)As semiconductor heterostructure. The figure
shows sweeps of the applied magnetic field B over an extended range around
12T at which the Landau level filling factor ν = 2. Despite the extensive lit-
erature on QHE breakdown, reviewed in ref. [2], the precise origin of the steps
has remained uncertain. In particular, they are not readily explicable in terms
of the properties of the magneto-conductivity tensor components, σxx and σxy ,
which describe successfully most of the macroscopic properties of the quantum
Hall fluid (QHF).
QHE breakdown corresponds to the disruption of dissipationless current flow
(I) along the voltage equipotentials and to an abrupt increase of Vx from a very
low value. The dissipative current, i, flowing perpendicular to the equipotentials
is given by 2e2 Vx /h. Prior to breakdown I/i can be as large as ∼ 106 -107 , values
approaching the Reynolds numbers achievable with cryogenic fluids. The dissi-
pative current i can arise from inter-Landau level tunnelling [6,7,8,9,10,11]. In
this process, an electron in the lower filled Landau level scatters elastically (or
quasi-elastically with the emission of a low energy acoustic phonon) across the
Landau level energy gap into the empty state of the upper unfilled Landau level.
Since the wavefunctions of the Landau states are strongly localised along the
direction perpendicular to the equipotentials and have strongly decaying gaus-
sian tails, a large electric field is required for significant spatial overlap between
the initial and final states of the same energy, i.e. the electric field E must be
∼ ωc /eλB , where ωc = eB/m is the cyclotron  frequency, m is the electron
effective mass (= 0.07me for GaAs) and λB = /eB is the magnetic length.
Alternatively, this condition can be expressed in the form mvd2 ∼ ωc , where
vd = E/B is the drift speed of the cyclotron orbit centre. For the experimental
conditions of Fig. 1, the average value of the component of the electric field E
along y (the average Hall field) is relatively small, Ey = VH /w ∼ 7 × 103 V m−1 ,
430 L. Eaves

Fig. 1. QHE breakdown in one of the Hall bar samples used to maintain the US Resis-
tance Standard at the National Institute of Standards and Technology, Gaithersburg.
(a) Plots of longitudinal voltage Vx versus magnetic field B at T = 1.3K, in the region
of the ν = 2 filling factor, I = 210μA. The arrows show the hysteresis for various
sweeps of B. (b) A detail of the breakdown curve, showing the large number of steps
in Vx . (Both figures from ref. [4], courtesy of NIST)

almost two orders of magnitude too small to produce significant wavefunction

overlap.
The essence of the model proposed here is that E and hence vd can be much
larger in localised microscopic regions of the Hall bar due to the proximity of the
edge of the Hall bar and the presence of charged impurities. It is proposed that
in these regions the QHF is unstable against inter-Landau level tunnelling. This
effect can be envisaged as a quantum analogue of the generation of vortex-pairs
in a von Karman vortex street, when a classical fluid flows past an obstacle, and
is also related to vortex pair formation in superfluids.

2 Model and Comparison with Experiment

For simplicity, let us assume a quadratic dependence of the electron potential
energy φ in the local region, as shown schematically in Fig. 2(a). Outside this
region, the Hall field is smaller. Over the distance ∼ s along the y-direction, we
write eφ ≈ mω02 y 2 /2. This model potential is a function of only one unknown
 Quantum Hall Effect Breakdown Steps 431

Fig. 2. a Model for spatial variation of energy of the two lowest Landau levels, n = 0
and n = 1 in a localised region of QHE breakdown. The eigenfunctions |α0 and |β1
are in the n = 0 (filled) and n = 1 (empty) Landau levels respectively, at filling factor
ν = 2. They have the same energy and spatially overlap each other. The length s is the
distance over which the cyclotron orbit centre jumps in an inter-Landau level tunnelling
transition. The inset shows schematically the geometry of the Hall bar sample. Vx is
measured between probes 1 and 3. (The Hall voltage VH is measured across 1 and 2 or
3 and 4). The small cross-hatched region indicates schematically a breakdown region
near the edge of the sample. The difference between chemical potentials at the two
edges, μ1 − μ2 = eVH . b Plot of the voltage equipotentials in the plane of the quantum
Hall fluid for a region of large velocity gradient in which breakdown occurs. The energy
variation of the Landau levels shown in box (a) above corresponds to that along the
line PQ.

parameter
 2  ω0 , and corresponds to a large local velocity gradient Ω = dvx /dy =
d φ/dy 2 /B = ω02 /ωc . The equipotentials shown in Fig. 2(b) give rise to this
type of velocity gradient. Due to the character of electron motion in crossed elec-
tric and magnetic fields, they also correspond to the directions of dissipationless
flow in this region. A potential distribution of this kind, where E is locally much
larger than Ey , can arise from the combined effect of a charged donor impurity
located current close to the QHF, and the large Hall field due to the charge dis-
tribution near the edge of the sample [10,12,13,14]. These charges are essentially
unscreened by the presence of the dissipationless QHF [14,15].
Fig. 2(a) also shows the wavefunction of a state |α0  in the lowest filled
Landau level (n = 0) and of a state |β1  in the n = 1 level. The two states have
432 L. Eaves

the same energy (the dispersion of the Landau level energies is due, of course,
to the spatial variation of φ). For simplicity, we neglect spin. The orbit centres
of |α0  and |β1  are separated by a distance Δy = s, given by

mω02 s2 /2 ≈ ωc . (1)

For small ω0 , spatial overlap between the wavefunctions is negligible. How-

ever, if ω0 is sufficiently large, overlap occurs (see Fig. 2a) and an electron in
state |α0  can undergo a quantum transition to the unfilled state |β1 . An inter-
Landau level transition of this kind leads to the dissipative breakdown current, i.
It requires a jump of orbit centre position which can be induced by the presence
of the charged impurity which destroys translational symmetry. To compare our
model with the data, in particular with the measured value of ΔVx , we need to
determine the size of dissipative current i generated by inter-Landau level tun-
nelling. In principle, the tunnelling rate could be calculated using Fermi’s Golden
Rule. The problem is that inter-Landau level tunnelling creates an electron in
the upper, previously empty, Landau level and an empty state, or hole, in the
previously filled lower Landau level. The two particles experience a Coulombic
attractive interaction and also tend to screen the local potential which generated
them, thereby perturbing the drift motion of other electrons in adjacent states
of the lowest Landau level. Since the interaction between the electron-hole pair
(or magneto-exciton) and the QHF is quite complicated, even in the absence of
impurity or edge charges [16], we cannot make a calculation based on a simple
independent particle picture. To avoid this difficulty, let us consider the follow-
ing semiclassical description, which seems to account for the essential features
of the breakdown mechanism.
At the critical breakdown values of current and magnetic field, we assume
that the electric field in the local breakdown region is just large enough to induce
inter-Landau level tunnelling. This process can be envisaged as electron-hole pair
formation, the electron occupying state |β1  and the hole corresponding to the
electron missing from state |α0  in the otherwise filled lower Landau level. The
presence of the electron-hole pair acts to screen the locally strong electric field,
thus temporarily inhibiting further generation of pairs by tunnelling. However,
due to the presence of the local electric field, the electron-hole pair drift away
from the region in which they are formed. For the potential distribution shown
in Fig. 1, assuming an independent particle description, the hole left in state
|α0  has a high E × B drift velocity, whereas the electron in |β1  has a lower
drift velocity, each given by the local value of the electric field. We account
qualitatively for the effect of the Coulombic attraction between the electron and
hole by assuming that the pair drift at a speed determined by the mean electric
field, which is mω02 s/2e. The mean drift speed is therefore vd  ≈ Ωs/2 and the
pair move a distance ∼ s from the formation region in a time τ = 2/Ωγ, where γ
is a numerical factor ∼ 1. At this separation, they no longer screen effectively the
strong local electric field around the impurity, so this is restored to its original
strength, leading to the creation of another electron-hole pair. Assuming that
the time τ is the controlling factor in the formation of electron-hole pairs, we
 Quantum Hall Effect Breakdown Steps 433

can take the rate of generation to be ≈ γΩ/2. On a longer timescale, given by

the intra-Landau level energy relaxation time τe ≈ 10−10 s, we assume that the
electron-hole pair breaks up, the two particles diffusing away from each other
with velocity components perpendicular to the equipotentials [17]. This process
can occur for each particle by emission of low energy acoustic phonons through
intra-Landau level scattering processes. The dissipative backscattering current
i ≈ eγΩ/2 causes an incremental increase in Vx given by

ΔVx = hi/2e2 = hγΩ/4e . (2)

The steps shown in Fig. 1 would then correspond to successive contributions

to the dissipative current of other breakdown regions with a similar local poten-
tial distribution due to the presence of a charged impurity, each adding ∼ eγΩ/2
to the dissipative current as B is swept away from the ν = 2 value. To test the
model, we now compare the observed value of ΔVx with the value of ω0 required
to generate inter-Landau level tunnelling. Using equation 1 and recalling that
Ω = ω02 /ωc , we write equation 2 in the form

eΔVx = πγ2 /ms2 . (3)

This equation allows us to relate the observed value of ΔVx = 5mV to the
orbit centre separation, s, rather than to ω0 . Setting γ ≈1, we then obtain
s2 /λ2B ≈ πωc /eΔVx , which gives s ≈ 3.5λB , where λB = /eB. This value
of s agrees quite well with the condition for alignment of the classical
√ turning

points of the wavefunctions |α0  and |β1 , namely when sT P = 3 + 1 λB .
This condition corresponds to the threshold of the strong increase in the matrix
element for inter-Landau level transitions [11].
The model can also be tested against the recently-reported voltage steps for
QH breakdown of hole gases [18]. Here the typical step height is ΔVx = 1mV
for breakdown at ν = 2. Using equation 3 and the measured value of the hole
effective mass, mh = 0.15me [19], we obtain s = 2.3λB , which is also quite close
to the value of sT P . The hole gas data therefore provide further support for the
model.
Recalling that Ω = ω02 /ωc we can use equation 2 to confirm that the size of
ω0 required for breakdown is consistent with that produced by an unscreened
charged impurity. For such an impurity located at a distance Δz from an idealised
strictly 2-dimensional QHF, it is easily shown that the step height of ΔVx = 5mV
observed at B ≈ 12T by Cage et al corresponds to a value of Δz given by
1/3
(h/16πεr ε0 BΔVx ) ≈ 13nm, where εr ≈ 12 is the relative permittivity of
GaAs. This value corresponds very closely to the mean distance of the QHF
from the GaAs/(AlGa)As interface [20] and indicates that charged interface
impurities can generate the local velocity required for our breakdown model.
To conclude this section, we note that the model may also be relevant to
the results of experiments by Bliek et al [21] in which the QHE breakdown was
induced by passing current through a narrow constriction, 1μm wide. The critical
current for breakdown in the constriction corresponds to the typical value of
electric field required for inter-Landau level scattering [9]. These measurements
434 L. Eaves

are also of interest because steps appear as regular values of resistance, Rx =

Vx /I, rather than of voltage, Vx . Briefly, the model proposed here can account
for this difference using the following argument (fuller details will be published
elsewhere). In a very narrow constriction, we can assume that the overall form
of the Hall voltage profile V (y) across the constriction, i.e. across the entire
region of current flow, is mainly determined by the edge charges. The form of
V (y) is a logarithmic function, with a prefactor determined by, and proportional
to, the total current I flowing
 through
 the constriction [10,12,14]. The local
velocity gradient, Ω (y) = d2 φ/dy 2 /B, is then also proportional to I. From
equation 2, it can be seen that this would then lead to increments, ΔVx , which are
proportional to I, i.e. to constant increments of ΔRx . Further analysis based on
this model and on the observed size of the steps, ΔRx ≈ 200Ω, indicates that the
breakdown region in the channel is confined to a thin (∼ 0.1μm) layer close to one
edge of the constriction, in reasonable agreement with model calculations [12].

3 Analogies with Classical and Quantum Fluids

Note that the proposed mechanism for the dissipative current i is closely analo-
gous to the formation of vortex-antivortex pairs which are created when a classi-
cal fluid flows past an obstacle, e.g. a cylinder, above a critical Reynolds number,
Re ≈ 60. In particular, the value of i/e, the electron-hole pair generation rate,
corresponds to the rate of vortex shedding in a classical von Karman vortex
street [22]. An insight into the analogy can be gained by applying Madelung’s
hydrodynamic formulation of the Schrödinger wave equation [23] to the QHF.
We can envisage the electron motion as comprising an E × B drift motion along
the equipotentials combined with an internal cyclotron motion with a vorticity
equal to ωc [24]. The formation of an electron-hole pair in an almost dissipa-
tionless QHF due to inter-Landau level tunnelling in the vicinity of a charged
impurity would then correspond to the creation of an electron (vortex) in the
previously empty n = 1 Landau level and the loss of an electron (or, creation of
hole = antivortex) in the previously filled lower Landau level. In this picture, the
creation of vortices of opposite sign, but equal magnitude, in the upper and lower
Landau levels can be compared with the conservation of circulation (vorticity)
in almost inviscid classical fluids.
Interestingly, using the classical concept of eddy viscosity for 2d flow [25]
with the tunnelling distance s playing the role of the mixing length, it is easily
shown that each step in Vx in Fig. 1 corresponds to an eddy viscosity increment
of ∼ s2 Ω/2 ∼ /m ∼ 10−3 m2 s−1 . This value is ∼ 105 times larger than the
corresponding value of the circulation quantum (/MHe ) for superfluid He; hence
a QHF is relatively stable against electron-hole pair (=vortex pair) generation
at high flow speeds, compared to the analogous case of an ideal superfluid He
film [3,26].
It is also interesting to note that the hysteresis and intermittent noise which
frequently accompany QHE breakdown (see Fig. 1) are features which are also
 Quantum Hall Effect Breakdown Steps 435

characteristic of the transition from laminar to turbulent flow in classical fluids

through pipes at high Reynolds numbers [22].
The hydrodynamic description proposed here may also be relevant to another
recent experiment on QHE breakdown [27], which suggests an analogy with
experiments on grid turbulence. Also, the recently reported non-local (delayed)
nature of breakdown in narrow Hall bars [28] is somewhat analogous to the
breakdown of laminar pipe flow in Reynolds’s classic experiments [22].

4 The Breakdown Steps and Their Relation

to Other Types of QHE Breakdown
Komiyama and Kawaguchi [17] have proposed a thermal instability or bootstrap
electron heating model to account for the type of QHE breakdown observed in
many samples. This model accounts successfully for the size of critical breakdown
current, Ic , of a wide variety of Hall bars of differing widths, studied by Kawaji
and co-workers [29], and by others. The model also successfully predicts the B 3/2
dependence of Ic , as observed by Kawaji and co-workers [29], [30]. However, the
type of breakdown described by the Komiyama-Kawaguchi model is probably
of a different type from that observed by Cage and co-workers. In particular,
the occurrence of bootstrap heating may lead to such a strong dissipation that
it masks the small breakdown steps which are best observed in high resolution
B-sweeps.
Finally, we note the rich, but seemingly diverse, variety of breakdown be-
haviour reported in recent publications: (i) the type described by the Komiyama-
Kawaguchi model [17,28]; (ii) the step- breakdown of the type discussed here
[4,5,9,18,21]; (iii) the breakdown observed at relatively low current density in
narrow, high mobility Hall bars, also probably due to inter-Landau level scat-
tering close to the sample edge [10]; (iv) the breakdown measurements which
show a relatively well-behaved dependence on magnetic field (filling factor), cur-
rent and temperature, which can be used to construct a phase diagram with a
phenomenologically similar form to that found for Type II superconductors [31].

5 Summary
A model has been proposed to explain the voltage steps observed in the break-
down of the QHE in two samples used to maintain the US Resistance Standard.
The model suggests that we can view this type of breakdown as an instability in
the flow of the QHF. This instability can arise from the generation of electron-
hole pairs by inter-Landau level tunnelling due to the presence of a charged
impurity in local regions of the Hall bar, where the electric field is large. This
type of process has analogies with vortex pair formation in classical fluids and
superfluids.
436 L. Eaves

Acknowledgement

I am grateful to Dr. M. Cage and NIST for permission to use Fig. 1, and to
my colleagues F. W. Sheard and K. A. Benedict for helpful discussions. This
work was partly supported by the Engineering and Physical Sciences Research
Council (UK).

References
1. K. von Klitzing, G. Dorda and M. Pepper, Phys. Rev. Lett. 45, 494-497 (1980).
2. For a comprehensive review of QHE breakdown, see G. Nachtwei, Physica E 4
79-101 (1991).
3. C. M. Muirhead, W. F. Vinen and R. J. Donnelly, Phil. Trans. R. Soc. Lond. A
311, 433-467 (1984).
4. M. E. Cage, J. Res. Natl. Inst. Stand. Technol. 98, 361-373 (1993). This reference,
and reference 4 below, describe the voltage steps observed in the two US Resistance
Standard samples with reference numbers GaAs(7) and GaAs(8).
5. C. F. Lavine, M. E. Cage, and R. E. Elmquist, J. Res. Natl. Inst. Stand. Technol.
99, 757-764 (1994).
6. H. L. Störmer, et al., in: J. D. Chadhi, W. A. Harrison (Eds.), Proc. of the 17th
Int. Conf. on the Physics of Semiconductors (17th ICPS 1984), Springer, Berlin,
1985 p.267.
7. O. Heinonen, P. L. Taylor and S. M. Girvin, Phys. Rev. 30, 3016-3019 (1984).
8. L. Eaves, P. S. S. Guimares and J.-C. Portal, J. Phys. C: Solid State Phys. 17,
6177-6190 (1984).
9. L. Eaves and F. W. Sheard, Semicond. Sci. Technol. 1, 346-349 (1986).
10. N. Q. Balaban, U. Meirav, H. Shtrikman and Y. Levinson, Phys. Rev. Lett. 71,
1443-1446 (1993).
11. C. Chaubet, A. Raymond and D. Dur, Phys. Rev. B 52, 11178-11192 (1995).
12. A. H. MacDonald, T. M. Rice and W. F. Brinkman, Phys. Rev. 28, 3648-3650
(1983).
13. D. J. Thouless, J. Phys. C 18, 6211-6218 (1985).
14. C. W. J. Beenakker and H. van Houten, in Solid State Physics: Semiconductor Het-
erostructures and Nanostructures, pp.1-228, eds. H. Ehrenreich and D. Turnbull,
Academic Press, 1991.
15. D. B. Chklovskii, B. I. Shklovskii and L. I. Glazman, Phys. Rev. B 46, 4026-4034
(1992).
16. C. Kallin and B. I. Halperin, Phys. Rev. B 30, 5655-5668 (1984).
17. S. Komiyama and Y. Kawaguchi, Phys. Rev. B 61, 2014-2027 (2000).
18. L. Eaves et al., Physica E 6,136-139 (2000).
19. B. E. Cole et al., Phys. Rev. B 55, 2503-2511 (1997).
20. F. Fang and W. E. Howard, Phys. Rev. Lett. 16, 797 (1966).
21. L. Bliek et al., Surf. Sci. 196, 156 (1988).
22. See for example Fluid Dynamics for Physicists, T. E. Faber (Cambridge University
Press, 1995).
23. E. Madelung, Z. Phys. 40, 322 (1926).
24. L. Eaves, Physica B 272, 130-132 (1999).
25. G. I. Taylor, Proc. Roy. Soc. London Series A, 85 685-705 (1932).
 Quantum Hall Effect Breakdown Steps 437

26. R. P. Feynman, Application of quantum mechanics to liquid helium, in Progress in

Low Temperature Physics I (ed. C. J. Gorter), North-Holland. Chapter 2 (1955).
27. I. I. Kaya, G. Nachtwei, K. von Klitzing and K. Eberl, Phys. Rev. B 58, R7536-
R7539 (1998).
28. S. Komiyama, Y. Kawaguchi, T. Osada and Y. Shiraki, Phys. Rev. Lett. 77, 558-
561 (1996).
29. S. Kawaji, Semicond. Sci. Technol. 11, 1546-1551 (1996).
30. S. Kawaji et al., J. Phys. Soc. Japan 63, 2303-2313 (1994).
31. L. B. Rigal et al, Phys. Rev. Lett. 82, 1249-1251 (1999).
Atomic Bose Condensate with a Spin Structure:
The Use of Bloch State

Hiroshi Kuratsuji

Department of Physics, Ritsumeikan University, Kusatsu City, 525-8577, Japan

Abstract. We study a novel aspect of the anisotropic bose condensates that is inspired
by the recently discovered alkari atom bose condensates. To formulate the theory, we
use the Bloch state which provides a suitable tool for describing the order parameter
possessing with the spin components. The Landau-Ginzburg type Lagrangian is re-
duced to that of ferromagnet fluid which contains the kinectic energy fluid as well as
the Heisenberg Hamiltonian. Specifically we scrutinize a possible vortex state for this
condensate possessing with spin structure; the profile for a single vortex as well as the
equation of motion.

1 Introduction
The discovery of the Bose-Einstein condensates (BEC) in alkari atomic vapours
has opened a new direction of condensed matter physics [1,2,3] besides the con-
ventional quantum condensates such as superfluid helium or quantum liquid.
The purpose of this report is to address a specific aspect of this newly estab-
lished BEC; namely, we want to explore the physical effect that is inherent in
the spin structure of the order parameter for the atomic condensate. The spin
nature of the order parameter is caused by the “hyperfine spin” of constituent
alkari atoms. Some aspects of the spin degree that arises from hyperfine spin was
first suggested by using the spinor field [4]. From experimental point of view, the
condensate incorporating spin degree can be realized in the optical trap [5] that
has been invented as a complement of the magnetic trap, since the spin degree
survives as an active degree, whereas in the magnetic trap the spin degree is
frozen along the magnetic field line. In the sequel to this experiment, the term
of the “spinor condensate”was introduced in [6], which indeed predicts the polar
state as well as the ferromagnetic state for the special case of the spin magnitude
1.
The bose condensates possessing with spin structure can be regarded as an
“anisotropic quantum fluid”. Apart from the atomic BEC, the superfluid He3-
A provides a typical anisotropic quantum fluid, which is indeed quite different
from the conventional superfluid He4 governed by the scalar order parameter.
The order parameter for He3A is written in terms of the vector order parameter
causing the so-called l-texture [7] that resembles the director in liquid crystals.
In what follows, we shall give a general formalism of the atomic bose conden-
sate accomodating spin texture and subsequently address some specific aspects
inherent in spin structure especially, vortex state. In order to carry out this, we
adopt the Landau-Ginzburg (LG) Lagrangian which has similar form with the

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 438–444, 2001.

c Springer-Verlag Berlin Heidelberg 2001
 Bose Condensate with Spin Structure 439

one developed for the dynamics of He3A [8]. Here the essential point is to use the
Bloch state [9] in order for describing the spin structure of the order parameter.
The use of the Bloch state can incorporate any magnitude of spin that includes
the condensate with spin 1 as a special case that has been previously given. The
LG Lagrangian is reduced to that of the ferromagnet fluid for which the La-
grangian consists of two terms; the fluid kinetic energy as well as the Heisenberg
ferromagnet. We consider the profile of a single vortex as well as its motion and
the resultant equation of motion gives the geometrical force that is caused by
the topological invariant known as the Mermin-Ho relation.

2 Order Parameter and Lagrangian

To describe the spin nature of the order parameter, it is natural to adopt the
spin states which is denoted by |J, M  with M = −J ∼ +J. Specifically, we are
concerned with the lowest state |J, −J. Using the usual vector notation, it is
written as (0, · · · , 1)T ≡ Ψ0 . From this vector, the general vector which directs
n = (sin θ cos φ, sin θ sin φ, cos θ) is constructed by rotating the Ψ0 by the rotation
operator; R(θ, φ) = exp[iθ(k · J)] where k = (cos φ, sin θ, 0). Thus, following the
general procedure, we have the vector
Ψξ = exp[η Jˆ+ − η ∗ Jˆ− ]Ψ0 = (1 + |ξ|2 )−J exp[ξ Jˆ+ ]Ψ0 (1)

with ξ = tan θ2 exp[−iφ] and Jˆ± the raising and lowering operator for spin. This
state vector is known to be the Bloch states (alias spin coherent state). For the
case of J = 1 , we have
θ √
Ψ = (1 + tan2 )(ξ 2 , 2ξ, 1)†
2
θ θ θ θ
= eiφ (sin2 eiφ , sin cos , cos2 e−iφ ) (2)
2 2 2 2
and for J = 12 it follows that Ψ † = (cos θ2 , sin θ2 e−iφ )† . Thus we regard equation
(1) as the general order parameter carrying the spin of arbitrary magnitude:
Ψ = Ψξ Δ0 exp[iα] (3)
Δ0 represents the magnitude of the condensate which is fixed to be constant
though it spacially varies in general. Actually, the order paramter is determined
up to a phase factor, exp[iα], but it can be chosen arbitrarily so we set α = 0 in
the following argument. Having defined the vector order parameter realized by
the Bloch state, we give the Lagrangian that governs the dynamics of this order
parameter. Here it is natural to set it in the similar form with the superfluid
He4: 
i
L = ( (Ψ † Ψ̇ − c.c) − H(Ψ † , Ψ ))dx (4)
2
and the Hamiltonian density
2
H= ∇Ψ † ∇Ψ + V (Ψ † , Ψ ) (5)
2m
440 H. Kuratsuji

Here V (Ψ † , Ψ ) represents the interaction potential that comes from the atomic
two-body interactions as well as the trapping potential which is chosen such that
the stability of the condensate Δ0 is maintained. The concrete form of this will
be given later.

3 Hydrodynamical Equation
Now we use the the Bloch form to reduce the Lagrangian. In order to carry out
this, we need to have some elementary but crucial calculations concerning the
derivative of the Bloch state, namely, for the time derivative
∂Ψξ
= Ṅ exp[ξ Jˆ+ ]Ψ0 + N ξ˙Jˆ+ exp[ξ Jˆ+ ]Ψ0 (6)
∂t
with N = (1 + |ξ|2 )−J . The gradient is also obtained by replacing the dot by ∇.
Then after some tedious manipulation, we get the first term of the Lagrangian,
LC (called the canonical term):

(ξ ∗ ξ˙ − c.c)
iΨ † Ψ̇ − c.c. = JΔ20 (7)
1 + |ξ|2
Hence the Hamiltonian term becomes
1 2 Δ20 ∇ξ ∗ ∇ξ
H= mΔ20 v2 + + V (|ξ|2 ) (8)
2 2Jm (1 + |ξ|2 )2
where
J i(ξ ∗ ∇ξ − c.c)
v= (9)
m 1 + |ξ|2
which reresents the velocity field (see below). In the above derivation, use is
made of the relation
2Jξ
N 2 Ψ0∗ exp[ξ ∗ Jˆ− ]Jˆ+ exp[ξ Jˆ+ ]Ψ0 = (10)
1 + |ξ|2
which is an expectation value of the spin operator with respect to the Bloch
state. This written in terms of the angular variable;

Lx = J sin θ cos φ, Ly = J sin θ sin φ, Lz = J cos θ, (11)

which defines the effective spin carried by the order parameter. If using the angle
polar coordinate, the hydrodynamical form of the Lagrangian becomes

L = [JΔ20 (1 − cos θ)φ̇ − H(θ, φ)]dr (12)

The first term of the reduced Hamiltonian is regarded as a kinectic energy where
the velocity field v is written in terms of the angular variable:
J
v= (1 − cos θ)∇φ (13)
m
 Bose Condensate with Spin Structure 441

The second term of the Hamiltonian has the same form as the continuous Heisen-
berg model of ferromagnet [10], namely, it is rewritten as

HS [(∇θ)2 + sin2 θ(∇φ)2 ]dr (14)

From the formal aspect, the above Lagrangian has the same form as the
superfluid
 He4 described by the scalar order parameter. The action principle
δ Ldt = 0 leads to the canonical equation of motion
dθ 1 δH dφ 1 δH
JΔ20 =− , JΔ20 = , (15)
dt sin θ δφ dt sin θ δθ
In terms of the spin variable, it turns out to be
δH
L̇ = L × (16)
δL
The stationary solutions are obtained by putting as θ̇ = 0, φ̇ = 0 and around
the stationary solutions one can get the small oscillation that is just realized
by spin wave. Besides this, we may have the case of“no flow”, namely, v = 0
leading to cos θ = 1 which gives the aligned spin L3 = J. Indeed the aligned
state is expected from the energy consideration if we take into account of the
proper form of the interaction potential.

4 Vortex State
We now consider the vortex state that is inherent in the spin texture. In the
following argument, it is assumed that the system is two-dimensional plane (x, y).

4.1 The Profile of a Single Vortex

We consider the following nonlinear potential which comes from the spin-spin
interaction; 
V (Ψ ∗ , Ψ ) = g Ψ ∗ Ŝ3 Ψ Ψ̇ ∗ Ŝ3 Ψ dx (17)

with the positive coupling constant g. In terms of the angular variable, this turns
out to be 
Ṽ = gΔ20 J 2 cos2 θd2 x (18)

The configuration of the single vortex is controlled by the angle variables (θ, φ).
A static solution for the one vortex is obtained by choosing the phase function
φ = n tan−1 ( xy ), with n = 1, 2, · · · . being the winding number, together with
the profile function θ that is given as a function of the radial variable r. Thus
the static Hamiltonian is written in terms of the field θ(r):

2 Δ20 J dθ n2 θ θ
H= [( )2 + 2 (2J + cos2 )4 sin2 + g  cos2 θ]rdr (19)
4m dr r 2 2
442 H. Kuratsuji

where g  = 4gJm
2 . The profile function θ(r) may be derived from the extremum
of H, namely, the Euler-Lagrange equation leads to

d2 θ 1 dθ n2
+ − (4J sin θ + sin 2θ) + sin 2θ = 0 (20)
dξ 2 ξ dξ 2ξ 2

where we adopt the scaling of the variable: ξ = g  /2r . In order to examine the
behavior of θ(ξ), we need a specific boundary condition at ξ = 0 and ξ = ∞. We
impose at the origin such that θ(0) = 0, whereas at ξ = ∞, we adopt two typical
possibilities: A) θ(∞) = π and B) θ(∞) = π/2. If we define the unit vector as
l = L/J we have l3 (0) = 1 for both casese A), B) and we have l3 (∞) = −1 for
case A) and l3 (∞) = 0 for case B). This feature indicates that the spin field
which directs upward gradually converts into the state of downward or outward.
First we consider the behavior near the origin ξ = 0, for which the differential
equation behaves like the Bessel equation, so we see the power law

θ(ξ) ξk (21)

with k 2 = n2 (2J + 1). Next we examine the behavior at ξ = ∞. This is simply

performed by checking the stability for two cases mentioned above: (A) and (B).
Now for the case (B), if putting θ(ξ) = π2 + α, with α the infinitesimal deviation,
so the linearized equation near ξ = ∞, α − α 0, which results in α exp[−ξ].
This suggests that the solution behaves as at ξ = ∞,
π
θ(ξ) = (1 − exp[−ξ]) (22)
2
On the other hand, for the case (A) we have α + α 0, which gives α exp[iξ]
meaning the oscillatory behavior. This simply implies that the solution with
θ(∞) = π does not converge to the stable solution and this feature means that
the case (A) is not relevant. Keeping mind of the above general feature, we here
give a numerical behavior of the function θ(ξ) (Fig.1).

4.2 Vortex Dynamics

We now consider a possible motion of a single vortex. The dynamics of a vortex is
described by the collective coordinates X(t) = (X, Y ) that represent the center of
the moving vortex, and the space coordinates in the angle variables are replaced
as x → x − X(t). Using the chain rule, ∂φ ∂Φ
∂t = ∂X Ẋ, the canonical term LC
becomes

LC = JΔ20 (1 − cos θ)∇φ · Ẋd2 x

= mΔ20 v · Ẋd2 x (23)

where v means the velocity field given above, in other words, the momentum
density p is defined to be canonically conjugate with X such that p = M Δ20 v
 Bose Condensate with Spin Structure 443

with ρ = M Δ20 (the mass density). The velocity field v does not bear any
singularity near the origin, and such vortices are called “soft-core” as in the case
of the superfluid He3-A. The equation of motion for a vortex is given by the
Euler- Lagrange equation
d ∂L ∂L
− = Fex (24)
dt ∂ Ẋ ∂X
This is rewritten as the balance of two types of forces; FC + FT = Fex where FT
is given by FT ≡ − ∂X∂T
, where T is the kinetic energy term of spin fluid (i.e. the
first term of eq.(8)). The RHS represents the force for balance that may come
from the other types of non-dissipative force as well as dissipative force. We get
the force from the canonical term:

FC = 2mΔ20 (k × Ẋ)[ (∇ × v)z d2 x] (25)

where the vector k is the z-directed unit vector. There is no contribution coming
from the kinetic energy term as is intuitively guessed. The force thus derived
is considered to be a counterpart of the Magnus-type force for the anistropic
bose condensates, that is the same as the force for a vortex in ferromagnet. We
consider the meaning of the integral in the (25). The integrand, which is nothing
but the continuous vorticity ω = ∇ × v

ω = J/m sin θ∇θ × ∇φ

J ∂l ∂l
= l·( × ) (26)
m ∂x ∂y

π
2
θ ( ξ )

Fig. 1. The profile of the function θ(ξ)

444 H. Kuratsuji

This relation is a counterpart of the Mermin-Ho relation for the case of the
superfluid He3A. The integral over the plane is convered into the one over the
field space 
σ=J sin θdθdφ (27)

so finally we get
F = 2mΔ20 σ(k × Ẋ) (28)
This is just the same form as the expression given for the vortex occuring in the
superfluid He3A [8].

5 Summary

In this report we have presented a hydrodynamical theory for the atomic bose
condensate accomodating the spin texture. This is naturally carried our by
adopting the Bloch state (or spin coherent state), which enables us to formulate
the anisotropic nature of the bose condensate. This is closely connected with the
superfluid He3A. The LG Lagrangian is brought to the fluid dynamical form for
the spin liquid. The spin vortex has been explicated from this fluid dynamical
Lagrangian in both static and dynamical aspects.

References
1. [Link], [Link], [Link], [Link] and [Link], Science
269, 198(1995).
2. [Link], [Link], J.J. Tollett, and [Link], [Link].75,
1687(1995).
3. [Link], [Link], [Link], [Link] Druten, [Link], [Link],
and [Link], [Link].75, 3969(1995).
4. [Link] and [Link], [Link].77,2595(1996).
5. [Link]-Kurn, [Link], [Link], [Link], [Link],
[Link], and [Link], [Link].80,2027(1998).
6. [Link],[Link].81,742(1998).
7. [Link] and [Link], [Link].59,533(1987).
8. [Link] and [Link], [Link].B59, 11175(1999).
9. [Link], [Link], [Link] and [Link], [Link].A6,2211(1972).
10. [Link] and [Link], [Link].186A,255(1994).
Quantum Dynamics of Vortex–Antivortex Pairs
in a Circular Box

Vittorio Penna

Dipartimento di Fisica and Unità INFM, Politecnico di Torino,

[Link] Duca degli Abruzzi 24, I-10129 Torino, Italy

Abstract. After reviewing the quantum dynamics of unbounded vortex pairs, the
case of a vortex-antivortex pair in a box with a circular boundary is investigated. The
appropriate algebraic framework in which performing the quantization is reconstructed
and the structure of the vortex pair spectrum is discussed.

1 Introduction
One of the early attempts to study the quantum dynamics of two-dimensional
(2d) pointlike massless vortices was performed in Ref. [1] within the canonical
quantization for an almost 2d system formed by an array of parallel vortex lines.
While the generalization of such a model to a gas of vortex lines that are closed
(and thus endowed with an arbitrarily complex 3d topology) has raised dramatic,
still open, formal problems [2], the simpler structure of 2d vortices has permitted
to investigate several aspects of vortex quantum dynamics (VQD), from the pos-
sible fractional character of their statistics [3], [4] and the influence of boundary
effects on VQD [5], to the emergence of point vortices, dynamically quantized,
within both a second quantized |Ψ |4 - field theory [6], and the quantum-phase
model for Josephson junctions arrays [7].
A renewed interest for 2d vortex dynamics (VD) has been prompted by the
recent, experimental as well as theoretical, developments in the context of both
superfluidity and superconductivity [8], [9]. The great improvements concerning
both the measurement techniques and the observations of microscopic processes
[10], [11] should render quantum aspects of VD accessible to experimental de-
tection. This is certainly true for the Bose-Einstein condensate (BEC) physics
which represents a rich source of new experimental scenarios in which vortex
creation is effected and VD can be studied [12].
This paper is devoted both to investigate the VQD of a vortex-antivortex
(VA) pair in the presence of a circular reflecting wall, and to illustrate the for-
mal tools one employs to study the spectral properties of vortex pair models. The
construction of the model Hamiltonian is based on the virtual charge method
which accounts for the boundary confining the fluid through an appropriate
configuration of virtual vortices [13], [5]. Although the quantization of the un-
bounded VA pair is apparently trivial (its spectrum essentially coincides with
that of a free particle), the interactions between the virtual vortices and the pair
vortices has the effect to confine the latter on closed trajectories that are thus
liable to be quantized. This aspect is particularly investigated in the sequel.

C.F. Barenghi, R.J. Donnelly, and W.F. Vinen (Eds.): LNP 571, pp. 445–452, 2001.

c Springer-Verlag Berlin Heidelberg 2001
446 V. Penna

The VQD originated by the VA pair Hamiltonian is studied by means of the

spectrum generating algebra method [14] applied in [15] to the vortex pair scat-
tering from a disk. This consists in identifying first a complete set of dynamical
degrees of freedom forming a Lie algebra (the dynamical algebra) and in deriving
then the Hilbert space of the system from the unitary irreducible representations
(UIR) of the associated Lie group. For the VA pair [15] such a group is SU(1,1)
whose UIR’s are completely classified. Here, we exploit repeatedly this formal
set-up to characterize the weakly excited quantum states of the VA dynamics.

2 Canonical Quantization of Planar Vortices

The Hamiltonian of N pointlike vortices in a frictionless fluid is given by [16]
H(z1 , ..., zn ) = −(ρ/4π) Σi=j ki kj ln[|zi − zj |/a] , (1)
where ρ is the fluid planar density, zj := xj + iyj ∈ C describes the jth vortex
position on the ambient plane, and the vorticity of the jth vortex kj = hnj /mH
(mH is the fluid atom mass) is quantized (nj ∈ Z) due to the Feynman-Onsager
condition [17]. The length a represents the vortex core size stating the minimum
distance between a vortex (kj > 0) and an antivortex (kj < 0) before coalescence
processes occur. The Hamiltonian equations are derived from Eq. (1) via the
Poisson brackets [16]
{F, G} = Σj [Fxj Gyj − Gxj Fyj ]/ρkj , (2)
(Ez := ∂E/∂z, E = F, G) involving, in turn, the ρkj -dependent canonical
brackets {xi , yj } = δij /(ρkj ). Then vortex coordinates can be regarded as a
complete set of canonically conjugate variables whose momenta are defined as
pj = ρkj yj . Also, one can easily check that Jz = −ρΣj kj (x2j + yj2 )/2 and J± =
ρΣj kj (xj ± iyj ) (J± := Jx ± iJy ) are constants of motion namely {F, H} = 0
with F = Ja , J∗ , (a = x, y, z), and fulfil the equation
, -
J∗ = −(2CJz + Jx2 + Jy2 )/ρ := Σi=j (ki kj /2) (xi − xj )2 + (yi − yj )2 , (3)
where C = ρ Σj kj is related to the total vorticity. J∗ identifies with the Casimir
function of the e(2)-like algebraic structure {Jz , J± } = ∓iJ± , {Jx , Jy } = C
(hereafter we denote it by e∗ (2)) which fully emerges when C = 0. Physically,
the conserved quantities Jx (Jy ), Jz represent, up to a constant factor, the x
(y) component of the fluid total momentum and the fluid total angular mo-
mentum, respectively. The quantization of the VD through the canonical rule
[xi , pj ] = δij i, (pj := ρkj yj ensures the canonical form) entails that the sym-
metry operators Ja form the algebra e∗ (2)
[Jx , Jy ] = iC , [Jz , Jx ] = iJy , [Jz , Jy ] = iJx . (4)
In view of the fact that the constants of motion can be used to integrate the
dynamical equation if they mutually commute, Eqs. (4) imply that the many-
body wave function for the 2d vortex gas is characterized by three macroscopic
quantum numbers two of which are, of course, H and J∗ , while the third one can
be chosen as an arbitrary linear combination I = aJx +bJy +cJz since [H, I] = 0.
 Quantum Dynamics of Vortex–Antivortex Pairs 447

2.1 The Spectrum of Unbounded Vortex Pairs

The spectrum of the vortex-vortex pair can be reconstructed within the quan-
tization procedure of the harmonic oscillator (HO) [3], [5]. Its generalization to
a pair with two arbitrary vortex charges k1 and k2 has been discussed in [15],
where a thorough analysis of the energy spectrum degeneracy is also given and
related to the choice of I. A special case (that relevant for the VA model we
are presenting) occurs when k1 = −k2 (pure VA case) which leads to the free
particle spectrum instead of the HO spectrum.
To explain the source of such a basic difference as well as for illustrative
purposes we first review the general case k1 = −k2 . The two-vortex Hamiltonian
given by Eq. (1) is H(z1 , z2 ) = −(ρk1 k2 /4π) ln[J∗ /(k1 k2 ρa2 )], where |z1 − z2 |2
has been replaced by ρk1 k2 J∗ due to Eq. (3). The set of energy eigenstates thus
coincides with the J∗ spectrum. Upon assuming 0 ≤ |k2 | ≤ k1 (such range suffice
to describe any possible pair), let us express J∗ through the canonical conjugate
variables P = Jy , X = Jx /C, p = [k1 k2 ρ/(k1 + k2 )](y1 − y2 ), x = x1 − x2 ,
satisfying the commutators [X, p] = [x, P ] = 0 and [x, p] = [X, P ] = i. Then
the logarithm argument can be written in the HO form
 
J∗ /(k1 k2 ρ) = |z1 − z2 |2 = p2 + w2 x2 /w2 , (5)

where the frequency w reads w = k1 k2 ρ/(k1 + k2 ). This implies that the states

Ψn (x; |w|) = C,n exp[−x2 /22 ]Hn (x/) , (6)

√ 
where C,n := 2n π n!, 2 = /|w|, that satisfies the secular equation p2 +
w2 x2 Ψn = |w|(2n+1)Ψn , represent the eigenstates of |z1 −z2 |2 with eigenvalues

Sn (w) = (/|w|)(2n + 1). (7)

A possible way to describe the energy-level degeneracy is that supplied by the
quantum number of the conserved quantity I ≡ Jz (this generates the rigid
rotations of vortices around the ambient plane origin which leave H unchanged).
The complete set of eigenstates Ψnm (x, X) = Ψn (x; |w|) Ψm (X; |C|) is readily
obtained by rewriting Eq. (3) as Jz = −(ρJ∗ + Jx2 + Jy2 )/2C. This makes evident
that the HO-like √ wave function Ψm (X; |C|) ≡ fL,m exp[−X 2 /2L2 ]Hm (X/L),
where fL,m := 2 πL m!, L = /|C|, diagonalizes Jx2 + Jy2 = P 2 + C 2 X 2 .
m 2

Hence the eigenvalues associated with Jz have the form

Λm (n; w) = −(sg(w)n + m + 1) , (sg(w) := w/|w|) (8)

where we have exploited the fact that ρJ∗ /C = (2n + 1)w/|w| and C =
ρ(k1 + k2 ) > 0, due to our initial assumptions. Such a description of the energy
degeneracy entails a significant geometric-quantum picture. One can show [15]
that the integers (n, m)
 state that the two vortices are confined along two circles
of radii Rj = |k1 k2 | Sn (w)/|kj ||C|, j = 1, 2, while the locus allowed for the
vorticity center is the circle of radius X 2 + Y 2 = [(2m + 1)/|C|]1/2 , labeled by
m. The uncertainty principle prevents one from getting any further information
448 V. Penna

on the angular position both of the vorticity center and of the vortex position
on their own circles. The effects of other choices of I are illustrated in Ref. [15].
For k2 → −k1 (pure VA case) the canonical scheme based on x, X, p, P breaks
down in that 1/(k1 +k2 ) → ∞. In particular, Jx and Jy tends to x and y = y1 −y2 ,
respectively, which now commute since ρk1 [x, y] ≡ [Jx , Jy ] = iC ≡ 0. Such a
situation imposes the use of the new (canonically conjugate) variables

x , y , Px = (y1 + y2 )/2σ , Py = −(x1 + x2 )/2σ, (σ := 1/ρk) (9)

where −k2 = k1 = k, which turns out to be completely disjoint from those em-
ployed in the case when k2 = −k1 , and obeys the standard relations [x, Px ] =
[y, Py ] = i, [x, Py ] = [y, Px ] = 0. Actually Px , Py (the vorticity center coordi-
nates, up to the factor kρ) and x, y do not play prefixed roles so that, depending
on the interactions involved by the dynamical problem, they can be regarded
either as momenta or as position variables. If H does not contain further inter-
actions, the simplest choice is to consider vortex wave functions in the momentum
space picture, namely to treat Px , Py as coordinates. Then plane waves

ΦK (Px , Py ) = ei(Qx Px +Qy Py ) /2π (10)

diagonalize x2 + y 2 namely H, Qx and Qy being the eigenvalues of x = i∂Px

and y = i∂Py , respectively. Information on Px and Py is, of course, completely
missing since the pair cannot localize anywhere in the ambient plane.

3 Pair Quantum Dynamics in a Circular Box

The potential flow function associated to a single, vortex placed at zj -inside a
circular box is given by V (z, zj ) = −(kj /2π) Re ln R(zj − z)/(R2 − zz̄j ) , where
R is the box radius. From the kinetic energy functional H = (ρ/2) d2 r v 2 (r),
where v(r) = ∇V ∧ ez is the velocity field, r is the planar vector version of z,
and ez · r = 0 (|ez | = 1), one derives [5], [13] the vortex-pair Hamiltonian
 
H∗ = (4π)−1 [ρk1 k2 ln|(R2 − z1 z̄2 )/R(z1 − z2 )|2 + Σj ρkj2 ln 1 − |zj |2 /R2 ]. (11)

In the VA case (k1 = k = −k2 ) H∗ can be written as H∗ = (ρk 2 /4π) lnA,

where

A = |z1 − z2 |2 S1 S2 /[R2 S1 S2 + R4 |z1 − z2 |2 ] . (Sj := R2 − |zj |2 ) (12)

This is also available in the form A = 1/{ν/[2(A3 −A1 ]+ν 2 /[(A3 −ν)2 −A24 ]}
with ν := R2 /2σ, whose new variables
, -
A± = (2σPy ± ix)2 + (2σPx ± iy)2 /8σ, A3 = σ[Px2 + Py2 + (x2 + y 2 )/4σ 2 ]/4

(A± := A1 ±iA2 furnish A1 ) are recognized to be the generators of the algebra

su(1, 1) since they satisfy the equations [A3 , A± ] = ±A± , [A− , A+ ] = 2A3 when
using the commutators pertaining variables (9).
 Quantum Dynamics of Vortex–Antivortex Pairs 449

Such a su(1,1) formulation is important for two reasons. First, it allows one to
recognize explicitly the constant of motion A4 ensuring the model integrability.
A simple calculation shows, in fact, that the so-called Casimir operator Q =
A23 − A21 − A22 can be written as Q = A24 − 1/4, where A4 = (xPy − yPx )/2 =
(x21 + σ12 p21 − x22 − σ22 p22 )/4 with σj = 1/ρkj . Consistently, one can check that
[As , Q] = 0, (s = 1, 2, 3) namely [As , A4 ] = 0 so that [H∗ , A4 ] = 0. Second, this
fact suffices to identify the Hilbert space basis relying on which the VQD of the
bounded pair can be developed. Observing that Ψm (xj ; σ), defined as in Eq. (6),
fulfils the secular equation [x2j + σj2 p2j ]Ψm = σ(2m + 1)Ψm (|σj | = σ), one finds

A4 Ψm (x1 , x2 ) = 2σ(m1 − m2 ) Ψm (x1 , x2 ) (m := (m1 , m2 )) ,

where Ψm (x1 , x2 ) = Ψm1 (x1 ; σ)Ψm2 (x2 ; σ) with mj ∈ N is an element of the

Hilbert space basis. The ensuing HO picture strongly differs from the plane
wave basis (10) of the unbounded case. Another macroscopic quantum effect is
that 4σA4 ≡ |z1 |2 − |z2 |2 (the difference, up to a factor π, of the disk areas
relative to the radii |z1 |, |z2 |) quantally ranges in Z because of the HO form of
|zj |2 = x2j + σj2 p2j with σ(2mj + 1) ≤ R2 owing to the boundary confinement.

3.1 Spectral Structure of Low Energy States

We concentrate now on two special cases related to the system ground-states
and their weakly excited states. In second case, we show how the macroscopic
exchange symmetry (ES) characterizing H∗ (A is symmetric under zi ↔ zj ,
i = j) seems to prelude a level splitting. A complete description of the spectrum
properties which necessarily requires a preliminary analysis of the phase space
structure will be developed elsewhere. From Eq. (12) one recognizes the two
(main) ground-state configurations z1 ≡ z2 (→ A4 = 0) and |zi | ≡ R, |zj | < R
(→ A4 = 0) (the latter can be realized in two independent ways linked via
the ES), we will denote with C1 and C2 , respectively. Real vortices endowed
with finite cores cannot satisfy such conditions in that, below the core size,
pair annihilations take place (notice how, close to the boundary, one of the two
vortices is virtual: actually this process might/should be modeled in a different
way to account for a possible boundary microscopic structure). Nevertheless,
C1 , C2 can be used as reference configurations to describe weakly excited states
exempt from VA annihilations. The VA dynamics is governed by the equations
, -
iżj = kj 1/(z̄j − z̄k ) + zk /(R2 − zk zj∗ ) − zj /(R2 − |zj |2 ) /(2π), (k = j) (13)

derived from H∗ by using Eq. (2). Based on them, one can show that when A4 = 0
(→ |z1 | ≡ |z2 |) any solution reduces (up a suitable rotation around the origin) to
a pair of curves z1 (t), z2 (t) such that z2 = −z1∗ for each time t. Such curves are
closed since the boundary deviates vortices approaching it. When A4 = 0, low
energy states related to C1 correspond to vortices that are very close initially
and placed, e.g., at the box center (|xj (0)|  R, yj (0) = 0). After travelling
towards the boundary with an almost constant VA distance, vortices scatter (at
y1 = y2 R) in opposite directions ”captured” by their virtual companion, and
450 V. Penna

start to move close to the boundary; finally the pair is reconstituted when they
reach the position y1 = y2 −R thus realizing a periodic motion (exchanging
vortex initial positions inverts their initial velocities and the motion directions
but the curves remain the same).
The double nature affecting this case is confirmed by the fact that A
|z1 − z2 |2 [1 − R2 |z1 − z2 |2 /S1 S2 ] for R  |zj |, [since Sj R2 , dynamics depends
on |z1 − z2 |2 which involves a free particle motion as that accounted by Eq.
(10)], whereas A S1 S2 [1 − S1 S2 /R2 |z1 − z2 |2 ] for |zj | R, z2 ≡ −z1∗ (Sj 0,
|z1 − z2 |2 ≈ 4R2 entails an HO-like motion of vortices along the boundary).
For deciding which behavior dominates quantally, let us introduce q 2 = (x2 +
y 2 )/R2 , p2 = σ 2 (Px2 + Py2 )/R2 to rewrite A as

A ≡ (1 − q 2 − p2 )2 − δ 2 /4 + A2 /(4q 2 − A). (δ/2 := 2σA4 /R2 ) (14)

For small value of A the implicit function p(q; A) defined on  a suitable q

interval IA practically coincides with the circle-arc p = [1 − q 2 − δ 2 /4 + A]1/2
since the term with A2 is negligible. The latter √ contains a singularity which
drastically makes p(q; A) go to zero, for q → A/2. Physically, this represents
the excaped annihilation of the two vortices when they reconstitute the pair,
after travelling along the boundary. A simple approximation of this (classical)
behavior is given by p∗ (q; A) [e − q 2 − d/4q 2 ]1/2 which gives e = q12 + q22 ,
d = 4q1 q2 by imposing p∗ (qi ) = 0 (IA = [q1 , q2 ] where q12 = A/4, q22 1 − (A +
2 2

δ 2 /4)1/2 ).
The fact that the resulting classical orbits well mimic the exact ones suggests
to approximate the Schrödinger problem (SP) (14) through the effective one [p2 +
q 2 + d/4q 2 ]Ψ (x, y) = ΛΨ (x, y) [Λ = e(A) must be imposed]. Solving exactly SP
(14) requires a separate work due to its complexity; a possible approach based on
deriving the dynamical algebra of the SP with eigenvalue-dependent generators
is provided in Ref. [18]. The same approach is easily applicable to the effective
SP thanks to the further su(1, 1) realization [18] K± = (r ∓ icP )2 /4c − cg/4r 2 ,
K3 = c(P 2 +g/r 2 )/4+r 2 /4c, where P = (Px , Py ), r = (x, y). The basic presence
of the Coulomb term 1/r 2 , allows one to write the effective SP as K3 Ψ (x, y) =
(R4 Λ/σ)Ψ (x, y) provided g ≡ dR4 /σ 2 , c ≡ σ. The reduction of this secular
equation to a 1d radial problem by inserting r2 := x2 + y 2 , θ := arct(y/x) shows
how it is equivalent to the SP [(d2 /dz 2 ) − g/z 2 − z 2 /c2 + 4(n − J)/c]ψnJ (z) = 0
whose solutions are

ψnJ (z) = (−1)n DnJ 2/z(z 2 /c)(α+1)/2 exp[−z 2 /2c]Lα 2
n (z /c), α := −2J − 1.

The eigenstates of the effective SP are easily written in terms of such solutions,
up to a factor einθ diagonalizing the angular part of P 2 . This proves how con-
figurations C1 are expected to have a discrete spectrum whose states may have
a free particle behavior just locally.
We conclude showing that C2 excited states have a strongly quantized spec-
trum. Also, we consider the circumstances that allows the level splitting (LS)
suggested by the ES. Classically, C2 states are characterized by a vortex running
counter-clockwise along the boundary while the other weakly oscillates at the box
 Quantum Dynamics of Vortex–Antivortex Pairs 451

center; the vortex exchange caused by ES places an antivortex moving clockwise

at the boundary but leaves A unchanged. The fact that the ES action connects
distinct orbits close to a pair of isoenergetic minima in the phase space suggests a
possible LS. For a particle with energy E0 in a two-well potential field [19] the LS
d
is given by ΔE = E− − E+ ∝ exp[− 0 dx p(x, E0 )/], where E± = E0 ∓ ΔE/2
are the energies of the symmetrized well states ψ± and x = d is the inversion
point related to E0 . Setting A S1 S2 [1 − S1 S2 /R2 |zi |2 ] if zi  zj , and using
the A4 eigenstates Ψm (x1 , x2 ) constrained by 2mr + 1 ≤ R2 /σ, r = 1, 2, the
eigenvalues of the A operator are found to be

λ(, k) = s1 s2 {1 − s1 s2 /[R2 σ(2 + 1)]} ,

where sr = 1 − σ(2mr + 1)/R2 , and mi =  (mj = k  ) for the vor-

tex close to (far from) the boundary. Even if the same eigenvalue belongs to
the state Ψm (x2 , x1 ), obtained by the ES action, symmetrized states such as
Ψ± = Ψm (x1 , x2 ) ± Ψm (x2 , x1 ) are physically prohibited since an infinite poten-
tial barrier is generated by A2 /(4q 2 − A) in the momentum p(q, A) derived from
Eq. (14). Such a problem could be skipped if the A4 symmetry is lost. Changes of
the constant of motion A4 during the system evolution (caused, e.g., by external
interactions as well as by a boundary with R = const) allow, in fact, to connect
the two energy minima through state paths avoiding the potential barrier.

References
1. A. L. Fetter: Phys. Rev. 162, 143 (1967).
2. M. Rasetti, T. Regge: Physica A 80, 217, (1975); G.A. Goldin, et al.: Phys. Rev.
Lett. 67, 3499 (1991); V. Penna, M. Spera: J. Math. Phys. 33, 901 (1992).
3. R. Y. Chiao, A. Hansen et al.: Phys. Rev. Lett. 55, 2887 (1985); F.D.M. Haldane,
Y.-S. Wu: ibid. 55, 1431 (1985); G.A. Goldin et al.: ibid. 58, 174 (1987).
4. J.M. Leinaas: Ann. Phys. (N.Y.) 198, 24 (1990).
5. V. Penna: Physica A 152, 400 (1988);
6. A.M. Thompson, J.M.F. Gunn: Physica C 235, 2953 (1994).
7. R. Fazio, G. Schön: Phys. Rev. B 43, 5307 (1991); see also references therein.
8. Proceedings of the 20th International Conference on Low Temperature Physics, ed.
by R.J. Donnelly, [Physica B 194-196, (1994)].
9. R. Iengo, G. Jug, Phys. Rev. B 54, 13207 (1996); H. H. Lee, J. M. F. Gunn: ibid
57, 7892 (1998); D.J. Thouless et al: Int. J. Mod. Phys. B 13, 675 (1999);
10. J.C. Davis et al.: Phys. Rev. Lett. 69, 323 (1992); G.G. Ihas et al.: ibid. 69, 327
(1992); see also the articles by E. Varaquaux and R. Packard in the present volume.
11. Proceedings of the ICTP Workshop on Josephson Junction Arrays, ed. by H. A.
Cerdeira, S. R. Shenoy [Physica B 222, 336-370 (1996)].
12. F. Dalfovo et. al.: Rev. Mod. Phys. 71, 463 (1999); M. R. Matthews et. al.: Phys.
Rev. Lett. 83, 2498 (1999); K. W. Madison et. al.: Phys. Rev. Lett. 84, 806 (2000);
13. R. Lupini, S. Siboni: Nuovo Cimento A 106, 957 (1991).
14. W.M. Zhang, D.H. Feng and R. Gilmore: Rev. Mod. Phys. 62, 761 (1990).
15. V. Penna: Phys. Rev. B 59, 7127 (1999).
16. P. G. Saffman: Vortex Dynamics (Cambridge University Press, Cambridge, 1992).
452 V. Penna

17. L. Onsager: Nuovo Cimento Suppl. 6, 249 (1949).

18. A. Inomata, H. Kuratsuji and C. C. Gerry, Path Integral and Coherent States of
SU(2) and SU(1,1), (World Scientific, Singapore, 1992).
19. L.D. Landau, E.M. Lifsits Quantum Mechanics (Pergamon, Oxford, 1957).
Index of Topics

The reference to each item is an article (indicate by a numeral) in a chapter

(indicated by a roman numeral). For example V.7 means article 7 in Section 5.

Aharonov - Bohm effect: III.3

action functional: V.7
Bernoulli: V.1, VI.1
Berry phase: III.3
Biot - Savart law: I.1, III.1, VII.1, IV.6
Bloch state: VI.4, VIII.4
Bose - Einstein condensation: V.1, VI.1, VI.2, VI.3, VI.4, VIII.4
boundary conditions: III.1, III.4, IV.4
boundary layer: II.1, IV.4
complexity: VII.3
coolant: II.3
counterflow: I.1, II.1, IV.1, IV.3
coupled fluids: I.1, IV.2
Couette flow: III.4
crisis: II.4
critical velocity: II.2, II.7, VI.2, IV.4, V.1, V.2, V.7, VI.1, VIII.1
cryogenics: II.3
current sheet: VII.5
cutoff: III.1
dispersion relation: V.1, V.3
double quantum: VIII.1
drag: II.4, II.7, VI.1
Ekman cell: III.4
Euler - Tricomi equation: V.2
field theory: VIII.5
films: VIII.2
fluctuations: VI.4
Fokker - Plank equation: IV.8, IV.10
friction: I.1, III.2, III.4, IV.2, IV.4, IV.6
Gaussian model: IV.7
Ginzburg - Landau equation: V.4
grid turbulence: II.1, II.5, II.6, IV.1, IV.5
Gross - Pitaevskii equation: III.3, V.1, V.2, V.3, V.5, V.6, VI.1, VI.2, VI.4, VII.1
Hamiltonian: III.2
healing: V.1
helicity: VII.2, VII.3, VII.6
3
He: VIII.1
hot wire: II.1
HVBK equations: I.1, III.2, III.4
intermittency: II.7
454 Index of Topics

ions: II.5, V.1, VI.1

Iordanskii force: III.3
jet: IV.2
Josephson effect: VI.4
Kelvin’s theorem: III.2, VII.1
Kelvin waves: I.1, II.5, IV.1, IV.2, IV.6
kinematic viscosity: II.1, II.4, IV.1
kinetics: VI.3
Kolmogorov’s law: II.1, IV.1, IV.5
Kosterlitz - Thouless transition: IV.10
Lagrangian: VI.4
Landau levels: VIII.3
Landau velocity: II.5
Langevin equation: IV.8
levitating sphere: II.7
LHC: II.3
linkage: VII.2, VII.3
local induction approximation (LIA): I.1, III.1, IV.3
Mach number: V.2
Madelung transformation: V.1
magnetic fields: VII.1, VII.4, VII.5
Magnus force: III.3, V.6
MHD: VII.2, VII.4
millikelvin temperatures: II.5
Navier - Stokes fluids: VII.1, VII.2, VII.6
NMR: VIII.1
nonlinear Schroedinger equation (NLSE): I.1, III.3, V.1, V.2, V.3, V.5, V.7, VI.2,
VI.3, VII.1
nonlocal model: V.3
nonlocality: VII.6
nonlocal nonlinear Schroedinger equation: V.3
numerical methods: III.1, IV.2, VII.1
Orr - Sommerfeld equation: IV.4
phase slips: II.2, VI.4
phase transition: IV.10
phonons scattering: III.3
pinning: II.2
pipe flow: II.1
Poiseuille flow: IV.4
quantum Hall effect: VIII.3
quantum tunnelling: II.2
quenched transition: IV.10
potential flow: II.7
Rankine vortex: VII.1
reconnections: III.1, IV.1, IV.3, V.1, VII.1, VII.2, VII.4, VII.5
 Index of Topics 455

refrigeration: II.4
remanent vorticity: II.2
rotating condensate: VI.2
rotons: V.3
second sound: II.1, II.6, IV.5
self - consistent equations: IV.2
skyrmion: VIII.1
sphere: II.4, II.7
spin up: V.4
stability: I.1, III.4, IV.4, V.2, V.4
stochastic behaviour: IV.8, IV.9
strain: VII.6
superconducting magnets: II.3
T1 and T2 turbulent states: II.1
Taylor - Couette flow: III.4
Taylor - Proudman theorem: III.2
Thomas - Fermi approximation: VI.2
towed grid: II.1, II.6, IV.5
trapped condensate: VI.2
tropicity: VII.3
turbulence decay: II.1, II.5, II.6, IV.5, IV.6
vibrating grid: II.5
Vinen equation: II.5, IV.3, IV.6
vortex dynamics: III.3, IV.6
vortex filaments: III.1
vortex mill: II.2
vortex nucleation: II.2, V.1, V.2, V.3, V.4, V.7, VI.4, VII.1
vortex pair: IV.10, VIII.5
vortex ring: II.2, IV.2, V.1, V.3, V.7, VII.1
vortex tension: III.2, III.4
waves: I.1, II.5, IV.1, IV.2, IV.6
weak link: VI.4
weak turbulence: V.5
wetting: VIII.2
wind tunnel: II.1
WKB method: V.5
writhing number: VII.3