Mathematics of Bose–Einstein condensation

Marcin Napiórkowski

February 2022
Chapter 1

Introduction

The phenomenon of Bose-Einstein condensation was predicted theoretically

by Bose and Einstein in 1924. It is a state of matter in which, at low
enough temperatures, a macroscopic number of particles occupies the lowest
energy state. It is caused by the quantum statistics of bosons only and is
therefore purely quantum mechanical. However, for more than ten years,
Einstein’s prediction has been neglected as purely mathematical conclusion
for a fictitious system of non-interacting ideal gas and with little relevance
to real physics. The idea remained abstract for a long time, even for the
authors themselves. Einstein wrote: The theory is pretty but is there also
some truth to it?
In 1938, Kapitza [1], and independently Allen and Misener, discovered
the phenomenon of superfluidity - a frictionless flow of fluid in liquid 4 He.
This remarkable phenomenon can be observed by naked eye. However, it
cannot be understood in terms of classical physics. In the same year, London
[2] had the idea that superfluidity could be an experimental manifestation
of Bose–Einstein Condensation.
In 1941, L.D. Landau developed a phenomenological theory of super-
fluidity in terms of the excitation spectrum of 4 He. The first microscopic
theory of interacting Bose gases in the context of BEC was formulated in
1947 by Bogoliubov. In spite of the successful development in understanding
superfluidity in its early years, it was only after the realization of atomic
BEC in 1995 (by Cornell and Wiemann and by Ketterle, all got the Nobel
Prize) that the theoretical concepts of Einstein and Bogoliubov have been
experimentally confirmed.
The relation between superfluidity and BEC is not obvious until today.
In fact, to discuss the connection between these two one first needs to define
them. This is related to the concept of the off-diagonal long-range order
introduced by Penrose and Onsager.
The experimental realisation of BEC led to a renewed interest in the
theoretical and mathematical study of bosonic many-body systems. It would

1
not be an exaggeration to say that the basis of much of the subsequent work
has been laid in the seminal 2002 paper of Lieb and Seiringer [3] in which the
authors proved for the first time that the ground state of a dilute, trapped
gas exhibits Bose–Einstein condensation.
The goal of this lecture is to discuss some of the topics mentioned above
in a mathematically precise way. We will define objects and set up problems
in a rigorous way. Some proofs we will be done in details, others will only
be sketched. From a mathematical point of view, most of the results we
will discuss have been obtained very recently (up to 15 years). During the
course we will formulate open problems which could lead to research and/or
master/PhD theses.

2
Chapter 2

Quantum many-body
systems

2.1 The Hamiltonian

The central object that we will study throughout the course is the non-
relativistic many-body Hamiltonian given by
N
X X


HN = −∆xj + V (xj ) + λ wN (xj − xk ). (2.1)
j=1 16j<k6N

It acts on the the Hilbert space HN = N 2 3

N
sym L (R ) which consists of func-
tions ψ(x1 , x2 , . . . , xN ) that are square integrable, i.e.
Z
|ψ(x1 , x2 , . . . , xN )|2 dx1 . . . dxN < ∞

and, since we consider bosons, are symmetric under the exchange of two
particles, i.e.

ψ(x1 , . . . , xi , . . . , xj , . . . , xN ) = ψ(x1 , . . . , xj , . . . , xi , . . . , xN )

for any i, j = 1, . . . , N . Here xj ∈ Rd stands for the coordinate of the j-th

particle. We have set the spatial dimension of the problem to be three as
we will focus on that case for the majority of the course. But, in principle,
physically relevant systems could be also one or two dimensional.
The Hamiltonian (2.1) describes the system of N particles that interact
through a two-body potential wN that in principle can depend on N . It
could be positive (repulsive) but it could also have an attractive part. A
purely repulsive potential is usually easier to handle. Most of the time we will
work with ’nice’ (smooth, quickly decaying), symmetric, repulsive potentials
which will make our live easier. The factor λ in front of the interaction terms

3
is assumed to be positive, i.e. λ > 0. It represents a coupling constant that
could also be N dependent. We will soon come back to this.
Finally, the one-body term consists of the kinetic energy operator −δxj
of the j-th particle and a external potential V (xj ) which acts on the j-th
particle too. On should think of the V as being a trapping potential which
keeps the particles located in a certain region of space. One example would
be a trapping potential of the form V (x) = x2 which represents a harmonic
trap. Another example would be a potential that is of the form

0 if x ∈ Ω
V (x) =
∞ if x 6∈ Ω
In this case the particles would be confined in Ω ⊂ R3 .

2.2 Quantities of interest

Given the Hamiltonian HN of a quantum system, there are many questions
one can try to address. The first one might concern its ground state energy
E0 (N ) = inf (ψ, HN ψ).
ψ∈D(HN ),kψk=1

If E0 (N ) is an eigenvalue, the corresponding ground state wave function ψ0

is determined by Schrödingers equation HN ψ0 = E0 (N )ψ0 .
One might be interested also in different observables. Recall, that the
expectation value of an observable A in the state ψ is given by
hAiψ = (ψ, Aψ).
More generally, if the system is at some positive temperature T > 0, one
would like to compute the free energy of the system, given by
F = −T ln Tr e−HN /T .
We choose units such that Boltzmanns constant equals 1, and shall often
write T = 1/β. The trace is over the physical Hilbert space, of course, re-
specting symmetry constraints arising from the indistinguishability of par-
ticles. The equilibrium state at temperature T is the Gibbs state
ρβ = e−β(HN −F ) .
It is normalized to have Tr ρβ = 1. The expectation value of an observable
A is then given by
 
Tr Ae−HN /T
hAiρβ = = Tr(Aρβ ).
Tr e−HN /T
For large particle number, it is usually hopeless to try to calculate ρβ
directly, but one can try to investigate properties of the reduced n-particle
density matrices, obtained by taking the partial trace of ρβ over N − n
variables. We will introduce some of these objects shortly.

4
2.3 Grand-canonical ensemble
It is often convenient not to fix the particle number N , but rather work in
the grand-canonical ensemble, where one takes a certain average over the
number of particles in the system. For simplicity, consider a system of just
one species of particles. The N -particle Hilbert space, HN , is then the set of
square-integrable functions that are totally symmetric under permutations.
In the grand-canonical ensemble, one has as Hilbert space the Fock space
∞
M
F= HN .
N =0

Here H0 = C by definition and the corresponding vector is called the vacuum

vector. As the Hamiltonian on Fock space one simply takes
∞
M
H= HN .
N =0

One typically sets H0 = 0 (vacuum has no energy).

For µ ∈ R, the grand canonical potential is defined as

J = −T ln TrF e−β(H−µN )

where N denotes the number operator, i.e.,

∞
M
N = n.
n=0

Since H is particle number conserving, we can also write this as

X
J = −T ln z N TrHN e−βHN ,
N ≥0

where z = eβµ is called the fugacity.

The grand-canonical Gibbs state is

ρβ,µ = e−β(H−µN −J) .

The chemical potential µ is adjusted to achieve a given average particle

number hN i. The latter equals

∂
hN i = Tr N ρβ,µ = − J.
∂µ

5
2.4 BEC in ideal gas
Let us now consider an non-interacting (ideal) Bose gas, that is confined in
a box of side length L, i.e. Λ = [0, L]3 . The Hamiltonian is given by
N
X
n-in
HN = −∆xi
i=1

and let us assume periodic boundary conditions. The spectrum of −∆ equals

 2
2π
(n2x + n2y + n2z ) (2.2)
L

with (nx , ny , nz ) ∈ Z3 . The corresponding eigenstates are plane waves eipx

3
with p ∈ 2π L Z . Let us order (and denote) the eigenvalues in (2.2) by

e0 ≤ e1 ≤ e2 ≤ . . . .

On Fock space we have X

H= ei a∗i ai
i≥0

and also X
β(H − µN ) = εi a∗i ai
i≥0

where εi = β(ei − µ).

We would like to calculate
∗
P
ln Tr e− i≥0 εi ai ai .

The spectrum of i≥0 εi a∗i ai is of the form i≥0 εi ni , with ni ∈ {0, 1, 2, . . .}.
P P
Summing over all possible occupation numbers is the same as summing over
all eigenstates, hence we have
P ∗ 1
Tr e−
YX Y
i≥0 εi ai ai = e−εi ni = .
n
1 − e−εi
i i

Here we have to assume that εi > 0 for all i for the geometric series to
converge. Then
∗
P
ln Tr e− i≥0 εi ai ai =
X
− ln(1 − e−εi ).
i

Thus the grand-canonical potential equals

2
X
J =T ln(1 − e−β(p −µ) ).
3
p∈( 2π
L )
Z

6
Note that εi > 0 can be achieved by taking µ < 0. This is not really
a restriction, however, as any particle number can be achieved even for
negative µ. In fact, the average particle number equals
∂ X 1
hN i = − J = β(p 2 −µ) .
∂µ e
3|
− 1}
2π
p∈( Z) L
{z
ha∗p ap i

Here the summands are ha∗p ap i, the average occupation number of momen-
tum [Link] µ varies between (−∞, 0), clearly hN i varies between (0, ∞).
We now perform a thermodynamic limit L → ∞. The sum over p can
then be interpreted as a Riemann sum for the corresponding integral. In
fact, Z
1 X 1
−→ dp
L3 3
(2π)3 R3
2π
p∈( L Z)

as L → ∞. The thermodynamic pressure of the system is thus

Z
J T 2
P = − lim 3 = − 3
ln(1 − e−β(p −µ) )dp
L→∞ L (2π) R3
and the average density equals
hN i
Z
1 1
ρ= = dp. (2.3)
L3 (2π)3 R3 eβ(p2 −µ) − 1
Notice that the density stays bounded as µ → 0! That is
Z
1 1
ρc (β) := lim ρ = 3 βp 2 dp < ∞.
µ→0 (2π) R3 e −1
Hence it appears that the density of the gas can never exceed ρc (β). What
is happening here? Recall that µ has to be chosen as to fix the density and,
hence, has to depend on L, in general. If ρ < ρc (β), then µ(L) → µ < 0 in
the thermodynamic limit. But when ρ > ρc (β), then µ(L) has to tend to
zero as L → ∞. In this case, the limits L → ∞ and µ → 0 must be taken
simultaneously and, in particular, do not commute!
In fact, if ρ > ρc (β), then µ is asymptotically equal to
µ = −(βL3 (ρ − ρc (β)))−1 as L → ∞.
For this value of µ we see that
ha∗0 a0 i 1 1
lim = lim 3 −βµ = ρ − ρc (β). (2.4)
L→∞ L3 L→∞ L e −1
Thus the zero momentum state is occupied by a macroscopic fraction of
all the particles. This phenomenon is called Bose–Einstein Condensation
(BEC). It occurs when ρ > ρc (β) or, equivalently, when
4π
T < Tc (ρ) = ρ2/3 (2.5)
ζ(3/2)2/3

7
since ρc (β) = ζ(3/2)(4π)−3/2 β −3/2 . Here ζ denotes the Riemann zeta func-
tion X 1
ζ(z) = .
kz
k≥1

In other words, BEC occurs below the critical temperature.

Problem 2.4.1. Derive (2.5).
We note that only the zero momentum mode is macroscopically occu-
pied, and the other occupations are much smaller. The smallest positive
eigenvalue of the Laplacian equals (2π/L)2 , and
1
∼ L2  L3 for large L.
eβ(2π/L)2 − 1
BEC represents a phase transition in the usual sense: the thermodynamic
functions exhibit a non-analytic behavior. Consider, for instance, the free
energy, which is given in a standard way as the Legendre transform of the
pressure. Specifically, the free energy per unit volume equals
Z
T 2
f (β, ρ) = µρ + ln(1 − e−β(p −µ) )dp,
(2π)3 R3
where µ is determined by (2.3) if ρ < ρc (β), and µ = 0 if ρ ≥ ρc (β). In
the latter case, we see that f (β, ρ) does not actually depend on ρ, and is
constant for ρ > ρc (β). In particular, f is not analytic. Intuitively, what is
happening as one increases the density beyond ρc (β) is that all additional
particles occupy the zero momentum mode and hence do not contribute to
the energy or the entropy, hence also not to the free energy.
Finally, let us just mention, that the arguments that we have mentioned
above do not work in dimensions d = 1, 2. In fact, in that case the ideal
Bose gas exhibits BEC only in the ground state (T = 0).

2.5 BEC in interacting systems - definition

2.5.1 Reduced densities
The one-particle reduced density associated with the N -body wave function
ψN is defined by
(1)
γψN := N Tr2,3,...,N |ψN ihψN |, (2.6)
where |ψN ihψN | denotes the orthogonal projection onto ψN and Tr2,3,...,N
is the partial trace over the last (N − 1) particles. In other words, the
one-particle reduced density γψ1 N is defined as the non-negative trace-class
operator on L2 (R3 ) with integral kernel
Z
(1)
γψN (x; y) = N dx2 . . . dxN ψN (x, x2 , . . . , xN )ψN (y, x2 , . . . , xN ). (2.7)

8
 (1)
Notice that we chose the normalization Tr γψN = N .
Analogously, for k = 2, 3, . . . , N , we can define the k-particle reduced
density associated with ψN by
 
(k) N
γψN := Trk+1,...,N |ψN ihψN |. (2.8)
k

The integral kernel of the k-particle density matrix is given by

(k)
γψN (x1 , . . . , xk ; y1 , . . . , yk ) =
 Z
N
dxk+1 . . . dxN ψN (x1 , . . . , xk , xk+1 . . . , xN )ψN (y1 , . . . , yk , xk+1 , . . . , xN ).
k
(k) N


The normalization is such that Tr γψN = k .
(k)
Clearly, for k < N , the k-particle reduced density γψN does not contain
(k)
the full information about the system. Still, γψN is enough to compute the
expectation of any k-particle observable. For example, if k = 1 and J (1) is
(1)
a one body operator J (1) = N
P
i=1 Ji , then

N
(1) (1) (1) (1)
X
(ψN , J (1) ψN ) = (ψN , Ji ψN ) = N (ψN , J1 ψN ) = Tr J1 γψN .
i=1

The reduced densities can be lifted to the Fock space setting. Note that
using (B.3) we see that the integral kernel (2.7) can be written as
(1)
γψN (x; y) = (ψN , a∗y ax ψN ).

This leads to the general definition that the one-body reduced density matrix
is the one-body operator (acting on the one-body Hilbert space) γ (1) defined
through
(g, γ (1) f ) = ha∗ (f )a(g)i, ∀f, g ∈ L2 (R3 ).
Here h·i denotes the expectation value in any state. In particular this defini-
tion applies to any state on Fock space, not only thermal equilibrium states.
One can also consider states of definite particle number, and hence recover
the definition for the canonical ensemble.

2.5.2 Definition of BEC

Let γ (1) be the reduced one-body matrix of a bosonic system. It is a positive,
trace class operator. Thus, it admits a spectral decomposition with respect
to an orthonormal basis {ui } (cf. (A.2)) of the form
X
γ (1) = λi |ui ihui |,
i

9
with λi > 0. According to the (generally accepted) definition of Penrose and
Onsager, Bose–Einstein Condensation occurs when γ (1) has an eigenvalue of
order N (or hN i). Note that this definition is independent of the fact,
whether the system under consideration is interacting or not.
Let us now check, how this definition relates to the concept of BEC
derived for the ideal Bose gas. In (2.4) we have shown that above the
critical density
ha∗0 a0 i supkf k=1 ha∗ (f )a(f )iβ,µ
0 < ρ − ρc (β) = lim < lim .
L→∞ L3 L→∞ L3
In particular, supkf k=1 ha∗ (f )a(f )iβ,µ is the largest eigenvalue of γ (1) so in-
deed it has to be of order N (as when taking the thermodynamic limit we
assume ρ = N/L3 ).

2.5.3 Off-diagonal long-range order

Consider the momentum distribution in a given state defined through
np := ha∗p ap i.
As, by definition, ap = dxax e−ipx /L3/2 , we have
R
Z Z
∗ 1 ∗ −ip(x−y) 1
hap ap i = 3 dxdyhax ay ie = 3 dxdyγ (1) (x, y)e−ip(x−y) .
L L
Let us assume that the system under consideration is translation invariant.
Then we have γ (1) (x, y) = γ (1) (x − y) and thus
Z
1
ha∗p ap i = 3 dxdyγ (1) (x − y)e−ip(x−y) . (2.9)
L
Inverting this relation (inverse Fourier transform), we obtain
1 X
γ (1) (x − y) = 3 hnp ieip(x−y) .
L 3
p∈(2π/L)Z)

Now, recall our analysis of BEC in the ideal case. It shows that when there
is BEC we have (in the thermodynamic limit)
hnp i = δ0 N0 + ñp
where the singular term arises from the macroscopic occupation of the zero
momentum state. Thus, taking the thermodynamics limit in (2.9), we obtain
Z
γ (1) (x − y) = N0 + dp ñp eip(x−y) .

By the expected regularity of ñp and the Riemann-Lebesgue lemma we ob-

tain
γ (1) (x − y) → N0 , as x − y → ∞. (2.10)
This behaviour is called off-diagonal (because x 6= y) long-range (because
x − y → ∞) order of the one-body density.

10
2.6 The curse of dimensionality
We shall now argue that, due to its high dimensionality, the many-body
Schrödinger equation is impossible to solve numerically at a high precision
for most physical systems of interest.
To this end let us consider a much simpler, discrete model - a quantum
spin system. We place a quantum (say, 1/2) spin on each of the N sites of a
square lattice. Since each spin is described the Hilbert space C2 , the Hilbert
space of the system is given by the tensor product ⊗x C2 where x enumerates
all the N sites. Thus, the dimension of the Hilbert space is 2N . In other
words, to encode a general quantum state we need 2N coordinates. Assume
we want consider a square lattice with length 20 and thus with 400 sites
altogether. Now, let us assume very optimistically, that one number can be
encoded using one bit which is carried on one elementary particle. Thus,
we would need 2400 elementary particles just to encode on the computer
one state of our quantum spin system. How much is 2400 ? According to
recent estimates of astrophysicists, there are 1086 elementary particles in
the universe. Since 2400 > 1086 , in principle we are not able to encode a
quantum state on the computer without even doing any operations on it.
That was a crude estimate for a spin system. The models we want to
consider are continuous and consist of thousands of particles (like in modern
experiments involving Bose–Einstein Condensates). To put the problem on
a computer one needs to discretize it and the example discussed above shows
that the complexity of this problem will be exponential.
All this forces us to consider effective theories. An effective theory is
a proposal to describe a system using less degrees of freedom. Of course,
there is price to pay - one looses information about the system. Let us
illustrate this idea using an example from classical physics. In classical
mechanics, the microscopic theory is given by Newton’s equations of motion.
Solving these equations would in principle give us information about all
positions and momenta of all particles involved. If one wants to understand
the behaviour of air in a room, then solving Newton’s equations is not a good
idea to say the least. But do we really need to know all the positions and
velocities of all particles? Maybe it would be enough to know the position
and momentum of a typical particle - that is, the probability that the particle
occupies a given very small region of the phase space (mathematically the
volume element dxdp) at an instant of time. We would thus be looking
for a probability distribution f (t, x, p) which now depends on only three
variables. Ludwig Boltzmann derived an equation for f which is now called
the Boltzmann equation. A rigorous justification of this equation starting
from Newton’s microscopic theory is still an active area of research and this
field of mathematical physics is called kinetic theory.
In this course we will try to understand to most important properties of
quantum many-boson systems by rigorously justifying effective theories. We

11
will be interested in effective descriptions of the the ground state properties,
excitation spectrum, effective dynamics. An effective theory usually is some
kind of limiting theory in a certain regime. In the next section we will
discuss the scaling regimes we will be studying.

2.7 Scaling regimes

Recall the Hamiltonian in (2.1) given by
N
X X


HN = −∆xj + V (xj ) + λ wN (xj − xk ),
j=1 16j<k6N

NN
defined on HN = 2 3
sym L (R ). We want to find an effective description
of the model. Keeping in mind that we are interested in thermodynamic
quantities, we have to remember that we we will be taking the N → ∞
limit. We observe that the kinetic energy (or, more generally, the one-body
operator) is of order N as it consists of N terms which are independent of N .
If we assume that the two-body interaction is N independent, i.e. wN ≡ w,
then the interaction term is of order N 2 (for large particle numbers) as there
are N (N − 1)/2 terms in the double sum. In the limit when N → ∞ the
interaction term would thus become dominant. But we want to keep the
kinetic term as, in some sense, it is more quantum than the pure interaction
(recall, BEC has been computed for the ideal gas). Thus, the idea is to
balance these two terms (at least, as far as the naive asymptotics in N
would suggest). This motivates setting λ = O(N −1 ) and this scaling is
called mean-field scaling. The simplest mean-field Hamiltonian is of the
form
N
X
1 X
HN = −∆xj + V (xj ) + w(xj − xk ).
N
j=1 16j<k6N

Note, that in this setup there are two length scales. The first one is given by
the trapping potential V and basically tells us where the particles can move.
It is important for the kinetic term, as it has implications in the spectral
gap of the one-body operator.
But for now let us focus on the length scale set by the interaction. When
wN ≡ w, then it is independent of N and therefore is of order O(1) (with
respect to N ). In fact, the length scale of the interaction is characterized
by a parameter called the scattering length. To define the scattering length,
we consider the two-body scattering problem. We assume the two-body
interaction is radial, positive and has a finite range. Consider the zero-
energy scattering equation
1
− ∆f + wf = 0 (2.11)
2

12
with the boundary condition f (x) → 1 as |x| → ∞. Under the assumption
of compact support of w one can show that for |x| sufficiently large, the
scattering solution satisfies
a0
f (x) = 1 − (2.12)
|x|
for a constant a0 > 0. This constant is called the scattering length. Equiv-
alently, the scattering length a0 can also be defined through the integral
Z
8πa0 = w(x)f (x)dx (2.13)

where f (x) is the solution of (2.11). From the point of view of physics, the
scattering length a0 measures the effective range of the interaction potential;
two quantum mechanical particles interacting through the potential w, when
they are far apart, feel the other particle as a hard sphere with radius a0 (in
particular, the scattering length of a hard sphere potential coincides with
the radius of the sphere).
Problem 2.7.1. Show that (2.13) holds true.
Problem 2.7.2. Consider the hard sphere potential. Show that the scatter-
ing length is in this case equal to the radius of the hard sphere.
Problem 2.7.3. Let f be the solution of the scattering equation for a posi-
tive interaction potential. Show that 0 ≤ f ≤ 1. Deduce that
R
w
a0 ≤ .
8π
Having defined the scattering length, let us come back to the mean-
field model. Assuming the trapping potential restricts the particles to a
volume of order O(1), we see that the density is of order N and the average
distance between the particles is N −1/3 . Thus it is much smaller that the
effective range of the interaction. In other words, each particle feels and
interacts with many other particles, but the strength of interactions is weak
(of order O(1)). This situation corresponds to a high density regime where
the particles meet very often but interact only a little bit each time.
To model a different situation, which will call the dilute regime (a typical
situation in experiments with ultracold gases), we can scale the interaction
and make it N dependent. Assume wN (x) = N 2 w(N x). By rescaling x =
N x̃ we see from (2.11) that
 
1 1
− 2 ∆x̃ + w(N x̃) f (N x̃) = 0
N 2
Thus f (N ·) solves the scattering equation
 
1
−∆ + N 2 w(N ·) f (N ·) = 0.
2

13
It follows that as |x| → ∞ we have

a0 a0 /N
f (N x) = 1 − =1−
N |x| |x|

and thus a = a0 /N is the scattering length of the rescaled potential wN (x) =

N 2 w(N x). This scaling corresponds to the dilute limit because the effective
range of the interaction is O(N −1 )  N −1/3 is much smaller than the av-
erage inter-particle distance. Thus, this scaling models a very short range
interaction that is very strong. Indeed, for a nice function w, the scaled wN
converges in a vague sense to a Dirac delta. Each particle encounters rare
but strong collisions. This scaling is called the Gross–Pitaevskii limit.
Finally, one can interpolate between these two extreme cases. This is
done by introducing β ∈ [0, 1] (do not confuse with inverse temperature).
For a general β we define

wN (x) = N 3β w(N β x).

Now, heuristically, the effective range of the interaction is of order N −β

(heuristically, because the scattering length does not scale nicely for inter-
mediate β’s). A special case corresponds to β = 1/3 (or, more generally
β = 1/d in d dimensions) because for β < 1/3 the average distance between
the particles is much bigger than the effective range of interaction (mean-
field limit), while for β > 1/3 the average distance between the particles
is much smaller than the effective range of interaction (GP/dilute limit).
These two regimes are very different but they lead to somewhat similar ef-
fects. This similarity is both very useful and confusing. Generally speaking,
the larger β is, the more difficult the problem is.
Finally, let us mention a slightly different scaling. While the setup de-
scribed above fits well when one wants to describe a trapped gas o bosons, a
different approach is usually considered when one wants to describe the ther-
modynamic in its classical formulation, that is when the system is restricted
to a box of length L and in the end one that the limit N → ∞, L → ∞ with
fixed ρ = N/L3 . In this case, the dilute limit is characterized by the relation

aρ1/3  1, (2.14)

which means that the effective range of the interaction a is much smaller
than the average distance between the particles ρ−1/3 .

14
Chapter 3

Ground state properties

We shall now discuss the properties of the ground state energy (i.e. T = 0)
of a system of interacting bosons, i.e.

E0 (N ) = inf (ψ, HN ψ)
ψ∈L2sym (R3N ),kψk=1

where
N
X
1 X
HN = −∆xj + V (xj ) + wN (xj − xk )
j=1
| {z } N − 1 16j<k6N
=:hxj

with wN (x) = N 3β w(N x).

3.1 Mean-field scaling

3.1.1 Ground state energy
Let us first consider the mean-field scaling, i.e. when β = 0. The mean-field
Hamiltonian is of the form
N
X
1 X
HN = −∆xj + V (xj ) + w(xj − xk ). (3.1)
N −1
j=1 16j<k6N

Recall, that the mean-field scaling correspond to a situation when the in-
teraction between the particles is frequent, but weak. We know that the
ground state of a non-interacting system (i.e. when w ≡ 0) is given by a
product state ũ⊗N (x1 , . . . , xN ) := ũ(x1 ) · · · ũ(xN ) where ũ is the (normal-
ized) ground state of the one-body operator. In the mean-field scaling the
interactions are supposed to be weak, so a simple minded argument suggests
that maybe the ground-state will not be changed dramatically and will be
close, in some sense, also to a product state.

15
 Let us make this ansatz, i.e. ψ(x1 , . . . , xN ) = u⊗N (x1 , . . . , xN ) for some
normalized function u ∈ L2 (R3 ). Computing the energy leads to the follow-
ing formula
(u⊗N , HN u⊗N ) = N E(u)
where
ZZ
1
E(u) = (u, hu) + w(x − y)|u(x)|2 |u(y)|2 dxdy. (3.2)
2 R3 ×R3

The functional (3.2) is called the Hartree functional.

Problem 3.1.1. Derive the Hartree functional.
Thus we obtain the following upper bound on the ground-state energy
E0 (N ) = inf (ψ, HN ψ) ≤ N inf E(u). (3.3)
ψ∈L2sym (R3N ),kψk=1 u∈L2 (R3 ),kuk=1

Problem 3.1.2. Let us assume that the Hartree functional has a minimizer
and let us denote it by u0 . It satisfies the so-called Hartree equation
hu0 + |u0 |2 ∗ w u0 = ε0 u0


(3.4)
R
where (f ∗ g)(x) = R3 f (x − y)g(y)dy denotes the convolution and ε0 is a
Lagrange multiplier due to the normalization constraint.
Note that in general there is no reason to expect that minimizers will be
unique. Uniqueness can be broken because of the one-body Hamiltonian h
or due to the interaction if it has an attractive part.
Let us denote by eH the infimum of the Hartree functional, i.e.
eH := E(u0 ).
From (3.3) we obtain
E0 (N )
≤ eH .
N
A natural question whether a matching lower bound holds. This is indeed
true.
Theorem 3.1.3 (Convergence of ground-state energy). Under appropriate
assumptions on h and w one has
E0 (N )
lim = eH .
N →∞ N
We will not state the most general assumptions on h and w that were
used to prove the most general version of Theorem 3.1.3 given by Lewin–
Nam–Serfaty–Solovej in 2014. Let us just mention that the one-body oper-
ator h can be much more general than in (3.1).
We will give a proof in a simpler case. We follow the proof of Lewin
(2015). Let us start with the simplest case when the following two conditions
are satisfied:

16
 • h is symmetric and positive preserving: (u, hu) ≥ (|u|, h|u|);

• w is positive-definite, i.e. w ≥ 0 and ŵ ≥ 0.

These two properties can be used through the following two lemmas.
Lemma 3.1.4 (Hoffmann–Ostenhof inequality). Let h be symmetric and
positive preserving. Then, for any bosonic N -body wave-function ψN we
have
N
X √ √
(ψN , hxi ψN ) ≥ N ( ρψN , h ρψN )
i=1
where ρψN is the one-body density defined by
Z
ρψN (x) = |ψN (x, x2 , . . . , xN )|2 dx2 . . . dxN .

Proof. We know from the definition of the reduced one-body matrix that
N
(1)
X X X
(ψN , hxi ψN ) = Tr(hγψN ) = nk (uk , huk ) ≥ nk (|uk |, h|uk |)
i=1 k k

(1) P
where we used the decomposition γψN = k nk |uk ihuk | and the assumption
on h. For real functions u1 , u2 we have
q q 
(u1 , hu1 ) + (u2 , hu2 ) = (u1 + iu2 , h(u1 + iu2 )) ≥ 2 2 2 2
u1 + u2 , h u1 + u2 .

Using this inductively we obtain

 
N
X X sX sX
(ψN , hxi ψN ) ≥ nk (|uk |, h|uk |) ≥  nk |uk |2 , h nk |uk |2  .
i=1 k k k

(1)
nk |uk |2 = γψN (x, x) = N ρψN (x) we get the desired
P
Using the fact that k
result.

Lemma 3.1.5 (Estimating the two-body interaction by a one-body term).

If 0 ≤ ŵ ∈ L1 (R3 ), the for all η ∈ L1 (R3 ) we have
N ZZ
X X 1 N
w(xi − xj ) ≥ η ∗ w(xi ) − w(x − y)η(x)η(y)dxdy − w(0).
2 R3 ×R3 2
1≤i<j≤N i=1

Proof. We use
ZZ Z
w(x − y)f (x)f (y)dxdy = (2π) 3/2
ŵ(k)|fˆ(k)|2 dk ≥ 0
R3 ×R3 R3
PN
with f = i=1 δxi − η and expanding all terms.

17
 We are ready to give the proof of Theorem 3.1.3.

Proof. Take η = N ρψN . Then the two lemmas above imply

√ N w(0) N w(0)
(ψN , HN ψN ) ≥ N E( ρψN ) − ≥ N eH − .
2(N − 1) 2(N − 1)
Minimizing over ψN and recalling the upper bound gives the theorem.

3.1.2 Ground state

Let us now give a sketch of the proof that the ground state in the mean-field
scaling exhibits Bose–Einstein Condensation. In fact, the proof below will
also give the statement about the ground state-energy as a by-product.
Before we start let us introduce a rescaled definition of the reduced
density matrices. Let us assume for the rest of this subsection that the
reduced density matrices are normalized differently than in (2.5.1), i.e. we
shall assume for now that
(k)
γψN (x1 , . . . , xk ; y1 , . . . , yk ) =
Z
dxk+1 . . . dxN ψN (x1 , . . . , xk , xk+1 . . . , xN )ψN (y1 , . . . , yk , xk+1 , . . . , xN ).

(k)
The normalization is then such that Tr γψN = 1.
Theorem 3.1.6 (Convergence of ground states). Assume that h and w
satisfy the inequality
−C1 (T1 + T2 ) − C ≤ w(x1 − x2 ) ≤ C2 (T1 + T2 ) + C
for some constants 0 ≤ C1 , C2 < 1. Assume V (x) → ∞ as |x| → ∞ and
that w is symmetric and smooth enough. Let ψN be the ground state for HN
(n)
and γψN its n-th reduced density matrix with the normalization that
Then there exists a subsequence and a probability measure µ on the set
of minimizers of the Hartree functional M such that
Z
(k)
γψN → |u⊗k ihu⊗k |dµ(u) (3.5)
j
M

strongly in the trace-class as Nj → ∞, for all k ≥ 1.

The probability measure µ describes the fragmented Bose–Einstein con-
densation. For instance, if there are only two Hartree minimizers, then µ
will give their relative occupations. The simplest case is when M = {u0 },
where there will always be complete Bose–Einstein condensation on u0 .
The main tool for proving the theorems is the quantum de Finetti theo-
rem. The latter is an abstract result which says that, at the level of reduced
density matrices, only product states remain in the limit N → ∞, for any
sequence of bosonic states.

18
Theorem 3.1.7 (quantum de Finetti). Let H be a separable Hilbert space
and let Γ(n) be a sequence of positive, trace-class operators satisfying

Γ(0) = 1, TrHnsym [Γ(n) ] = 1, Trn+1 [Γ(n+1) ] = Γ(n)

where Trn+1 is the partial trace with respect to the last variable in Hn+1 .
Then there exists a unique Borel probability measure µ on the sphere SH =
{u ∈ H : kuk = 1} of H, invariant with respect to multiplication by a phase,
such that Z
(n)
Γ = |u⊗n ihu⊗n |dµ(u)
SH

We will not give a proof of this statement now.

Proof of Theorem 3.1.6. We start by writing

E0 (N ) 1 (1) 1 (2)
= (ψN , HN ψN )HN = TrH [T γψN ] + TrH2sym [wγψN ]
N N sym 2
1 (2)
= TrH2sym [(T1 + T2 + w)γψN ]
2
Notice that after this rewriting the the N -dependence now appears only in
(2)
γψN . To describe the limit of this object we will use the quantum de Finetti
theorem.
(n)
To this end notice that, by definition, the sequences γψN are bounded
uniformly (by 1) in trace class. This allows us to use the Banach–Alaoglu
theorem, to find a converging subsequence. Indeed, the Banach–Alaoglu
theorem states that, if X is a separable Banach space, then to any sequence
(x∗n ) in X ∗ (the dual of X) which is bounded, i.e. kx∗n k ≤ M , there exists a
weak-∗ convergent subsequence (x∗nk ). Weak-∗ convergent means that there
exists a x∗ ∈ X ∗ such that x∗nk (x) → x∗ (x) as k → ∞ for all x ∈ X.
Moreover, kx∗ k ≤ M .
In our case X ∗ is the space of trace class operators. X is then the space
of compact operators. Thus for any compact operator Kn on Hn we have
(n)
TrHn [γψN Kn ] → TrHn [γ (n) Kn ]

along a subsequence. We will show that this convergence is actually strong.

To this end it is enough (this is easy to prove in Hilbert space setting, for
trace class it is not so obvious, but still true) to show that Tr[γ (n) ] = 1, i.e.
no mass is lost in the limit.
Using the assumptions of the theorem, we compute
1 (2)
eH ≥ Tr 2 [(T1 + T2 + w)γψN ]
2 Hsym
1 (2) (2)
≥ (1 − C1 ) TrH2sym [(T1 + T2 )γψN ] − C Tr[γψN ]
2
(1)
= (1 − C1 ) TrH [T γψN ] − C.

19
 (1)
It follows that TrH [T γψN ] is uniformly bounded and thus so is TrH [(T +
(1)
C0 )γψN ] which by the cyclicity if the trace yields (possibly, up to a further
subsequence) that
(1)
(T + C0 )1/2 γψN (T + C0 )1/2 *∗ (T + C0 )1/2 γ (1) (T + C0 )1/2

for some C0 . Consequently,

(1) (1)
1 = TrH [γψN ] = TrH [(T + C0 )−1 (T + C0 )1/2 γψN (T + C0 )1/2 ]
→ TrH [(T + C0 )−1 (T + C0 )1/2 γ (1) (T + C0 )1/2 ] = TrH [γ (1) ]

since (T + C0 )−1 is by the assumptions on V (x) a compact operator. The

fact that TrHn [γ (n) ] = 1 follows from a similar argument using the fact that
 
n
(1) 1 X (n)
TrH [T γψN ] = TrHn  Tj γψN 
n
j=1

is also uniformly bounded in N and that nj=1 Tj also has compact resolvent
P
which allows for similar argument.
Thus we have proven for any n ∈ N that
(n)
γψN → γ (n)

strongly in trace class. Testing this convergence against a bounded operator

Bn+1 = Bn ⊗ 1 we obtain
(n+1)
TrHn+1 [γψN (Bn ⊗ 1)] → TrHn+1 [γ (n+1) (Bn ⊗ 1)] = TrHn [(Trn+1 γ (n+1) )Bn ]

but also
(n+1) (n+1) (n)
TrHn+1 [γψN (Bn ⊗1)] = TrHn [(Trn+1 γψN )Bn ] = TrHn [γψN Bn ] → TrHn [γ (n) Bn ].

This implies
Trn+1 γ (n+1) = γ (n)
and thus we can use Theorem 3.1.7. Before that, using the fact that (by
assumptions) T1 + T2 + w is bounded below by, say, CT , we write
1 (2) 1 (2)
lim inf TrH2sym [(T1 + T2 + w)γψN ] = TrH2sym [(T1 + T2 + w − 2CT )γψN ] + CT
N →∞ 2 2
1
≥ TrH2sym [(T1 + T2 + w − 2CT )γ (2) ] + CT
2
1
= TrH2sym [(T1 + T2 + w)γ (2) ]
2

20
where we used Fatou’s lemma for positive operators. Using the quantum de
Finetti theorem we obtain
Z
E0 (N ) 1
eH ≥ lim inf ≥ TrH2sym [(T1 + T2 + w)|u⊗2 ihu⊗2 |]dµ(u)
N →∞ N 2
Zu∈SH
≥ EH (u)dµ(u) = eH .
u∈SH

The second part of the theorem follows from the fact that all inequalities
become equalities.

Thus we have the following Corollary:

Corollary 3.1.8 (Bose–Einstein condensation in the mean-field limit). As-

sume there exists a unique minimizer u0 for the Hartree functional. Then
for a sequence of ψN of (3.1) satisfying

(ψN , HN ψN ) = N eH + o(N )

we have
(1)
γψN → |u0 ihu0 |.
Thus the ground state exhibits Bose–Einstein condensation.

Proof. It follows from taking the partial trace in (3.5).

3.2 Gross–Pitaevskii scaling

One would like to know, whether statements similar to Theorems 3.1.3 and
3.1.6 hold true also in the case when the interaction becomes short range,
i.e. when
N
X
1 X
HN = −∆xj + V (xj ) + wN (xj − xk ). (3.6)
N −1
j=1 16j<k6N

with wN (x) = N 3β w(N x) and β ∈ (0, 1]. For strictly positive β, the function
wN provides an approximation to the identity (up to normalization), i.e.
Z 
3β
wN (x) = N w(N x) →N →∞ w δ0

in the sense that

Z 
2
|u0 | ∗ wN (x) → w |u0 |2 (x).

21
Consequently, in the scaling regime when β > 0 the Hartree equation (3.4)
has to be modified and the new effective equation reads
Z 
hu0 + w |u0 |2 u0 = ε0 u0 . (3.7)

This equation is sometimes called the nonlinear Schrödinger (NLS) equation.

It turns out that the NLS equation provides a good effective description only
when β < 1. When β = 1, then a subtle phenomenon takes place which
results in the appearance of the scattering length in the effective equation,
the so-called Gross–Pitaevskii equation

hu0 + 8πa0 |u0 |2 u0 = ε0 u0 .

The corresponding functional is the Gross–Pitaevski functional

Z
E GP (u) = (u, hu) + 4πa0 |u(x)|4 dx. (3.8)
R3

If one assumes the two-body interaction is positive, then Lieb and Seiringer
proved that statements analogous to Theorems 3.1.3 and 3.1.6 hold true
when β = 1 with the effective theory now being given by the Gross–
Pitaevskii functional. The presence of the scattering length is, from a physics
perspective, crucial as it is a parameter which experimentalists can manip-
ulate.
One can interpret these results as a universality result. The effective
theory does not depend on the details of the underlying microscopic system,
but rather on an quasi-macroscopic quantity that is accessible in the lab.
Finally, let us mention that the Gross–Pitaevski functional cannot be
derived using a product wave-function as an ansatz. This would to the NLS
functional. It follows that the trial wave-function must include correlations
between the particles. Those correlations appear on very short length scales
and are therefore difficult to capture. We will not discuss the related details
in this course and refer to the original papers.

22
Chapter 4

Bogoliubov theory

4.1 Landau criterion for superfluidity

It is observed experimentally that a moving superfluid doesn’t dissipate
its kinetic energy. Landaus theory of superfluids is based on the Galilean
transformation of energy and momentum.
Let E and P be the energy and momentum of the fluid in a reference
frame K. If we try to express the energy and momentum of the same fluid
but in a moving frame K 0 , which has a relative velocity V with respect to a
reference frame K, we have the following relations:

P 0 = P − M V,
1 1 (4.1)
E 0 = P 02 /(2M ) = |P − M V |2 = E − P V + M V 2
2M 2
where M is the total mass of the fluid.
We first consider a fluid at zero temperature, in which all particles are in
the ground state and flowing along a capillary at constant velocity v. If the
fluid is viscous, the motion will produce dissipation of energy via friction
with the capillary wall and decrease of the kinetic energy. We assume that
such dissipative processes take place through the creation of elementary
excitation. Let us first describe this process in the reference frame K which,
rather confusingly, moves with the same velocity v of the fluid. In this
reference frame, the fluid is at rest and its energy is the ground state energy
that we denote by E0 . If a single elementary excitation with a momentum
p appears in the fluid, the total energy of the fluid in the reference frame
K is E0 + (p), where (p) is the energy of the excitation with momentum
p. Let us move to the moving frame K 0 in which the fluid moves with a
velocity v but the capillary is at rest. In this moving frame K 0 which moves
with the velocity v with respect to the fluid, the energy and momentum of

23
the fluid are given by setting V = −v in (4.1). We obtain

P 0 = p + M v,
1
E 0 = E0 + (p) + pv + M v 2 .
2
The above results indicate that the changes in energy and momentum caused
by the appearance of one elementary excitation are (p) + pv and p, respec-
tively.
Spontaneous creation of elementary excitations, i.e. energy dissipation,
can occur if and only if such a process is energetically favorable. This re-
quires
(p) + pv < 0.
This is satisfied when |v| > (p)|p| and pv < 0, i.e. when the elementary
excitation has the momentum p opposite to the fluid velocity v and the fluid
velocity |v| exceeds the critical value
(p)
vcr = min . (4.2)
p |p|
Thus, superfluidity will occur only if the critical velocity is strictly positive.
In particular, the ideal Bose gas has (p) = p2 and thus vcr = 0 so it is
not superfluid. In particular, the particle-particle interaction is a crucial
requirement for the appearance of superfluidity in a bosonic system.

4.2 Bogoliubov approximation

Our starting point is the Bose gas enclosed in a cubic box of length L with
periodic boundary conditions. The second quantized Hamiltonian is given
by (we do not introduce any scaling of the interaction at this point)
X 1 X
H= p2 a∗p ap + ŵ(k)a∗p+k a∗q−k ap aq .
p
2L3
p,k,q

We are interested in the excitation spectrum of the system, that is the

structure of the energies above the ground state energy.
If there were no interaction then we know that the ground state would
be given by all particles in the zero momentum mode. The interaction w has
the property that it converts a pair of particles with momenta p and q into
a pair with momenta p + k and q − k. Starting with all N particles having
momentum zero, we would first get (N − 2) with momentum zero, together
with one pair having momenta k and k. When the potential is applied again
we could get two possibilities: one would be two pairs k, k and q, q; the
other would be a genuine triplet k, q, r, such that k + q + r = 0. But the
probability of the former relative to the latter would be (N − 2)(N − 3)/4

24
because there are (N − 1) particles with zero momentum and only 2 with
nonzero momentum. Applying w over and over again we will ultimately get
a finite fraction of triplets, quartets, etc. as well as pairs, but hopefully if
the interaction is weak enough we need consider explicitly only pairs in the
ground state wave function. This suggests that only terms with pairs should
be relevant in the ”effective” Hamiltonian.
These were the arguments that led N. N. Bogoliubov to drop all terms
in the Hamiltonian involving more than two creation/annihilation operators
of a non-zero mode. Doing this we obtain
X a∗0 a0

ŵ(0) ∗ ∗
H≈ a a a0 a0 + k + 3 (ŵ(k) + ŵ(0) a∗k ak
2
2L3 0 0 L
k6=0
X ŵ(k)
+ (a∗0 a∗0 ak a−k + a∗k a∗−k a0 a0 )
2L3
k6=0
ŵ(0)ρ
= (N − 1) + HBog + R
2
where
ρ = N/L3 ;
X 1X
HBog := (k 2 + ρŵ(k))a∗k ak + ρŵ(k)(a∗k a∗−k + ak a−k );
2
k6=0 k6=0
−ŵ(0) X ŵ(k)
(a∗0 a∗0 − N )ak a−k ) + a∗k a∗−k (a0 a0 − N ) .


R := 3
(N − N 0 )(N − N0 − 1) + 3
2L 2L
k6=0

We used
a∗0 a∗0 a0 a0 = N0 (N0 − 1) = N (N − 1) − 2N0 (N − N0 ) − (N − N0 )(N − N0 − 1).
We argue that R is small, because
a∗0 a∗0 ≈ a0 a0 ≈ N0 ≈ N.
The last step is called the c-number substitution. Thus the effective Hamil-
tonian is given by a quadratic operator (quadratic in creation and annihi-
lation operators) that does not conserve the number of particles. It can
however be diagonalized. To this end we use a Bogoliubov transformation
By introducing
bp = cp ap + sp a∗−p , b∗p = cp a∗p + sp a−p
with appropriate cp , sp (see exercises) such that c2p − s2p = 1. Then
X
HBog = EBog + (p)b∗p bp (4.3)
p6=0

with p
(p) = |p| p2 + 2ρŵ(p).

25
Problem 4.2.1. Derive (4.3)
Note that it follows that
(p) p 2
= p + 2ρŵ(p)
|p|
and thus the critical velocity vcr > 0. Thus Bogoliubov’s computations
shows that the interacting Bose gas satisfies Landau’s criterion for superflu-
idity.

4.3 Validity of the Bogoliubov approximation

4.3.1 Exciting the Hartree state
We have shown already, that a mean-field Bose gas exhibits Bose–Einstein
condensation at zero temperature with the condensate wave-function given
by the minimizer of the Hartree functional. Thus the Hartree state plays the
role of the macroscopically occupied state that in Bogoliubov’s computation
was represented by the zero momentum mode. The excited states should
therefore correspond to the situation when a particle is outside the Hartree
state. To implement this idea rigorously, we will introduce the so-called
excited Fock space.
Consider now any (real-valued) orthonormal basis u0 , u1 , . . . of L2 (Ω)
containing the Hartree minimizer u0 . Then, it is known that {ui1 ⊗s ui2 ⊗s
. . . uiN } is an orthogonal basis of the symmetric space HN , where
1 X
u1 ⊗s u2 . . . ⊗s uN = √ uσ(1) (x1 ) · · · uσ(N ) (xN ).
N ! σ∈S
N

Let us define
HN
0 := Span(u ⊗ . . . × u0 )
| 0 {z }
N times
and for k > 0
k
⊗(N −k)
O
HN
k := Span(u0 ⊗ . . . × u0 ) ⊗s H+ = u0 ⊗s Hk+
| {z }
N −k times sym

where H+ = {u0 }⊥ = Span{u1 , u2 , . . .} ∈ L2 (Ω). It follows that the many-

body Hilbert space HN can be written as a direct sum of the form

HN = HN N N
0 ⊕ H1 ⊕ · · · ⊕ HN .

Thus, any N -body wave function Ψ ∈ HN can be written as

⊗(N −1) ⊗(N −2)
Ψ = ψ0 u⊗N
0 + u0 ⊗s ψ1 + u0 ⊗s ψ2 + · · · + ψN

26
with ψk ∈ Hk+ . It is easy to check that
⊗(N −k) ⊗(N −l)
hu0 ⊗s ψk , u0 ⊗s ψl iHN = δkl hψk , ψl iHl

from which it follows that

N
X
2 2
kΨk = |ψ0 | + kψk k2Hk .
+
k=1

Therefore we see that the linear map

N
≤N
M
UN : HN → F+ := Hn+
n=0

given by
UN (Ψ) = ψ0 ⊕ ψ1 ⊕ ψ2 ⊕ . . . ⊕ ψN
≤N
is a unitary operator from HN onto the truncated, excited Fock space F+ .
The latter can always be seen as being embedded in the the full excited Fock
space given by
M∞
F+ := Hn+ .
n=0
The full Fock space of excited particles appears as the limit of the truncated
excited Fock space when N → ∞.
The operator UN has some important properties. In particular we have
that UN can be written as
N
!
N −j
M a
UN (Ψ) = Q⊗j p 0 Ψ
j=0
(N − j)!

for all Ψ ∈ HN . Here Q = 1 − |u0 ihu0 | is the projection onto the excited
space. Similarly  
N N
∗ 
M X (a∗ )N −j
UN ψj =
 p 0 φj
j=0 j=0
(N − j)!
≤N
for all φj ∈ Hj+ . These operators satisfy the following indentities on F+ :

UN a∗0 a0 UN
∗
= N − N+ ,
UN a∗ (f )a0 UN
∗
= a∗ (f ) N − N+ ,
p
(4.4)
UN a∗0 a(f )UN
∗
p
= N − N+ a(f ),
UN a∗ (f )a(g)UN
∗
= a∗ (f )a(g).
∗
P
Here N+ = m≥1 am am is the operator counting the number of excited
particles.

27
4.3.2 The Bogoliubov Hamiltonian
We define the Bogoliubov Hamiltonian in the following way:
X 1 1
H= hum , (h + K1 )un ia∗m an dxdy + hum ⊗ un , K2 ia∗m a∗n + hK2 , um ⊗ un iam an
2 2
m,n≥1
(4.5)
where K1 : H+ → H+ and K2 : H+ → H+ are operators defined by
ZZ
hu, K1 vi = u(x)v(y)u0 (x)u0 (y)w(x − y)dxdy,
Z ZΩ⊗Ω (4.6)
hu, K2 vi = u(x)v(y)u0 (x)u0 (y)w(x − y)dxdy
Ω⊗Ω

for all u, v ∈ H+ . Finally, h is the one-body operator given by

ZZ
2 1
h = −∆ + V + |u0 | ∗ w − eH − |u0 (x)|2 |u0 (y)|2 w(x − y)
2
which comes from the Hartree equation.
While we shall treat K1 as a one-body operator, we should really think
of K2 as its integral kernel K2 (x, y) = (Q ⊗ Q)(u0 ⊗ u0 w(. − .))(x, y) which
is the two-body function obtained by projecting the symmetric function
u0 (x)u0 (y)w(x − y) onto H2+ .

4.3.3 Rigorous statement about the Bogoliubov approxima-

tion
We shall now make several assumptions that will be used to prove the validity
of the Bogoliubov approximation.
A1 - interaction. Assumptions on the interaction potential are the same
as in Theorem 3.1.6.
A2 - Hartree theory. We assume that the Hartree minimizer is unique
and nondegenerate.
A3 - Bose–Einstein Condensation. We assume that there is BEC on u0
in the sense that for any constant R > 0, there exists a function R : N →
[0, ∞) with limN →∞ R = 0 such that, for any wave function ΨN ∈ HN
satisfying hΨN , HN ΨN i ≤ E(N ) + R we have
(1)
hu0 , γΨN u0 i
≥ 1 − R .
N
Note that during the course we have already showed that these assumptions
are satisfied in many situations.
Using assumptions (A1) and (A2) one can prove that

cdΓ(h + 1) − C ≤ H ≤ dΓ(h + C) + C (4.7)

28
for some constants c, C > 0. We will now state the main result about
Bogoliubov approximation, which heuristically can be written as
∗
UN (HN − N eH )UN → H.

Here it is.
Theorem 4.3.1 (Validity of Bogoliubov approximation). Under the as-
sumptions (A1)-(A3) the following holds true:
1 - weak convergence to H. For any Φ and Φ0 in the (quadratic form)
domain of H we have

lim hΦ0 , UN (HN − N eH )UN

∗
Φi = hΦ0 , HΦiF+ .
N →∞

2 - convergence of eigenvalues. Let λi (A) denote the i-th eigenvalue of

the operator A. Then

lim (λL (HN ) − N eH ) = λL (H).

N →∞

3 - convergence of eigenvectors. Let ΨL L

N and Φ denote the eigenvectors
corresponding to the L-th eigenvalue of HN and H respectively. Then

lim UN ΨL
N =Φ
L
N →∞

and the convergence is strong in F+ .

Let us now sketch the main ideas behind the proof of this theorem.
Roughly, it contains two main steps. First, one has to compute UN (HN −
∗ and rewrite it as
N eH )UN
∗
UN (HN − N eH )UN =H+R (4.8)

where R will be treated as a rest term.

In the second step one has to show that the rest term is in some sense
small. This will then lead to the desired conclusion by the min-max principle,
which roughly states states that if self-adjoint operators A and B satisfy
A ≤ B, then the n-th eigenvalue of A is smaller than the n-th eigenvalue of
B.
Let us now discuss how to get (4.8). First we write
X 1 X
HN = Tmn a∗m an + Wmnpq a∗m a∗n ap aq
2(N − 1)
m,n≥0 m,n,p,q≥0

where Tmn = hum , (−∆ + V )un i and Wmnpq = hum ⊗ un , w(x − y)up ⊗ uq i.
A tedious but straightforward computation using the relations (4.4)
shows that
X4
∗
UN (HN − N eH )UN = Aj
j=0

29
where
1 N+ (N+ + 1)
A0 = W0000 ,
2  N −1 
X N − N+ − 1 p
A1 = T0m + W000m N − N+ am + h.c.,
N −1
m≥1
X
A2 = Tmn a∗m an − (T00 + W0000 )N+
m,n≥1
X N − N+
+ hum , (|u0 |2 ∗ w + K1 )un ia∗m an
N −1
m,n≥1
 p 
1 X (N − N+ )(N − N+ − 1)
+ hum ⊗ un , K2 ia∗n a∗m + h.c. ,
2 N −1
m,n≥1
1 X
Wmnp0 a∗m a∗n ap
p
A3 = N − N+ + h.c.,
N −1
m,n,p≥1
1 X
A4 = Wmnpq a∗m a∗n ap aq .
2(N − 1)
m,n,p,q≥1

Now we have to identify the Bogoiubov Hamiltonian. This we do by

looking at A2 . We see that
1 − N+
A2 − H = dΓ(Q(|u0 |2 ∗ w)Q + K1 ) +
N −1 !!
p
X (N − N + )(N − N + − 1)
+< hum ⊗ un , K2 i a∗n a∗m −1 ) .
N −1
m,n≥1

≤M
We will show that when evaluated on a state Φ ∈ F+ , then A2 − H can be
made small. To this end we will use to following lemma whose proof follows
basically from (4.7).
Lemma 4.3.2. Under the assumptions on the interaction we have the fol-
lowing estimates
1
dΓ(QT Q) ≤ H + CN+ + C,
1 − α1 (4.9)
α2
dΓ(Q(|u0 |2 ∗ w)Q ≤ H + CN+ + C
1 − α1
where 1 > α1 > 0 and α2 > 0.
≤M
Let Φ ∈ F+ . Then
D N+ − 1 E
hA2 iΦ − hHiΦ = − dΓ(Q(|u0 |2 ∗ w)Q + K1 )
N −1 Φ
X
∗ ∗
(4.10)
+< hum ⊗ un , K2 ihan am XiΦ
m,n≥1

30
 √
(N −N )(N −N −1)
+ +
where X = N −1 − 1.
Let us rewrite w = w+ − w− where w± ≥ 0. By linearity, using the
triangle inequality, we have
D N+ − 1 E D N+ − 1 E
− dΓ(Q(|u0 |2 ∗ w)Q) ≤ dΓ(Q(|u0 |2 ∗ w+ )Q)
N −1 Φ N −1 Φ
D N+−1
E
+ dΓ(Q(|u0 |2 ∗ w− )Q) .
N −1 Φ
Since the convolution of two non-negative functions is a non-negative func-
tion and as the multiplication operator by a non-negative function is non-
negative operator, we have that
dΓ(Q(|u0 |2 ∗ w± )Q) := A± ≥ 0
is a positive operator that commutes with N+ . It follows that A± =
(A∗± A± )1/2 and we can write
N+ − 1 q ∗ N+ − 1 p
A± = A± A± .
N −1 N −1
√
Using the fact that A± (and therefore A± ) does not change the number
of particles of a given state we have
Dq N+ − 1 p E M
A∗± A± ≤ hA± iΦ
N −1 Φ N −1
≤M
for Φ ∈ F+ . Thus
D N+ − 1 E M D E
− dΓ(Q(|u0 |2 ∗ w)Q) ≤ dΓ(Q(|u0 |2 ∗ (w+ + w− ))Q)
N −1 Φ N −1 Φ
M α2
≤ h H + CN+ + CiΦ
N − 1 1 − α1
where in the last step we used (4.9).
Furthermore, since for any ψ with kψk = 1 we have by the Cauchy-
Schwarz inequality
|hψ, ±K1 ψi| ≤ kK1 ψk ≤ kK1 k
it follows that
kK1 k1 + K1 ≥ 0.
Thus, similarly as above we get
D N+ − 1 E D N+ − 1 E
− dΓ(QK1 Q) = dΓ(Q(kK1 k1 + K1 −kK1 k1)Q)
N −1 Φ | {z } N −1 Φ
≥0
M D E M D E
≤ dΓ(Q(kK1 k1 + K1 )Q) + dΓ(QkK1 k1Q)
N −1 Φ N −1 Φ
3M kK1 k
≤ hN+ iΦ .
N −1

31
 It follows that for the first term on the RHS of (4.3.3) we get the bound
D N+ − 1 E CM
− dΓ(Q(|u0 |2 ∗ w)Q + K1 ) ≤ hH + N+ 1iΦ
N −1 Φ N −1
where the constant C > 0 is independent of N . √
+ (N −N )(N −N −1)
+
It remains to look at the second term in (4.3.3). Recall X = N −1 −
1. By the Cauchy-Schwarz inequality we have
X X X
hum ⊗ un , K2 iha∗n a∗m XiΦ ≤ ( |hum ⊗ un , K2 i|2 )1/2 ( |ha∗n a∗m XiΦ |2 )1/2
m,n≥1 m,n≥1 m,n≥1
ZZ
1/2
X
≤( |K2 (x, y)|2 dxdy)1/2 ( ha∗n a∗m am an iΦ )1/2 hX 2 iΦ .
m,n≥1

Here we used that K2 is a Hilbert-Schmidt operator. Now, one can check

that
(N+ + 1)2
X2 ≤ C .
(N − 1)2
Furthermore, since a∗n a∗m am an = a∗n an a∗m am − δnm a∗n am , we get
X
ha∗n a∗m am an iΦ ≤ hN+2 iΦ .
m,n≥1

Altogether this gives

1/2
(N+ + 1)2

X 1/2 CM
hum ⊗un , K2 iha∗n a∗m XiΦ ≤ CkK2 k2 hN+2 iΦ ≤ hN+ +1iΦ .
(N − 1)2 N −1
m,n≥1

Thus we see that

CM
|hA2 iΦ − hHiΦ | ≤≤ hH + N+ 1iΦ
N −1
and thus is relatively small as compared with H if M  N . Similar estimates
∗
P4 be obtained for other terms that are left from UN (HN − N eH )UN =
can
j=0 Aj after removing A2 . Then one can indeed show that the contribu-
tions coming from sectors of the Fock space with large particle numbers are
indeed small. The details require some more subtle arguments and can be
found in the 2015 CPAM paper of Lewin, Nam, Serfaty and Solovej (see also
the paper of Grech and Seiringer CMP 2013).

32
Appendix A

Hilbert spaces and operators

A.1 Operators
Recall that a Hilbert space H is a vector space endowed with a sesquilinear
map (·, ·) : H × H → C (i.e., a map which is conjugate
p linear in the first
variable and linear in the second) such that kφk = (φ, φ) defines a norm
on H which makes H into a complete metric space.
One basic property of Hilbert spaces that we will use is the fact that for
any closed subspace V ⊂ H there corresponds the orthogonal complement
V ⊥ such that V ⊕ V ⊥ = H.
Another one goes under the name of the Riesz representation theorem:
to any continuous linear functional f : H → C there is a unique ψ ∈ H such
that f (φ) = (ψ, φ) for all φ ∈ H.
We shall always assume that our Hilbert spaces are separable and there-
fore that they have countable orthonormal bases.
Definition A.1.1. By an operator (or more precisely densely defined oper-
ator) A on a Hilbert space H we mean a linear map A : D(A) → H defined
on a dense subspace D(A) ⊂ H. Dense refers to the fact that the norm
closure D(A) = H.
Definition A.1.2. If A and B are two operators such that D(A) ⊆ D(B)
and Aψ = Bψ for all ψ ∈ D(A) then we write A ⊂ B and say that B is an
extension of A.
Note that the domain is part of the definition of the operator. In defining
operators one often starts with a domain which turns out to be too small
and which one then later extends.
Definition A.1.3. We say that A is a symmetric operator if

(φ, Aψ) = (Aφ, ψ) (A.1)

for all ψ, φ ∈ D(A).

33
 It is a fact that (A.1) holds if and only if (ψ, Aψ) ∈ R for all ψ ∈ D(A).
This property is of great importance in quantum mechanics as expectations
values of observables need to be real.
Definition A.1.4. An operator A is said to be bounded on the Hilbert space
H if D(A) = H and A is continuous, which by linearity is equivalent to
kAk = sup kAφk < ∞.
φ,kφk=1

The number kAk is called the norm of the operator A. An operator is said
to be unbounded if it is not bounded.
Definition A.1.5. If A is an operator we define the adjoint A∗ of A to be
the linear map A∗ : D(A∗ ) → H defined on the space
D(A∗ ) = {φ ∈ H| sup |(φ, Aψ)| < ∞}
ψ∈D(A),kψk=1

and with A∗ φ defined such that

(A∗ φ, ψ) = (φ, Aψ)
for all ψ ∈ D(A). The existence of A∗ φ for φ ∈ D(A∗ ) is ensured by the
Riesz representation theorem. If D(A∗ ) is dense in H then A∗ is an operator
on H.
Definition A.1.6. A bounded operator K : H → H is called compact if
and only if for every bounded sequence {φn } ⊂ H, {Kφn } has a subsequence
convergent in H.
An equivalent definition says, that compact operators are the closure of
the set of finite rank operators.
Theorem A.1.1. Let K be a compact operator on H. Then there exist
orthonormal bases {un } and {vn } of H and a sequence of real numbers {λn }
with λn →n→∞ 0 such that
X
Kφ = λn (un , φ)vn
n

for all φ ∈ H.
Definition A.1.7. An operator A defined on a subspace D(A) of H is said
to be positive (or positive definite) if (ψ, Aψ) > 0 for all non-zero ψ ∈ D(A).
It is said to be positive semi-definite if (ψ, Aψ) ≥ 0 for all ψ ∈ D(A). In
particular, such operators are symmetric.
Definition A.1.8. If A and B are two operators with D(A) = D(B) then
we say that A is (strictly) less than B and write A < B if the operator
B − A (which is defined on D(B − A) = D(A) = D(B) is a positive definite
operator. We write A ≤ B if B − A is positive semi-definite.

34
 We conclude that if K is a positive, compact operator then
X
K= λi |ui ihui | (A.2)
i
where now λi ’s can be chosen positive.
Among compact operators two special classes are of particular impor-
tance: trace class operators and Hilbert-Schmidt operators.
Definition A.1.9. Let H be a separable Hilbert space and {φn } its orthonor-
mal basis. Then for any positive operator A we define its trace to be
X
Tr A = (φn , Aφn ).
n

It is easy to see that if 0 ≤ A ≤ B then Tr A ≤ Tr B. Also, Tr(U AU −1 ) =

Tr A for any unitary operator U . Finally, if A is trace class and B is bounded,
then Tr(AB) = Tr(BA).
Definition A.1.10. Any (bounded)√ operator A on H is called trace class if
and only if Tr |A| < ∞. Here |A| = A∗ A.
A trace class operator is compact. It follows from the representation
P in
Theorem A.1.1 that a compact operator is trace class if and only if i |λi | <
∞. In particular, a positive, trace class operator has a representation of the
form (A.2).
Theorem A.1.2. Let k · k1 be defined for trace class operators by kAk1 =
Tr |A|. Then the space of trace class operators equipped with the norm k · k1
is a Banach space. Furthermore kAk ≤ kAk1 .
Let us now mention the other important class of compact operators.
Definition A.1.11. An operator A on H is called Hilbert–Schmidt if and
only if Tr A∗ A < ∞.
As in the case of trace class operators, a Hilbert–Schmidt operator is
compact. As before this leadsPthe condition that a compact operator is
Hilbert–Schmidt if and only if i λ2i < ∞.
p Among Hilbert–Schmidt
p operators one can introduce the norm kAk2 =
∗
Tr(A A) = (A, A)2 . With this scalar product the space of Hilbert–
Schmidt becomes a Hilbert space. We also have kAk ≤ kAk2 ≤ kAk1 .
Another important fact is the following
Theorem A.1.3. Let (M, µ) be a measure space and H = L2 (M, dµ). Then
A is Hilbert–Schmidt if and only if there is a function K ∈ L2 (M × M, dµ ⊗
dµ) with Z
(Af )(x) = K(x, y)f (y)dµ(y).

Moreover, Z
kAk22 = |K(x, y)|2 dµ(x)dµ(y).

35
A.2 Tensor product of Hilbert spaces
Let H and K be two Hilbert spaces. Consider Z = H × K. We form a vector
space which has Z as a basis. To this end we consider the vector space
of functions from Z → C and identify (v, w) with the function that takes
the value 1 on (v, w) and 0 otherwise. Let Z0 be the subspace spanned by
elements of the form
(u, v1 + v2 ) − (u, v1 ) − (u, v2 ),
(u1 + u2 , v) − (u1 , v) − (u2 , v),
(λu, v) − λ(u, v),
(u, λv) − λ(u, v)
where λ ∈ C.
Definition A.2.1. The algebraic tensor product is defined by
H ⊗al K = Z/Z0
and the image of (v, w) in this quotient is denoted v ⊗ w and is called the
tensor multiplication.
We introduce the inner product that satisfies
(u1 ⊗ v1 , u2 ⊗ v2 )H⊗al K = (u1 , u2 )H (v1 , v2 )K .
We set
H ⊗ K := (H ⊗al K)cpl .
Thus, Span{u ⊗ v|u ∈ H, v ∈ K} is dense in H ⊗ K. We call the vectors of
the form u ⊗ v pure tensor products.
If we have an operator A on the Hilbert space H and an operator B on
the Hilbert space K, then we may form the tensor product operator A ⊗ B
on H ⊗ K with domain
D(A ⊗ B) = Span{φ ⊗ ψ|φ ∈ D(A), ψ ∈ D(B)}
and acting on pure tensor products as
A ⊗ B(φ ⊗ ψ) = (Aφ) ⊗ (Bψ).
The tensor product may in a natural way be extended to more than two
Hilbert spaces. N
In particular, we may for N = 1, 2, . . . consider the N -fold
N
tensor product H of a Hilbert space H with itself. On this space we
have a natural action of N N group SN . I.e., if σ ∈ SN , then we
the symmetric
have a unitary map Uσ : N H → N H defined uniquely by the following
action on the pure tensor products
Uσ u1 ⊗ . . . ⊗ uN = uσ−1 (1) ⊗ . . . ⊗ uσ−1 (N ) .
NN NN
We shall denote by Ex : H → H the unitary corresponding to a
simple interchange of the two tensor factors.

36
Definition A.2.2. Let P+ be the orthogonal projection given by
X
P+ = (N !)−1 Uσ .
σ∈SN

We define the symmetric tensor of Hilbert spaces as

N
O ON
H = P+ ( H).
sym

Recall that in (2.1) we defined the Hamiltonian that in a more general

notation can be written as
N
X X
HN = hj + wjk .
j=1 16j<k6N

The one-body operators hj should be identified with

hj = 1 ⊗ . . . ⊗ 1 ⊗ hj ⊗1 ⊗ . . . ⊗ 1.
|{z}
jth slot

The operator
n-in
HN = h1 + . . . + hN
may then be defined on the domain
n-in
D(HN ) = Span{φ1 ⊗ . . . ⊗ φN |φ1 ∈ D(h1 ), . . . , φN ∈ D(hN )}

which is dense in N H if D(hi ) is dense in H for each i.

N
Similarly, the two-body operator wkj is acting on the Hilbert space
Hj ⊗ Hk and in the context of the many-body Hamiltonian wjk has to
be understood as being tensored with identities on the remaining variables.

37
Appendix B

Fock spaces and second

quantization

B.1 Fock space

Recall that we introduced the Fock space
∞
M
F= HN .
N =0

For Ψ, Φ ∈ F we define the inner product to be

X
(Ψ, Φ)F = (ψ (n) , φ(n) )Hn .
n≥0

For Ψ ∈ F we denote by kΨk the corresponding norm given by

X
kΨk2F = kψ (n) k22 .
n≥0

The Fock space allows to describe states of the system where the number
of particles is not fixed. A normalized vector Ψ = {ψ (n) }n≥0 ∈ F describes
a state that with probability kψ (n) k22 has n particles. In particular, states
with exactly N particles are embedded in the Fock space; they are described
by vectors of the form {0, . . . , 0, ψN , 0, . . .} ∈ F having only one non-zero
component. These vectors are eigenvectors of the particles number operator
N defined by
(N Ψ)(n) = nψ (n)
for any Ψ = {ψ (n) }n≥0 such that
X
n2 kψ (n) k22 < ∞.
n≥0

The vacuum vector Ω = {1, 0, 0, . . .} plays a special role; it is an eigenvector

of N with eigenvalue zero and describes a state with no particles.

38
B.2 Creation and annihilation operators
For any one-particle wave-function f ∈ L2 (R3 ) we define the creation oper-
ator a∗ (f ) and the annihilation operator a(f ) by setting
n
∗ (n) 1 X
(a (f )Ψ) (x1 , . . . , xn ) = √ f (xj )ψ (n−1) (x1 , . . . , xj−1 , xj+1 , . . . , xn ),
n
j=1
√
Z
(a(f )Ψ)(n) (x1 , . . . , xn ) = n + 1 f (xn+1 )ψ (n+1) (x1 , . . . , xn , xn+1 )dxn+1

In the case of the creation operator we have to assume n ≥ 1. To get

a physically consistent interpretation, in the definition we need to assume
that the vacuum (n = 0) component of the Fock space vector is zero after
the action of the creation operator. Also,

a(f )Ω = 0.

Creation and annihilation operators satisfy canonical commutation relations

(CCR)

[a(f ), a(g)] = [a∗ (f ), a∗ (g)] = 0, [a(f ), a∗ (g)] = hf, gi, ∀f, g ∈ L2 (R3 ).
(B.1)

Problem B.2.1. Check (B.1).

It is also convenient to introduce operator-valued distributions a∗x and

ax so that
Z Z
∗ ∗
a (f ) = f (x)ax dx, a(f ) = f (x)ax dx, ∀f ∈ L2 (R3 ). (B.2)
R3 R3

Expressed through these operator-valued distributions, the CCR take the

form
[a∗x , a∗y ] = [ax , ay ] = 0, [ax , a∗y ] = δ(x − y), ∀x, y ∈ R3 .
The interpretation is straightforward: a∗ (f ) creates a new particle with wave
function f , while a(f ) annihilates such a particle. Creation and annihilation
operators are closed densily defined operators on F; moreover, a∗ (f ) is the
formal adjoint of a(f ) (as the notation suggests). Notice also that a∗ (f ) is
linear while a(f ) is antilinear.

Problem B.2.2. Check that a∗ (f ) and a(f ) are formal adjoints, i.e.

(Ψ, a∗ (f )Φ)F = (a(f )Ψ, Φ)F

for Φ, Ψ ∈ F and f ∈ L2 (R3 ).

39
 Despite the bosonic creation and annihilation operators being unbounded,
usually domain questions are unproblematic since one can take the domain
of a sufficiently large power of the number operator to easily make sense
of most expressions - see (B.5). One has to be more careful in this respect
when working with the operator-valued distributions as formally ax is ob-
tained form a(f ) by taking f to be the Dirac δ-function which is not an
element of the one-body Hilbert space.
Notice that if Ψ = {ψ (n) }n≥0 , then
√
(ax Ψ)(n) (x1 , . . . , xn ) = n + 1ψ (n+1) (x, x1 , . . . , xn ). (B.3)

Hence
X Z (n)
(Ψ, N Φ) = n dx1 . . . dxn ψ (x1 , . . . , xn )φ(n) (x1 , . . . , xn )
n≥0
XZ (n−1)
= dxdx1 . . . dxn−1 (ax Ψ) (x1 , . . . , xn−1 )(ax Φ)(n−1) (x1 , . . . , xn−1 )
n≥1
Z
= dx(ax Ψ, ax Φ).

It follows that as a quadratic form we can write

Z
N = dx a∗x ax . (B.4)

This expression for N suggests that, although creation and annihilation

operators are unbounded operators, they can be bounded with respect to
the square root of the number of particles operator.
Indeed, in norm we have the bounds

ka(f )ΨkF ≤ kf kkN 1/2 ΨkF ,

(B.5)
ka∗ (f )ΨkF ≤ kf kk(N + 1)1/2 ΨkF

for any f ∈ L2 (R3 ). To prove the first bound, we observe that by the
Cauchy–Schwarz inequality
Z Z 1/2
2
ka(f )ΨkF ≤ dx|f (x)|kax ΨkF ≤ kf k dxkax ΨkF = kf kkN 1/2 Ψk.

The second estimate follows from the first one and the canonical commuta-
tion relations.

B.3 Second quntization of operators

Similarly to the expression (B.4), we can express the second quantization
of any one-particle operator in terms of the operator-valued distributions

40
a∗x and ax . Let J (1) be an operator on the one-particle space L2 (R3 ). The
second quantization of J (1) is the operator dΓ(J (1) ) on F defined by the
requirement that
n
(1)
X
(dΓ(J (1) )Ψ)(n) = Ji ψ (n)
i=1
(1)
where Ji denotes the operator acting on L2 (R3n ) as J (1) on the i-th particle
and as the identity on the other (n−1) particles. If the one-particle operator
J (1) has the integral kernel J (1) (x; y), we can write
n
(1)
XX
(Φ,dΓ(J (1)) Ψ) = (φ(n) , Jj ψ (n) )
n≥1 j=1
X Z
= n dxdydx2 . . . dxn φ(n) (x, x2 , . . . , xn )J (1) (x; y)ψ (n) (y, x2 , . . . , xn )
n≥1
XZ
= dxdydx2 . . . dxn J (1) (x; y)(ax Φ)(n−1) (x2 , . . . , xn )(ay Ψ)(n−1) (x2 , . . . , xn )
n≥1
Z
= dxdyJ (1) (x; y)(ax Φ, ay Ψ).

Thus, as a quadratic form, we have

Z
dΓ(J ) = dxdyJ (1) (x; y)a∗x ay .
(1)

Problem B.3.1. Let J (2) be an operator on the two-particle Hilbert space

L2sym (R6 ). We define the second quantization dΓ(J (2) ) of J (2) by
X (2)
(dΓ(J (2) )Ψ)(n) = J{i1 ,i2 } ψ (n)
{i1 ,i2 }

where the sum runs over all sets {i1 , i2 } of two different indices in {1, . . . , n}
(2)
and where J{i1 ,i2 } denotes the operator on L2sym (R3n ) acting as J (2) on the
variables i1 , i2 and as identity on the other (n − 2) variables. Show that if
J (2) has the integral kernel J (2) (x1 , x2 ; y1 , y2 ) then
Z
dΓ(J ) = dx1 dx2 dy1 dy2 J (2) (x1 , x2 ; y1 , y2 )a∗x1 a∗x2 ay1 ay2 .
(2)

Problem B.3.2. Consider the Hamiltonian

N
X X
HN = −∆i + w(xi − xj ).
i=1 i<j

Show that its second quantization equals

Z Z
∗ 1
H = dx ∇x ax ∇x ax + w(x − y)a∗x a∗y ax ay dxdy.
2

41
Problem B.3.3. Consider the many-body Hamiltonian of the form
N
X X
HN = Ti + Wij .
i=1 i<j

Let {un } be an orthonormal basis of the one-body Hilbert space. Show that
the second quantization of HN equals
X 1 X
H= (um , T un )a∗m an + (uk ⊗ ul , W um ⊗ un )a∗k a∗l am an
m,n
2
k,l,m,n

where a∗m = a∗ (um ).

Problem B.3.4. Consider the operator on the Fock space of the form
X
A= ei a∗i ai .
i

Show that the spectrum of A is of the form

X
ei ni , ni ∈ {0, 1, 2, . . .}.
I

42
Bibliography

[1] Kapitza, P. (1938). ”Viscosity of Liquid Helium Below the λ-Point”.

Nature 141 (3558)

[2] F. London (1938). ”The λ-Phenomenon of Liquid Helium and the Bose-
Einstein Degeneracy”. Nature. 141 (3571)

[3] E. H. Lieb and R. Seiringer, Proof of Bose-Einstein Condensation for

Dilute Trapped Gases, Phys. Rev. Lett., 88 (2002), p. 170409.

43