Draft version July 5, 2021

Typeset using LATEX twocolumn style in AASTeX631

Gravitational lensing formalism in a curved arc basis: A continuous description of observables and
degeneracies from the weak to the strong lensing regime
1, 2, ∗
Simon Birrer
1 Kavli Institute for Particle Astrophysics and Cosmology and Department of Physics,
Stanford University, Stanford, CA 94305, USA
2 SLAC National Accelerator Laboratory, Menlo Park, CA, 94025, USA
arXiv:2104.09522v2 [[Link]] 1 Jul 2021

ABSTRACT
Gravitationally lensed curved arcs provide a wealth of information about the underlying lensing
distortions. Extracting precise lensing information from extended sources is a key component in many
studies aiming to answer fundamental questions about the Universe. To maintain accuracy with
increased precision, it is of vital importance to characterize and understand the impact of degeneracies
inherent in lensing observables. In this work, we present a formalism to describe the gravitational
lensing distortion effects resulting in curved extended arcs based on the eigenvectors and eigenvalues
of the local lensing Jacobian and their directional differentials. We identify a non-local and non-linear
extended deflector basis that inherits these local properties. Our parameterization is tightly linked to
observable features in extended sources and allows one to accurately extract the lensing information of
extended images without imposing an explicit global deflector model. We quantify what degeneracies
can be broken based on specific assumptions on the local lensing nature and assumed intrinsic source
shape. Our formalism is applicable from the weak linear regime, the semi-linear regime all the way
up to the highly non-linear regime of highly magnified arcs of multiple images. The methodology
and implementation presented in this work provides a framework to assessing systematics, to guide
inference efforts in the right choices in complexity based on the data at hand, and to quantify the
lensing information extracted in a model-independent way. ‡

Keywords: Gravitational lensing (670) — stong gravitational lensing (1643) — weak gravitational
lensing (1797)

1. INTRODUCTION In the very close vicinity of massive over-densities,

Gravitational lensing displaces the observed position such as galaxies or galaxy clusters, the lensing effect
and distorts the shape of apparent objects on the sky can lead to highly distorted images and even the ap-
due to intervening inhomogeneous matter along the line pearance of multiple images of the same source. In
of sight. In the cosmological context, the lensing effect these regimes, the second order perturbations do not
can mostly be well approximated with a first order dis- accurately describe the observed distortions of extended
placements and second order perturbations on the shape lensed sources anymore.
of the lensed source (see e.g., Blandford & Narayan 1992; One way to bridge the gap between the linear and
Kaiser & Squires 1993; Kaiser et al. 1995; Mellier 1999; non-linear lensing distortion regimes is with third order
Bartelmann & Schneider 2001). The displacement ef- polynomial perturbations on the lensing potential (flex-
fect is not an observable as the intrinsic position of the ion) as an octopole signal in the measured shape (e.g.
objects can not be determined. The distortion of the Goldberg & Natarajan 2002; Goldberg & Bacon 2005;
shapes of extended objects do contain statistical signal Irwin & Shmakova 2005, 2006; Bacon et al. 2006). The
due to the correlation of the apparent shapes of differ- flexion measurement has been employed, for example,
ent objects along similar lines of sights, known as cosmic in the Hubble Ultra Deep Field (Irwin et al. 2007), and
shear. in combination with shear and strong lensing conjugate
points in cluster models, both in parametric and non-
parametric form (e.g., Leonard et al. 2007). A Taylor
∗ sibirrer@[Link] expansion determination of lensing quantities to fourth
2 S. Birrer

order has been investigated by Wagner & Bartelmann stant, H0 , from time-delay cosmography measurements
(2016), and a generalized weak lensing effect by Fleury (e.g. Schneider & Sluse 2013; Birrer et al. 2016; Sonnen-
et al. (2019). Overall, there has been only moderate suc- feld 2018; Kochanek 2020; Birrer et al. 2020). We refer
cess and applicability of flexion corrections in providing to Birrer et al. (2020) for the latest constraints on H0
model-independent local lensing constraints. by the TDCOSMO collaboration when only using MST-
In certain regimes, a fourth order approximation with invariant imaging observables by effectively allowing an
a carefully chosen coordinate system can match some additional MST degree of freedom in the mass profiles
further positional and local constraints in quadruply im- and anchoring the radial density profiles by stellar kine-
aged lenses (Wagner 2019) but still does not allow one matics measurements.
to describe extended arcs accurately. The reason for the On the other hand, a certain model may be insufficient
limits in polynomial extensions is the non-perturbative in describing the wealth of data available. This can par-
nature beyond the second order (shear and convergence) ticularly be the case in the galaxy cluster regime where
of the matter distribution in the Universe per se, lead- parameterized models are currently limited to match
ing to non-linear lensing effects deviating from a Tay- conjugate points within the astrometric measurement
lor expansion (e.g., Schneider & Er 2008). The most uncertainty and are incapable to describe the relative
prominent and abundant signature of non-linear lensing distortions observed in extended sources to the noise
effects beyond shear and convergence are curved arc dis- level of high resolution data (Yang et al. 2020; Dai et al.
tortions. Thought, in the infinitesimal small sources, the 2020). Another example is the interpretation of anoma-
alignment of quadrupole and octupole moments induce lous quadruply lensed quasar flux ratios. In some cases,
curvature locally (e.g., Irwin & Shmakova 2005), more observed flux ratios may not exclusively be due to dark
extended observed effects can not be described by third matter substructure but instead might have contribu-
order flexion terms (or even higher order ones) and re- tions from from larger scale baryonic components in the
quire a non-perturbative treatment of the lensing effect. lensing galaxy that a simplified lens model may have ne-
In the absence of a clean data-driven approach in glected (see e.g., Hsueh et al. 2016, 2018; Gilman et al.
the non-linear regime, the use of explicit deflector mass 2017). Similar effects can be observed when quantifying
models to provide the link between the observables and distortions in extended arcs, e.g. in Birrer et al. (2017a)
the lensing deflection field became the standard in many larger scale potential corrections had to be applied be-
analyzes involving strong gravitational lensing. One ex- fore substructure investigations could proceed.
ample of a lens model family widely employed is the The key to extract maximal precision while maintain-
singular power-law mass density profile. For the spher- ing accuracy in the non-linear regime of gravitational
ical case, there are theoretical studies quantifying and lensing is to allow for freedom in the lensing description
discussing how well observables are able to constrain the where data is able to constrain it and to have transpar-
global mass profile slope of the imposed power-law pro- ent priors in regimes where the data does not provide
file (e.g. Suyu 2012; O’Riordan et al. 2019). information to the problem at hand. The aim of this
The constraints derived from parameterized models paper is to provide a theoretical formalism that allows
may not in all circumstances reflect the observational one to quantify the invariant observables in gravitational
information on the deflection field. When employing lensing and a practical implementation to extract this
a specific parameterized model, constraints can be de- information from extended lensed images.
rived within the specific lens model family parameters We introduce a formalism to describe the distortion ef-
only. In the case of the constant power-law slope mass fects of curved extended arcs based on the eigenvectors
profile, the constraints on the logarithmic slope from ex- and eigenvalues of the local lensing Jacobian and their
tended imaging observables are are only possible due to directional differentials. The eigenvectors and their dif-
the demanding constraints of the global deflector model. ferentials are describing particular aspects of observa-
Neither the local slope nor the average of the sloe within tional lensing features. We identify specific bases for
a certain range are observables themselves per se. a non-local non-linear extension of the local properties
On one hand, specific functional forms may only probe to accurately predict and describe the detailed shape of
a sub-set of possible lensing configuration allowed by the extended sources at and around the location of interest
data, leading to over-constrained deflector inferences. without the need of a globally defined deflector model.
For instance, an imposed functional form on the deflec- Degeneracies inherent in lensing, such as the MST and
tor profile can artificially break the mass-sheet degen- it’s generalization, the Source Position Transform (SPT)
eracy (MST, Falco et al. 1985; Gorenstein et al. 1988) (Schneider & Sluse 2014; Unruh et al. 2017; Wertz et al.
and potentially bias the inference of the Hubble con- 2018) pose limits on the extractable lensing information.
 Gravitational lensing formalism in a curved arc basis 3

In the most general form, the SPT is not restricted to The lens equation, which describes the mapping from
curl-free deflector fields. The arc basis introduced in the source plane β to the image plane θ, is given by
this work is a suited approach to explore the curl-free
components of the SPT with some minimal, but well β = θ − α(θ), (1)
motivated, broad assumptions on the local lensing dis-
tortions. The method presented in this work effectively where α is the angular deflection as seen on the sky
allows one to extract lensing informations from extended between the original unlensed and the lensed observed
sources mitigating degeneracies inherent in lensing. position of an object.
Our formalism is applicable in all cosmological regimes
2.1.2. First and second order Cartesian differentials
of gravitational lensing, from the weak linear regime,
the semi-linear regime, up to the fully non-linear regime The differential of the lens equation between the
of highly magnified arcs and Einstein rings of multiple source position and its lensed appearance, the Jacobian,
images. is
∂βi ∂αi
The paper is organized as follows: In Section 2 we Aij ≡ = δij − . (2)
introduce the formalism of radial and tangential distor- ∂θj ∂θj
tions in the eigenvector basis of the lensing Jacobian The Jacobian describes the local linear distortions of a
and their differentials. We then discuss local lensing in- small extended source or likewise the magnification of
variances and degeneracies in the context of the curved an unresolved small source. The magnification µ is the
arc basis in Section 3. In Section 4 we discuss the ob- change in differential area from the source to the image
servables in curved arcs that allow us to constrain ra- position and can be expressed as the determinant of the
dial and tangential aspects of a global mass distributions inverse Jacobian
and demonstrate how our formalism is able to extract all
relevant information without being over-constraint. In µ = det(A)−1 . (3)
Section 5 we elaborate about applications of the method-
ology that can benefit from the approach we introduce The components of the Jacobian can be decomposed into
in this work, its limitations, and provide a specific ex- the convergence
ample. We conclude in 6.  
1 ∂αx ∂αy
All figures and inferences can be reproduced using κ= + , (4)
2 ∂θx ∂θy
code available at this repository ‡1 . All numerical com-
putations are performed with lenstronomy2 (Birrer & the shear components
Amara 2018; Birrer et al. 2021) version 1.8.2.  
1 ∂αx ∂αy
γ1 = − , (5)
2. LENSING FORMALISM FOR CURVED ARCS 2 ∂θx ∂θy
In this section, we first review the lensing formalism  
in general terms, in particular the polynomial Carte- 1 ∂αx ∂αy
γ2 = + , (6)
sian expansion in second and third order differentials of 2 ∂θy ∂θx
the lensing potential (2.1). We then introduce the for- and the curl component
malism of the differentials in the eigenvector basis (2.2).
We identify the local differentials in eigenvector space at-
 
∂αx ∂αy
tributed to curved arcs which provide a continuous map- curl = − . (7)
∂θy ∂θx
ping from the weak lensing to the strong lensing regime
(2.3). We use the eigenvector basis to define a mini- The next to leading order polynomial expansion of the
mal local lens model able to describe extended curved lens equation is known as flexion (Goldberg & Natarajan
arcs preserving the key differential quantities over an 2002; Goldberg & Bacon 2005; Bacon et al. 2006) and
extended area around the localized position (2.4). describes the gradients of the Jacobian

2.1. Lensing formalism basics ∂Aij

Dijk ≡ . (8)
∂θk
2.1.1. Lens equation
The lens equation (Eqn 1) to second polynomial order
in θ = θ 0 + ∆θ is given by
1 [Link] arcs
2 [Link] 1
βi ≈ θi − αi (θ 0 ) + Aij ∆θj + Dijk ∆θj ∆θk . (9)
2
4 S. Birrer

To this stage, no symmetry on the form of the lens equa- center of the structure. We can associate the eigenvalues
tion (Eqn 1) or the Jacobian (Eqn 2) have been invoked. as the inverse radial and tangential stretch of an image
In the case of a single lensing plane, the source term of exhibited by the massive structure
the gravitational deflection field is the convergence field,
κ(θ), with zero curl, and there exists a scalar lensing A · erad = λ−1
rad erad
potential, ψ, given by A · etan = λ−1
tan etan ,
Z
1 where we noted erad to be the radial component and
ψ(θ) = d2 θ 0 κ(θ 0 ) ln |θ − θ 0 |, (10)
π etan to be the tangential component of the Jacobian A
with their corresponding eigenvalues λ−1 −1
rad and λtan . In
such that
this form, λrad corresponds to the stretch factor of the
α(θ) = ∇ψ(θ). (11)
source in radial direction and λtan in tangential direc-
The Jacobian (Eqn 2) is symmetric, without any curl tion, corresponding to
component, and can be decomposed into a trace (con-
∂θr ∂θt
vergence κ) and trace-free (shear γ1 , γ2 ) term as λrad = , λtan = , (18)
∂βr ∂βt
" #
∂2ψ 1 − κ − γ1 −γ2 where ∂βr (∂βt ) correspond to the directional differen-
Aij = δij − ≡ . (12)
∂θi ∂θj −γ2 1 − κ + γ1 tials in the source plane corresponding to the reflected
radial (tangential) direction in the image plane. The
The magnification µ (Eqn 3) can be written as magnification is the product of the orthogonal stretches
1
µ= . (13) µ = λrad λtan . (19)
(1 − κ)2 − γ12 − γ22
We define the scalar angle φtan (φrad ) as the angle
The flexion terms can be compactly written as (see between the eigenvector etan (erad ) and a specific polar
e.g. Kaiser 1998) coordinate system of choice (e.g. centered at the massive
" # structure for convenience) such that
−2γ1,1 − γ2,2 −γ2,1
Dij1 = . (14)
−γ2,1 −γ2,2 cos (φtan ) = etan · e0 (20)
" # with e0 is the unit vector in the direction of the coordi-
−γ2,1 −γ2,2 nate center. A convenient coordinate center is the center
Dij2 = . (15)
−γ2,2 2γ1,2 − γ2,1 of a mass distribution.
In general terms, we can associate the tangential di-
2.2. Differentials in eigenvector space notation
rection to be along the major shear direction and the
2.2.1. Jacobian in eigenvector space radial component orthogonal to it. The directions of
We describe the Jacobian A (Eqn. 2) in terms of its the eigenvectors themselves are independent of the co-
two eigenvectors ei with corresponding eigenvalues λi ordinate center.
2.2.2. Third order differentials in eigenvector space
A · ei = λi ei . (16)
Analogously to the polynomial flexion as the differ-
In the case of a symmetric Jacobian, the eigenvectors entials of the Jacobian in Cartesian direction, we can
are orthogonal and the eigenvalues are real. The mag- introduce differentials along the eigenvectors of the tan-
nification µ (Eqn. 3) can be written as gential and radial eigenvalues as well as the differentials
Y of the direction of the eigenvectors themselves.
µ= 1/λi . (17) The differential of the eigenvector directions along its
i
own direction provides a measure of curvature. We de-
In the weak lensing regime, the two eigenvectors and the fine the tangential curvature, stan , as
direction provide a complete and equivalent description
∂φtan
to shear and convergence and we can state the properties stan ≡ (21)
∂etan
by referring to the major and minor eigenvector.
In the vicinity of a collapsed over-dense structure, such and the curvature in the radial direction, srad , as
as a galaxy or a galaxy cluster, the two eigenvectors
∂φrad
are to good approximation radial and tangential to the srad ≡ . (22)
∂erad
 Gravitational lensing formalism in a curved arc basis 5

The curvature terms stan and srad are coordinate sys-

tem independent. We note that the directional differ-
entials in the two directions etan and erad are the same
when the eigenvectors are orthogonal.
For the differentials of the eigenvalues we introduce tan
the following notations. The gradient of the tangential
magnification in tangential direction is
∂λtan
∂t λtan ≡ , (23)
∂etan
stan rad
and in radial direction
∂λtan
∂r λtan ≡ . (24)
∂erad
The gradient of the radial magnification in radial direc-
tion is
∂λrad
∂r λrad ≡ (25)
∂erad
and in tangential direction is
Figure 1. Illustration of eigenvectors and curvature along
∂λrad
∂t λrad ≡ . (26) the tangential direction for a single arc feature. The direc-
∂etan tional differential along the tangential direction stan marks
2.3. Tangential arcs and their eigenvector components a radius with radius r = s−1 tan on which the tangential and
radial eigenvalues, λtan and λrad are constant and pointing
In the following we focus on a single, considered most either in the direction or orthogonal to the center. ‡
prominent, third order eigenvector differential, the cur-
vature in the direction of the tangential direction stan
require a non-local expression covering the extents of
(Eqn. 26). We can describe in a minimal form a tan-
individual arcs or images observed. We demand the fol-
gential arc by considering radial and tangential stretch,
lowing conditions to be satisfied by the local deflector
λrad and λtan , radial direction, φrad , and the curvature
model around a pre-specified location θ0 , such as the
in the tangential direction, stan (Eqn. 26). Figure 1 pro-
center of an arc:
vides an example of an arc fully described by these four
components. The eigenvalues remain constant along a 1. The differentials at θ0 result in λrad , λtan , stan and
circle defined by the inverse curvature r = s−1 tan .
direction φtan , as specified.
Figure 2 illustrates tangential arcs as a function of 2. The curvature stan is constant along the tangential
tangential-to-radial stretch ratio, λtan /λrad , and curva- direction, effectively letting the path integral along
ture. Any other differential or higher order is set to zero the tangential eigenvector direction go around in
in this illustration. The linear regime is fully described a circle. We denote the radius of this circle as the
by three parameters in this notation, namely the radial curvature radius.
and tangential stretches, and the orientation of one of
the eigenvectors, while the curvature is zero (stan = 0). 3. Constant tangential stretch λtan on the curvature
The highly non-linear regime requires only the addition radius.
of one parameter, namely the curvature along the larger 4. Constant radial stretch on the curvature radius.
eigenvector (now called tangential stretch). Even a com-
pletely round Einstein ring can be fully described in this 5. No curl component on the deflection field.
notation by setting λtan = ∞ with stan > 0. The expres- 6. Zero deflection shift at the location θ0 , α(θ0 ) = 0.
sion of tangential arcs does allow one to locally describe
the lensing phenomenology from the weak lensing to the We identify the following deflector model that uniquely
strong lensing regime. satisfies the criteria mentioned above, of which the de-
flection angle is given by
2.4. A lens model basis for extended tangential arcs   
λtan − λrad θ − θc θ0 − θc
To this point in the manuscript, the discussion around α(θ) = s−1tan −
λrad λtan |θ − θc | |θ0 − θc |
tangential arcs have been made at the infinitesimal dif-
+ 1 − λ−1


ferential limit. Applications to describe extended arcs rad (θ − θ0 ) , (27)
6 S. Birrer

20
tan/ rad

10
Shear stretch

Curvature stan
Figure 2. Illustration of tangential arcs as a function of tangential to radial eigenvector stretch ratio λtan /λrad and tangential
curvature stan . The description of curved arcs in the eigenvector components allow us to describe distortions of lensed object
from the weak lensing regime continuously to the highly-magnified and distorted strong lensing regime. ‡

with θ c is the centroid position of the curvature radius of tangential-to-radial stretch ratio, λtan /λrad , and the
MST term matches the inverse of the radial stretch λrad .
θ c = θ 0 − s−1
tan erad . (28) We emphasize that this expression is only valid locally,
such as around an image of an arc, and is not meant
Equivalently, the deflector model above can be ex-
to cover an entire deflection field with multiple images.
pressed as a singular isothermal sphere model (SIS) in
We refer to Section 5 where we use a local tangential arc
combination with an MST as
parameterization basis separately on multiple images to
α(θ) = λMST [αSIS (θ) − αSIS (θ 0 )]+(1 − λMST ) (θ − θ0 ) , constrain more complex global deflector models.
(29)
with λMST = λ−1
rad and 3. OBSERVATIONAL INVARIANCES

θ − θc Having introduced the formalism of tangentially

αSIS (θ) = θE , (30) curved deflectors in describing curved arcs, it is essential
|θ − θc |
to understand and characterize lensing invariances and
with Einstein radius assumptions for extracting general lensing constraints.

λrad
 We thus dedicate this section to lensing degeneracies and
θE = s−1
tan 1 − . (31) their invariances in the characterization of curved arcs
λtan
within the locally tangential curved deflector model. In
The centroid matches the curvature radius, the Ein- Section 3.1 we formulate the general class of lensing in-
stein radius is adjusted such as to matches the ratio variances in an operator notation. We then discuss the
 Gravitational lensing formalism in a curved arc basis 7

specific class of the MST in Section 3.2 and how this tion operator Lj ◦ L−1
i between two images of the same
degeneracy translates to constraints of curved arcs. In source3 .
Section 3.3 we discuss shape noise degeneracies in the In the following, we sequence the general degeneracy
regime of curved arcs. operator J in a scalar component, λMST , a linear distor-
tion component, Γ, and a third component O encapsu-
3.1. Operator notation of general lensing invariances lating any higher order components not captured in the
To characterize general lensing degeneracies inherent previous two components, as
in gravitational lensing, we define the following notation:
L is the lensing operator distorting the source, effectively J ≡ λMST ΓO. (36)
mapping the lensed coordinates, θ, to the coordinates The scalar component in this transform is the special
prior to lensing, β. In general terms, L describes a co- case of the MST. The shear component Γ characterizes
ordinate mapping. The lens equation (Eqn. 1) can be the shape noise, while the non-linear component O char-
written in this notation as L(θ) = θ − α(θ). Given an acterizes any higher order distortion of the SPT.
intrinsic source morphology S, such that S(β) describes In Section 3.2 we further discuss the MST component
the intrinsic surface brightness at position β, the dis- within the framework of curved arcs, and in Section 3.3
torted image, D(θ), can be written as we discuss the shape noise aspects of the SPT, with a
brief discussion on higher order terms.
D(θ) = S(L(θ)). (32)
3.2. Mass-sheet transform (MST)
In terms of the operator notation, L is acting on S re-
sulting in D, stating as The mass-sheet transform (MST) is the scalar com-
ponent of the general SPT (Eqn. 36). This scalar com-
D = L ◦ S. (33) ponent is a multiplicative transform of the lens equa-
tion (Eqn. 1) preserving image positions (and thus any
With this notation, we can describe the general invari- higher order differentials too) under a linear source dis-
ance between lensing operator L and source morphology placement β → λβ and was introduced by Falco et al.
S resulting in the same image D with one single addi- (1985); Gorenstein et al. (1988) as such
tional mapping operator J, by expanding expression (33)
with the unity operator (now written as J−1 J) as
λMST β = θ − λMST α(θ) − (1 − λMST )θ. (37)
D = L ◦ S = L◦1◦S = L ◦ (J−1 J) ◦ S
The term (1 − λMST )θ in the equation above describes
= (L ◦ J−1 ) ◦ (J ◦ S) ≡ L̃ ◦ S̃. (34) an infinite sheet of convergence (or mass) and hence the
name mass-sheet transform. The corresponding trans-
In the last line, we defined the transformed deflection op-
form of the convergence profile is given by
erator, L̃ ≡ L ◦ J−1 , and transformed source, S̃ ≡ J ◦ S,
resulting in the same image D. The only formal require- κMST (θ) = λMST κ(θ) − (1 − λMST ). (38)
ment on J in Equation (34) above is that the mapping
is bijective and the inverse J−1 is uniquely defined over The MST can be described as a global transform of
the extent of image D. the convergence and hence it can lead to physical solu-
In summary, for any bijective angular mapping oper- tions for a wide range of values of λMST . A fact that
ator J, there exists an alternative solution to the lens makes the MST a prominent and relevant degeneracy
equation, D = L̃ ◦ S̃ with source S̃ = J ◦ S and lens for many applications, in particular the measurement of
L̃ = L ◦ J−1 . This statement is an operator formulation the Hubble constant with time-delay cosmography (e.g.,
of the Source Position Transform (SPT) (Schneider & Schneider & Sluse 2013; Birrer et al. 2016; Sonnenfeld
Sluse 2014; Unruh et al. 2017; Wertz et al. 2018). 2018; Kochanek 2020; Blum et al. 2020; Birrer et al.
In the presence of two or more images, Di , Dj , the 2020). Only observables related to the absolute source
relative operator translating one image into another can size, intrinsic magnification of the lensed source, the ab-
be determined without the knowledge of the intrinsic solute lensing potential, or the relative time delay when
source S: imposing a known cosmology with absolute distances,
are able to break this degeneracies.
Dj = Lj ◦ S = Lj ◦ L−1
i ◦ Di = L̃j ◦ L̃−1
i ◦ Di . (35)
3
In short, the measurable quantity in lensing under the We refer to Tessore (2017) for the explicit notation of this invari-
full consideration of the SPT is the relative distor- ance in the linear regime of a matix with shear and convergence.
8 S. Birrer

The differentials of the lens equation (e.g. Jacobian 2 flection field, however, contains a significant curl contri-
and flexion 8) transform under an MST as bution. For the off-axis distortions (Figure 4), even at
the center of the arc, significant curl contributions arise
A0 = λMST A , D0 = λMST D. (39) from the SPT.
The coefficients in the Jacobian and higher order differ- In the next approach, we restrict the lensing transform
entials are not constrained by imaging observables unless L̃ to a curl-free curved arc (Eqn. 27), while demanding
other constraints or assumptions on the lensing profile the source morphology to be sheared. Figure 5 shows
are inserted and thus do not serve themselves to be ob- the approximate SPT with a curl-free curved arc for on-
servables. axis distortions, re-fit to give the best possible fit to the
Equivalently, the MST scales the radial and tangential original arc generated with a round source. While the
eigenvectors as curved arc parameter fit follows the same infinitesimal
properties as for the SPT at the center of the arc, resid-
λ0tan = λ−1
MST λtan , λ0rad = λ−1
MST λrad . (40) uals in the extent of the arcs remain. Thus, within the
assumption of a curl-free tangentially curved deflector
The quantities that remain locally invariant under the model, the shape noise can be constrained. Off-axis dis-
MST is the ratio of tangential to radial eigenvalue tortions, as illustrated in Figure 6, are more constrained,
λtan /λrad that describes the relative distortions and any as the remaining residual patterns indicate. This feature
directional quantities (eigenvector direction)4 . Consid- can also be linked to the missing curl component in the
ering the third order derivatives, the curvature stan and center of the arc, as expected by the SPT. The closest
srad remain invariant under the MST. The derivatives approximations to the exact SPT within the curved arc
of the eigenvalues follow the same scaling with the MST lens model family does not allow us to adequately de-
as the eigenvalues themselves. scribe the observed arcs to the signal-to-noise level of
the simulation for substantial distortions of the source,
3.3. Shape degeneracies in curved arcs
thus restricting shape noise.
Beyond the MST, the remaining aspects of a linear dis- Higher order SPT components (O in Eqn. 36) in gen-
tortion are the reduced shear components (Γ in Eqn. 36). eral lead to source transforms that deviate from elliptical
These components are changing the ellipticity of the in- shapes. A subset of these transforms can lead to curl-
trinsic source. In the regime where the lensing operator free mappings L̃. One mathematically possible case is
L is linear, any linear SPT, Γ, leads to a linear transform when there is no lensing (L̃ = 1), then the shape of the
of L̃ = L ◦ Γ−1 and is thus indistinguishable from the source, S̃ is a curved arc itself. However, the physical
reduced shear. This degeneracy is generally known as plausibility of galaxies resembling in an intrinsic arc-
shape noise (see e.g. Bernstein & Jarvis 2002). We refer like shape needs to be considered and the likelihood of
to Appendix B for a shear and intrinsic shape notation higher-order morphological shapes can be estimated em-
convenient in transforming according to a linear SPT. pirically from the shapes of the entire galaxy population
However, if the lensing operator L is non-linear, such in low lensing environments. We further refer to Schnei-
as in the regime of curved arcs, the shape noise trans- der & Sluse (2014) for a discussion on higher order SPTs
formed lensing operator couples the differentials non- in the axi-symmetric case for global mass distributions,
linearly and can give rise to a curl component in the and to Unruh et al. (2017) for non-axi-symmetric cases.
deflection operator L̃. In this section, we did not discuss the impact of a point
We illustrate the non-linear coupling by performing spread function (PSF), In the example in Section (5.1)
an SPT on a curved arc with a round source and an we incorporate a PSF corresponding to a HST observa-
extended curved tangential deflector model given by ex- tion. We point out that uncertainties in the ellipticity of
pression 27. In Figure 3 the shear transform is per- the PSF can also lead to degeneracies related to shape
formed along the tangential axis and in Figure 4 the noise and thus accurate and precise PSF estimates are
transform is performed along the orthogonal shear mod- essential for studies of gravitational lensing, in particular
ulus. By construction, the SPT results in a perfect when extracting significant information from individual
match of the original arc for all cases. For the on-axis objects.
SPT (Figure 3), the local eigenvectors and tangential We also note that when multiple arcs of the same
curvature are transformed by the expected relative tan- source are present and the local lensing distortions are
gential and radial size of the source. The extended de- simultaneously reconstructed, this will add further con-
straining power on the SPT components depending on
4 This is equivalent to the reduced shear expression. the relative alignment of the different curved arcs. So
 Gravitational lensing formalism in a curved arc basis 9

even if there is a curl-free SPT allowed to reproduce one ent lens models. The round model exhibits, imposed by
arc by transforming the source morphology in a partic- its symmetry, identical curved arc structure along the
ular way, the lensing operator of an additional image azimuth with the curvature radius and direction point-
may require significant curl components to match the ing towards the center of the deflector profile. The el-
observations. Given that off-axis shape distortions are liptical mass model, here described as a power-law el-
better constrained than on-axis components, multiple liptical mass distribution (PEMD, see Appendix A for
images of arcs that are asymmetrically aligned, mean- details), causes a change in the tangential stretch λtan
ing on- and off-axis directions in the individual arcs cor- along the azimuth with a 180◦ symmetry imposed by the
respond to different axes in the intrinsic source plane, lens model symmetry. The curvature radius and direc-
do suppress the shear-ellipticity degeneracy more effi- tion, however, remain centered on the deflector mass. In
ciently. Fully connected Einstein rings further enhance the third case, we illustrate the azimuthal behavior of a
the suppression of the shear-ellipticity degeneracy. round mass density with an addition of an external shear
component. While the change in the tangential stretch
4. CONSTRAINING GLOBAL MASS
varies almost identically as for the case of an elliptical
DISTRIBUTIONS
mass distribution, the additional unambiguous feature
Theoretical discussions in the literature in regards to of the shear component is the fact that the direction
mass profile constraints are primarily using positional of the curvature in the arc is offset from the deflector
constraints and magnification ratios and are often tied center with an altered curvature radius.
and applicable to a specific mass profile family. In this The example illustrated in Figure 7 demonstrate how
section, we discuss and illustrate which observational extended resolved arcs are able to break the ellipticity-
features extractable by curved arcs allow us to constrain shear degeneracy. The formalism of curved arcs is able
what specific aspects of global mass distributions in the to capture these constraints. We do not discuss az-
non-linear regime of gravitational lensing. We first dis- imuthal structure beyond a dipole and external shear
cuss the tangential constraints related to ellipticity and but expect that the curved arc formalism and approach
external shear of a mass distribution (Section 4.1) and is also able to effectively describe and present observa-
then in a second step we separately discuss the radial tional signatures in more complex regimes of azimuthal
constraints provided by observed curved arcs (Section structure5 .
4.2). This section is accompanied by Appendix A where
we state the specific functional form of the global lens
models we use in this work as an example. 4.2. Radial constraints
4.1. Azimuthal constraints The primary radial constraint from gravitational lens-
ing of a mass profile is the Einstein radius θE . In the
Tangential distortions in strong gravitational lensing
round case, the Einstein radius marks the radius where
imprint signal about the asymmetric mass distribution
the tangential stretch λtan diverges and changes its sign,
in the main deflector and along the line of sight. To
known as the critical curve. The next-order leading
first order, the asymmetry can be described by an el-
term characterizing the radial profile is the radial stretch
liptical mass distribution and external shear. Positional
eigenvalue λrad . This value, however, is not an observ-
constraints of quadruply imaged sources can only par-
able due to the MST and only ratios of eigenvalues are
tially break the shear-ellipticity degeneracy (see e.g.,
observable. The leading order measurable quantity by
Schechter & Wynne 2019; Luhtaru et al. 2021) under
gravitational lensing observables is the normalized dif-
fixed radial profile constraints. O’Riordan et al. (2020)
ferential radial stretch of ∂r λrad /λrad measured as the
studies positional and magnification constraints on the
average finite differential between two arcs at differ-
joint ellipticity-power law radial slope, not considering
ent radial distance from the critical curve. The quan-
degeneracies with external shear.
tity ∂r λrad /λrad can be equivalently expressed as radial
In the formalism of extended curved arcs, the fol-
derivatives of the deflection angle α or the lensing po-
lowing shape quantities provide information about the
tential ψ (Eqn. 10)
azimuthal structure of the lens: (i) the change in the
tangential stretch along the azimuth of the deflector, 00 000
∂t λtan , (ii) change in the curvature direction along the ∂r λrad α ψ
= = , (41)
azimuthal direction, (iii) change in the curvature radius λrad 1 − α0 1 − ψ 00
along the azimuthal direction.
5
Figure 7 illustrates the curved arc properties at a fixed see e.g. a study with multi-pole moments and their impact on
radial distance along the azimuthal axis for three differ- Hubble constant measurements by Van de Vyvere et al. in prep
10 S. Birrer

Source
Lens
Image
Residual
Curl

Figure 3. Demonstration of the shape noise component of the SPT applied on a curved arc on-axis relative to tangential
eigenvector. The middle column corresponds to a reference example of a round intrinsic source (top) being distorted by a
curl-free curved arc (Eqn. 27, second row) resulting in the lensed curved arc (third row). The re-fitting with an SPT mapping
leads, by construction, to a perfect fit (fourth row indicates reduced residuals of the fit) without curl (fifth row). The other
columns correspond to an enforced different elliptical shape of the intrinsic source (S̃) with a lensing operator (L̃) to perfectly
describe the SPT. The resulting fit to the data is perfect but the required curl is non-zero. The off-axis distortions with the
SPT is presented in Figure 4. ‡
0
where denotes the radial derivative6 . where κE is the convergence at the Einstein radius.
The invariant quantity at the Einstein radius when The only difference between the expression in this work
the radial differential is scaled relative to the Einstein and by Kochanek (2020) is the representation of the
radius, is given by MST, either by the absolute radial stretch eigenvector
or the convergence at the Einstein radius, respectively.
∂r λrad (θE )
ξrad ≡ θE . (42) Both expressions allow for a model-independent inter-
λrad (θE ) pretation and translation of lensing constraints from one
We note that the quantity ξrad is effectively equivalent mass-profile family to another. This relation has been
in the constraining power to the expression introduced used, for example, by Shajib et al. (2021) to derive con-
by Kochanek (2020) straints on a family of more flexible mass density pro-
files based on original constraints derived with power-
∂r λrad (θE ) α00 (θE ) law density profiles.
θE ∝ θE , (43)
λrad (θE ) (1 − κE ) In the following, we discuss what aspects of curved
arcs allow us to constrain ∂r λrad /λrad . We identified
6 We also refer to Sonnenfeld (2018) to the use and derivation of three distinct aspects; (i) relative arc thickness measure-
the right hand side of Equation 41.
 Gravitational lensing formalism in a curved arc basis 11

Source
Lens
Image
Residual
Curl

Figure 4. Demonstration of the shape noise component of the SPT applied on a curved arc off-axis relative to tangential
eigenvector. The middle column corresponds to a reference example of a round intrinsic source (top) being distorted by a
curl-free curved arc (Eqn. 27, second row) resulting in the lensed curved arc (third row). The re-fitting with an SPT mapping
leads, by construction, to a perfect fit (fourth row indicates reduced residuals of the fit) without curl (fifth row). The other
columns correspond to an enforced different elliptical shape of the intrinsic source (S̃) with a lensing operator (L̃) to perfectly
describe the SPT. The resulting fit to the data is perfect but the required curl is non-zero. The on-axis distortions with the SPT
is presented in Figure 3. Off-axis shape distortions are better constraint by curved arcs than on-axis distortions as illustrated
in the difference in the residuals between this figure and Figure 3. ‡

ments, (ii) relation of ∂r λrad /λrad to tangential stretch ference of radial differentials but instead subdominant to
due to underlying symmetries, and (iii) positional con- the aspects mentioned in the following paragraphs.
straints of arcs.
4.2.2. Differential tangential extent of arcs
4.2.1. Differential radial thickness of arcs Differentials in the tangential extent of arcs do also
The most direct constraints on the radial differentials allow us to constrain the radial differentials when im-
can be made by measuring the relative thickness of mul- posing symmetries between the differential quantities.
tiply imaged arcs appearing at different radial distances Specifically, an azimuthally symmetric deflection field
from the critical curve. This measurement is demanding, obeys the following relation between tangential stretch
as arcs are usually not stretched along the radial direc- and relative source and image position in radial direc-
tion (λrad ≈ 1) and thus thin. Relative thickness dif- tion:
ferences of a few percent are often below the resolution θr
λtan = . (44)
of the instrument. We emphasize that radial differential βr
thickness, though the most intuitive constraining aspect, This relation is simply reflecting the fact that when ro-
is often not the dominating constraining factor in the in- tating the source position around the center of the de-
12 S. Birrer

Source
Lens
Image
Residual
Curl

Figure 5. Demonstration of the shape noise component of the curl-free curved arc approximation of the SPT applied on a
curved arc on-axis relative to tangential eigenvector. The middle column corresponds to a reference example of a round intrinsic
source (top) being distorted by a curl-free curved arc (Eqn. 27, second row) resulting in the lensed curved arc (third row).
The re-fitting with a tangentially curved deflector model best approximating the SPT leads, by construction, to a perfect fit
(fourth row indicates reduced residuals of the fit). The tangentially curved deflector models have, by design, no curl components
(fifth row). The other columns correspond to an enforced different elliptical shape of the intrinsic source (S̃) with a lensing
operator (L̃) of a curved arc (Eqn. 27) approximating the SPT. The resulting fit to the data is not perfect and the enforced
curl-free nature of the model leads to distinguishable intrinsic source shape features. The on-axis distortions with the curved
arc approximated SPT is presented in Figure 6. ‡

flector, the image positions are demanded to rotate with Imposing this relation allows one to derive constraints
the same angle. This symmetry argument leads to an on the radial density profile while utilizing measure-
imposed relation between the differential of the tangen- ments of tangential stretch differences. Relative tan-
tial stretch in radial direction, dλtan /dr, and the radial gential stretch differences are often easier to measure as
eigenvector λrad . In particular, differentiating relation the extent of the arc is larger in the tangential direction,
(44) along the radial direction results in well beyond the seeing limit.
  In Figure 8 we illustrate the differences of tangential
1 θr dβr λtan λtan
∂r λtan = − 2 = 1− , (45) arcs relative to the scale at the Einstein radius for three
βr βr dθr θr λrad different values of the power-law slope of a constant
where in the last equality above we substituted βr = power-law mass profile. The differentiability between
θr /λtan (Eqn. 44) and dθr /dβr = λrad . A version of the different power-law slopes is provided in relative radial
MST invariant relation of expression (45) reads stretch and relative tangential stretch.
  We emphasize that the relative tangential stretch re-
∂r λtan 1 λtan lation along specific directions can also be caused by
= 1− . (46)
λtan θr λrad
 Gravitational lensing formalism in a curved arc basis 13

Source
Lens
Image
Residual
Curl

Figure 6. Demonstration of the shape noise component of the curl-free curved arc approximation of the SPT applied on a
curved arc off-axis relative to tangential eigenvector. The middle column corresponds to a reference example of a round intrinsic
source (top) being distorted by a curl-free curved arc (Eqn. 27, second row) resulting in the lensed curved arc (third row). The
re-fitting with a tangentially curved deflector model best approximating the SPT leads, by construction, to a perfect fit (fourth
row indicates reduced residuals of the fit) without curl (fifth row). The curve arc models have, by design, no curl components
(fifth row). The other columns correspond to an enforced different elliptical shape of the intrinsic source (S̃) with a lensing
operator (L̃) of a curved arc (Eqn. 27) approximating the SPT. The resulting fit to the data is not perfect and the enforced
curl-free nature of the model leads to distinguishable intrinsic source shape features. The off-axis distortions with the curved
arc approximated SPT is presented in Figure 5. ‡

azimuthal structure. Specific assumptions, such as the the deflector density and we ignore this image in this
absence or presence of an azimuthal twist as a function discussion, as it is often unobserved.
of radius may impact radial constraints on the profile, Image pairs satisfy the lens equation (Eqn. 1). The
if they are primarily derived from the tangential scale lens equation (Eqn. 1) for the two solutions θin and θout
ratio, a statement also made by Kochanek (2021). arising from the same source position β demands that

4.2.3. Positional constraints

θin − α(θin ) = β = θout − α(θout ). (47)
Positional constraints of image pairs of the same
source also contain information about the radial differ-
ential stretch ∂r λrad /λrad . In this section we discuss the To investigate radial dependences on the relative so-
round deflector case where two magnified images appear, lution of the lens equation, we expand the solution rel-
one inside the Einstein radius, θin , and one outside the ative to the Einstein radius, where the solution is given
Einstein radius, θout . A third de-magnified solution of by θE − α(θE ) = 0. We can write the solution of the lens
the lens equation is at or very close to the center of equation in an integral form of the source displacement
14 S. Birrer

Round Elliptical Shear

Figure 7. Illustration of curved arc properties at a fixed radial distance along the azimuthal axis for three different lens models.
Left: Round lens model, resulting in a fully symmetric appearance of arcs. Middle: elliptical mass distribution, causing a change
in the tangential stretch ∂t λtan along the azimuth with a 180◦ symmetry imposed by the lens model symmetry. The curvature
radius and direction, however, remain centered as it is the case for a round mass distribution. Right: Round mass density with
an addition of an external shear component. While the change in the tangential stretch varies almost identically as for the case
of an elliptical mass distribution, the additional unambiguous feature of the shear component is the fact that the direction of
the curvature in the arc is offset from the mass distribution center with an altered curvature radius. ‡

= 1.8
=2
= 2.2

1.2
=2

X = rad
1.0 X = tan
X/X

X = tan rad
0.8
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
/ E
Figure 8. Illustration of the differences of tangential arcs relative to the scale at the Einstein radius for three different values
of the power-law slope of a power-law mass profile, as specified by the colors in the legend. Top: Curved arcs at different radii
for a fixed intrinsic source size normalized to match the width at the Einstein radius. Bottom: Difference in the tangential
(dotted-dashed), radial (dashed) and magnification (solid) of the arcs relative to the isothermal density profile (black). The
differentiability between different constant power-law slopes is provided in both, relative radial stretch, and relative tangential
stretch. Positional constraints on the appearance of multiple images are not part of this figure and are covered in Figure 9. ‡

from the origin as write the radial solution of the lens equation as
θin θout
dβ(θ0 ) 0 dβ(θ0 ) 0
Z Z
dθ = β = dθ . (48) Z ∆θin Z ∆θout
dθ0 dθ0 1 1
θE θE dθ0 = dθ0 .
0 λrad (θE − θ0 )0 λrad (θE + θ0 )
Defining the relative radial distance from the Einstein (49)
radius for the two images as ∆θin ≡ θE −θin and ∆θout ≡ Writing λrad (θ)−1 as a Taylor expansion around θE and
θout − θE and noting that dβr (θ)/dθr = λ−1rad (θ), we can only considering first and second order terms in ∆θ,
 Gravitational lensing formalism in a curved arc basis 15

Equation 49 can be approximated by 1.2

= 1.8
1
∆θin +
1 ∂r λrad 2
∆θin ≈
1
∆θout −
1 ∂r λrad 2
∆θout . 1.1 =2

in
λrad 2
2 λrad λrad 2 λ2rad = 2.2
(50)
1.0

out/
We further simplify the expression above explicitly stat-
ing the asymmetry in the image appearance ∆θout /∆θin
as a function of the mean displacement of the image 0.9 exact
pair relative to the critical curve, (∆θout + ∆θin )/2. Ap- approx
proximating ∆θout 2
/∆θin ≈ ∆θout , Expression 50 can be 0.8
expressed by 1.05

approx/exact
∆θout ∂r λrad ∆θin + ∆θout
∆θin
≈1+
λrad 2
. (51) 1.00
This relation shows that the radial asymmetry in the 0.95
appearance of images relative to the Einstein radius 0.0 0.2 0.4 0.6 0.8
1(
(or in more general terms the critical curve) is directly
2 out + in)/ E
linked to the reduced derivative of the radial stretch,
∂r λrad /λrad at the Einstein radius, and is linear as a Figure 9. Ratio of relative radial distances of an image pair
function of mean radial separation. Relation 51 is effec- relative to the Einstein radius for different radial power-law
tively equivalent to the relation presented by Sonnenfeld density slopes. ∆θout is the distance from the outer image to
(2018) expressed in terms of differentials of the lensing the Einstein radius and ∆θin is the distance from the inner
potential. image to the Einstein radius. Top panel: Distance ratio of
To investigate the validity of the approximation in ex- ∆θout /∆θin as a function of average distance normalized to
pression (51), we compare in Figure 9 the relative radial the Einstein radius, 1/2(∆θout + ∆θin )/θE . Solid lines in-
dicate the exact solution of the lens equation, while dashed
image position for different slopes of a power-law ra-
lines show the approximated linear solution ignoring terms
dial density profile. For the isothermal density profile beyond ∂r λrad /λrad given by Equation 51. Lower panel: Ra-
(γ = 2), λrad is constant and the exact solution as well tio of exact to approximate pair asymmetry. Imprints of the
as the approximation predicts an exact symmetry in the distortions of extended sources are illustrated in Figure 8. ‡
image pair appearance. For shallower and steeper slopes
∂r λrad as well as higher order terms are non-zero and an Here we provide an example of utilizing the curved
asymmetry in the appearance is observed. The approx- arc formalism to derive macro-model independent con-
imate solution proves to be accurate to one percent in straints on the deflector model for a quadruply imaged
the inferred power-law slope out to about 0.4 × θE in extended source. We assess this alternative to fitting a
mean separation of the images. global deflector model and discuss what constraints are
data-driven and what constraints are model-driven.
5. EXAMPLE AND DISCUSSIONS
In the previous sections, we have introduced the for- 5.1.1. Model set up and fitting procedure
malism to describe local curved arcs and have elaborated Our input deflector model is a PEMD profile with a
lensing degeneracies and constraints from a theoretical circular Gaussian source. We are using a Hubble Space
point of view. The goal of this section is to outline Telescope typical point spread function (PSF) width (as
potential practical applications and outline extensions. a Gaussian kernel), pixel scale and noise level (Figure 10
We provide an example of deriving macro-model inde- top left). We explicitly chose an example of a macro-
pendent lensing constraints from a multiply imaged ex- model that can not be represented globally by the de-
tended source in Section 5.1. In Section 5.2 we provide grees of freedom we allow for with individual tangen-
suggestions in the usage of the presented formalism for tially curved deflector models.
different science cases, and in Section 5.3 we discuss limi- For the model fitting, we define four regions of the
tations and possible extensions of the current formalism. image that capture the individual distorted images and
chose four independent extended tangentially curved de-
5.1. Example: Model-independent extraction of lensing flector models in the reconstruction process. The local
information of a quadruply imaged extended deflector models have the parameterization of the tan-
source gential and radial stretch eigenvalues, λtan and λrad , the
16 S. Birrer

direction φrad , tangential curvature stan , and a tangen- the model-independent inference of the tangential and
tial eigenvalue differential ∂t λtan . The underlying de- radial stretch eigenvectors (λtan , λrad ) at the positions
flector model is stated in Appendix A.4. In addition of the images (subscript 0-3) of the example displayed in
to the distortions, each deflector model patch has two Figure 10. In addition to the eigenvalues, we also show
additional uniform deflection displacement parameters the eccentricity moduli of the source shape (e1 and e2 ,
that effectively map the center of the curved arc to the see Appendix B.2). Not displayed are the direction φrad ,
center of the intrinsic source and contain the positional curvature stan and tangential differential ∂t λtan parame-
information. Per curved arc, there are seven free pa- ters for the individual local tangentially curved deflector
rameters. For the source morphology, we allow for a models. The sampling is performed under flat priors in
free ellipticity, as parameterized with the eccentricity the parameters stated. The posteriors are consistent
moduli (see Appendix B.2). This description ensures a with the input truth (black line, evaluated from the in-
full exploration of the shape-noise degeneracy discussed put macro-model at the positions of the curved arcs).
in Section 3.3. We fix the intrinsic source size for the We notice a significant degeneracy between the intrin-
purpose of an efficient sampling and the fact that the sic shape parameter e1 and the eigenvalues of all the lo-
MST adds an additional full degeneracy in the overall cal arcs. The direction of e1 corresponds to horizontal
scales of the inferred eigenvalues (see Section 3.2). We and vertical distortions and are almost on-axis with the
use lenstronomy in the joint-linear mode, meaning tangential direction of all the four images. We showed in
that the likelihood of the different patches and different Section 3.3 that the shape noise is less well constrained
deflector models are evaluating given the same source on-axis to the tangential arc than off-axis (comparison
morphology surface brightness amplitude. This mode of e.g. the residuals of Figure 5 for on-axis and Figure 6
has been used by Yang et al. (2020, 2021) to reconstruct for off-axis shape noise). Thus, we expect a stronger
the intrinsic sources of multiply lensed galaxies in the breaking in the off-axis direction of the shape noise (e2
cluster lensing environment. PSF and noise properties in this example), than in on-axis direction (e1 in this ex-
are matched to the input simulation during the infer- ample). On-axis shape noise is also degenerate with the
ence. tangential-to-radial stretch ratio (e.g. Figure 5). The
In total, the sampling contains 30 non-linear param- degeneracies and relative uncertainties in this example
eters. For the parameter posterior sampling we follow are a reflection and confirmation of the discussion pre-
Birrer et al. (2015). We first find a maximum likeli- sented in Section 3.3.
hood position using a Particle Swarm Optimizer (PSO;
Kennedy & Eberhart 1995) exploring a large volume of 5.1.3. Global model-dependent constraints
parameter space (200 particles for up to 500 iterations).
We can compare the constraints on the same quanti-
We then use the obtained best-fit value as a starting
ties as measured by the curved arc inference when per-
point with significantly narrower proposal distribution
forming an inference on a global deflector model and
to perform a Monte Carlo Markov Chain (MCMC) us-
then evaluating the local quantities from the global pos-
ing emcee (Foreman-Mackey et al. 2013) (with 300 par-
terior model.
ticles for 2000 burn-in and 2000 sampling iterations to
In our example, we chose as a global macro-model
ensure convergence of the chain).
as an elliptical power-law mass density (PEMD, Ap-
pendix A) model with external shear (Appendix B.1)
5.1.2. Model-independent curved arc constraints
with flat priors on all the parameters. The
Figure 10 presents the best fit reconstruction using the PEMD+shear model is a popular model of choice in
curved arc formalism. The local curved arc deflectors many applications on galaxy-scale strong gravitational
centered at the appearances of the arcs reproduce the lensing modeling applications. In addition to the source
observables to the noise level of the input data, without shape parameters, we also allow the source size param-
relying on specific assumptions on the functional form eter to vary in this scenario to be agnostic to MST
of the global macro lens model. Thus, we expect from breaking effects. The red contours in Figure 11 corre-
this modeling procedure an accurate extraction of the spond to the post-processed posterior predictions from
lensing information independent of the underlying global the global PEMD+shear model inference of the same
deflector properties. data for the same quantities as derived for the curved
Beyond the best fit, the posteriors on the curved arc arc inference. The differences in the posterior widths
parameters capture effectively the lensing information in between the curved arc measurements and a global lens
the extended data that go beyond the positional infor- model inference is attributed to the specific assumptions
mation. The blue contours in Figure 11 correspond to imposed by the choice of the macro-model parameteriza-
 Gravitational lensing formalism in a curved arc basis 17

Observed Reconstructed Normalized Residuals 6

0 0
4
1 1
2

(fmodel - fdata)/
log10 flux

log10 flux
2 2 0
3 3 2
4 4 4
5 5 6
10.0 10.0
Input Truth Best fit
7.5 7.5
5.0 5.0
2.5 2.5
det (A 1)

det (A 1)
0.0 0.0
2.5 2.5
5.0 5.0
7.5 7.5 Input Truth
Best fit
10.0 10.0

Figure 10. Example of applying curved arc description in deriving macro-model independent constraints on the lens model.
The input mock data is generated with a PEMD mass profile (top left: input mock data, lower left: local curved arc differentials
from the input truth at the arc positions and magnification map). The individual arcs are fit within separate cut-out regions
with independent tangentially curved deflector models. Only the source is demanded to share the same morphology among the
different curved arcs. The local curved deflector formalism allows us to describe the input data to the noise level (middle top:
best fit reconstruction of the arcs, top right: reduced residuals of the model - data). The lensing constraints derived from the
local tangentially curved deflector models at the position of the arcs are accurate (bottom right). The set of local tangentially
curved deflector models do not require to fully describe the global macro model in regions absent of data constraints (bottom
middle: local magnification predictions). ‡

tion and the translation of the prior space. Global mass The parameterization we chose inherently contains the
profile assumptions can be discussed in terms of required input truth and, thus, allows for an accurate recovery of
radial and tangential symmetries demanded by a cer- the input quantities. Had we chosen a different parame-
tain model (see Section 4). In tangential direction, the terization of the macro-model, the general expectation is
PEMD+shear model allows only specific configurations that the posteriors are within the margins of the curved
of the curvature direction and strength, and the tan- arc measurement, modulo an overall MST re-scaling not
gential differentials along the azimuthal direction (Sec- represented in the displayed curved arc posteriors, to be
tion 4.1). In addition, the symmetry requires a specific consistent with the data. However, any narrowing of
relation of tangential and radial stretch (Eqn. 46). The the posterior due to further implied constraints on the
asymmetry in the appearance of the images further al- macro model might lead to biases within the boundaries
lows to add constraints on the relative radial stretch dif- of the curved arc posterior.
ferential ∂r λrad /λrad beyond the explicitly measured dif- We will discuss certain aspects of this example in Sec-
ferential width in radial direction reflected in the curved tion (5.3) in more broader terms in light of possible ap-
arc posteriors. These assumptions and symmetry con- plications and limitations.
siderations allows the imposed model to break the shape
noise and the related degeneracies present in the curved 5.2. Science cases
arc inference.
In this section we highlight several science cases where
Furthermore, the PEMD+shear model imposes a
our formalism may find beneficiary applications. A more
one-to-one relation between the measurable quantity
uniform approach to quantifying lensing constraints
∂r λrad /λrad and the power-law slope (Eqn. A10). This
across different scientific studies and analyses may also
assumption imposed by the model effectively breaks the
result in an overall better ability to utilizing constraints
MST and simultaneously allows the model to constrain
obtained for originally addressing a specific science ques-
the source size.
tions and then translating the constraints to other in-
vestigations.
18 S. Birrer
rad, 0 = 1.09+0.08
0.07

Curved arcs: local data constraints

PEMD+shear: imposed macro model
tan, 0 = 3.72+0.27
0.24
0.8 0.9 1.0 1.1 1.2 4.0 4.5 5.0 5.5 6.0 0.9 1.0 1.1 1.2 1.3 3.2 3.6 4.0 4.4 4.8
tan, 0

rad, 1 = 1.10+0.07
0.07
rad, 1

tan, 1 = 4.74+0.32
0.30
tan, 1

rad, 2 = 0.94+0.07
0.06
rad, 2

tan, 2 = 4.22+0.27
0.26
4.8 4.4 4.0 3.6
tan, 2

rad, 3 = 0.92+0.07
0.06
0.8 0.9 1.0 1.1
rad, 3

tan, 3 = 2.77+0.19
0.21
3.3 3.0 2.7 2.4
tan, 3

e1 = 0.10+0.07
0.07
0.2 0.1 0.0 0.1
e1

e2 = 0.02+0.04
0.04
6 8 0 8 6
0.1 0.0 0.0 20.0 0.1
e
0.9
1.0
1.1
1.2
1.3
3.2
3.6
4.0
4.4
4.8
0.9
1.0
1.1
1.2
1.3
4.0
4.5
5.0
5.5
6.0
0.8
0.9
1.0
1.1
1.2
4.8
4.4
4.0
3.6

0.8
0.9
1.0
1.1

3.3
3.0
2.7
2.4

0.2
0.1
0.0
0.1

6
0.08
0
8
6
0.1
0.0

0.0
0.1
rad, 0 tan, 0 rad, 1 tan, 1 rad, 2 tan, 2 rad, 3 tan, 3 e1 e2

Figure 11. Comparison of a model-independent and model-dependent inference of the tangential and radial stretch eigenvectors
(λtan , λrad ) at the positions of the images (subscript 0-3). The example input and fit with curved arcs is shown in Figure 10. e1
and e2 correspond to the intrinsic source eccentricity moduli (see Appendix B.2). Blue contours: Posteriors of the extended
local arc inference. The different images are constrained by independent curved arc parameters. Only the intrinsic source is
demanded to be identical when predicting the individual arc surface brightness. The source size is held fixed, thus the posteriors
reflect a slice within the MST (i.e. re-scaling all eigenvalues results in an equally valid model with re-scaled source). Not
shown are the direction φrad , curvature stan and tangential differential ∂t λtan parameters for the individual local tangentially
curved deflector models. Uncertainties quoted in the figures correspond to these blue contours. Red contours: Post-processed
predictions of the same quantities derived from a global PEMD+shear model inference of the same data. The intrinsic source
size was a free parameter. The assumptions on the chosen global mass profile breaks the MST. Black lines: Truth input values
computed from the input lens model (PEMD+shear). The eigenvalues (modulo an overall scaling) of the blue contours can
be considered as a measurement provided by the data. No continuity in the deflection field between the curved arc locations
is required. The additional constraints between blue and red contours do solely come from the specific imposed global model
assumptions, in addition to the MST breaking in the PEMD+shear scenario. Accuracy in the red contours is only guaranteed
if the chosen lens model assumptions are valid. ‡
 Gravitational lensing formalism in a curved arc basis 19

5.2.1. Dark Matter: Locally resolved vs unresolved variant quantities from modeling imaging data, i.e. as
small-scale distortions quantified in expression 42 on the radial profile. In a
Unresolved flux ratio statistics of multiply imaged second step, one can translate these constraints with ad-
quasars is a powerful probe of small scale dark mat- ditional data, such as kinematics. Due to the tight cou-
ter clustering and constraining the nature of dark mat- pling between radial and tangential constraints, careful
ter (Dalal & Kochanek 2002; Hsueh et al. 2020; Gilman assessment of the tangential structure assumptions need
et al. 2020a,b). Interpretations of the flux ratios re- to be taken as well (Kochanek 2021). The Fermat po-
quire reference flux ratios predicted by a smooth macro tential prediction can then be re-scaled by a factor of
model. Current flux ratio statistics constraints are de- the relative local convergence at the Einstein radius be-
rived from quadruply lensed quasars only (Hsueh et al. tween the initial model used in extracting the lensing
2020; Gilman et al. 2020a) and the positional constraints information, and the one constrained by external data.
of the images and the deflector light are the primary A special case of such an analysis is presented by Birrer
sources of information to establish a macro-model ref- et al. (2020) in using the most direct parameterization
erence prediction. Assessments of potential systemat- relevant for the time-delay prediction, the MST itself,
ics in regard to an assumed macro-model parameteriza- in translating constraints from the PEMD models to a
tion have been studied by Hsueh et al. (2016); Gilman more general form of mass density profiles constrained
et al. (2017); Hsueh et al. (2017, 2018) and is a potential by stellar kinematics observations.
source of noise. Physically interpretable mass models can be well ap-
The curved arc formalism, applied in similar way as proximated by a pure MST within a range exceeding
for the example in Section 5.1, allows one to establish 10% in the MST (Kochanek 2020; Blum et al. 2020;
a reference local flux-ratio prediction based on the ex- Birrer et al. 2020). Higher-order radial differentials can
tended host galaxy, without relying on assumptions on potentially distinguish variations among the families of
the macro model. The example in Section 5.1 translates models but are hard to measure in practice. We refer to
in a 1% flux ratio prediction, below the current measure- Section 2 of Birrer et al. (2020) for a detailed discussion
ment errors of the fluxes (Nierenberg et al. 2020), thus of data constraints and physical descriptions of density
making the statistical not limited by macro-model un- profiles following approximately an MST relative to a
certainties. On one hand, such an approach requires suf- baseline model.
ficient host galaxy light components around the quasars, 5.2.3. Large scale structure and the statistics of
potentially restricting such an analysis to a subset of the gravitational lenses
quadruply lensed quasar systems. On the other hand, Searches for strong gravitational lenses in current
this approach can be also employed in using doubly and ongoing large area imaging surveys, such as the
lensed quasar, a much larger population of lenses, for Dark Energy Survey (DES) and the Hyper-Supreme-
systems with extended host information equivalently. Cam survey (HSC) have resulted in hundreds of promis-
In the fully resolved regime of extended arcs in the ab- ing galaxy-galaxy scale candidate lenses (see e.g., Jacobs
sence of quasars, a perturbative description of extended et al. 2019; Sonnenfeld et al. 2018). With the next gen-
arc might be able to replace global model fitting in char- eration large area ground and space based surveys (Vera
acterizing the abundances of small scale structure in the Rubin Observatory LSST, Euclid, Nancy Grace Roman
lens and along the line of sight, as done in the litera- Space Telescope), of order 105 galaxy-galaxy lenses will
ture (e.g., Vegetti et al. 2012; Hezaveh et al. 2016; Bir- be discovered (e.g., Collett 2015). The number of curved
rer et al. 2017a). Substructure signal is generally an arcs, where non-linear curvature can be detected, even
anomaly of required local lensing perturbations to match in the absence of a detectable counter image, may well
the appearances of multiply imaged sources with a sin- be up to an order of magnitude larger, simply by the
gle, yet inherently unknown, morphological structure of argument of lensing cross-section.
the source. The advantage of reduced shape noise in the strong
lensing regime relative to the linear lensing regime and
5.2.2. Time-delay cosmography and Hubble constant image multiplicity, combined with the expected num-
measurement
ber of curved arcs, is that we gain significant infor-
The relevant radial quantity to derive from the mass mation about the galaxy-halo connection from cluster
density profile to achieve an accurate time-delay pre- down to galaxy scales. Proposed statistical studies on
diction is the local convergence at the Einstein ra- the radial density profiles of galaxies using positional
dius, which is not a direct observable from lensing data and magnification information (see e.g., Blandford &
(Kochanek 2002). We recommend to derive solely in- Kochanek 1987; Kochanek & Blandford 1987; Sonnen-
20 S. Birrer

feld & Cautun 2021) can be enhanced with the full infor- rection, which have not been considered in this work,
mation encompassed in curved arcs. Strong lenses may may be required. Extensions to higher orders can be
also be able to provide significant cosmic shear informa- implemented within the provided framework and do not
tion (Birrer et al. 2017b, 2018; Arjun Kuhn et al. 2020; impact the general methodological questions and con-
Fleury et al. 2021), potentially even in tomographic clusions presented in this work. Continuity constraints
mode. These are only two specific examples utilizing and priors can also be added, either directly in the in-
partial information contained in the non-linear lensing ference of curved arc constraints on the data, or in post-
observables. processing on the posterior level. Continuity can be de-
The description introduced in this work may also help manded in both, the lensing differentials, as well as the
with simulations and calibrations of large scale weak total deflection. The latter is effectively demanding the
lensing surveys. In particular, in investigations into positional constraints on the arcs to be a solution of a
next-to-leading order lensing effects and potential sys- global macro model for a single source position.
tematics in the shape measurements as a cause (see In the case of fully connected arcs, effectively Einstein
Schneider & Er 2008, for such a discussion in regard rings, require full curl-free continuity in the local deflec-
to flexion) may be needed for the next generation weak tion field in azimuthal direction. In this regime, where
lensing surveys. a subset of the source is displayed along a continuous
All in all, a continuous formalism to describe observ- rotation of the tangential direction, the shape-noise de-
ables from the weak to the strong lensing regime allows generacy is most effectively been broken.
one to self-consistently combine the currently distinct However, we stress that invariances under the MST
cosmological probes gaining synergies and complemen- remain even in the regime of fully connected arcs, par-
tarity in systematics and constraints. ticularly impacting the constraining power in the radial
direction.
5.3. Discussion of limitations and extensions
6. CONCLUSION
The focus of this work is primarily to present a frame-
In this work, we introduced a formalism to describe
work and methodology to allow the science investigator
the gravitational lensing distortion effects of curved ex-
to assess impacts of certain assumptions on specific sci-
tended arcs based on the eigenvectors and eigenvalues
ence cases and to translate constraints on lensing quan-
of the local lensing Jacobian and their directional differ-
tities beyond a given family of mass models. This work
entials. We identified a set of non-linear extended de-
does not state whether or not certain assumptions on
flector descriptions that inherit the local properties able
the global deflector mass distribution are valid for spe-
to describe the extent of individual lensed images. Our
cific science investigations and their stated uncertain-
parameterization is tightly linked to observable features
ties. One globally imposed constraint that is valid for
in extended sources and allows one for an accurate ex-
any physical deflector mass distribution is the continu-
traction of the relevant information of extended images
ity in the deflection field. A set of local arc models do
without imposing an explicit global deflector model.
not demand this continuity between the different arcs,
We re-formulate the most general lensing invariance
as, for example, illustrated for the example presented
in an operator notation and subsequently quantify what
in Section 5.1 in Figure 10 by the disconnected criti-
aspects can be broken based on specific assumptions on
cal curves. When the individual curved arcs are suffi-
the local lensing nature and assumed intrinsic source
ciently separated from each other without constraining
shape.
data in between, dropping the continuity assumption is
Our main findings are:
a practical convenience for being agnostic to the deflec-
tion field behavior outside the data constraining region, 1. The non-linear lensing nature in curved arcs allows
and counting on a physical model that is able to contin- one to partially break the shape-noise degeneracy.
uously connect the different local regions. In particular, shape noise off-axis to the eigen-
In many real lensing configurations, arcs extend over vector directions can be constrained while on-axis
large azimuthal angles, effectively adding more con- shape distortions are more degenerate with lensing
straints on the azimuthal structure of the deflector and eigenvalues and curvature.
physically demanding stronger assumptions on the con-
tinuity of the global deflection field. More extended arcs 2. Elliptical mass distributions lead to tangential
most likely require also more and higher order local dif- stretch gradients but keep the curvature radius
ferentials to match the observations. In particular, dif- along the azimuth constant. External shear dis-
ferentials associated with the curvature strength and di- tortions do, in addition to tangential stretch gra-
 Gravitational lensing formalism in a curved arc basis 21

dients, lead to offsets in the curvature radius and framework to assess systematics, and to guide an in-
direction along the azimuth (Section 4.1). ference effort in complexity choices based on the data at
hand. Implementations of all the aspects presented in
3. Information on the radial mass profile can be ob- this work are available in lenstronomy. The specific
tained by measuring the differential thickness of examples and discussions provided in this work can serve
arcs along the radial direction, the radial distance as a baseline for more extended theoretical and practi-
ratio of image pairs, or, when imposing azimuthal cal investigations and assessments in different regimes of
constraints, by the tangential stretch change along gravitational lensing and for different scientific investiga-
the radial direction (Section 4.2). tions. We outline applications and implications for dark
matter substructure inferences, measuring the Hubble
4. Imposing symmetries on the global form of the
constant, and large scale structure inferences from the
tangential behavior of the deflector profile can
statistics of gravitational lenses.
break the shape noise while imposing functional
forms on the radial deflector profile can break the ACKNOWLEDGMENTS
MST.
SB thanks Anowar Shajib, Dominique Sluse, Daniel
Our formalism is applicable in all regimes of gravi- Gilman, Tommaso Treu, Martin Millon, Lyne van de
tational lensing, from the weak linear regime, the semi- Vyvere and Roger Blandford for useful feedback in the
linear regime up to the highly non-linear regime of highly process of writing this manuscript. Support for this
magnified arcs and Einstein rings of multiple images. work was provided by the National Science Foundation
The methodology presented in this work provides a through NSF AST-1716527.

APPENDIX

A. POWER-LAW ELLIPTICAL MASS DISTRIBUTION (PEMD)

A.1. Parameterization
The elliptical power-law mass distribution can be defined as7
!γ 0 −1
3 − γ0 θE
κ(θ1 , θ2 ) = p , (A1)
2 qθ12 + θ22 /q

where q is the semi-minor to semi-major axis ratio, θE is the Einstein radius, and γ 0 is the logarithmic slope of the
three-dimensional mass profile. γ 0 = 2 is an isothermal profile, the limit of γ 0 → 3 results in a point mass and γ 0 → 1
describes a uniform critical mass sheet. The coordinates (θ1 , θ2 ) are rotates such that θ1 is along the semi-major axis.
Alternatively, the same profile can be defined as
!γ 0 −1
3 − γ0 0
θE
κ(θ) = p , (A2)
2 θr 1 −  cos(2φ)

with θr is the radial distance to the center, φ is the angle relative to the major axis, and  is the ellipticity, which is
related to the axis ratio, q, by
1 − q2
= . (A3)
1 + q2
To provide an identical normalization of the deflection angles, the Einstein radii of expression (A1), θE , and of
0
expression (A2), θE ,need to follow the relation
 0
2
θE 2q
= . (A4)
θE 1 + q2

7 This is the current lenstronomy convention with version 1.8.1

and previous versions.
22 S. Birrer

The Einstein radius definition of expression (A1) is such that the square average of the deflection angle along the
0
semi-major and semi-minor axis corresponds to θE , while expression (A2) matches the Einstein radius θE in directions
half-way between the semi-major and semi-minor axes.
Computations for deflection angles and lensing potential are provided by Barkana (1998); Tessore & Metcalf (2015)8 .

A.2. Eigenvectors in spherical case

In the spherical case of the PEMD profile (Eqn. A1, A2), the deflection deflection angle and differentials are simple
analytical expressions. The deflection angle in radial direction is given by
 γ 0 −2
θE
α(r) = θE (A5)
r

with θE is the Einstein radius and γ 0 is the three-dimensional power-law slope of the mass profile.
The tangential and radial eigenvalues are given by
 γ 0 −1
1 θE
=1− (A6)
λtan r
and  γ 0 −1
1 0 θE
= 1 + (γ − 2) . (A7)
λrad r
The radial differential of the tangential eigenvalue, ∂r λtan , is given by

γ 0
(1 − γ 0 ) θrE
∂r λtan = 
γ 0 −1 2 , (A8)
θE 1 − θrE

and the radial differential of the radial eigenvalue, ∂r λrad , is given by

γ
(1 − γ 0 )(2 − γ 0 ) θrE
∂r λrad = 
γ 0 −1 2 . (A9)
θE 1 + θrE (γ 0 − 2)

At the Einstein radius θE , we can express the MST invariant quantity ∂r λrad /λrad as

∂r λrad (θE ) γ0 − 2
= , (A10)
λrad (θE ) θE

and the overall lensing scale invariant quantity ξrad (Eqn. 42) is given by

ξrad = γ 0 − 2. (A11)

This relation reflects the fact that the MST-invariant observational constraint captured by ξrad , when interpreted as
a constant power-law mass density, constrain the power-law slope and effectively breaks the MST.

A.3. Eigenvectors in elliptical case

In the elliptical case, the eigenvectors and directions remain the same as for the round case, substituting the r by
the circularized expression r0 (i.e. denominator of expression A1 or A2).
In addition, a non-zero second order differential emerges, namely the tangential stretch differential in tangential
direction, ∂t λtan . Using the chain rule, we can write
∂λtan ∂r0
∂t λtan = . (A12)
∂r0 ∂etan

8 Both computational methods to compute lensing properties are

implemented in lenstronomy.
 Gravitational lensing formalism in a curved arc basis 23

Adopting the ellipticity definition of the form of expression (A2), such that
p
r0 = r 1 −  cos(2φ), (A13)

we can write
∂r0 ∂r0 ∂φ  sin(2φ)
= =p . (A14)
∂etan ∂φ ∂etan 1 −  cos(2φ)
Combining Equations (A12) and (A14), we can compactly write
∂λtan  sin(2φ)
∂t λtan = 0
p , (A15)
∂r 1 −  cos(2φ)
with the first term given by the round case of expression (A8).

A.4. Curved arc description with tangential stretch differential

The constant radial and tangential eigenvalue deflector model with constant tangential curvature is presented in
Section 2.4 and can be written as a combination of a SIS and an MST. A convenient way to introduce a model
satisfying the same quantities locally and has an additional tangential differential component ∂t λtan is the singular
isothermal ellipsoid (SIE) replacing the SIS profile. The SIE is the special case of the PEMD for γ 0 = 2 (Appendix A)
and ∂t λtan is given by expression A15. To satisfy locally constraints on ∂t λtan with the next leading order is set to
zero is satisfied with an eccentricity off-axis by π/4 to the curvature direction. The off-axis direction is determined by
the sign of imposed ∂t λtan .

B. ELLIPTICITY AND SHEAR

B.1. Shear
Shear distortions are fully characterized by the constant γ1 and γ2 values (Eqn. 5, 6) and lead to deflection filed
" #
γ1 γ2
α(θ) = (θ − θ0 ), (B16)
γ2 −γ1

where θ0 is a, somewhat arbitrary, zero point of the deflection field, only impacting an overall constant shift of the
deflection angle. In polar coordinates, we can also equivalently parameterize the shear distortions with an absolute
shear strength γ and an orientation relative to the first axis φγ . The Cartesian shear components are then given by

γ1 = γ cos(2φγ ) γ2 = γ sin(2φγ ). (B17)

A pure shear distortions do have a magnification effect µ = (1 − γ12 − γ22 )−1 and thus do alter the size of the object
in addition of causing distortions. Shape distortions agnostic to the intrinsic source size are generally referred to as
reduced shear, and given by
0 γ1,2
γ1,2 = . (B18)
1−κ
In terms of a descriptive lens model, we require the knowledge of kappa in this notion. However, we can introduce
a reduced shear model which, by design, has magnification µ = 1. Such a model, when defined by reduced shear
components (γ10 , γ20 ), requires a convergence of
1
κ=1− p 0 0
. (B19)
1 − γ12 − γ22
We define the normalized reduced shear model, Lnrs , as the linear distortion model with parameters γ10 and γ20 where
the convergence term is set by Equation (B19). This specific linear distortion parameterization preserves the total
magnification. The inverse lensing operator is given by the same operator with flipped signs in the reduced shear
components
L−1 0 0 0 0
nrs (γ1 , γ2 ) = Lnrs (−γ1 , −γ2 ). (B20)
This specific from of the shear description becomes relevant in Appendix B.2 when discussing intrinsic surface brightness
ellipticity and degeneracies with shear distortions.
24 S. Birrer

B.2. Ellipticity
A convenient way to describe elliptical surface brightness distributions is by the axis ratio q and orientation φI
of annuli of constant surface brightness. A surface brightness profile with constant ellipticity can be described as
a distortion transform of a radial surface brightness profile Ir (r), with ellipticity operator E(x, y), such that Ie =
Ir (E(x, y)). Different ellipticity operators are used in the literature. Differences exist in the definition of the ellipticity
as well as the overall size change. We are using the operator
√ √
E(q) : (x, y) → ( qx, y/ q) (B21)

where x is in the orientation of the major axis. This is the same ellipticity operator as used in the PEMD profile
defined by expression (A1). The operator form of expression (B21) conserves the product-averaged radius.
A convenient basis to express the axis ratio q and the orientation angle φI is with the eccentricity moduli
1−q 1−q
e1 = cos(2φI ) e2 = sin(2φI ). (B22)
1+q 1+q
B.3. Shape noise
In the basis of the eccentricity moduli (Eqn. B22), the ellipticity operator E is identical to the lensing distortion
operator Lnrs (see Section B.1) by identifying γ1 = e1 and γ2 = e2 .
Using these bases for shear and ellipticity, we can identify the shape noise component of the SPT as

1 = Lnrs (−e1 , −e2 )◦E(e1 , e2 ), (B23)

and enables a separability of the MST component and shape noise component.

REFERENCES
Arjun Kuhn, F., Bruderer, C., Birrer, S., Amara, A., & Birrer, S., Shajib, A. J., Galan, A., et al. 2020, A&A, 643,
Réfrégier, A. 2020, arXiv e-prints, arXiv:2010.08680. A165, doi: 10.1051/0004-6361/202038861
[Link] Birrer, S., Shajib, A. J., Gilman, D., et al. 2021, Journal of
Bacon, D. J., Goldberg, D. M., Rowe, B. T. P., & Taylor, Open Source Software, 6, 3283, doi: 10.21105/joss.03283
A. N. 2006, MNRAS, 365, 414, Blandford, R. D., & Kochanek, C. S. 1987, ApJ, 321, 658,
doi: 10.1111/j.1365-2966.2005.09624.x doi: 10.1086/165660
Barkana, R. 1998, Astrophysical Journal v.502, 502, 531. Blandford, R. D., & Narayan, R. 1992, ARA&A, 30, 311,
[Link] query? doi: 10.1146/[Link].30.1.311
bibcode=1998ApJ...502..531B&link type=ABSTRACT Blum, K., Castorina, E., & Simonović, M. 2020, ApJL, 892,
Bartelmann, M., & Schneider, P. 2001, PhR, 340, 291, L27, doi: 10.3847/2041-8213/ab8012
doi: 10.1016/S0370-1573(00)00082-X Collett, T. E. 2015, ApJ, 811, 20,
Bernstein, G. M., & Jarvis, M. 2002, AJ, 123, 583, doi: 10.1088/0004-637X/811/1/20
doi: 10.1086/338085 Dai, L., Kaurov, A. A., Sharon, K., et al. 2020, arXiv
Birrer, S., & Amara, A. 2018, Physics of the Dark Universe, e-prints, arXiv:2001.00261.
22, 189, doi: 10.1016/[Link].2018.11.002 [Link]
Birrer, S., Amara, A., & Refregier, A. 2015, ApJ, 813, 102, Dalal, N., & Kochanek, C. S. 2002, ApJ, 572, 25,
doi: 10.1088/0004-637X/813/2/102 doi: 10.1086/340303
—. 2016, JCAP, 08, 020, Falco, E., Gorenstein, M., & Shapiro, I. 1985, Astrophysical
doi: 10.1088/1475-7516/2016/08/020 Journal, 289, L1.
—. 2017a, JCAP, 2017, 037, [Link] query?
doi: 10.1088/1475-7516/2017/05/037 bibcode=1985ApJ...289L...1F&link type=ABSTRACT
Birrer, S., Refregier, A., & Amara, A. 2018, ApJL, 852, Fleury, P., Larena, J., & Uzan, J.-P. 2019, PhRvD, 99,
L14, doi: 10.3847/2041-8213/aaa1de 023526, doi: 10.1103/PhysRevD.99.023526
Birrer, S., Welschen, C., Amara, A., & Refregier, A. 2017b, —. 2021, arXiv e-prints, arXiv:2104.08883.
JCAP, 2017, 049, doi: 10.1088/1475-7516/2017/04/049 [Link]
 Gravitational lensing formalism in a curved arc basis 25

Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, Leonard, A., Goldberg, D. M., Haaga, J. L., & Massey, R.
J. 2013, PASP, 125, 306, doi: 10.1086/670067 2007, ApJ, 666, 51, doi: 10.1086/520109
Gilman, D., Agnello, A., Treu, T., Keeton, C. R., & Luhtaru, R., Schechter, P. L., & de Soto, K. M. 2021, arXiv
Nierenberg, A. M. 2017, MNRAS, 467, 3970, e-prints, arXiv:2102.08470.
doi: 10.1093/mnras/stx158 [Link]
Gilman, D., Birrer, S., Nierenberg, A., et al. 2020a, Mellier, Y. 1999, ARA&A, 37, 127,
MNRAS, 491, 6077, doi: 10.1093/mnras/stz3480 doi: 10.1146/[Link].37.1.127
Nierenberg, A. M., Gilman, D., Treu, T., et al. 2020,
Gilman, D., Du, X., Benson, A., et al. 2020b, MNRAS, 492,
MNRAS, 492, 5314, doi: 10.1093/mnras/stz3588
L12, doi: 10.1093/mnrasl/slz173
O’Riordan, C. M., Warren, S. J., & Mortlock, D. J. 2019,
Goldberg, D. M., & Bacon, D. J. 2005, ApJ, 619, 741,
MNRAS, 487, 5143, doi: 10.1093/mnras/stz1603
doi: 10.1086/426782
—. 2020, MNRAS, 496, 3424, doi: 10.1093/mnras/staa1697
Goldberg, D. M., & Natarajan, P. 2002, ApJ, 564, 65, Schechter, P. L., & Wynne, R. A. 2019, ApJ, 876, 9,
doi: 10.1086/324202 doi: 10.3847/1538-4357/ab1258
Gorenstein, M. V., Falco, E. E., & Shapiro, I. I. 1988, ApJ, Schneider, P., & Er, X. 2008, A&A, 485, 363,
327, 693, doi: 10.1086/166226 doi: 10.1051/0004-6361:20078631
Hezaveh, Y. D., Dalal, N., Marrone, D. P., et al. 2016, ApJ, Schneider, P., & Sluse, D. 2013, A&A, 559, A37,
823, 37, doi: 10.3847/0004-637X/823/1/37 doi: 10.1051/0004-6361/201321882
Hsueh, J.-W., Despali, G., Vegetti, S., et al. 2018, MNRAS, —. 2014, A&A, 564, A103,
475, 2438, doi: 10.1093/mnras/stx3320 doi: 10.1051/0004-6361/201322106
Hsueh, J. W., Enzi, W., Vegetti, S., et al. 2020, MNRAS, Shajib, A. J., Treu, T., Birrer, S., & Sonnenfeld, A. 2021,
MNRAS, doi: 10.1093/mnras/stab536
492, 3047, doi: 10.1093/mnras/stz3177
Sonnenfeld, A. 2018, MNRAS, 474, 4648,
Hsueh, J. W., Fassnacht, C. D., Vegetti, S., et al. 2016,
doi: 10.1093/mnras/stx3105
MNRAS, 463, L51, doi: 10.1093/mnrasl/slw146
Sonnenfeld, A., & Cautun, M. 2021, arXiv e-prints,
Hsueh, J. W., Oldham, L., Spingola, C., et al. 2017, arXiv:2102.08973. [Link]
MNRAS, 469, 3713, doi: 10.1093/mnras/stx1082 Sonnenfeld, A., Chan, J. H. H., Shu, Y., et al. 2018, PASJ,
Irwin, J., & Shmakova, M. 2005, NewAR, 49, 83, 70, S29, doi: 10.1093/pasj/psx062
doi: 10.1016/[Link].2005.01.032 Suyu, S. H. 2012, MNRAS, 426, 868,
—. 2006, ApJ, 645, 17, doi: 10.1086/504100 doi: 10.1111/j.1365-2966.2012.21661.x
Irwin, J., Shmakova, M., & Anderson, J. 2007, ApJ, 671, Tessore, N. 2017, A&A, 597, L1,
1182, doi: 10.1086/522819 doi: 10.1051/0004-6361/201629947
Jacobs, C., Collett, T., Glazebrook, K., et al. 2019, ApJS, Tessore, N., & Metcalf, R. B. 2015, A&A, 580, A79,
243, 17, doi: 10.3847/1538-4365/ab26b6 doi: 10.1051/0004-6361/201526773
Unruh, S., Schneider, P., & Sluse, D. 2017, A&A, 601, A77,
Kaiser, N. 1998, ApJ, 498, 26, doi: 10.1086/305515
doi: 10.1051/0004-6361/201629048
Kaiser, N., & Squires, G. 1993, ApJ, 404, 441,
Vegetti, S., Lagattuta, D. J., McKean, J. P., et al. 2012,
doi: 10.1086/172297
Nature, 481, 341, doi: 10.1038/nature10669
Kaiser, N., Squires, G., & Broadhurst, T. 1995, ApJ, 449, Wagner, J. 2019, Universe, 5, 177,
460, doi: 10.1086/176071 doi: 10.3390/universe5070177
Kennedy, J., & Eberhart, R. 1995, in Proceedings of Wagner, J., & Bartelmann, M. 2016, A&A, 590, A34,
ICNN’95 - International Conference on Neural Networks, doi: 10.1051/0004-6361/201527540
Vol. 4, 1942–1948 vol.4, doi: 10.1109/ICNN.1995.488968 Wertz, O., Orthen, B., & Schneider, P. 2018, A&A, 617,
Kochanek, C. S. 2002, ApJ, 578, 25, doi: 10.1086/342476 A140, doi: 10.1051/0004-6361/201732240
—. 2020, MNRAS, 493, 1725, doi: 10.1093/mnras/staa344 Yang, L., Birrer, S., & Treu, T. 2020, MNRAS, 496, 2648,
—. 2021, MNRAS, 501, 5021, doi: 10.1093/mnras/staa4033 doi: 10.1093/mnras/staa1649
Yang, L., Roberts-Borsani, G., Treu, T., et al. 2021,
Kochanek, C. S., & Blandford, R. D. 1987, ApJ, 321, 676,
MNRAS, 501, 1028, doi: 10.1093/mnras/staa3713
doi: 10.1086/165661