The formation of a soliton gas condensate for the focusing

Nonlinear Schrödinger equation

Aikaterini Gkogkou1 , Guido Mazzuca1 , and Kenneth D. T-R McLaughlin1
1
Department of Mathematics, Tulane University, New Orleans, LA
arXiv:2502.14749v1 [math.AP] 20 Feb 2025

February 21, 2025

Abstract
In this work, we carry out a rigorous analysis of a multi-soliton solution of the focusing
nonlinear Schrödinger equation as the number, N , of solitons grows to infinity. We discover
configurations of N -soliton solutions which exhibit the formation (as N Ñ 8) of a soliton gas
condensate. Specifically, we show that when the eigenvalues of the Zakharov - Shabat operator
for the NLS equation accumulate on two bounded horizontal segments in the complex plane
with norming constants bounded away from 0, then, asymptotically, the solution is described
by a rapidly oscillatory elliptic-wave with constant velocity, on compact subsets of px, tq. This
is to be distinguished from previous analyses of soliton gasses where the norming constants
were tending to zero with N , and the asymptotic description only included elliptic waves in the
long-time asymptotics.

1 Introduction
The focusing Nonlinear Schrödiner (NLS) equation is one of the fundamental examples of an
integrable partial differential equation for a complex valued unknown ψpx, tq:
1
iBt ψ “ ´ Bx2 ψ ´ |ψ|2 ψ , px, tq P R. (1.1)
2
This equation is a canonical model for the deformation of a laser pulse traveling down an optical
fiber transmission line [35], for one dimensional wave propagation in deep water, and several
other physical phenomena [41].
The NLS equation (1.1) was discovered to be integrable by Zakharov and Shabat [33]. The
equation arises as the compatibility condition of the pair of two linear differential equations, the
so-called Lax pair:
ˆ ˙
´iz ψ
Bx Φ “ Φ,
´ψ iz
´iz 2 ` 2i |ψ|2 zψ ` 2i ψx
¨ ˛

Bt Φ “ ˝ ‚Φ
i 2 i 2
´zψ ` 2 ψ x iz ´ 2 |ψ|

where z P C is a spectral parameter. In order for there to exist a simultaneous solution to both
of these differential equations, ψpx, tq must solve the NLS equation. But more importantly, as
ψpx, tq evolves according to the NLS equation, the scattering data associated to the differential
equation (1.2) evolve in an explicit way, which yields a solution procedure for the NLS equation:

1
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 2

from the initial data one computes the scattering data (to be defined momentarily), whose
evolution in t is explicit. For any t ą 0, the solution ψpx, tq is determined via the inverse
scattering transform. It is then the analysis of the direct scattering transform, and the inverse
transform at later times, which provides meaningful information about the behavior of solutions
to the NLS equation.
The NLS equation possesses a broad family of fundamental solutions. First among these is
the family of soliton solutions:
2
´η 2 qts ´iϕ0
ψs “ 2η sech p2ηpx ` 2ξt ´ x0 qq e´2irξx`pξ e ,

parametrized by four parameters, ξ, η, ϕ0 , and x0 . Soliton solutions are exponentially localized

traveling wave solutions, and represent canonical nonlinear phenomena which has been observed
and exploited in many areas of application, including fluid dynamics, nonlinear optics, and other
areas in the quantum realm [21, 22, 26, 36]. But there are many more solutions in the catalog,
including periodic and quasi-periodic solutions described using algebro-geometric function the-
ory, dispersive shock waves, breathers, as well as the object of study in this paper: multi-soliton
solutions, which we will refer to as N -soliton solutions. The positive integer N refers to the
number of fundamental solitons that are in some sense present in the solution.
An N -soliton solution typically exhibits the following asymptotic behavior as |t| Ñ 8: the
solution decomposes into a collection of well-separated, exponentially localized traveling wave
solutions, each of which asymptotically approaches a soliton solution as in (1.4) with predeter-
mined parameters.
The soliton, and N -soliton solutions, were part of the original discovery of integrable non-
linear partial differential equations, and there emerged very quickly a dual interpretation of the
“solitonic component” of a solution as a collection of particles, at least in the |t| Ñ 8 asymptotic
regime. Of course, for intermediate times, it is not easy to identify any individual solitons in a
solution of the PDE, but nonetheless the particle interpretation persisted, with Zakharov [40]
proposing a kinetic theory of solitons in 1971. The kinetic theory of solitons is a qualitative
theory describing the evolution of a density function for solitons, supposing them to be indi-
vidual particles which asymptotically interact in a pair-wise fashion. In the last decade, there
has emerged new experimental, numerical, and analytical justification for this qualitative theory
(see [36] for a review of these developments).
In this paper we develop the analytical description of N -soliton solutions in a new region of
space-time which has been numerically observed (and conjectured) to correspond to a special
type of soliton gas called a soliton condensate. The paper is organized as follows. In Section 2
we present a very brief introduction to the kinetic theory of solitons, explain the characteriza-
tion of N -soliton solutions in the soliton gas condensate scaling via Riemann-Hilbert problems,
formulate a Riemann-Hilbert problem for the limiting soliton gas condensate solution, and de-
scribe our results regarding the behavior of N -soliton solutions of the NLS equation in the
condensate scaling. In Section 3-6 we carry out the asymptotic analysis of the Riemann-Hilbert
problem for the soliton gas condensate. In Section 7 we present the proof of our main results,
using the previously mentioned asymptotic analysis. Some technical results are deferred to the
appendices.

2 N -soliton solutions, soliton gasses, and soliton gas con-

densates
2.1 Direct and Inverse Scattering for the NLS equation
The direct scattering transform is a mapping from a function ψpxq to the scattering data asso-
ciated to the differential equation (1.2). We will give a very brief description here, and refer the
reader to [6] for a more detailed description. One studies two sets of Jost solutions Φ˘ px, zq of
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 3

this differential equation, which are uniquely determined by a normalization condition

ˆ ˙
1 0
˘
Φ px, zqe izxσ3
Ñ I as x Ñ ˘8 , σ3 “ .
0 ´1

The matrix-valued function Φ` px, zq is a fundamental solution to (1.2), as is Φ´ px, zq. This
fact yields the scattering relation between these two solutions,
ˆ ˙ ˆ ˙
apzq ´bpzq apzq ´bpzq
Φ´ px, zq “ Φ` px, zq , det “ |apzq|2 ` |bpzq|2 “ 1 ,
bpzq apzq bpzq apzq

which in turn yields the reflection coefficient rpzq “ bpzq{apzq and the transmission coefficient
1
T pzq “ apzq .
It turns out that the Jost solutions have analyticity properties in the variable z, which follows
from the usual existence theory. Specifically, the first column of Φ´ px, zq (and the second column
of Φ` px, zq) has an analytic extension to C` , the upper-half of the complex z plane, where it
exhibits exponential decay for x Ñ ´8 (and the second column of Φ` px, zq decays exponentially
for x Ñ `8). Similarly, the second column of Φ´ px, zq and the first column of Φ` px, zq have
analytic extensions to C´ .
The transmission coefficient also has an analytic extension to C` , and each pole of T pzq in
C` corresponds to an L2 eigenfunction for the differential equation (1.2). For reasonable classes
of functions ψpxq, there are at most a finite number of poles of T pzq in C` . At each pole λj ,
the first column of Φ´ px, λj q and the second column of Φ` px, λj q are linearly dependent, and
the proportionality constant cj relating them is referred to as a normalization constant, and is
defined by
` ˘p1q ` ˘p2q
Φ´ px, λj q “ cj Φ` px, λj q .

The scattering data (also referred to as the spectral data) corresponding to ψpxq consists of
the reflection coefficient rpkq, the eigenvalues tλj uN
j“1 , and the normalization constants tcj uj“1 .
N

Following [6], we let D denote the scattering data:

␣ (
D “ rpzq, tλj , cj uN
j“1 .

As ψpx, tq evolves in time according to the NLS equation (1.1), the scattering data also evolves
in time, but that evolution is quite simple since the direct scattering transform linearizes the
flow: the eigenvalues do not change in time, and the reflection coefficient and normalization
constants evolve explicitly, so that
␣ ( ! 2itz 2 2
)
Dptq “ rpz, tq, tλj , cj ptquN
j“1 “ rpzqe , tλj , cj e2itλj uN
j“1 .

We note in passing that the direct scattering transform as discussed above is defined for
generic initial data. However, for non-generic data certain spectral singularities exist which
require additional information and analysis. As we are interested in reflectionless potentials for
which rpzq ” 0, spectral singularities are not an issue.
The solution ψpx, tq is uniquely determined by the scattering data, and the mapping from
Dptq to ψpx, tq is referred to as the inverse scattering transform. The inverse scattering transform
can be characterized in terms of a system of integral equations, referred to as Gelfand-Levitan-
Marcenko equations. Equivalently, it can be characterized in terms of a Riemann-Hilbert prob-
lem, which is the approach we take here. Riemann-Hilbert formulations of the inverse problem
have over the last 30 years yielded a remarkable number of different asymptotic results concern-
ing the behavior of solutions of integrable nonlinear partial differential equations, starting with
the work of Deift and Zhou [11] for the modified KdV equation.
For the NLS equation, the Riemann-Hilbert problem is to determine a 2 ˆ 2 matrix M “
M pzq “ M pz; x, tq. We will usually suppress the dependence on x and t, and simply write M
or M pzq.
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 4

RHP 2.1. Given spectral data Dptq find a 2 ˆ 2 matrix-valued function M pzq such that
• M pzq is meromorphic in CzR, with simple poles at the points tλj , λj uN
j“1 .
• For z P R, M has boundary values M` pzq and M´ pzq, which satisfy the jump relation
ˆ ˙
1 ` |rpzq|2 rpzqe´2iθpz;x,tq
M` pzq “ M´ pzq JM pzq, JM “ , (2.4)
rpzqe2iθpz;x,tq 1

where θpz; x, tq “ zx ` 2z 2 t.
• At each of the poles λj in the upper half-plane, M pzq satisfies the residue condition
„ ˆ ˙ȷ
0 0
resλj M pzq “ lim M pzq , (2.5)
zÑλj cj e2iθpλj ;x,tq 0

and at each pole λj in the lower half-plane:

„ ˆ ˙ȷ
0 ´cj e´2iθpλj ;x,tq
resλj M pzq “ lim M pzq . (2.6)
zÑλj 0 0

• M pzq “ I ` Opz ´1 q as z Ñ 8.
Existence and uniqueness for this Riemann-Hilbert problem follows from what are by now
standard arguments: uniqueness follows from Liouville’s theorem, and existence follows from a
careful application of a vanishing lemma as described in [42]. The solution ψpx, tq to the NLS
equation corresponding to the initial data ψpxq and determined by the scattering data Dptq is
obtained from M pz; x, tq through the z Ñ 8 normalization:
ˆ ´8 ˙ ˆ ˙
1 ´ x |ψps, tq|2 ds ψpx, tq 1
M pz; x, tq “ I ` ´8 2 ` O .
2iz ψpx, tq |ψps, tq| ds z 2
x

We mention that the RHP approach to solve non-linear integrable PDEs is not just a relevant
analytical tool, but it is also numerical one. Indeed, several authors [4, 5, 20, 39] employed this
method to numerically analyze various non-linear integrable PDEs.
Notation: We will use D to denote the scattering data at t “ 0 (i.e. D “ Dp0q). In order to
emphasize the dependence on the scattering data, we will occasionally write

ψpx, t; Dq

to represent the solution to the NLS equation determined by the initial scattering data D.

2.2 N -soliton solutions and soliton gasses

In this paper, we are interested in N -soliton solutions, for which rpzq ” 0. Under this assump-
tion, the above Riemann-Hilbert problem simplifies. Indeed, the jump relation becomes trivial:
M` pzq “ M´ pzq for z P R, and this implies that now M is a meromorphic function of z. The
Riemann-Hilbert problem is referred to as a meromorphic Riemann-Hilbert problem, for which
asymptotic analysis started in [25] where the authors studied a semi-classical scaling of the NLS
equation. (See also [1], where the authors studied the asymptotic behavior of discrete orthogonal
polynomials, through the analysis of meromorphic Riemann-Hilbert problems).
The original set of kinetic equations in [40] were derived under the loose assumption of a
dilute gas of solitons, described by an evolving density function f pz, x, tq. This density function
yields (in some sense) the number of solitons in a small volume element around pz, x, tq. Of course
the notion of a soliton in a small space-time window, let alone the idea that there might be a
large number of solitons with slightly different eigenvalue parameters in that same space-time
window, while at the same time supposing there is enough space between solitons to consider
them to be dilute, was quite challenging from the point of view of analysis.
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 5

One experimental prediction in [40] is that a tracer soliton with arbitrary eigenvalue param-
eter, launched into this gas, should interact with the gas in such away that its velocity deforms
as it propagates, and the explicit computation of this time-dependent velocity, as a function of
density of the gas as well as the tracer’s initial eigenvalue parameter, yielded an experimental
prediction which could in principle be tested. About 30 years later, El [16] extended the kinetic
theory to a dense gas, in which a coupled system of equations was derived, with the evolving
density coupled to the evolving velocity of a tracer. Both [40] and [16] focused on the KdV
equation, but this was extended by El and Kamchatnov in [13] to other integrable nonlinear
partial differential equations, including the NLS equation. Significant progress regarding the
understanding of the kinetic equations arising in different settings, as well as more detailed
analysis of the kinetic equations themselves, has continued, for example in [14, 27, 37, 38].
However, only recently has there been any validation of the kinetic theory, in the form of
robust numerical [7, 8, 32] and experimental [29, 30] evidence (see the review manuscript [36]).
There have also been a few limited analytical results, particularly [17–19, 24].
In [17] the authors considered the N Ñ 8 limit `of an ˘N -soliton solution for the KdV equation,
with normalization constants uniformly of size O N ´1 . The N Ñ 8 analysis was carried out
under the assumption that x and t were bounded. The fact that the normalization constants
were vanishingly small as N Ñ 8 provided a necessary foothold for the rigorous analysis. This
can be understood (intuitively) by considering the connection, in the case of a single soliton,
between the normalization constant and the initial location x0 of the soliton’s peak:
ˇ ˇ
1 ˇ c ˇ
x0 “ log ˇˇ ˇˇ
2η 2η

(here c is the normalization constant, and λ “ ξ ` iη is the eigenvalue), so if c « N ´1 , then x0 is

shifted a distance log N to the left. One could then imagine that if one has an N -soliton solution
with all the normalization constants of size O N ´1 , all the solitons are shifted far to the left,
` ˘

and what remains for finite values of x is the accumulated tails of all these solitons. The results
in [17] are consistent with this intuition. They prove convergence of the N -soliton solutions (as
N Ñ 8) to a “soliton gas solution” of the KdV equation, but only for x and t in compact sets.
They subsequently carry out an asymptotic analysis of the soliton gas solution for |x| Ñ 8, and
also for t Ñ 8 with |x| t bounded. In the large-time analysis of the soliton gas solution, they
established the existence of an evolving density function, depending on the spectral parameter,
space, and time, which describes the local, evolving solitonic content of the solution, providing
a first analytical corroboration of the kinetic theory, and in the process showing that in the
long-time regime, what emerges is a type of dense soliton gas, called a condensate. However,
in [17] they did not consider the behavior of the original N -soliton solution on space-time scales
that would overlap with their asymptotic analysis of the soliton gas solution.
More recently, in [2,3,15] the authors considered the soliton gas and the breather gas for the
NLS equation under the assumptions that the eigenvalues λj accumulate on a domain D Ă C`
(and its complex conjugate D) and norming constants uniformly OpN ´1 q.
In [18], the authors considered the mKdV equation and studied the interaction of a tracer
soliton with a similar soliton gas with vanishingly small normalization constants. In [19] the
authors developed a probabilistic analysis of a soliton gas for the `NLS ˘equation, again under
the assumption that the normalization constants are uniformly O N ´1 . In other words, the
“bulk” asymptotic behavior of N -soliton solutions, with normalization constants chosen to be
bounded (away from both 0 and 8), has not been considered in the context of the kinetic theory
of solitons. (However, the works [25, 31] concerning the semi-classically scaled NLS equation,
could be thought of as examples of this scaling.)
Remark 2.1. Similarly, if c is large, then x0 is shifted far to the right, and so the above intuitive
suggests that if all the normalization constants are O pN q, then again for x and t on compact
sets, one only sees the accumulated tails of all the solitons.
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 6

2.3 N -soliton solutions: condensate scaling

The present manuscript considers the asymptotic behavior of N -soliton solutions of the NLS
equation, with eigenvalues accumulating on a horizontal line in C` (and its complex conjugate
contour in C´ ), with normalization constants that are bounded, and bounded away from 0.
Assuming the limit N Ñ 8, such a configuration has been referred to (see, for example, [36]) as
a soliton condensate, accompanied by the conjecture that soliton condensates achieve maximal
soliton density.
We will refer to such an N -soliton solution, or a sequence of such N -soliton solutions (indexed
by N ), as N -soliton condensate(s), and we will refer to the associated scattering data as N -
soliton condensate scattering data. Specifically, we consider the following assumptions.
Assumption 2.2 (N -soliton condensate scattering data). For a positive integer N , we generate
the scattering data as follows:
1. There are two positive constants a and b, which define a line segment η1 “ r´a ` ib, a ` ibs,
which we orient from right to left, and its complex conjugate η2 “ r´a ´ ib, a ´ ibs, which
we orient from left to right. We define
A “ a ` ib
to be the right-most endpoint of η1 .
2. There is a density function ρpzq which is analytic in a neighborhood of the line segment
η1 (and the neighborhood’s Schwarz reflection), satisfying
ρpzqis analytic in a neighborhood of each line segment η1 and η2 ,
ρpzq P R for z “ t ` ib, t P p´a, aq,
ˆ a
ρpt ` ibqdt “ 1,
´a

ρpzq “ ρpzq.

3. The eigenvalues tλj uN

j“1 on η1 are chosen to satisfy
ˆ ℜpλj q
2j ´ 1
ρpx ` ibqdx “ , ℑpλj q “ b, j “ 1, . . . , N.
´a 2N
See Figure 1 for an example with uniform density.
4. There is a “normalization constant sampling function” hpzq which is analytic in a neigh-
borhood of the line segment η1 (and the neighborhood’s Schwarz reflection), satisfying
hpzq “ hpzq.
5. For each positive integer N , the normalization constants tcj uN
j“1 are chosen to satisfy

cj “ hpλj q, j “ 1, . . . , N .

6. There are two contours, Γ1 P C` and Γ2 P C´ , which are Schwarz reflections of each other
(and so they are determined by Γ1 ). The contour Γ1 is a simple, closed, analytic curve
that encircles the interval η1 . See Figure 2.
We therefore have, for every positive integer N , our N -soliton scattering data
! )
N
DN “ r ” 0, tλj , cj uj“1 ,

where λj is defined in (2.15) and the normalzation constants cj are defined in (2.16). For each
positive integer N , this scattering data corresponds to an N -soliton condensate solution to the
NLS equation, which we will denote by
ψN px, tq “ ψpx, t; DN q.
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 7

λj

Figure 1: Poles λj and their conjugates

`
Γ1 `
´ λj ´
´ ´
` `
R
` `
´ ´
´ ´
` `
Γ2

Figure 2: Example of contour Γ1 , Γ2

In what follows, unless otherwise specified, we will assume that the scattering data under con-
sideration is a sequence of N -soliton condensate scattering data, with N large and increasing
towards 8. We will refer to this collection of assumptions by saying, “assuming a sequence
of N -soliton condensate scattering data”, or “under the N -soliton condensate scattering data
assumption”.

2.4 Riemann-Hilbert problem for ψN px, tq

Given the scattering data DN , the Riemann-Hilbert problem 2.1 simplifies significantly, since
rpzq ” 0 for all real z. Indeed, since the jump relation across the real axis (2.4) becomes
the trivial relation M` pzq “ M´ pzq for all real z, Morera’s Theorem tells us that M pzq is
then analytic across the real axis, and hence M pzq is meromorphic in C. We will make one
standard transformation, from M pzq to a new matrix Apzq, following (for example) [17]. We
use the contours Γ1 and Γ2 which are provided to us under the N -soliton condensate scattering
assumption, Γ1 encircling the interval η1 and Γ2 its Schwarz reflection. Each of these contours
is taken to be oriented clockwise, as shown in Fig. 2. We then define Apzq as follows:
$ ˜ ¸
’ 1 0
M pzq , z inside Γ1
’
’ řN cj
e2iθpλj q 1
’
’
’
’ ´ j“1 z´λ j
’
’
’
’
’
& ˜ řN cj
¸
´2iθpλj q
Apzq “ 1 j“1 z´λj e
’
’ M pzq , z inside Γ2
’
’
’
’ 0 1
’
’
’
’
’
’
% M pzq, otherwise.

Now M pzq has simple poles at the eigenvalues λj and λj , but the definition above is chosen so
that Apzq is analytic inside the contours Γ1 and Γ2 . Indeed, the reader may check from the
residue conditions (2.5) and (2.6) that (for example) the pole of M11 pzq at λk is cancelled since

ck M12 pzqe2iθpλk q
A11 pzq “ M11 pzq ´ ` ⟨analytic at λk ⟩ ,
z ´ λk
and the first two terms have the same exact residue at λk . Similar arguments show that the
matrix Apzq is analytic inside Γ1 and inside Γ2 , and it is also obviously analytic for z outside
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 8

the two contours. Finally, A clearly has boundary values as z approaches the contours Γ1 and
Γ2 (with different boundary values as z approaches each contour from inside or outside).
Notation: boundary values for z on a contour: Following standard convention, for z
in the contour Γ1 (or Γ2 ), we will let A` pzq denote the boundary value of A obtained by letting
z 1 approach z from outside Γ1 (or outside Γ2 ):

A` pzq “ lim Apz 1 q, z P Γj ,

z 1 Ñz
z1 P exterior of Γj

and we will let A´ pzq denote the boundary value from inside Γ1 (or Γ2 ). (Note that this is
consistent with the usual convention that the ` side of a contour is on the left of the contour
as one traverses it according to its orientation).
The new matrix-valued function Apzq satisfies a Riemann-Hilbert problem, which we will
refer to as the N -soliton condensate Riemann-Hilbert problem.
For this Riemann-Hilbert problem, the poles are accumulating on the segments η1 , η2 which
we parametrize for future convenience:
! ˇ ) ! ˇ )
η1 “ z P C ˇ z “ a ´ 2at ` ib, t P p0, 1q , η2 “ z P C ˇ z “ ´a ` 2ta ´ ib, t P p0, 1q .
ˇ ˇ

In particular, η1 is the horizontal segment from A to ´A, and η2 is the one from ´A to A.
Under the N -soliton scattering assumption, we have the following Riemann-Hilbert problem.
RHP 2.3 (N -soliton Condensate Riemann-Hilbert Problem). Find a 2 ˆ 2 matrix function A “
Apzq “ Apz; x, t, N q such that
• A is analytic on CztΓ1 Y Γ2 u
• A` pzq “ A´ pzqJA pzq where the jump matrix JA pzq is given by
$˜ ¸
’ 1 0
’
’ řN cj 2iθpλj q , z P Γ1
j“1 z´λj e 1
’
’
’
&
JA pzq “ (2.21)
’ ˜ řN cj
¸
´2iθpλj q
’
’
’ 1 ´ j“1 z´λj e
’
’
% , z P Γ2
0 1
ˆ ´8 ˙ ˆ ˙
1 ´ |ψN ps, tq|2 ds ψ px, tq 1
• Apzq “ I ` x ´8 N 2 `O , as z Ñ 8.
2iz ψN px, tq x
|ψ N ps, tq| ds z2
Note that the above z Ñ 8 asymptotic condition describes exactly how to extract the
N -soliton condensate solution ψN px, tq from the solution A of the Riemann-Hilbert problem.
Usually the asymptotic normalization condition is described in the weak form A “ I ` O z1 ,
` ˘

but in this case, since the contours Γ1 and Γ2 are bounded, A possesses a complete asymptotic
expansion in powers of z1 .

2.5 Riemann-Hilbert problem for the soliton gas

The N -soliton condensate Riemann-Hilbert problem behaves singularly as the number of eigen-
values grows to infinity, because the sums appearing in the jump relationships (2.21) do not
converge as N Ñ 8. However, their asymptotic behavior can be determined:
N ˆ
cj e2iθpλj q hpλqρpλqe2iθpλq dλ
ˆ ˙
ÿ 1
“ ´N `O ,
j“1
z ´ λ j η1 z ´ λ N
N ˆ ˆ ˙
ÿ cj hpλqρpλqe´2iθpλq ds 1
e´2iθpλj q “ N `O ,
j“1 z ´ λj η2 z ´ λ N
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 9

where we recall that the contour η1 is oriented ` ˘from right to left, while η2 is oriented from left
to right, and in both cases the error term O N1 is uniform for all z in the respective contour Γ1
or Γ2 . Indeed, under the N -soliton condensate assumption, a complete asymptotic expansion
for this sum can be calculated, because the eigenvalues tλj uN j“1 have been selected to arrive at a
uniform mid-point approximation. The next term in the expansion can be computed explicitly,
we perform such computation in Appendix A.
If we replace the finite sum by its leading order asymptotic behavior, we arrive at a Riemann-
Hilbert problem which still contains the large parameter N . It is this Riemann-Hilbert problem
which we study as N Ñ 8. The associated solution to the NLS equation, ψSG “ ψSG px, tq “
ψSG px, t; N q, is the soliton gas (or soliton condensate) solution.
RHP 2.4 (Soliton gas condensate Riemann-Hilbert problem). Find a 2 ˆ 2 matrix function
à “ Ãpzq “ Ãpz; x, t, N q such that
• Ã is analytic on CztΓ1 Y Γ2 u
• Ã` pzq “ ô pzqJà pzq where the jump matrix Jà pzq is given by
$˜ ¸
1 0
´ hpλqρpλqe2iθpλq dλ
’
’
’ , z P Γ1
& N η1 1
’
’
’ λ´z
JÃ pzq “ (2.24)
’
’ ˜ ´ hpλqρpλqe´2iθpλq ds ¸
’
’ 1 N η2 λ´z
’
’
% , z P Γ2
0 1
ˆ ´8 ˙ ˆ ˙
1 ´ |ψSG ps, t; N q|2 ds ψ px, t; N q 1
• Ãpzq “ I ` x ´ 8 SG 2 `O , as z Ñ 8.
2iz ψSG px, t; N q x
|ψSG ps, tq; N | ds z2
This Riemann-Hilbert problem arises formally, by replacing the jump matrices (2.21) with
(2.24). Since it is not related to Riemann-Hilbert problem 2.3 by explicit transformations, the
existence and uniqueness of a solution must be established independently. Note that this state-
ment extends to the soliton gas condensate solution ψSG px, t; N q. Uniqueness for the Riemann-
Hilbert problem 2.4 (regardless of whether or not the solution is known to exist) is completely
straightforward since detJÃ ” 1. See, for example, the first paragraph of the proof of Theorem
3.1 on P. 1503 of [9]. Regarding existence, the crucial fact is that the jump matrices obey the
symmetry
ˆ ˙ ˆ ˙
0 1 0 1
JÃ pzq “ JÃ pzq .
1 0 1 0

This permits the direct application of an existence theory via the connection between Riemann-
Hilbert problems and Fredholm singular integral operators (see Theorems 9.1, 9.2, and 9.3
in [42], or the proof of Theorem 5.3 in [10]). Thus we state without further discussion the
following result:
Theorem 2.5. Under the N -soliton condensate scattering data assumption, the Riemann-
Hilbert problem 2.4 possess a unique solution, and determines ψSG px, t; N q, a solution of the
NLS equation.

2.6 Statement of the results

In Sections 3 - 6 we determine the asymptotic behavior of the solution à of the Riemann-Hilbert
problem 2.4. The main application of this work is to determine the asymptotic behavior of the
soliton gas condensate ψSG px, t; N q, which is described in the following theorem.
Theorem 2.6. Under the N -soliton condensate scattering data assumption, for all px, tq in
a compact set K, there is N0 so that for all N ą N0 , ψSG px, t; N q satisfies the asymptotic
description (7.4). Specifically, for |ψSG |, we have
2 N -SOLITON SOLUTIONS, SOLITON GASSES, AND SOLITON GAS CONDENSATES 10

´ ¯´ ¯
Θ ´ ζpx,N q 1´τ
Θ ´ ζpx,N q τ ´1
˘˘2
2 ;τ 2 ;τ
` ` τ `1
Θ 1 ` ` ˆ ˙
2 2 2 ;τ 2π 2π 1
|ψSG px, t; N q| “ c ´ ¯ `O ,
Θpτ ; τ qΘp0; τ q Θ2 ´ ζpx,N q log N
2π ; τ

´ ¯ (2.26)
1
where the error term O is uniform for all px, tq in the compact set K. Here Θpz; τ q is
log N
the Jacobi Theta-3 function as defined in [12]. The constants τ, c and ζpx, N q are given by
˜ˆ ¸´1 ˆ ´A
´A
1 1 1
c“ ds , τ “ 2c ds ,
2 A R` psq ´A Rpsq
˜ˆ ˆ ´A ¸
´A
τ ∆pxq ´ 2 lnpN q s2 s2
ζpx, N q “ ´ ds ´ τ ds ,
2cπ ´A Rpsq A R` psq
1 1 1 1
where Rpzq “ pz ´ Aq 2 pz ` Aq 2 pz ´ Aq 2 pz ` Aq 2 , the brunch-cuts are the canonical ones, and
the integration is performed on the straight line connecting the endpoints. The function ∆pxq is
defined in terms of the functions ρpzq, hpzq as follows
˜ˆ ¸´1 ˜ ˆ ˆ A ¸
´A ´A
1 log p2πhpsqρpsqq ` 2isx log p2πhpsqρpsqq ´ 2isx
∆pxq “ ds ´ ds ` ds .
´A Rpsq A R` psq ´A R` psq

Remark 2.2. It is interesting to observe that the leading order asymptotic behavior of the
modulus of the soliton gas condensate solution ψSG px, t; N q (2.26) does not depend on time.
The explanation for this fact is that our assumption on the scattering data, the poles are
accumulating on symmetric contours. Essentially, the soliton gas condensate is in a sort of
equilibrium, and can be interpreted as an elliptic wave with zero velocity. This could also be
interpreted as another form of soliton-shielding, similar to [2, 3]. In particular, we want to
emphasize that the solution (2.26) and the solution [3, eq. (4.44)] are similar, indeed they both
arise as solutions of similar RHPs, the two main differences are that in our case the contours
where the poles are accumulating are horizontal, and the norming constants cj are bounded
away from zero, while in [3] the contours are vertical and the constants are uniformly OpN ´1 q.
In appendix B we consider the non-symmetric setting where the line segments η1 and η2
are shifted so that they are not symmetric with respect to the origin. An argument involving
a translation of the spectral plane reveals a new explicit dependence on time. Finally, it is
worth noticing that the solution depends just on the product of hpzq, ρpzq and not on each
function separately. Fig. 3 shows plots of the leading order asymptotic of the solution of the
NLS equation for different values of N .
Proposition 2.7. Let Apzq, Apzq
r be the solutions of RHP 2.3 and RHP 2.4 respectively. Under
the N -soliton condensate scattering data assumption, there is N0 so that for all N ą N0 , we
have ˆ ˆ ˙˙
1
Apzq “ I2 ` O Apzq
r .
N p1 ` |z|q
´ ¯
where the error term, O N p1`|z|q
1
is uniform in the entire complex plane.
The proof of this proposition is presented in Appendix A.3, and relies on the complete
asymptotic description of the solution à to the Riemann-Hilbert problem 2.4, which as men-
tioned above is developed in Sections 3 - 6.
It is an immediate consequence of this uniform approximation that the N -soliton solution is
extremely well approximated by the soliton gas condensate solution.
Corollary 2.8. Under the N -soliton condensate scattering data assumption, for all px, tq in a
compact set, there is N0 and a constant c0 so that for all N ą N0 , we have
c0
|ψN px, tq ´ ψSG px, t; N q| ď .
N
3 DRESSING TRANSFORMATIONS 11

In the next four sections, we develop the asymptotic analysis of the solution of the Riemann-
Hilbert problem 2.4, to establish, in Section 7, the asymptotic behavior of the soliton gas
condensate solution ψSG px, t; N q. We prove Proposition 2.7, Corollary 2.8 and other technical
results in Appendix A.

|ψ(x,t)|, N=2000 |ψ(x,t)|, N=1000

0.30 0.30

0.25 0.25

0.20 0.20

0.15 0.15

0.10 0.10

0.05 0.05

-4 -2 2 4
x -4 -2 2 4
x

Figure 3: Solution to the NLS equation (1.1) in assumptions 2.2. Here A “ 1 ` i, and N is specified
in the plots

3 Dressing transformations
In this section, we make two transformations, starting with RHP 2.4, which yield a new Riemann-
Hilbert problem. This new Riemann-Hilbert problem is not yet in the class of small-norm
problems, but is a stepping-stone toward that objective. Recall that we have already turned the
residue conditions of M at λj into suitable jump conditions for the function A r across Γ1 Y Γ2
(see RHP 2.4).
Following the procedure highlighted in Proposition 2.3 and Lemma 2.4 of [17], we move the
jump condition from Γ1 , Γ2 to η1 , η2 . To accomplish this, we define the function
$ ˜ ¸
1 0
´
’
Apzq , z inside Γ1
’
’ r
N η1 hpλqρpλq 2iθpλq
’
λ´z e dλ 1
’
’
’
’
’
’
’
´ hpλqρpλq ´2iθpλq ¸
’
& ˜
Bpzq “ 1 N λ´z e dλ
’ Apzq 0
’
’ r η2
, z inside Γ2
’
’ 1
’
’
’
’
’
’
otherwise
’
% Apzq,
r

Γ1
η1 ´ ´
` `
R

η2 ` `
´ ´
Γ2

Figure 4: Contours Γ1 , Γ2 , η1 , η2
3 DRESSING TRANSFORMATIONS 12

The reader may verify that Bpzq has no jumps across Γ1 , Γ2 , but it has jumps across η1 , η2 .
Specifically, the matrix Bpzq solves the following RHP.
RHP 3.1. We look for a square matrix function Bpzq such that:
• Bpzq is analytic in Cz pη1 Y η2 q. It possesses smooth boundary values B` and B´ as z
approaches either contour from the ` or ´ side.
• The boundary values satisfy the jump relation Bpzq` “ Bpzq´ JB pzq across the contour
η1 , η2 , where
$˜ ¸
’ 1 0
’
’
2iθpzq
, z P η1
2πiN hpzqρpzqe 1
’
’
’
&
JB pzq “ ˜ ¸
’
´2iθpzq
1 2πiN hpzqρpzqe
’
’
’
’
’ , z P η2
0 1
%

• Bpzq “ I2 ` Op1{zq as z Ñ 8.
The next classical step in the analysis of the RHP is to modify the previous one to control the
N dependence of the jumps, in order to derive the N Ñ 8 asymptotic behavior of ψSG px, t; N q
[17, 18]. First, we introduce a vertical line segment γ which connects the points ´A and ´A, as
shown in Figure 5, and then we introduce two new functions gpzq and ypzq, both being analytic
in Cztγ Y η1 Y η2 u. As usual, these functions achieve their boundary values g˘ and y˘ in the
sense of smooth functions on the contours γ, η1 and η2 . Then, we define the function Cpzq as
follows

Cpzq “ eplnpN qgp8q`yp8qqσ3 Bpzqe´plnpN qgpzq`ypzqqσ3 , (3.1)

where gp8q, yp8q are the values of gpzq and ypzq at 8, which we assume are finite (this
assumption will be verified later, since these two quantities will be explicitly constructed). The
function Cpzq satisfies the following RHP.
RHP 3.2. We look for a function Cpzq P MatpC, 2q such that:
• Cpzq is analytic on Cztη1 Y η2 Y γu, and achieves its boundary values C˘ on η1 and η2 in
the sense of smooth functions.
• The boundary values satisfy C` pzq “ C´ pzqJC pzq, where the jump matrix JC pzq is given
as follows:
$¨ ` ˘ ˛
lnpN q g´ pzq´g` pzq `y´ pzq´y` pzq
’
’
’ e 0
’
’ ˚ ‹
’
’
’ ˚ ‹ , z P η1
’
’ ˝ ` ˘ ` ˘ ‚
2iθpzq ´ lnpN q g´ pzq`g` pzq ´py´ pzq`y` pzqq lnpN q g` pzq´g´ pzq `y` pzq´y´ pzq
’
’ 2πiN hpzqρpzqe e e
’
’
’
’
’
’
’
’
’ ` ˘ ` ˘
’ ¨ lnpN q g´ pzq´g` pzq `py´ pzq´y` pzqq
’ ˛
& e 2πiN hpzqρpzqe´2iθpzq elnpN q g´ pzq`g` pzq `y` pzq`y´ pzq
JC pzq “ ˚ ‹
’
’ ˚ ‹ , z P η2
’
’ ˝ ` ˘ ‚
’ lnpN q g` pzq´g´ pzq `y` pzq´y´ pzq
’
’
’
’ 0 e
’
’
’
’
’
’ ¨ ` ˘ ˛
’
lnpN q g´ pzq´g` pzq `py´ pzq´y` pzqq
’
e 0
’
’
’ ‚,
’
’ ˝ ` ˘ zPγ
elnpN q g` pzq´g´ pzq `y` pzq´y´ pzq
’
% 0

• Cpzq “ I2 ` Op1{zq, as z Ñ 8.
3 DRESSING TRANSFORMATIONS 13

´ η1 ´
` `
γ
` η2 `
´ ´

Figure 5: Non-analyticity contour for the function Cpzq

As mentioned, we introduce the function gpzq to control the N dependence in the jump
matrices. We require that gpzq satisfies the following RHP.
RHP 3.3. We look for a scalar function gpzq such that:
• gpzq is analytic on Cztη1 Y η2 Y γu
• gpzq satisfies the conditions:

g` pzq ` g´ pzq “ 1, z P η1 (3.2)

g` pzq ` g´ pzq “ ´1, z P η2
2
g` pzq ´ g´ pzq “ ´ , z P γ (3.3)
τ

• gpzq “ gp8q ` Op1{zq, as z Ñ 8,

where τ, gp8q are two constants independent of z still to be determined.
Moreover, we wish to make the off-diagonal entries of the jump matrix JC constants (for z
on the contours η1 and η2 ). This is accomplished by requiring that the function ypzq satisfies
the following RHP.
RHP 3.4. We look for a scalar function ypzq such that:
• ypzq is analytic on Cztη1 Y η2 Y γu
• ypzq satisfies the conditions:

y` pzq ` y´ pzq “ lnp2πhpzqρpzqq ` 2iθpzq, z P η1

y` pzq ` y´ pzq “ ´ lnp2πhpzqρpzqq ` 2iθpzq, z P η2 (3.4)
y` pzq ´ y´ pzq “ ∆, zPγ

• ypzq “ yp8q ` Op1{zq, as z Ñ 8,

where yp8q is a constant still to be determined.
Assuming that such gpzq, ypzq exist, then the jump matrix JC pzq reduces into:
$¨ ` ˘ ˛
lnpN q g´ pzq´g` pzq `y´ pzq´y` pzq
’
’
’ e 0
’
’ ˚ ‹
’
’˚
’ ‹ , z P η1
’
’ ˝ ` ˘ ‚
’ lnpN q g` pzq´g´ pzq `y` pzq´y´ pzq
’
’
’
’ i e
’
’
’
’
’
’ ` ˘
& ¨ lnpN q g´ pzq´g` pzq `y´ pzq´y` pzq
’ ˛
JC pzq “ ˚ e i (3.5)
‹
’
’
’
’
˚
˝
‹
‚ , z P η2
’ ` ˘
’ lnpN q g` pzq´g´ pzq `y` pzq´y´ pzq
’
’
’
’ 0 e
’
’
’
’
’
’ ˜ 2 ¸
’
e τ lnpN q´∆ 0
’
’
’
’
’ 2 , zPγ
% 0 e´ τ lnpN q`∆
3 DRESSING TRANSFORMATIONS 14

3.1 The gpzq function

To solve RHP 3.3, we proceed as follows. First, consider the RHP for g 1 pzq. Directly from the
RHP for gpzq, we get the following.
RHP 3.5. We look for a scalar function g 1 pzq such that
• g 1 pzq is analytic on Cztη1 Y η2 u
• g 1 pzq satisfies the conditions:
1 1
g` pzq ` g´ pzq “ 0, z P η1
1 1
g` pzq ` g´ pzq “ 0, z P η2

• g 1 pzq “ Opz ´2 q, as z Ñ 8.
The previous RHP can be solved explicitly as
2c 1 1 1 1
g 1 pzq “ ´ , Rpzq “ pz ´ Aq 2 pz ` Aq 2 pz ` Aq 2 pz ´ Aq 2 ,
τ Rpzq

for some c, τ P C, where for each factor p¨q1{2 we use the principal branch, so that Rpzq is
analytic in Cztη1 Y η2 }, and is normalized such that

Rpzq „ z 2 , z Ñ 8.

Define the function ˆ z

2c 1 1
gpzq “ ´ ds ` , (3.6)
τ A Rpsq 2
then it is easy to check that g satisfies the following conditions:

g` pzq ` g´ pzq “ 1, z P η1
ˆ
4c 1
g` pzq ` g´ pzq “ ´ ds ` 1, z P η2
τ γ Rpsq
ˆ
4c 1
g` pzq ´ g´ pzq “ ´ ds, z P γ
τ η1 R` psq

Imposing
ˆˆ ˙´1 ˆ
1 1 1
c“ ds , τ “ 2c ds (3.7)
2 η1 R` psq γ Rpsq
gpzq in (3.6) solves RHP 3.3.
For later usage, we list a series of properties that the function R and its integrals satisfy.
Since the proof is technical, we defer it to the appendix A.
Proposition 3.6. The function R defined as
1 1 1 1
Rpzq “ pz ´ Aq 2 pz ` Aq 2 pz ` Aq 2 pz ´ Aq 2 ,

where for each factor p¨q1{2 we use the principle branch, satisfies the following properties
2 2
1. Rpzq “ z 2 ´ A `A ` O z12 , as z Ñ 8,
` ˘
2
2. Rpzq “ Rpzq,
3. Rp´zq “ Rpzq,
´
4. η1 R`1pzq dz P R´ ,
´ 1
5. γ Rpzq dz P iR´ .

Remark 3.1. As a corollary of Proposition 3.6 we deduce that c P R´ and τ P iR` , for c, τ
defined in (3.7). So 2c
τ P iR` .
3 DRESSING TRANSFORMATIONS 15

3.2 The ypzq function

We now find the explicit solution to RHP 3.4.
Proposition 3.7. The solution y : C Ñ C to the Riemann-Hilbert Problem 3.4 is given by
ˆˆ ˆ ˆ ˙
Rpzq log p2πhpsqρpsqq `2iθps, x, tq log p2πhpsqρpsqq ´2iθps, x, tq ∆px, tq
ypzq “ ds ´ ds ` ds ,
2πi η1 R` psqps ´ zq η2 R` psqps ´ zq γ Rpsqps ´ zq
(3.8)
with ∆px, tq defined as
ˆˆ ˙´1 ˆ ˆ ˆ ˙
1 log p2πhpsqρpsqq ` 2iθps, x, tq logp2πhpsqρpsqq ´ 2iθps, x, tq
∆px, tq “ ds ´ ds ` ds .
γ Rpsq η1 R` psq η2 R` psq
(3.9)

Furthermore, ∆px, tq is independent of t, so that

ˆˆ ˙´1 ˆ ˆ ˆ ˙
1 log p2πhpsqρpsqq ` 2izx log p2πhpsqρpsqq ´ 2ixz
∆pxq “ ds ´ ds ` ds .
γ Rpsq η1 R` psq η2 R` psq
(3.10)
Lastly, yp8q “ limzÑ8 ypzq is finite.

Proof. The reader may verify using the Cauchy-Plemelj-Sokhotski formula that y defined in
(3.8) satisfies conditions (3.4).
It is straightforward to verify, using the asymptotic behavior Rpzq “ z 2 1 ` O z12 as
` ` ˘˘

z Ñ 8, that
ˆ ˆ
z2 ´ 1 log p2πhpsqρpsqq ` 2iθps, x, tq 1 log p2πhpsqρpsqq ´ 2iθpx, t, sq
ypzq “ ´ ds ` ds
2πi z η1 R` psq z η2 R` psq
ˆ
1 ∆px, tq ¯
´ ds ` Opz ´2 q
z γ Rpsq

and (3.9) guarantees that ypzq Ñ y8 as z Ñ 8. Finally, we notice that ∆px, tq is independent
of time. Indeed, the coefficient of t in ∆px, tq is given by
˜ˆ ˆ ¸
s2 s2
´4i ds ` ds
η1 R` psq η2 R` psq

which can be shown to be zero by residue calculation and by leveraging the symmetry of the
endpoints of η1 . Therefore, (3.9) reads
ˆˆ ˙´1 ˆ ˆ ˆ ˙
1 log p2πhpsqρpsqq ` 2ixz logp2πhpsqρpsqq ´ 2ixz
∆pxq “ ds ´ ds ` ds .
γ Rpsq η1 R` psq η2 R` psq

For later use, we prove the following proposition.

Proposition 3.8. Define ∆pxq as in (3.10), then ∆pxq P iR.
´ 1
Proof. Applying Proposition 3.6, we know that γ Rpsq ds P iR; so to conclude it remains to
show that
ˆ ˆ ˆ ˙
log p2πhpsqρpsqq ` 2ixz log p2πhpsqρpsqq ´ 2ixz
´ ds ` ds P R . (3.11)
η1 R` psq η2 R` psq

r ´ ib, the second integral can be rewritten as

Setting z “ x
4 OPENING LENSES 16

ˆ a ˆ a
x ´ ibqρpr
log p2πhpr x ´ ibqq ´ 2ixpr
x ´ ibq log p2πhpr x ` ibqρprx ` ibqq ` 2ixpr
x ` ibq
dr
x“ dr
x
´a R` pr
x ´ ibq ´a R ` x
pr ` ibq
ˆ A
s“r
x`ib log p2πhpsqρpsqq ` 2ixs
“ ds
´A R` psq
ˆ
log p2πhpsqρpsqq ` 2ixs
“´ ds ,
η1 R` psq

from which (3.11) follows.

4 Opening lenses
We next introduce two further transformations. The first involves identifying a contour defor-
mation, so that the jump matrices (on the new contours) have entries that are bounded, but
exhibit rapid oscillations. The second transformation is referred to as “opening lenses”, whose
objective is to deform the Riemann-Hilbert problem so that rapidly oscillatory terms become
exponentially decreasing. As with the transformations from the previous section, these further
transformations are invertible, so that knowledge of the solution of this new Riemann-Hilbert
problem will be equivalent to knowledge of the original Riemann-Hilbert problem 3.2.
To achieve the first transformation, we aim to find two contours η3 , η4 such that the analytic
extension of g` pzq´g´ pzq is purely imaginary for z P η3 , η4 . The existence of these two contours
is a result of the following Lemma.
Lemma 4.1. Let the function gpzq be the unique solution to RHP 3.3. For z P η1 , we have the
representation ˆ
4c z 1
g` pzq ´ g´ pzq “ ds.
τ A R´ psq
The quantity g` ´ g´ has an analytic extension to C` z pη1 Y γq. We define upzq to be that
analytic extension, ˆ
4c z 1
upzq “ ds, z P C` z pη1 Y γq , (4.1)
τ A Rpsq
where the path of integration is chosen to avoid η1 Y η2 Y γ, so that
u´ pzq “ g` pzq ´ g´ pzq, z P η1 .
There exists a contour η3 connecting A with ´A such that
upzq P iR, z P η3 .
The contour η3 lies above η1 and we take it to be oriented from A to ´A (like η1 ).
Similarly, for z P η2 , we have
ˆ ˆ
4c ´A 1 4c z 1
g` pzq ´ g´ pzq “ ds ` ds.
τ A R´ psq τ ´A R´ psq
As with (4.1), we define
ˆ z
4c 1
ũpzq “ ds, z P C´ z pη2 Y γq , (4.2)
τ A Rpsq
where again the path is chosen to avoid η1 Y η2 Y γ. Note that
ˆ ˆ
4c ´A 1 4c z 1
ũ´ pzq “ ds ` ds, z P η2 ,
τ A R´ psq τ ´A R´ psq
4 OPENING LENSES 17

so that
ũ´ pzq “ g` pzq ´ g´ pzq, z P η2 .
There exists a contour η4 connecting ´A with A such that

ũpzq P iR, z P η4 .

The contour η4 lies below η2 and we take it to be oriented from ´A to A (like η2 ).

Furthermore, the following hold
ˆ ´A ˆ ´A ˆ 8
2c 1 2c 1 1 2c 1 1
ds “ 1, ds “ , ´ ds “ ´ 1 .
τ ´A Rpsq τ A R` psq τ τ A Rpsq τ

Finally, define Ω` `
3 to be the open region enclosed by η1 , η3 and Ω2 the open region enclosed by
η2 , η4 . Then we have
# #
ℜpupzqq ă 0 z P Ω` upzqq ą 0 z P Ω`
ℜpr
3
, 2
(4.3)
ℜpupzqq ą 0 z P C` zΩ` 3 ℜpr
u pzqq ă 0 z P C `
´ zΩ2

where Ω` `
3 and Ω2 denote the closure of those sets.

The contours η3 , η4 , γ are displayed in figure 6. Since the proof is technical, we defer it to
the appendix A.
Remark 4.1. For future reference, note that
2
u
rpzq “ ˘ ´ up´zq , z PR
τ
where is the rectangle formed by the real axis and the points A and ´A.
Define the function yr as follows:
$ ` ˘
& ´ypzq ` log `2πhpzqρpzq˘ ` 2iθpzq,
’
’ z P Ω`
3
yrpzq “ ´ypzq ´ log 2πhpzqρpzq ` 2iθpzq, z P Ω`
2
’
z elsewhere
’
% ypzq,

which is analytic in Cz pη3 Y γ Y η4 q and satisfies the following conditions:

` ˘
yr` pzq ` yr´ pzq “ log 2πhpzqρpzq ` 2iθpzq, z P η3
` ˘
yr` pzq ` yr´ pzq “ ´ log 2πhpzqρpzq ` 2iθpzq, z P η4
yr` pzq ´ yr´ pzq “ ∆pxq, zPγ

Moreover, define the matrix function D as follows

#
Cpzq, z R Ω` `
3 Y Ω2
Dpzq “
CpzqJC pzq, z P Ω` `
3 Y Ω2 .

Here the matrix JC in (3.5), which was originally defined only on the contours η1 , η2 , and γ,
possesses an analytic extension to the domains Ω`
3 and Ω2 , using y
`
r, u, and ũ where needed.
Applying Lemma 4.1 and using the definition of D, we can check that D satisfies the following
RHP.
RHP 4.2. Find a matrix function D P Matp2, Cq holomorphic in Cztη3 Y η4 Y γu such that
4 OPENING LENSES 18

Ω` η3
3
η1

η2
Ω`
2 η4

Figure 6: Jump contours for RHP 4.2 and for yrpzq

1. D` pzq “ D´ pzqJD pzq

$¨ ˛
’
’ e´ lnpN qupzq`ry´ pzq´ry` pzq 0
’
’
‚, z P η3
’
’ ˚ ‹
’
’ ˝
lnpN y y
’ qupzq`r pzq´r pzq
’
’
’
’ i e ` ´

’
’
’
’
’
’
& ¨e´ lnpN qũpzq`ry´ pzq´ry` pzq
’
i
˛
JD pzq “ ˚
‚, z P η4
’˝ ‹
’
’
’ lnpN qũpzq`r
y` pzq´ry´ pzq
’
’
’
’ 0 e
’
’
’
’
’
’ ˜ ¸
’ e τ2 lnpN q´∆
’
0
’
’
’
’ 2 , zPγ
% 0 e´ τ lnpN q`∆

2. Dpzq “ I2 ` Op1{zq, as z Ñ 8.
3. upzq P iR for z P η3 , and ũpzq P iR for z P η4 .

4.1 Jump matrix factorization

On η3 , the jump matrix JD factors as follows
¨ ´ lnpN qupzq`ry´ pzq´ry` pzq ˛
e 0
JD pzq “ ˝ ‚
lnpN qupzq`r
y` pzq´r
y´ pzq
i e
ˆ ´ lnpN qupzq`r
y´ pzq´r
y` pzq
˙ˆ ˙ˆ ˙
1 ´ie 0 i 1 ´ielnpN qupzq`ry` pzq´ry´ pzq
“
0 1 i 0 0 1

and on η4 as

e´ lnpN qpũpzqq`ry´ pzq´ry` pzq

¨ ˛
i
JD pzq “ ˝ ‚
0 elnpN qpũpzqq`ry` pzq´ry´ pzq
ˆ ˙ˆ ˙ˆ ˙
1 0 0 i 1 0
“ .
´ielnpN qũpzq`ry` pzq´ry´ pzq 1 i 0 ´ie ´ lnpN qũpzq`r
y´ pzq´r
y` pzq
1
4 OPENING LENSES 19

Thanks to the conditions that the function yr satisfies on η3 and η4 , we can rewrite the jump for
D:
$˜ ¸˜ ¸˜ ¸
e2yr´ pzq e2yr` pzq
’ 1 ´ie´ lnpN qupzq 2πhpzqρpzqe 2iθpzq 0 i 1 ´ielnpN qupzq 2πhpzqρpzqe 2iθpzq
’
’
’ , z P η3
’
’
’ 0 1 i 0 0 1
’
’
’
’
’
’
’
’ ˜ ¸˜ ¸˜ ¸
& 1 0 0 i 1 0
JD pzq “ e´2yr´ pzq e´2yr` pzq , z P η4
’
’
’ ´ielnpN qũpzq 2πhpzqρpzqe ´2iθpzq 1 i 0 ´ie´ lnpN qũpzq 2πhpzqρpzqe ´2iθpzq 1
’
’
’
’
’
’
’ ˜ τ2 lnpN q´∆
’ ¸
’
’ e 0
’
’ 2 , zPγ
0 e´ τ lnpN q`∆
%

We define the matrices

˜ ¸ ˜ ¸
´ lnpN qupzq e2ypzq
r
lnpN qupzq e2ypzq
r
1 ´ie 1 ´ie
M´ pzq “ 2πhpzqρpzqe2iθpzq , M` pzq “ 2πhpzqρpzqe2iθpzq
0 1 0 1
˜ ¸ ˜ ¸
´
1 0 `
1 0
N pzq “ e´2ypzq , N pzq “ e´2ypzq .
´ie´ lnpN qũpzq 2πhpzqρpzqe ´ielnpN qũpzq 2πhpzqρpzqe
r r
´2iθpzq 1 ´2iθpzq 1
(4.6)
The off-diagonal entries of the matrices M are piecewise analytic functions with jumps across
˘

the contours η1 and η3 . Notice that because u changes signs across the contour η1 , we have
` `˘ ` ˘
M ´ pzq “ M´ ` pzq, z P η1 .
Analogously, the off-diagonal entries of N˘ are piecewise analytic, with jumps across the
contours η2 and η4 . Notice that because ũ changes signs across η2 , we have
` ´˘ ` ˘
N ´ pzq “ N` ` pzq, z P η2 .
We can rewrite the jump matrix JD for the matrix D as follows.

$ ˜ ¸
’ ´ 0 i
’ pM q´ pzq
’ pM` q` pzq, z P η3
i 0
’
’
’
’
’
’
’
’
’
’ ˜ ¸
’
& 0 i
JD pzq “ pN` q´ pzq pN´ q` pzq, z P η4
’
’
’ i 0
’
’
’
’˜
’
’ ¸
’ e τ2 lnpN q´∆
’
’ 0
’
’ 2 , zPγ
0 e´ τ lnpN q`∆
%

The next transformation is from Dpzq to Epzq, the so-called “opening of lenses” transfor-
mation. Our aim is to arrive at a new RHP for a function Epzq with constant jumps on the
contours η3 , η4 , γ and “small” jumps on the other 4 contours ω1 , ω2 , ω3 , ω4 that we construct
according to the following proposition.
Proposition 4.3. Fix minpℜpAq, ℑpAqq ą 3ε ą 0, let UA , UA , U´A , U´A be 4 disks of radius ε
centered at A, A, ´A, ´A respectively and define V “ CztUA Y UA Y U´A Y U´A u. Consider
the matrices M´ , N` (4.6), then there exist four contours ω1 , ω2 , ω3 , ω4 , a κ ą 0 and a N0 such
that for all N ą N0
ˆ ˙ ˆ ˙
` 1 0 ´κ ´ 1 0
N pzq “ `OpN q, z P pω1 Y ω2 qXV; M pzq “ `OpN ´κ q, z P pω3 Y ω4 qXV .
0 1 0 1
4 OPENING LENSES 20

` `
In particular, the contours ω1 , ω2 Ă C´ zΩ2 lie on opposing sides of η4 , and ω3 , ω4 Ă C` zΩ3 lie
on opposing sides of η3 . An example of these contours is shown in Fig. 7.

Proof. We focus on M´ , the proof in the other situation is analogous. First, we notice that in
view of the definition (4.6), we must show that there exist two contours ω3 , ω4 connecting A
with ´A, a κ ą 0 and a N0 such that for all N ą N0

e2rypzq
´ie´ lnpN qupzq “ OpN ´κ q , z P pω3 Y ω4 q X V .
2πhpzqρpzqe2iθpzq
`
Consider the set C` zpΩ3 Y γq. We notice that in these regions the quantities hpzq, ρpzq,
θpzq and yrpzq are all analytic and bounded. Moreover, (4.3) holds. (See Assumptions 2.2
and Proposition 3.7, and Lemma 4.1 to verify these properties.) Therefore, we can choose
`
ω3 , ω4 Ă C` zpΩ3 Y γq (it is convenient to pick ω3 , ω4 to be analytic arcs). Finally, since ℜpupzqq
is continuous on pω3 Y ω4 q X V, there exists κ ą 0 such that
min ℜpupzqq “ κ ,
zPω3 XV

so there exists N0 such that for all N ą N0

e2rypzq
´ie´ lnpN qupzq “ OpN ´κ q , z P ω3 X V ,
2πhpzqρpzqe2iθpzq
proving the result.

Given these 4 contours, we can define a new function Epzq as

$ `
’ Dpzq N pzq, ´1
’ z P Ω1
z P Ω`
’
’
’
’
’ Dpzq pN´ pzqq , 2
´1
`
z P Ω´
’
& Dpzq pN pzqq ,
’
’ 2
(4.7)
´1
Epzq “ Dpzq pM` pzqq , z P Ω`
3
’ ´1
Dpzq pM´ pzqq , z P Ω´
’
’
’
’ 3
’
’
’
’
’ Dpzq M´ pzq, z P Ω4
elsewhere
%
Dpzq,

where the sets Ω`

2 , Ω2 , Ω3 , Ω3 are as in figure 7 and M , M , N , N are defined in (4.6). The
´ ` ´ ` ´ ` ´

matrix Epzq solves the following RHP.

RHP 4.4. Find a matrix function Epzq P Matp2, Cq analytic in Cztη3 Yη4 Yγ Yω1 Yω2 Yω3 Yω4 u
such that
1. E` pzq “ E´ pzqJE pzq
$˜ ¸
’ 0 i
, z P η3 Y η4
’
’
’
’
’
’
’ i 0
’
’
’
’
’
’ ˜ 2 ¸
’
τ lnpN q´∆
& e 0
’
’
, zPγ
’
2
lnpN q`∆
JE pzq “ 0 e´ τ
’
’
’
’
’
’
’ ´
’ M pzq,
’
’
’ z P ω3 Y ω4
’
’
’
’
’
’
’
% N` pzq, z P ω1 Y ω2

2. Epzq “ I2 ` Op1{zq, as z Ñ 8.
5 GLOBAL PARAMETRIX 21

ω4
Ω4

Ω` η3
3
Ω´ η1
3
ω3
γ
ω2
Ω´
2 η2
Ω`
2 η4

Ω1
ω1
Figure 7: Contour for RHP 4.4

5 Global Parametrix
In this section, we define and solve the model problem that yields the leading order asymptotic
behavior for the Riemann-Hilbert problem 2.4 for z away from the four branch points. The
following RHP comes from the small norm approximation that we have done in the previous
section.
RHP 5.1. Find a matrix function Λpzq P Matp2, Cq holomorphic in Cztη3 Y η4 Y γu such that
1. Λ` pzq “ Λ´ pzqJΛ pzq
$˜ ¸
’ 0 i
, z P η3 Y η4
’
’
’
’
’
’ i 0
&
JΛ pzq “
’
’ ˜ 2 lnpN q
¸
’
’ e´∆` τ 0
, zPγ
’
’
’ 2 lnpN q
0 e∆´
% τ

2. Λpzq satisfies the asymptotic behavior: Λpzq “ I2 ` Op1{zq, as z Ñ 8

3. Λ has 41 -root singularities (each entry of Λ) at z “ A, ´A, A, ´A.
Remark 5.1. Note that in all previous RHPs, the solution was bounded at each branch point.
Here, local analysis shows that we must have 1{4-root singularities at each branch point.
Following the procedure in [6], we factor the matrix Λpzq as

Λpzq “ ξpzqeiν0 σ3 Ξpzqeiνpzqσ3 , (5.1)

where ξ, ν are two scalar functions, Ξpzq P Matp2, Cq, and ν0 is a constant to be determined.
Specifically, ξ is the solution to the following RHP.
RHP 5.2. Find a scalar function ξpzq holomorphic in Cztη3 Y η4 u such that
1. ξ` pzq “ i ξ´ pzq, z P η3 Y η4
5 GLOBAL PARAMETRIX 22

2. ξpzq “ 1 ` Op1{zq, as z Ñ 8
3. ξ has 14 -root singularities at z “ A, ´A.
Directly by applying the Sokhotski-Plemelj formula we get
ˆ ˙1{4 ˆ ˙1{4
z`A z´A
ξpzq “ .
z´A z`A

Moreover, the function νpzq satisfies the following RHP.

RHP 5.3. Find a scalar function νpzq holomorphic in Cztη3 Y η4 Y γu such that
1. ν` pzq ` ν´ pzq “ 0, z P η3 Y η4
´ ¯
2. ν` pzq ´ ν´ pzq “ i ∆ ´ 2 lnpN
τ
q
, z P γ.
1 1 1 1
We can solve the previous RHP explicitly. Let Spzq “ pz ´ Aq 2 pz ´ Aq 2 pz ` Aq 2 pz ` Aq 2
and choose the branch-cuts in such a way that S is analytic in Cztη3 Y η4 u, and Spzq „ z 2 at
infinity. Consider νpzq “ Spzqϕpzq, where ϕ is analytic everywhere except on γ. Directly from
the jump conditions that the functions S and ν satisfy, we can deduce the jump condition for ϕ
´ ¯
i ∆ ´ 2 lnpN
τ
q

ϕ´ pzq ´ ϕ` pzq “ , zPγ

Spzq
which implies that
ˆ lnpN q
1 1 ∆´2 τ
ϕpzq “ ds.
2π γ Spsq s´z
From there, we can deduce the solution to the RHP for ν
ˆ lnpN q
Spzq 1 ∆´2 τ
νpzq “ ds.
2π γ Spsq s´z

From the last expression we determine the large-z asymptotics for the function νpzq

νpzq “ a1 z ` a0 ` Op1{zq, zÑ8

ˆ ˆ
1 ∆ ´ 2 lnpN q
τ ∆ ´ 2 lnpN q 1 ∆ ´ 2 lnpN q
a1 “ ´ τ
ds “ ´ , a0 “ ´ τ
s ds. (5.2)
2π γ Spsq 4cπ 2π γ Spsq
Considering all the above, one can deduce the RHP for the function
1 ´iν0 σ3
Ξpzq “ e Λpzqe´iνpzqσ3 .
ξpzq

RHP 5.4. Find a square matrix function Ξpzq analytic in Cztη3 Y η4 u such that
ˆ ˙
0 1
1. Ξ` pzq “ Ξ´ pzq , z P η3 Y η4
1 0
2. Ξpzq “ rI2 ` Op1{zqs e´ipa1 z`a0 `ν0 qσ3 , zÑ8
3. Ξ has 12 -root singularities (each entry of Ξ) at z “ A, ´A.
The goal is to find an explicit representation for Ξ, which yields an explicit representation
for Λ, since the functions ξ and ν are known. Consider now the double-sheeted Riemann surface
R which consists of two copies of the complex plane, SI and SII , which are glued together along
the two branch cuts η3 and η4 . Every point z P C has an image in SI via some projection map
PI pzq, and it has an image in SII via some projection map PII pzq. Conversely, any point p P SI
is the image of some point z P C through the map PI pzq, and there exists another point p1 P SII
5 GLOBAL PARAMETRIX 23

SI

´A α A
PI pzq C
β
´A
A
´A A

´A
SII A

´A A PII pzq
β
´A
A

Figure 8: 2-sheeted Riemann Surface R

which is the image of the same point z in C through PII pzq. We introduce a canonical homology
basis with the α-cycle encircling η3 counterclockwise on SI , and the β-cycle going from η3 to η4
on SI and coming back to η3 on SII . For more details, see Fig. 8.
Notational convention: To avoid a flurry of symbols, we will often use f pzq instead of
f pPI pzqq, or f pPII pzqq, where needed. Readers beware!
Let p be a point on R. We seek a column vector function W on the Riemann surface defined
as ˆ ˙
w1 ppq
W ppq “
w2 ppq
which is used to express the matrix-valued function Ξ as follows:
´ ¯
Ξpzq “ W pPI pzqq , W pPII pzqq . (5.3)

One can check that (5.3) is consistent with the jump condition that Ξ satisfies. Note that since
Ξ has 21 -root singularities at the points A, ´A in the z-plane, the function W has simple poles
at the points A, ´A on R. Moreover, from the large-z asymptotics of the function Ξ, we deduce
that „ˆ ˙ ȷ
1
W ppq “ ` Op1{pq e´ipa1 p`a0 `ν0 q , p Ñ 81
0
„ˆ ˙ ȷ
0
W ppq “ ` Op1{pq eipa1 p`a0 `ν0 q , p Ñ 82
1
where 81 and 82 are the images of 8 in SI and SII , respectively. Following [6], we solve RHP
5.4 in terms of Theta functions. To this end, we define the function ρ as follows
#
Sppq, p P SI
ρppq “
´Sppq, p P SII

¸ c
Now, consider the integral α ρpsq ds. One can check that for c as in (3.7), it follows that
¸ c
α ρpsq
ds “ 1. Furthermore, we can express τ defined in (3.7) as an integral on the β-cycle as
˛
c
τ“ ds ,
β ρpsq
5 GLOBAL PARAMETRIX 24

and we recall that it is positive imaginary, see Remark 3.1. Next, we introduce the Abel map
related to R as follows ˆ p
c
Appq “ ds. (5.5)
A ρpsq
This is typically multi-valued, and if one wants to create a single-valued function, one may
restrict the path of integration to avoid PI pγq Y PII pγq Y PI prA, 8sq Y PII prA, 8sq. Moreover,
the function A satisfies the following conditions

A` ppq ` A´ ppq “ 0, p P η3 ,
A` ppq ´ A´ ppq “ ´1, p P PI pγq,
A` ppq ´ A´ ppq “ 1, p P PII pγq,
A` ppq ` A´ ppq “ τ, p P η4
Next, we consider the Jacobi Theta-3 function [12, 20.2]
ÿ 2
Θpz; τ q “ e2πinz`πn iτ , z P C
nPZ

which is an even function of z satisfying the following properties

Θpz ` 1; τ q “ Θpz; τ q
Θpz ` τ ; τ q “ e´πiτ ´2πiz Θpz; τ q
2
τ ´2πimz
Θpz ` n ` mτ ; τ q “ e´πim Θpz; τ q, n, m P Z

2 . We use the Θ-function to construct W ppq for p P R, and then

and has a simple zero at z “ τ `1
through (5.3) we find Ξpzq solving RHP 5.4. Recall that the function W ppq has simple poles at
the points A, ´A and that two entries of W ppq vanish at 81 and 82 .
We first construct two meromorphic scalar functions on R, each having simple poles at
A, ´A, with one function vanishing at 81 and the other at 82 . We guess that such functions
have the following form
` ˘ ` ˘
Θ Appq ´ Ap82 q ` τ `1
2 ; τ Θ Appq ´ ApQ1 q ` τ `1
2 ;τ
F1 pp; τ q “ ` ˘ ` ˘
Θ Appq ´ ApAq ` τ `1 τ `1
2 ; τ Θ Appq ´ Ap´Aq ` 2 ; τ
` ˘ ` ˘
Θ Appq ´ Ap81 q ` τ `1
2 ; τ Θ Appq ´ ApQ2 q ` τ `1
2 ;τ
F2 pp; τ q “ ` ˘ ` ˘
Θ Appq ´ ApAq ` τ `1 τ `1
2 ; τ Θ Appq ´ Ap´Aq ` 2 ; τ

where Q1 , Q2 P R are zeros of F1 pp; τ q and F2 pp; τ q, respectively, on R, which will be determined.
The structure of F1 and F2 ensures that the number of poles of each matches the number of
zeros (which must be true for meromorphic functions in this Riemann surface). Additionally,
they do not have jumps across η3 , η4 , but they do across γ. Thus, they nearly solve the RHP
for W ppq, except for the asymptotic behavior of F1 at 81 , F2 at 82 , and their jump across γ.
To fix the asymptotic behavior of W ppq, we introduce the function δppq on R as follows
ˆ p
a2 s2 ` r a1 s ` r
a0
(5.10)
r
δppq “ ds
A ρpsq
where ra2 , r a0 have to be determined. Below, we present a proposition that is useful for
a1 , r
understanding the analytic properties and the asymptotic behavior of δppq.
Proposition 5.5. Consider the Riemann surface R previously defined and let δppq be as in (5.10).
Setting
ˆ ´A
s2
a0 “ ´2ca1
r ds , r
a1 “ 0 , r
a2 “ a1 ,
A S` psq
ˆ ´A ˆ 8 2
s2 s ´ Spsq
ν0 “ a1 p1 ´ τ q ds ´ a1 A ´ a0 ` a1 ds ,
A S` psq A Spsq
5 GLOBAL PARAMETRIX 25

then δppq satisfies the asymptotic behavior

ˆ ˙
1
δppq “ a1 p ` a0 ` ν0 ` O , p Ñ 81
p
where a1 , a0 are given in equation (5.2), and ν0 appears in the asymptotic behavior of W ppq in
(5.4). Then, δpzq is analytic in Cztη3 Y η4 u (recalling our slight abuse of notation) and satisfies
the following conditions

δ` pzq ` δ´ pzq “ 0, z P η3
δ` pzq ` δ´ pzq “ ζpx, N q, z P η4
where
˜ˆ ˆ ¸
´A
τ ∆pxq ´ 2 lnpN q s2 s2
ζpx, N q “ ´ ds ´ τ ds P iR.
2cπ ´A Rpsq η1 R` psq

Proof. Let p P SI , then

ˆ ˆ ˙˙
2 1
ρppq ” Sppq „ p 1`O , p Ñ 81
p2
Therefore
a2 p2 ` r
r a1 p ` r
a0 a2 p2 ` r
r a1 p ` r
a0
“ ´ ´ ¯¯
ρppq p2 1 ` O p12
ˆ ˙ˆ ˆ ˙˙
a1 r
r a0 1
“ r a2 ` ` 2 1`O
p p p2
ˆ ˙
a1
r 1
“ra2 ` `O .
p p2
We can then write
ˆ p ˆ pˆ ˆ pˆ
a2 s2 ` r a2 s2 ` r
˙ ˙
r a1 s ` r
a0 r a1 s ` r
a0 a1
r a1
r
ds “ a2 ´
´r ds ` a2 `
r ds
A ρpsq A ρpsq s A s
ˆ p
a2 s2 ` r
ˆ ˙
r a1 s ` r
a0 a1
r
“ a2 ´
´r ds ` pr
a2 p ´ r
a2 A ` r
a1 lnppq ´ r
a1 lnpAqq .
A ρpsq s
To match the asymptotics of W ppq as p Ñ 81 in equation (5.4), we require
ˆ 8 2
τ ´1 s ´ Spsq
a2 “ a1 ,
r a1 “ 0 ,
r ν0 “ a0 ´ a1 A ´ a0 ` a1
r ds ,
2c A Spsq
´8 1
where we used Lemma 4.1 to compute A Spsq ds. Additionally, if p P γ, then
ˆ ´A
a1 s2 ` r
a0
δ` ppq ´ δ´ ppq “ 2 ds
A S` psq
which we require to be zero, so δppq is analytic in γ. This fixes r
a0 as
ˆ ´A
s2
a0 “ ´2ca1
r ds .
A S` psq
Plugging r
a0 into the previous expression for ν0 , we get
ˆ ´A ˆ 8 2
s2 s ´ Spsq
ν0 “ a1 p1 ´ τ q ds ´ a1 A ´ a0 ` a1 ds .
A S` psq A Spsq
5 GLOBAL PARAMETRIX 26

Finally, a straightforward computation shows that

δ` ppq ` δ´ ppq “ 0, p P η3
ˆ ´A
a1 s2 ` r
a0
δ` ppq ` δ´ ppq “ 2 ds, p P η4
´A Spsq

where we can rewrite the second jump relation (on η4 ) as follows

˜ˆ ˆ ¸
´A
s2 s2
δ` ppq ` δ´ ppq “ 2a1 ds ´ τ ds
´A Spsq η3 S` psq
˜ˆ ˆ ¸
´A
τ ∆pxq ´ 2 lnpN q s2 s2
“´ ds ´ τ ds
2cπ ´A Spsq η3 S` psq
˜ˆ ˆ ¸
´A
τ ∆pxq ´ 2 lnpN q s2 s2
“´ ds ´ τ ds “: ζpx, N q .
2cπ ´A Rpsq η1 R` psq

We used Lemma 4.1 to simplify the expression. We notice that ζpx, N q depends on N and x
since a1 does. Furthermore, it is easy to check that ζ P iR. This concludes the proof of the
proposition.

5.1 Solution on the Riemann surface

We are now in position to explicitly solve the RHP for the function W ppq (5.4) on the Riemann
surface R. Define the functions
` ˘ ` ˘
1 Θ Appq ´ Ap82 q ` τ `1
2 ; τ Θ Appq ´ ApQ1 q ` τ `1
2 ; τ ´iδppq
w1 ppq “ ` ˘ ` ˘e ,
F1 p81 ; τ q Θ Appq ´ ApAq ` τ `1 τ `1
2 ; τ Θ Appq ´ Ap´Aq ` 2 ; τ
` ˘ ` ˘
1 Θ Appq ´ Ap81 q ` τ `1
2 ; τ Θ Appq ´ ApQ2 q ` τ `1
2 ; τ ´iδppq
w2 ppq “ ` ˘ ` ˘e .
F2 p82 ; τ q Θ Appq ´ ApAq ` τ `1 τ `1
2 ; τ Θ Appq ´ Ap´Aq ` 2 ; τ

We notice that defining W ppq “ pw1 ppq, w2 ppqq⊺ , it has the right asymptotic behavior for
p Ñ 8, but, in general, it is not single-valued on the Riemann surface. Indeed, with the
notational convention that p ` β means adding a β-cycle, we have

ˆ ˆ ˙˙
ζpx, N q
w1 pp ` βq “ exp 2πi Ap82 q ` ApQ1 q ´ ApAq ´ Ap´Aq ´ w1 ppq ,
2π
ˆ ˆ ˙˙ (5.11)
ζpx, N q
w2 pp ` βq “ exp 2πi Ap81 q ` ApQ2 q ´ ApAq ´ Ap´Aq ´ w1 ppq .
2π

Therefore, to have W ppq single valued on the Riemann surface, we must set

ζpx, N q
Ap82 q ` ApQ1 q ´ ApAq ´ Ap´Aq ´ “ 0,
2π
ζpx, N q
Ap81 q ` ApQ2 q ´ ApAq ´ Ap´Aq ´ “ 0,
2π
which implies that

ζpx, N q
ApQ1 q “ ApAq ` Ap´Aq ` ´ Ap82 q ,
2π
ζpx, N q
ApQ2 q “ ApAq ` Ap´Aq ` ´ Ap81 q .
2π
6 LOCAL PARAMETRICES 27

Note that this determines the zeros Q1 , Q2 of F1 pp; τ q and F2 pp; τ q, respectively, although
the path of integration from A to Q1 or Q2 will necessarily involve many traverses of the cycle
β, since ζpx, N q is growing in N . We can further simplify the expression for W ppq applying the
following proposition, which is a Corollary of Lemma 4.1.
Proposition 5.6. Consider the Riemann surface R, and define the Abel map Appq (5.5); then
τ 1 τ ´1
ApAq “ , Ap´Aq “ , Ap81 q “ ´Ap82 q “ .
2 2 4
Therefore, we can rewrite the solution W ppq as
¨ ζpx,N q
˛
1 ΘpAppq` 1`3τ
4 ;τ qΘpAppq´ 2π ` 1´τ
4 ;τ q ´iδppq
ˆ
w1 ppq
˙ e
˚ F1 p81 ;τ q 1 τ
ΘpAppq` 2 ;τ qΘpAppq` 2 ;τ q
W ppq “ ‚.
‹
“˝ ΘpAppq` 3`τ
ζpx,N q
` τ ´1
w2 ppq 1 4 ;τ qΘpAppq´ 2π 4 ;τ q ´iδppq
F2 p82 ;τ q 1 τ
ΘpAppq` 2 ;τ qΘpAppq` 2 ;τ q
e

5.2 Solution on the Complex plane

We now have to project the solution from the Riemann surface R to the complex plane. Consider
the two projections PI pzq, PII pzq defined from the complex plane to the first and the second
sheet of R respectively, see Fig. 8. Then, one can rewrite the solution of the RHP 5.4 as
ˆ ˙ ˆ ˙
w1 pPI pzqq w1 pPII pzqq s pz; x, tq s12 pz; x, tq
Ξpzq “ “: 11 , (5.12)
w2 pPI pzqq w2 pPII pzqq s21 pz; x, tq s22 pz; x, tq
from the definition of w1 ppq, w2 ppq (5.11) it is easy to see that the previously defined function
solves RHP 5.4.
For future use, we write the previous solution explicitly, making use of our previously fore-
warned notational abuse in which Apzq and δpzq are considered to be evaluated on the complex
plane, or, equivalently, on the first Riemann sheet.
First, we explicitly write the norming constant F1 p81 ; τ q, F2 p82 ; τ q
´ ¯
Θ pτ ; τ q Θ ´ ζpx,N
2π
q
; τ
F1 p8q :“ F1 p81 ; τ q “ ` ˘ ` ˘ , (5.13)
Θ 3τ4´1 ; τ Θ τ `1 4 ;τ
´ ¯
Θ p0; τ q Θ ´ ζpx,N2π
q
; τ
F2 p8q :“ F2 p82 ; τ q “ ` ˘ ` 1`τ ˘ , (5.14)
Θ 3´τ 4 ; τ Θ 4 ;τ

then, the entries of the solution (5.12)

` 1`3τ
˘ ´ ζpx,N q 1´τ
¯
1 Θ Apzq ` 4 ; τ Θ Apzq ´ 2π ` 4 ; τ
s11 pz; x, tq “ ` ˘ ` ˘ e´iδpzq
F1 p8q Θ Apzq ` 21 ; τ Θ Apzq ` τ2 ; τ
` 1`3τ
˘ ´ ζpx,N q 1´τ
¯
1 Θ ´Apzq ` 4 ; τ Θ ´Apzq ´ 2π ` 4 ; τ
s12 pz; x, tq “ ` ˘ ` ˘ eiδpzq
F1 p8q Θ ´Apzq ` 12 ; τ Θ ´Apzq ` τ2 ; τ
˘ ´ ¯ (5.15)
` 3`τ ζpx,N q τ ´1
1 Θ Apzq ` 4 ; τ Θ Apzq ´ 2π ` 4 ; τ
s21 pz; x, tq “ ` 1
˘ ` τ
˘ e´iδpzq
F2 p8q Θ Apzq ` 2 ; τ Θ Apzq ` 2 ; τ
` 3`τ
˘ ´ ζpx,N q τ ´1
¯
1 Θ ´Apzq ` 4 ; τ Θ ´Apzq ´ 2π ` 4 ; τ iδpzq
s22 pz; x, tq “ ` ˘ ` ˘ e .
F2 p8q Θ ´Apzq ` 21 ; τ Θ ´Apzq ` τ2 ; τ

6 Local parametrices
In the previous section, we developed a solution to RHP 5.1 which serves as an outer approx-
imation to the solution of RHP 4.4 when we are not too close to the four branching points
6 LOCAL PARAMETRICES 28

A, A, ´A, ´A. However, in the vicinity of these points, this approximation is no longer valid.
Therefore, we must construct a solution in a small neighborhood around each of these four
points to better approximate the solution to RHP 4.4 locally. The global solution to RHP 4.4
will then be given in terms of the error function that describes the difference between the global
solution and the approximate solutions, inside and outside the four neighborhoods, which as we
will see in this section, exists since it satisfies a small norm RHP.

6.1 Local parametrix near z “ A

Recall from proposition 4.3 that UA is a small neighborhood of z “ A. Let us denote EA the
solution to the local problem near z “ A. Below, we outline the process we follow to obtain the
solution EA . First, inside UA , we seek EA satisfying the Riemann-Hilbert jump relations

EA` pzq “ EA´ pzq JEA pzq

$˜ ¸
’ 0 i
& i 0 , z P η4
’
’
(6.1)
’
JEA pzq “
’
’
’
’
% `
N pzq, z P ω1 Y ω2

Following [28], we construct EA in terms of special functions. For this purpose, we define
the function Ψpξq as follows

$¨ i
˛
’
’ I0 p2ξ 1{2 q π K0 p2ξ
1{2
q
’
‚, argpξq P p´π{4, π{4q
’
’ ˚ ‹
’˝
’
’ 1{2 1 1{2 1{2 1 1{2
’ 2πiξ I0 p2ξ q ´2ξ K0 p2ξ q
’
’
’
’
’
’
’
’
’ ¨ 1 p1q 1 p2q
˛
1{2 1{2
2 H0 p2p´ξq 2 H0 p2p´ξq
’
’
’
& q q
Ψpξq “ ˝ ‚, argpξq P pπ{4, πq (6.2)
˚ ‹
’ p1q 1 p2q 1
1{2 1{2 1{2 1{2
πξ pH0 q p2p´ξq q πξ pH0 q p2p´ξq q
’
’
’
’
’
’
’
’
’
’ ¨ ˛
1 p2q p1q
2 H0 p2p´ξq
1{2
´ 12 H0 p2p´ξq1{2 q
’
’
’
’ q
’
‚, argpξq P p´π, ´π{4q
’
’ ˚ ‹
’
’ ˝
’ p2q 1 p1q 1
% 1{2 1{2 1{2 1{2
´πξ pH0 q p2p´ξq q πξ pH0 q p2p´ξq q

where H0 and H0 are the Hankel functions and K0 and I0 are the modified Bessel functions,
p1q p2q

see [12, Chapter 10]. In [28], it is shown that Ψ satisfies the jumps given in Fig. 9. For brevity,
we introduce the notation Ypzq “ 2πhpzqρpzqe´2iθpzq .
Proposition 6.1. The matrix valued function EA given by the expression

2c
´z ds
´ ¯´σ3 {2
EA pzq “ QA pzqeiπσ3 {4 Ψ pφA pzqq e´iπσ3 {4 elnpN q τ σ
A Rpsq 3 e2rypzq Ypzq (6.3)

satisfies the jump relations (6.1). Here Ψ is given by equation (6.2), QA is an analytic function
inside UA and the map φA is defined as
ˆ
c z ds
φA pzq1{2 “ ´ lnpN q . (6.4)
τ A Rpsq
6 LOCAL PARAMETRICES 29

´ ¯
1 0
1 1

´ ¯
0 1
´1 0
0

´ ¯
1 0
1 1

Figure 9: Jumps for Ψpξq

Proof. Notice that we can factorize the matrix N` pzq as follows

´ ¯σ3 {2 ˆ 1
˙
0 ´ 2rypzq ¯´σ3 {2
` y pzq
2r
N pzq “ e Ypzq e Ypzq
´ielnpN qũpzq 1
and therefore, using conditions (4.5) and the definition (4.2), we can rewrite the jump of EA as
$ ˆ ˙ ,
0 i
, z P η4 /
’
’ /
/
¯σ3 {2 & i 0
’
’ /
´ ’ /
.´ ¯´σ3 {2
JEA pzq “ e2ry´ pzq Ypzq e2ry` pzq Ypzq .
’ ˆ ˙ /
’
’ 1 ´z
0 /
/
’ 4c ds , z P ω1 Y ω2 //
´ielnpN q τ
’
1
% A Rpsq
-

Next, we consider the function

´ ¯σ3 {2
P pzq “ EA pzq e2rypzq Ypzq

and it is easy to check directly from the jumps of EA , that P satisfies the following jump relation
$˜ ¸
’ 0 i
, z P η4
’
’
’
’
’
’ i 0
&
P` pzq “ P´ pzq
’
’ ˜ ¸
’
’ 1 ´ z ds
0
, z P ω1 Y ω2
’
’
% ´ielnpN q 4c
’
τ A Rpsq 1
´ z ds
Analysis of the local behavior of the function 4c τ A Rpsq near z “ A , prompts to introduce
locally the map φA : z ÞÑ φ defined via the relation
ˆ
1{2 c z ds
pφA pzqq ” φ1{2 “ ´ lnpN q .
τ A Rpsq
The jumps of P then become
$˜ ¸
’ 0 i
, p´8, 0q
’
’
’
’
’
’ i 0
&
JP pφA pzqq “
’
’ ˜ ¸
’
’ 1 0
, φA pω1 q Y φA pω2 q
’
’
% ´ie´4φ1{2
’
1
6 LOCAL PARAMETRICES 30

where φA pω1 q and φA pω2 q are the images of ω1 and ω2 under the map φA . We then define the
function
1{2
ΨA pφA pzqq “ P pφA pzqq e2φ σ3

which satisfies the jump relation

$˜ ¸
’ 0
’ i
’
’ , p´8, 0q
’ i
’
’ 0
&
ΨA` pφA pzqq “ ΨA´ pφA pzqq (6.5)
’
’ ˜ ¸
’ 1 0
’
’
% ´i 1 , φA pω1 q Y φA pω2 q
’
’

Finally, consider the function

ΨA pφA pzqq “ eiπσ3 {4 ΨpφA pzqq e´iπσ3 {4 (6.6)
where Ψ is given by equation (6.2). We can check directly by using the jump relation of Ψ
that ΨA in (6.6) satisfies the desired jumps. Unraveling the transformations, we see that (6.3)
satisfies the desired jump relations near z “ A, for any analytic and invertible matrix valued
function QA . The quantity QA will be determined at a later stage.

6.2 Local parametrix near z “ ´A

Recall that U´A denotes a small neighborhood of z “ ´A and E´A is the solution to the local
problem near z “ ´A. Inside U´A , E´A satisfies the jump relation
$˜ ¸
’ 0 i
’
’ , z P η4
i 0
’
’
’
’
’
’
&
E´A` pzq “ E´A´ pzqJE´A pzq, JE´A pzq “ 2 (6.7)
’
’
’ ep´∆` τ lnpN qqσ3 , zPγ
’
’
’
’
’
’
% `
N pzq, z P ω1 Y ω2

Proposition 6.2. The matrix valued function E´A given by the expression
˜ ¸σ3 {2 #
´z 1
iπσ3 {4 ´iπσ3 {4 lnpN q 2c ds
σ3 1 e´prypzq´ τ lnpN qqσ3 , zPΩ
E´A pzq “ Q´A pzq e Ψpφ´A pzqq e e τ ´A Rpsq
1
Ypzq e´prypzq` τ lnpN qqσ3 , zPΩ
r

satisfies the jump relation (6.7). Here, Ψ is as in (6.2), Q´A is an analytic function inside U´A ,
the map φ´A is defined as
ˆ
1{2 c z ds
φ´A pzq “ ´ lnpN q ,
τ ´A Rpsq

and the domains Ω, Ω

r are as in Figure 10.

Proof. We introduce the function m:

$
pyrpzq´ τ1 lnpN qqσ3 ,
& E´A pzq e zPΩ
’
’
mpzq “ .
’
% E pzq epyrpzq` τ1 lnpN qqσ3 ,
’
zPΩ
r
´A
6 LOCAL PARAMETRICES 31

Ω
ω2
η4
Ω̃

ω1

Figure 10: Contour for RHP 6.7

Using the jump of E´A and yr across γ, we can check that m is analytic across γ. Moreover,
it satisfies the following jump relations across η4 Y ω1 Y ω2

m` pzq “ m´ pzqJm pzq,

and the jump matrix Jm pzq can be factored as follows

$ ˆ ˙ ,
0 i
’
’ , z P η4 /
/
& i 0
˜ ¸´σ3 {2 ’
’ /
/ ˜ ¸σ3 {2
1 . 1
Jm pzq “ ,
Ypzq ’
’ ˆ
1 ´ 0
˙ /
/ Ypzq
’ /
’ 4c z ds , z P ω1 Y ω2 -/
´ielnpN q τ ´A Rpsq 1
%

and we define the function

˜ ¸´σ3 {2
1
P pzq “ mpzq ,
Ypzq

which satisfies the jump relations

$˜ ¸
’ 0
’
’
i
, z P η4
& i 0
’
’
’
P` pzq “ P´ pzq .
’
’ ˜ ¸
’
’ 1 ´ 0
’
’
% lnpN q 4c z ds , z P ω1 Y ω2
´ie τ ´A Rpsq 1
´z
Analysis of the local behavior of the function 4c
τ ´A
ds
Rpsq near z “ ´A , prompts to introduce
locally the map φ´A : z ÞÑ φ defined via the relation
ˆ z
1{2 c ds
pφ´A pzqq ” φ1{2 “ ´ lnpN q ,
τ ´A Rpsq

and therefore the jumps of P become

$˜ ¸
’ 0 i
, p´8, 0q
’
’
’
’
’
’ i 0
&
JP pφ´A pzqq “
’
’ ˜ ¸
’
’ 1 0
, φ´A pω1 q Y φ´A pω2 q.
’
’
% ´ie´4φ1{2
’
1
6 LOCAL PARAMETRICES 32

We then define the function

1{2
Ψ´A pφ´A pzqq “ P pφ´A pzqq e2φ σ3
,

which satisfies the jump relation

$˜ ¸
’ 0 i
, p´8, 0q
’
’
’
’
’
’ i 0
&
Ψ´A` pφ´A pzqq “ Ψ´A´ pφ´A pzqq .
’
’ ˜ ¸
’
’ 1 0
% ´i 1 , φ´A pω1 q Y φ´A pω2 q
’
’
’

Notice that this is the jump relation that ΨA satisfies. Therefore, the function Ψ´A is given by

Ψ´A pφ´A pzqq “ eiπσ3 {4 Ψpφ´A pzqq e´iπσ3 {4 ,

and we can check directly by using the jumps of Ψ, that Ψ´A above, satisfies the desired jumps.
Tracking the subsequent transformations we have completed the proof of the proposition.

6.3 Local parametrix near z “ A

Recall that UA is a small neighborhood of z “ A and we denote by EA the solution to the local
problem near z “ A. Inside UA , EA satisfies the jump relation
$˜ ¸
’ 0 i
’
& i 0 , z P η3
’
’
EA` pzq “ EA` pzqJEA pzq, JEA pzq “ . (6.8)
’
’
’
’
% ´
M pzq, z P ω3 Y ω4

Proposition 6.3. The matrix valued function EA given by the expression

´z
2c ds
´ e2rypzq ¯´σ3 {2
EA pzq “ QA pzqΨA pφA pzqq elnpN q τ σ
A Rpsq 3 , (6.9)
Ypzqe4iθpzq

satisfies the jump relation (6.8). Here, the function ΨA is given by the expression
ˆ ˙
0 ´i
ΨA pφA pzqq “ ´σ2 ΨA pφA pzqq σ2 , σ2 “ , (6.10)
i 0

and satisfies the jumps shown in Fig. 11. Moreover, ΨA is as in (6.6) and satisfies the jump
relation (6.5), QA is an analytic function inside UA , and the map φA is defined as
ˆ
c z ds
φA pzq1{2 “ ´ lnpN q .
τ A Rpsq

Proof. Similar analysis as before, simplifies JEA into

$ ˆ ˙ ,
0 i
, z P η3 /
’
’ /
& i 0
˜ ¸σ3 {2 ’
’
’
/
/
/ ˜ ¸´σ3 {2
e2ry´ pzq . e2ry` pzq
JEA pzq “ ´ z ds ˙
.
Ypzqe4iθpzq ’ ˆ ´ lnpN q 4c
/ Ypzqe4iθpzq
1
’ τ
/
’
’
’ ´ie A Rpsq
, z P ω3 Y ω4 /
/
/
0 1
% -
6 LOCAL PARAMETRICES 33

We define the matrix function

˜ ¸σ3 {2
e2rypzq
P pzq “ EA pzq ,
Ypzqe4iθpzq

and we introduce locally the map φA : z ÞÑ φ via the relation

ˆ
1{2 c z ds
pφA pzqq ” φ1{2 “ ´ lnpN q .
τ A Rpsq
The jumps of P now become
$˜ ¸
’ 0 i
,
’
’
’ p0, ´8q
’
’
’ i 0
&
JP pφA pzqq “ .
’
’ ˜ 1{2
¸
1 ´ie4φ
’
’
, φA pω3 q Y φA pω4 q
’
’
’
% 0 1

Finally, we define the function

1{2
ΨA pφA pzqq “ P pφA pzqq e2φ σ3
,
which satisfies the jumps shown in Fig. 11 with the prescribed orientation. We claim that the
function ΨA given by the expression
ΨA pφA pzqq “ ´σ2 ΨA pφA pzqq σ2
satisfies the jumps shown in Fig. 11, where ΨA is as in (6.6) and satisfies the jump relation
(6.5). To show this claim, let us denote by Σ the contour of non-analyticity for ΨA , and Σ
r the

´ ¯
1 ´i
0 1

´ ¯
0 i
i 0
ΣN
0

´ ¯
1 ´i
0 1

Figure 11: Jumps for ΨA on Σ

r

contour of non-analyticity for ΨA . Note that the two contours are oriented the opposite way. It
is easy to check that
ˇ ˇ
ΨA˘ pzqˇ “ ΨA¯ pzqˇ
ˇ ˇ
Σ r Σ

and therefore it is immediate that

ˇ ´ ˇ ¯´1
JΨA pzqˇ “ JΨA pzqˇ
ˇ ˇ
Σ
r Σ
6 LOCAL PARAMETRICES 34

´ ¯
1 0
´i 1

´ ¯
0 i
i 0
Σ
0

´ ¯
1 0
´i 1

Figure 12: Jumps for ΨA on Σ

´ ¯
1 0
i 1

´ ¯
0 ´i
´i 0
ΣN
0

´ ¯
1 0
i 1

Figure 13: Jumps for ΨA on Σ

r

which implies that ΨA satisfies the jumps shown on Fig. 13 along the contour Σ. r Using the
jumps of ΨA on Σ, it is easy to check that the function defined in (6.10) satisfies the desired
r
jump relation on Σ.
r Tracking the subsequent transformations we obtain the local solution given
in (6.9).

Similarly, one can construct the local solution of the problem near z “ ´A, the details of
which we omit.

6.4 Error function and global parametrix

In this subsection, we define the global approximation to E, and introduce the error function E.
The global approximation is defined separately within each of the four disks and in the region
exterior to all the disks.

$
’
’
’ EA pzq, z P UA
z P UA
’
& EA pzq,
’
’
Eapprox pzq “ E´A pzq, z P U´A (6.11)
’
’ E´A pzq, z P U´A
’
’
’
otherwise.
’
% Λpzq,

The error matrix, Epzq is defined as follows

6 LOCAL PARAMETRICES 35

´1
Epzq “ Epzq pEapprox pzqq .

Here, recall that E is given by (4.7) and Λ is the outer approximate solution given in (5.1).
We choose the functions QA , QA , Q´A , Q´A appearing in the definitions of the local solutions
so that the quantity Epzq is in the small norm setting. Specifically, we compute the jump of Epzq
across the disk boundaries, and then choose these quantities to ensure that the jump matrices
are of the form I ` small.
We start with the jump relationship satisfied by Epzq for z P BUA where BUA (as all the four
neighborhoods) has been chosen with counterclockwise orientation:
´1
JE pzq “ E´ pzq E` pzq
´ ¯´1
´1
“ EpzqΛ´1 pzq EpzqEA pzq
´ ¯σ3 {2 ´
2c z ds
“ Λpzq e2rypzq Ypzq e´ lnpN q τ A Rpsq σ3 eiπσ3 {4 Ψ´1 pφA pzqq e´iπσ3 {4 Q´1
A
pzq.

Next we utilize the asymptotic behavior of the function Ψ defined in (6.2), as presented
in [28, p. 369]
¯´σ3 {2 1 ˆ ˙
1 ` Opξ ´1{2 q i ` Opξ ´1{2 q 2ξ1{2 σ3
´
Ψpξq “ 2πξ 1{2 ? ´1{2 e
2 i ` Opξ q 1 ` Opξ ´1{2 q

uniformly as ξ Ñ 8. Using this, we find

? ¯σ3 {2 ´
2 ´ 2c z ds 1{2
JE pzq “ Λpzq e2rypzq Ypzq e´ lnpN q τ A Rpsq σ3 eiπσ3 {4 e´2pφA pzqq σ3
¨ 2 ¨ ´ ¯ ´ ¯˛ ˛
´1{2 ´1{2
ˆ ˙ O φA pzq O φA pzq ˆ ˙´ ¯σ3 {2
˚I2 ´ 1 1 ´i ˚ ‹ ` . . .‹ 1 ´i 2πφ1{2 pzq
˚ ˚ ‹ ‹ ´1
e´iπσ3 {4 QA pzq .
˝ 2 ´i 1 ˝ ´ ´1{2 ¯ ´
´1{2
¯‚ ‚ ´i 1 A
O φA pzq O φA pzq

Using (6.4), one can further simplify the previous expression because
ˆ
2c z ds
ˆˆ ˙ ˙
iπ 1{2
exp ´ lnpN q ` ´ 2φA pzq σ3 “ eiπσ3 {4 .
τ A Rpsq 4

Therefore, JE pzq becomes

? ¯σ3 {2
2 ´
JE pzq “ Λpzq e2rypzq Ypzq eiπσ3 {4
¨ 2 ¨ ´ ´ ¯¯˛ ˛
´1{2 ´1{2
ˆ ˙ O φA pzq O φA pzq ˆ ˙´ ¯σ3 {2
˚I2 ´ 1 1 ´i ˚ ‹ ` . . . ‹ 1 ´i 2πφ1{2 pzq
˚ ‹ ‹
˚ e´iπσ3 {4 Q´1 pzq .
˝ 2 ´i 1 ˝ ´ ´1{2 ¯ ´
´1{2
¯‚ ‚ ´i 1 A A
O φA pzq O φA pzq
(6.13)

At this point, we require that JE behaves like the identity matrix plus small corrections
across BUA . Therefore, matching the leading order terms in both sides in the last relation, we
must define QA as follows:
? ¯σ3 {2 ˆ ˙´ ¯σ3 {2
2 ´
1 ´i 1{2
QA pzq “ Λpzq e2rypzq Ypzq eiπσ3 {4 2π pφA pzqq e´iπσ3 {4 . (6.14)
2 ´i 1
6 LOCAL PARAMETRICES 36

Proposition 6.4. The function QA defined in (6.14) is analytic for z P UA . Moreover, using this
function in (6.3), we have
ˆ ˙
1
JE “ I ` O , z P BUA .
ln pN q

Proof. To establish the analyticity of the function QA pzq inside the disk UA , observe that all
the functions comprising QA pzq are analytic within UA , except for the functions Λpzq, yrpzq,
1{2
and φA pzq which have jumps across p´8, 0q. Computing the jump of QA pzq on p´8, 0q by
using the respective jumps of these functions and the definition of QA pzq, it is straightforward
to verify that these jumps cancel each other. Consequently, QA pzq is analytic on p´8, 0q, and
therefore analytic within UA . Returning to (6.13), we see immediately from the definition (6.4)
that (6.15) holds.

Remark 6.1. Note that we can apply a similar approach to that used in proposition 6.4 to
identify the analytic functions Q present in all local approximate solutions. By utilizing these
functions and employing an argument analogous to that in proposition 6.4 we can establish that
ˆ ˙
1
(6.16)
` ˘
JE pzq “ I ` O , z P B UA Y UA Y U´A Y U´A .
log N

Since the determination of each of QA , Q´A , and Q´A is achieved by similar considerations,
we omit these calculations for the sake of brevity. Next, we are interested at the jump of the
function E across the remaining contours. As observed previously, E has no jumps within the
four disks. From its definition, we can check that E is also analytic across η3 Y η4 Y γ. There
remains the jumps across the contours ω1 Y ω2 Y ω3 Y ω4 :
#
Λpzq M´ pzq Λ´1 pzq, z P ω3 Y ω4
JE pzq “ ` ´1
Λpzq N pzq Λ pzq, z P ω1 Y ω2 .

From proposition (4.3) we can conclude that JE behaves like identity plus small across the
contour ω1 Y ω2 Y ω3 Y ω4 . We have established all the necessary ingredients to conclude that E
defined in (6.11) satisfies a small norm Riemann-Hilbert problem. The contour for this Riemann-
Hilbert problem consists of the four disk boundaries, and the contours tωj X Vu4j“1 , where we
recall that V “ CztUA Y UA Y U´A Y U´A u, as shown in Fig. 14. On each of the contours,
the jump matrix for E is uniformly close to I. Specifically, we have (6.16), and on each of the
contours ωj , we have the uniform bound
` ˘
JE pzq “ I ` O e´c log N , z P ωj X V, j “ 1, . . . , 4.

Then, following (for example) the statement and proof of Theorem 7.10 on P. 1532 of [9], we
may conclude that Epzq not only exists, it also satisfies
ˆ ˙
1
Epzq “ I ` O (6.17)
log N p1 ` |z|q

for all z P C. Therefore, the global solution E exists and is given by the expression
ˆ ˆ ˙˙
1
Epzq “ I ` O Eapprox pzq, z P C.
log N p1 ` |z|q

By unraveling all the transformations we have introduced in Sections 3 and 4, we have established
a complete asymptotic description of the solution to RHP 2.4. In the next section we will carry
out this unraveling in order to determine the asymptotic behavior of the solution to the NLS
equation.
7 SOLUTION TO THE NLS EQUATION 37

ω4

U´A UA
ω3
ω2

U´A UA
ω1
Figure 14: Jumps for the matrix JE

7 Solution to the NLS equation

In this section, we enforce all the previous constructions to prove Theorem 2.6.
Remark 7.1. To obtain ψSG px, t; N q (and its conjugate) we see from RHP 2.4 that we need the
behavior of Ãpzq as z Ñ 8. Specifically, we have:

ψSG px, t; N q “ lim 2iz Ã12 pzq, ψSG px, t; N q “ lim 2iz Ã21 pzq. (7.1)
zÑ8 zÑ8

Therefore, an expression for the matrix-valued function Ãpzq for large-z is required. Observe
that once z is large enough, we are outside all of the various domains, which gives us

Ãpzq “ ξpzq e´plog N gp8q`yp8qqσ3 Epzq eiν0 pzqσ3 Ξpzqeiνpzqσ3 eplog N gpzq`ypzqqσ3 . (7.2)

To see this, recall that in the exterior region, we have:

Epzq “ Dpzq “ Cpzq, Bpzq “ Ãpzq, Eapprox pzq “ Λpzq,

which, when combined with the definitions of Cpzq, Λpzq and Epzq (see relations (3.1), (5.1) and
(6.11)) leads to (7.2).
Now, we prove Theorem 2.6, which we repeat for the convenience of the reader.
Theorem 7.1. Under the N -soliton condensate scattering data assumption, for all px, tq in
a compact set K, there is N0 so that for all N ą N0 , ψSG px, t; N q satisfies the asymptotic
description (7.4). Specifically, for |ψSG |, we have

´ ¯ ´ ¯
Θ ´ ζpx,N q
` 1´τ ; τ Θ ´ ζpx,N q
` τ ´1
˘˘2
;τ
` `
1 τ `1 ˆ ˙
2 2 Θ 2 ; τ 2π 2 2π 2 1
|ψSG px, t; N q| “ c ´ ¯ `O ,
Θpτ ; τ qΘp0; τ q Θ2 ´ ζpx,N q log N
2π ; τ
´ ¯
where the error term O log1 N is uniform for all px, tq in the compact set K. Here Θpz; τ q is
the Jacobi Theta-3 function as defined in [12]. The constants τ, c and ζpx, N q are given by
˜ˆ ¸´1 ˆ
´A ´A
1 1 1
c“ ds , τ “ 2c ds ,
2 A R` psq ´A Rpsq
˜ˆ ˆ ´A ¸
´A
τ ∆pxq ´ 2 lnpN q s2 s2
ζpx, N q “ ´ ds ´ τ ds ,
2cπ ´A Rpsq A R` psq
7 SOLUTION TO THE NLS EQUATION 38

1 1 1 1
where Rpzq “ pz ´ Aq 2 pz ` Aq 2 pz ´ Aq 2 pz ` Aq 2 , the branch-cuts are the canonical ones, and
the integration is performed on the straight line connecting the endpoints. The function ∆pxq is
defined in terms of the functions ρpzq, hpzq as follows
˜ˆ ¸´1 ˜ ˆ ˆ A ¸
´A ´A
1 log p2πhpsqρpsqq ` 2isx log p2πhpsqρpsqq ´ 2isx
∆pxq “ ds ´ ds ` ds .
´A Rpsq A R` psq ´A R` psq

Proof. Using equations (7.1), (7.2) and (6.17), we have:

ˆ ˙
1
ψSG px, t; N q “ lim 2iz ξpzq s12 pz; x, tq exp p´ lnpN qpgp8q ` gpzqq ` iν0 ´ iνpzq ´ ypzq ´ yp8qq ` O
zÑ8 log N
ˆ ˙
1
ψSG px, t; N q “ lim 2iz ξpzq s21 pz; x, tq exp p` lnpN qpgp8q ` gpzqq ´ iν0 ` iνpzq ` ypzq ` yp8qq ` O
zÑ8 log N
(7.4)

Therefore, since the two previous limits are finite, we can compute the modulus of the solution
as

` 1`3τ
˘ ´ ζpx,N q 1´τ
¯
4z ξ pzq2 2Θ ´Apzq ` 4 ; τ Θ ´Apzq ´ 2π ` 4 ; τ
|ψSG px, t; N q|2 “ lim ´ ` ˘ ` ˘
zÑ8 F1 p8qF2 p8q Θ ´Apzq ` 12 ; τ Θ ´Apzq ` τ2 ; τ
` ˘ ´ ζpx,N q
¯
Θ Apzq ` 3`τ4 ; τ Θ Apzq ´ 2π ` τ ´1
4 ; τ ˆ
1
˙
ˆ ` ˘ ` ˘ `O ,
Θ Apzq ` 12 ; τ Θ Apzq ` τ2 ; τ log N
(7.5)

where we used the definition of s12 , s21 in (5.15). We notice that the terms in the previous
equations have the following asymptotics for large-z

ξpzq „ 1 ` Opz ´1 q
` ˘ ´ ζpx,N q
¯
Θ ´Apzq ` 1`3τ4 ; τ Θ ´Apzq ´ 2π ` 1´τ 4 ;τ θ1 pN q
` 1
˘ ` τ
˘ „ ` Opz ´2 q
Θ ´Apzq ` 2 ; τ Θ ´Apzq ` 2 ; τ z
` ˘ ´ ζpx,N q
¯
Θ Apzq ` 3`τ4 ; τ Θ Apzq ´ 2π ` τ ´1
4 ; τ θ2 pN q
` 1
˘ ` τ
˘ „ ` Opz ´2 q
Θ Apzq ` 2 ; τ Θ Apzq ` 2 ; τ z

where θ1 pN q, θ2 pN q are defined as

´ ¯
ζpx,N q 1´τ
ˆ
τ `1
˙ Θ ´ 2π ` 2 ; τ
θ1 pN q “ ´c Θ1 ;τ ` ˘ ` 1`τ ˘
2 F1 p8qΘ 3´τ 4 ; τ Θ 4 ;τ
´ ¯
ζpx,N q τ ´1
ˆ
τ `1
˙ Θ ´ 2π ` 2 ; τ
θ2 pN q “ c Θ1 ;τ ` ˘ ` 3τ ´1 ˘ ,
2 F2 p8qΘ 1`τ 4 ; τ Θ 4 ;τ

and where the constant c P R´ was previously computed is independent of N . In this notation,
we can rewrite (7.5) as
´ ¯ ´ ¯
Θ ´ ζpx,N q
` 1´τ ; τ Θ ´ ζpx,N q
` τ ´1
` 1 ` τ `1 ˘˘2
Θ ; τ 2π 2 2π 2 ;τ ˆ
1
˙
2 2 2
|ψSG px, t; N q| “ c ´ ¯ `O
Θpτ ; τ qΘp0; τ q Θ2 ´ ζpx,N q ; τ log N
2π

where we also used eqs. (5.13)-(5.14).

A TECHNICAL RESULTS 39

Remark 7.2. Note that the asymptotic description of the solution is valid for px, tq in compact
sets. The solution is a pure periodic elliptic wave and the period is independent of the distribu-
tion ρpzq, the norming constants cj , and the length of the interval p´A, Aq. Finally, we remark
that the modulus of the solution does not depend on time.
We show in figure 3 several plots of the solution (7.4) for different values of N .

Acknowledgments The authors thank Robert Jenkins for valuable discussions. G.M. was
partially supported by the Swedish Research Council under grant no. 2016-06596 while the
author was in residence at Institut Mittag-Leffler in Djursholm, Sweden during the fall semester
of 2024.

A Technical Results
A.1 Proof of Proposition 3.6
Proposition A.1. The function Rpzq defined as
1 1 1 1
Rpzq “ pz ´ Aq 2 pz ` Aq 2 pz ` Aq 2 pz ´ Aq 2 ,
where for each factor p¨q1{2 we use the principle branch, satisfies the following properties
2 2
1. Rpzq “ z 2 ´ A `A ` O z12 , as z Ñ 8,
` ˘
2
2. Rpzq “ Rpzq
3. Rp´zq “ Rpzq
´
4. η1 R`1pzq dz P R´
´ 1
5. γ Rpzq dz P iR´

Proof. Properties 1.-3. follows from direct computations. Regarding 4., we rewrite the integral
as

ˆ ´A ˆ a ˆˆ 0 ˆ a ˙
1 1 1 1
ds “ ´ dx “ ´ dx ` dx
A Rpsq ´a Rpx ` ibq ´a Rpx ` ibq 0 Rpx ` ibq
˜ˆ ˆ a ¸
a
2.-3. 1 1
“ ´ dx ` dx .
0 Rpx ` ibq 0 Rpx ` ibq

We must use some geometry to prove that the equation integral is negative. First, we rewrite
it in terms of the angles

φ1 “ argpx`ib´Aq “ ´π , φ2 “ argpx`ib`Aq “ 0 , φ3 “ argpx`ib`Aq , φ4 “ argpx`ib´Aq ,

as ˜ˆ ¸
a
1
ℑ 1 1 1 1 φ3 `φ4 dx . (A.1)
|x ´ a| 2 |x ´ a ` 2ib| 2 |x ` a| 2 |x ` a ` 2ib| 2 ei 2
0

We notice that the angles φ3 P arctan ab , arctan 2b and φ4 P π2 , π ´ arctan 2b ; there-

` ` ˘ ` ˘˘ ` ` ˘˘
a a
fore, since a, b ą 0
φ3 ` φ4 ´ π ¯
P ,π ,
2 2
which implies 4. With analogue computations, one can show 5.
A TECHNICAL RESULTS 40

A.2 Proof of Lemma 4.1

In this subsection, we give a proof of Lemma 4.1. First, we prove the following auxiliary result.
Proposition A.2. The g-function given in (3.6) satisfies the following symmetry relation

ℜ pgpzqq “ ´ℜ pgpzqq , z P C` . (A.2)

Proof. We can rewrite the g-function (3.6) as

ˆ
1 2c z ds
gpzq “ ´ , z P C`
2 τ A Rpsq

ˆ z
1 2c ds
gpzq “ ´ ´ , z P C´ .
2 τ A Rpsq

Let z P C , then
`

ˆ z ˆ 1
1 2c ds 1 2c pz ´ Aqdω
gpzq “ ´ ´ “´ ´
2 τ A Rpsq 2 τ 0 R pspωqq
ˆ ˆ (A.3)
1
1 2c pz ´ Aq 1 2c z dζ
“´ ´ dω “ ´ ` “ ´gpzq
2 τ 0 Rpζpωqq 2 τ A Rpζq
where we have used the change of variables

spωq “ zω ` p1 ´ ωqA, ζpωq “ zω ` p1 ´ ωqA, 0ďωď1

and the properties Rpzq “ Rpzq, and spωq “ ζpωq. Equation (A.3) shows the desired symmetry.

We are now ready to prove the next lemma that proves the first part of Lemma 4.1.
Lemma A.3. Let the function gpzq be the unique solution to RHP 3.3. For z P η1 , we have the
representation ˆ
4c z 1
g` pzq ´ g´ pzq “ ds. (A.4)
τ A R´ psq
The quantity g` ´ g´ has an analytic extension to C` z pη1 Y γq. We define upzq to be that
analytic extension, ˆ
4c z 1
upzq “ ds, z P C` z pη1 Y γq , (A.5)
τ A Rpsq
where the path of integration is chosen to avoid η1 Y η2 Y γ, so that

u´ pzq “ g` pzq ´ g´ pzq, z P η1 .

There exists a contour η3 connecting A with ´A such that

upzq P iR, z P η3 .

The contour η3 lies above η1 and we take it to be oriented from A to ´A (like η1 ).

Similarly, for z P η2 , we have
ˆ ´A ˆ z
4c 1 4c 1
g` pzq ´ g´ pzq “ ds ` ds.
τ A R´ psq τ ´A R´ psq

As with (A.5), we define

ˆ z
4c 1
ũpzq “ ds, z P C´ z pη2 Y γq ,
τ A Rpsq
A TECHNICAL RESULTS 41

where again the path is chosen to avoid η1 Y η2 Y γ. Note that

ˆ ´A ˆ z
4c 1 4c 1
ũ´ pzq “ ds ` ds, z P η2 ,
τ A R´ psq τ ´A R´ psq

so that
ũ´ pzq “ g` pzq ´ g´ pzq, z P η2 .
There exists a contour η4 connecting ´A with A such that

ũpzq P iR, z P η4 .

The contour η4 lies below η2 and we take it to be oriented from ´A to A (like η2 ).

Furthermore, the following hold
ˆ ´A ˆ ´A ˆ 8
2c 1 2c 1 1 2c 1 1
ds “ 1, ds “ , ´ ds “ ´ 1 . (A.6)
τ ´A Rpsq τ A R` psq τ τ A Rpsq τ

Proof. We prove Lemma for upzq, and the proof for u rpzq is analogous.
First, we verify (A.4), for z P η1
ˆ ˆ ˆ
2c z 1 2c z 1 4c z 1
g` pzq ´ g´ pzq “ ´ ds ` ds “ ds ,
τ A R` psq τ A R´ psq τ A R´ psq
where we used the definition of gpzq (3.6). As we did for the function gpzq, the extension in the
upper-half plane of upzq “ g` pzq´g´ pzq for z P η1 must have an extra contour of non-analyticity
and we impose that such contour is exactly γ. Local analysis of upzq in a neighborhood of
z “ A shows that there exists a single arc emerging from A along which ℜpupzqq P iR. For a
formal discussion, see [34, P.31], or [23, Chapter 3]. Now this arc can be extended maximally,
and there are only a finite number of possibilities which we consider in what follows (using
either [34, Chapters 3 and 4] or [23, Theorem 3.5, p. 37] ), eventually arriving at the conclusion
that it must terminate at ´A. Note that properly speaking we should extend this function to the
Riemann surface R, where we would be considering level trajectories of a quadratic differential.
Following [23, Theorem 3.5, p. 37], the maximally extended arc could either diverge to 8,
or remain in a compact subset of C. If it remains in a compact subset of the plane it must
terminate somewhere, and the only other possibilities are the branch points ´A, A, and ´A.
To show that it connects A with ´A, we rule out the other possibilities, i.e. we prove that
η3 cannot connect A with 8 or ´A or A.
If the contour η3 would connect A with A, then the integral
ˆ A
2c 1
´ ds ,
τ A Rpsq
should be purely imaginary. By deformation, one can rewrite the previous integral as
ˆ ˜ˆ ˆ ´A ˆ A ¸
2c A 1 2c ´A
1 1 1
´ ds “ ´ ds ` ds ` ds .
τ A Rpsq τ A R` psq ´A Rpsq ´A R` psq

Applying Proposition 3.6, we deduce that

ˆ ´A ˆ A
1 1
ds “ ´ ds .
A R` psq ´A R` psq
Therefore, by (3.7),
ˆ A
2c 1 1
´ ds “ ´ .
τ A Rpsq 2
So, η3 cannot connect A with A.
A TECHNICAL RESULTS 42

If η3 would connect A with ´A then

ˆ ´A
2c 1
´ ds ,
τ A Rpsq

should be purely imaginary. By deformation of the contour, we obtain the following equality
ˆ ˜ˆ ˆ ´A ¸
2c ´A 1 2c ´A
1 1
´ ds “ ´ ds ` ds ,
τ A Rpsq τ A R` psq ´A Rpsq

therefore, by applying Proposition 3.6 and (3.7) we deduce that

ˆ ´A
2c 1 1 1
´ ds “ ´ ´ , (A.7)
τ A Rpsq 2τ 2
which has a non-trivial imaginary part, so η3 cannot connect A with ´A.
Finally, if η3 would connect A with 8 then
ˆ
2c 8 1
´ ds ,
τ A Rpsq
would be purely imaginary. By Proposition 3.6 and contour deformation, the previous integral
is equivalent to

ˆ 8
˜ˆ
8 ˆ ´A
¸ ˆˆ ˆ ˙
2c 1 c 1 1 c 1 1 1 1
´ ds “ ds ` ds “ ds ` ds “ ´ ,
τ A Rpsq τ A Rpsq 8 Rpsq τ η1 R´ psq γ Rpsq 4τ 4
where in the last equality we used (A.7). Therefore, we showed that η3 must connect A with
´A and the equalities (A.6). We now prove that the contour η3 is always above η1 . We notice
that in this region, we are extending the function
ˆ
4c z 1
u´ pzq “ ds z P η1 .
τ A R´ psq
As a consequence of Proposition 3.6 and equation (3.7)
ˆ x0 `ib ˆ ´A
1 1 2c
ds “ ds , P iR`
A R´ psq ´x0 `ib R´ psq τ
for all x0 P p0, aq. The previous equality implies that
ˆ
1
ds P R.
η1 R´ psq

Proceeding as in the previous proof, see (A.1), we can show that

˜ˆ ¸ ˜ˆ ¸
x0 `ib x0 `ib
1 1
ℜ ds ą 0 , ℑ ds ą 0
A R´ psq A R´ psq

for all x0 P p0, aq, so ℑpupx0 ` ibqq ą 0, ℜpupx0 ` ibqq ă 0

Consider now a point z0 “ x0 ` iy0 for x0 P p0, aq and y0 ą b, we claim that
ˆˆ x0 `iy0 ˙
1
ℑ ds ă 0 (A.8)
x0 `ib Rpsq
and it’s decreasing as a function of y0 .
First, for y0 P pb, `8q, we rewrite the previous integral in terms of the angles
A TECHNICAL RESULTS 43

φ1 “ argpx0 `iy0 ´Aq , φ2 “ argpx0 `iy0 `Aq , φ3 “ argpx0 `iy0 `Aq , φ4 “ argpx0 `iy0 ´Aq ,

as
˜ˆ φ1 `φ2 `φ3 `φ4
¸
y0
e´i 2
ℜ 1 1 1 1 dy .
b |x0 ` iy ´ a ´ ib| 2 |x0 ` iy ´ a ` ib| 2 |x0 ` iy ` a ´ ib| 2 |x0 ` iy ` a ` ib| 2
We notice that Φ “ φ1 ` φ2 ` φ3 ` φ4 is a continuous function, which has its minimum value
when` y˘0 “ b, where Φ “ π and its maximum when y0 “ `8, where Φ “ 2π. Therefore,
cos Φ2 ď 0 and it is equal to zero only when y0 “ b. This implies (A.8) and that the integral
is a decreasing function of y0 .
To conclude, let x0 P p0, aq and consider the integral
ˆ x0 `iy0 ˆ x0 `ib ˆ x0 `iy0
1 1 1
ds “ ds ` ds .
A Rpsq A Rpsq x0 `ib Rpsq
From the previous proof, we know that the first integral has positive imaginary part, and the
second one has negative imaginary part, which is a decreasing function of y0 . Therefore, for any
x0 P p0, bq there exists a value of yrpx0 q such that
˜ˆ ¸ ˜ˆ ¸
x0 `ib x0 `ir
y
1 1
ℑ ds “ ´ℑ ds ,
A Rpsq x0 `ib Rpsq

and
$ ´´ ¯
& ℑ x0 `iy0 1
´´A Rpsq ds¯ ą0 b ă y0 ă yrpx0 q
% ℑ x0 `iy0 1
A Rpsq ds ă0 y0 ą yrpx0 q
This implies that
#
ℜpupx0 ` iy0 qq ă 0 x0 P p0, aq , b ă y0 ă yrpx0 q
.
ℜpupx0 ` iy0 qq ą 0 x0 P p0, aq , y0 ą yrpx0 q

By applying Proposition 3.6, we deduce an analogous result for x0 P p´a, 0q; therefore, we show
that η3 can be rewritten as
␣ ˇ (
y px0 q, x0 P p´a, aq ,
η3 “ z P C ˇ z “ x0 ` ir
and it lies above η1 .

Finally, we prove the following proposition, which implies the final part of Lemma 4.1.
Proposition A.4. The g-function given in (3.6) satisfies the following inequalities
1
0 ď ℜ pgpzqq ď z P C` zΩ`
3 , (A.9)
2
1
ď ℜpgpzqq ă 1 z P Ω`
3 .
2
rpzq (4.1)-(4.2) satisfy the following inequalities
Furthermore, the functions upzq, u
`
ℜpupzqq ă 0 z P Ω`
3 ℜpupzqq ą 0 z P C` zΩ3 (A.10)
`
upzqq ą 0 z P Ω`
ℜpr 2 ℜpr
upzqq ă 0 z P C´ zΩ2

and the equalities hold for z P η3 or z P η4 respectively. Ω`

2 , Ω3 are the open regions enclosed
`

by η1 , η3 and η2 , η4 respectively, see Figure 6.

A TECHNICAL RESULTS 44

Proof. To prove (A.9), we note that

• gpzq is an analytic function in C` zpη1 Yγq, where γ here denotes the portion of the vertical
contour γ in C` .
• On η1 , g satisfies g` pzq ` g´ pzq “ 1, and on γ, g satisfies g` pzq ´ g´ pzq “ ´2
τ (see
(3.2)-(3.3)).
• From (A.2), it follows that ℜ pgpzqq “ 0 for z P R.
• Since gp8q P iR, then ℜ pgpzqq Ñ 0 as z Ñ 8.
We introduce the related function g̃, defined as follows:
#
1 ´ gpzq, z P Ω`
g̃pzq “ 3
(A.11)
gpzq, z P C` zΩ`
3.

One may verify that in the upper half-plane, g̃ is analytic in C` z pη3 Y γq. Furthermore, it is a
consequence of Lemma A.3 that ℜ pg̃˘ pzqq ” 21 for all z P η3 .
We therefore know that ℜpg̃pzqq is harmonic for z P C` z pη3 Y γq. Therefore, by the maxi-
mum modulus principle, ℜpg̃pzqq can only achieve its maximum and minimum on the boundary
of its domain of harmonicity. But this function is identically 0 on R, vanishes at 8, and is
identically 1{2 on η3 .
We note that ℜpg̃pzqq ě 0 for all z P γ. Indeed, suppose to the contrary that there is a point
ẑ P γ such that ℜpg̃pẑq ă 0. This would imply a minimum z ˚ P γ, with g̃pz ˚ q ă 0. But since
τ P iR, the quantity
g̃` ´ g̃´ “ ´2
#
g̃pzq, for z to the right of γ,
g̃pzq ´ τ2 , for z to the left of γ,

is analytic in a neighborhood of z ˚ , has the same real part as g̃pzq, and therefore has a critical
point at z ˚ , which is impossible since g̃ 1 pzq “ g 1 pzq for z outside Ω`
3 and g pzq only vanishes at
1

8. Thus ℜpg̃pzqq ě 0 for all z P γ.

Now applying the maximum modulus principle, we have that

0“ min ℜ pg̃psqq ă ℜ pg̃pzqq ă max ℜ pg̃psqq “ 1{2, z P C` zpη3 Y γq.

sPRYη3 Yγ sPRYη3 Yγ

Now for z P C` zΩ`

3 , gpzq “ g̃pzq, so that

1
0 ă ℜ pgpzqq ă , z P C` zΩ`
3. (A.12)
2
However, within the set Ω`
3 , using the definition (A.11), we have

1
ă ℜ pgpzqq ă 1, z P Ω`
3. (A.13)
2
We now show (A.10). We notice that upzq “ 1´2 gpzq. Therefore, using the proof of proposition
A.4, and specifically relations (A.12) and (A.13), it follows that

0 ă ℜpupzqq ă 1, z P C` zΩ`
3
´1 ă ℜpupzqq ă 0, z P Ω`
3

rpzq, therefore we have com-

as desired. Because of Remark 4.1, an analogous result holds for u
pleted the proof.
A TECHNICAL RESULTS 45

A.3 Proof of Proposition 2.7

Now, we can prove Proposition 2.7.
Proposition A.5. Let Apzq, Apzq
r be the solutions of RHP 2.3 and RHP 2.4, respectively, then
ˆ ˆ ˙˙
1
Apzq “ I2 ` O Apzq
r , (A.14)
N p1 ` |z|q

uniformly in z.

Proof. Consider the function EA pzq “ Apzqô1 pzq which is analytic on CzΓ1 Y Γ2 . Its jump on
Γ1 equals
˜ ¸
1 0 ´ ¯´1
´1 ´1
JEA pzq “ ô pzqJA pzqJà pzqà pzq “ ô řN cj e2iθpλj q ´ 2iθpλq ô
j“1 z´λj ` N η1 hpλqρpλqe
z´λ
dλ
1
ˆ ˙
0 0 ´ ¯´1
“ I2 ` ô q1 pz;N q ô
N 0

where we have introduced the quantity q1 pz; N q:

˜
N ˆ ¸
ÿ cj e2iθpλj q hpλqρpλqe2iθpλq dλ
q1 pz; N q “ N `N .
j“1
z ´ λj η1 z´λ

The jump for EA on Γ2 is similarly evaluated, and one finds that

ˆ ˙´ ¯
0 q2Npzq ´1
JEA pzq “ I2 ` ô pzq ô
0 0

where
˜
N ˆ ¸
ÿ cj hpλqρpλqe´2iθpλq ds
q2 pz; N q “ N e´2iθpλj q ´ N .
j“1 z ´ λj η2 z´λ

Note that for z P Γ2 , we have via symmetry that

q2 pz; N q “ q1 pz; N q.

Since both A and à solve Riemann-Hilbert problems on the same collection of contours, the
quantity EA also satisfies a Riemann-Hilbert problem:
` ˘ EA pzq that is analytic in Cz pΓ1 Y Γ2 q, satisfies
RHP A.6. Find a 2 ˆ 2 matrix valued function
the normalization condition EA pzq “ I ` O z1 as z Ñ 8, has smooth boundary values on the
` and ´ side of the contours Γ1 and Γ2 , and satisfies the jump relation

pEA q` pzq “ pEA q´ pzqJEA pzq , z P Γ1 Y Γ2 .

We will prove that this Riemann-Hilbert problem is a so-called "small norm Riemann-Hilbert
problem", in the sense that the jump matrix JEA is uniformly close to the trivial jump matrix
I2 . We focus on the quantity q1 pz; N q, which we will show is analytic in an annular region
surrounding the contour Γ1 (but bounded away from the interval η1 ), where it satisfies

|q1 pz; N q| ď const

unformly in the annular region. (The constant depends in particular on the distance between
the annular region and the interval η1 ). Because of the relation (A.19), the same estimate holds
for q2 pz; N q in an annular region surrounding Γ2 .
A TECHNICAL RESULTS 46

First, from the definition (A.16), the analyticity of q1 pz; N q in an annular region surrounding
Γ1 is clear. In order to establish the uniform estimate (A.20), we introduce the map Φ : λ ÞÑ x
defined via the relation
ˆ λ
Φpλq “ ρpsqds.
λ0

Recall that the discrete eigenvalues λj are defined as

ˆ λj
2j ´ 1
ρpsqds “ , j “ 1, 2, . . . , N
λ0 2N

which we can rewrite as xj :“ Φpλj q “ 2j´12N from the definition of Φ. We observe that in the
λ-plane, the discrete eigenvalues λj are not equidistant, and therefore they do not correspond
to the midpoints of intervals that start at λ0 “ ´a ` ib and end at λN `1 “ a ` ib. However,
in the variable x, the points xj defined above become the midpoints of intervals rξj , ξj`1 s, with
equidistant endpoints ξj defined as
j´1
ξj “ , j “ 1, . . . , N ` 1,
N
where we note that ξ1 “ 0 and ξN `1 “ 1. One also has that ξj`1 is the midpoint of the interval
rxj , xj`1 s.
We introduce the following notation:

hpsqe2iθpsq ` ˘
Hpsq “ , F pxq “ H Φ´1 pxq
z´s
where h is the normalization constant sampling function. We have:

ÿN ˆ A N
ÿ ˆ 1 N
ÿ
˜ ˆ ξj`1
¸
Hpλj q ´ N Hpsqds “ F pxj q ´ N F pxqdx “ F pxj q ´ N F pxqdx .
j“1 ´A j“1 0 j“1 ξj

Now we expand F psq using a Taylor series expansion around xj . The first nonzero contribution
arises from F 2 pxj q, and so we find
N ˆ A N
ÿ 1 ÿ 2
Hpλj q ´ N Hpsqds “ ´ F pxj q ` OpN ´3 q
j“1 ´A 24N 2 j“1

where the error term is OpN ´3 q because the function F is analytic in a neighborhood of the
interval r0, 1s. The sum on the right hand side of the last relation is again a Riemann sum on
the same grid, and so it can be uniformly approximated:
N ˆ ˙
ÿ ` ˘ 1
F 2 pxj q “ N F 1 p1q ´ F 1 p0q ` O .
j“1
N

Now using (A.16), (A.21), and some calculus, we find that

˜ ¸
1 ph1 p´Aq ` 2ihp´Aqθ1 p´Aqqe2iθp´Aq hp´Aqe2iθp´Aq 1
q1 pz; N q “ ` (A.23)
24 z`A pz ` Aq 2 ρp´Aq
1 ph1 pAq ` 2ihpAqθ1 pAqqe2iθpAq hpAqe2iθpAq
ˆ ˙ ˆ ˙
1 1
´ ` `O .
24 z´A pz ´ Aq2 ρpAq N

Therefore we have established that q1 pz; N q is analytic in the desired annular region, and satisfies
(A.20). Although such an explicit formula is not entirely necessary, (A.23) provides the leading
A TECHNICAL RESULTS 47

order asymptotic behavior of q1 pz; N q for N large. As mentioned above, this yields the uniform
boundedness of q2 pz; N q in an annular region surrounding Γ2 .
Next, we return to JEA pzq on Γ1 (see formula (A.15)). At this point, it is convenient to
assume that the contour Γ1 encircles not only the interval η1 , but also the collection of contours
η3 , ω3 , and ω4 as well (see Figure 4.4). This assumption simplifies the representation of ô
which we use to estimate the jump matrix JEA pzq. Indeed, then for z P Γ1 we have the following
representation for Ã:
˜ ¸
1 0
Ãpzq “ e´plnpN qgp8q`ypzqqσ3 EpzqΛpzqeplnpN qgpzq`ypzqqσ3 ´
´N η1 hpλqρpλq
λ´z e
2iθpλq
dλ 1

where Epzq is the error function

ˆ and˙ Λpzq is the outer approximate solution. By direct calcula-
0 0
tion, the product ô pzq q1 pzq ´ pzq in (A.15) becomes
ô1
N 0
˜ ¸
´plnpN qgp8q`ypzqqσ3 ypzqσ3
0 0 ´1
e EpzqΛpzqe q1 pzqe´2 lnpN qgpzqσ3 e´ypzqσ3 pEpzqΛpzqq eplnpN qgp8q`ypzqqσ3 .
N 0

We notice that:
1. gp8q P iR and therefore e˘ lnpN qgp8q is bounded in N .
2. the scalar q1 pz; N q, as well as the matrices E, Λ, and eypzqσ3 are all uniformly bounded in
N , for all z P Γ1
and we have
´ ¯
1 ´2 lnpN qgpzq ´ lnpN q 1`2 ℜpgpzqq`ip1`2 Impgpzqqq
e “ e´ lnpN qp1`2gpzqq “ e .
N
Therefore we need an estimate on 1 ´2 lnpN qgpzq
Ne ,
and we have
ˇ ˇ ´ ¯
ˇ 1 ´2 lnpN qgpzq ˇ ´ lnpN q 1`2 ℜpgpzqq C
ˇ e ˇ“e ď , z P Γ1 ,
ˇN ˇ N

which follows because of (A.9). Finally then, we have established that

ˆ ˙
1
JEA pzq “ I2 ` O , (A.24)
N

where the error term is uniform for all z in an annular region surrounding Γ1 . The analysis of
JEA pzq on Γ2 Ă C´ is entirely similar, and we leave the reader to verify that
ˆ ˙
1
JEA pzq “ I2 ` O , (A.25)
N

uniformly for all z in an annular region surrounding Γ2 as well.

Returning to the Riemann-Hilbert problem A.6, we have established that the jump matrix
satisfies (A.24) and is analytic in an annular region around Γ1 , and similarly it satisfies (A.25)
and is analytic in an annular region surrounding Γ2 . Then, following the statement and proof
of Theorem 7.10 on P. 1532 of [9], we may conclude that EA pzq not only exists, it also satisfies
(A.14).
B NON-SYMMETRIC CASE 48

B Non-symmetric case
In this section, we study RHP 2.1 where rpzq ” 0 and the poles tλj uN j“1 accumulate on two
segments ηp1 , ηp2 which are not symmetric with respect to the imaginary axis. We refer to this
situation as the non-symmetric case. For this purpose, we consider initial data generated under
the following assumptions.
Assumption B.1 (non-symmetric N -soliton gas condensate). For a positive integer N , we gen-
erate the scattering data as follows:
1. There are three positive constants a1 ă a2 and b, which define a line segment ηp1 “
ra1 ` ib, a2 ` ibs, which we orient from right to left, and its complex conjugate ηp2 “
ra1 ´ ib, a2 ´ ibs, which we orient from left to right. We define

A1 “ a1 ` ib, A2 “ a2 ` ib

to be the left and right endpoints of ηp1 .

2. There is a density function ρppzq which is analytic in a neighborhood of the line segment
η1 (and the neighborhood’s Schwarz reflection), satisfying

ρppzq is analytic in a neighborhood of each line segment ηp1 and ηp2 ,

ρppzq P R for z “ t ` ib, t P pa1 , a2 q,
ˆ a2
ρppt ` ibqdt “ 1,
a1

ρppzq “ ρppzq.

3. The eigenvalues tλj uN

j“1 on η
p1 are chosen to satisfy
ˆ ℜpλj q
2j ´ 1
ρppx ` ibqdx “ , ℑpλj q “ b, j “ 1, . . . , N.
a1 2N

4. There is a "normalization constant sampling function" p

hpzq which is analytic in a neigh-
borhood of the line segment ηp1 (and the neighborhood’s Schwarz reflection), satisfying
hpzq “ p
p hpzq.
5. For each positive integer N , the normalization constants tcj uN
j“1 are chosen to satisfy

cj “ p
hpλj q, j “ 1, . . . , N .

6. There are two contours, Γ p 1 P C` and Γ p 2 P C´ , which are Schwarz reflections of each other
(and so they are determined by Γ p 1 ). The contour Γ p 1 is a simple, closed, analytic curve
that encircles the interval ηp1 .
We can now state the analog to theorem 2.6 for the non-symmetric case.
Theorem B.2. Consider RHP 2.1 under assumptions B.1, then for any x, t in a compact set,
the modulus of the N -soliton solution of the NLS equation (1.1) ψN px, tq is such that
ˆ ˙
1
|ψN px, tq ´ ψN,SG px, t, N q| “ O
p
log N

where ψpN,SG px, t; N q is the N -soliton gas condensate limit of the NLS equation under assump-
tions B.1 and is such that

ψpN,SG px, t; N q “ ψN,SG px ` 2pa1 ` a2 qt, t; N qe´ipa1 `a2 qpx`pa1 `a2 qtq ,
and ψN,SG px, t; N q is the N -soliton gas condensate solution described in Theorem 2.6.
B NON-SYMMETRIC CASE 49

Remark B.1. We notice that in this case the N -soliton condensate solution is an elliptic wave
with constant speed v “ 2pa1 ` a2 q.
Our stategy to prove the previous result is to link the solution of RHP 2.1 under the (sym-
metric) N -soliton condensate scattering data assumption (see Assumptions 2.2) to the non-
symmetric one (see Assumptions B.1).
Lemma B.3. Let M1 pz, x, tq, M2 pz, x, tq` be the solutions
˘ of RHP`2.1 under ˘assumptions 2.2 -
a2 ´a1 a1 `a2
B.1 respectively. If a “ 2 , ρpzq “ ρ̂ z ` 2 and hpzq “ ĥ z ` a1 `a2
2
then
ˆ ˙
a1 `a2 a1 ` a2 a1 `a2
M2 pz, x, tq “ e´i 2 px`pa1 `a2 qtqσ3 M1 z ´ , x ` 2pa1 ` a2 qt, t ei 2 px`pa1 `a2 qtqσ3 .
2

Proof. We start by considering RHP 2.1 under Assumptions B.1 and through a series of trans-
formations, we connect it to the one under Assumptions 2.2.
Define κ as
a1 ` a2
κ“ ,
2
and consider the following translation of the eigenvalues and of the variable z

λ
p j “ λj ´ κ , zp “ z ´ κ . (B.8)
Then, the RHP 2.1 becomes
RHP B.4. Let λ pN be N discrete eigenvalues in C` , and let c1 , . . . , cN be the correspond-
p1 , . . . , λ
ing norming constants, we look for M2 pp z , x, tq such that
• M2 p¨, x, tq P MatpC, 2q
• M2 pp
z , x, tq is analytic on the complex plane except at λ
p1 , . . . , λ
pN
z , x, tq satisfies the residue conditions
• M2 pp
„ ˆ ˙ȷ
0 0
resλp j M pp
z , x, tq “ lim M2 pp z , x, tq ,
cj e2iθpλj `κq 0
p
zpÑλ pj
« ˜ ¸ff
´2iθpλ
p j `κq
res p M2 pp z , x, tq “ lim M2 pp z , x, tq 0 ´cj e
λj 0 0
zpÑλpj

• M2 pp
z , x, tq „ I2 ` Op1{zq as z Ñ 8.
Let’s focus on the function θpλ
pj ` κq, defining x
p as

x
p “ x ` 4κt ,
one immediately obtains that

θpλ
pj ` κq “ θp
pλ pj q ` κpp
x ´ 2κtq . (B.9)
where θpzq
p “x pz ` 2tz 2 .
Using the previous equality (B.9), we can rewrite the two residue matrices as

ˆ ˙ ˆ ˙ ˜ ¸ ˜ ¸
0 0 0 0 0 ´cj e´2iθpλj `κq ´cj e´2iθpλj q´2ikppx´2κtq
“ 0
p p
“ ,
cj e2iθpλj `κq 2iθp
pλ x´2κtq
p j q`2ikpp
0 cj e 0
p
0 0 0 0
which can be further factorized as
REFERENCES 50

ˆ ˙ ˆ ˙
0 0 x´2κtqσ3
´iκpp 0 0 iκppx´2κtqσ3
“e e
cj e2iθpλj q`2iκppx´2κtq 0 cj e2iθpλj q 0
p p p p
˜ ¸ ˜ ¸
0 ´cj e´iθpλj q´ikppx´2κtq “ e´iκppx´2κtqσ3 0 ´cj e´iθpλj q
p p
eiκppx´2κtqσ3 .
0 0 0 0

Given the previous factorization, we define the function

M1 pp p, tq “ eiκppx´2κtqσ3 M2 pp
z, x z , x, tqe´iκppx´2κtqσ3 . (B.10)
Then, M1 pp p, tq solves the following RHP
z, x
RHP B.5. Let λ1 , . . . , λ
p pN be N discrete eigenvalues in C` , and let c1 , . . . , cN be the correspond-
ing norming constants, we look for M pp p, tq such that
z, x
• M1 p¨, x
p, tq P MatpC, 2q
p, tq is analytic on the complex plane except at λ
• M1 p¨, x p1 , . . . , λ
pN
• M
xppz, xp, tq satisfies the residue conditions
„ ˆ ˙ȷ « ˜ ¸ff
0 0 0 ´c e ´2iθp
pλ pj q
resλp j M z, x
xpp p, tq “ lim M z, x
xpp p, tq , res p M pp
x z, x
p, tq “ lim M pp
z, x
p, tq j
cj e2iθpλj q 0 λj
p p
zpÑλj zpÑλj 0 0

• M
xppz, x
p, tq „ I2 ` Op1{p z q as zp Ñ 8.
where θpzq
p “x pz ` 2tz 2 .
Furthermore, in view of the change of coordinates (B.8) and the assumptions of the Lemma,
Hypotheses B.1 are mapped to Hypotheses 2.2. Therefore, given (B.10) we proved that

M2 pz, x, tq “ e´iκpx`2κtqσ3 M1 pz ´ κ, x ` 2κt, tq eiκpx`2κtqσ3 ,

as claimed.

In particular, the previous Lemma and Theorem 2.6 imply Theorem B.2.

References
[1] J. Baik, T. Kriecherbauer, Kenneth D T, and P. D. Miller, Discrete orthogo-
nal polynomials. (AM-164), Annals of Mathematics Studies, Princeton University Press,
Princeton, NJ, Jan. 2007.
[2] M. Bertola, T. Grava, and G. Orsatti, Soliton Shielding of the Focusing Nonlinear
Schrödinger Equation, Phys. Rev. Lett., 130 (2023), p. 127201.
[3] M. Bertola, T. Grava, and G. Orsatti, B̄-problem for focusing nonlinear Schrödinger
equation and soliton shielding, arXiv preprint arXiv:2409.14825, (2024).
[4] D. Bilman, P. Nabelek, and T. Trogdon, Computation of large-genus solutions of
the Korteweg–de Vries equation, Physica D: Nonlinear Phenomena, 449 (2023), p. 133715.
[5] D. Bilman and T. Trogdon, On numerical inverse scattering for the Korteweg–de Vries
equation with discontinuous step-like data, Nonlinearity, 33 (2020), p. 2211–2269.
[6] M. Borghese, R. Jenkins, and K. D. T.-R. McLaughlin, Long time asymptotic
behavior of the focusing nonlinear Schrödinger equation, Ann. Inst. H. Poincaré C Anal.
Non Linéaire, 35 (2018), pp. 887–920.
[7] F. Carbone, D. Dutykh, and G. El, Macroscopic dynamics of incoherent soliton en-
sembles: Soliton gas kinetics and direct numerical modelling, Europhysics Letters, 113
(2016), p. 30003.
REFERENCES 51

[8] T. Congy, G. El, and G. Roberti, Soliton gas in bidirectional dispersive hydrodynamics,
Physical Review E, 103 (2021), p. 042201.
[9] P. Deift, T. Kriecherbauer, K. T. McLaughlin, S. Venakides, and X. Zhou,
Strong asymptotics of orthogonal polynomials with respect to exponential weights, Commu-
nications on Pure and Applied Mathematics, 52 (1999), p. 1491–1552.
[10] P. Deift, T. Kriecherbauer, K. T.-R. McLaughlin, S. Venakides, and X. Zhou,
Uniform asymptotics for polynomials orthogonal with respect to varying exponential weights
and applications to universality questions in random matrix theory, Communications on
Pure and Applied Mathematics, 52 (1999), p. 1335–1425.
[11] P. Deift and X. Zhou, A Steepest Descent Method for Oscillatory Riemann–Hilbert
Problems. Asymptotics for the MKdV Equation, The Annals of Mathematics, 137 (1993),
p. 295.
[12] NIST Digital Library of Mathematical Functions. https://dlmf.nist.gov/, Release 1.2.1
of 2024-06-15. F. W. J. Olver, A. B. Olde Daalhuis, D. W. Lozier, B. I. Schneider, R. F.
Boisvert, C. W. Clark, B. R. Miller, B. V. Saunders, H. S. Cohl, and M. A. McClain, eds.
[13] G. El and K. A, Kinetic equation for a dense soliton gas, Physical review letters, 95
(2005).
[14] G. El and A. Tovbis, Spectral theory of soliton and breather gases for the focusing
nonlinear Schrödinger equation, Phys. Rev. E, 101 (2020), p. 052207.
[15] G. Falqui, T. Grava, and C. Puntini, Shielding of breathers for the focusing nonlinear
Schrödinger equation, arXiv preprint arXiv:2412.16696, (2024).
[16] E. Gennady, Resolution of a shock in hyperbolic systems modified by weak dispersion,
Chaos: An Interdisciplinary Journal of Nonlinear Science, 15 (2005).
[17] M. Girotti, T. Grava, R. Jenkins, and K. D. T.-R. McLaughlin, Rigorous Asymp-
totics of a KdV Soliton Gas, Communications in Mathematical Physics, 384 (2021),
p. 733–784.
[18] M. Girotti, T. Grava, R. Jenkins, K. T. McLaughlin, and A. Minakov, Soli-
ton versus the gas: Fredholm determinants, analysis, and the rapid oscillations behind
the kinetic equation, Communications on Pure and Applied Mathematics, 76 (2023),
p. 3233–3299.
[19] M. Girotti, T. Grava, K. D. McLaughlin, and J. Najnudel, Law of Large Numbers
and Central Limit Theorem for random sets of solitons of the focusing nonlinear Schrödinger
equation, arXiv preprint arXiv:2411.17036, (2024).
[20] A. Gkogkou, B. Prinari, and T. Trogdon, Numerical inverse scattering transform for
the defocusing nonlinear Schrödinger equation with box-type initial conditions on a nonzero
background, arXiv preprint arXiv:2412.19703, (2024).
[21] A. Hasegawa, Optical soliton: Review of its discovery and applications in ultra-high-speed
communications, Frontiers in Physics, 10 (2022).
[22] H. Hiyane, T. Busch, and T. Fogarty, Quantum soliton-trains of strongly correlated
impurities in Bose-Einstein condensates, Physical Review Research, 6 (2024).
[23] J. A. Jenkins, Univalent Functions and Conformal Mappings, Springer-Verlag, 1958.
[24] R. Jenkins and A. Tovbis, Approximation of the thermodynamic limit of finite-
gap solutions of the focusing NLS hierarchy by multisoliton solutions, arXiv preprint
arXiv:2408.13700, (2024).
[25] S. Kamvissis, Kenneth D T, and P. D. Miller, Semiclassical soliton ensembles for
the focusing nonlinear Schrödinger equation, Annals of Mathematics Studies, Princeton
University Press, Princeton, NJ, Aug. 2003.
[26] E. Kengne, W.-M. Liu, and B. A. Malomed, Spatiotemporal engineering of matter-
wave solitons in Bose–Einstein condensates, Physics Reports, 899 (2021), p. 1–62.
REFERENCES 52

[27] A. Kuijlaars and A. Tovbis, On minimal energy solutions to certain classes of in-
tegral equations related to soliton gases for integrable systems, Nonlinearity, 34 (2021),
p. 7227–7254.
[28] A. B. J. Kuijlaars, K. R. McLaughlin, W. Van Assche, and M. Vanlessen,
The Riemann–Hilbert approach to strong asymptotics for orthogonal polynomials on [-1,1],
Advances in mathematics, 188 (2004), pp. 337–398.
[29] M. D. Maiden, N. K. Lowman, D. V. Anderson, M. E. Schubert, and M. A.
Hoefer, Observation of dispersive shock waves, solitons, and their interactions in viscous
fluid conduits, Physical review letters, 116 (2016), p. 174501.
[30] S. M. Mossman, G. C. Katsimiga, S. I. Mistakidis, A. Romero-Ros, T. M.
Bersano, P. Schmelcher, P. G. Kevrekidis, and P. Engels, Observation of dense
collisional soliton complexes in a two-component Bose-Einstein condensate, Communica-
tions Physics, 7 (2024), p. 163.
[31] J. Robert and K. D. T. McLaughlin, Semiclassical limit of focusing NLS for a family
of square barrier initial data, Communications on Pure and Applied Mathematics, (2014).
[32] G. Roberti, G. El, A. Tovbis, F. Copie, P. Suret, and S. Randoux, Numerical
spectral synthesis of breather gas for the focusing nonlinear Schrödinger equation, Physical
Review E, 103 (2021), p. 042205.
[33] Z. Shabat, Exact theory of two-dimensional self-focusing and one-dimensional self-
modulation of waves in nonlinear media, Sov. Phys. JETP, 34 (1972), pp. 118–134.
[34] K. Strebel, Quadratic Differentials, Springer-Verlag, 1984.
[35] P. Suret, A. Picozzi, and S. Randoux, Wave turbulence in integrable systems: nonlin-
ear propagation of incoherent optical waves in single-mode fibers, Optics Express, 19 (2011),
p. 17852.
[36] P. Suret, S. Randoux, A. Gelash, D. Agafontsev, B. Doyon, and G. El, Soliton
gas: Theory, numerics, and experiments, Physical Review E, 109 (2024), p. 061001.
[37] A. Tovbis and F. Wang, Recent developments in spectral theory of the focusing NLS
soliton and breather gases: the thermodynamic limit of average densities, fluxes and cer-
tain meromorphic differentials; periodic gases, Journal of Physics A: Mathematical and
Theoretical, 55 (2022), p. 424006.
[38] , Soliton Condensates for the Focusing Nonlinear Schrödinger Equation: a Non-Bound
State Case, Symmetry, Integrability and Geometry: Methods and Applications, (2024).
[39] T. Trogdon, S. Olver, and B. Deconinck, Numerical inverse scattering for the Ko-
rteweg–de Vries and modified Korteweg–de Vries equations, Physica D: Nonlinear Phenom-
ena, 241 (2012), p. 1003–1025.
[40] Zakharov, Kinetic equation for solitons, Sov. Phys. JETP, 33 (1971), pp. 538–540.
[41] V. E. Zakharov, What Is Integrability?, Springer Berlin Heidelberg, 1991.
[42] X. Zhou, The Riemann–Hilbert Problem and Inverse Scattering, SIAM Journal on Math-
ematical Analysis, 20 (1989), p. 966–986.