Sunteți pe pagina 1din 85

` degli studi del Salento

Universita
Facolt`a di Scienze MM.FF.NN
Corso di laurea in MATEMATICA
Anno Accademico 2008/2009

Initial-value problem for the


nonlinear Schr
odinger equation:
the Inverse Scattering Transform

Relatore

Candidata

Prof. Giorgio Metafune

Barbara Prinari

Lecce - April 27, 2010

Contents
Introduction

1 Solitons and NLS Solitons

2 Focusing NLS with decaying boundary conditions


2.1

13

Direct scattering problem . . . . . . . . . . . . . . . . . . . . . . . . 14


2.1.1

Jost solutions and integral equations . . . . . . . . . . . . . . 14

2.1.2

Scattering data . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.2

Inverse scattering problem . . . . . . . . . . . . . . . . . . . . . . . . 23

2.3

Time evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.4

Soliton solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3 Defocusing NLS with non-decaying boundary conditions


3.1

3.2

3.3

29

Direct scattering problem . . . . . . . . . . . . . . . . . . . . . . . . 30


3.1.1

On the analyticity of the Jost solutions . . . . . . . . . . . . . 35

3.1.2

Scattering data . . . . . . . . . . . . . . . . . . . . . . . . . . 45

Inverse scattering problem . . . . . . . . . . . . . . . . . . . . . . . . 61


3.2.1

Riemann- Hilbert problem via uniform coordinate . . . . . . . 61

3.2.2

Trace formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.2.3

Reflectionless solutions . . . . . . . . . . . . . . . . . . . . . . 70

Time evolution and soliton solutions . . . . . . . . . . . . . . . . . . 72

4 Conclusions

76

A A relevant lemma

78

Bibliography

82

Introduction
Many nonlinear evolution equations (NLEEs), namely nonlinear partial differential
equations (PDEs) of first order with respect to time, admit solitary waves, i.e. exponentially localized traveling wave solutions. For most equations, these waves do not
survive mutual interaction. Nonetheless, a special class of equations exists, usually
called integrable systems, for which solitary waves pass through one another nondestructively as if they were particles whence the name solitons. One of these
equations is the Korteweg-de Vries (KdV) equation: ut + 6uux + uxxx = 0 (subscripts x and t denote partial differentiation throughout). Another such equation
is the nonlinear Schrodinger (NLS) equation: iqt = qxx 2 |q|2 q, where = 1

and = 1 denote, respectively, the so-called focusing and defocusing dispersion


regimes. The class of integrable NLEEs includes many other systems, comprising
not only PDEs, but also ordinary differential equations (ODEs), nonlocal equations
and difference equations. Remarkably, these integrable NLEEs possess a rich and
beautiful mathematical structure. Each of these equations can be written as the
compatibility condition of a Lax pair, which is an overdetermined system of linear
ODEs containing an eigenvalue, or scattering parameter, and in which the solution
of the NLEE plays the role of a potential. The first half of the Lax pair is called
the scattering problem. In turn, the Lax pair is directly related to the existence
of a method, called the Inverse Scattering Transform (IST), that allows the initialvalue problem (IVP) for the integrable equation to be solved exactly. The direct
problem in the IST associates to the solution of an integrable NLEE the reflection
coefficients, discrete eigenvalues and norming constants of the scattering problem
collectively referred to as the scattering data. The inverse problem (that is,
the reconstruction of the solution of the NLEE in terms of the scattering data) then
reveals that the solitons are exactly the portion of the solution associated with the
3

discrete eigenvalues. Moreover, in a precise sense the IST is the analogue of the
Fourier transform for nonlinear problems, and reduces to it in the linear limit.
Other remarkable features of integrable NLEEs are: (i) They admit an infinite
number of conserved quantities in involution. (ii) They are completely integrable
infinite-dimensional Hamiltonian systems. In fact, the IST can be seen as a canonical
transformation to action-angle variables. (iii) They are a member of an infinite hierarchy of commuting flows. (iv) They possess a bi-Hamiltonian structure. (v) Exact
solutions exist that describe the elastic interactions of an arbitrarily large number
of solitons. Also, direct methods exist for generating these multi-soliton solutions.
(vi) Auto-Backlund transformations (called Darboux transformations) exist, some
of which have the effect of adding or subtracting one soliton to the solution.
The study of integrable systems has blossomed into an active and diverse field of
research, and several monographs are now dedicated to various aspects of it [1]-[6].
Many properties of integrable systems connect them to other branches of mathematics: (i) the characterization of the space of existence of solutions for these integrable
equations is naturally carried out within the framework of classical and functional
analysis; (ii) the inverse problem in the IST is typically formulated as a RiemannHilbert problem (RHP); (iii) the integrability of a NLEE is deeply connected to the
analyticity properties of its solutions; (iv) the Ablowitz-Ramani-Segur conjecture
states that a necessary condition for a NLEE to be integrable is that its solutions
possess the Painleve property, namely, that the only movable singularities are poles,
and the solution is single-valued around them (or, in other words, that the location
of any algebraic, logarithmic or essential singularities is independent of the initial
conditions); (v) at the heart of the various mathematical miracles of integrable
NLEEs is their relation with infinite-dimensional Kac-Moody algebras. Still other
properties and other connections exist. Importantly, the theory of solitons and integrable systems also allows one to obtain concrete information about the solutions of
physically relevant NLEEs, since, typically, equations solvable by IST are the result
of asymptotic reductions of fundamental equations of physics.

The nonlinear Schr


odinger equation
The nonlinear Schrodinger (NLS) equation that is the object of this thesis is a prototypical dispersive nonlinear PDE that has been derived in many areas of physics
and analyzed mathematically for over 40 years. Historically the essence of NLS-type
equations can be traced back to the 1950s, with the early work of Ginzburg and
Landau [7] and Ginzburg [8] and their study of the macroscopic theory of superconductivity, and also of Ginzburg and Pitaevskii [9] who subsequently investigated
the theory of superfluidity. Nonetheless, it was not until the works of Chiao et al
[10] and Talanov [11] that the wider physical importance of NLS equation became
evident, especially in connection with the phenomenon of self-focusing and the conditions under which an electromagnetic beam can propagate without spreading in
nonlinear media. In the general situation, an optical beam in a dielectric broadens
due to diffraction. However, in materials whose dielectric constant (or refractive
index) increases with the field intensity, the critical angle for internal reflection at
the beams boundary can become greater than the angular divergence due to diffraction and, as a consequence, the beam does not spread and can, in some situations,
continue to focus into extremely high intensity spots.
Beside the fact that NLS systems have direct applications in many physical
problems, the importance of the NLS equation is also due to its universal character
[12]. Generically speaking, most weakly nonlinear, dispersive, energy-preserving
systems give rise, in an appropriate limit, to the NLS equation. Specifically, the NLS
equation provides a canonical description for the envelope dynamics of a quasimonochromatic plane wave propagating in a weakly nonlinear dispersive medium
when dissipation can be neglected.
Mathematically, the NLS equation attains broad significance since, in one transverse dimension, it is integrable via the Inverse Scattering Transform (IST, for
brevity), it admits multisoliton solutions, it has an infinite number of conserved
quantities, etc. Solving the initial value problem for the NLS equation by means of
the IST will be the main object of this thesis. Below we give a general description
of the method. Chapter 1 provides a brief overview on NLS solitons. The details of
the IST as it applies to the specific case will be developed in Chapter 2 (NLS with
decaying boundary conditions) and Chapter 3 (NLS with non-decaying boundary
conditions).
5

The Inverse Scattering Transform


The IST is a method that allows one to linearize a class of nonlinear evolution
equations. An essential pre-requisite of the IST method is the association of the
nonlinear PDE with a pair of linear problems. We say that the operator pair X, T
is a Lax pair for the nonlinear equation
qt = F [x, t, q, qx , qxx , . . . ] ,

q = q(x, t)

[F is a nonlinear function of q and (some of) its spatial derivatives, and it may
or may not depend explicitly on the independent variables x, t] if the compatibility
condition of the over-determined linear system of differential equations
vx = Xv,

vt = T v

[i.e., the equality of the mixed derivatives vxt = vtx ] is identically satisfied provided
q solves the nonlinear PDE [13]. Here X, T are in general matrix functions of
q, qx , qxx , . . . and of a complex parameter k which is assumed to be time-independent.
The solution of the initial-value problem by IST proceeds in three steps, as
follows (cf. Fig.1):
(i) the direct problem - the transformation of the initial datum from the original physical variables (q(x, 0)) to the transformed scattering variables
(S(k, 0));
(ii) time dependence - the evolution of the transformed data often according to
simple, explicitly solvable evolution equations (i.e., finding S(k, t));
(iii) the inverse problem - the recovery of the evolved solution (q(x, t)) from the
evolved solution in the transformed variables (S(k, t)).
Both the direct and the inverse problem make use of the first operator in the
Lax pair, so-called scattering problem, while the time evolution is determined by
the second operator in the Lax pair.
In the direct problem, the first step is to construct eigenfunction solutions of
the associated linear problem vx = Xv. These eigenfunctions depend both on the
original spatial variable and on the spectral parameter (eigenvalue) k. Second, with
6

Figure 1: The scheme of the Inverse Scattering Transform for solving the initial
value problem for an integrable evolution equation.

these eigenfunctions, one determines scattering data S(k, 0) that are independent of
the original spatial variable. The time-evolution of the scattering data is determined
by the second operator in the Lax pair, vt = T v, and it usually amounts to solving
a simple linear ODE. In the inverse problem, the first step is the recovery of the
eigenfunctions from the (evolved) scattering data S(k, t). Finally, one recovers the
solution in the original variables q(x, t) from these (evolved) eigenfunctions.
The analyticity properties of the eigenfunctions as functions of the scattering
parameter k are key to the formulation of the inverse problem. For the NLS equation discussed above, when decaying boundary conditions are considered (i.e., q 0
rapidly enough as x ), the eigenfunctions are sectionally analytic functions of
the scattering parameter and the inverse problem is therefore formulated as a generalized Riemann-Hilbert problem (RHP with poles), which in turn is transformed into
a system of linear algebraic-integral equations. The problem is significantly more
involved for potentials that are not decaying as x , and it will be described
in Chapter 3.
It is important to mention that with the IST machinery one can do more than
solve the initial-value problem; one can also construct special solutions by positing
an elementary ansatz for the transformed variables S(k, 0) and then applying the inverse transformation to obtain the corresponding solution in the physical variables.
In general, the soliton and multisoliton solutions are constructed in this way. Also,
the IST provides an effective way to study the asymptotic (long-time) behavior of
the solutions (e.g., via the nonlinear steepest descent technique), as well as a way
to study their stability (using the so-called squared eigenfunctions).

While the IST for integrable systems with decaying boundary conditions is wellunderstood, in general a lot less is known regarding problems where the boundary
conditions play an important role. Examples of this kind of problems are: (i) initialvalue problems with non-decaying boundary conditions, in which the solution does
not tend to zero as x ; (ii) boundary value problems, where the spatial
domain is only a portion of the real line; (iii) (2+1)-dimensional soliton equations in
which the solutions do not vanish at infinity along certain directions. These kinds
of problems are notoriously more difficult, and many of them are still open. In some
cases, their solution requires the development of new mathematical methods.
This thesis will focus on the IST for the nonlinear Schrodinger equation in the defocusing dispersion regime under non-decaying boundary conditions, so as to include
dark/grey solitons (see Chapter 2). The plan of the work is as follows. Chapter 1
gives a description of NLS soliton solutions. In Chapter 2 we discuss the IST for the
focusing NLS with decaying boundary conditions, in order to establish the notations
and illustrate the concepts in a simpler case. Chapter 3 describes the IST for the defocusing NLS equation with non-decaying boundary conditions. Some conclusions,
perspectives and open problems will finally be given.

Chapter 1
Solitons and NLS Solitons
Localized (nonoscillatory) solitary waves have been known to wave dynamics researchers since the observation of the Scottish engineer John Scott Russell in 1834.
While conducting experiments to determine the most efficient design for canal boats,
he made a remarkable discovery, described in his Report on waves [14].
I was observing the motion of a boat drawn along a narrow channel by a pair
of horses, when the boat suddenly stopped - not so the mass of water in the channel.
It accumulated round the prow of the vessel in a state of violent agitation, then
suddenly leaving it behind, it rolled forward with great velocity, assuming the form
of a large solitary elevation, a rounded, smooth and well-defined heap of water,
which continued its course along the channel apparently without change of form or
diminution of speed. I followed it on horseback, and overtook it still rolling on at a
rate of some eight or nine miles an hour, preserving its original figure some thirty
feet long and a foot to a foot and a half in height. Its height gradually diminished,
and after a chase of one or two miles I lost it in the windings of the channel. Such,
in the month of August 1834, was my first chance interview with that singular and
beautiful phenomenon. [...]
It was not to be supposed that after its existence had been discovered and phenomena
determined, endeavors would not be made to reconcile it with existing theory, or to
show how it ought to have been predicted from the known equations of fluid motion.
In other words, it now remained for the mathematician to predict the discovery after
it had happened...
Despite Russells detailed observation (and further studies by Airy, Boussinesq,
9

Stokes), it was only in 1895 that Korteweg and de Vries, derived, as a model for
the evolution of long, unidirectional waves in shallow water, a nonlinear PDE (now
known as KdV equation) that approximately described the wave elevation [15].
In 1965, the pioneer paper by Zabusky and Kruskal showed that solitary wave
solutions of KdV interact elastically, i.e. they pass through one another and subsequently retain their characteristic form and velocity [16]. Because of this particle-like
property, they termed these solutions solitons.
There are several types of solitons. A bright soliton is one which vanishes well
away from its peak (center) point. On the contrary, dark/gray solitons tend to a
nontrivial background state away from their center point. Indeed, for the dark/gray
solitons, the center point is located at the minimum in amplitude, unlike the
bright case where the center point is a maximum.
The NLS equation
iqt = qxx 2 |q|2 q ,

(1.1)

where 2 = 1 and = 1 for the case of anomalous dispersion (also referred to


as focusing, or attractive NLS) and = 1 for the case of normal dispersion (defocusing or repulsive NLS) admit both these types of soliton solutions depending
on the dispersion regime.
The focusing NLS (Eq. (1.1) with = 1) admits bright solitons of the form
2 v 2 )t+

q(x, t) = a sech[a(x 2vt xo )] ei[vx+(a

o]

(1.2)

where v, a and xo , o are arbitrary real parameters. A bright soliton has the classical bell-shaped profile, rapidly decaying to zero away from its center point (cf.
Fig.1.1a).
As discussed earlier, the remarkable property of solitons is to emerge from a
collision preserving their form and velocity: Fig. 1.1b shows a two-soliton collision.
Note the position shift arising from the nonlinear interaction.
Gray solitons manifest themselves when the dispersion is in the normal regime.
In particular, classical dark solitons are solution of the de-focusing NLS (Eq. (1.1)
with = 1) of the form
2

q(x, t) = q0 e2iq0 t tanh q0 (x x0 )


10

(1.3)

A2

0.4
0.3
0.2
0.1

-4

-2

Figure 1.1: Left: A bright soliton, with A2 the square modulus of the solution,
the coordinate in the co-moving frame. Right: A two-soliton solution of the NLS
equation (in optics coordinates, where the retarded time t is the space-like variable
and the propagation distance z is the evolution variable).
with the beam intensity profile dropping to zero at its center and
2

q(x, t) q0 e2iq0 t

as

x .

More general soliton solutions with nontrivial boundary conditions are also possible.
These are the so-called gray solitons, in which the minimum of intensity is not zero:
2

q(x, t) = q0 e2iq0 t [ tanh (q0 (x 2q0 t x0 )) i] ,

2 + 2 = 1.

(1.4)

In order to make the comparison with solutions from the inverse problem easier,
we note that this solution can also be written as
2

q(x, t) = q0 e2iq0 t ( i)

( + i)2 + e2q0 (x2q0 tx0 )


.
1 + e2q0 (x2q0 tx0 )

(1.5)

Note that the factor i in the second line can be arbitrarily modified, provided
its modulus is kept equal to 1. As x
2

q(x, t) q (t) = q0 e2iq0 t [|| i] q0 ei+2iq0 t ,

|| i = ei

so that, denoting by the asymptotic phases at x , one has + = + 2iq02 t,


= + 2iq02 t and
+ i||
ei(+ ) =
.
(1.6)
i||
11

A2
1
0.8
0.6
0.4
0.2

-4

-2

Figure 1.2: A gray soliton, A2 being the square modulus of the solution, the
coordinate in the co-moving frame.

From the previous expression it is also evident that the asymptotic phases change
in time, but in such a way that their difference remains constant and given by (1.6)
in terms of the soliton amplitude (q0 ) and velocity (2q0 ).
Note that there are more general gray soliton solutions
q(x, t) = q0 ei(x+

2 t)+2i|q

0|

2t

tanh (2|q0 |(x + 2t) + x0 ).

(1.7)

The plane wave behavior can be eliminated by using the Galileian invariance of NLS,
namely by means of a transformation
q(x , t ) = q(x, t)ei(x+t)
with
x = x + 2t ,

t = t.

However, this procedure cannot be used if there is more than one soliton of this
type, with different wave numbers. Note that one could also consider potentials
with asymmetric boundary conditions in the amplitude, as well as in the phase, i.e.
q(x, t) q e2i|q |

2t

with, possibly, |q+ | =


6 |q | [31].

12

(1.8)

Chapter 2
Focusing NLS with decaying
boundary conditions
In this chapter we will illustrate the IST for the focusing NLS equation under the
assumption that the solution q(x, t) decays as x sufficiently rapidly (at least
q L1 , as we shall see). This problem was solved almost 40 years ago by Zakharov
and Shabat [17]. The purpose of including it here is to establish the notations,
illustrate the concepts in a simpler case, and to be able to compare and contrast
them with the case of non-decaying boundary conditions to be discussed in Chapter
3.
We start by formulating the IST procedure for the somewhat more general system
(cf. [18])
iqt = qxx 2rq 2

irt = rxx 2qr 2

(2.1a)
(2.1b)

and then consider the reductions r = q with = +1, 1, giving (1.1), as a special
case.
The Lax pair associated to the system (2.1) is given by
!
ik q
vx =
v,
r ik

13

(2.2a)

usually referred to as ZS-AKNS scattering problem, and


vt =

2ik 2 + iqr
2kr + irx

2kq iqx

2ik 2 iqr

(2.2b)

T
where v is a 2-component vector, v(x, t) = v (1) (x, t), v (2) (x, t) .1 The equality of
the mixed derivatives of v, i.e. vxt = vtx , is equivalent to the statement that q and
r satisfy the evolution equations (2.1a)- (2.1b), if k, the scattering parameter, is
independent of x and t. For this reason, the evolution equations (2.1) are sometimes
referred to as the compatibility condition of the equations (2.2a)-(2.2b). We refer
to the equation with the x derivative, Eq. (2.2a), as the scattering problem and the
equation with the t derivative, Eq. (2.2b), as the time dependence.
For future convenience, we rewrite the scattering problem (2.2a) as
vx = (ikJ + Q) v
where
J=

1 0
0 1

Q=

(2.3)

0 q
r 0

(2.4)

I will denote the 2 2 identity matrix.

2.1
2.1.1

Direct scattering problem


Jost solutions and integral equations

We refer to solutions of the scattering problem (2.2a) as eigenfunctions with respect


to the parameter k. When the potentials q, r 0 sufficiently rapidly as x
(in the following we will require q, r at least in L1 (R) with respect to x, for all t 0),
the eigenfunctions are asymptotic to the solutions of
!
ik 0
vx =
v
0 ik
1

Here and in the following, unless otherwise specified, superscripts () with = 1, 2 denote the
-th component of the corresponding two-component vector. Also, T denotes matrix transpose.

14

when |x| is large. Therefore, it is natural to single out the solutions of Eq. (2.2a)
(eigenfunctions) defined by the following boundary conditions:
!
!
1
0
k)
(x, k)
eikx ,
(x,
eikx
as x
(2.5a)
0
1
!
!
0
1
k)
(x, k)
eikx ,
(x,
eikx
as x +.
(2.5b)
1
0
Note that here and in the following the bar does not denote complex conjugate, for
which we will always use the notation. It is convenient to introduce eigenfunctions
with constant boundary conditions by defining
M(x, k) = eikx (x, k),
N(x, k) = eikx (x, k),

k)
(x, k) = eikx (x,
M
(x, k) = eikx (x, k).
N

(2.6a)
(2.6b)

k) are solutions
In terms of the notations (2.4), the Jost solutions M(x, k) and N(x,
of the differential equation
x (x, k) = ik (J + I) (x, k) + (Q) (x, k),

(2.7)

(x, k) satisfy
while the Jost solutions N(x, k) and M
x (x,
k) = ik (J I) (x,
k) + (Q)
(x, k),
with the constant boundary conditions
!
1

M(x, k)
,
M(x,
k)
0
!
0
k)
N(x, k)
,
N(x,
1

(2.8)

0
1

as x

(2.9a)

1
0

as x +.

(2.9b)

Solutions of the differential equations (2.7) and (2.8) can be represented by means
of the following integral equations
Z +
(x, k) = w +
G(x x , k)Q(x )(x , k) dx

Z +
x , k)Q(x )(x
(x, k) = w +
G(x
, k) dx

15

where w = (, 0)T , w = (0, )T and the (matrix) Greens functions G(x, k) and

G(x,
k) satisfy the differential equations
[Ix ik (J + I)] G(x, k) = (x),

[Ix ik (J I)] G(x,


k) = (x) ,
where (x) is the Dirac-delta (generalized) function. The Greens functions are
not unique, and, as we will show below, the choice of the Greens functions and
the choice of the inhomogeneous terms w, w together uniquely determine the Jost
solutions and their properties of analyticity in k.
By using the Fourier transform method, one can represent the Greens functions
in the form
!
Z
1/p
0
1
G(x, k) =
eipx dp
2i C
0 1/ (p 2k)
!
Z
1/
(p
+
2k)
0
1

eipx dp
G(x,
k) =
2i C
0
1/p
where C and C will be chosen as appropriate contour deformations of the real p-axis.
(x, k) defined by
It is natural to consider G (x, k) and G
!
Z
1
1/p
0
G (x, k) =
eipx dp
2i C
0 1/ (p 2k)
!
Z
1/
(p
+
2k)
0
1
k) =
Gx,
eipx dp
2i C
0
1/p
where C and C ) are the contours from to + that pass below (+ functions)
and above ( functions) both singularities at p = 0 and p = 2k, or respectively,
p = 0 and p = 2k. Contour integration then gives
!
!
e2ikx 0
1
0
,
G (x, k) = (x)
, (2.10)
G (x, k) = (x)
0 e2ikx
0
1
where (x) is the Heaviside function ((x) = 1 if x > 0 and (x) = 0 if x < 0). The
+ functions are analytic and bounded in the upper half-plane of k and the -
functions are analytic in the lower half-plane. By taking into account the boundary
16

conditions (2.9), we obtain the following integral equations for the Jost solutions:
! Z
+
1
M(x, k) =
+
G+ (x x , k)Q(x )M(x , k)dx
(2.11a)
0

! Z
+
0
+ (x x , k)Q(x )N(x , k)dx
N(x, k) =
+
G
(2.11b)
1

! Z
+
0

(x , k)dx
(x x , k)Q(x )M
M(x, k) =
+
G
(2.11c)
1

! Z
+
1

(x , k)dx .
N(x, k) =
+
G (x x , k)Q(x )N
(2.11d)
0

Eqs. (2.11) are Volterra integral equations, whose solutions can be sought for in
the form of Neumann series iterates. In the following Lemma we will show that if
q, r L1 (R), the Neumann series associated to the integral equations for M and
N converge absolutely and uniformly (in x and k) in the upper k-plane, while the
and N
converge absolutely and
Neumann series of the integral equations for M
uniformly (in x and k) in the lower k-plane. This implies that the Jost solutions
M(x, k) and N(x, k) are analytic functions of k for Im k > 0 and continuous for

k) are analytic functions of k for Im k < 0 and


Im k 0, while M(x,
k) and N(x,
continuous for Im k 0.
Lemma 1. If q, r L1 (R), then M(x, k), N(x, k) defined by (2.11a) and (2.11b)
(x, k)
are analytic functions of k for Im k > 0 and continuous for Im k 0, while M
(x, k) defined by (2.11c) and (2.11d) are analytic functions of k for Im k < 0
and N
and continuous for Im k 0. Moreover, the solutions of the corresponding integral
equations are unique in the space of continuous functions.
Proof. We prove the result for M(x, k). The proofs are similar for the remaining
eigenfunctions. The Neumann series
M(x, k) =

j=0

where
m0 (x, k) =

1
0

,
17

mj (x, k)

(2.12)

mj+1 (x, k) =

G+ (x x , k)Q(x )mj (x , k)dx

is, formally, a solution of the integral equation (2.11a). Taking into account the
explicit expression of G+ (x, k) and Q(x) and looking separately at each component,
we can write
Z x
(1)
(2)
q(x )mj (x , k)dx
mj+1 (x, k) =
Z
x

(2)
(1)
mj+1 (x, k) =
e2ik(xx ) r(x )mj (x , k)dx .

(2)

Because m0 = 0, for any non-negative integer j one has


(1)

(2)

m2j+1 (x, k) 0,

m2j (x, k) 0.

Moreover, one can check that for any f L1 (R) the following identities hold:
1
j!

|f ()|

Z

|f ( )| d

j

1
d =
(j + 1)!
1
=
(j + 1)!

Z x

d
d

Z

|f ( )| d

|f ()| d

j+1

j+1

(2.13)

Then one can show by induction that for Im k 0, such that |e2ik(xx ) | 1 for any
x x ,
hR
ij hR
ij+1
x
x



|q(x
)|
dx
|r(x
)|
dx

(2)

m2j+1 (x, k)
j!
(j + 1)!
hR
ij hR
ij
x
x



|q(x
)|
dx
|r(x
)|
dx

(1)

.
m2j (x, k)
j!
j!

Therefore, if q, r L1 (R) and Im k 0, the Neumann (2.12) series is dominated


in norm by a uniformly convergent series and it is itself uniformly convergent. It
follows that M(x, k) is analytic for Im k > 0 (a uniformly convergent series of analytic functions converges to an analytic function). For the same reason, M(x, k) is
continuous for Im k 0.
A proof of the uniqueness of the solution for the integral equation (2.11a) can be
obtained as follows. Let us suppose that M1 (x, k) and M2 (x, k) both solve the
18

integral equation. Then their difference m(x, k) = M1 (x, k) M2 (x, k) is such that
Z

Z x
(1)
2ik( )

(1)

m (x, k) =
q()
e
r( )m ( , k)d d

Z x
(2)
m (x, k) =
e2ik(x) r()m(1) (, k)d.

The first equation yields the bound


Z
Z x
(1)

m (x, k)
|q()|


(1)


|r( )| m ( , k) d d.

Assume that m(1) (x, k) is bounded, i.e., there exists a real and positive constant C


such that m(1) (x, k) C. Iterating the above bound n times we then obtain
R
n R
n

|q()|
d
|r()|
d
(1)

m (x, k) C
.
n!
n!

Since the right-hand side tends to zero as n , m(x, k) is identically zero and
the solution of the integral equation (2.11a) is unique in the space of continuous
functions. 
Note that simply requiring q, r L1 (R) does not yield analyticity of the eigenfunctions on the real k-axis, for which more stringent conditions must be imposed.
Having q, r vanishing faster than any exponential as |x| implies that all four
eigenfunctions are entire functions of k C.

From the integral equations (2.11) we can compute the asymptotic expansion at
large k (in the proper half-plane) of the eigenfunction. Integration by parts yields
!
Rx
1

1 2ik
q(x
)r(x
)dx

M(x, k) =
+ O(1/k 2 )
(2.14a)
1
2ik r(x)
!
R +
1

1
+
q(x
)r(x
)dx
2ik x
k) =
N(x,
+ O(1/k 2)
(2.14b)
1
2ik r(x)
!
1
q(x)
2ik
R +
N(x, k) =
+ O(1/k 2)
(2.14c)
1

1 2ik x q(x )r(x )dx


!
1
q(x)
(x, k) =
R x2ik
M
+ O(1/k 2).
(2.14d)
1

1 + 2ik q(x )r(x )dx


19

2.1.2

Scattering data

The two eigenfunctions with fixed boundary conditions as x are linearly


independent, and the same holds for the two eigenfunctions with fixed boundary
T
conditions as x +. Indeed, let u(x, k) = u(1) (x, k), u(2) (x, k) and v(x, k) =
T
v (1) (x, k), v (2) (x, k) be any two solutions of (2.2a) and let us define the Wronskian
of u and v, Wr(u, v), as
Wr(u, v) = u(1) v (2) u(2) v (1) .

(2.15)

Using (2.2a) it is easy to verify that


d
Wr(u, v) = 0
dx

(2.16)

and therefore, from the prescribed asymptotic behavior for the Jost solutions, it
follows


k) = 1
Wr , = lim Wr (x, k), (x,
(2.17a)
x


k) = 1
Wr , = lim Wr (x, k), (x,
(2.17b)
x+

which proves that the solutions and are linearly independent, as are and
As a consequence, we can express (x, k) and (x,
k) as linear combinations of
.
k), or vice-versa, with the coefficients of these linear combinations
(x, k) and (x,
depending on k only. Hence, the relations
k)
(x, k) = b(k)(x, k) + a(k)(x,
k) = a
k)
(x,
(k)(x, k) + b(k)(x,

(2.18a)
(2.18b)

hold for any k C such that the four eigenfunctions exist. In particular, (2.18) hold
for Im k = 0 and define the scattering coefficients a(k), a
(k), b(k), b(k) for k R.
as x with Eqs. (2.18) shows that
Comparing the asymptotics of Wr(, )
the scattering data satisfy the following characterization equation:
a(k)
a(k) b(k)b(k) = 1

k R.

(2.19)

The scattering coefficients can, in turn, be represented as Wronskians of the Jost


solutions. Indeed, from Eqs. (2.18) it follows that
)

a(k) = Wr(,

a(k) = Wr(, ),
20

(2.20a)


b(k) = Wr ,
.

),
b(k) = Wr(,

(2.20b)

Therefore, if q, r L1 (R), Lemma 1 and Eqs. (2.20) imply that a(k) admit analytic
continuation in the upper k-plane, while a(k) can be analytically continued in the
lower k-plane. In general, b(k) and b(k) cannot be extended off the real k-axis. It
is also possible to obtain integral representations for the scattering coefficients in
terms of the eigenfunctions, but since it is not necessary for our present purposes
(no such representation is known in the case of non-decaying boundary conditions),
we will skip the derivation.
Eqs. (2.18) can be written as
k) + (k)e2ikx N(x, k)
(x, k) = N(x,
k)

(x, k) = N(x, k) + (k)e2ikx N(x,

(2.21a)
(2.21b)

where we introduced
(x, k) = M(x, k)/a(k),

(x, k)/

(x, k) = M
a(k)

(2.22)

(k) = b(k)/
a(k).

(2.23)

and the reflection coefficients


(k) = b(k)/a(k),

Note that from the representation of the scattering data as Wronskians of the eigenfunctions and from the asymptotic expansions (2.14), it follows
a(k) = 1 + O(1/k),

a
(k) = 1 + O(1/k)

(2.24)

in the proper half-plane, while b(k), b(k) are O(1/k) as |k| on the real axis.
Proper eigenvalues and norming constants
A proper (or discrete) eigenvalue of the scattering problem (2.2a) is a (complex)
value of k corresponding to a bounded solution v(x, k) such that v(x, k) 0 as
x (usually one requires v L2 (R) with respect to x). Suppose that kj =
j + ij (with j > 0) is such that a(kj ) = 0. Then from (2.20a) it follows that
Wr((x, kj ), (x, kj )) = 0 and therefore j (x) := (x, kj ) and j (x) := (x, kj ) are
linearly dependent, that is, there exists a complex constant cj such that
j (x) = cj j (x).
21

(2.25)

Hence, by (2.5a) and (2.5b) it follows that


!
1
j (x)
ej xij x x
as x
0
!
0
j (x) = cj j (x)
ej x+ij x x
as x +,
1
and therefore kj is a proper eigenvalue. On the other hand, if a(k) 6= 0, then any
solution of the scattering problem blows up in one or both directions. We conclude
that the proper eigenvalues in the region Im k > 0 are precisely the zeros of the
scattering coefficient a(k). Similarly, the eigenvalues in the region Im k < 0 are
kj ) =
given by the zeros of a
(k), and these zeros kj = j + i
j are such that (x,

J
kj ) for some complex constant cj . The coefficients {cj }
cj (x,
cj }Jj=1 are
j=1 and {
called norming constants. In terms of the Jost solutions, the norming constants are
defined by
j (x) = cj e2ikj x N
j (x) ,
Mj (x) = cj e2ikj x Nj (x) ,
M
(2.26)
j (x) = M(x,

j (x). As
where Mj (x) = M(x, kj ), M
kj ) and similarly for Nj (x) and N
we shall see in the following, discrete eigenvalues and associated norming constants
are part of the scattering data (i.e., necessary to uniquely solve the inverse problem
and reconstruct q(x, t) and r(x, t)).
Note that if the potentials q, r are rapidly decaying, such that Eqs. (2.18) can
be extended off the real axis, then
cj = b(kj )
cj = b(kj )

for j = 1, . . . , J

for j = 1, . . . , J.

Symmetry reductions
The NLS equation is a special case of the system (2.1) under the symmetry reduction
r = q . This symmetry in the potentials induces a symmetry between the Jost
solutions analytic in the upper k-plane and the ones analytic in the lower k-plane.
In turn, this symmetry of the Jost solutions induces symmetries in the scattering
data.
T
Indeed, if v(x, k) = v (1) (x, k), v (2) (x, k) satisfies equation (2.2a) and r = q ,

then v(x, k) = v (2) (x, k ), v (1) (x, k ) (where denotes conjugate transpose) also
22

satisfies the same equation (2.2a). Taking into account the boundary conditions
(2.5), we get
!
!
(2)

(2)

(x,
k
)

(x,
k
)
k) =
k) =
(x,
,
(x,
,
(2.27a)
(1) (x, k )
(1) (x, k )
!
!
(2)

(2)

N
(x,
k
)
M
(x,
k
)
k) =
(x, k) =
N(x,
,
M
.
(2.27b)
N (1) (x, k )
M (1) (x, k )
Then, from the Wronskian representations for the scattering data (2.20) it follows
b(k) = b (k ) ,

a
(k) = a (k ) ,

(2.28)

which implies that kj is a zero of a(k) in the upper half k-plane if and only if kj is
a zero of a
(k) in the lower k-plane and vice-versa. As a consequence, the zeros of
a(k) and a(k) are paired, their number is the same (i.e., J = J) and
kj = kj ,

2.2

cj = cj

j = 1, . . . , J.

(2.29)

Inverse scattering problem

The inverse problem consists of constructing a map from the scattering data, that
is: (i) the reflection coefficients (k) and (k) for k R, defined by (2.23), (ii)
the discrete eigenvalues {kj }Jj=1 (zeros of the scattering coefficient a(k) in the upper
 J
half plane of k) and kj
(zeros of the scattering coefficient a
(k) in the lower half
j=1

plane of k), and (iii) the norming constants {cj }Jj=1 and {
cj }Jj=1, cf. (2.26); back to
the potentials q and r. First, we use these data to reconstruct the eigenfunctions
(x, k)) and then we recover the potentials from the large
(for instance, N(x, k) and N
k asymptotics of the eigenfunctions, cf. Eqs. (2.14). Note that the inverse problem
is solved at fixed t, and therefore the explicit time-dependence is omitted. In fact,
in the inverse problem both x and t are treated as parameters. If the data are the
evolved data, i.e., data at time t > 0, then the reconstruction through Eqs. (2.14)
yields q(x, t) and r(x, t).
k)
In the previous section, we showed that the eigenfunctions N(x, k) and N(x,
exist and are analytic in the regions Im k > 0 and Im k < 0, respectively, if
q, r L1 (R). Similarly, under the same conditions on the potentials, the functions
23

(x, k) and
(x, k) introduced in (2.22) are meromorphic in the regions Im k > 0
and Im k < 0, respectively, with poles at the zeros of a(k) and a(k). Therefore, in
the inverse problem we assume these analyticity properties for the unknown eigen (x, k)) or modified eigenfunctions ((x, k) and
functions (N(x, k) and N
(x, k)).
With these assumptions, Eqs. (2.21) can be considered as the jump conditions of
a Riemann-Hilbert problem. To recover the sectionally meromorphic functions from
the scattering data, we will convert the Riemann-Hilbert problem into a system of
linear integral equations.
Suppose that a(k) and a
(k) have a finite number of simple zeros in the regions
Im k > 0 and Im k < 0, respectively, which we denote as {kj , Im kj > 0}Jj=1 and

J
. We shall also assume that a() 6= 0 and a() 6= 0 for R (i.e.,
kj , Im kj < 0
j=1

a and a have no zeros on the real axis).

Let f (), R, be an integrable function and consider the projection operators


Z +
1
f ()

P [f ](k) =
d.
(2.30)
2i (k i0)
If f+ (resp. f ) is analytic in the upper (resp. lower) k-plane and f (k) 0 as
|k| for Im k 0, then
P [f ] = f ,

P [f ] = 0

[whence P are projection operators into the upper/lower half k-planes]. Let us now
apply the projector P to both sides of Eq. (2.21a) and P + to both sides of (2.21b).
Taking into account the analyticity properties of N, N, ,
and the asymptotics
(2.14) and using (2.26), we obtain
!
Z +
J C e2ikj x
P
1
1
()e2ix
j
N(x, k) =
+
Nj (x) +
N(x, )d (2.31a)
2i (k i0)
0
j=1 k kj
!
Z +
J C
j e2ikj x
P
0
1
()e2ix

N(x, k) =
+
N
(x)

N (x, )d (2.31b)
j

2i (k + i0)
1
j=1 k kj
j (x) = N(x,
kj ) and we introduced
where Nj (x) = N(x, kj ), N
Cj =

cj

a (k

Cj =

cj

a
(kj )

j)

24

for j = 1, . . . , J

(2.32a)

for j = 1, . . . , J

(2.32b)

with denoting derivative of a(k) and a


(k) with respect to k. Note that the equations
k) depend on the extra terms
defining the inverse problem for N(x, k) and N(x,


j (x) J . In order to close the system, we evaluate (2.31a) at


{Nj (x)}Jj=1 and N
j=1
thus obtaining
k = kj for j = 1, . . . , J and (2.31b) at k = kj for j = 1, . . . , J,
!
Z +
J C e2ikj x
P
1
()e2ix
1
j

N (x) =
+
N(x, )d
(2.33a)
N
(x)
+
j

2i k
0
j=1 k kj
!
Z +
J C
m e2ikm x
P
0
()e2ix
1

Nj (x) =
+
N
(x)

N (x, )d .(2.33b)
j

2i
kj
1
m=1 kj km
Eqs. (2.31) and (2.33) together constitute a linear algebraic-integral system of equations which, in principle, solve the inverse problem for the eigenfunctions N(x, k)
k).
and N(x,
By comparing the asymptotic expansions at large k of the right-hand sides of
(2.31) with the expansions (2.14), we obtain
Z
J
P
1 +
(2)
2ikj x
()e2ix N (2) (x, )d (2.34a)
r(x) = 2i
e
Cj Nj (x) +

j=1

+
J
P

(1) (x) + 1
(1) (x, )d (2.34b)
q(x) = 2i
e2ikj x Cj N
()e2ix N
j

j=1

which reconstruct the potentials in terms of the scattering data and thus complete
the formulation of the inverse problem (as before, the superscript () denotes the
-component of the corresponding vector).

We mention that the issue of establishing existence and uniqueness of solution


for the equations of the inverse problem is usually carried out by converting the
inverse problem into a set of Gelfand-Levitan-Marchenko integral equations. This
problem has also been considered, in a more general context but always in the L1
class (i.e., for decaying potentials), in [19, 20].

2.3

Time evolution

We will now show how to determine the time dependence of the scattering data.
Then, by the method of Sec. 2.2, one will be able to establish the time evolution of
q and r.
25

The operator (2.2b) determines the evolution of the Jost solutions, which can be
written as
!
A B
t v =
v
(2.35)
C A
where B, C 0 as x (since q, r L1 (R)). Then the time-dependent
eigenfunctions must asymptotically satisfy the differential equation
!
A
0
t v =
v
as x
(2.36)
0 A
with
A = lim A(x, k) = 2ik 2 .
|x|

(2.37)

The system (2.36) has solutions that are linear combinations of


!
!
A t
0
v+ =
,
v =
.
0
eA t
However, such solutions are not compatible with the fixed boundary conditions of
the Jost solutions, i.e. Eqs. (2.5a)-(2.5b). Therefore, we define time-dependent
functions
(x, t; k) = eAt (x, t; k) ,
(x, t; k) = eAt (x, t; k) ,

t; k)
t; k) = eA t (x,
(x,
t; k)

(x,
t; k) = eA t (x,

(2.38a)
(2.38b)

to be solutions of the differential equation (2.35). Then, the evolution for and
becomes
!
!
A A
B
A
+
A
B

t =
,
t =
(2.39)
C
A A
C
A + A
so that, taking into account Eqs. (2.18) and evaluating (2.39) as x +, we obtain
t a = 0,

t a
=0

t b = 2A b,

tb = 2iAb

or, explicitly,
a(k, t) = a(k, 0),

a
(k, t) = a
(k, 0)
26

(2.40a)

b(k, t) = b(k, 0)e4ik t ,

b(k, t) = b(k, 0)e4ik2 t .

(2.40b)

From (2.40a) it follows that the discrete eigenvalues (i.e., the zeros of a and a
)
are constant as the solution evolves. Not only the number of eigenvalues, but also
their locations are fixed. Thus, the eigenvalues are time-independent discrete states
of the evolution. In fact, this time invariance is the underlying mechanism of the
elastic soliton interaction for the NLS equation and in integrable soliton equations
in general. On the other hand, the evolution of the reflection coefficients (2.23) is
given by
2
2
(k, t) = (k, 0)e4ik t ,
(k, t) = (k, 0)e4ik t
(2.41)
and this also gives the evolution of the norming constants:
2
Cj (t) = Cj (0)e4ikj t .

Cj (t) = Cj (0)e4ikj t ,

(2.42)

The expressions for the evolution of the scattering data allow one to solve the initialvalue problem for the NLS. The procedure is as follows: (i) The scattering data are
calculated from the initial data q(x) = q(x, 0) and r(x) = r(x, 0) at time t = 0
according to the procedure described in Sec. 2.1; (ii) the scattering data at a later
time t > 0 are determined from Eqs. (2.40)-(2.42); (iii) the solutions q(x, t) and
r(x, t) are recovered from the scattering data using, for instance, the reconstruction
formulas (2.31)-(2.34).

2.4

Soliton solutions

In the case where the scattering data comprise proper eigenvalues but (k) = (k)
0 for all k R (corresponding to the so-called reflectionless solutions), the algebraicintegral system (2.31) and (2.33) reduces to a simple linear algebraic system, namely
!
J C e2ikj x N (x)
P
1
j
j
l (x) =
N
+

kl kj
0
j=1
!
J C
m e2ikm x N
m (x)
P
0
Nj (x) =
+
,
kj km
1
m=1
that can be solved in closed form. The one soliton solution, in particular, is obtained
for J = J = 1 (i.e., one single discrete eigenvalue k1 = + i and corresponding
27

norming constant C1 ). In the relevant physical case, when the symmetry r = q


holds, using (2.28) and (2.29) in the above system, we get
"
#
2 2i(k1 k1 )x 1
C1
|C
|
e

1
=
,
e2ik1 x 1
k1 k1
(k1 k1 )2
"
#
2 2i(k1 k1 )x 1
|C
|
e
1
(2)
N1 (x) = 1
(k1 k1 )2

(1)
N1 (x)


T
(1)
(2)
where, as usual, N1 (x) = N1 (x), N1 (x) . Then, if we set
e2 =

k1 = + i,

|C1 |
,
2

from (2.34) it follows that


q(x) = 2i

C1 2ix
e
sech (2x 2) .
|C1 |

Taking into account the time-dependence of C1 as given by (2.42), one finally gets
the well-known bright soliton solution of the NLS equation
q(x, t) = 2ie2i+4i(

2 2

)ti0 sech [2 (x 4t) 2 ]


0

with
C1 (0) = 2e20 +i0 .
The comparison with (1.2) shows that from a spectral point of view v + ia represents
the discrete eigenvalue of the scattering problem (in particular, the real part of the
discrete eigenvalue determines the soliton velocity, and its imaginary part fixes the
soliton amplitude), and xo , o , related to the center of the soliton and its overall
phase, correspond to modulus and argument of the associated norming constant.

28

Chapter 3
Defocusing NLS with
non-decaying boundary conditions
As discussed in Chap. 1, in the normal dispersion regime the NLS equation only
admits dark/gray soliton solutions, which are not decaying as x . In order to
solve the initial value problem for a meaningful class of solutions that also includes
solitons, one is therefore forced to consider non-decaying boundary conditions.
The IST for the defocusing scalar NLS equation with non-decaying boundary
conditions as x was first proposed by Zakharov and Shabat [21] (see also
[4] and [22]-[29]). However, many important issues (among which establishing the
analyticity of the eigenfunctions with respect to the scattering parameter k, or
formulating and solving the inverse problem in terms of a Riemann-Hilbert problem)
have never been addressed. These will be the goals of this chapter.
In the following we consider the scattering problem (2.3), i.e.
vx = (ikJ + Q) v
where
J=

1 0
0 1

(3.1)

Q=

0 q
r 0

as

(3.2)

with a matrix potential Q such that


Q Q (t) =

0
q (t)
r (t)
0
29

x .

(3.3)

The second operator in the Lax pair, as before, is given by


!
2ik 2 + iqr
2kq iqx
vt =
v.
2kr + irx
2ik 2 iqr

(3.4)

The boundaries can, in principle, be asymmetric, i.e. q+ , r+ need not to be equal


to q , r (cf. [31]). However, the spectral properties of the operator (3.1) depend
on the behavior of the potentials at large |x| and the problem with non decaying
boundary conditions significantly simplifies if q+ (t)r+ (t) = q (t)r (t) q02 , where
q0 R (and positive, for definiteness) and = 1 or = 1 in the focusing and
de-focusing cases, respectively.
As we shall see in the following, when = 1 (de-focusing case) and nonzero
boundary conditions are imposed, the continuous spectrum of the operator (3.1)
acquires branch points. In the fully asymmetric case, when q+ (t)r+ (t) = |q+ |2
and q (t)r (t) = |q |2 with |q+ |2 6= |q |2 , it has 4 branch points corresponding to
k 2 |q |2 = 0. Since in the de-focusing regime r = q , the scattering operator
is self-adjoint and a finite number of non degenerate discrete eigenvalues will be
located on the real axis, outside the continuous spectrum.
After the IST for the defocusing NLS equation was first addressed in [21], it
was shown that in order to apply correctly the method proposed by Zakharov and
Shabat, the NLS equation must be renormalized by adding a linear term [22, 4].
Such renormalization is not necessary if one adopts the method suggested in this
work.

3.1

Direct scattering problem

As mentioned above, in this Chapter we will assume r = q and


q (t) = q0 ei (t) ,

r (t) = q0 ei (t)

(3.5)

where q0 > 0 and 0 < 2. The constant q02 plays the role of density and are
called asymptotic phases. The boundary conditions (3.5) are said to be satisfied in
Schwartz sense if q q is of Schwartz type in the vicinity of , i.e. if q(x, t) is a
smooth function such that limx x x |q(x, t) q (t)| = 0 for any , N.
30

The boundary conditions (3.5) are compatible with the time evolution in the
sense that q0 and = + are time independent.
As x , the eigenfunctions of the scattering problem (3.1) asymptotically
satisfy the first order system of ODEs
!
!
!
v1
ik q (t)
v1
=
v2
r (t) ik
v2
x

or the second order ODE


x2 vj = (k 2 q02 )vj

j = 1, 2.

Each of the two equations has the two linearly independent solutions eix and eix
where
q
= k 2 q02 .
(3.6)
Therefore it is natural to introduce (for real k such that |k| q02 and real )
the eigenfunctions defined as the solutions of (3.1) with the following boundary
conditions:
(x, k) weix ,
(x, k) veix ,

k) we
(x,
ix

as x

(3.7a)

k) veix
(x,

as x +.

(3.7b)

In order to satisfy asymptotically the first order system (3.1), the boundaries have
to be chosen in such a way that
(iI + ikJ + Q ) w = 0,

(iI + ikJ + Q+ ) v = 0

and
(iI + ikJ + Q+ ) v = 0,

(iI + ikJ + Q ) w = 0

where I denotes the 2 2 identity matrix and with Q given in (3.3). Consider, for
instance, the homogeneous system corresponding to w, i.e.
( k)w1 iq w2 = 0

ir w1 + ( + k)w2 = 0.

31

The system has zero determinant, so of course it admits nontrivial solutions. There
are two quite natural choices
!
!
+k
iq
w=
or
w=
.
ir
k
Obviously, the two vectors are proportional to each other. The equation defining v
can be obtained from the previous ones by the exchange q , r q+ , r+ , whence
the possible choices for v are
!
!
+k
iq+
v =
or
v =
.
ir+
k
Similarly, one would have
v=

k
ir+

w
=

k
ir

or

or

v=

iq+
+k

w =

iq
+k

and could take any two pairs.


Following, for instance, [17] and [4], we will fix the boundary conditions as follows
!
!
+k
iq
w=
,
w =
(3.8a)
ir
+k
!
!
iq+
+k
v=
,
v =
(3.8b)
+k
ir+
but mention that the choice that more easily can be generalized to a multicomponent
NLS [32] corresponds to
!
!
+k
k
w=
,
w =
ir
ir
!
!
+k
k
.
,
v =
v=
ir+
ir+

32

In the following analysis, it is convenient to consider functions with constant (i.e.


x-independent) boundary conditions. Hence, we define the Jost solutions as follows
k),
(x, k) = eix (x,
M
k)
(x, k) = eix (x,
N

M(x, k) = eix (x, k),


N(x, k) = eix (x, k),

(3.9a)
(3.9b)

such that
M(x, k)

+k
ir

N(x, k)

iq+
+k

,
,

(x, k)
M

iq
+k

(3.10a)

k)
N(x,

+k
ir+

x +.

(3.10b)

The Jost solutions can be represented by means of the following integral equations
Z +
M(x, k) = w +
G (x x , k) (Q(x ) Q ) M(x , k)dx
(3.11a)

Z +
N(x, k) = v +
G+ (x x , k) (Q(x ) Q+ ) N(x , k)dx
(3.11b)

Z +
(x, k) = w +
(x , k)dx
(x x , k) (Q(x ) Q ) M
M
G
(3.11c)

Z +

+ (x x , k) (Q(x ) Q+ ) N
(x , k)dx
N (x, k) = v +
G
(3.11d)

are solutions of the following differential equawhere the Greens functions G , G


tions
(Ix i(I + kJ iQ )) G (x, k) = (x)I
(x, k) = (x)I ,
(Ix i(I + kJ iQ )) G

(3.12a)
(3.12b)

(x) being the Diracs delta function. The Greens functions are not unique, and,
as we show below, the choice of the Greens function (and the choice of the inhomogeneous terms in the integral equations above) determines the Jost solutions and
their analytic properties. In this work, we plan to derive integral equations for the
Jost solutions and to establish decay conditions on q(x) q , q(x) q+ as x
such that the Jost solutions are analytic functions of the spectral parameter k in
appropriate regions of the Riemann surface of 2 = k 2 q02 .
33

Using the Fourier transform method, i.e. looking for a solution in the form
Z
1
G (x, k) =
eipx G (p, k) dp
2i
and substituting into (3.12a), it is easy to find
1
G (p, k) =
p(p 2)

pk
iq
ir
p+k

Assume 6= 0 (i.e., k not at a branch point). Then the matrix G (p, k) has two
distinct poles, both on the real p-axis, i.e. p = 0 and p = 2 if k (, q0 )
(q0 , ). Therefore, it is natural to define
!
Z
pk
iq
1
eipx
dp
G (x, k) =
2i C p(p 2)
ir
p+k
where C is a contour on the complex p-plane indented so as to passes below both
poles p = 0 and p = 2 along arbitrarily small semi-circles (cf. Fig. 3.1).
Im p

p=0

C-

p=2
o

Re p

Figure 3.1: The contour C obtained indenting around the poles p = 0 and p = 2
.

Contour integration then yields




1 + e2ix
e2ix 1
G (x, k) = (x)
I+
(ikJ + Q )
2
2i
34

(3.13)

or explicitly

e2ix 1
q

2i 2ix
G (x, k) = (x)
.
e
1
1 2ix
(e
+ 1) + ik
2
2i
(3.14)
Note that this matrix Greens function is also well-defined in the limit 0.
Indeed, as 0, one has
1 2ix
(e
2

e2ix 1
+ 1) ik
2i
e2ix 1
r
2i



G (x, k) (x) I + x(ikJ + Q )

for k q0 .

Analogously, one would have

(x, k) = 1
G
2i

(p, k) dp
eipx G

where
1
(p, k) =
G
p(p + 2)

p+k
iq
ir
p++k

and C is the contour passing below both poles at p = 0 and p = 2, i.e.




2ix
2ix
1
+
e
1

e
(x, k) = (x)
G
I+
(ikJ + Q ) .
2
2i

(3.15)

The remaining Greens functions are obtained in a similar way, by taking contours
passing above the singularities at p = 0 and p = 2


1 e2ix
1 + e2ix
G+ (x, k) = (x)
I+
(ikJ + Q+ )
(3.16)
2
2i


1 + e2ix
e2ix 1

G+ (x, k) = (x)
I+
(ikJ + Q+ ) .
(3.17)
2
2i
Note that
(x, k) = e2ix G (x, k).
G

3.1.1

(3.18)

On the analyticity of the Jost solutions

All Greens functions, as functions of k and , have been defined so far for k R,
|k| > q0 . One expects to be able to prove that they can be analytically extended
35

to either one or the other sheet of an appropriate Riemann surface associated to


2 = k 2 q02 . More precisely, once an appropriate Riemann surface of the function
(k) has been introduced, such that on one sheet Im 0 and on the other one
Im 0, the functions M(., k) and N(., k) can be analytically continued on the
k) and N(.,
k) can be
sheet of the Riemann surface where Im 0, while M(.,
analytically continued on the other sheet, where Im 0. Indeed, as far as M(x, k)
and N(x, k) are concerned, they are defined through integral equations whose kernel
has terms of the type (x)e2ix and (x)e2ix respectively, hence we expect Im
0 to be a necessary condition for the integrals to converge. Therefore, we will fix a
branch for the function (k) (k q0 )1/2 (k + q0 )1/2 as follows.
We introduce the local polar coordinates
k q0 = r1 ei1
k + q0 = r2 ei2

0 1 < 2
2 <

and the magnitudes r1 , r2 uniquely fixed by the location of the point k : r1 = |k q0 |,


r2 = |k + q0 | (cf. Fig. 3.2). Then we define
(k) = (r1 r2 )1/2 ei(1 +2 )/2 .

(3.19)

If = (1 +2 )/2, then varies continuously between 0 and both in the upper and
in the lower k-planes and Im 0 (cf. Appendix); so this branch corresponds to
the upper sheet of the Riemann surface, and has a cut in the region (, q0 )
(q0 , ). Conversely, on the lower sheet of the Riemann surface we define
(k) = (r1 r2 )1/2 ei(1 +2 )/2

(3.20)

and this will give Im 0 (cf. Appendix), and again a branch cut in the region
(, q0 ) (q0 , ).

In the following, we will therefore work on the genus 0 two-sheeted Riemann


surface R of the algebraic function
2 = k 2 q02

(3.21)

formed by gluing two copies C1 and C2 of the extended complex plane C {}, cut
along the curves (, q0 ), (q0 , +). The positive sides of the cuts on C1 are glued
with the negative sides of the curves on C2 and vice versa (see Fig. 3.3a). Such a
36

Im k

.k
r2

r1

1
=
=0

=0

q0

q0

Re k

Figure 3.2: A branch cut for = (k 2 q02 )1/2 in the complex k plane. Here =
(1 + 2 )/2.
0I

Sheet I: Im >0

Im k<0 (back)

Sheet I: Im >0

Im k>0

cI

--

b II

8 8

a II
--

d II

a II

q0

d II

Im k<0 (front)

8 8

c II

--

aI

Im k>0 (back)
c II

0 II q 0

-- q0

Im k>0

-- q 0

bI

Sheet II: Im <0

b II

dI
q0

Im k<0

-+

--

-- q0

dI

aI

q0

0I

-- q 0

cI

8 8

bI

+
--

Im k>0 (front)

Im k<0

Sheet II: Im <0

0 II

Figure 3.3: Left: The two-sheeted Riemann surface R. Right: The genus 0 surface
topologically equivalent to R.

37

two-sheeted Riemann surface is topologically equivalent to a sphere (cf. Fig. 3.3b.)


Then the function becomes single-valued on the surface R:
(
p
1 (k) = k 2 q02
k C1
p
(k) =
2 (k) = k 2 q02
k C2
p
p
where k 2 q02 is the branch defined in (3.19) (such that Im 0), and k 2 q02
corresponds to the branch in (3.20), for which Im 0. When not otherwise clear
from the context, we will denote a point on the surface R with affix k on C1 by the
pair (k, 1 (k)), and its counterpart on C2 by (k, 2 (k)).
Remark 1. From now on, we will always think of k as being complex, on the appropriate sheet (which will be specified when not otherwise clear from the context),
and for the limiting values on R/[q0 , q0 ] one would have to specify from which side
(kIm 0) the limit is being taken. Note that, by virtue of the way the two cuts are
glued together, taking the limit kIm 0+ on one sheet corresponds to kIm 0 on
the other one and vice-versa.
Neumann series
In order to establish the analytic continuation of the Jost solutions in the appropriate
sheets of the Riemann surface, the starting point is given by the defining integral
equations. We already observed that the Greens functions defining M(x, k) and
N(x, k) are analytic and exponentially decaying at the infinite limit of integration
provided (k, ) C1 , i.e., on the upper sheet of the Riemann surface, while the
(x, k) and N
(x, k) are analytic and exponentially deGreens functions defining M
caying in the lower sheet of the Riemann surface. These are obviously only necessary
conditions to establish the analyticity of the respective solutions. Consider the integral equation defining M(x, k), and let m(x, k) = M(x, k)/2. We will prove that
m(, k) can be analytically continued on the upper sheet of the Riemann surface.
Then this will establish analyticity of M(, k) as well, apart from a pole at infinity (to be expected, given the asymptotic behavior of M(x, k) as x ) and,
possibly, the branch points of , i.e. q 0.
We define the solution m(x, k) through a Neumann series:
m(x, k) =

j=0

38

mj (x, k)

where
1
m0 (x, k) =
2

+k
ir

and the other terms in the series are defined iteratively as follows
Z x
mj+1 (x, k) =
G (x x , k) (Q(x ) Q ) mj (x , k) dx ,

with G (x, k) given by (3.13). Note that when q0 = 0, Q 0 and k. In this


case, m0 = (1, 0)T and
!
Z x
0
q(x )
mj+1 (x, k) =
mj (x , k) dx .
2ik(xx )
r(x )e
0

Let x = (x1 , x2 ) be a two component vector, and define


kxk = |x1 | + |x2 | .
For any 2 2 matrix A, one then has
kA xk

2
P

i,j=1

(3.22)

|aij | kxk .

Thus, if Im k 0, so that |e2ik(xx ) | 1 for any x x , we have


Z x
kmj+1 (x, k)k
(|q(x )| + |r(x )|) kmj (x , k)k dx

and this can be used to obtain a bound on kmj+1 k proceeding by induction as in


Lemma 1. Note first that ||m0 || 1. Also, if
Z x
j
1

kmj (x, k)k


(|q(x )| + |r(x )|) dx
,
j!

then the same set of identities as in Lemma 1 gives


Z x
j+1
1

kmj+1 (x, k)k


(|q(x )| + |r(x )|) dx
(j + 1)!

which completes the induction and allows us to conclude that the Neumann series
for m(x, k) converges absolutely and uniformly x R and Im k 0, as long as
q, r L1 (R).
39

Let us now consider the general case, i.e. Q 6= 0 and 6= k. Recall Im 0 on

the upper sheet of the Riemann surface. Thus, on the upper sheet, |e2i(xx ) | 1
for any x x and we have

2q0
2i(xx )

Z x
z }| {
e
1

kmj+1 (x, k)k


2+
2 |k| + |q | + |r |
2

(|q(x ) q | + |r(x ) r |) kmj (x , k)k dx .

In the following Lemma 2 we will show that


2ix

e

1
(|k| + q0 ) K0 + K1 |x |
2 +

with K0 , K1 independent of x and k. As a consequence, we get


Z x
kmj+1 (x, k)k
(K0 + K1 |x x |) (|q(x ) q | + |r(x ) r |) kmj (x , k)k dx .

Then, if
1
kmj (x, k)k
j!

Z

(K0

K1

|x x |) (|q(x ) q | + |r(x ) r |) dx

j

with K = max {K , km0 k} for = 1, 2, the same argument as above yields


1
kmj+1 (x, k)k
(j + 1)!

Z

(K0

K1

|x x |) (|q(x ) q | + |r(x ) r |) dx

j+1

We conclude that the Neumann series for m(x, k) converges absolutely x R on


the upper sheet of the Riemann surface as long as q q and r r are in L11 (, a)
for all a R, where
 Z b

1
L1 (a, b) = f :
(1 + |x|) |f (x)| dx < .
a

A similar construction can be applied to the other eigenfunctions.


Lemma 2. There exist constants K0 and K1 such that
2ix


e

1
(|k| + q0 ) K0 + K1 |x | .
2 +

40

Proof. First note that as 0 (i.e., as k q0 )


2i(xx )

e

1

(|k| + q0 ) 4q0 |x x |

and therefore by continuity there exists a > 0 such that for any with || <
2i(xx )

e

1

(|k| + q0 ) < 4q0 (1 + ) |x x | .

Also, if || (recall Im 0 and x x)

q

2i(xx )
||2 + q02 + q0
e
|k| + q0
1

=
(|k| + q0 ) 2 || 2

||
q

q

2
2
2
=2
1 + q0 / || + q0 / 2
1 + q0 / 2 + q0 / .
Thus, we have the desired result, with
K1 = 4q0 (1 + )


q
2
2
K0 = max 2, 2
1 + q0 / + q0 /
.
Re-normalization approach
An alternative approach to establish the analyticity of the eigenfunctions takes advantage of the re-normalization of the potentials suggested in [33]. Since analyticity
needs to be established at fixed time, say, t = 0, we can, without loss of generality, omit the time dependence. Let us introduce new smooth potentials q(x) and
r(x) = q (x) such that
q(x) q ,

q(x)
r (x) = q02

r(x) r

x R

(3.23a)
(3.23b)

i.e., a function q(x) with constant modulus and whose asymptotic phases are prescribed. Following [33], we define the matrices
!
i
q (x) k
(3.24a)
U1 (x; k, ) =
k i
r (x)
41

and also
U2 (x; k, ) =

+k k
i
r (x) i
r (x)

(3.24b)

By construction, both matrices asymptotically diagonalize the scattering problem


(3.1) and one can easily check that
!
1
i
r
(x)

k
U1
(3.25a)
1 (x; k, ) =
2 ( k) k i
q (x)
and
1
U1
2 (x; k, ) =
2i
r (x)

i
r (x)
k
i
r (x) ( + k)

(3.25b)

The determinant of U1 is independent of x, and U1


1 is well-defined for all k except
at the branch points of . On the other hand, due to (3.23b), the determinant of
U2 is nonzero for all x R and for all k except at the branch points of . Let us
now introduce
v = U1 v1 ,
v = U2 v2
with v a solution of (3.1), so that v1 and v2 satisfy the scattering problems
j vj ,
x vj = L

j = U1 LUj U1 x Uj
L
j
j

for j = 1, 2 ,

(3.26)

with L = ikJ + Q the original scattering operator. Let us compute the modified
j .
scattering operators L
U1
1 LU1 =
1
=
2 ( k)

i ( k) [22 + q(x)r(x) + r(x)q(x) 2q02 ] 2k ( k) r(x) + ( k)2 r(x) + r2 (x)q(x)


2k ( k) q(x) + ( k)2 q(x) + q2 (x)r(x) i ( k) [22 + 2q02 q(x)r(x) r(x)q(x)]

and
1
U1
1 x U1 =
2 ( k)

r (x)
qx (x) i ( k) rx (x)
i ( k) qx (x)

q (x)
rx (x)

Now notice that from (3.23b) it follows qx /


q =
rx /
r and therefore
h(x) =

1
q(x)
1 qx
1 rx
log

.
4 x
r(x)
2 q
2 r

(3.27a)

Let us also introduce


f (x) =


1
q(x)r(x) + r(x)q(x) 2q02
2
42

(3.27b)

g(x) =

1
[q(x)
r (x) r(x)
q (x)]
2

(3.27c)

and note that


f, g, h 0

as x .

1 the
Then, after a few manipulations, we obtain for the new scattering operator L
following expression:
1 = iJ + W
L
(3.28)
with
if (x) ( + k)h(x)

r (x) [g(x) + kf (x) + iq02 h(x)] /q02


q(x) [g(x) kf (x) iq02 h(x)] /q02
if (x) + ( + k) h(x)
(3.29)
In this form, the scattering operator is such that all potentials f, g, h are decaying as
x . The price to pay, though, is that the potential matrix W also depend on
k [a situation known in the literature as energy-dependent potentials]. However,
all matrix entries are bounded functions of k away from the branch points of ,
which allows us to establish analyticity of the corresponding eigenfunctions.
1 is given by
The integral equation for the modified eigenfunction associated to L

1
W(x, k) =

1
0

m(x, k) =

1
0
2i(xx )
0 e

W(x , k)m(x , k)dx

and we can define a solution through a Neumann series

m(x, k) =

mj (x, k)

j=0

with
m0 (x, k) =

1
0

mj+1 (x, k) =

1
0
2i(xx )
0 e

W(x , k)mj (x , k)dx .

Therefore, since on the upper sheet of the Riemann surface Im 0, we have


!
Z x
2
P
||mj+1 (x, k)||
|Wi (x , k)| ||mj (x , k)||1 dx ,

i,=1

43

the norm being the one defined in (3.67). Now notice that all entries of the energy
dependent potential matrix W are bounded functions of k away from the branch
points of . In fact, if |k q0 | for some small > 0, we have
1
1
,
||

| + k|
1 + max {1, 1/} ,
||

|k|
max {1, 1/}
||

and therefore if we introduce L = 1 + max {1, 1/},




2
P
1
|Wi | 2L |f | + 2 |h| + 2 (|f | + |g|) .
q0
i,=1

Hence, the same estimates as in Lemma 1 and in the previous section guarantee
that, provided f, g, h L1 (R), then m(x, k) (and consequently M(x, k)) can be
analytically continued on the entire upper sheet of the Riemann surface, except,
possibly, at the branch points of . The results for the remaining eigenfunctions can
obviously be proved analogously.
2 introduced in (3.26) can be calculated in a similar
The scattering operator L
way. One has
U1
2 LU2 = iJ +

if (x)
i [g(x) + kf (x)] / ( + k)
i [g(x) kf (x)] / ( k)
if (x)

and
h(x)
U1
2 x U2 =

k
( k)
( + k)
+k

so that
if (x) + ( k) h(x)
i [g(x) + kf (x) + iq02 h(x)] / ( + k)
i [g(x) kf (x) iq02 h(x)] / ( k)
if (x) + ( + k) h(x)
(3.30)
In this case, however, it does not seem possible to uniformly bound all matrix entries
with respect to k, due to the factors 1/( k) in the off-diagonal terms. It is worth
mentioning that
!
i( + k)/
q (x) 0
U2 = U1 ,
=
(3.31)
0
1

2 = iJ+ 1
L

and therefore
2 = 1 L
1 1 x ,
L
44

which we can write as


with



2 = 1 L
1 (x )1
L
(x )1 =

2h(x) 0
0
0

1 is uniformly bounded in k (except at the branch points), and so is obviNow L


2 is not. However, the corresponding
ously the additional term (x )1 , while L
eigenfunctions are proportional to each other, which suggests that, in order to prove
analyticity of the eigenfunctions through a Neumann series approach starting from
an analytic (or meromorphic, since the asymptotic eigenvectors can have simple
poles at infinity) initial value, it is sufficient to prove that the entries of the scattering operator are uniformly bounded in k up to a similarity transformation. In other
words, uniform boundedness of the energy dependent potential matrix with respect
to k, is clearly sufficient, but it does not seem to be a necessary condition for the
eigenfunctions to be meromorphic functions of the spectral parameter.

3.1.2

Scattering data

The two eigenfunctions with fixed boundary conditions as x are linearly independent, as are the two eigenfunctions with fixed boundary conditions as x +.
T
Indeed, if u(x, k) = u(1) (x, k), u(2) (x, k) and v(x, k) = (v (1) (x, k), v (2) (x, k))T are
any two solutions of (3.1), they satisfy
d
Wr(u, v) = 0
dx
where the Wronskian of u and v, Wr(u, v), is defined as usual:
Wr(u, v) = u(1) (x, k)v (2) (x, k) u(2) (x, k)v (1) (x, k).
From the asymptotics (3.7a)-(3.7b) it follows


k) = 2( + k)
Wr , = lim Wr (x, k), (x,
x


k) = 2( + k).
Wr , = lim Wr (x, k), (x,
x+

(3.32)

(3.33)

(3.34a)
(3.34b)

For real k and real these Wronskians are non zero provided 6= 0 (i.e., k = q0 ),

which proves that the functions , are linearly independent, and so are and ,
only the branch points being excluded.
45

Remark 2. Note that in the previous equations there is an abuse of notations.


because, given the way the cuts on the Riemann surface are introduced, the Jost
functions are discontinuous for |k| > q0 . However, the meaning of the previous
equations is the following. Take a value of k on aI and consider the function at
that point, (k + i), with an arbitrarily small > 0 (cf. Fig. 3.3). Since aI is
i) and
glued to dII , the corresponding eigenfunction to consider would be (k
one can check by looking at the large x asymptotics and taking for the value that
the square root assumes on aI (which is the same it takes on dII , i.e. the positive
p
arithmetic square root, k 2 q02 ) that indeed in the limit 0+
i0)) = 2( + k)
Wr((k + i0), (k

and the same holds for values on any point above/below the cut.
Since the two sets of eigenfunctions are linearly independent, we can write (x, k)
k) as linear combinations of (x, k) and (x,
k), or vice-versa. The coeffiand (x,
cients of these linear combinations depend on k. Hence, the relations
k i)
(x, k + i) = b(k + i)(x, k + i) + a(k + i)(x,
k + i) = a(k + i)(x, k i) + b(k + i)(x,
k + i)
(x,

(3.35a)
(3.35b)

hold for any k such that all four eigenfunctions exist (in other words, is arbitrarily
small, either positive or negative, and it is meant to identify on which edge of the
cut one is considering eigenfunctions and scattering data). Note that if the values
of the Jost solutions at the edges of the cut are considered as limiting values from
the appropriate sheet, one has to take into account the way the cuts are connected,
meaning that they will have the same k R and opposite sign of for functions
analytically continued on different sheets. In the following, with this in mind we
will omit writing explicitly.1

Comparing the asymptotics of Wr , as x with Eqs. (3.35a)(3.35b)
shows that the scattering data satisfy the following characterization equation
a(k)
a(k) b(k)b(k) = 1.
1

(3.36)

Suppose, for instance that in Eq. (3.35a) k bI (cf. Fig. 2a), i.e. > 0 and k lies in the upper
half plane of the upper sheet, second quadrant, just above the cut. Then, for we have to consider
the value at the corresponding point on cII , i.e. with < 0 and k lies in the lower half plane of the
lower sheet. Eq. (3.35a) then defines the values of a(k) and b(k) precisely at this point, belonging
to both bI and cII .

46

The scattering coefficients can be represented as Wronskian of the Jost solutions.


Indeed, from Eqs. (3.35a)(3.35b) it follows
Wr((k), (k))
,

Wr((k),
(k))


Wr (k), (k)
b(k) =
,

Wr((k),
(k))

a(k) =

Wr((k),
(k))

Wr((k),
(k))


Wr
(k),
(k)
b(k) =

Wr((k),
(k))

a(k) =

(3.37a)
(3.37b)

or, taking into account (3.34a)(3.34b)


Wr((k), (k))
,
2( + k)


Wr (k), (k)
,
b(k) =
2( + k)

a(k) =

Wr((k),
(k))
2( + k)


b(k) = Wr (k), (k)


2( + k)

a
(k) =

(3.38a)
(3.38b)

where takes on the value that pertains to the edge of the cut one is considering.
Note that from (3.38a) and from the results of Sec. 3.1.1 on the analyticity of
the eigenfunctions, it follows that a(k) can be analytically continued where and
are analytic, i.e. on the upper sheet, except, possibly, at the branch points k = q0 ,
and a
(k) can be analytically continued on the lower sheet, again with the possible
exception of the branch points.
Remark 3. Note that in the decaying case there is also an alternative way for relating the scattering coefficients to the eigenfunctions (for instance, cf. [1]). Namely,
one can consider the components of the equations defining the scattering coefficients,
(1) (x, k) = b(k) (1) (x, k) + a(k)(1) (x, k)
(2) (x, k) = b(k)(2) (x, k) + a
(k) (2) (x, k)

(3.39a)
(3.39b)

or, equivalently, in terms of the modified eigenfunctions (3.9),


(1) (x, k)
M (1) (x, k) = b(k)N (1) (x, k)e2ikx + a(k)N
(2) (x, k) = b(k)N
(2) (x, k)e2ikx + a
M
(k)N (2) (x, k).

(3.40a)
(3.40b)

Then one can take the limit as x + on both sides of the two equations and use
that in the decaying case the constant boundaries are such that N (1) (x, k) 0 and
(2) (x, k) 0, while N (2) (x, k) 1 and N
(1) (x, k) 1 as x +, to obtain
N
a(k) = lim (1) (x, k)eikx
x+

47

(3.41a)

a
(k) = lim (2) (x, k)eikx .
x+

(3.41b)

These relations allow the analytic extension of the scattering coefficients in the proper
half-plane, under the assumption that the limits with respect to x of the analytic
k) are still an analytic function of k.
functions (, k) and (,
In the non-decaying case this procedure has to be conveniently modified as follows.
Taking into account the boundary conditions (3.10b), we have
!
!
!
i+
(1) (x, k)eix
iq
e

+
k
0
b(k)e2ix
+ a(k)
as x +
(2) (x, k)eix
+k
iq0 ei+
hence, if we multiply the first component by + k, the second one by iq0 ei+ iq+ ,
add them up and use (3.6), we obtain


2( + k)a(k) = lim eix ( + k)(1) (x, k) + iq+ (2) (x, k) .

(3.42a)



2( + k)
a(k) = lim eix ir+ (1) (x, k) + ( + k)(2) (x, k) .

(3.42b)

x+

Since both components of are analytic on the upper sheet of the Riemann surface,
this relation allows to analytically extend also a(k) on the same sheet. Similarly,
one can show that
x+

For future convenience we rewrite Eqs. (3.35a) and (3.35b) as


(x, k) + (k)e2ix N(x, k)
(x, k) = N
k)

(x, k) = N(x, k) + (k)e2ix N(x,

(3.43a)
(3.43b)

where
(x, k) = M(x, k)/a(k),

(x, k)/

(x, k) = M
a(k)

(3.44)

(k) = b(k)/
a(k).

(3.45)

and the reflection coefficients are defined as


(k) = b(k)/a(k),

Introducing the 2 2 matrices m+ (x, k) = ((x, k), N(x, k)) (meromorphic in the

k),
upper sheet of the Riemann surface) and m (x, k) = N(x,
(x, k) (meromorphic in the lower sheet), the jump conditions (3.43) can be written as
m+ (x, k) m (x, k) = m (x, k)V(x, k)
48

(3.46)

where
V(x, k) =

(k)
(k)
(k)e2ix

(k)e2ix
0

(3.47)

Note that (3.46) can be looked at as a canonical Riemann-Hilbert problem on


the sphere in Fig. 3.2, in the following sense: m+ is a (matrix) function which is
defined and meromorphic on the upper hemisphere, m is a meromorphic on the
lower hemisphere, and (3.46) specifies the jump across the equator.
Behavior at large k of eigenfunctions and scattering data
In order to determine the asymptotic behavior at large k of the eigenfunctions, we
first note the following. On the upper sheet of the Riemann surface (Im 0), one
has:
above the cut (kIm > 0)
+ k 2k + O(1),

q02 1
+ o(1/k)
2k

|k| , Im > 0, kIm > 0


(3.48a)

and below the cut (kIm < 0)


q02 1
+ o(1/k),
+k
2k

k 2k + O(1)

|k| , Im > 0, kIm < 0.


(3.48b)
On the other hand, on the lower sheet of the Riemann surface (Im 0) one has:
above the cut (kIm > 0)
+k

q02 1
+ o(1/k),
2k

k 2k + O(1)

|k| , Im < 0, kIm > 0


(3.48c)

and below the cut (kIm < 0)


+ k 2k + O(1),

q02 1
+ o(1/k)
2k

|k| , Im < 0, kIm < 0.


(3.48d)
The asymptotic behavior at large k for the Jost solutions can be obtained from the
integral equations (3.11a)(3.11d). Indeed, for instance from (3.11b) it follows
!
!
N1 (x, k)
iq+
=
N2 (x, k)
+k
49

1
2

!
( k) N2 (x , k) (q(x ) q+ ) iq+ (r(x ) r+ ) N1 (x , k)
dx
ir+ N2 (x , k) (q(x ) q+ ) + ( + k) (r(x ) r+ ) N1 (x , k)
!
( + k) N2 (x , k) (q(x ) q+ ) + iq+ (r(x ) r+ ) N1 (x , k)

e2i(xx ) dx .

ir+ N2 (x , k) (q(x ) q+ ) + ( k) (r(x ) r+ ) N1 (x , k)

If one assumes
N1 (x, k)
N2 (x, k)

(x)
( + k)(1 + (x))

as |k| ,

substituting into the right hand side yields


!
!
(x)
iq+

( + k)(1 + (x))
+k
!
Z
(2 k 2 )(1 + (x )) (q(x ) q+ ) iq+ (r(x ) r+ ) (x )
1

dx
2 x
ir+ ( + k)(1 + (x )) (q(x ) q+ ) + ( + k) (r(x ) r+ ) (x )
!
Z
( + k)2 (1 + (x )) (q(x ) q+ ) + iq+ (r(x ) r+ ) (x )
1

e2i(xx ) dx + . . .

2 x
ir+ ( + k)(1 + (x )) (q(x ) q+ ) + ( k) (r(x ) r+ ) (x )
!
!
Z
iq+
q02 (1 + (x )) (q(x ) q+ ) iq+ (r(x ) r+ ) (x )
1
=

dx
2 x
+k
ir+ ( + k)(1 + (x )) (q(x ) q+ ) + ( + k) (r(x ) r+ ) (x )
!
Z
( + k)2 (1 + (x )) (q(x ) q+ ) + iq+ (r(x ) r+ ) (x )
1

e2i(xx ) dx + . . .

2 x
ir+ ( + k)(1 + (x )) (q(x ) q+ ) + ( k) (r(x ) r+ ) (x )
where . . . indicate terms of higher order with respect to the preceding one. Integrating by parts the last term yields
!
iq+
rhs
+k
!
Z
1
q02 (1 + (x )) (q(x ) q+ ) iq+ (r(x ) r+ ) (x )

dx
2 x
ir+ ( + k)(1 + (x )) (q(x ) q+ ) + ( + k) (r(x ) r+ ) (x )
!
( + k)2 (1 + (x )) (q(x) q+ ) + iq+ (r(x) r+ ) (x)
1 1
+
+ ....
2 2i ir+ ( + k)(1 + (x )) (q(x) q+ ) + ( k) (r(x) r+ ) (x)
and therefore we see that, on the upper sheet, if kIm > 0, in the lower component
the first term provides the leading order contribution (hence (x) = 0), while in the
upper component of the last term one has also an O(1) contribution, hence
(x) = iq+ i (q(x) q+ ) iq(x),
50

kIm > 0.

On the other hand, if kIm < 0 the first term provides the leading order contribution
in the first component, hence (x) = iq+ , while in the second component one has
a contribution also from the last term, i.e.


k
k
( + k) i
(r(x) r+ ) (x) = ( + k) 1 q+ (r(x) r+ ) 2
42
4 ( + k)


q+
( + k) 1 + 2 (r(x) r+ )
q0
and consequently
r(x) r+
,
kIm < 0.
r+
We conclude that as |k| on the upper sheet of the Riemann surface

iq(x)
+
o(1)

kIm > 0

+ k + O(1)

ix

!
(x, k) = v (x, k)e ,
v (x, k) =
.

iq
+
o(1)
+

kIm < 0

( + k) r(x) + o(1/|k|)
(x) =

r+

(3.49a)
Let us now consider the integral equation (3.11a), which we can write in component
form
!
!
M1 (x, k)
+k
=
M2 (x, k)
ir
!
Z x

( + k)M2 (x , k)(q(x ) q ) + iq M1 (x , k)(r(x ) r )


1
+
dx

ir M2 (x , k)(q(x ) q ) + ( k)M1 (x , k)(r(x ) r )


2
!
(

k)
M
(x,
k)(q(x)

q
)

iq
M
(x,
k)(r(x)

r
)
1
2

+ h.o.t.
2
4i
ir M2 (x, k)(q(x) q ) + ( + k) M1 (x, k)(r(x) r )
where h.o.t. denote higher order terms in k as |k| (note that the last term

comes from an integration by parts of the term with e2i(xx ) ). Assuming


M1 (x, k) ( + k) (1 + (x))
M2 (x, k) (x)

and substituting in both sides of the previous equation we can determine (x), (x).
Indeed, the same procedure as above yields
(x) 0,

(x) = ir(x)
51

kIm > 0

(x) =

q(x) q
,
q

(x) = ir

kIm < 0

and so, as |k| on the upper sheet of the Riemann surface,

+
k
+
O(1)

kIm > 0

ir(x) + o(1)

ix

!
(x, k) = w (x, k)e
,
w (x, k)
.
q(x)

(
+
k)
+
o(1/|k|)

kIm < 0

ir + o(1)
(3.49b)
In a similar way one obtains

!
q(x)

(
+
k)
+
o(1/|k|)

q+

kIm > 0

ir+ + o(1) !

ix

(x, k) = v (x, k)e


,
v (x, k)

+ k + O(1)

kIm < 0

ir(x) + o(1)
(3.49c)

iq + o(1)

kIm > 0

( + k) r(x)
+ o(1/|k|)
r

ix

k) = w (x, k)e ,
!
(x,
w (x, k)
.

iq(x)
+
o(1)

kIm < 0

+ k + O(1)
(3.49d)
In order to obtain the asymptotic behavior of the scattering data a(k), a
(k), instead of the Wronskian representations, it is more convenient to use the alternative
representation (3.42a), i.e.




1
ix (1)
ix (2)
a(k) =
( + k) lim e (x, k) + iq+ lim e (x, k) .
x+
x+
2( + k)
Assuming that is possible to interchange the limits as |k| and x +, and
taking into account (3.49b), we obtain



1

( + k)2 + iq+ (ir+ )


kIm > 0

2 ( + k)


a(k)
1
q+

( + k)2
+ iq+ (ir )
kIm < 0

2 ( + k)
q

i.e., since q+ r+ = 2 k 2 and q+ r = q+ r+ ei(+ ) (2 k 2 ) ei(+ ) ,


(
1
kIm > 0
a(k)
.
(3.50a)
i(+ )
e
kIm < 0
52

It is noteworthy that both limits are finite and, in particular, unimodular. A similar
relation holds for a
(k), for which (3.42b) yields



1
r

+
2

r+ q + ( + k)
kIm > 0
2( + k)
r
a
(k)


1

r+ q+ + ( + k)2
kIm < 0

2( + k)
i.e.

a
(k)

kIm < 0
.
kIm > 0

i(+ )

(3.50b)

From (3.38b) it also follows that



b(k) = O |k|1 ,

b(k) = O |k|1

as

|k| , k R.

We will say more about the asymptotic behavior of the scattering coefficients in
Sec. 3.2.1.
Behavior at the branch points of Jost solutions and scattering coefficients
Since the Greens functions are well defined in the limit 0, the Jost solutions
also admit a well defined limit at the branch points k = q0 .

Following [4], we now discuss the behavior of the scattering coefficients in the
vicinity of the branch points, k = q0 . From (3.38a) it follows that if and are
linearly independent at k = q0 or k = q0 (i.e., at = 0), then a(k) has a singularity
of the form
a
a(k)|k=q0 =
+ O(1)
(3.51)

with non-zero a
Wr((x, q0 ), (x, q0 ))
a =
.
2q0
However, it may also happen that (x, k) and (x, k) become linearly dependent at
k = q0 or k = q0 , or both. In this case, either a+ or a , or both, vanish, so that
a(k) is non-singular near the corresponding branch point. In scattering theory, when
this happens k = q0 or k = q0 are called virtual levels [4]. The behavior of b(k)
in the neighborhood of k = q0 is the same as a(k). Indeed, from (3.34a)-(3.34b) it

53

become linearly dependent at k = q0 ,


follows that and (as well as and )
hence there exist constants and such that
q0 ),
(x, q0 ) = (x,

q0 ).
(x, q0 ) = (x,

Evaluating the previous equations as |x| and taking into account (3.7a)-(3.7b)
and (3.8a)-(3.8b), we obtain

q0 ),
(x, q0 ) = ei(+ 2 ) (x,

q0 ).
(x, q0 ) = ei( 2 ) (x,

(3.52)

Then, from (3.38b) it follows that

b(k)|k=q0 = ei(+ 2 )

a
+ O(1),

(3.53)

i.e.
b = ia ei+
and consequently the behavior of the reflection coefficient(k) at the branch points
is given by
lim (k) = iei+ .
(3.54)
kq0

From these relations it follows that b = 0 iff a = 0.


In a similar way one obtains that if and are linearly independent at k = q0 ,
then
a

a
(k)|k=q0 =
+ O(1)
(3.55)

i+ a
b(k)
=
ie
+ O(1)
(3.56)
k=q0

and consequently
lim (k) = iei+ .
(3.57)
kq0

Symmetry reductions
The NLS equation (1.1) is a special case of the system (2.1a)(2.1b) under the
symmetry reduction
r = q ,
2 = 1.
(3.58)
Since we are interested in the defocusing case, we will consider = 1. This symmetry in the potential induces a symmetry among the Jost solutions. In turn, this
54

symmetry of the Jost solutions induces a symmetry in the scattering data. Indeed,
if v(x, k) = (v (1) (x, k), v (2) (x, k))T satisfies Eq. (3.1) and symmetry (3.58) holds with
= 1, v(x, k) = (v (2) (x, k ), v (1) (x, k ))H also satisfies the same equation. Taking
into account the boundary conditions (3.7a)(3.7b), we get
!
!
(2)

(2)

(x,
k
,

(x,
k
,

)
k, ) =
k) =
(x,
,
(x,
(3.59a)
(1) (x, k , )
(1) (x, k , )
!
!
(2)

(2)

N
(x,
k
,

)
M
(x,
k
,

)
(x, k, ) =

N
,
M(x,
k) =
(3.59b)
N (1) (x, k , )
M (1) (x, k , )
where is considered on the appropriate sheet. Note that is a function of k, so
the explicit dependence on has so far being omitted; however, when discussing
symmetry reductions and symmetries, it becomes important to specify the values of
both k and .
In order to define this symmetry relation more precisely, let us label by j = 1
the upper sheet of the Riemann surface and j = 2 the lower sheet. The involution
P (k, j) = (k , j + 1),

j = 1, 2

where j + 1 is intended mod 2, permutes the two sheets in the sense that
1 (k) = (2 (k ))
where 1 is the value of the square root on the first sheet, 2 on the second. Indeed,
if k k , then 1 2 1 and 2 2 so that (cf. Fig. 3.2) and
since on the second sheet the square root is defined by changing the overall sign, the
result is that only the sign of the imaginary part is changed if one exchanges k and
k and goes from one sheet to the other one. As a consequence, from (3.38a)(3.38b)
it follows
a(k, ) = a (k , )
b(k, ) = b (k , ).

(3.60a)
(3.60b)

In particular, on the edges of the cuts this gives


b(k + i0) = b (k i0)

a
(k + i0) = a (k i0),
55

(3.61)

i.e., values of a, a
and b, b on opposite edges are conjugate to each other (which is
consistent with the way the two sheets of the Riemann surface are glued together,
cf. Fig. 3.3.) Then, taking into account (3.45), one has
(k + i0, ) = (k i0, )

(3.62)

wherever all four scattering coefficients are defined (and takes the value that
pertains to the corresponding edge).
There is another involution sending k k and (if k k , then, on
each sheet, 1 2 1 , 2 2 , cf. Fig. 3.2; in this case, takes opposite real
part and the same imaginary part, i.e. on each sheet one has (k) = ((k )) .) A
more formal definition of this involution would be
R(k, j) = (k , j)

j = 1, 2

and one can check that


k
1 (x, k, )
ir
k
((x, k , )) =
1 (x, k, )
iq+

((x, k , )) =

(3.63a)
(3.63b)

and
k
1 (x, k, )
iq

k, )
k , ) = k 1 (x,
(x,
ir+
k , )
(x,

(3.63c)
(3.63d)

where 1 is the Pauli matrix

1 =

0 1
1 0

(3.64)

Note that these symmetries relate the values of the two components of each Jost
solution at points k and k on the same sheet. In particular, the previous equations
provide a relationship between the values of the Jost solutions on the upper and
lower edges of the cuts on the corresponding sheets of analyticity, i.e.
((x, k i0)) =

k
1 (x, k + i0)
ir

56

k
1 (x, k + i0)
iq+

k i0) = k 1 (x,
k + i0)
(x,
iq

k i0) = k 1 (x,
k + i0).
(x,
ir+
((x, k i0)) =

Furthermore, from (3.35a)(3.35b) it follows

q+
b(k, ) ei(+ + ) b(k, )
r
r+
(a(k , )) =
a(k, ) ei(+ ) a(k, )
r

(b(k , )) =

(3.65a)
(3.65b)

and
r+
(b(k , )) = b(k, ) ei(+ + )b(k, )
q
q+


(
a(k , )) =
a
(k, ) ei(+ ) a
(k, ).
q

(3.65c)
(3.65d)

In particular, we finally obtain


b (k i0) = ei(+ + ) b(k + i0),
b (k i0) = ei(+ + )b(k + i0),

a (k i0) = ei(+ ) a(k + i0)

a
(k i0) = ei(+ ) a
(k + i0).

(3.66a)
(3.66b)

Proper eigenvalues and norming constants


Since the spectral operator (3.1) associated to the de-focusing NLS is self-adjoint
(r = q ), its spectrum lies on the real axis. Also, due to (3.60a)(3.60b) and (3.36),
on either side of the cuts
|a(k)|2 = 1 + |b(k)|2 ,

|k| > q0

(3.67)

and therefore the zeros of a(k) can only lie on the segment q0 < k < q0 . In fact,
as far as the branch points are concerned, one has either |a(q0 )| = (the general
case) or |a(q0 )| < (the case of virtual levels) in which case still |a(q0 )| 1
by the unitarity relation (3.67). Hence, a(k) has only a finite number of zeros lying
inside the segment q0 < k < q0 .
A proper eigenvalue of the scattering problem (3.1) is a value of k corresponding
to a bounded solution v such that v 0 as x . The zeros of a(k) correspond
57

to the discrete eigenvalues of the scattering problem. Indeed, suppose kj R ,


|kj | < q0 for j = 1, . . . , J are such that a(kj ) = 0. Then, from (3.38a) it follows that
for k = kj the eigenfunctions and are proportional to each other, i.e., there exist
(complex) coefficients bj such that
(x, kj , (kj )) = bj (x, kj , (kj )),

j = 1, . . . , J.

(3.68)

As a consequence, since both and are defined on the upper sheet of the Riemann
surface, one has
2 2
j (x) (x, kj , ij ) ei kj q0 x wj ej x wj
x
2 2
j (x) = bj (x, kj , ij ) bj ei kj q0 x vj bj ej x vj
x +
where wj and vj are the two-component vectors
!
kj + ij
wj =
,
vj =
ir
and we introduced
j =

iq+
kj + ij

q02 kj2 R+ .

(3.69)

The complex constants bj are called norming constants. According to the symmetry
(3.60a), on the lower sheet of the Riemann surface a(kj , ij ) = 0 and therefore
kj , ij ) = bj (x,
kj , ij ),
(x,

j = 1, . . . , J.

(3.70)

Then, taking (3.59a) into account, from (3.68), (3.70) it follows that for any j =
1, . . . , J
bj = b .
(3.71)
j
Also, from (3.63a)-(3.63d), one has
bj =

q+
bj ,
r

b = r+ bj .
j
q

(3.72)

Following [4], one can also prove that the zeros of a(k) are simple. Indeed, let
us introduce the following notation for the derivative with respect to the spectral
parameter
v
v =
.
k
58

Any solution v of the scattering problem (3.1) is such that


v
= iJv + (ikJ + Q)v .
k
In particular, for the solutions and we have

(Wr ( , )) = i Wr (J, ) ,
x

(3.73)

where Wr denotes the usual Wronksian, cf. (3.33). Indeed,


 




(Wr ( , )) = Wr
, + Wr ,
x
x
x

= i Wr (J, ) + Wr ((ikJ + Q) , ) + Wr ( , (ikJ + Q))

and (3.73) follows because, introducing for shortness L = ikJ + Q,



Wr (L , ) = det L Wr , L1 = Wr ( , L)

due to the fact that

1
L.
det L
Therefore, evaluating (3.73) at k = kj and taking into account (3.68), we obtain
L1 =

(Wr ( (x, kj ), (x, kj ))) = i Wr (J(x, kj ), (x, kj )) = ibj Wr (Jj (x), j (x)) .
x
(3.74)
Similarly

(Wr (, )) = i Wr (J, )
(3.75)
x
and therefore

(Wr ((x, kj ), (x, kj ))) = ibj Wr (Jj (x), j (x)) .


x

(3.76)

Let us now introduce


(x, k) = Wr (J(x, k), (x, k)) ,

(x, kj ) = Wr (Jj (x), j (x))

(3.77)

and note that, since j and j are decaying at both space infinities, one has
(x, kj ) 0 as |x| . Then from (3.76) it follows
Z x

Wr ((x, kj ), (x, kj )) = ibj


(x , kj )dx

59

and from (3.74)

Wr ( (x, kj ), (x, kj )) = ibj

(x , kj )dx .

From (3.38a), recalling that Wr (, ) |k=kj = 0, we obtain





da
1

a (kj )
=
(Wr (, ) + Wr ( , ))
dk k=kj
2( + k)
k=kj
Z
ibj
(x, kj )dx
=
2ij (kj + ij )
R
and then a (kj ) 6= 0 iff (x, kj )dx 6= 0. The symmetry relation (3.63b) evaluated
at k = kj , q0 < k < q0 gives
((x, kj )) =
hence
Jj (x) =
where 2 is the second Pauli matrix

ij + kj
1 (x, kj )
iq+

q+
2 (j (x))
kj ij

2 =

0 i
i 0

(3.78)

Substituting into (3.77) we conclude that


(x, kj ) = Wr (Jj (x), j (x)) =

q+
iq+
Wr ( 2 (j (x)) , j (x)) =
kj (x)k2
kj ij
kj ij

with the usual definition for the Euclidean norm of a two-component complex vector.
Hence
Z
Z
iq+
(x, kj )dx =
kj (x)k2 dx 6= 0
kj ij

and

ibj
iq+
a (kj ) =
2ij (kj + ij ) kj ij

bj
kj (x)k dx
2ij r+

which is different from zero, thus completing the proof.

60

kj (x)k2 dx
(3.79)

3.2

Inverse scattering problem

3.2.1

Riemann- Hilbert problem via uniform coordinate

In order to formulate a Riemann-Hilbert problem on the complex plane, instead of


a Riemann surface, we introduce the conformal mapping defined by
z = z(k) = k + (k) .

(3.80a)

The inverse mapping is given by




1
q02
k = k(z) =
z+
2
z
(as follows from a direct check) and also


1
q02
z
,
(k) = z k =
2
z

(k) k =

(3.80b)

q02
.
z

(3.80c)

It is not difficult to show that:


i the cut L on the Riemann surface is mapped onto the real axis of the complex
z-plane. More precisely, if we write, according to Fig. 3.2,
L = bI cI dI aI ,
in this precise order, then
k bI (, q0 ) z (, q0 )

k cI = (q0 , ) z (q0 , 0)

k dI = (+, q0 ) z (0, q0 )

k aI (q0 , +) z (q0 , +)
so that the variable z describes its entire real axis;
ii the sheets C1 /C2 are mapped onto the upper/lower half planes of z
iii a neighborhood of k = on C1 with kIm > 0 is mapped onto a neighborhood
of z =
61

iv a neighborhood of k = on C1 with kIm < 0 is mapped onto a neighborhood of


z=0
v a neighborhood of k = on C2 with kIm > 0 is mapped onto a neighborhood of
z=0
vi a neighborhood of k = on C2 with kIm < 0 is mapped onto a neighborhood of
z=
vii the symmetry k i0 k + i0 on L (used to relate the limiting values of
eigenfunctions and scattering data at the edges of the cut) transforms into
z q02 /z on the real z-axis.
In the following we will therefore express all eigenfunctions and scattering data
intervening in the inverse problem in terms of the uniform variable z.
As a function of q
z, M/a is defined for all zIm 0, analytic except at the points
zj = kj + ij (j = q02 kj2 0) where it has simple poles. Taking into account
the large k behavior of the eigenfunctions and scattering data (cf. (3.49b), (3.50a)),
one also obtains

1
z
a(z)
(3.81a)
q
+ ei(+ ) z 0
q
and

M(x, z)

a(z)
M(x, z)

a(z)

z
ir(x)

z q(x)/q+
ir+

z
!

z 0.

z) and a
Similarly, N(x,
(z) are analytic in the lower half plane (zIm 0)

1
z
a
(z)
r+

ei(+ ) z 0
r

(3.81b)
(3.81c)

(3.81d)

and

z)
N(x,

z
ir(x)

62

(3.81e)

z q(x)/q+
ir+

z)
N(x,

z0

and N(x, z) is analytic in the upper half plane (zIm 0) and


!
iq(x)
N(x, z)
z
z
!
iq+
N(x, z)
z 0.
z r(x)/r+

(3.81f)

(3.81g)
(3.81h)

/
Finally, M
a, as a function of z, is defined in the lower half plane (zIm 0),
where q
it is meromorphic with simple poles at the points zj = kj + (kj ) = kj ij
and a
, one
(j = q 2 k 2 0). Taking into account the large k behavior of M
0

/
can check that M
a has the same behavior as N both at z = 0 and as z but
in the lower half plane, i.e.
!

iq(x)
M (x, z)

z
(3.81i)
a
(z)
z
!

iq+
M (x, z)

z 0.
(3.81j)
a
(z)
z r(x)/r+
The jump conditions (3.43a)-(3.43b) at the real z-axis can be written as
M(x, z)
(x, z) = (z)ei(zq02 /z)x N(x, z)
N
a(z)
(x, z)
M
2
z).
N(x, z) = (z)ei(zq0 /z)x N(x,
a
(z)
Note that
i

z) = N
(x, z)e 2i (zq02 /z)x
(x,

(x, z) = N (x, z) e 2 (zq0 /z)x ,

and that the symmetries (3.59a)-(3.59b) read


z),
((x, z )) = 1 (x,

(x, z)
(N(x, z )) = 1 N

(3.82a)

( 1 is the Pauli matrix defined in (3.64)) and the symmetries (3.63b), (3.63d) relating the values of the eigenfunctions below and above the cut on L become
x, q02 /z



q02 /z
ir+
1 (x, z)
1 (x, z)
iq+
z
63

(3.82b)

x, q02 /z



iq+
q02 /z
1 (x, z)
1 (x, z).
ir+
z

(3.82c)

In order to take into account the assigned behavior of the eigenfunction at z = 0,


let us divide the jump conditions by z and consider
(x, z)
M(x, z) N
2
N(x, z)

= (z)ei(zq0 /z )x
za(z)
z
z
(x, z) N(x, z)
z)
M
N(x,
2

= (z)ei(zq0 /z)x
.
z
a(z)
z
z
In this way, the functions will be bounded at infinity but have an additional pole at
z = 0. However, the pole contribution can be subtracted out since the value of all
the functions at z = 0 is given. Note that the pole at z = 0 is along
but
R the contour,
R 

the integral is intended in the principal value sense, as lim0+ + dz and


2

the integrals of such functions as ei(zq0 /z)x are taken in the sense of distributions.
Taking into account (3.81b)-(3.81j), one can write
!
!
#
"
J
P
1
0
1
M(x, zj )
M(x, z)

(3.83a)

za(z)
z
0
ir+
j=1 (z zj )zj a (zj )
!
!
#
"
J b e2j x N(x, z )
z)
P
1
0
N(x,
1
(z) i(zq02 /z)x
j
j

=
e
N(x, z)

z
z
z
0
ir+
j=1 (z zj )zj a (zj )
"
!
!
#
J

(x, zj )
P
M(x,
z)
0
1 iq+
M

(3.83b)
z
a(z)
z
j )
zj a
(
zj )
1
0
j=1 (z z
"
!
!
#
J
(x, zj )
P
0
N(x, z)
1 iq+
bj e2j x N
(z) i(zq02 /z)x

=
e
N(x, z)

z
z
j )
zj a (
zj )
z
1
0
j=1 (z z

where we also used (3.68) and (3.70) to express M(x, zj ) and M(x,
zj ) in terms of

N(x, zj ) and N(x, zj ).


Let us now consider the standard projectors in the z-plane, i.e.
Z
1
f ()
P [f ](z) =
d,
2i (z i0)

(3.84)

defined for any f (), R that is decaying at large and integrable on the real
line. P [f ](z) is analytic in the upper/lower half plane and, if f admit analytic
continuation in the upper/lower half plane, one has
P [f ](z) = f (z),
64

P [f ](z) = 0.

Hence, if we apply the projector P to eq. (3.83a) and P+ to eq. (3.83b), taking into
account the analyticity properties and decay of the involved functions, we obtain
!
!
Z
J b e2j x N(x, z )
z)
P
1
0
N(x,
1
1
() i(q02 /)x
j
j
=
+
+
+
d
e
N(x, )

z
z
2i
( z)
0
ir+
j=1 (z zj )zj a (zj )
!
!
Z
J
zj )
P
0
N(x, z)
1 iq+
bj e2j x N(x,
1
() i(q02 /)x
=
+
+

d
e
N (x, )

z
z
j )
zj a
(
zj )
2i
( z)
1
0
j=1 (z z

i.e.,

z) =
N(x,

z
ir+

z bj e2j x N(x, zj )
z
+
+

zj a (zj )
2i
j=1 z zj
J
P

() i(q02 /)x
e
N(x, )
( z)
(3.85a)

(3.85b)
N(x, z) =

iq+
z

Z
(x, zj )
z
() i(q02 /)x
z bj e2j x N

d
e
N(x, ) .
+

j
zj a
(
zj )
2i
( z)
j=1 z z
J
P

The previous system of equations is closed by evaluating the first equation at z = zm


and the second one at z = zm for each m = 1, . . . , J, so that, introducing the shorthand notations
j (x) = N
(x, zj ) ,
N

Nj (x) = N (x, zj )

j = 1, . . . , J

(3.86)

we have
m (x) =
N

Nm (x) =

zm
ir+

iq+
zm

zm bj e2j x Nj (x)
zm
+
+

m zj
zj a (zj )
2i
j=1 z

J
P

()
2
ei(q0 /)x N(x, )
( zm )
(3.87a)

Z
j (x)
zm bj e2j x N
zm
()
2
(x, ).
+

d
ei(q0 /)x N

(
z
)
2i

z
)
m
j
j
j
m
j=1

J
P

(3.87b)

Note that from (3.81e), looking at the large z limit of the second component of
z), one gets
N(x,
Z
J b e2j x N (2) (x, z )
P
1
() i(q02 /)x (2)
j
j
ir(x) = ir+ +

d
e
N (x, ). (3.88)
zj a (zj )
2i

j=1
65

z) according to (3.81f) gives


Also, the limit as z 0 of the first component of N(x,
Z
J 1 b e2j x N (1) (x, z )
P
q(x)
1
()
2
j
j
= 1
+
d 2 ei(q0 /)x N (1) (x, ) (3.89)

q+
zj a (zj )
2i

j=1 zj

and one can verify, using the symmetries of eigenfunctions and scattering data, that
the two previous equations give the same result for the potential, up to complex
conjugation. Indeed, the complex conjugate of Eq. (3.89) gives
Z
J
 ir+ () i(q2 / )x

P
1 bj e2j x
(1)
(1)
0
ir(x) = ir+ ir+
N
(x,
z
)

d
e
N
(x,
)
,
j

2i
2
j=1 (zj ) (a (zj ))
(3.90)
to be compared with (3.88). From (3.82b)-(3.82c) one has

z
1 N x, q02 /z
ir+

(x, z) = z 1 N
x, q 2 /z .
N
0
iq+

(3.91a)

N(x, z) =

Also, note that the points zj = kj + ij , with q0 < kj < q0 and j =


such that

(kj , (kj )) (kj , ij ) kj , (kj )

(3.91b)
q

q02 kj2 are

and therefore from symmetry (3.82a) one has


N

(1)

(x, zj )

Also, from (3.65b), i.e.

zj (2)
ij kj (2)
N (x, zj ) =
N (x, zj ).
=
iq+
iq+
a (k , ) =

it follows

(3.92)

r+
a(k, )
r

d
r+ d
a (k , ) =
a(k, ).
dk
r dk

(3.93)

Then, taking into account that


d
d
=
dz
+ k dk

(3.94)

it hence follows that




da(z)
ij
da(k, )
a (zj )
=
.
dz z=zj
kj + ij
dk k=kj

66

(3.95)

Similarly, one gets


ij d
a(k, )
a
(
zj ) =
.
kj ij
dk k=kj

Therefore, one has

(a (zj )) =

(
a (
zj )) =

(3.96)

r+ zj
a (zj )
r zj

(3.97)

q+ zj
a (
zj )
q zj

(3.98)

where the derivatives are meant with respect to the variable z. Furthermore, one
also has (3.65a)-(3.65b) and (3.72), i.e.
bj =
and
(k , ) =

q+
(k, )
r+

q+
bj
r

(3.99)

(q02 /z) =

q+
(z).
r+

Performing in the integral the change of variables q02 /, such that


R R 0
+ + and d (q02 / 2) d, one can rewrite (3.90) as
0
 ir+
1 bj e2j x
(1)
ir(x) = ir+ ir+
N
(x,
z
)

2i
j=1 (zj ) (a (zj ))
J
P

(3.100)
R0

R
0


d 2  i(q02 / )x
(1)
q
/
e

N
(x,
q
/)
0
0
q02

and taking into account (3.99), (3.97), (3.92), (3.100) and (3.91a) one can check that
it coincides with (3.88).
As mentioned above, the integrals of exp [i(z q02 /z)x] over z are taken in the
sense of distributions, namely
Z +
Z 0
Z
2
i(zq02 /z)x
i(zq02 /z)x
e
dz =
e
dz +
ei(zq0 /z)x dz

0


Z
Z
2
q0
i(zq02 /z)x
=
e
1 + 2 dz =
eipx dp = 2(x)
z
0

where in the second equality we have performed the change of variables z q02 /z.
In a similar manner one can show that
Z
Z
dz
2
2
i(zq02 /z)x dz
e
= 0,
ei(zq0 /z)x 2 = 2 (x).
z
z
q0

67

Let us investigate in detail the decay of (z) as |z| and z 0 (obviously,


the same will hold for (z)), from the definitions (3.38a) and (3.38b). Taking into
account (3.49b)-(3.49c), one has
!
!!
z
+
a
z
+
a
1
2
e2i(z)x Wr
,
ir(x) + b1 /z
ir(x) + b2 /z
b(z)
|z|
2
z
i.e.

b2 b1 + ir(x) (a1 a2 ) i(zq02 /z )x


e
z2
and since a(z) 1 as |z| , it follows that
b(z)

(z)ei(zq0 /z )x = O(z 2 )
2

|z|

|z| .

Similarly, taking again into account (3.49b)-(3.49c), one has


!!
!
q(x)
q(x)
2
2
z
+
a
z
+
a
z
z
3
4
q
q+
b(z) q02 e2i(z)x Wr
,
ir+ + b4 z
ir + b3 z

z 0,

i.e.
b(z)

q02

and, since








r+ r
b4
b3
2
2
z iq(x)

+ z q(x)

+ i (a3 r+ a4 r ) ei(zq0 /z )x
q q+
q q+
r+
q

r
q+

= 0 and a(z) q+ /q as z 0, it follows that


(z)ei(zq0 /z )x = O(z 2 )
2

z 0.

In conclusion, the result at the end of Sec. 2.2.1 for b(k) can be strengthened to
b(k) = o(|k|1) as |k| on either side of the cut, and one can conclude that the
decay properties of and at both z = 0 and |z| = are sufficient to ensure that
the integral equations derived for the inverse problem are well defined.

3.2.2

Trace formula

Recall that as functions of the unform variable z, a(z) is analytic in the upper half
plane and a(z) in the lower. Assume that a(z) and a(z) have simple zeros at points
q
n
oJ
zj = kj + ij , |kj | < q0 , j = q02 kj2 > 0

j=1

68

z0

and

q
n
oJ
zj = kj ij , |kj | < q0 , j = q02 kj2 > 0

j=1

respectively. Therefore, if we define


(z) =

J
Y
z zm
a(z),
z

z
m
m=1

(z)
=

J
Y
z zm
a(z),
z

m
m=1

(3.101)

(z) is analytic in the upper z-plane, where it has no zeros, while (z)
is analytic in
the lower z-plane, where it has no zeros. Moreover, from (3.81a),(3.81d) it follows

that (z), (z)


1 as |z| in the respective half-planes, hence
Z
1
log ()
log (z) =
d
Im z > 0
2i z
Z

1
log ()
d = 0
Im z > 0
2i z
and
Z

1
log ()

log (z) =
d
Im z < 0
2i z
Z
1
log ()
d = 0
Im z < 0.
2i z
Subtracting the equations from one another and using (3.101), we obtain


Z
J
P
z zm
1
log (a()
a())
log a(z) =
log
+
d
Im z > 0
z zm
2i
z
m=1


Z
J
P
z zm
1
log (a()
a())
log a
(z) =
log

d
Im z < 0.
z zm
2i
z
m=1

This allows one to recover a(z) and a


(z) from knowledge of {zj = kj + ij }Jj=1 and
of a()
a(), which, according to (3.67), for R can be written as a()
a()
2
2 1
|a()| = 1 |()|
. Explicitly, one can write
"
 #

Z
J 
Y
log 1 |()|2
z zm
1
a(z) =
exp
d
Im z > 0 (3.102)
z zm
2i
z
m=1

and recalling that, according to (3.81a), a(z) q+ /q as z 0, we conclude that


the potential satisfy
"
 #
Z
J
Y
log 1 |()|2
q+
zm
1
=
exp
d .
(3.103)
q m=1 zm
2i

69

Following [4], we refer to (3.103) as -condition, since q+ /q gives the asymptotic


phase difference + of the potential. In particular, in the reflectionless
case one has that the asymptotic phases and the soliton amplitudes and velocities
are not independent from each other. Specifically, they are related via the following
condition
J
Y
q+
km + im
=
.
(3.104)
q m=1 km im

3.2.3

Reflectionless solutions

In the reflectionless case, with one single discrete eigenvalue k1 (q0 , q0 ), the
system (3.85) reduces to an algebraic system
!

C1 21 x
1
1 (x) =
e
N1 (x)
N
+ 1
2i1
ir+
!
iq+
C1 21 x
N1 (x) =
+ 1
e
N1 (x)
2i1
1
where
C1 =

b1
,

i1 a (k1 , i1 )

C1 =

1 = k1 + i1 ,

1 =

and

b1
i1 a (k1 , i1 )
q

(3.105)

q02 k12 .

Therefore, taking into account that 1 1 = q02 = ir+ (iq+ ), we obtain


!
C1 r+ 21 x
1 + 2
e
iq+

1

N1 (x) =
q0 |C1 | 21 x
q0 |C1 | 21 x
1
1
e
1+
e
21

(3.106)

21

where we used C1 = C1 and 1 1 /q+ r+ . Now note that from the definition of Cj ,
taking into account (3.93) and (3.99), the second symmetry yields for the norming
constants
bj
1 q+
r

=
bj
,
Cj =

ij (a (kj , ij ))
ij r r+ a (kj , ij )
i.e.
Cj =

q+
Cj e2i+ Cj .
r+
70

(3.107)

Note also that the previous equation is equivalent to


Cj = |Cj |ei+

(3.108)

i.e., either arg C1 = + or arg C1 = + . Now, recalling the definition of (3.105)


and taking into account the relation (3.79), one gets
Z
1
i+
2
Cj = 2q0 e
||j (x)|| dx
(3.109)

which yields
arg Cj = + .
Therefore, substituting into (3.106), we obtain the regular solution, for which the
denominator never vanishes,
!
iq+
1
N1 (x) =
(3.110)
q0 |C1 | 21 x
1
1 + 2 e
1

which corresponds to the lower sign in (3.108).


Inserting (3.110) into (3.85) and going back from the uniform variable z to the
original variables k, we obtain
!

+
k
e21 x
(x, k) =
N1 (x),
N
+ C1
k + 1
ir+
which, as expected, is analytic in the entire lower sheet. If we now consider the
large k asymptotics for, say, kIm < 0, recalling (3.48d) and (3.49c) we obtain from
the second component
21 x

ir(x) = ir+ + C1 e

(2)
N1 (x)

21 x

= ir+ + C1 e

Then we note that as x +, since 1 =


lim r(x) = r+ + 2i1

1+
1

1 1 1 21 x
C
e
21 q+
.

C1 C1 41 x
q02 (2
2e
1)

p
q02 k12 > 0, r(x) r+ and also

1
q 2 + 2i1 (k1 + i1 )
(k1 + i1 )2
2
= 0
=
= 1 ei+
q+
q+
q+
q0

and one can check that this coincides with r = q0 ei provided the -condition
(3.103), which, in the case of one-soliton reduces to
k1 + i1
1
= ei(+ ) ,
k1 i1
1
71

(3.111)

is satisfied. Taking into account the symmetries, the reflectionless solution becomes
ir(x) = ir+ +

3.3

C1 1 e21 x
1+

q0 |C1 | 21 x
e
21

(3.112)

Time evolution and soliton solutions

The operator (3.4) determines the evolution of the Jost solutions, which can be
written as
!
A B
t v =
v.
(3.113)
C A
We assume that, as x , q q (t) q0 ei (t) and, since r = q , r r (t)
q0 ei (t) ; also, we assume that qx , rx 0 as |x| (this last condition is clearly
compatible with the choice of the non-vanishing boundaries.) As a consequence, we
obtain the following system of equations fixing, at large space infinities, the time
dependence of the eigenfunctions

and


(1)
v i 2k 2 + q02 v (1) 2kq (t)v (2)
t

(2)
v 2kr (t)v (1) i 2k 2 + q02 v (2)
t


(1)
v i 2k 2 + q02 v (1) 2kq+ (t)v (2)
x +
t

(2)
v 2kr+ (t)v (1) i 2k 2 + q02 v (2) .
t

In order to decouple the previous systems, we can use the scattering problem, according to which as x
q (t) v (2) vx(1) + ikv (1) ,

r (t) v (1) vx(2) ikv (2)

and consequently, at both space infinities, i.e. as |x|


(1)
v iq02 v (1) 2kvx(1)
t
(2)
v iq02 v (2) 2kvx(2) .
t
72

(3.114a)
(3.114b)

Note that the eigenfunctions themselves, whose boundary values at space infinities
are given by (3.7a)-(3.7b), are not compatible with this time evolution. Therefore,
one introduces time-dependent eigenfunctions
t)
t) = eiB t (x,
(x,
t)

(x,
t) = eiD t (x,

(x, t) = eiA t (x, t),


(x, t) = eiC t (x, t),

to be solutions of the time-differential equation (3.113). Then, one has, for instance
(1)
(1)
= iA (1) + eiA t
,
t
t

(2)
(2)
= iA (2) + eiA t
t
t

(3.115)

and taking into account that the components of satisfy, asymptotically as |x|
, (3.114a)-(3.114b) and that (since q0 , and hence also , are constant in time, and
the only dependence of r (t) on t is via the phase (t))
!
!
+k
0

(x, t)
eix ,

eix
x
t
ir (t)
(t)r (t)
we obtain, substituting into (3.115), from the first component
A = 2k + q02

(3.116)

and from the second component


A = q02 + 2k.
We conclude that, in order for the two of them to be compatible, one must have
(t) = 2q02

(t) = 2q02 t + ,

i.e.

(3.117)

which completely fixes the time evolution of the asymptotic phase for the potential.
From (3.116) it then follows that the time evolution of the scattering eigenfunction is given by
!
A iA
B

=
.
(3.118)
t
C
A iA
In a similar way, one obtains that
B = A = q02 2k
73

provided (3.117) holds, and, consequently,



=
t

A + iA
B
C
A + iA

(3.119)

and
C = A ,

D = A

provided
+ (t) = 2q02

i.e.

+ (t) = 2q02 t + +

(3.120)


=
t

so that

=
t

A + iA
B
C
A + iA

A iA
B
C
A iA

(3.121)
Note that, as a consequence of (3.117) and (3.120) one also has that the phase
difference is, as expected, time-independent, i.e.
d
d
(+ )
= 0.
dt
dt
From the scattering equation (3.35a) we then obtain
!
A iA
B
b
a

=
= b + b
+
+a
t
t
t
t
t
C
A iA
A iA
B
C
A iA


 b
b + a = b + b
t

a
+
b + a
t

i.e.
2iA b =

A + iA
B
C
A + iA
A iA
B
C
A iA

b
a
b +
b
t
t

which gives, evaluating it as x +,


a
= 0,
t

b
= 2iA b.
t

We conclude that a(k, t) = a(k, 0), i.e., a is time-independent, while


2

b(k, t) = b(k, 0)e2i(q0 +2k)t .


74

(3.122a)

The eigenvalues kj , being the (real) zeros of a(k) along the segment |k| < q0 of the
real axis on the upper sheet of Riemann surface, are also constant of the motion.
Since at those zeros the eigenfunctions and are proportional to each other (cf.
(3.68)), in those cases when b can be analytically continued in the neighborhood of
each kj , one has bj (t) = b(kj , t) and therefore
2

bj (t) = bj (0)e2iq0 t+4kj j t .

(3.122b)

The norming constants Cj will have precisely the same time dependence, i.e.
2

Cj (t) = Cj (0)e2iq0 t+4kj j t .

(3.122c)

From (3.112) we can now reconstruct the pure one-soliton solution, taking into
account that the time-dependent terms are q+ and C1 , which, due to (3.120) and
(3.122c), evolve according to
2

q+ (t) = q0 ei+ +2iq0 t ,

C1 (t) = C0 e2iq0 t+4k1 1 t


p
where k1 is the discrete eigenvalue (|k1 | < q0 ), and 1 = q02 k12 . Then from
(3.112) it follows that
2

q(x, t) = q0 ei+ +2iq0 t +

i.e.
2iq02 t

q(x, t) = q0 e
If we introduce

iC1 (0)1 e2iq0 t+4k1 1 t21 x


q0 |C1(0)| 21 x+4k1 1 t
e
1+
21

i
e + +

iC1 (0)
1+

1
q0

e21 x+4k1 1 t

q0 |C1 (0)| 21 x+4k1 1 t


e
21

q0 |C1 (0)|
21
and take into account (3.108), the previous relation can be written as

2i1 1 21 (x2k1 tx0 )


e
2
q2
q(x, t) = q0 e2iq0 t ei+ 1 0 21 (x2k1 tx0 ) ,
1+e
e21 x0 =

(3.123)

(3.124)

which coincides, up to an overall phase factor, with the dark soliton solution (1.5),
with = k1 /q0 and = 1 /q0 .
75

Chapter 4
Conclusions
Over the past fifty years, a large body of knowledge has been accumulated on NLS
systems, which continue to be extensively studied worldwide. Nonetheless, problems in which the boundary conditions play a significant role still pose significant
challenges, and a comprehensive picture is still elusive.
This thesis has dealt with the solution of the initial value problem for the nonlinear Schrodinger equation in the defocusing dispersion regime via the Inverse Scattering Transform, when certain special non-decaying boundary conditions are imposed.
The choice of the boundary conditions is such that it allows for initial data that include dark/grey solitons.
In Chapter 1 we have given a description of NLS soliton solutions. In Chapter 2
we have discussed the IST for the focusing NLS with decaying boundary conditions,
in order to establish the notations and illustrate the concepts in a simpler case. In
Chapter 3 we have described the IST for the defocusing NLS equation with nondecaying boundary conditions. Special attention has been devoted to the direct
problem, where we have obtained sufficient conditions on the initial data such that
the eigenfunctions are analytic away from the branch points and the direct problem
is well-posed. As far as the inverse problem is concerned, we have formulated it as
a Riemann-Hilbert problem with respect to a suitable uniform variable, derived the
asymptotic behavior of the scattering data and obtained necessary conditions for
the well-posedness of the inverse problem.
The study of sufficient conditions for the inverse problem to be well-posed, and
the investigation of the functional class of the corresponding NLS solutions will be
76

considered in a future work.

77

Appendix A
A relevant lemma
The purpose of this appendix is to prove the following lemma.
Lemma 3. The choice of the branch cut for in Sec. 3.1.1 is such that Im(k) 0
on the entire upper sheet of the Riemann surface (sheet I), and Im( k) 0 on
the entire lower sheet (sheet II).
Proof. Assume for simplicity q0 = 1 (the result is obviously independent of this
choice, since one can always normalize the potentials in the NLS equation). In this
case,
2 = (k 1) (k + 1) = r1 r2 ei1 +i2
where r1 , r2 and 1 , 2 are defined like in Fig. 3.2 and
0 1 < 2,

2 < .

On sheet I of the Riemann surface:


= (r1 r2 )1/2 ei(1 +2 )/2
On sheet II of the Riemann surface:
= (r1 r2 )1/2 ei(1 +2 )/2+i .
Of course, r1,2 and 1,2 are not independent, only two of them are. Note first that,
by the law of sines,
r1 sin 1 = r2 sin 2 = Im k
(A.1)
78

Also, by the law of cosines


r22 = r12 + 4r1 cos 1 + 4

(A.2)

r2 cos 2 = 2 + r1 cos 1 = Re k + 1.

(A.3)

and finally
Of course (A.1),(A.2),(A.3) are not all independent either.
Note that Im k is only discontinuous on the positive (or negative) real k-axis
and Im is only discontinuous on the branch cut (Im k = 0 and |Re k| 1). Thus,
except for these locations, Im ( k) is continuous, and therefore it can change
sign only by passing through a zero. So, we look for the locus of points satisfying
Im ( k) = 0. We have
Im = (r1 r2 )1/2 sin [(1 + 2 ) /2]
on sheet I. Thus we look for solutions of the following equation
(r1 r2 )1/2 sin [(1 + 2 ) /2] = r1 sin 1 .

(A.4)

If (A.4) is satisfied, then (Im )2 = (Im k)2 , i.e.


r2 sin2 (1 + 2 ) /2 = r1 sin2 1

(A.5)

Of course, squaring may introduce spurious solutions. However, what is important


is that the solutions of (A.4) are (for each choice of sign) a subset of those of (A.5).
From (A.5) we have
r2 (1 + cos (1 + 2 )) r2 (1 cos 1 cos 2 + sin 1 sin 2 ) = 2r1 sin2 1
and, using (A.1) and (A.3)
r2 = r1 + 2 cos 1 .
Squaring again (and again possibly introducing spurious solutions) we have
r22 = r12 + 4r1 cos 1 + 4 cos2 1
i.e., using (A.2),
cos2 1 = 1.
79

(A.6)

Thus, the only solutions to Im ( k) = 0 are for 1 = 0, i.e. on the real k-axis.
Note, however, that solutions with 1 = and 0 < r1 < 2 are spurious, since they
do not satisfy (A.6) (r1,2 are positive by definition). Thus, Im ( k) = 0 only on
the branch cut, which means that Im ( k) are both sign-definite on either sheet
of the Riemann surface.
So, it is sufficient to check the value of Im ( k) at one point. For k = i, one

has = 2ei/2 on sheet I and = 2ei/2 on sheet II. Thus, on sheet I





i/2
Im ( k) = Im
21 e
= 21>0

which means that Im ( k) > 0 in all of sheet I of the Riemann surface. Similarly,
on sheet II, for k = i, one has




i/2
Im ( k) = Im
21 e
= 21<0
and Im ( k) < 0 in all of sheet II.

80

Acknowledgments
This process has taken a long time, and it has started and ended in two different
lives. I would like to dedicate it to my parents, Costantino and Marisa, whose
presence and support has been invaluable to me, on either side of the ocean, and
to my brother Marco, who has kept me busy with having to learn more math stuff
than I needed at that time. But one never knows...
Aunt Rosa and uncle Luciano, as well as Lella, Nora, Mario and the new additions to
the family, Giulio and Simone, deserve a special mention for their constant affection.
I would like to express my gratitude to Mark and Gino for the many discussions and
help on some of the results in this work. I am also grateful to Giorgio, for advising
an unusual student on a topic unusual to him, and allowing a long-distance thesis
relationship, as well as to Marco and Flora, with whom this all started many years
ago. A special acknowledgment goes to Sarby, Radu, George and Greg at UCCS,
for allowing me to take a trip to Italy before the end of the semester.
Many thanks to two very good friends of mine for their help. One of them recently,
and rightfully, said to me that this small accomplishment should be qualified as a
degree in stubbornness (and I am sure the other one would have agreed). Thanks,
Anna and Enzo, for reminding me, occasionally, what a silly idea that was! :)
Thanks also to Giorgia, Nadia, Monika, and all the other friends, for putting up with
me when this brilliant idea of mine did not leave me as much time as I would have
liked to spend with them. And uncountable thanks and thoughts to another one,
whose most precious pebbles have helped me make it through all the discouraging
days and made my sun shine brighter.
It is a long, yet non-exhaustive list, much longer than what I originally had in mind.
Which reminded me, if necessary, that there is always a lot of people to be thankful
to, more than one can list and, in some cases, name.
81

Bibliography
[1] Ablowitz M.J. and Segur H., Solitons and the Inverse Scattering Transform,
Society for Industrial and Applied Mathematics (1981)
[2] Calogero A. and Degasperis F., Spectral Transform and Solitons, North-Holland
(1982)
[3] Novikov S. P., Manakov S.V., Pitaevskii L.P. and Zakharov V.E., Theory of
Solitons: the Inverse Scattering Transform, Plenum (1984)
[4] Faddeev L.D. and Takhtajan L.A., Hamiltonian Methods in the Theory of Solitons, Springer-Verlag (1987)
[5] Ablowitz M.J., Clarkson P.A., Solitons Nonlinear Evolution Equations and Inverse Scattering, London Mathematical Society Lecture Notes Series 149 Cambridge University Press (1991)
[6] Ablowitz M.J., Prinari B., Trubatch A.D., Discrete and Continuous Nonlinear
Schrodinger Systems, London Mathematical Society - Lecture Notes Series 302
Cambridge University Press (2004)
[7] Ginzburg V.L. and Landau L.D., Sov. Phys. JETP 20, 1064 (1950)
[8] Ginzburg V.L., Sov. Phys. JETP 2, 589 (1958)
[9] Ginzburg V.L. and Pitaevskii L.P., Sov. Phys. JETP 7, 858 (1958)
[10] Chiao R.Y., Garmire E. and Towner C.H., Phys. Rev. Lett. 15, 479 (1964)
[11] Talanov V.I., Radiophysics 7, 254 (1964); Talanov V.I., Sov. Phys. JETP Lett.
109, 138 (1965)
82

[12] Benney D.J. and Newell A.C., J. Math. Phys. 46, 133 (1967)
[13] Lax, P. D., Integrals of nonlinear equations of evolution and solitary waves,
Commun. Pure Appl. Math. 21, 467 (1968)
[14] Scott Russell, J., Report of the committee on waves, Rep. Meet. Brit. Assoc.
Adv. Sci. 14, 311 (1844)
[15] Korteweg, D. J. and de Vries, G., On the change of form of long waves advancing
in a rectangular canal, and on a new type of long stationary waves, Philos. Mag.
Ser. 5, 422 (1895)
[16] Zabusky, N. J. and Kruskal, M. D., Interaction of solitons in a collisionless
plasma and the recurrence of initial states, Phys. Rev. Lett., 15, 240 (1965)
[17] Zakharov V.E. and Shabat, A.B., Exact theory of two-dimensional self-focusing
and one-dimensional self-modulation of waves in nonlinear media, Sov. Phys.
JETP 34, 62 (1972)
[18] Ablowitz M.J., Kaup D.J., Newell A.C. and Segur H., The inverse scattering
transform - Fourier analysis for nonlinear problems, Stud. Appl. Math., 53
(1974) 249-315
[19] Beals R. and Coifman R.R., Scattering and inverse scattering for first order
systems, Commun. Pure Appl. Math., 37 (1984) 39
[20] Beals R. and Coifman R.R., Inverse scattering and evolution equations, Commun. Pure Appl. Math., 38 (1985) 29
[21] Zakharov V.E. and Shabat A.B., Interaction between solitons in a stable
medium, Sov. Phys. JETP 37, 823 (1973)
[22] Kulish P.P., S.V. Manakov and Fadeev L.D., Comparison of the exact quantum
and quasiclassical results for a nonlinear Schr
odinger equation, Theor. Math.
Phys. 28, 625 (1976)
[23] Zakharov V.E. and Manakov S.V., JETP Lett. 18, 243 (1973)
[24] Zakharov V.E. and Manakov S.V., The theory of resonant interaction of wave
packets in nonlinear media, Soviet Phys. JETP 42, 842 (1976)
83

[25] Zakharov V.E. and Manakov S.V., Asymptotic behavior of nonlinear wave systems integrable by the inverse scattering method, Soviet Phys. JETP 44, 106
(1976)
[26] Gerdjikov V.S. and Kulish P.P., Completely integrable Hamiltonian systems
connected with the non self-adjoint Dirac operator, Bulg. J. Phys. 5 (1978) 337
(in russian)
[27] Leon J., The Dirac inverse spectral transform: kinks and boomerons J. Math.
Phys. 21, 2572 (1980)
[28] Asano N. and Kato Y., Non-self-adjoint Zakharov-Shabat operator with a potential of the finite asymptotic values. I. Direct spectral and scattering problems,
J. Math. Phys. 22 (1981) 2780
[29] Asano N. and Kato Y., Non-self-adjoint Zakharov-Shabat operator with a potential of the finite asymptotic values. II. Inverse Problem, J. Math. Phys. 25
(1984) 570
[30] Klaus M., Shaw J.K., On the eigenvalues of Zakharov-Shabat systems, SIAM
Jour. Math. Anal. 34 (4): 759-773 (2003)
[31] Boiti M. and Pempinelli F., The spectral transform for the NLS equation with
left-right asymmetric boundary conditions, Nuovo Cimento 69, 213 (1982)
[32] Prinari B., Ablowitz M. J. and Biondini G., Inverse scattering transform for the
vector nonlinear Schrodinger equation with non-vanishing boundary conditions,
J. Math. Phys. 47, 063508 (2006)
[33] Kawata T. and Inoue H., J. Phys. Soc. Jpn. 44 (1979) 1722

84

S-ar putea să vă placă și