IN D I I Springer Series in Nonlinear Dynamics

M. Lakshmanan (Ed.)

Introduction and Applications

Proceedings of the Winter School,
Bharathidasan University, Tiruchirapalli,
South India, January 5-17, 1987

With 29 Figures

Springer-Verlag
Berlin Heidelberg New York Tokyo
Professor Dr. Muthusamy Lakshmanan
Department of Physics, Bharathidasan University, Tiruchirapalli-620 024,
Tamil Nadu, India

ISBN-13: 978-3-642-73195-2 e-ISBN-13: 978-3-642-73193-8

001: 10.1007/978-3-642-73193-8

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broad-
casting, reproduction on microfilms or in other ways, and storage in data banks. Duplication of this
publication or parts thereof is only permitted under the provisions of the German Copyright Law of
September 9, 1965, in its version of June 24, 1985, and a copyright fee must always be paid. Violations
fall under the prosecution act of the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1988
Softcover reprint of the hardcover 1st edition 1988
The use of registered names, trademarks, etc. in this publication does not imply, even in the absence of
a specific statement, that such names are exempt from the relevant protective laws and regulations and
therefore free for general use.

2153/3150-543210
Preface

It is more than 20 yeal's since the concept of the soliton was introduced into
nonlinear dynamics by Zabusky and Kruskal in their now famous numerical
experiments on the Kortweg-deVries equation. Since then the field has grown
in an almost exponential manner and has now entered a period of stability and
high respectability. It has attracted the attention of researchers in all mod-
ern areas of mathematics, physics, engineering and biology. On the one hand
various mathematical concepts such as prolongation structures, jet bundles,
space curves and surfaces, gauge-equivalence, Lie-algebraic properties includ-
ing Kac-Moody and Virasoro algebras, symplectic structures, Lie-Backlund
symmetries, singularity structures, and so on have been attributed to soli-
ton properties, while on the other hand numerous applications have been
found in such wide areas as fluid dynamics, lattice theory, plasma physics,
condensed matter physics, superconductivity, magnetism, nonlineal' optics,
particle theory, general relativity, aerodynamics, meteorology and electrical
networks.
The Science and Engineering Research Council of the Government of In-
dia, Department of Science and Technology (DST), has formulated a pro-
gramme of annual summer/winter schools to encourage research by younger
scientists in the frontier al'eas of nonlinear phenomena. A winter school in
this series on the topic "Solitons" was held at the Bharathidasan University,
Tiruchirapalli, South India, January 5-17, 1987. This book contains the pro-
ceedings of this winter school. It includes eighteen articles by the speakers
at the winter school (of which two articles were given in absentia) and six
contributions by participants.
The book consists of five sections. The first section (Part I) deals with
introductory remarks on integrability and dynamics and historical aspects of
the solitary wave which eventually led to the concept of the soliton. Part
II deals with the mathematical theory of solitons in both 1+1 and 2+ 1 di-
mensions. The inverse scattering transform (1ST), Lie-Bacldund symmetry,
singularity structure and integrability aspects of nonlinear evolution equa-
tions are discussed here. Then in the next section (Part III) lattice solitons
are considered. The quantum field theoretical and statistical mechanical as-
pects of solitons are described in Part IV. Finally in Part V a few selected
physical and biological applications are considered.
We al'e extremely grateful to the DST for its financial support and to
Professor A. Gnanam, Vice-Chancellor, Bharathidasan University for his en-

v
thusiastic encouragement. The school was organized in collaboration with
Professor P.K. Kaw, Institute for Plasma Research, Gandhinagar, and his
support is also gratefully acknowledged. My colleagues offered me unflinch-
ing cooperation in this endeavour and I am particularly indebted to Dr. R.
Sahadevan, Messrs. S. Raj as ekar, K. Porsezian, and S. Parthasarathy for
their help. Finally, I record my appreciation of the very efficient typing of
Mr. S. Venugopal of the entire manuscript.

Tiruchirapalli, India ivI. Lakshmanan

August, 1987

VI
Contents

Part I Introduction
Inaugural Address - The Dynamics of Dynamics
By E.C.G. Sudarshan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
"The Wave" "Par Excellence" , the Solitary Progressive Great Wave of
Equilibrium of the Fluid: An Early History of the Solitary Wave
By R.K. Bullough (With 8 Figures) . . . . . . . . . . . . . . . . . . . . . . . . 7

Part II Mathematical Theory: 1ST, Symmetries, Singularity

Structure and Integrability

Topics Associated with Nonlinear Evolution Equations and Inverse

Scattering in Multidimensions
By M.J. Ablowitz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Inverse Problems and a Unified Approach to Integrability in 1, 1+1
and 2+ 1 Dimensions *
By A.S. Fokas and V. Papageorgiou (With 3 Figures) ........... 66
Gauge Unification of Integrable Nonlinear Systems
By A. Kundu (With 3 Figures) . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Prolongation Structure in One and Two Dimensions
By A. Roy Chowdhury . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Integrable Equations in Multi-Dimensions (2+1) are Bi-Hamiltonian
Systems
By A.S. Fokas and P.M. Santini ............. . . . . . . . . . . . . .. 118
Painleve Analysis and Integrability Aspects of Nonlinear Evolution
Equations
By M. Lakshmanan and K.M. Tamizhmani .................. 145
Generalised Burgers Equations and Connection Problems for Euler-
Painleve Transcendellts
By P.L. Sachdev . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

VII
Backlund Transformations and Soliton Wave Functions*
By J .A. Rao and A.A. Rangwala .......................... 176
Comparison of Some Numerical Schemes for the K-dV Equation*
By A. Hasan and M.S. Kalra ........... . . . . . . . . . . . . . . . . .. 184
K-dV Like Equations with Domain Wall Solutions and Their
Hamiltonians*
By Bishwajyoti Dey ................................... 188

Part III Lattice Solitons

Lattice Solitons and Nonlinear Diatomic Models
By P.C. Dash ........................................ 196
Recent Results in Toda Lattice
By Z. Popowicz 212
Construction of Exact Invariants for One- and Two-Dimensional
Classical Systems*
By R.S. Kaushal ...................................... 226
Nonlinear Chains and Kink-Impurity Interactions*
By Bishwajyoti Dey ................................... 233

Part IV Statistical Mechanics and Quantum Aspects

Quantum Solitons: An Overview *

By R. Rajaraman ..................................... 240
Soliton Statistical Mechanics: Statistical Mechanics of the Quantum
and Classical Integrable Models
By R.K. Bullough, D.J. Pilling, and J. Timonen ............... 250
Exactly Solvable Models in Statistical Mechanics
By M. Wadati and Y. Akutsu (With 12 Figures) 282

Part V Applications: Physics and Biology

Solitons and Some Other Special Solutions in Field Theory

By K. Babu Joseph. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 308
Solitary Waves of the "2-Dimensional Ferromagnet" *
By R. Rajaraman ..................................... 321
Soliton Propagation in Optical Fibres
By A. Kumar (With 2 Figures) ........................... 328

* Contribution by the Participants

VIII
Davydov's Soliton
By A.C. Scott . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 343
Generalized Nonlinear Schrodinger Equations in Quantum Fluid
Dynamics*
By B.M. Deb and P.K. Chattaraj ......................... 359

Index of Contributors . . . . . . 367

IX
 Part I

Introduction
Inaugural Address - The Dynamics of Dynamics
E.G.G. Sudarshan
The Institute of Mathematical Sciences, Madras 600113, India

A dynamical system is defined by a collection of configurational coordi-

nates and equations of motion obeyed by them. These equations of motion
may be generated by a suitable principle or may themselves be postulated.
Given such equations of motion we would like to solve them so that the
dynamical variables at any time may be determined as a function of
the initial variables and time. For a system which is a generalization
of a Newtonian system these would be even in number. The central problem
of dynamics is the determination and characterization of the solutions.
Naturally if we can solve the problem completely then we could consider
various aspects of the solutions including the dependence of the solution
on the initial" data. Except for the really trivial cases, even in
relatively elementary examples there are interesting dependences and
qualitatively new features emerging. For example if we consider the
elementary problem of uniform acceleration, say a projectile moving
vertically in terms of the initial position so' initial velocity u
and the acceleration due to gravity -g, the distance s travelled in
time t i s :

s (, )

But if we ask the time t, at which the distance s, is reached we have

to solve a quadratic equation in which s,-so' g and u appear as coeffi-
cients. We ma~ get no solution, one solution or two solutions depending
on the set of initial values.

Returning to the question of a general dynamical system, wnat con-

stitutes a solution? Most of the time "reducing to quadratures" is
considered as having solved the problem, though we may not get the
solutions in terms of elementary functions. In most cases singular
integrals and singularities may appear. One immediate consequence of
this is that the solution may become multiple valued and thus imply
an unexpected richness of the dynamics.

2
 If the equations of motion of a system can be reduced to a polynomial
in a single dynamical variable and its time derivative equated to zero,
the problem is reducible to quadratures in terms of Abelian integrals.
These have pole and branch point (including logarithmic) singularities.
Correspondingly, the dynamical system develops qualitatively new physi-
cal behaviour. For simple harmonic motion the singularity corresponds
to the limits of the simple harmonic motion.

In a recent paper M. Lakshmanan and R. Sahadevan [1 ] have given

a succinct exposition of nonlinear dynamics from the point of view of
integrability and Painleve analysis with many standard examples and
applied the method to two,three and N~oupled quartic anharmonic oscil-
lators [2].

There is a close connection between integrable systems and Lie

groups. Gelfand and Kirillov have shown that under very general condi-
tions the generators of a Lie algebra of rank y and dimension N=2n+ y
can be realized in terms of rational functions of n pairs of canonical
variables and y unknowns. If one of these generators turns out to
be the Hamiltonian, then the dynamical evolution can be viewed as a
one-parameter group of transformations generating an orbit. Naturally
this is true not only for polynomial Hamiltonians but also Hamiltonians
which are rational functions.

But few systems of practical interest are integrable; and so we

must resort to a study of qualitative dynamics. The state of the system
can be plotted as a point in phase space; and the evolution in a small
but finite time maps this point to another point, unless singularities
intervene. These phase space maps can be viewed as an alternate form
of (discrete time!) dynamics, and one could ask questions about longtime
behaviour and other qualitative aspects. Among dynamics so defined
we may identify systems which "mix" (in which the long-time evolute
is independent of the initial conditions) and those in which the number
of constants of motion are much less than normally expected.

Another class of questions of interest involve stability and secular

behaviour. These questions are expected to be complicated for nonlinear
systems, but they can become nontrivial even for innocent looking systems.
For example, consider a Hamiltonian system with n degrees of freedom
whose Hamiltonian is quadratic in the phase space variables. The equa-
tions of motion are all linear and finite time solutions all exist.
However the diagonalization is nontrivial (though already done by

3
Williamson [3]). It is only in one of the Williamson classes is the
solution properly bounded.

Just because the equations of motion of a dynamical system have

been integrated it does not follow that the system behaves in a satis-
factory manner. If we consider a system with a central inverse cube
force added to an inverse square force, the orbit would be a rosette
made from a precessing ellipse. There are more startling examples
of integrable chaotic systems. In fact the most acceptable forms of
chaos are generated by such mechanisms.

We are thus led to the unexpected and unsettling recognition that

the lawfulness and the determinism embodied in the orbits of an integ-
rable dynamical system does not automatically imply our intuitive notions
of continuity and stability. Slight disturbances could dramatically
alter the orbits; and the conserved quantities of dynamics vary irregu-
larly over the phase space. Long term trends cannot be predicted on
ideas abstracted from simple dynamical systems.

When a dynamical system has nonlinear equations Qf motion, the

dynamic inertia of the system becomes dependent on the configuration.
If it happens that this dynamic inertia tends to vanish these are points
of maximum fluctuation where even a miniscule change in the configuration
can cause a substantial change in the outcome. Gone is the smooth
dependence of the outcome on the initial conditions.

Such irregular configurations could also result from constraints.

Given a system with holonomic constraints we can get rid of the cons-
traints, at least locally, by choosing generalized coordinates. But
with nonholonomic constraints we have to evolve a whole new theory of
constrained systems. The extraction of the true degrees of freedom
and the true equations of motion are nontrivial tasks and the question
of uniqueness must be investigated in each case.

The elimination of constraints, even when they are holonomic, may

not be possible globally. A simple example is provided by a charged
particle moving in a monopole magnetic field. In this case the configu-
ration space is ]R3 with the origin deleted, and the equa·tions of motion
can be written down and seen to be explicitly rotationally invariant.
However, the Lagrangian formulation of the problem requires. the introduc-
tion of the vector potential. This vector potential cannot be chosen
in a rotationally invariant manner; and any choice involves a "line
singularity" along which the potential cannot be defined. The proper

4
way to handle this problem is to use fibre bundle formalism and identify
the choice of any vector potential as a particular section of the fibre
bundle. In this case the structure is quite simple; the group space
of SU(2) may be seen as a fibre bundle on s2, the unit sphere, as base.
Such dynamics on manifolds is not any longer a curiosity, but appropriate
to many problems of dynamics particularly to gauge field theories con-
sidered as dynamical systems with infinite number of degrees of freedom.

The invariance group of the differential equations of motion leads

to integrals of motion, and serves to simplify the problem of complete
integration of the equations of motion. In the Lagrangian form if
the Lagrangian is invariant under a group of transformations, so will
the equations of motion be invariant. The reverse is not necessarily
true. The equations of motion may exhibit quasi-invariance and change
of the Lagrangian by a total derivative which mayor may not require
a redefinition of the generators of change. For example, the system
moving in two dimensions having the Lagrangian

1 ·2·2 ..
2 m(x + y ) + eB(xy - xy) (2 )

possesses a modified translation invariance with generators a +eBa ,

x Py
ay
which commute amongst themselves. On the other hand the

system with Lagrangian

..., 1··2 .. 1 2
~= 2 m(x - y) + e(xy - xy) + 2(x - y) (3)

is only quasi invariant under the translation ax + a y ; but by

the addition of ~t {;(x - y2)} this becomes strictly invariant under
2
the same translation. Of course, the equations of motion are unchanged
by this addition.

In the study of mechanics we have come a long way. From the idealized
free particles and the two-body celestial mechanics problems we get
the impression that mechanics is an exact strictly casual discipline
with solutions which may be difficult to compute but which are generally
well behaved. This good behaviour includes smooth variations of the
trajectory with regard to specification of initial data, and regular
longtime behaviour. For nonlinear systems, even relatively simple
ones, neither aspect of good behaviour may obtain. Catastrophes and
singularities may vitiate the first tendency; and the discovery of
completely integrable chaotic systems puts in evidence unexpected possi-
bilities in longtime behaviour. It is a classical result of celestial

5
mechanics that present knowledge is unable to predict whether the moon
would escape or undergo capture in the moon-earth-sun system; but
many more new features come to light in simple dynamical systems with
nonlinear interactions, including the existence of mixing systems.

The generalization of mechanics to general topological manifolds

also introduces a new flavour to dynamics itself; and reminds us that
we have not always been dealing with the most natural frameworks in
mechanics. We are in a period where older results and problems are
restudied with new insights and points of view.

It is therefore appropriate that Bharathidasan University has

organized this winter school on 'Solitons'. I am honoured to inaugurate
it.

REFERENCES

[1] M. Lakshmanan and R. Sahadevan, Painlev~ analysis, Lie-symmetries,

and integrability of coupled nonlinear oscillators of polynomial
type, Physics Reports, to be published; R. Sahadevan, Painlev~
analysis and integrability of certain coupled nonlinear oscillators,
Ph.D. Thesis, University of Madras, 1986 (Unpublished).
[2] M. Lakshmanan and R. Sahadevan, Phys. Rev. A31 (1985) 861;
R. Sahadevan and M. Lakshmanan, Phys. Rev. A33 (1986) 3563.
[3] J. Williamson, Amer. J. Math. 58 (1936) 141.

6
"The Wave" "Par Excellence", the Solitary Progressive
Great Wave of Equilibrium of the Fluid:
An Early History of the Solitary Wave
R.IC Bullough
Department of Mathematics, U.M.I.S.T., P.O. Box 88,
Manchester M601QD, United Kingdom

It is shown how in 1876 Rayleigh resolved the conflict between Russell

on the one hand, and Stokes and Airy on the other, about the nature
of the solitary ~ave and the formula for its velocity of propagation
c. However, Boussinesq had already done this in 1872 when he published
the Boussinesq equation and gave its solitary wave solution. The
fundamental articles by Russell of 1840 and 1844 which first introduced
the solitary wave and gave the formula c = !g(h+k) for its velocity
are surveyed. They show that he understood the collision properties
of solitons in 1835, though these objects and their mathematics emerged
only some 130 years later.

1. INTRODUCTION

This lecture is concerned with a few points surrounding the early history
of the solitary wave and of the soliton it gave birth to.

Because the 1973 review of solitons [1] made its now well known
quotation from John Scott Russell's 1844 paper [2], everybody knows
how in the month of August 1834 he rode his horse along the banks of
a certain Scottish canal [3] in pursuit of a disturbance of the water
which in particular Stokes [4], and more latterly ourselves have called
a 'solitary wave'. Recall Russell 'was observing the motion of a boat
which was rapidly ~rawn along a narrow channel by a pair of horses
when the boat suddenly stopped--not so the mass of water in the channel
which it had put in motion: it accumulated round the prow of the vessel
in a state of violent agitation, then suddenly leaving it behind, rolled
forward with great velocity, assuming the form of a 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 (He) followed it on horseback and overtook it still rolling
at 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

7
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 encounter with the singular and beautiful
phenomenon which I have called the Wave of Translation, a name which
it now generally bears: which I have since found to be an important
element in almost every case of fluid resistance, and ascertained
to be of the type of that great moving elevation of the sea, which,
with the regularity of a planet, ascends our rivers and rolls along
our shore~. This 'first chance encounter' was actually reported first
of all in an earlier paper [5] and in a rather more prosaic fashion
( Re f. [5], p. 6 1 ) .

Notice that Russell is using 'Wave of Translation' not 'solitary

wave'. As we shall see it was certainly he who introduced the word
'solitary' to describe it, and on the p. 61 of Ref. [5] he used 'a
large solitary progressive wave' (Russell's italics) propably the
first such use. But generally he preferred 'Wave of Translation',
or 'The Wave' ([5], p. 61) as it is used in the title of this lecture.
It was Wave of Translation he continued to use throughout his life
and finally in the posthumous book [6], 'The Wave of Translation in
the Oceans of Water, Air and Ether' published in 1885, on which I
comment shortly. However, I shall follow Stokes [4], Rayleigh [7],
and indeed current practice, and use the term 'solitary wave' for
Russell's 'Wave' here; and this lecture is concerned with the history
of the solitary wave from August 1834, when Russell first saw it,
upto 1876 when Rayleigh [7] calculated its profile.

Of course the history of the solitary wave, or even of this parti-

cular solitary wave, does not stop in 1876. In 1895 Korteweg and
de Vries published their paper [8] in which they gave their now famous
equation. And they also gave the solitary wave solution of that equation.
It was probably only then, as we shall see, that the controversy sur-
rounding the whole idea of the solitary wave as Russell conceived it
ceased, and during the period 1876-95 a number of papers contributed
to its discussion--two by McCowan [9,10] amongst others.

These days we usually quote the KdV equation in a scaled form

such as [11]

o , ( 1. 1 )

au
where u t means at' etc. And its solitary wave solution is then

8
 u = (1. 2)

in which ~ is a free, real, parameter. Both the equation (1.1) and

its solution (1.2) actually refer to a frame translating at the 'sound'
speed, but Russell was not concerned with subtleties like that. However
he was very much aware of the formula for that sound speed, namely

(1. 3)

in which g is the acceleration due to gravity and h, using Russell's

terminology [2], is the depth of undisturbed water in the canal.
Equation (1.1) is thus in a frame translating at Co with Co set equal
to unity.

Following the discoveries of Zabusky and Kruskal [12], and Gardner,

Greene, Kruskal and Miura [13], we now know that (1.2) is not just
a solitary wave: it is a soliton with the remarkable collision proper-
ties of solitons [12]. Viewed in these terms, and without any computer
power at his disposal, Russell's mathematical capabilities were rela-
tively weak. He knew [2] Euler's 'general formula for the motion of
fluids in the Memoirs of the Academy of Sciences of Berlin' [14]. And
he seems to have had a useful knowledge of classical texts like
Lagrange's [15] and Laplace's [16], and evert Poisson's [17], while
he particularly appreciated [2] a work [18] by the two brothers Weber,
Professors at Leipzig and Halle respectively. But it was On his acute-
ness of observation and a strong if variable physical insight and under-
standing on which he relied; and from his experiments on water waves
in channels constructed in the laboratory during 1834-40 [2], as well
as the earlier experiments he performed on the canal [5] in the period
1834-35, he already knew that the solitary wave he had first seen
generated in 1834 had this collision property. Perhaps what prevented
him from stressing this particular, and to us so much more remarkable,
feature was the m~re compelling need he faced to justify to his scienti-
fic peers the idea of the solitary wave itself. It is with this aspect
that this historical note is concerned.

2. JUSTIFICATION OF THE SOLITARY WAVE:

BOUSSINESQ'S PAPER

,Although, as I shall show, Rayleigh's paper [7] should have settled the
whole argument in 1876, it was settled even before this in a remark-
able paper by Boussinesq presented in 1871 [19] and published in 1872
[20]. Moreover, and despite the discussion between 1876 and 1895 cul-

9
minating in the paper by Korteweg and de Vries [8], this paper might
in one respect have opened up the problem again rather than closed it.
This was because much of Russell's own case for the existence of the
solitary wave rested on the exact truth of his formula for the velocity
of propagation of that wave

c = !g(h+k) (2•1)

which he first gave as such in [2]: again the terminology is Russell's,

and k is the height of the peak of the solitary wave above the surface
of the undisturbed water. Obviously (2.1) is a natural generalisation
of (1.3) and I believe this is how Russell found it. However he then
went on to demonstrate its truth by experiment [2]. The point now
is that the formula (2.1) is not quite the velocity of the solitary
wave solution (1.2) of the KdV equation (1.1), even when that solution
is placed in the laboratory, rather than the moving, frame. I establish
the actual error shortly. Later I show how Korteweg and de Vries [8]
really did settle the whole matter, however.

Still the matter was settled by Boussinesq in 1871-72. For in his

paper [20] he gave the analysis by which, starting from Euler's equation
for the conservation of momentum under pressure gradients Vp and body
forces vn in an incompressible fluid of density p

DOt pu
-
= p(u t + (u.V)u)
- - -
= -V
p
- vn, u (u,v,w) (2.2)

(in a modern notation), together with the equation for conservation

of mass

v.u o , (2.3)

he reached the partial differential equation

(2.4)

h is the depth of undisturbed water and Co is given by (1.3).

Boussinesq also gave the solitary wave solution of this equation, namely,

u(x,t) = k sech 2 (~ $ (x - ct») (2.5)

The height at peak of this solitary wave is indeed k, while its velocity
c proves to be given by (2.1) exactly.

10
 One easily scales (2.4) to

called the Boussinesq equation. The solitary wave solution of (2.6) is

and

2 (2.8)
c

The Boussinesq equation is integrable and (2.7) is a soliton; but the

spectral problem which solves it is a third order (or 3 x 3 matrix)
spectral problem [21], and so it is rather more complicated than the
2
Schrodinger spectral problem -~xx + u~ k ~, already familiar from
wave mechanics, which solves the KdV equation (1.2) [13]. It is there-
fore fascinating to speculate on the history of the soliton itself if
Boussinesq's paper of 1872 had then, and subsequently, received the
attention it deserved. It is just possible that this meeting at
Tiruchirapalli would not have taken place had it done so!

Of course the Boussinesq equation (2.6) 'contains' the KdV equation

(1.1). ror in reaching (2.4) terms effectively of order O(k 2 /h 2 ) are
dropped. And to this order (2.4) can be 'factorised' so that (2.6)
is written as

o • (2.9)

A solution of

U
t + u X + -2
1 (l2U 2 + u I
XX) X
o (2.10)

solves (2.9), and such a solution is

221
u = >,. sech 2" >,. (x - c 1t ) (2.11)

with

1+2.>,.2 (2.12 )
2

Evidently

(2.13)

11
and c~ differs from c 2 , equation (2.8), by O{A 4) = 0{k 2 /h 2 ). On the
other hand in a moving frame, so that a/at + a/ax ~a/at, (2.10) easily
scales to the KdV equation (1.1), so that (2.13) makes the point that
the KdV soliton moves at a speed slightly different from Russell's
formula (2.1)! Still it was at 0{A 2 ) that the argument raged, as we
shall see, and A2 ~ 1/5 in Russell's first experiments [5] on his canal.
Larger values of A2 arise in the more detailed laboratory experiments
reported in [2], however: the largest value quoted seems to be A2 = 1/3
(Ref. [2] Table VII, p. 336) but values of A = 1 were examined for
Russell [2] certainly knew that the waves broke in this case as he points
out (e.g. [2], p. 340). Rayleigh [7] quotes Airy [22] as saying the
wave always broke in this case and shows why. Rayleigh's actual remarks
in this connection are given at the very end of this paper.

In the remainder of this note I survey some aspects of Russell's

work, that due to Airy [22] which disagreed with Russell's, some due
to Stokes [4] which did likewise, and Rayleigh's paper [7] which confirmed
both the solutions (2.5) or (2.7) (upto (k 2 /h 2 )!) for Russell's solitar~
wave and his key formula (2.1) for the velocity of that solitary wave
exactly.

3. THE BOOK 'THE WAVE OF TRANSLATION' AND

RUSSELL'S PAPERS OF 1840 AND 1844

Although almost everybody alive must now know of Russell's 1844 paper
[2], few have read it, and only the well-known quotation from it survives.
Still fewer people are aware that the 1844 paper was preceded by the
longer paper [5] published in 1840 in the Transactions of the Royal
Society of Edinburgh. These two papers, [2] and [5], contain Russell's
permanent contribution to the evolution of our science. In a working
lifetime of some 56 years (he was born in 1808, had studied at three
of the four Scottish universities, Edinburgh, St. Andrews and Glasgow,
before graduating from the last at the age of 16, and died in 1882)
he published some 49 scientific articles [ll]--including 21 British
Association Reports, 4 Edinburgh Royal Society Transactions or Pro-
ceedings, and one Royal Society (of London) Proceedings [23]. (He
was elected FRS in 1847.) He also wrote at least three books, namely
the posthumous [6], an original, and indeed seminal, contribution to
Naval Architecture 'The Modern System of Naval Architecture' [24],
and a pedagogical contribution a 'Systematic Technical Education' [25].
But he never had a permanent academic or scientific appointment {Sir
w. R. Hamilton, the great Hamilton, wrote of Russell as a 'person of
active and inventive genius' in support of his application for the

12
Chair of Mathematics at Edinburgh University in 1842; but this was
not enough to get Russell the job). Russell spent the period upto
1846 on his scientific researches often in his spare time [5). By
1855 he was certainly building the great 12,000 ton iron ship 'The
Great Eastern' at his own shipyard at Millwall on the Thames, for the
engineer Isambard Kingdom Brunel; and in 1856 he published his major
work on Naval Architecture 'The Modern System'. Further details of
a remarkable life are given in the Appendix to [11) and the biography
[26). He seems to have reflected on his solitary wave throughout that
life and on June 16, 1881, one year before his death on June 8, 1882,
he gave his paper [3) at the Royal Society entitled 'The Wave of trans-
lation and the work it does as the carrier wave of sound'. This was
followed by the posthumous book [6) of 1885.

In the paper [23) Russell apparently argued, as his book certainly

does, that since in general dispersion will obviate any possibility
of transmitting information by sound, the solitary wave, which, virtually
by definition has permanent shape, must be used instead. The 'Wave
of translation' is Russell's solitary wave so the formula (2.1) applies.
-1 -2
From c - 1100 ft .sec g "" 32 ft sec , one finds h+k - 8 miles;
Russell corrects for varying density by a factor 2/3 so the equivalent
depth of the atmosphere is about 5 miles. From there Russell [6) used
the velocity of light for c, the same g, and certain apparently arbitrary
factors -5 x 10 4 , to reach a 'depth of the universe' of 5 x 10 17 miles:
this is in error by only 5 orders of magnitude--quite a respectable
estimate in my view given all the circumstances!

The permanent work in the book [6) is the Appendix. This is simply
a reprint of the 1844 paper [2). That paper is primarily concerned
to report the formula (2.1) which it then validates pragmatically by
showing that the speeds of the solitary waves actually measured by
Russell in the laboratory fit to it: no theory leading to (2.1) is pre-
sented anywhere in Russell's work. The paper also contains a categori-
sation of waves wqich was very much Russell's concern at the time. He
here gives four 'orders': First, the wave of translation--wave of
first order and its two species, positive and negative, though Russell
well recognised the important difference between the positive and nega-
tive cases--see below; second, oscillatory waves, positive and negative
and second order; third, surfaces agitated to minute depth, that is
papillary waves, third order; and fourth, corpuscular Waves, sound waves
in fluids reflecting the existence of molecular forces. There appears
the Table I "System of Water Waves". First order is as above: it is
solitary in character, it can be free or forced, and includes the 'wave

13
of resistance', the 'tide wave', and the aerial sound wave (as explained
already). Waves of the second and third orders are gregarious: order
two includes stream ripples, wind waves, ocean swell, stationary or
progressive, free or forced; third order includes 'dentate waves' and
'zephyral waves', free or forced. The corpuscular wave of fourth order
is solitary and includes the water and sound waves.

That Russell knew about soliton break-up is evidenced by the Fig.6

in the Plate 47 of the 1844 paper [2]: this Plate 47 is now reproduced
here as the Fig. 1. One sees the first example of break-up is positive
and so really is soliton break-up. However, the second example shows
break-up into a pOlsitive wave leading and a negative one following.
Still, Russell certainly knew the negative wave is not really a soliton
at all--or even a solitary wave: its speed did not conform to the formula
(2.1) as the experiments he reports in the Fig. 5 of Plate 48 of [2]
showed, while in the Plate 52 of [2] he shows how the negative wave
though at first solitary breaks up into gregarious oscillating waves
of order two. The Plates 48 and 52 of [2] are here reproduced as the
Figs. 4 and 7.

The 'wave of resistance' of first order referred to in the Table I

and appearing already in the quotation from [2], is a reference to
Russell's work in the earlier paper [5]. In this long paper he is not
so much concerned with the velocity formula (2.1), which he quotes in
words only, but with the very practical problem of the resistance of
boats towed (or travelling) through water of finite depth. One should
recall that Russell was actually on his horse on the banks of canal
in the month of August 1834 charged by the Union Canal Company to deter-
mine the efficiency of canals for the possible transport of steam driven
barges £,3]. The conventional wisdom of the time, due to Newton,
Bernouilli and d'Alembert to whom he refers, was that the resistance
R was simply proportional to the square v2 , of the velocity V of the
boat through t~e water. Being aware as he says [5] 'of the very imper-
fect state of that part of Theoretical Hydrodynamics which relates
to the Resistance of Fluids to the Motion of Floating Bodies, and that
there had been found in its application to the solution of practical
questions, discrepancies so wide between predicted results and the
observed phenomena, as render the principles of the theory exceedingly
false guides, when followed as maxims of art' Russell determined to
investigate the situation by a series of experiments which he carried
out in the summers of 1834 and 1835. He found instead ·of R ocv 2 that
resistance X plotted against velocity Y followed curves like those

14
 ' :.lI +-l~_
' _ _ __ --.

1t---=--===---===x
.i=:- --
Jd=:==c===--
1 r~
-- -------. --:;- -.

'0 '---
' --- ~~--~
__--- l:J0TIDJ~i~----
'l ll - -

'lte:-
" ~
r' - .-
------------------
.!I I
~

.:
r--------~--------
I--l
;~

--~~""-

--
1'-' ~ ~n,,'.
-----==-=----"'"--- lt~-~--~====~.~~====~---'
.,
----
~ 11
.1 == .' ~-------------------~~~=.~---------~
------ -~------~>~----
----------

Fig. This Figure is the Plate 47 of Ref. [2]. This plate shows the
ways in which Russell created solitary waves in troughs some 6 inches
deep in the laboratory. For soliton theory the two most interesting
Figures in this plate are the two at the bottom (Fig. 6) as explained
in the text. Fig. 5 on the plate shows in its four diagrams the motion
of individual water particles during the passage of a solitary wave
(the particles were actually material particles suspended in the
water). The first diagram describes simultaneous motions at different
points; the second, motion of four particles throughout the whole pas-
sage of the solitary wave from A to X (the curves are semi-ellipses
according to Russell and show that there is a net translation of posi-
tion which he takes to be the volume of water to create the wave divid-
ed by the breadth of the channel); the third shows the motions of ver-
tical planes during the passage (small particles were suspended from
stalks in equally spaced vertical planes in the undisturbed water) ;
the fourth shows movement of initially equally spaced horizontal planes
of particles during the passage.

15
 x

----~y;!o-------.,.l: A~=-----'JT~-------Y

AX and AY rectangular. co-ordinates.

Velocity measured on AY, and resistance on AX.
AP the parabola resulting from tho;) squares of the velocities.
AMm R the line of resistance, M the point of fll'st ma:wnum, and In the
s;}cceeding point of minimum.

Fig_ 2 Two sets of observations made by Russell on his canal

of the resistance X of a boat plotted against its speed Y. The
curves P are the parabola of the conventional view at the time
and the curves R are what Russell found in 1834-35(taken from [5]).

shown in Fig. 2: curves Pare R ocv 2 curves R with a maximum at M and

a minimum at m, both depending on the depth h of the fluid, are followed
by the rising curves to R; but, later, for velocities -29 mph, he
argues that these curves too fallaway from a second maximum. In this
later discussion Russell gives an explanation of the whole form. Its
essentials are these: a crucial feature is the emersion (i.e. emer-
gence) of the whole boat: this explains the second maximu~ of resis-
tance at V ~ 29 m.p.h. (which Russell [5] computes as 4g/3) and its
subsequent fall where at 43.8 m.p.h. (V = 2g) the 'floating body emerges
wholly from the fluid and skims its surface'. However, the more pro-
found influence arises, as Russell was the first to realise, from the
generation and propagation of waves. These waves travelled with speed
c given by (2.1), according to Russell, so were unconnected with the
form of the vessel. It was this speed which coincided with the first
maximum at M. When V < c the effect was to generate Russell's 'Great
Anterior Wave' [5], a wholly positive disturbance, near the prow and
a depression, the 'Great Posterior Wave' near the stern. Thus the
vessel tilted and increased its resistance. But for V > c the vessel
could ride horizontally on the water diminishing the resistance to m.
Russell noticed that in this region the destructive power due to what

16
he called the 'stern surge', destructive to the banks of the channel
and dangerous in navigating shallow water, disappeared. Whilst the
great anterior wave of so called displacement was bell shaped, the great
posterior wave of so called replacement was oscillatory, breaking into
the 'surge' or 'breaking' wave. At this stage he therefore distinguished
four species of wave, the 'Ripple or Dentate', the 'Oscillatory', the
'Surge', and '"The Wave" "par excellence", the solitary progressive
great wave of equilibrium of the fluid'. For the vessel itself the
Great Primary Wave of displacement i.e. the great anterior wave, was
the only example of 'The Wave'. 'The Wave' is the solitary wave as
I use it here--so thus the title of this lecture!

Russell noticed the independence of 'The Wave' from its mode of

genesis and that its speeds (8 m.p.h. for the channel 5.5 ft deep he
was concerned with) were independent of the speed of the generating
bodies (e.g. 2, 5, 6 and 12 m.p.h.). Moreover 'Another observation
equally simple served to show that a large or high wave had a greater
velocity than a small one. When a small wave preceded a large one,
the latter invariably overtook the other, and when the large wave was
before the less, their mutual distance invariably became greater' (Ref.
[5], p. 6). This remark certainly supports the formula (2.1), but it
also supports the concept of the soliton also. Moreover, reference
to Plate II of [5] shows that Russell had already drawn the shapes
of different observed examples of his "Wave". And he also draws there
three different examples of soliton break-up which he must therefore
have known about in 1835, and strictly from his observations on the
canal not from the experiments of [2].

Russell was naturally concerned with the problem of moving a boat

into the minimum resistance region m and noted some of the possible
difficulties. 'It appears', he says, 'that increased force applied
gradually to the vessel for the purpose of rendering the velocity of
the body equal to or greater than that of the wave, has the effect
at the same time of increasing at a more rapid rate the retarding forces,
and a limit is soon reached, which it has in many case been found impos-
sible to pass' (later he shows how a singularity may develop at M).
'It is the circumstance of ~he very rapid increase of the resistance
in approximating to the velocity of the wave that has led to the false
idea that there is a final and low limit to velocity in shallow water.
There are circumstances in which this limit is final, the channel being
very shallow, and the boat very bluff in its formation, I have seen
in such an extreme case, when the depth of the channel was about five
feet, the channel laid bare in the stern hollow behind the wave, so

17
that the stern of the vessel no longer floated but rested on the bottom,
while the bow was elevated and buried in a large anterior wave, rising
more than two feet above the level, and overflowing the banks, and
the posterior wave rushed furiously on behind, roaring and foaming,
tearing up the banks of the channel, and threatening the destruction
of the vessel, which, indeed, on stopping, it nearly accomplished,
In such a case the persons in the vessel were not visible from the
shore, being sunk in the hollow between the great anterior and posterior
waves. '

He then goes on to discuss the passage beyond the point M, namely

to m, where the boat rides horizontally upon the wave. Figures 7 and
8 from [5] appearing in Fig. 3 show exactly what he intends. 'But

Fig.3.
(UI rcsl.)

.-........_.._.... __.... --..........~.................. -

~--...-.-.---.---.-----..

Fig. 4.
(nl , miles un hour.)

--=-==-~~-...-..-.. --~-:
... ~- =
.
Fig.6.
(n16 lIlil"" nn I,our.)

.m .. _.. _~~~=--._-
Fig.O.
(AI7t mil .. IIlI bour.)

Fig. 7.-llcbind lb. Wave.

Fig•. B.--TJprn lb. Wave.

_~.g=;=~=====--~.=
___.=._=-n.~_......m:::;2:

Fig. 3 Figs. 3-8 taken from Russell [5]. Figs. 7 and 8 show how a
maximum resistance (Fig. 7) is replaced by a minimum as the vessel rides
on the solitary wave.

18
it will be inquired, how is a vessel to be placed in such circumstances?
How is the extreme resistance of the anterior wave to be vanquished,
and the vessel planted on its summit? This is admitted to be a practical
problem, often of extreme difficulty; sometimes it is impracticable.
There are forms of vessel that do not admit of a position of stable
equilibrium on the top of a wave. Still, however, it is a practical
problem solved everyday on all canals navigated on the Scotch system.
Vassels of greater length than the wave, having a fine entrance, built
of light materials, and drawn by well trained highly bred horses, and
guided by experienced postillions, are raised by a sudden and powerful
jerk to the top of the wave (at from 6 to 8 m.p.h.) and are drawn along
on the summit of the wave with greater ease at 10 or 12 m.p.h., than
at 6 or 7.' (see §IX of [5]).

I have given these various quotations from [5] at length in order

to show Russell's grasp of the situation and incidentally to exhibit
the lively and imaginative character of his writing: I hope it will
stimulate others to search out his articles and read them. They reflect
a much more leisured scientific age, and amaze, stimulate, and infuriate
all at the same'time. In the rest of this paper I want to show- why
Russell had such difficulty in establishing the existence of the object
we now call the solitary wave. Herschel's comment "It is merely half
of a common wave that has been cut off" [27] seems typical of his cri-
rl:ics. To simplify the account here I shall focus only on the objections
raised by Airy [22] and Stokes [4]. This leads to some historical
comment on the shape of Russell's wave. Then, in a final section, I
shall quickly show how Rayleigh [7], in particular, fully justified
Russell's position.

4. THE OBJECTIONS OF AIRY AND STOKES AND

SOME COMMENTS ON THE SHAPE OF THE WAVE

In publishing the 1844 paper [2], which just preceded the appearance
of Airy's treati~e [22], Russell already knew that Airy disagreed with
his formula (2.1) for the velocity c of propagation of a wave of per-
manent shape. Such a wave is necessarily a wave u(x,t) = u(x - ct).
Russell believed his 'Wave' was in this category and, by implication
was the only such wave being necessarily positive and lump shaped i.e.
solitary. Airy believed to the contrary. Rayleigh [7] quotes from
Airy [22] "We are not disposed to recognize this wave as deserving
the epithets 'great' or 'primary' . • . and we conceive that ever since
it was known that the theory of shallow waves of great length was con-
tained in the relation d 2X/dt 2 = gkd 2 X/dx 2 , ... the theory of the solitary

19
wave has been well known". And again "Some experiments were made by
Mr. Russell on what he calls a negative wave--that is, a wave which
is in reality a progressive hollow or depression. But (we know not
why) he appears not to have been satisfied with these experiments,
and has omitted them in his abstract. All the theories of our IV Sec~

tion, without exception, apply to these as well as to positive waves, f

the sign of the coefficient only being changed." That the second comment
is unfair in the light of the material published in [2] is clear--
though Airy might not have seen that. However, it is immediately
obvious that the linear equation given in the first quotation can only
have harmonic solutions albeit, since it is not dispersive, they are
unique in speed as well as of permanent shape: But they are nowise
solitary. More generally Airy [22] found the velocity formula

2 .9. e mh - e- mh
c .9. tanh mh (4.1)
m emh + e-mh m

2 -1
which reduces to co' (1.3), for wave number m« h . However he also ,
included the effect of finite disturbance k to reach

g(h + 3k) . (4.2)

As we shall see Rayleigh [7] subsequently found an argument for the

same result; but, finding it for a distorting wave, he then developed
the very different argument that arrives at (2.1). Boussinesq [19,20]
quotes Bazin as confirming (2.1), for (2.5), and (4.2) with ~k not
3k, for a step function. But in 1844 Russell ~2] was simply very much
concerned to show that Airy's formulae did not fit his actual observa-
tions. And the Fig. 4, reproduced from Plate 48 of [2] shows in its
Figs. 3 and 4 how well he succeeds in doing this. Russell also knew
a result by Kelland [28] quoted in [2] p. 333 as [29]

ah - e -ah }
{ 1 - e: ae (4.3)
e ah + e- ah

in which e: is given as the semi-elevation, h the depth of water in

repose, and A ~he length of the wave. He suggests (Ref. [2],p.334)
that Airy [22] has simply followed Kelland 'over the same ground in
an elaborate paper on waves' and concludes 'a theory of the first order
accurately representing this characteristic phenomenon is still wanting,
a worthy object for the enterprise of a future wave mathematician'.
There proved to be at least two of these, Bousinnesq [19,20] and
Rayleigh [7].

20
 ...
.IWAVES_ (}n(v l--'Ji:m.'l:n~n ifF !iqtUlJII.,II.,

- ..
,..W... , {'...... ' .. /I, ..... ;UI

.
ILI~ 1M
Jl!j.J I". ~•
Yr- T

" " "

~
TO

?~ _..---.1
. /
V'-
/"
.f.- I--"
,
.- .- r-

..
/ ./ .- I-- I-- -
,

/"
•
I-- r- - V'
/"

V
/ /"
r--
V
-- 1--"
~
. .
V V

-----
V ~i'.
.- --
/ V . "'"'-.. f
->- V
V
/ V .
' r--
V
•
l .'
I V
/ V. ~ r-
; / /
/; I--

VI I- :~ '--

Fig. 4 This Figure is the Plate 48 of [2]. Its Fig.l is a check of

the formula (2 . 1) against observations for channels 1.0 to 8.0 inches
deep and wave heights up to 0.54 inches; Fig. 2 is for depths 1.62-8.0
inches deep and wave heights up to 0.96 inches. Figs. 3 & 4 compare
experimental points against (2.1) (bold line) and against four different
formulae due to Airy which Russell does not specifically quote (dotted
lines). Fig. 5 compares the formula (2.1) (curve AB) against observa-
tions for a negative wave in a rectangular channel and formula c 2 =
1 / 2 g(h+k) (curve AB') for a negative wave in a triangular channel.
Fig. 6 plots general results for c (in units of one foot per second)
against (h+k) (in units of five inches): AB is (2.1) for a rectangular
channel; AB' is c 2 = 1/2 g(h+k) for a triangular one.

At this stage it is useful to summarize what Russe ll appa r ent l y

knew. Lagrange [15) f irst integrated the cons e rv . _ion of momentum
equa tion (2.2) in the case of irrotational motion u = V~. This int egra l
is

~t(x,t) + F(t) + 21 ~
2 (4.4)

o (4.5)

21
(by choosing the zero forn ) and this result contains both Bernouilli's
theorem [30] and Torricelli's [31]. Russell well understood Torricelli's
theorem, which he quotes in [5], as w,ell as (4.5) and its application
in the pitot tube. He built a shallow bottomed boat 'The Skiff' [5]
in which he placed a series of pitot tubes to measure the velocity
profile of water passing a boat in connection with the 1834-35 experi-
ments.

The formula (1.3) quoted by Russell in [2] is apparently due to

Lagrange who used the 'method of long waves' to find it [15]. consider
motion in the (x,y)-plane only with an undisturbed surface y = Yo'
and a disturbed one y = n(x,t) + Yo. At an arbitrary point (x,y) within
the disturbed fluid the pressure p there is Po the pressure at the
disturbed surface enhanced by the potential term gp (y 0 + n -y) so that

p - Po = gp(yo +n- y) (4.6)

The content of the method of long waves is to identify y and Yo so

that p is independent of y: then from (2.2), all particles in a plane
perpendicular to x get the same acceleration, and all particles in
this plane remain in the same plane and the horizontal velocity u depends
on x, t only, u u(x,t). Equation (2.2) is

u
t
+ uu
x
= - lp ax
ap (4.7)

and the nonlinear term is simply dropped as being of the second order.
Though this is not a priori a necessary feature of the method of long
waves, there is no dispersion term left in the method to balance it.
From (4.6) ap/ax = gp'ln/clx and so

(4.8)

If E; = I u dt, E;tt = -gn x ' and continuity (2.3) means

~(E; h b)ox n box, (4.9)

ax

b being the breadth of the canal--both expressions being the volume

of fluid which, up to time t, has entered the space between x, x+ox.
From (4.9)

gh E;xx and

(4.10)

22
 2 2
with c~ given by (1.3). This is the relation d 2 X/dt 2 gkd X/dx of
Airy [22] quoted by Rayleigh [7].

The argument given here is really that as presented by Lamb [32].

George Green (of Green's Theorem) considered long waves in a rectangular
canal with slowly varying breadth 8 and depth y and found a wave height
1 1
oc p-2y~' and the velocity formula (1.3) as c =;gy [33]. For a
1 1
triangular canal with one side vertical [35] Green found c = (2gy )2,
a result Russell [2] also applies to the isosceles triangle and confirms
experimentally (see Fig. 6 line AB' of the Plate 48 of [2] in our Fig.4).
Kelland [28] found c = I9A7b for a canal of cross-sectional area A
and breadth b at the surface. But this result Russell [2] apparently
confirmed only for the rectangular and triangular cases disagreeing
otherwise (Ref. [2], p. 355). Our Fig. 5, which is Plate 49 of [2],
shows the different triangular cross-sections Russell worked with.
Nevertheless Stokes [36] says Kelland's formula 'agrees with the experi-
ments of Mr. Russell' simply.

We turn now to the work of Stokes himself. In the British Associa-

tion Report of 1846 [36] Stokes says "It is the opinion of Mr. Russell
that the solitary wave is a phenomenon sui generis in nowise deriving
its character from the circumstances of the generation of the wave.
His experiments seem to render this conclusion probable. Should it
be correct the analytical character of the solitary wave remains to
be discovered". Note that he expects to agree with Russell, and that
already he firmly uses the designation 'solitary wave'. However, in
a paper read on 1 March 1847 and pubiished in the Trans. Camb. Phil.
Soc. [4] "On the theory of oscillatory waves" he effectively concludes
(for axes x horizontal and y vertically downwards) that

y a cos m(x - ct) - Ka 2 cos 2m(x - ct) (4.11 )

-1
(with a = -m Ac g and m arbitrary, but K then fixed and depending
on a, m and h, and c given by (4.1))" is the only form of wave which
possesses the property of being propagated with a constant velocity
and without change of form--so that a solitary wave cannot be propaga-
ted in this manner. Thus the degradation observed by Russell is not
due to the imperfect fluidity of the fluid and its adhesion to the
sides and bottom of the canal but is an essential characteristic of
the solitary wave".

Stokes's argumentation is as follows: He starts from (4.4) in

the form

23
 WAV£ S _ I)rtbrl .

B~ .Iuft!!" ;'" .,F ,lt~ Z-rptniNf'lu.

111. 1.
~ ' ~ :
~ : "~.~! '
'~'~:'- ii"I!•
•
. I. '
~: I !! I
~
• .. I , .T 1 I' ~
___ t
I

'~ . 'l

II
Ii
.0;,"

..
II
...-----

Fig. 5 This Figure is the Plate 49 of [2]. The Figs. 3,4,5 show the
triangular profiles used by Russell in his experiments. The curves AB
describe the crest of the wave and in Fig. 5 this crest is breaking in
the regions aA and Bb. Fig. 1 is the smooth curves drawn through observ-
ed wave heights in order to remove observational error, and similar
curves were drawn for the velocities. These smoothed results were then
plotted in Fig. 2 and compared against (2.1) plotted as the curve AB.

(4.12 )

with continuity

o , (4.13 )

and with

£5£. 0 when y h , (4.14 )

dy
and

~ + £5£. Q£ + ~ Q£ o when p 0 , (4.15 )

dt dx dx dy dy

24
the condition for the free surface remaining the free surface throughout
the motion. The velocities u = d~/dx and v = d~/dy, and p, depend
on x - ct, the condition for permanent profile, together with y. From
(4.12) we can admit in ~ a term linear in t to fix p so

f(x - ct) + Ct (4.16 )

and (x - ct) is called x, while C is set to -gk. Then (4.12) is

(4.17)

so that for u = v 0, p o when y -k. Also, from dependence on

x - ct, (4.15) is

~~ _ c) dp + ~ dp o when p o , (4.18 )
dx dx dy dy

and (4.17) into (4.18 ) gives

dy
2 2
g ~- c d 02cp + 2c
dx
(~ i.....CfJ.
dx dx 2
2
+ ~ d
dy dy
2
i) -(~f d2
_..5e.
dx 2

2
2 ~ ~!LsL _
dx dy dxdy (~)2 d 2p
dy2
0 (4.19 )

when

2
g(y+k) + c ~ -
dx ~ {(~) + (~)2} o . (4.20)

At this stage the argument is exact, but Stokes now first of all
neglects all the (nonlinear) terms in the second row of (4.19), retaining
those in the first row however. He sets

(4.21 )

(y 0 is the undisturbed surface) so that (4.20) implies

g(y + k) + c ( ~x + y~xy) - 21 ( ~x2 + ~y2) o (4.22a)

and by one iteration

Y = -k gc ~x + g
c ~xy (k + _c
g '"x)
or
+ 2g
1 ( ~x2 + ~y2 ) (4.22b)

25
in which y evidently describes the ordinate of the disturbed surface.
He now puts y = -(k + (c/g)~x) in the first two terms of (4.19) and
y = 0 in the second two terms so that

g~y - c2~xx - g(~yy - c2~ xxy ) (k + £~

g x) + 2c(~ x ~ xx + ~ y ~ xy ) = O.
(4.23 )
Thus ~ satisfies (4.13), with (4.14) at y = h, and with (4.23) at y
= O. When ~ is determined (4.22b) gives the ordinate y of the surface
and k is fixed by the condition that the mean value of y shall be zero.

The first approximation to (4.23) is

g~ c 2 d 2 <p = 0 , when y o . (4.24)

dy
dx 2

Put ~ A e(mx+ny) which satisfies (4.13): the boundary

m2 +n 2 =0
conditions are chosen to make m imaginary (so the behaviour along x
is matched to a Fourier series and periodic boundary conditions have
entered). Then

~(em(h-Y) + e-m(h- y )) (A sin mx + B cos mx) (4.25)

since the choice satisfies (4.14). Then one non-trivial contribution

to (4.25) satisfies

mh -mh
c
2 .<I e - e (4.26 )
m e mh + e mh '

which is of course the formula (4.1). Stokes then proves that

d(log c 2 )/d (2mh) < 0 which implies a unique m satisfies (4.26).

By choosing the origin of x

~ = A ( e m(h-y) + e-m(h-y)) sin mx , (4.27)

and this, put into (4.20), yields

y -m Ac g-l (e mh + e- mh ) cos mx (4.28 )

(he sets k = 0 since the mean value y 0) and this is the equation
for the free surface.

For a second approximation he uses ~ from (4.28) in the small

terms of (4.23) retained, so that

26
 o . (4.29 )

Stokes continues "The general value of ~ given by (4.25), which is

derived from (4.13) and (4.14), must now be restricted to satisfy (4.29).
It is evident that no new terms in ~ involving sin mx and cos mx need
be introduced, since such terms may be included in the first approximate
value, and the only other term which can enter is one of the form

B (e 2m (h- y ) + e- 2m (h- y )]sin 2mx . (4.30 )

Substituting this term in (4.29) and simplifying by means of (4.26)

(4.31 )

Moreover, since the term in ~ containing sin mx must disappear from

(4.29), the equation (4.26) will give c to a second approximation"--
so the velocity formula (4.26) is actually unchanged.

If we now set a for the coefficient of cos mx in the first approxi-

mation (4.28),

a = (4. 32a)
and

A -ca/(e mh - e -mh ), (4.32b)

while

em(h-y) + e-m(h-y) 2 (e 2m (h-y) + e- 2m (h-y)) .

-ac sin mx + 3ma c mh -mh 4 s~n2mx.
em(h-y) _ e-m(h-y) (e - e )
(4.33 )

Thus the ordinate of the surface to the second approximation from (4.22b)
is
2 (4.34 )
y a cos mx ~ Ka cos 2mx

with x -+ x - ct, c 2 given by (4.26) , and

-2mh
1 (e 2mh + e + 4 ) coth mh (4.35 )
K m
2 (e mh - e -mh
)
2

which is the statement (4.11).

To proceed further Stokes simplifies to h » A(depth h »length A

of the wave). Then (4.33), (4.34) and (4.35) are

27
 cp - ace-my sin mx

y 1 ma 2 cos 2mx
a cos mx - "2

1
K "2 m 11/>" (4.36 )

and c 2 = g>../211; the parameter k o. He then finds at next order that

(with h »>..),

cp -ac (1

1 2 3 2 3
Y a cos mx - "2 rna cos2mx + 8 m a cos3mx

c = (*) 1/2 (1 + 2:~a2) (4.37)

and only at this order does the wave amplitude a enter the expression
for c. Stokes draws the form of y for a = 7>../80 (for some reason)
but notes the term in a retainer "is almost insensible". His most
significant result is that the figure is not symmetrical above and
below y = O.

Stokes also notes that "It is remarkable that this expression(for

y) coincides with that of the prolate cycloid, if the latter equation
be expanded according to ascending powers of the distance of the tracing
point from the centre of the rolling circle, and the terms of the fourth
order be omitted. The prolate cycloid is the form assigned by Mr.
Russell to waves of the kind considered here (Reprints of the British
.Association Vol. VI p. 448 (Ref. [34], p. 448)". For h/>" not great
the form of the surface is not a prolate cycloid even to second approxi-
mation.

Before discussing further Stokes's results (4.11) and their exten-

sion (4.36) ang (4.37), it is interesting to take up his comment on
the form of Russell's wave. The reference by Stokes to [34] is correct
for 1837 but not for 1847. For in [2] (of 1844) Russell already des-
cribed the shapes of his waves as curves of versines not cycloids(see
below). The paper [34] (of 1837) is the report of a "Committee on
Waves" set up by the British Association in 1836. The committee had
two members--Russell and Sir John Robinson, Secretary to the Royal
Society of Edinburgh. It was charged to determine: first 'What is
a Wave'; second what is the nature of the 'Waves of the Sea'; third
does the behaviour of the 'Tidal Elevation' obey the same laws as other

28
waves; and is its propagation affected by 'Local Winds'. The report
is plainly Russell's report on these questions.

In it he was most concerned to establish what he had already believed

to be true--namely that the tidal wave was an example of his solitary
wave. In September 1836 he carried out experiments on the River Dee,
near Chester, where there is a 5 mile channel of the river straight
and uniform in depth and width. Then he carried out experiments on
the Firth of Clyde, where conditions were less uniform. From these
experiments, and a further series in the laboratory, the Committee
concluded: 1. the certain existence of a 'Great Primary Wave' of
fluid in its laws different from that of the oscillatory waves treated
hitherto; 2. that its velocity c satisfied (2.1); 3. that c is indepen-
dent of the mode of generation; 4. the motion of the particles of
water involves their actual net translation (refer Fig. 5 in our Fig.l);
5. the form of the wave is cycloidal, being first a prolate cycloid,
becoming a cusped 'common' cycloid as the height k increases to h,
and breaking beyond that height; 6. in channels of arbitrary cross-
section c is 'that due to gravity acting through a height equal to
the depth of the centre of gravity of the transverse cross-section
(so (2.1) applies to a rectangular channel and c 2 = ~9(h+k) in a tri-
angular one (refer [35))); 7. that the increased height k in a wedge
shaped narrowing of the channel may follow k.~ b- 1 / 2 (b the breadth
(refer [33))); 8. in a uniformly shelving channel the wave breaks
when k = h; 9. the solitary wave can be normally reflected without
any change but reversal of direction; 10. the solitary waves cross
each other without change of any kind (solitons again); 11. Waves
of the Sea are not of this type but are second order oscillatory; 12.
these waves nevertheless become solitary waves as they approach the
shore and break for k < h; 13. waves at the surface of the sea do not
move with the velocity due to the total depth h+k; 14. that at sea
every 3rd or 7th or 9th wave can be the largest (this is in fact a
reference to waves in deep water governed by the nonlinear Schrodinger
equation) and these, in particular, will break on the shore; 15. the
Tide Wave is the only wave of the ocean which is a wave of first order
(is a solitary one); 16. the tide itself is a compound wave; 17. a tidal
bore is created when the water is so shallow at low water that 'the
first waves of flood tide move with velocity so much less than that
due to the succeeding part of the tidal wave, as to be overtaken by
the subsequent waves, or wherever the tide rises so rapidly, and the
water on the shore or the river so shallow, that
the height of the first wave of the tide is greater

29
than the depth of the fluid at that place. Hence in deep water vessels
are safe from the waves of rivers which injure those on the shore';
18. that because the Tidal Wave is solitary (2.1) means that deepening
a channel means an advance in the time of observation of high water;
19. that likewise spring tides (large k) travel faster than neap tides
(small k) with a consequent effect on tide tables.

This report was followed (37) by a further brief note on the form
of the tidal wave in the Firth of Forth. Again the note identifies
the Tidal Wave with Russell's solitary wave, while 'It appears that,
like the great wave of Translation, tidal waves could not only meet
each other without losing their individuality, but they could pass
over each other when going in the same direction'.

Thus Russell confirms yet again that he had observed the soliton
property. However, he never found the analytical sech 2 formula for
the soliton's shape. His shapes are cycloids in (34); his Wave was
a trochoid in [5], p. 85; and it becomes a 'curve of versines' in [2]
whilst in that paper his cycloid refers to his gregarious oscillatory
waves of order two. Russell shows some of his laboratory observations
of wave shapes in the Plates 50 and 51 of [2]. The latter is reproduced
here as the Fig. 6 and shows two waves breaking. The Plate 52 of [2],
reproduced here as Fig. 7, shows that Russell's 'curve of versines'
1
is a ik(l + cosB) curve, i.e. it is harmonic but rendered positive.
At first sight Russell's insight has really failed him here. But note
that for large k (Figs. 4 and 5 in the Plate) he is already correcting
this harmonic shape by an effect due to the translation of the particles.
The translation in the Fig. 4, AA', is the volume of water used to
create the wave divided by the breadth of the channel, and Russell
has already demonstrated this is the net translation of planes of water
particles (cf. our Fig. 1). The Fig. 6 in our Fig. 7 shows a single
particle path in the case of Fig. 5, where k~h, and three successive
positions of the wave in regard to it.

Evidently Russell would have been in a much stronger position in

1844 if he could have given an analytical form to his wave by analytical
methods. I now return to Stokes's [4] result of 1847, namely (4.11)
and its extension, and indicate what was wrong with it.

The passage to h»A (i.e. m h ~oo) involved in (4.36) and (4.37)

prejudices any discussion with respect to Russell. But Stokes's results
at second order are at first quite general and in particular, for the

30
 - -------
----------------~~-.~-"--------------~~

..
~~~~--=~:
. 2 ====
======. ===
!:::::=: s~=~~=--"-"- - _____ .'
Fig. 6 This Figure is Plate 51 of [2]. The waves are 'drawn by them-
selves' by using pairs of equal waves travelling in opposite directions
and registering their wetting profile in collision. The second four
curves are all drawn this way; the first two, which are breaking, are
apparently found similarly. The last four curves are also apparently
drawn in the same way but the troughs have the sloping bottoms gX, kX,
tX and mX with gradients one in twelve. Otherwise h = 2 inche"s for all
the curves. The dimensions were reduced by Russell by a factor 2/3 and
the parallel lines are really one inch apart (the preceding Plate 50 is
similar but is drawn full size).

'long wave' condition mh small, Ka 2 in (4.35) reduces to 3a 2 /4m 2 h 3 .

Since Stokes's Fourier series must converge he then requires 3a 2 /4m 2 h 3
< a and (a/h)< m2 /h 2 . According to Stokes this resolves an apparent
discrepancy with Airy [22] (Art. 198, etc.) who working under the long
wave condition mh« 1 finds the form y of his wave alters at second
order as the wave proceeds while c depends on wave height at that order
(Airy may even reach his expression (4.2) for c at this point). Evidently
Stokes has (a/h)< m2 h 2 and Airy has mh small with a/h > (mh)2. Thus
the two assume "different physical circumstances".

31
 YAVlla __ (I"u, I .

~ :;::/>;,,, "':r.~~
:~f ~ -
-~ ~-;;.~-~-;~,;-...:..
.~ ..
t ~.l:..--=-,_ _,- - ....:::
.... .

~, ------------------~~~.------------------

~.~~----------------------------------------------------------

~----------~~---1..I~~--,.~--~~~~-----
~al~s 'l __ .

.
Fig. 7 This Figure is the Plate 52 taken from [2]. Figs. 3,4,5 show
Russell's curve of versines description of the shapes of his Waves
and Figs. 4,5 show the corrections he made for the translation of the
particles. Fig. 6 shows the single particle path. Figs. 1 and 2 are
each superb examples of soliton break up: Fig. 2 is generated from a
long low column of water, Fig. 1 by a plate with variable force and
longitudinal velocity. The lower Figs. show the creation of a solitary
negative wave and its break-up into gregarious oscillatory waves of
Order II.

This is almost but not quite the nub of the matter. To second
approximation Stokes has no error and everything depends on the approach
to the long wave limit A =2TIm-1~ 00. In this respect both Stokes and
Airy, who assume A = 00 a priori but then work incorrectly, are wrong.

Certainly Stokes avoids the assumption of p independent of y, charac-

teristic of the method of long waves, and nonlinear terms are retained
at sufficient accuracy in (4.19) at his second approximation. For mh
small the two terms of (4.11) then prove to be the first two in the Fourier expansion
of the square cn 2 of the Jacobian elliptic function cn. And in later
work (see Stokes in [39]) Stokes finds the third term also. Since cn 2

32
Fig. 8 This Figure is Plate 53 of [2] and is not referred to in the
text. It shows reflex ion of a solitary wave from the surface R for dif-
ferent angles of incidence measured from R. The reflected wave decreases
in amplitude with decreasing angle of incidence until at about 15°
incidence it prefers to travel as a single wave by increasing its
height in the plane of R.

~ sech 2 as the period A finally goes to infinity Stokes's real error

has been to impose periodic boundary conditions in all of his analyses.
It was Korteweg ~nd de Vries [8] who pointed out that Stokes [4] had
found the first two terms of the cn 2 function and it was in this way
they finally settled the matter. No such escape seems available to Ai.ry
however. The critically different physics described by periodic boundary
conditions is still not always appreciated [38].

Later Stokes acknowledged his error [39] but Airy never did. His
opinion carried weight well beyond Rayleigh's paper [7] and Korteweg
and de Vries [8] seem to have written their paper in 1895 essentially
for that reason. They quote Boussinesq [19,20], Rayleigh [7] and St.

33
Venant [40) as establishing the theory of the solitary wave but noted
that (in 1895) treatises by Lamb ([32), 2nd edition 1895) and Basset
still assert that Airy was correct in his opinion. The same is true
of the Encyclopaedia Brittanica article on Russell of 1886. Moreoever,
Korteweg and de Vries say that even Rayleigh [7) and McCowan [9) do
not directly refute Lamb's and Basset's assertions and "It is the desire
to settle this question definitively which has led us into the somewhat
tedious calculations which are to be found at the end of this paper".
The KdV equation itself, though noted as "very important" in [8) gets
rather less discussion.

5. RAYLEIGH'S PAPER AND HIS RESOLUTION OF THE PROBLEM

These various remarks in [8) seem quite extraordinary when we actually

look at what Rayleigh did in his paper [7) in 1876. It is somewhat of
a relief to turn to this paper after those of Stokes which are also
"somewhat tedious". The paper [7) is 'On Waves', and in it Rayleigh
exploits his physical insight as usual rather than any extraordinary
manipulative mathematical skills. In these terms his paper [7) should
have settled the matter in England at least, and Russell, a Scot by
birth should have benefitted for he had long since moved to London [26).
Boussinesq [20) working in Paris could be ignored but Rayleigh scarcely.
Yet he seems to have been if Korteweg and de Vries [8), working in
Holland however, were right. McCowan's papers [9,10) suggest very much
that they were.

Rayleigh uses h for the height of the proposed solitary wave above
the undisturbed water, and he uses t for the depth of undisturbed water.
At the risk of confusing, I shall at first keep Rayleigh's usage and
change it to Scott Russell's only when I turn, with Rayleigh, to his
actual analysis for the solitary wave.

Rayleigh first notes (in effect by appeal to (4.5)) that a steady

water velocity U o along x in the undisturbed region will lead to an

increase in pressure inside the solitary wave where, by continuity,

if h > 0, u drops to u. On the other hand the pressure will decrease
91
from the p n = gh, and the possibility of a balance exists. He first
of all assumes the postulate of the theory of long waves, i.e., A»h,
so that the velocity v (in the y-direction) is negligible and u (the
velocity along x) is independent of y: u is of course uniform along
z, i.e., across the canal. Continuity then means

u = (5.1)

34
and

(5.2)

'The principle of hydrodynamics' (namely (4.5)) now means that the

increase in pressure due to the fall from Uo to u is

1 2 2 1 2 2th+h 2
(5.3)
2" p(u o - u ) 2" pU o (Hh)2

while the net gain is therefore

2 2
1 + h/2R.
- gp} 0:. [P:o - gp Jh . (5.4)
r:o (1+h/t)2

Implicitly he now moves to a frame where the velocity Uo is reduced

to still water so the region where the velocity is u moves up the x-
direction (actually he treats a stationary wave and points out this
also gives the velocity of the wave in still water). The condition
for a free surface (no net gain in pressure) is to the order just realised
then u~ = gt, which is Lagrange's (1.3). To second approximation how-
ever it follows that

6p g ph { (1+h/2t) 1}0:._lgPh 2 (5.5)

(1+h/t)2 2 t

so that p is defective wherever h * 0, and it is impossible for a long

wave of finite height to be propagated in still water 'without change
of type'. Evidently (from (5.4)) if h > 0 one obtains a better result
if Uo is increased, thus making the left hand expression less negative.
But if h < 0 one obtains a better result if Uo is decreased, making
the curly bracket there less positive. The wholly positive wave Russell
conceived therefore has a velocity c >c o ' while a negative wave has
c < co.

Rayleigh now notes that (5.1) is

u' _ u - Uo (g/t)1/2 h (5.6)

(since he has just shown h is infinitesimal). For vanishingly small

h, as required, (5.6) is the condition that the disturbance moves up
the positive x direction. If the condition is violated the wave emerges
in the negative x direction. Rayleigh adopts this condition as a local
condition for the profile at any finite distance h above the undisturbed

35
water so that, with velocity and height changing gradually, the local
condition for no negative wave is

du' 19/tJl,+h) dh (5.7)

and the resultant net condition at final height h is that the total
discrepancy u' = u - Uo will be

u' 2/9 Lh+h - II} (5.8)

Now the formula (1.3) means that at total depth (h+JI,), Uo = /gth+JI,)
so that u, the velocity of the peak relative to still water is

'" ;g /H3h . (5.9)

This is actually Airy's formula (4.2). However, the condition is an

always positive moving profile and this profile necessarily distorts
(as Rayleigh has proved already).

Rayleigh next uses the same line of reasoning to regain Kelland's

formula c ;gAlE as well as those of Green. Let A be the area of
the cross section below the undisturbed level, b the breadth at that
level. Then continuity means

(A + bh)u Au (5.10)
o

for small enough h; so instead of (5.2)

u
2 - u2 (2bh/A)u 2 (5.11 )
o o

2 2
But, since 'by dynamics' (namely (4.5) u u 2gh if the upper surface
o
is free,

gAlb , (5.12 )

which is Kelland's formula.

Rayleigh then goes on to argue as follows: "The energy of a long

wave is half potential and half kinetic. If we suppose that initially
the surface is displaced, but that the particles have no velocity, we
shall evidently obtain (as in the case of sound) two equal waves tra-

36
velling in opposite directions". Evidently the total energies of each
are equal and make up the original energy of displacement, these derived
waves have each one half the elevation of the original and potential
energies one quarter. The energy of each wave is one half so potential
and kinetic energies are equal in the two derived waves. Now apply
this result to the case of gradually changing band A. The potential
energy oc{length x breadth x (height)2} of the wave which is then true
of the total energy. Since wave-length ~ propagation velocity bc/A75
from (5.12)), energy

EoclTA76T x (height) 2 x b (5.13 )

and, since this is constant for a slowly varying canal,

(5.14 )

1 1
This is oc ,-ib- i for a rectangular canal, namely Green's result [33]
for a canal of slowly varying depth and breadth. Both (5.12) and (5.14)
are also given by Airy [22] (Art. 260) and Stokes [36] points out that
his proof of (5.12) is actually simpler than Kelland's in [28]. However,
Rayleigh simply refers them jointly to Green, Kelland and Airy by
reference to Stokes [36].

Next Rayleigh proves that h cannot exceed the value due to the velo-
city U o by general ising (5.3), a result he notes as 'otherwise obvious'

(which it is). He then shows that if u is less than the velocity of

2 0
the free wave (gAo> bu o ) a contraction of area Ao to A produces a
depression of the surface while an expansion from A to A produces an
2 0
elevation the effects being reversed for gAo < buD. He notes that a
stationary wave can sustain itself "in a stationary position without
requiring a variation in the channel; and the effects of such a variation
are naturally much intensified" and goes on to show that the situation
2 -1
when U o > gAob is unstable.

At this point Rayleigh starts a wholly new section entitled "The

Solitary Wave". He says "This is the name given by Mr. Scott Russell
to a peculiar wave described by him" (in [2]). Since Russell's wave is
6 or 8 times the canal depth in length it should be treatable by the
theory of longwaves. However "there are several circumstances observed
by Mr. Russell which indicate it has a character distinct from that
of long waves"--notably the different behaviours of positive and negative
waves, the former having a "remarkable permanence", the latter being
"soon broken up and dissipated".

37
 Rayleigh remarks that Airy "appears not to recognize anything dist-
inctive in the solitary wave" and quotes from Airy [22] "We are not
disposed . in the form of the two quotes given above. On the other
hand he then also quotes the earlier view expressed by Stokes in [36]
"It is the opinion of Mr. Russell (also given above). After dis-
missing a paper by Earnshaw [41], which appeals to experiment in order
to validate the postulates of the theory of long waves and thereby runs
into a problem of matching rotational motion in the wave to irrotational
motion beyond it, Rayleigh goes on to consider whether there can be
compensation between pressure variation at the surface from the finite-
ness of height and variation due to the departure from the law of uniform
velocity proper to very long waves. This is the crucial consideration.

To analyse this he introduces stream lines ~ const. and a velocity

potential ~ setting

F(x + iy} e iy d F(x} (5.15 )

dx

this choice being motivated by the fact that one stream line, at the
bottom of the canal, is straight. Thus

F - ~
2!
F" + ~
4!
F(iv}

3 5
~ yF' - L F'" + L F (v) (5.16 )
3! 5!

~ = 0 on the bottom of the canal, and at this point by changing back

to Russell's notation and choosing h for the depth of undisturbed water,
~ = -ch at the free surface. On this surface p is uniform so that
(from (4.5})

2 2 2 - 2g(y-h)
u + v c (5.171

Then
2 4
F' - L F' + L F (v) + (5. 18a)
u 2!
II
4!
and

v -yF" + (5.18b)

so that

(F' }2 y2F'F'" + i(F,,}2 + c 2 - 2g(y - h} (5.19 )

on the free surface. But the free surface is also ~ -ch so that

38
 3
yF' - L F'" + ••• -ch . (5.20 )
3!

The procedure now is to eliminate F from (5.19) and (5.20) to get

a differential equation for y, the ordinate of the free surface. If
F varies slowly one can solve these two equations for y by iteration
so that

F' ch + .1. y2F'" +

Y 6

-ch {ly + 1
6 y2(.l)"
Y + ••• } (5.21 )

and (5.19) becomes

1 2 1 2 1
- - -3 Y (-)" + Y (-)' (5.22 )
/ y y

or
2
3 (y - h) (1 _ 9.Y ) (5.23 )
h2 c2
from which, when y' = 0, Y
2
h or y = c /g. Then if 1 - gy/c
2 ~O,

y = c 2 /g is maximum, y = k + h (say). Thus

2
c = g(k + h) (5.24 )
Also
1
y - h k sech 2 2"(x/b) (5.25 )
and

b2 h 2 (h + k)/3k (5.26 )

The formula (5.24) for c 2 is of course precisely Russell's (2.1). Also

(5.25) is

u(x, t) y - h =k sech 2 [ ~ ~~)i (~+~ ixJ (5.27 )

(at t = 0) and the maximum height is Russell's k. However the formula

1
differs from Boussinesq's result (2.5) by the scaling (h/h+k)2 on
x. This scaling has the same order of discrepancy as does the KdV
! 1
solitary wave (2.11), for [1/(1 +k,th)P .. l - 2"(k/h). But now the exact
result (5.24) is found for c 2 (to within the order worked). Thus,
in this very direct way, Rayleigh further justifies Boussinesq's analysis
·and wholly justifies (2.1) and therefore, for Russell, his whole position.

It would be tempting now to trace the further developments to modern

methods which, for example, identify symplectic manifolds and infinite

39
dimensional Lie algebras within the theory, quantise it and find Yang-
Baxter integrability conditions. I leave this fascinating series of
developments to other contributions to this meeting.

I therefore finish with Rayleigh's own summing up of his work on

the solitary wave. He says [7] "The velocity of propagation is given
by (5.24), which is Scott Russell's formula exactly. In words, the
velocity of the wave is that due to half the greatest depth of water.

Another of Russell's observations is now readily accounted for:-

'It was always found that the wave broke when its elevation above the
general level became equal, or nearly so, to the greatest depth. The
application of mathematics to this circumstance is so difficult, that
we confine ourselves to the mention of the observed fact [22,42]'
When the wave is treated as stationary it is evident from dynamics
that its height can never exceed that due to the velocity of the stream
in the undisturbed parts; that is, k is less than U~/2g. But u~ =
g(h + k), and therefore k is less than ~(h + k) or k is less than h.
When the wave is on the point of breaking, the water at the crest is
moving with the velocity of the wave".

Rayleigh then goes on to consider Periodic Waves in Deep Water.

REFERENCES

[1] A. C. Scott, F. Y. F. Chu, D. W. McLaughlin, Proc. IEEE 61 (1973)

1443.
[2] J. S. Russell, "Report on Waves", British Association Reports
(1844).
[3] The canal is believed to be the Union Canal which passes below
the campus of Heriot-Watt University, Edinburgh, Scotland. The
Union Canal Company certainly paid for the boats Russell construc-
ted in 1835 for the experiments described in [5] below. However,
Encyclopaedia Brittanica 1886 says "Having been consulted as to
the possibility of applying steam-navigation to the Edinburgh
and Glasgow Canal, he replied that the question could not be ans-
wered without experiment, and that he was willing to undertake
such if a portion of the canal were placed at his disposal".
Still the 100th anniversary of Russell's death in 1882 was held
at Heriot-Watt University. Despite perfect arrangements in all
other respects, the organisers were unable to generate a It foot
solitary wave on this occasion despite the several powerful boats
stopped for this purpose! In the light of such difficulties,
Russell's experiments on this canal in 1834-35 described in [5]
below are astonishing.
[4] Sir George Stokes, Trans. Camb. Phil. Soc. 8 (1847) 441.
[5] J. S. Russell, "Experimental Researches into the Laws of Certain
Hydrodynamical Phenomena that accompany the Motion of Floating
Bodies and have not previously been reduced into conformity with
the Laws of Resistance of Fluids" Trans. ROY. Soc. Edinburgh
14 (1840) 47-109.
[6] J. S. Russell, The Wave of Translation in the Oceans of Water,
Air and Ether (Trubner, London, 1895).
[7] Lord Rayleigh (J. W. Strutt), "On Waves", Philos. Mag. 1 (1876)257.

40
 [8] D. J. Korteweg, G. de Vries, "On the Change of Form of Long Waves
Advancing in a Rectangular Canal, and on a New Type of Long
Stationary Waves", The London, Edinburgh, and Dublin Philosophical
Magazine and Journal of Science Series 5, 39, No. 241, June 1895,
pp. 422-443 (Philos. Mag. 39 (1895) 422).
[9] J. McCowan, "On the Solitary Wave", Philos. Mag. 31 (1891) 45.
[10] J. McCowan, "On the Highest Wave of Permanent Type", Philos. Mag.
38 (1894) 351.
[11] R. K. Bullough and P. J. Caudrey, eds., Solitons, Topics in
Current Physics 17 (Springer, Heidelberg, 1980), Chap. 1.
[12] N. Zabusky, M. D. Kruskal, Phys; Rev. Lett. 15 (1965) 240.
[13] C. S. Gardner, J. M. Greene, M. D. Kruskal, R. M. Miura, Phys.
Rev. Lett. 19 (1967) 1095.
[14] L. Euler, Memoirs of the Academy of Sciences of Berlin (1755).
[15] J. L. Lagrange, Mechanique Analytique pp. XII, 512 (Chez La
Veuve Desaint, Paris, 1788); Nouvelle Edition Augmente Par
L'auteur 2 Vols. Paris 1811-15.
[16] P. S. Laplace (Marquis de Laplace): Traite de Mechanique Celeste
(Chez, J. B. M. Duprat, Paris an VII, 1798-1823).
[17] S. D. Poisson, "Memoire sur la theorie des ondes" Mem. de l'Acad.
Royale des Sciences (i) (1816).
[18] E. H. Weber, W. Weber, Wellenlehre auf Experimente gegrunder,oder
uber tropfbarer Flussigkeiten mit Adwendung auf die Schall und
Licht-Wellen (Leipzig, 1825).
[19] Reference [20] was presented on 13 Nov. 1871 and published in 1872.
Rayleigh [7] refers to the note by Boussinesq (J. Boussinesq,
"Theorie de l'intumesence liquid appelee onde solitaire ou de
translation, se propagent dans un canal rectangulaire" Comptes
Rendus LXX II, 755(1871)) and acknowledges Boussinesq's priority.
[20] J. Boussinesq, "Theorie des ondes et des remous qui se propagent
Ie long d'un canal rectangulaire horizontal, en communiquant au
liquide contenu dans ce canal des vitesses sensiblement pareilles
de la surface au fond" J. Math. Pures et Appliquees 2(1872)55.
[21] P. J. Caudrey, "The Inverse Problem for a General N x N Spectral
Equation", Physica 60 (1982) 51-66.
[22] Sir G. B. Airy (Astronomer Royal), "Tides and Waves", Encyclo-
paedia Metropolitana (1845).
[23] J. S. Russell, Proc. Roy. Soc. 32 (1881) 382.
[24] J. S. Russell, The Modern System of Naval Architecture (Day &
Son, London, 1865).
[25] J. S. Russell, Systematic Technical Education for the English
People (Day & Son, 1869).
[26] G. S. Emerson, John Scott Russell: A Great Victorian Engineer
and Naval Architect (John Murray, London, 1977).
[27] Sir John Herschel, quoted by Russell [24], p. 208.
[28] The Reverend P. Kelland, "On the Theory of Waves Par~ I" Trans.
Roy. Soc. Edinburgh 14 (1839) 497-545. The result c
-1
gAb quoted below (4.10) appears on the p. 530.
[29] Reference to Kelland [28] shows that Kelland's formula in his
notation is~ first of all (p. 514)
-CLh
2lT 2 e CLh - e (A.l )
A c g eCLh + e -CLh 2 2
1- (2~aJ [eCLh - e-CLh]
CLh -CLh-l
with a the amplitude of the wave given by a = (A/lT)(e +e ) .
However in a Sec. III devoted to "Solitary Wave Motion" Kelland
defines this motion to have a velocity along x positive at all
points along the length of the wave and reaches the formula (4.3)
Wl th the seml-elevation E: = (bl c a) (e CLh - e -CLh) where u (the velo-
city along x) defines b through u = b(e CLY - e- CLY )(l + sin(x - ct)).

41
 (Evidently Russell mixes a and a in (4.3)). Kelland (28) acknow-
ledges that his formula seems to mean that larger wave height
k means longer wavelength A contrary to Russell. Russell (2)
p. 334 plainly appreciates the candour of Kelland's "my solution
can only be regarded as an approximation, nor does it very accu-
rately agree with observation", and in contrast with his response
to Airy on the same page.
[ 30) D. Bernouilli, Hydrodynamica (Argentorati, 1738).
[ 31 ) E. Torricelli, De motu gravium natural iter accelerato (Firenze,
1643) .
[ 32) Sir Horace Lamb, Hydrodynamics, 6th ed. (Cambridge University
Press, 1952).
[ 33) George Green, Trans. Camb. Phil. SOT. 6 (1837) 457. Russell (34)
supported the breadth dependence B-' of the wave height (see
(35) as well as 7 in our remarks on (34) in the text). He obser-
ved that the height of the wave increased as the depth of the
fluid decreased, but that variation in wave height was very slow
compared with variation in depth across the canal.
[ 34) Sir John Robinson, K. H., John Scott Russell, "Report on the
Committee on Waves" British Association Reports 6 (1837) 417-
496.
[ 35) George Green, Trans. Camb. Phil. Soc. 7 (1839) 87. Russell [ 34)
gave the formula c = (~gy) i, but Green shows that Russell's
1 1
experimental data better fit c = ("2 gy),-
(36) Sir George Stokes, British Association Reports (1846).
(37) Sir John Robinson, J. S. Russell, "Report, 1840" British Associa-
tion Reports, pp. 441-443 (1840).
( 38) R. K. Bullough, D. J. Pilling and J. Timonen, Soliton Statistical
Mechanics, p. 276 in this volume.
( 39) I cannot now find actual reference to this. But Rayleigh (7)
in a footnote added in 1899 (to his collected papers) quotes
Stokes's supplement to [4) in Stokes Mathematical and Physical
Papers Vol. I (Cambridge University Press, Cambridge, 1880) indi-
cating that like the work of KdV (8) Stokes's later work confirms
the existence of absolutely permanent waves of finite height.
In this supplement to [4) Stokes develops the analysis at finite
depth to 3rd order. He finds

2 D1 1 2
mc + (S + 2S2 + 12)b (A.2)
y 51 SO
1 1 4

where S.~ e imk/ c +e -imkc (.~-1,2,

= _ ... ) , D i _- e imk/c - e -imkc
k = ch and bD 1 = a. In this paper he introduces the stream
function ~ as well as the velocity potential ~ = -k (a different
k) at the canal bottom.
(40) B. De Str Venant, Comptes Rendus, Vol. ci (1885) reference taken
from (8).
[41) S. Earnshaw, The Mathematical Theory of the Two Great Solitary
Waves of the First Order, Trans. Camb. Phil. Soc. VIII (1849)
326-341.
[42) Beyond the analysis which Rayleigh gives in (7) and despite the
paper (10) by McCowan this quotation from Airy [22) is still
surely right: the problem is of current interest, but the broken
wave is dramatically turbulent, and deterministic chaos theory
(for example) does not handle such a case yet. (For a picture
see e.g., front cover to 'Physics Bulletin' 38, No.5, May 1987,
and the article by I. Grant "Flow Measurement by Pulsed Laser
(Speckle)" ibid. pp. 175-177 which describes the methods used
to observe it.)

42
 Part II

Mathematical Theory: 1ST, Symmetries,

Singularity Structure and Integrability
Topics Associated with Nonlinear Evolution Equations
and Inverse Scattering in Multidimensions
M.J. Ablowitz
Department of Mathematics and Computer Science, Clarkson University,
Potsdam, NY 13676, USA

In recent years the basic structure required to implement the inverse

scattering transform in 1+1 and 2+1 dimensions has been clarified and
extended. Aspects involved with fully multidimensional problems have
also been treated. In particular the inverse scattering associated
with various multidimensional operators and generalizations of the
sine-Gordon and self-dual Yang-Mills equations have been studied. A
review of some of this work will be discussed in this review.

The Inverse Scattering Transform (I.S.T.) is a method to solve certain

nonlinear evolution equations. There has been wide ranging interest
in this method for many reasons. A review of earlier work can be found
in [1 J. A surprisingly large number of physically interesting nonlinear
equations can be solved via 1ST; there are many applications in physics
including: surface waves, internal waves, lattice dynamics, plasma
physics, nonlinear optics, particle physics and relativity. Mathemati-
cally speaking the field is also quite rich, with nontrivial results
in the areas of analysis, group theory, algebra, differential and alge-
braic geometry being used by various researchers. From our point of
view, 1ST allows us to solve the Cauchy problem for these nonlinear
systems. We shall concentrate on questions in infinite space. All of
the nonlinear equations discussed below arise as the compatibility
condition of certain linear equations, one of which is identified as
a scattering (direct and inverse scattering is required) problem and
the other(s) serves to fix the "time evolution" of the scattering data.

In one spatial dimension the prototype problem is the (KdV) equation

o. ( 1)

The KdV equation is compatible with

vxx + u(x,t)v AV (2 )

(3)

44
i.e., v xxt = v txx implies (1). Equation (2) is the time independent
Schrodinger scattering problem, A the eigenvalue (y = const. in (3)).
The solution of (1) on the line: -oo<x<oo for initial values u(x,t=O)
vanishing sufficiently rapidly at infinity is obtained by studying
the associated direct and inverse scattering problem of (2) and using
(3) to fix the time evolution of the scattering data. It turns out
that the inverse problem accounts to solving a matrix Riemann-Hilbert
boundary value problem (RHBVP) whose jump discontinuity depends expli-
. 2 -ikx
citly on the scattering data. Calling A =-k , v(x,k) = ~(x,k)e
the RHBVP takes the following form,

~ _ ( x , t , a ( k ) ) V ( x , t , k) on Eo ~ ± .... 1, 1 k 1.... 00 , (4 )

where V(x,t,k) = r(k,t)e 2ikx , a(k) = -k, ~ = {k:kEJR}, and ~± are

the limiting boundary values, as Imk .... 0 ±, of meromorphic functions
in the upper (+) lower (-) half plane. (4) may be converted into a
linear integral equation by taking a minus projection and the potential
is reconstructed via

u(x,t) 1 a J (x,t,-k)V(x,t,k)dk,
- 11 ax ~ (5 )
C

where the contour is taken above all poles of r(k,t); of which there
is at most a finite number, k. = i~, KJ.> 0 , j = 1, ... ,N. The scattering
J J
data: the reflection coefficient, r(k,t) evolves simply in time

r (k,t) r(k, 0) (6 )

The above scheme may be extended so as to solve a surprisingly large

number of interesting nonlinear evolution equations. There are two
scattering problems of particular interest in one dimension:

(i) Scalar scattering problems:

n
~ A v,
j=2
v(x,k), uj E a::

(ii) First order systems - generalized AKNS

dv
dx i k J v + q v

v(x,k), q(x) E a: NxN , J = diag(J 1 , ••• ,J n )

Ji* Jj, i * j
q ii O.

45
Via an appropriate transformation the inverse problem associated with
(i), (ii) can be expressed as a matrix RHBVP of the form (4). The
potentials u., q can be shown to satisfy nonlinear evolution equations
J
by appending to (i) and (ii), suitable linear time evolution equations.
One then finds that the scattering data V(x,t,k) evolves simply in time.
Well-known solvable nonlinear equations include the Boussinesq, modified
KdV, sine-Gordon, nonlinear Schrodinger, and three wave interaction
equations. The reader may wish to consult for example [2a-e] for a
detailed discussion of some of this material.

It is most significant that these concepts can be generalized to

bospatial plus one time dimension. Here the prototype equation is the
Kadomtsev-Petviashvili (K-P) equation:

(7)

which is the compatibility equation between the following linear problems:

ov
y
+ v
xx
+ u(x,y,t)v o (8 )

x
v t + 4v xxx + 6uv x + 3(u x -0 f u dx')v + yv o (9 )
-00
y

(y = const.). We shall consider the question of solving (7) for u(x,y,O)

decaying sufficiently rapidly in the plane r2 x 2 + y2 ~ 00. Physi-
cally speaking, both cases 0 2 -1 (KPI) 0 2 +1 (KPII) are of interest.
Whereas KPI can be related to a RHBVP of a certain type (nonlocal; see
ref. [3]) I KPII turns out to require new ideas. Letting

v ~(x,y,k) exp(ikx + k 2 y/o)

o = oR + i0 1 , OR *
O. Then there exist functions ~ bounded for all
x,y satisfying ~ ~ 1 as Ikl ~ 0 0 . However such a function turns out
to be nowhere ~nalytic in k, rather it depends nontrivially on both the
real and imaginary parts of k(k=k R + ik I ). ~ = ~(x,y,kR,kI).

In fact ~ satisfies a generalization of a RHBVP--namely a a(DBAR)

problem where ~ satisfies,

( 10)

where a l( a + i ~k ) and V has the structure

ak 2 ak R a I

46
 sgn(k O ) exp[i8(x,y,kR,kr'~0)1

21TloRI

k
(x + 2y ~)(~ - kR)
OR 0

( 11 )
~o

(10-11) may be converted into a linear integral equation by employing

the generalized Cauchy formula. T(kR,k r ) is viewed as the "nonphysical"
data (i.e., inverse scattering data or inverse data) and the potential
is reconstructed via

u(x,y)
2i a ( 12)

The basic ideas used in order to derive these equations is as follows.

We convert the equation for ~ = ~(x,y,k):

o ( 13)

into an integral equation

~(x,y,k) 1 + G(u,~) ( 14)

where

G( f) G*f = ffG(x-x', y-y' ,k) f(x' ,y' )dx'dy', ( 15)

the Green's function kernel being given by (k=kR+ik r ):

-ff
1 exp [ i (~ x+yy) 1
2 d~dy
f (ion-~
(211 -2k~)

sgn(y) fd~ exp[ix~+~(~+2k)y/ol

2110

( 16)

where

{1: x > 0, 0: x < O} •

47
The a derivative of the Green's function is especially simple,

sgn(k O )
exp[i8(x,y,k R ,k r )] (17 )
2nl ~ I
where

a 1 a a
"2(ak + i ~), and
ai< R r

kr
8(x,y,k ,k ) = -2(x + 2y iJ)k O.
R
Taking the d derivative of ( 14)

( 18)

and using (17) shows that

( 19)

u(x' ,y' )w(x' ,y' ,kR,kr)dx'dy']. (20)

Multiplying (20) by expr-i8(x,y,k R ,k r )] and employing the following

symmetry condition on the Green's function

(21 )

where ~O

(22 )

whereupon (10-11) follow. The eigenfunction Il is recovered with the

generalized Cauchy formula

48
 1 +
1T
(23 )

noting that using (10-11), (23) becomes a linear integral equation

for~. The potential u(x,y) is recovered by taking k~oo in (13) or (14)
and (23). For the K-P the evolution of the data obeys (Y=4ik 3 in (9))

(24 )

where kO

Special cases include °= oR + io I :

aT (25 )
at
1, ~I

aT (26 )
at
These formulae allow us in principle to solve the Cauchy problem for
K-P and in particular the limit (ii) discussed above allows us to give
an alternative solution for KP I via a and not via a nonlocal RHBVP.

Similar ideas apply to higher order scalar problems

( iii) o

where v, u.E ~ and to first order systems

J

.)
( lV ° aayv + 'J aaxv + q( )v
x,y 0

Interested readers may consult reference 4a, and review 4b for more
details.

The notion of a extends to higher dimensional scattering and inverse

scattering problems. However as we shall mention, despite the fact
that the inverse scattering problem is essentially tractable there does

49
not appear to be any local nonlinear evolution equations in dimensions
greater than 2+1 associated with multidimensional generalizations of
( i i i) or (i v) .

Our prototype scattering problem will be

ov + ~v + u(x,y)v 0
y

n a2
E --2' ( 27)
R,= 1 aXR,

Letting

v lJ(x,y,k) exp( ikx + k 2 y/o)

k kR + ik r , k E a:: n

n
k.x E kjX j , 0 OR + iO r
1

Then there exist functions lJ bounded for all x, y satisfying lJ -+ 1,

as I kjl -+ 0 0 , j 1, ..• ,n. When OR * 0, lJ turns out to be nonanalytic
in each of the var iables k, i. e., lJ lJ (x,y,k R , ••. ,k R ' kr , ..• ,k r )
1 n 1 n
and satisfies a a problem linear in lJ , in each of the variables kji
i.e., we shall show that lJ satisfies an equation of the form,

j 1, ... ,n (28 )

where Tj is an appropriate linear integral operator.

The basic idea in order to derive (28) follows a similar format to

the two dimensional case described earlier. From the definition of
lJ(x,y,k) below (27) we see that it satisfies

o. (29 )

We convert to an integral equation

(30)

where the Green's function kernel is given by

50
 exp [ i ( x . ~ +YTl ) 1
---:-:;-, ff 2 d~ d Y (3' )
(211)n+ iay-~ -2k.~

sgn(y)

(32 )

Taking the a derivative of (30)

(33 )

and using

- - - laRlfexp[iS(x,y,kR,kI,~)l
(211) n

(34 )

where

p(~) (35 )

shows that

(36 )

where

and w satisfies

(38 )

MUltiplying (37) by exp(-is) and using the symmetry condition

(39 )

yields

51
 (40 )

and hence (36) gives

We see that T. is an integral operator which depends on a scalar scat-

J
tering function T = T(kR,kI'~)~ being effectively (n-1) integration
parameters (due to the delta function in (41) in the nonlocal operator
T j) .

One can use a generalized Cauchy formula such as (23) in order to

obtain a linear integral equation to reconstruct ~. However due to the
redundancy of the data discussed below, we find that an alternative
method is more useful. The inverse problem is redundant, i.e., we are
given T(kR,kI'~) (3n-1 parameters) and we must reconstruct a local
potential u(x,y) (n+1 parameters). A serious issue is how to characterize
admissible inverse data T, i.e., data that really arises from a local
potential (small generic changes in T(kR,kI'~) cannot be expected to arise
from a local p~tential u(x,y)). Insight into this question is obtained
by noting that T must satisfy a nonlinear constraint, one which is ob-
2 --
tained by requiring a ~/aki ak j a2~/ak.ak. (it]). The form of this
J l.
constraint is given by

(42 )

where £
iJ' is a linear operator and N .. a nonlinear (quadratic.) nonlocal
l.J
operator. These operators are given by

:r ij (~.-k'R)(_a- +..!. a
J J a k. 2 a~ i
) _ (~.-k. )(_a_ + ..!._a_)
l. l.R a k. 2 a~ j
(43 )
l. J

(44 )

There is, in fact, an explicit transformation of variables

52
which simplifies (42). Namely,

ko1)°

~,
n w,
E w)oXR)o + 2
j=2

~j (45)

transforms (42) into

(46 )

using the generalized Cauchy formula (23) we have

~o o(T)(j(,w,w O )
T ( X ,w ,w O) - fJ 1) , dX ' dX '
11 - R ·1
X - X

(47)

where

SS exp[-i(ywO+x.w)]u(x,y)dxdy. (48)

We have used the fact that when Wo = 2k1.(~-kR)/crR and w = ~-kR are
kept fixed, T(X'w,w O) ~ u(w,w O) (The Fourier Transform of u(x,y» for
large Xj(w,*O); this is the analogue of the Born approximation.

We expect that for suitably "small" u (i.e., no homogeneous solutions

to the relevant integral equations) if 1 is independent of x,j and decays

53
sufficiently fast for Iwl ,lwol~oo , then T(kR,kI'~) is admissible. More-
over (47) gives a formula to reconstruct the potential by quadratures.
Limits to case a = i and reductions to stationary potentials u(x,y) = u(x)
can be carried out. Details can be found in Ref. [Sa,b]. It should also
be noted that in recent work Nachman and Lavine [Sc] have extended their
ideas to situations where there are homogeneous solutions to the relevant
integral equations. (42) also suggests why simple local nonlinear evolu-
tion equations have not been associated with equation (27). Namely, in
the previous lower dimensional (2+1 and 1+1) problems the time evolution
of the scattering data obeyed a particularly simple equation (e.g.,
~~ = w(kR,kI)T). However in this case such a simple flow will not be
maintained due to the nonlinear constraint (42).

These ideas can be generalized to first order systems:

av n J. av
(v) ay + a L J aX j qv
j=l
k R.
J.*J.,k*P,
J J

with many similar results obtained [6a,b,c};though there are some impor-
tant differences as well: see ref. [6c]. Again the scattering data
satisfies a nonlinear constraint. In general, there is no compatible
local nonlinear evolution equation associated with (v). However when
certain restrictions are put on J. then the constraint equation becomes
J
linear and the so-called N wave interaction equations are compatible
with the system (v). Nachman and Ablowitz [6a] showed that at most,
the system would be 1+1 dimensional, and Fokas [6b] showed that indeed
the system is reducible to 2+1 dimensions by a transformation of inde-
pendent variables (characteristic variables). In [6c] Fokas studies
the inverse scattering of (v). For 0= i he finds an equation similar
to (42). However its integrated form shows that in order for the poten-
tial to be reconstructed one must solve a reduced system of equations
of the form (v): i.e., for N = 2. This is in contrast to the scalar
problem where reconstruction is via quadratures.

Beals and Coifman have an alternative but similar formulation [7a,b]

for multidimensional scalar problems.
There is an n-dimensional problem which also fits within the frame-
work of 1ST: The so-called generalized wave and generalized sine-Gordon
equation (GWE and GSGE). These equations arise in the context of dif-
ferential geometry and serve to extend the classical results of Backlund
for the sine-Gordon equation to n-dimensions [8]. The n-dimensional
Backlund transformation is given by:

54
 t
+ XA X A - XB, (49 )

where

n lL
L a dx·,
j=' Xj J

A ..
1J
s . (Z ) a. J. dx J. ,
1 1

aa l i
B .. -~-- dx., '.:::. i , j .:::. n, (50 )
1J a'j aX j 1

and a = {a ij }ElR nxn . Equations (49-50) reduce to the Backlund trans-

formation for the generalized sine-Gordon equation (GSGE) when

s . (z)
1
( z 2 + (28 i' - ,)) /2 z, ( 5' )

and for the generalized wave equation (GWE) when

s 1. (z) _ At z) • (52 )

The compatibility condition required for the existence of solutions

to these Backlund transformations results in a system of second-order
partial differential equations for an orthogonal nxn matrix a = {a .. }
1J
in (49) which is a function of n independent variables a = a(x"x 2 , ... ,
x n ). The equation has the form

(_,a _ 5.J)
a
+ _a_(_,_ aa l i
ax a -ax.)
'i xi j'k J

i ~ j,

i,j,k distinct,

aa ji aa'k
a
,i
ax:-'
1
i ~k, (53 )

where E for the GSGE and E o for the GWE.

55
 We observe that when n 2 and e: 1 (GSGE) , the orthogonal matrix
a = {a .. } given by
1J

1
cos'2u sin'2u
a [ 1 (54 )
. 1
-s1n'2u
1
cos'2u
1
for the function u = u(x,t) reduces the GSGE to the classical sine-
Gordon equation ( K=-1),

U tt - u xx - Ksin u o. (55 )

On the other hand when n = 2 and K = 0, then with (54) the GWE reduces
to the wave equation .( 55). When n > 3 the generalization of the wave
equations discussed here is nonlinear.

The Backlund transformations (49) described above are in fact matrix

Riccati equations. Linearizations of such a system can be performed
in a straightforward manner. Introducing the transformation

x (56 )

where U, V and nxn matrix functions of x 1 , ... ,x n , the following linear

system is deduced:

( 57)

with the components of A, B given by (50). Compatibility ensures that

the orthogonal matrix a = {a .. } satisfies the GSGE with (51) and GWE
1J
with (52). Alternatively, if we call

I/! , (58 )

the following'linear system of 2n o.d.e.'s are obtained:

(59)

where Aj , C j are 2nx2n matrices with the block structure

A. (60 )
J

56
Here aj , Yj are nxn matrices having the following structure:

ae. (61 )
J

where e. {ej}ik is the unit matrix

J

i = k = j

{~ otherwise,
(62)

and in component form y. takes the form

J

(1 -6 ) _1_~ 6 . - (1- 6 . )1- ~ 6 kJ· . (63 )

kj a~k a Xk ~J ~J a1~ ax~

In (61) a is the orthogonal matrixlRn-+SO(n) associated with the GWE

when 6 = A and with the GSGE when 6 = ~(z + 1 /z) , A ~(z - 1/z) ,
and y. is the matrix (63): lR -+ M (lR), y~ + y. = O. Equations (53)
J n n J J
arise as the compatibility condition associated with (58). More expli-
citly, for the GWE the scattering problem takes the form [~=~(x'A)l

.2.lL AAj~ + C.~ (64)

aX j J

with

A.
J [ :~ :jJ
, (65 )

and C j given by (60,63).

For the GSGE the scattering problem for ~ ~(x,z) takes the form

6(z)

(66 )

6(Z), A(Z), C. given above, or equivalently

J

+ + C.~ , (67 )
J

57
where

B. ( to ua j ) ,
u = diag(+l,-l, ... ,-l). (68 )
]
aju 0
In [8] it is shown how these linear problems may be viewed as a direct
and inverse scattering problem for the GWE and GSGE. Namely, the direct
and inverse problem may be solved for matrix potentials, depending on
the orthogonal matrix a, tending to the identity sufficiently fast in
certain "generic" diiections. It should be noted that solving the n-
dimensional GWE and GSGE reduces to the study of the scattering and
inverse scattering associated with a coupled system of none-dimensional
o.d.e. 's. This is in marked contrast to other attempts described earlier
to isolate solvable (local) multidimensional nonlinear evolution equation
which are compatibility conditions of two Lax-type operators, e.g.,

L1jJ (69 )

(70 )

where L is a partial differential operator with the variable t entering

only parametrically. Although as we have seen nonlinear evolution equa-
tions in three independent variables can be associated with such Lax
pairs (e.g., the K-P, Davey-Stewartson, three wave interaction equations,
etc.) little progress via this route has been made in more than three
dimensions. As discussed earlier one has to overcome a serious constraint
inherent in the scattering/inverse scattering theory for higher dimen-
sional partial differential operators in order to be able to isolate
associated solvable nonlinear equations, i.e., the scattering data
generally satisfies a nonlinear equation (eq. (42)). The analysis
associated with the GWE and GSGE avoids these difficulties since the
GWE and GSGE problems are simply a compatible set of nonlinear one-
dimensionalo.e.e.'s. The results in ref. [8] demonstrate that the
initial value problem is posed with given data along lines and not
on (n-1) dimensional manifolds.

Similar ideas apply to certain n-dimensional extensions of the so-

called anti-self-dual Yang-Mills equations (SDYM) [9]. In two complex
variables the self-dual Yang Mills equations take the form (see [10])

0, (71 )

where n is a positive matrix valued function of (x 1 ,X 2 )E , 2 • Alter-

natively SDYM takes the form

58
 a A, a A2
+ 0 (72)
-
ax, a X2

a A, a A2
+ [A, ,A 2 ] 0, (73 )
a X2 ax,

where

A. - n
-, an
(74)
J ax.
J
The SDYM may be obtained via the compatibility condition of the following
linear system

~
ax,
- z ~ - A,m
a X2

~ + z ~- A2 m. (75 )
aX2
ax,
Multidimensional extensions may be obtained. For example, cOnsider
the linear system

', .•. ,n (76)

+ zs.
J
a ( 77)

and

Compatibility (commutativity) implies:

o (78 )

aAj
+ [A. ,A.] o (79)
aXi 1 J

o. (80 )

A potential n may be introduced as before:

A. (8l)
J

59
to obtain

a~
s ._a__ (~-l __ )
o. (82 )
J - ax;
aX j + 1 L

Clearly when n 2 this system reduces to the classical SDYM equation.

Solutions to these equations may be constructed via the a method.

Define

(83)

with

a s. _a_
ax.'
J
J -
aX j + 1

We shall show that the

-a integral equation

m(x,z) ff (mV)(x,s) dSAd~ (84 )

r + 2lTi
s-z

satisfies (76) (dSAd~ = -zidsRdsr). Operating on (84) with D~ yields,

(85 )

where

1 ZL~(mV)
J 2 lTi f
/;;-Z

sL~(mV)
+ 2lTi f ---=---
s-z
ds Ad~. (86 )

Putting (85),(86) together gives

A. j + 2lTi
f -(D~m)V + m(D~V)
- " - - - - - - . ? - - - d s '\d~ (87 )
s - Z

where

'A. (x) (88 )

J

60
We shall require V(x,z} to satisfy

DjV 0 (89 )
Z

in which case using (84 ) in ( 87) by writing

mV
'A. 'Aj(m - 2ni f s-z dr; Ad~} (90 )
J

we find

( (D~m) - 'A.(x}m}V
(Djm 'A.m} f J dt;l\d~. ( 91)
Z J 2ni
r; - Z

For V suitably chosen (84) has a unique solution in which case

Djm - 'A.m O. (92 )

Z J

Thus 'A. A. and solutions of the extended SDYM are obtained.

J J

The condition (89) is satisfied if we take V(x,z} = V(u(x},z},

Then

L V' (uR.'z) (ZOjR. + S j Sj+1 Z0 jR.) o (93 )

R.=1

by virtue of s. = (-l}j. In ref. [9] other examples of multidimensional

J
extensions of SDYM and a rigorous derivation of the foregoing is given.

ACKNOWLEDGEMENTS

I am most pleased to acknowledge the many crucial contributions of my

colleqgues: A. S. Fokas, D. Bar Yaacov, A. I. Nachman, R. Beals and
K. Tenenblat. This work was supported in part by the National Science
Foundation under grant number DMS-8202117, the Office of Naval Research
under grant number N00014-76-C-0867, and the Air Force Office of
Scientific Research under grant number AFOSR-84-0005.

61
REFERENCES

[1] M. J. Ablowitz and H. Segur, Solitons and the Inverse Scattering

Transform (SIAM, Philadelphia, 1981).
[2a] M. J. Ablowitz and A. S. Fokas,"Comments in the Inverse Scattering
Transform and Related Nonlinear Evolutions Equations," in Lecture
Notes in Physics, 1B9, ed. K. B. Wolf (Springer, Berlin, 1982).
[2b] R. Beals and R. Coifman, Commun. Pure Appl. Math., 1984, pp.39-90.
[2c] R. Beals, "The Inverse Problems for Ordinary Differential Operators
on the Line," to appear, Amer. J. Math.
[2d] A. B. Shabat, Func. Annal and Appl. 9 (1975); Diff. Eq. 15
(1979) 1824.
[2e] P. Caudrey, Physica 6D (1982) 51.
[3a] S. V. Manakov, Physica 3D (1981) 420.
[3b] A. S. Fokas and M. J. Ablowitz, Stud. Appl. Math. 69 (1983) 211.
[4a] M. J. Ablowitz, D. Bar Yaacov and A. S. Fokas, Stud. Appl. Math.
69 (1983) 135.
[4b] A. S. Fokas and M. J. Ablowitz, "The Inverse Scattering Transform
for Multidimensional (2+1) Problems," in Lecture Notes in Physics,
1B9, ed. K. B. Wolf (Springer, Berlin, 1982).
[5a] A. I. Nachman and M. J. Ablowitz, Stud. Appl. Math. 71 (1984) 243.
[ 5b] M. J. Ablowitz and A. I. Nachman, Physica 1BD (1986) 223.
[5c] A. I. Nachman and R. Lavine, "On the Inverse Scattering Transform
for the n-Dimensional Schr~dinger Operator," to be published as
Proceedings of Conference on Nonlinear Evolution Equations,
Solitons and the Inverse Scattering Transform, held at Mathematische
Forschungsistitut, Oberwolfach, W. Germany.
[6a] A. I. Nachman and M. J. Ablowitz, Stud. Appl. Math. 71 (1984) 251.
[6b] A. S. Fokas, Phys. Rev. Lett. 57 (1986) 159.
[6c] A. S. Fokas, J. Math. Phys. 27 (1986) 1737.
[7a] R. Beals and R. Coifman, "Multidimensional Inverse Scattering on
Nonlinear PDE," Proc. Symposium Pure Math. 43 (1985) 45.
[ 7b] R. Beals and R. Coifman, Physica 1BD (1986) 242.
[ 8] M. J. Ablowitz, R. Beals and K. Tenenblat, Stud. Appl. Math.
74 (1986) 177.
[ 9] M. J. Ablowitz, D. G. Costa and K. Tenenblat, Solutions of Multi-
dimensional Extensions of the Anti-Self-Dual Yang-Mills Equations,
preprint, INS #66, 1986.
[ 10] K. Pohlmeyer, Commun. Math. Phys. 72 (1980) 37.

62
Inverse Problems and a Unified Approach
to Integrability in 1, 1 + 1 and 2+ 1 Dimensions *
A.S. Fokas and V. Papageorgiou
Department of Mathematics and Computer Science and Institute for
Nonlinear Studies, Clarkson University, Potsdam, NY 13676, USA

A unified approach for solving initial value problems for equations in

0+1, 1+1 (one spatial and one temporal), and 2+1 (two spatial and one
temporal) dimensions is given. Illustrative examples in each of these
cases are provided. Some remarks on inverse problems in higher than
two spatial dimensions are made in the context of inverse scattering.

1. INTRODUCTION

The aim of this paper is to emphasize that there exists a unified ap-
proach for solving initial value problems for equations in 1, 1+1 (i.e.,
one spatial and one temporal), and 2+1 (i.e., two spatial and one tem-
poral) dimensions. Furthermore it remarks on inverse problems in higher
than two spatial dimensions. Although these inverse problems are not
related to physically significant nonlinear evolution equations, they
are useful in the context of inverse scattering. In this presentation
we emphasize the main ideas. The detailrofurmalisms can be found in
the cited papers.

It turns out that solving the initial value problem for some equa-
tions for q(t), or q(x,t), or q(x,y,t) is equivalent to solving an
inverse problem for some related eigenfunction ~(z;t), or ~(Z;x,t),

or ~(Z;Xly,t). The inverse problem takes the form of a Riemann-Hilbert

(RH) problem for equations in 1 and 1+1, and the form of a non local RH
problem or of a a(DBAR) problem for equations in 2+1 (a DBAR problem
is generalization of a RH problem). To define the relevant RH or DBAR
problems one needs to study the analyticity properties o f ' with
respect to z. Furthermore those problems are uniquely defined in terms
of certain asymptotic data of the underlying linear system satisfied
by~: Monodromy data in 1, scattering data in 1+1 and some cases of

*This article consists of expanded material of six lectures presented

by one of us (A. S. Fokas) at this Winter School on "Solitons".

63
2+1, and inverse data in some cases of 2+1. We use the Painleve rV(PIV),
modified KdV (mKdV) and the Davey-Stewartson (DS) as illustrative exam-
ples of equations in 1, 1+1, and 2+1 respectively.

The above inverse problems can be naturally generalized to higher

than two spatial dimensions. For example, the generalization of the
inverse problem associated with the DS equation leads to an inverse
problem for a matrix valued function 'I'(z,x O,x), z e: Cn , xOe:]R.1,
x e: ]R.n, n > 1. However, while the associated potential q(xO'x) depends
on n+1 variables, the inverse data T(zR,zI,m 2 , ... ,m n ), zR e: ]R.n, zre: ]R.n,
mR, e:]R., depends on 3n-1 var iables. This has important implications: (a)
The inverse data must be appropriately constrained. This "characteri-
zation" of the inverse data is conceptually analogous to the charac-
terization of the inverse scattering data in the multidimensional
Schrodinger equation [11. (b) The existence of "redundant" scattering
parameters can be used to reduce the above problem to one in two spatial
dimensions. This is in contrast to the case of the multidimensional
Schrodinger equation where the inverse problem can be solved in closed
form. (c) Since the inverse problem for 'I' is reduced to one in two
spatial dimensions, it follows that, if one allows'!', q to d~pend

parametrically on t, q(xO,x,t) satisfies an evolution equation reducible

to two spatial dimensions. In particular, the N-wave interaction equa-
tion in n+1 spatial dimensions can always be reduced to two spatial
dimensions. Thus a genuine three-spatial-dimensional nonlinear evolution
equation, related to an inverse problem, remains to be found. (It should
be noted that several other "multidimensional" problems can be reduced
to one or two spatial dimensions, see M. J. Ablowitz's contribution
in these proceedings.)

We first define the standard RH and DBAR problems.

2• RH AND DBAR PROBLEMS

Let C be a simple, smooth closed contour dividing the complex z-plane

into two regions D+ and D (the positive direction of C will be taken
as that for which D+ is on the left).

A function ~(z) defined in the entire plane, except for points

on C, will be called sectionally holomorphic if: (i) the function ~(z)
+ -
is holomorphic in each of regions D and D except, perhaps, at z = 00,
(ii) the function ~ (z) is sectionally continuous with respect to C,
approaching the definite limiting values ~+(~), ~-(~) as z approaches

64
 Fig. 1

a point ~ on C from D+, or D-, respectively. The classical homogeneous

RH problem is defined as follows [2]. Given a contour C, and a function
G(~) which is Holder on C and det G(~) *
0 on C, find a sectionally
holomorphic function ~(z), with finite degree at ~ , such that

(2.1)

where ~±(~) are the boundary values of ~(z) on C. If G(~) is scalar,

(2.1) is solvable in closed form. If G(~) is a matrix valued function,
then (2.1) is in general solvable in terms of a system of Fredholm
integral equations. Various generalizations of the above RH problem
are possible. For example: (i) The contour C may be replaced by a
union of intersecting contours. (ii) G(~) may have simple discontinui-
ties at a finite number of points; in this case one allows ~(z) to
have integrable singularities in the neighbourhood of these points.
(iii) RH problems may be considered in other than Holder spaces (e.g.
[3]): (iv) One may consider inhomogeneous RH problems ~+(~) = G(~)~-(~)
+ F(~) on C.

The DBAR problem can be defined as follows: Given a~/az, find ~.

If a~/az = 0 everywhere except on a curve, then the DBAR problem reduces
to a RH problem (since a~/az = ~+ - ~-, in a distribution sense). The
DBAR problem can be solved via the following extension of Cauchy's
formula [4]

12.
-7T~ f
lR 2 d(; A d~ ..,r -z
(2.2)

It is interesting that the first RH problem was formulated in connec-

tion with an inverse problem (see [12] for references). Actually RH
problems are intimately related to solutions of inverse problems in
1+1, 2+1, and 1 dimensions:

65
3. INVERSE PROBLEMS IN 1+1

We recall that a necessary condition for a given nonlinear equation

for q(x,t) to be solvable via 1ST is that this equation is the compati-
bility condition of a Lax pair of linear equations. Let us consider
the modified KdV equation

0, (3 •1)

as an illustrative example. Equation (3.1) is the compatibility condi-

tion of

'¥x(z;x,t) iZ[J 2 , '¥(z;x,t)] + Q'¥(z;x,t);

Q (3. 2a)

[Uo,'¥(z;x,t)] + Q'¥(z;x,t) (3.2b)

dz
We first note that the above Lax pair is isospectral, i.e. , dt = O.
Also it turns out that equation (3.2a) is of primary importance; equa-
tion (3.2b) plays only an auxiliary role. To solve the initial value
problem for initial data decaying as Ixl~oo , one first formulates an
inverse problem for '¥(z;x,t): Given appropriate scattering data
reconstruct 'l'.

By studying the analytic properties of '¥ with respect to z, where

'¥ satisfies (3.2a) one establishes the existence of a '¥ which is
a sectionally meromorphic function of z, with a jump along the Re z
axis. This jump as well as the residues of the poles, are given in
terms of appropriate scattering data. Thus the inverse problem is
equivalent to a matrix, regular, continuous RH problem defined along
the Re z axis and uniquely specified in terms of scattering data.

Since in the above discussion we have only used (3.2a), it is evident

that one may pose an inverse problem for any function q(x). However,
this result is useful for solving the initial value problem for q(x,t)

66
only if q evolves in such a way in t, that the scattering data is known
for all t. If ~ evolves in t according to (3.2b) (i.e., if q solves
(3.1» then it turns out that the evolution of the scattering data with
respect to t is simple. Hence, the above RH problem is specified in
terms of initial scattering data; its solution yields ~(z;x,t) and
then (3.2a) gives q(x,t).

We summarize the results of(S,13] concerning mKdV in the case of

solitonless potentials.

Proposition 3.1 (Bounded eigenfunctions). A solution of (3.2a) bounded

for all complex values of z = zR + iZ I and tending to I as Z ... co is
given by
~+( z;x), zI > 0
",(z;x) { -
(3.3)
~ (z;x) , zI < 0

where ~±(z;x) satisfy the following integral equations:

I +f X d~eiZ(X-~)J n±Q(~)~±(Z;~)
-co

(3.4)

where if F and Yare 2 x 2 matrices then

eYF Y -Y
e Fe , n+F T
[: :12) n F
[
0

F21
0

0 ) (3.5)

nOF Diag(F 11 ,F 22 )·

Proposition 3.2 (Departure from Holomorphicity-Scattering Equation).

~+, ",- are holomorphic functions of Z for zI > 0, zI < 0 respectively.

The departure from holomorphicity for z = zR is given by

(3.6)

where

B(z) T I + f co
-co
d~e-iz~Jn+(Q~+), b(z) T I + f cod~e-iz~Jn_(Q~-)
-co
so,

~
+ (z;x)e iz~J-1
(B (z)b(z» (3.7)

67
Proposition 3.3 (Inverse Problem-Reconstruction of Q)

Q(x) is obtained from

1
Q(x) = [ J, 211
fco dz''I'(z';x)e i z' x; (I-B -1 (z')b(z'))], (3.8)
-co
where 'I'(z;x) solves the following Riemann-Hilbert boundary value problem:

'I'(z;x) = I +
1 fco dZ''I'(z';x)e iz 'XJ(I-B- 1 (z')b(z'))
(3.9)
211i
-co z' - (z - i 0 )

Using equation (3.2b) we obtain:

Proposition 3.4 (Evolution of Scattering Data). The evolution of the

scattering data from B(z;O), b(z;O) is given by

B(z;t) b(z;t)

Since B (resp. b) is a strictly upper (resp. lower) triangular matrix

the evolution of the scattering data is given by

(3.10)

4. INVERSE PROBLEMS IN 2+1

Let us consider the Davey-Stewartson equation (a two dimensional ana-

logue of the nonlinear SChrodinger equation)

A = ± 1 (4.1)

as an illustrative example. A Lax pair for (4.1) is given by

'I'x iz (J'I',-'I'J) + q'l' + aJ'I'y' J -.-

G -:J' q ~
(~A : J (4.2a)

(4.2b)

where A1 , A2 , A3 , A30 are appropriate matrix functions of Q, Q (The bar

denotes complex conjugate).

The situation is conceptually similar to the case of 1+1: To solve

the initial value problem for q(x,y,t) one first formulates an inverse

68
problem for ~(z;x,y,t). Depending on the value of a there exist two
different cases (for brevity of presentation we assume non-existence
of poles, i.e., non-existence of lumps): (i) 0 1. There exists a
~ which is a sectionally holomorphic function of z and which has a
jump along the Re z axis. This jump is also given in terms of scattering
data but it depends on them in a non-local way. Thus the inverse pro-
blem is equivalent to a non-local, matrix continuous RH problem defined
along the Re z axis and uniquely specified in terms of scattering data.
(ii) 0 = - i. There exists a ~ which is bounded for all complex z,
but which is analytic nowhere in the complex z plane. However, its
departure from holomorphicity a~/az can be expressed in terms of appro-
priate inverse data. Thus, now the inverse problem is equivalent to a
a (OBAR) problem: Given a~/az reconstruct ~ .

Using (4.2b), again one shows that the inverse scattering and the
inverse data evolve simply in time. Hence, the above RH and a problems
are specified in terms of initial data; their solutions yield ~(z;x,y,t)

and then (4.2a) gives q(x,y,t).

We summarize the results of [6) concerning OSI (0 1, Proposition
4.1.-4.4) and OSII (0 = -i, Propositions 4.5-4.8).

Proposition 4.1 (Bounded Eigenfunctions) A solution of (4.2a) with

o = 1 bounded for all complex values of z zR + iZ I and tending to I
as z ..... 00 is given by

~+(z;x,y), o
~(z;x,y)
{- ~ (z;x,y), o
(4.3)

+
where ~-(z;x,y) satisfy the following integral equations:

~±(z;x,y) = I + ~n fX d~ eiZ(X-~);foodnf~m eim[(y-n)I+(x-~)J).

-00 -00- -00

2n foo d~ eiZ(X-~);fOO dnfoo dm eim[(Y-n)I + (x-~)J)

x -0) -co

(cf. (3.5) for notation).

+
Assuming that the linear integral equations (4.4)- have no homoge-
neous solutions, it follows that:

69
Proposition 4.2 (Departure from Holomorphicity). ~+, ~ are holomorphic
functions of z for zI > 0, zI < 0, respectively. Hence the function
~(z;x,y) defined by (4.3) is a sectionally holomorphic function of z.
.
I n partlcu 1 ar, --
H o for all z, with zI * o and a~
-- = ~
+ (z;x,y) -
az az
~ (z;x,y) for z = zR. The departure from holomorphicity is given by:

+ - OO d Z
~ (z;x,y) - ~ (z;x,y) f
'\11- (
T Z ,x,y ) e i z 'Jx+ i z 'y
'.

-OOf (_'
L.. , Z
) e -izJx-izy I (4.5)

for z = zR' where the scattering data f(z' ,z) are given by:

- foo dm f 1 2 ( z ' ,m) f 2 1 (m, z ) ,

- 00

inf'" dsf'" dn Q~;2 e-(z+z' )s+i(z-z')n (4.6)

-00 -co

f21 (z' ,z) - 2n f'" ds f'"

-co -co
-
dnA Q~ +11 e i(z+z' )s+i(z-z')n , f 22 =0 . (4.7)

Proposition 4.3 (Inverse Problem-Reconstruction of the potential q)

q(x,y) is obtained from:

q(x,y) = - 2n[J,
1 f'" dz' f'" dz~
- (z';x,y)e
iz'Jx
f(z',z)e
-izJx+i(z'-z)Yj
,
-00 -co
(4.8)

where ~-(z;x,y) solves the following integral equation:

U/'("
z ;x,y ) e iz"Jx f ("
z ,z ') e -iz'Jx+i(z"-z')y
f'" f'"
T

~-(z;x,y) + 2ni
-00
dz"
-00
dz'
z' - z + iO

1. (4.9)

Finally from (4.2b) we obtain the following:

Proposition 4.4 (Evolution of the Scattering Data). The evolution of

the scattering data from t = 0, f(z' ,z;O) is given by:

,2 A z2 tA
f (z' ,z; t) e- z t 30 f(z',z;O)e 30 (4.10)

where

f(z' ,z;O) is given by (4.7) and A30 diag(i,-i).

70
Proposition 4.5 (Bounded Eigenfunctions). A solution of (4.2a) with
a = -i bounded for all complex values of Z zR + iZ I and tending to I=
as Z ~ ~ satisfies the following Fredholm linear integral equation

'I' (Zix,y) (4.11 )

where

{(G zR ,ZI,q'l'(Zi·'·»}lj * ~(f:d~f~jZIdmr~dn -(d~l~jZIdmr~n) x

{exp [ (m+ i ( 1- J j ) Z ) (x- ~) + im (y- n) 1[q ( ~, n) 'I' ( Z i ~, n) 1}1 j' ( 4 . 12 ) 1 j

and

{exp [ - (m+ i ( 1+J . ) Z ) (x-~ ) + im ( y-n ) 1[q ( ~ , n ) 'I' ( Z i ~ , n ) ]} 2 . , (4.12) 2j

] ]

Proposition 4.6 (Departure from Holomorphicity). For every Z E C

where the matrix Q is defined by: Q 11 o

Qij * Tij(z)exp0ij(Zix,y), i * j (4.14 )

~Tl rd~r-co dn{q(~,n)'I'(Zi~,n)}ij exp{-0ij(Zi~,n)},

-00
i *j

Propostion 4.7 (Inverse Problem-Reconstruction of q). q(x,y) is ob-

tained from

(4.15 )

where 'I'(Zix,y) satisfies:

'I'(Zi x ,y) 1
2Tli ff 'I'C(Zi X ,y) Q (ZR,ziix,y) dz '" dz '
z'-z
I. (4.16 )

Finally from equation (4.2b) we obtain:

71
Proposition 4.8 (Evolution of the Inverse Data). The inverse data
at time t, n(zR,zI;x,y,t), is given by

(4.17)

where n(zR,zI;x,y,O) is given by (4.14) using the initial condition

q(x,y,O) and A30 = diag(i,-i).

5. INVERSE PROBLEMS IN 0+1

The Lax pair associated with the PIV equation

~ ~(iY)2 + 3 y 3 + 4ty 2 + 2(t 2 + cdy + B (5. 1 )

dt 2
2y dt "2 y'
is given by

-~
1
8 0 -V
2
Y
Z
(z)
+ [:,v-e
u 0
-8 J
co
-t
u + [
2v
-(v-28 ) -(8 0 -v) Z
uy 0
] 1]
- Y(z),

(5. 2a)

1
[:'V-8
U Y( z). (5.2b)
o
u 0 -9co J

Indeed Yzt = Ytz implies

iY - 4v + y 2 + 2ty + 48 0 , du -u{y+2t),
dt dt
(5.3 )
48
dv 2 2
v + {_a - y)v + {8 0 + 8 co )y,
dt y y

where,

As in the cases of 1+1 and 2+1, solving the initial value problem
of PIV reduces to solving an inverse problem for Y: Reconstruct Y(z;t)
in terms of appropriate monodromy data. Again this inverse problem will
be solved in terms of a RH problem. Thus it is essential to study the
analytic properties of Y with respect to z. However, in contrast to
the analogous problem in 1ST for 1+1 and 2+1, this task here is straight-
forward: Equation (5.2a) is a linear ODE in z, therefore its analytic

72
structure is completely determined by its singular points. In this
particular case z = 0 is a regular singular point and z = is an irregu-
lar singular point of rank 2. Complete information about z = 00 is
provided by the monodromy matrix Moo and by the Stokes multipliers
a, b, c, d. Solutions of (5.2a), YO and Y" normalized at zero and
infinity respectively are related via a connection matrix EO with
entries u o' SO' YO' 00· Taking into consideration the above singula-
rities, there exists a sectionally holomorphic function Y, with jumps
_ n n 3n 5n
across the four rays, arg z - - 4' 4' ~, ~ and with singularities
at z = 0, z = 00. The jumps are specified by a, b, c, d and the nature
of singularities by MO' Moo. This leads to a matrix, singular, dis-
continuous RH problem, defined on the above rays and specified in terms
of the monodromy data

Monodromy Data (MD)

A consistency condition of the above RH problem yields

4 -1 -1
II G.) M = EO MO EO'
j= 1 J 00

where G. are the Stokes matrices uniquely defined in terms of the Stokes
J
multipliers. Using (5.5) and certain similarity arguments it can be
shown that all MD can be expressed in terms of two of them. Furthermore,
equation (5.2b) implies that the MD are time invariant. Hence the
above basic RH is specified in terms of two initial parameters (these
two initial parameters are obtained from the two initial data of PIV).
The solution of this RH problem yields Y(z;t) and hence (5.2a) yields

y(t) .

The above basic RH problem can be simplified considerably: (i)

Assume 0 ~ 00 < 1, 0 ~ 000 < 1,°0 *
~; then the above RH problem is
regular. It is interesting that the basic RH problem can be used to
obtain Schlessing~r transformations which shift 00 and 000 by a half-
integer. By using these transformations the general case is reduced
to the regular case. (ii) The basic RH problem can be mapped to a
sequence of two RH problems, one on the line arg z = %and the other
on the line arg z = - %. The first one is continuous (both at x = 0
and x = 00); furthermore, it can be solved in closed form. The second
one is discontinuous both at x = 0 and x = 00. By using standard auxi-
liary functions one maps the discontinuous problem to a continuous
one. Then the theory of continuous RH problems on simple contours
can be used to establish uniqueness and existence of solutions. Elemen-

73
tary solutions of PIV, expressible in terms of Weber-Hermite functions
are natually obtained within the above formalism. We summarize the
results of [7] concerning PIV.

Proposition S.l (Direct Problem). Let YO be the solution of (S.2a)

analytic in the neighbourhood of z = 0 and normalized by the require-
ments that det YO = , Let Y j , j =
and that YO also solves (S.2b).
', ... ,4 be solutions of (5.2a) analytic in the neighbourhood of infinity
such that det Y. = 1 and Y.- Y as Ixl ~ 00 in S., where denotes
J J 00 J
asymptotics, Yoo is the formal solution matrix of (5.2a) in the neigh-
bourhood of infinity, and the sectors 8. are given by
J

11 11 11 311
8, : 4 < arg z < 4' 82:
4 < arg z < '4 ,
311 511 511 711
S3: '4 < arg z < '4 , 84:
'4 < arg z < '4

The rays C, , ... ,C 4 are defined by arg z = 11 11 311 , 511

4' 4' '4 '4 respectively.

Fig. S. 1

Then the analytic functions YO' Y" ... 'Y 4 satisfy:

(i)

where YO(z) is holomorphic at z = O. n/2, YO(z) has

a logarithmic singularity.)

D
( i i) Y. (z) Y (z)eQ(z)(,/z) 00 as ~ 00, z 'i' Diag(8 ,-8 ),
J 00 00 00

2
Q(z) ~ Diag(q,-q), q(z,t) 'i' 2z + zt, Yoo(z) is holomorphic at

z = 00.

74
( iii) Yo (ze 2in ) YO(z)M O' MO
""
e XP (21 •• 0 ' 2i .Jexp( 21'·0 'j , (5.4)
exp(-2in0 0 )

n n
J o i f °0* 2' J 1 if °0 = 2·

(iv) Y2 (z) Y1 (z)G 1 , Y3 (z) Y2 (z)G 2 , Y4 (z) Y3 (z)G 3 ,

2in
Y1 (z) Y4 (ze )G 4 M"", (5.5)

where

G1
""
(1
a J ' G2 ...... ( J , G3 ... (1
c
OJ,

G4
"" ( J M
"" "" exp(2inD ). (5.6)

1. (5.7l

Furthermore, the parameters

(5.8)

satisfy the following consistency condition.

4
(vi) (II G.)M (5.9)
j=l ] ""

Proposition S.2 (Properties of Monodromy Data)

(i) The monodromy data, MD, given by (5.8) and defined in Proposition
5.1, are time-invariant.
(ii) All of the MD can be expressed in terms of two of them.

(iii) (1+bc)exp(2in0J + [ad + (1+cd)(1+ab)]exp(-2in0",,) = 2 cos2n0",

(5.10)
In what follows we formulate a RH problem for the case that
0< 00 <1, 0 < 0"" < 1. This assumption leads to a regular RH problem.
The general case follows by considering this result and Schlessinger
transformations.

75
Theorem 5.1 (Inverse Problem). Consider the following matrix, regular
homogeneous RH problem along the four rays C 1 , ... ,C 4 (Figure 5.1):
Determine the sectionally holomorphic function ~(z), ~(z) = ~j(z) if
z is in Sj' j = 1, ... ,4, from the following conditions:

1. ~,satisfy the jump conditions

J

2'
~1(Z;;) = ~4(Z;; e ~1T)g4(Z;;) (5.11 )
along the rays C2 , C 3 , C4 , C 1 respectively, where

gj 7 eQGje- Q , j = 1,2,3, g4 7 eQG4e-QM~. (5.12 )

D
2. ~(z) - (~) ~(I + o(~)) as I zl -+ ~. (5.13 )

3. ~(z) has at most an integrable singularity at the origin with a

monodromy matrix given by

(5.14 )

In the above, G j , Q, M~, D~, MO are defined in Proposition 5.1.

4. The monodromy data MD, given by (5.8), satisfy the properties given
in Proposition 5.2(ii). Then:
(i) The above RH problem is discontinuous both at the origin and at
infinity. Actually

4 4
II g, z -+ 0; II g, - M , z -+ ex). (5.15 )
j=l J J ~
j=l

Fig. 5.2

76
(ii) To obtain the solution of the above RH problem consider the
following RH problem along the contour e, + e 3 : Determine the
sectionally holomorphic function K(z), K(z) = K,(z) if z in
5, + 52' K(z) = K2 (z) if z in 53 + 54' from the following
conditions:

,. K. satisfy the jump condition

J
2q
de )
-ale

-be'q ~ -,
h on e3
-ale

p (z)
f (5. '6)

(If h" h2 denote h in 52 + 53 and 54 + 5, respectively then h=h,

on e" h = h2 on e 3 .)

2. K( z) (5.17)

3. K(z) has at most an integrable singularity at the origin with a

monodromy matrix given by

(5. '8)

The above RH is discontinuous both at the origin and at infinity.

Actually if gK ,gK denote the jump matrices along e" e 3 respectively
, 3
then

M , z ~(X) (5. '9)

00

However, the above RH problem can be mapped to a continuous one using

the auxiliary functions

(5.20 )

to remove the above singularities.

77
 ~ is related to K via:

KhM, M 7 Diag(l,-a/c), (5.21 )

(i.e., ~ 1

Proposition 5.3 (The Solution of PIV). Let ~(z) be the solution

matrix of the inverse problem formulated ~n Theorem 5.1. Then y(t),

y(t) = -(u1 du
dt + 2t), u ~ -21im ~21(z)e-2q(Z), (5.22 )
I zl"oo
solves PIV.

6. INVERSE PROBLEMS IN n SPATIAL DIMENSIONS, n > 2

Consider the inverse problem associated with the following system of

N first-order equations in n+1 dimensions:

n
~ +0 L J j/, ~ (6.1)
Xo l/,=1 xl/,

where q(xO'x) is an N x N matrix-valued off-diagonal function in m n + 1

decaying suitably fast for large x o' x, and the Jl/, are constant real
1
diagonal N x N matrices (we denote the diagonal entries of Jj/, by Jl/,'
N
... ,J!I,). Alternatively, using the transformation

n
~(z,xO,x) = ~(z,xO,x)exp[i L z!l,(x!I,- oxOJ!I,)]' z €C n , (6.2)
j/,=1

equation (2.13) becomes

n
~x + 0 L (Jj/,~x + iZj/,[Jj/,'~]) (6.3)
o j/,=1

We assume tha~ n ~ N, otherwise the entries of the Jj/, matrices will be

linearly related and one can always reduce n by a change of coordinates.
An inverse problem in this case is defined as follows: Given appropriate
inverse data T, where T is an N x N matrix-valued off-diagonal function
of suitable inverse parameters, reconstruct the potential q. Again
there exists a p which is bounded for all complex z, z €C n • a~/az
n n
depends on appropriate inverse data T(zR,zI,m 2 , ... ,m ), ZR€m, Z{m,
a2T a 2T n
mj/,€ R. T satisfies - - - = ---.Using this equation and introducing
Born variables,
az.az.
l. J
az.az.
J l.

78
 z, m <= =:> y E (6.4)

one obtains a characterization equation for the inverse data:

d Xp d Xp N ~ ~ [ T] (w 0 ' w , X P
~ij ij __~R~~I~_________________ , (6.5)
T (wO,w) 7' T (w O ,w, X)
11

where N is a quadratic function of T. That is, Tij(z,m) is appropriate

inverse data iff the right-hand side of (6.5) is independent of X.
Hence, equation (6.5) serves as both characterizing T ij and defining
;ij. This equation was first introduced by Nachman and Ablowitz [8].
Using equation (6.5) and taking the limit of )l as Ixl'" we show that
the general problem of reconstructing an N x N potential q in n+' spatial
dimensions, is reduced to one of reconstructing a 2 x 2 potential with
entries qij, qji in two dimensions. The inverse data needed for this
reconstruction is precisely ;ij,;ji. This reduction makes crucial use
of the existence of redundant scattering parameters. In this sense it
is the analog of the Born approximation. However, the crucial difference
is that while in the inverse scattering of the multidimensional Schro-
dinger equation one can reconstruct the potential in closed form, here
one can only reduce the general problem to one for 2x 2 matrices in
two dimensions. This reduced problem was solved in [6]. In the follow-
ing, we summarize the results of [9].

Proposition 6.1 (Bounded Eigenfunctions). The function)l (xO,x,z)

defined below, solves equation (6.3), is bounded for all complex values
of z and tends to I for large ~:

f
JR
2 dt; Odt;,
exp [ is i j (x O-t; 0' x, -t; 1 ' z ) ]
i
(x,-t;,) - OJ,(xo-t;O)

where Sij is defined by

n Ji-J j [ xl ( oz ~) I
l:
-L....&
~=,
01 Ji
1

79
Equivalently \.I .. satisfies
1J

i
oij + sgn(or J 1)
2n i

(6.7)

where

(6.8)
(2n)n

Proposition 6.2 (Departure from Holomorphicity). Let \.Iij be defined

byeq. (6.5). Then

fJR n-1
2
dm exp[la
. i
(x,m)]T
ij
(z,m)\.I(xO,X,A
ij
(z,m))E ij , (6.9)

i j i
where B (x O,x 1 ,z), a
(x,m) are defined by (6.6), (6.8) respectively; E ..
. . .th . 1J
is an N x N matrix with zeros in all its entr1es except the 1J , Wh1Ch
equals 1; and A ij and Tij are given by

Aij(
1 z,m )",,(ij
.. zl (z +m, zr ); r=2, .. , n.
R rR r 1

Tij(z,m) '" fJR n + 1 dsOds eXP[-iB ij (sO,Sl,z)-ia i (s,m)](q\.l)i j (so's,z).

(6.10)

Proposition 6.3 (Characterization of T)

(a) Assume that a\.l/az p is given by Eq. (6.9) and the Tij(z,m) is given
by (6.10). Then

(6 • 11 )

80
where

(6.12 )

(b) Assume that a~/azp is given by Eq. (6.9) and that a2~/az az
is symmetric with respect to r, p. Then Tij(z,m) solves (6.11). r p

Following A. Nachman and M. J. Ablowitz we introduce appropriate

Born variables. Then equation (6.11) can be integrated. Furthermore,
we can compute the limit of T ij in the new coordinates as IXpl ~ 00

(see below):

ij ij ,n, be
Let Wo ' w 1 ' wI' 1 2, ... , n £ JR 1 and Xl £C 1 1 2" •••
defined by

n Ji-J j n Ji-J j n Ji
Wo ...
ij r r 2 ij r r E m
r
'T E 10 1 zr ' w 1 :i= E i (ozr)r- r Ji'
r=1 or r r=1 °rJ1 r=2
1

2, ..• , n. (6.13 )

Assume that

(J~-J?)(J~-Jg) ~ (J~-J?)(J~-Jg), for all distinct i,j,r and p~1.

(6.14 )

For convenience of writing we usually suppress the superscripts, i,j

in wO,w 1 ' X. Let z denote z1, ... ,zn' m denote m2 , ... ,m n , X denote
X2"'" Xn ' w denote w1 , .•. ,w n ' Then we have the following.

(a) The inverse of the transformation,z,m ~ wO' w, X is given by

n
+ E
r=1
(6.15 )

(b) rn the new coordinates, Eq. (6.11) with r = 1 becomes

Nij [ 1 - (6.16 )
1p T (wO,w,X), P - 2, ... ,n.

81
(c) In the new coordinates,

where

(6.17)

(d) Let

~.Q, j
11·l (6.18)

~.Q,j
Then the Ili satisfy

' dX ' dW2 exp[i{(xO-xO')wO+wO+(x-x')w}] ..

dX O ..
. q lJ( xO,x
I I)~lJ(
11· I
xo,x I ,wo,w ) ,
x 1 -x 1I - a J l1 ( x O- Xo' ) l

2 .. . .
dx'dx'dw qJl(Xl x' )~lJ(X' x' w w)
o O' .... i 0' , 0' ~.Q,j
11·l

(6.19 )

ij
(e) lim T (wO,w,X)
IXpl-+oo

(6.20 )

(f) The basic characterization equation is given by

dX ' dX ' Nij[T](w W XPl )

PR PI 1 p O ' ' .
------------------------, (6.21)
X - XI
P P
p'
where X denotes

82
 It follows from the above that as
.... p
Ix I~oo, the ~ij,s decouple.
Furthermore, the 07 J ,
. . . .
O~J satisfy a system of two equations depending
1 1
on qlJ, qJ1. It turns out that: (a) By introducing appropriate spatial
variables ~,the 07J, 0 jj satisfy equations in two spatial dimensions.
1 1 . . . . . .
(b) The inverse data needed to reconstruct ~~J, ~IJ (and hence q1J,
qji) can be obtained from ~ij:

Proposition 6.4 (Reconstruction of q). Let

JjJi - JjJi JiJj - JjJi

2 r r 2 , r , r
r = 1, ... ,n, (6.22 )
{lr
* JiJj - JjJi'
, 2 , 2
8r
JiJj - JjJi'
, 2 , 2

where for convenience of writing we have suppressed the dependence of

n
{l r''''Q r on 1,
••
J•
L t
e '"I;' 0 e: lR , '"I;' e: lR ,

x , = ~, , x2 = (6.23 )

3, ... ,n.

Then we have the following:

(a) The system (6.'9) becomes

i
. .(
dJ ~) _ sgn (OI J ,)
~i ~O,~,z - 2ni

[ .~ij(1;' ~)l ij~jj( , 1;"

"'0-"'0'''''-'''' ,z q ~i ~O''''' '~2 -
1;" I;' 1;"
X exp 115

(6.24)

where

n Jj - Jj
2 ,
Z
r='
E (z (l
r r
+
Jj , , 8
Ji zr r)'

, ,
Ji_ Jj
(6.25 )
°1

83
(b) ;ij in the new coordinates becomes

;ij(z,m) = fF n+1 d~Od~1 exp[-i~ij(~O'~l'z) +

Ji n
~) + m 'l]qijjJ~~j(" ,I z) (6.26)
re, r le,O'e",
Jl r=3
1
where
n
3, ... , n. (6.27)

(c) The inverse data associated with (6.24) and the analogous problem
for ~ji
... j , ~ii
... j are given by fij, fji. Let
Ji
",i j 2
~. ( z , ~ 2 -~ 1 J i
1

+ i (6.28 )

Then

Ji
x(qijcp)(~O'~1'~2 - (~1-~1) ), ~3'···'~n'z). (6.29 )
1
Equations (6.1 )-(6.3) with a -1 lead to a system which appears
to be physically more interesting: (a) Since the system is hyperbolic one
may consider the physically important question of inverse scattering (IS);
i.e., given a scattering amplitude function S(\,k) find the potential
q(xO'x). (b) A special case of the above system, namely if the JiS are
constrained by

J~
p
- Jjp Ji - Jj
p p
p, r 1, ... ,n, 1,j,~ 1 , •.• , N, (6.30)
JQ, - Jj Ji - Jj
r r r r

is associated with the N-wave interaction in n+1 spatial and one temporal
dimensions [10]. The above system can be considered as a limiting case
of (6.1)-(6.3) [8]. Alternatively, it can be considered on its own right
[11]; the problem of reconstruction can be reduced to one for a 2 x 2
matrix problem in two spatial dimensions.

84
ACKNOWLEDGEMENTS

This work was partially supported by the Office of Naval Research under
Grant Number N00014-76-C-0867 and the National Science Foundation under
Grant Number DMS-8501325. A. S. Fokas thanks Professor Lakshmanan and
his school for their warm hospitality.

REFERENCES

[1] R. G. Newton, Scattering Theory of Waves and Particles (Springer,

Berlin, 1982).
[2] F. D. Gakhov, Boundary Value Problems (Pergamon, New York, 1966);
N. P. Vekua, Systems of Singular Integral Equations (Gordon and
Breach, New York, 1967).
[3] S. Prossdorf, Some Classes of Singular Equations (North-Holland,
Amsterdam, 1978).
[4] L. Hormander, Complex Analysis in Several Variables (Van Nostrand,
New York, 1966).
[5] R. Beals and R. R. Coifman, Comm. Pure Appl. Math. 37 (1984) 39.
[6] A. S. Fokas, Phys. Rev. Lett. 51 (1983); A. S. Fokas and M. J.
Ablowitz, J. Math. Phys. 25 (1984) 2505.
[7] A. S. Fokas, U. Mugan, M. J. Ablowitz, A Method of Solution for
Painleve Equations: PIV, PV, preprint INSi73 (1987).
[8] A. I. Nachman, M. J. Ablowitz, Stud. Appl. Math. 71 (1984) 251.
[9] A. S. Fokas, J. Math. Phys. 27 (1986) 1737.
[10] M. J. Ablowitz and R. Haberman, Phys. Rev. Lett. 35 (1975) 1125.
[11] A. S. Fokas, Phys. Rev. Lett. 57 (1986) 159.
[12] A. S. Fokas and M. J. Ablowitz, Comm. Math. Phys. 91 (1983) 381.
[13] A. V. Mikhailov, Physica 3D (1981) 1&2, 73-117.

85
Gauge Unification of Integrable Nonlinear Systems
A. Kundu
Theoretical Nuclear Physics Division, Saha Institute of Nuclear Physics,
92 A.P.C. Road, Calcutta 700009, India

Gauge equivalence of generalized NLS type equations is established with

its possible application to find out soliton solutions and with the
generation of new integrable systems. Through certain gauge choices
some realistic models are exactly solved. Explicit auto BT for different
classes of equations are also obtained by gauge transformation.

1• INTRODUCTION

Some twenty years back discovery of nonlinear integrable field models

created immense excitements. In fact when in 1967 Kruskal et ale [1]
showed the Korteweg-de Vries (KdV) equation to be completely integrable
with soliton solutions having beautiful properties, it was not considered
by all to have any universal appeal. Subsequently, however, mainly
with the works of Zakharov and Shabat [2] and AKNS [3] the existence
of a whole class of such systems was revealed. Recent years have now
witnessed a rapid growth of such members in the family of integrable
systems. In the present day, however, the situation is somewhat reversed.
There is already a large collection of members in the integrable circle,
seemingly all with their own originality and independence, demanding
individual care and analysis. Besides these bonafide members there
is also a vast number of candidates from the real world with their
nonlinearity, awaiting their recognition at least 'near' to some integ-
rable members. In this complicated affair of today, there is then a
natural urgenoy to work out some sort of unification scheme to group
together integrable members of the same class by finding out their inter-
relationship and pinpointing some genuine representatives, from which
others originate. Thus individual treatment of each one of them, as
is the standard practice of today, would become unnecessary and full
information about only a few basic equations would be sufficient to
solve the rest, generated from them. The aim of our article is to present
such a unification scheme through gauge equivalence , to trace out
genuinely independent equations, find out gauge relations between
different systems and recognize real world candidates for their member-

86
ship of the integrable family. Though our ambitious programme is not
yet completed and we confine here only to nonlinear Schrodinger (NLS)
type of equations, we are able to achieve a breakthrough in this line
recently [4-8]. Thisscheme also enabled us to use the same machinary
to generate a number of new integrable equations, some of them coinci-
ding with the known nonlinear realistic systems or coming near to them,
thus solving exactly few long standing hydrodynamic problems. Along
with the above idea of gauge transformation (GT) between different
systems we also use it for another aspect, i.e., for finding Backlund
transformation (BT) between different solutions of the same equation.

Interest in gauge equivalence between different nonlinear dynamical

systems and the corresponding models was boosted up after the remark-
able works of Lakshmanan [9], Pohlmeyer [10] and Zakharov and Takhtajan
[11]. On the one hand, sine-Gordon (SG) and nonlinear a model was
linked [10] and extended [12] and on the other hand Landau-Lifshitz
equation (LLE) was gauge related to nonlinear Schrodinger equation (NLSE)
and generalized [13-14]. However, besides the nonlinear systems men-
tioned above, there are various other NLS like equations in the integ-
rable family, e.g., NLSE of repulsive type [15], attra;tive-repulsive
NLSE [16] and also systems like derivative NLSE [17], mixed NLSE [18],
Chen-Lee-Liu equation [19], modified derivative NLS [20], Gerdjikov-
Ivanov equations [21], etc. It is natural to ask whether they are inter-
related and whether there exist any Landau-Lifshitz type equation,
gauge equivalent to them. Besides, these integrable members, there
are also some well-known hydrodynamic equations with high nonlinearity,
such as Johnson equation [22], Beney's first [23] and second [24] kind
of long-short wave interaction equations. Whether integrability of
these real models, which mostly received approximate treatment, may be
recognized? Our aim is to answer to the above questions (Fig. 1) by
generalizing the gauge equivalence scheme and proposing a LLE with non-
compact Grassmannian manifold SU(p,q)/S(v(r,s) x v(u,v)). Through
H-gauge transformation we are also able to connect realistic systems
with integrable models and thus solve them exactly.

The organization of the paper is as follows. In Sec. 2 the general

scheme of gauge equivalence (GE) is outlined and demonstrated for NLS
and LLE. In Sec. 3, GE is applied to noncompact manifolds recovering
old and yielding new results. In Sec. 4 we demonstrate the applicabi-
lity of GE for extracting soliton solutions and other informations.
Section 5 finds generalized LLE equivalent to derivative and mixed NLSE's.
In Sec. 6 we generate through H-gauge transformation a hierarchy of
higher-order equations, connecting a number of integrable systems. In

87
 Fig. 1 Are there
any interrelation
between these mem-
bers? Are there some
genuinely fundamen-
tal members?

~
(:) FAMILY

Sec. 7 we connect realistic equatiomwith integrable systems and solve

them exactly. Section 8 demonstrates the applicability of GE for finding
BT even where earlier attempt [25] failed. Section 9 is the concluding
section.

2. GAUGE EQUIVALENCE OF NONLINEAR

EVOLUTIONARY EQUATIONS

Let the linear problem or Lax pair associated with the given integrable
equation tq = 0 be expressed as

¢ U¢ , V¢, (2.1)
x

where the Jost function ¢ and U, V are complex matrix functionsof the
field q(x,t), qx,the independent variables x and t and the spectral
parameter A. The compatibility of system (2.1), i.e., ¢xt = ¢tx:
Ut - Vx + [u,vl = 0 yields the original nonlinear system tq 0 by appro-
priate construction of U and V. For real A, ¢ E G, G being a compact
or noncompact Lie group and under the local gauge transformation

¢(x,t; A) l E G (2.2)
A =A
o

the Jost function c-hanges as

-1
g (x,t; AO) ¢ (X,t;A) (2.3)

88
and the corresponding new linear system is

<1>' U'<I>' , <1>' V'.j> , (2.4)

x t
with
-1 -1 -1 -1
U' 9 Ug 9 gx 9 (U-Uo)g, gx g = U = U
0
1\=\
0
-1 -1 9 -1 (V -1
V' 9 Vg 9 gt - VO)g, gt g V V ( 2.5 )
0
1\=\
0

The compatibility of (2.4) gives now the GE equation

o
(2.6)

relative to a new field S. There exists also another type of spectral

parameter dependent GT: $(x,\) G(q,q' ,A) <I>(x,\), which generates
an auto Backlund transformation (BT) mapping a solution q to a different
solution q' of the same system.

Let us now look into the well-established [11] GE between NLSE and
LLE to make the picture clear. Scalar attractive NLSE

o (2.71

is given by the Lax pair

U V (2.8)

where A1 = ia 3 , AO = ~a+-~a-, B1 = 2A O ' B2 = 2A1 and BO = -i(l~fa3+

~x a+ +~xa-), where a± = a 1 ±ia 2 and ai' i = 1,2,3 are Pauli matrices.
If $ = g-l<1> with 9 =\~=\o ' where <I>± is the Jost functions with

<I>±
, exp(ia 3 x) and scattering matrix T(\) <1>-1<1> , then defining
x -). ±ro +
-I
S+ = 9 a 3 g we get the GT Lax operators to be

U' = 9 -1 [(A 0 + \A1 ) - (A + \oA1) ]g i (\ \ )S

0 0

and

V' (2.9)

where we have used

SS -S S
x x

89
 1
(2.9) yields the GE LLE: St = 2i[S,Sxx1 - 8A OSX. Note that for the
-1
choice g =~_(X'AO) one similarly gets (2.10) S g_ 03 g _ which is
-1 -1
related to S+ as S_ = To S+To. But if we try to set ~± = g± ~±,

would satisfy different equations like ~±x - U±~± thus violating

Jost function condition. On the other hand with the choice ~± =
we immediately get the relation between the scattering matrices:

T' (A) T(A) , ( 2 . 11 )

which helps to bypass the individualIST investigation of a system, if

one knows the corresponding information of its gauge equivalent counter-
parts. The integrability property is also obviously preserved. Another
way of looking at GT for 2x2 matrices is through the trihedral Ea'
as used by Pohlmeyer, Eichenherr and others, where one introduces Ea=
eo
a
= g-1 0 g, a = 1,2,3 with the relation E Eb = g b + if bE, E+ =
a a a acc a
03Eao3' where gab is a symmetric and fabc is an antisymmetric tensor
and with the equation E A bEb' E t = B bEb. Hence in our case
-1 ax ~T a a
S = E3 So = g 03 g , E± g O±g with the equation

Sx 2(~E+ + 1jJE), E+x -2iAE+ - 1jJS, Ex 2iAE - ~S (2. 12a)

and

St 2(-i~
x + 2A~)E+ +2 (i1jJx +
2A1jJ)E ,

E+t = 2i(11jJ 12 - 2A2)E+ - (i1jJx + 2A1jJ)S. (2.1 2b)

The 0(3)-invariants are tr(s2),

x tr(St'S x ) and tr(S2)
t .
Eliminating
E± the final equation for S is again given by LEE (2.10').

3. GENERALIZED LLE WITH NONCOMPACT GRASSMANNIAN

MANIFOLD AND GAUGE EQUIVALENT NLSE

We propose a generalized Landau-Lifshitz equation (LLE) with non-compact

manifold:

SEM G/H,

G SU(p,q), H S(U(u,v) ® U(r,s)), (3.1)

where u+v = m, s+r = n, u+r p, v+s = q, p+q = m+n = N and find its
gauge equivalent system using the above scheme. The corresponding
linear system is

90
 u i A5, v (m+n)/2mn (3.2)

with 5 satisfying 5 2 = aI + b5, 5-

f, diag(Iu,-I v ) and f2 = diag(Ir,-I s )' a, b are constants and Ii
is a ixi matrix. G/H being a symmetric space, it is always possible
to express 5 = gEg
-, ,
9 E G and E = diag(I 1m, -I In), which gives
, _, , m n _,
on using 5 = 9[E,L ]g- , L = 9 g , -2[5,5] = 29E[E,L,]g the trans-
)J )J )J )J x
formed operators

U' 9
-, Ug iH - A,

V' 9
-, Vg (3.3)

B. Taking A and B in the explicit form

A B
~x +~A2-A, ~ 1
b2

with W
+
f,W f 2 ,
-
Ai = -Ai'
i
tr(A,+A 2 ) = 0, Ai = Hx and b, =
-
(ww- ,
)JIm) +H t ,
b 2 = -(W~-)J(m/n)In) + H~. The compatibility of (3.3) yields the gauge
equivalent matrix NL5E

(3.4)

It is worth mentioning that in deriving (3.4) we have set simply A,=

iAoIm/m, A2 = -iAoIn/n. In general, contrary to the usual belief,
LLE is GE not only to NL5E but also to all its H-gauge equivalent part-
ners (see Fig. 2), which may also be obtained by different choices of
function Hi. We, therefore, conclude that LLE (3.') with noncompact
manifold is gauge equivalent to matrix NL5E (3.4) with internal symmetry
group U(u,v) ® U(r,s) along with all its H-GT systems. Note that for
trivial boundary condition on fields W, AO is arbitrary real and may
be trivial ["], but for a nontrivial boundary condition, which is impor-
tant in noncompact cases AO is nontrivial, since it should always be
from the real spectrum [6]. The following particular cases recover
old results, yield new relations and answer some of the questions raised
in 5ec. ,.

91
 G=SU(p,q) /r- _ _ _ ~~___H_L,S-
H= S(U(p)(x)U(q) ) • ~
r---------:--:::.-----=-:-::-'S"f--' - - - - - - -- -- MNLS
SE:G/H p= 1

VNLS-

N= 2

~-- - -------- SNLS

SNLS
+ +_
H = S(U(m)(x)U(p,m,q»
Sf SU(p,q) ,- __ ~f~ _____~~*
S
'---m-=--:1:---------( U-(-P- I..,q-s:)(-v,CO-- - - - -- ~:ts.LS·-

-'~~ -~---~~---~~~:-\_s-
Fig. 2 Gauge equivalence between generalized LLE with noncompact Grass-
mannian manifold and NLSE along with its H-gauge equivalent systems and
their different reductions. (±) represents 'attractive' or 'repulsive'
while (+-) represents 'attractive-repulsive' type cases

3.1 Examples

Let u = m, v = 0, r = p-m > 0, that is,

S E SU(p,q)/S(U(m)xU(p-m,q» (3.5)

We further suppose that (i) p = N, q = 0, r 1 = r 2 ~ 1, hence S = S+

recovering equivalence of LLE with compact manifold S E SU(N)/S(U(m)®
U(n)) [12] and consequently for m = 1 with S E CpN [13] and for N = 1
with standard S E S2 [11]. For extracting the relevance of noncompact-
ness we suppose that (ii) p = m, q = n giving r1 = I, r2 = -I, -
~=-~
t
,
that is LLE with SU(p,q)/S(U(F) ®U(q) is GE to matrix NLS of repulsive
type. p = 1 connects LLE with SU(1,N-1)/U(N-1) and vector NLS (VNLS)
of repulsive type

o. (3.6)

N=2 gives a SU(l,l )/U(l) version of LLE and related repulsive scalar
NLS [15]. (iii) m = 1 reduces (3.5) to LLE with noncompact manifold
SU(p,q)/U(p-1,q) and GE attractive-repulsive VNLS [16]

0, (3.7)

92
which in limiting cases leads again to VNLS of attractive or repulsive
type. The above equivalence is schematically depicted in Fig. 2.

It may also be shown that [4] SU(2) LLE with easy axis (6)0) aniso-
tropy (ALLE) is gauge equivalent to attractive SNLS, while the aniso-
tropic SU(I,I) LLE to repulsive SNLS and easy plane (6<0) case to AKNS
system.

4. APPLICATION OF GE FOR SOLITON SOLUTION

Most of the early works, besides establishing the GE, largely ignored
the possible applications of such a beautiful relationship. We aim to
utilize the GE for extracting information of LEE without the associated
ted ius 1ST calculation but only using well-investigated properties of
its GE NLSE. As shown in §2, the Jost function of LLE ¢± may be
-1
expressed through that of NLSE ¢a as ¢± = ¢o ¢±' where

¢1 2*

[
± E¢±
¢±
¢2
±
1*
¢± 1
is the Jost solution of spectral problems connected with NLSE and E =+1
correspond to NLSE of attractive and repulsive type, respectively rela-
ting to LLE with S E SU(2)/U(1) and SU(l,l )/U(l) manifolds. As shown
before the scattering matrices of GE system are identical. The field
solution of LLE may be expressed through NLSE Jost solutions as

S
-1 1 ( S3 0 3 + S - 0 - + S+o+) , 6 det¢
¢ 0+ 0 3¢ 0+ r0 0 0'

_ 2¢ 1¢ 2 +*
S3 1¢ ~12 +E 1¢ 021 2, S o 0'
S ES . (4.1)

4.1 SU(2)/U(1) LLE Solution from

NLSE of Attractive Type

The solution to LLE is normally extracted through tedious 1ST calcula-

tions [26]. We, however, recover the soliton solution through GE using
the known results of NLSE [2] given by the Jost function for N-soliton
solution:

N C ¢
L -x=x-
n n
exp(iAnx), (4.2)
n=l n

93
where ~n ~(X.An)' An being the discrete spectrum. For simplicity
we consider N = 1 giving 1-s01iton solution of NLSE as

N
l\I(x) -2 E C ~' (x)exp(iA x) I
= -2in exp(iy)sech y, (4.3)
n=l n n n N=l

where A1 = ~+in, y = 2~x + 4(~2-n2)t + ~o and y

the other hand from (4.2) one obtains

~1 _ ptA ) exp[i(A-A )x)sech y,

o 0 0

~2 (4.4)
o

with p(A O ) in I(A-A o )' which gives directly from (4.2) the LLE soliton
solution

sech 2 y,

arg(S+) 2 A x - Y - tan -1{ [(A ~) In ) coth y } (4.5)

o 0

coinciding for AO =0 with the result of ref. 26 found through direct 1ST.

4.2 Noncompact SU(l,l}/U(l} LLE Solution

Through Repulsive Type NLSE

The noncompact LLE model may be given by the Hamiltonian

2
H S 1 )dx (4.6)
x

1 and S = SaT E SU(l,l )/U(l). This is a new model,

a
which may have physical applications. For finding its soliton solution
we may apply again the GE with NLSE of repulsive type established here
and use the known 1ST information of the latter system [15). The 1ST
programme of repulsive NLSE is rather complicated due to nontrivial
"
b oun d ary con d ltlon l'lm 11\1 12 ~~.
2 Here ~ = ± (2
A - ~ 2)i actually serves
the role of spectrall~1tameter and ~(A) is defined on a 2-sheeted
Riemann surface with cuts at (-oo,-~) and (~,oo). The bound states are
, at ~n = (2
An - ~ 2)i = lV
' ' h 1 12 2
glven n Wlt An < ~ • The Jost solution cor res-
ponding to 1-s01iton of NLSE is given by [15)

exp ( - i ~x ) [ 1 J
A-~ [[: ]- ~ [:; (4.7)

94
where

K1 (y) = vf(y), K2 (y) = v(y-iv)f(y), f(y) = [1+exp(2y) ]-1,

2 2
1',;2 = A2_ 1, A1 = y, v = 1 - Y and y

l\I(x,t) [ (y + i v ) 2 + exp ( 2 y) ] f ( y ) . (4.8)

Using € = 1 and (4.7) one gets from (4.1) the soliton solution for (4.6)
LLE in the form

(4.9)

where we have chosen AO - 1',;0 = y for simplicity, which, however puts

restriction on the soliton velocity u 2y = 2(A O - 1',;0)' 2 >u > O.
We also have the restriction I Ad> >L 1 since AO must be from the
continuous spectrum.

5. NLS TYPE EQUATIONS AND

GAUGE EQUIVALENT EXTENDED LLE

Using the technique similar to that applied above for standard NLSE, we
may also find new extended LLE, gauge equivalent to various known NLS
type equations [5].

5.1 DLL Gauge Generated from DNS

The scalar derivative NLS (DNS)

o a > 0 (5•1)

may be given by the linear system

v (5.2)

+ - - 2 + 2-
where A = qo ± q*o , B = (iq
+ alql q)o + (±iq* +alql q*)o .
x -1 x -1 2
Defining as before S = g 03 g one gets SS = 2A a gAg, SS =
2 2 1 2 2 ~1 0 x
-4 Aoa g- 03A g and S(St-2aAoSx) = 2aAog Bg. Repeating now the
above procedure for GT (2.5) one obtains the new system

95
 U'

V'

+ i

A-A
+ __ 0 (SS - 2uA 2SS ) (5.3)
21.. t 0 x
o
yielding the extended LLE (DLL) [5]:

(DLL) : --2
S3x 0, U >0 (5.4)
4uAo
with S E SU(2)/U(1 )[SU(1,1 )/U(1)] corresponding to +(-) signs. Note
that the integrable system (5.4) has also been found recently through
an altogether different method [27].

5.2 XLL Gauge Generated from XNS

The known mixed NLS (XNS), a hybrid of DNS and NSE,

(5.5)

given by somewhat complicated U, V operators [18] is gauge trallsformed

similarly to [5]

-1
U' g {o03 + aA}g

oS + a SS
c x
-1 2
V' g . { k03 + bA + aB +yo3A }g

kS + E.ss (5.6)
c x

where c,o, a, k, b,y, b are different expressions depending on A, Ao'u

and B [5]. The flatness condition of (5.6) yields gauge equivalent LLE
type equation in the form (XLL)t

tXLL given in refs. 5,6 may be simplified to this form.

96
 0, (5.7)

where y = 41.
o
{/2B - 0.1. } and p
0
0./(20.1.
0
- 12]3)2. The ± sign as before
corresponds to S E SU(2)/U(1) (SU(l,l)/U(l)) cases. One gets also
the invariant relations

(5.8)

Note that XLL (5.7) has exactly the same form as DLL (5.4) but only with
different coefficients, which clearly reduces to the corresponding coeffi-
cients of LLE and DLL in particular cases a = 0, a '" 0 and a '" 0, a= 0,
respectively. The coincidence of XLL and DLL reflects the fact that
their gauge equivalent Schrodinger type equations, e.g., XNS and DNS
are U(l )-gauge related (see§6.1), which does not change the corresponding
LLE system as shown in §6.

5.3 MDLL Equivalent to DMNS

A modified DNS [MDNS) given by

(MDN~): iqt + (q/p)xx 0, p (5.9)

corresponds to the linear system

U (5.10)

with A as in (5.2) and

-1 -i q ].
B D p [
±q* i

Repeating the above procedure we now get the gauge transformed operators
as

U'

( 5 . 11 )

97
yielding a new LLE type integrable equation (MDLL) [5]

(MDLL) : (5.12 )

2 2-!
with X = (l+tr(Sx)/8A o ) , with S E SU(2)/U(l)[SU(l,1)/U(l)] corres-
ponding to +(-) signs in (5.9).

6. UNIFICATION THROUGH H-GAUGE TRANSFORMATION

It is interesting to note that from the matrix NLSE (3.4) one can generate
-1
through the local H-gauge transformation ~ ~ h2 ~ hl relative to group
element h = diag(h 1 ,h 2 ) E H = S(U(u,v)® U(s,t» a class of new higher-
order equations

0, (6 •1 )

where Hi h~la h .. Since under h(x,t) E H-gauge transformation, the

]1 1]1 1 1 1 1
model field is invariant: S' = g'~ g'-l = gh~h- g- = g~g- = S, the
LLE system remains unchanged under such transformation, whereas the
, -1 . -1
equivalent matrix NLSE changes as Ui = h Uih+(lA~-h hi). We consider
below the simplest case of h = exp(i00 3 ) EU(l), G = SU(2)(SU(l,1» and
show how certain new hierarchy of integrable equations can be generated
unifying different systems for various particular choices.

6.1 U(l)-Gauge Generated Higher Order

Integrable Systems

The XNS (5.5) is transformed to the higher-order equation

2i0
Q e q. (6.2)

It is evident, that a choice of 0 = ¥(x + ¥t) and a Galilian transfor-

mation with V = -28/a reduces (6.2) to DNS (5.1) and thus establishes
gauge equivalence between XNS and DNS while for a = 0 we directly get
NLS. We may, therefore, consider gauge generation from later two systems
only.

98
 (i) Gauge generation from NLS: One interesting choice for B ,
~
e.g., B~j) = P j , B~j) = ~j where (Pk)t
(~k)x yielding conservation
laws from the Ricatti-equation generates a hierarchy of integrable
systems [8]

iQt + Qxx+BIQI2Q + 2 {(~j-2P~-iPjx)Q - 2i P j Qx} 0,

j = 1,2, ....
One may easily construct for these systems exact solutions, Lax pairs,
infinite conservation laws, etc. [8]. In the simplest case of j = 1
we have a(l) = - olqI 2, a t(l) = io (qq* - q*q ) yielding the new integ-
x x x
rable system with fifth-order nonlinearity

reducible to a linear equation iqt + qxx = 0 for B = O. For choice

a = a(x) one gets a NLS with variable dependent coefficients. Other
choices of a will be discussed in the next section.
(ii) Gauge generation from DNS: Similar choice as the above gives
again another hierarchy of integrable systems generated from DNS [8].
The simplest case gives Johnson type equation

which for the choice 0 - %reduces to Chen-Lee-Liu [19] equation (CLL).

and for 0 ~ to Gerdjikov-Ivanov (GI) equation [21]

Thus we have established the gauge relation between a large number of

systems. Consequently, the soliton solutions and Lax pairs for these
systems may also be trivially constructed from those of DNS thus avoiding
lengthy sophistic,ated approaches [28,31].

7. EXACT SOLUTION OF REALISTIC MODELS THROUGH GE

Johnson derived [22] the following fifth order nonlinear equation (JE)
for the amplitude of the fundamental wave.

(JE): iAT-alA~~-a2IAI2A + a 3 1AI4A + ia 4 1AI2A -iaSA( IAI2)~-a6AaT=o,

(7.1)

99
 2
where and a i are real numbers. Previous attempt [22] to solve
8~=oIAI
this equation was both tedious and approximate. We, however, notice
that (7.1) may be gauge transformed to XNS

(7.2)

oa 1+a 4 +a 5 +3a 6 +4 = O. Thus

with a = oa 1 (a 6 +4) + a 5 and a constraint
(7.1) can be solved exactly extracting the Lax pair and different soliton
solutions [8].

In describing the mechanism of interaction between long and short

waves Benney has derived the following famous equations. The first
of this kind may be given by [23]

(LS1): Qt = -iQxx + k 3AxQ + k3AQx-ik4A2Q+2isIQI2Q 0, (7.3)

where A(x,t) describes the long wave, while Q(x,t) is associated with
the short wave. The second equation due to Benney is [24]

d a ( IS I
2
(LS2) : )x + cpLL x ),

St + C 2
g S x - ioLS = E(iSS xx + iylsI s + mLS x. + nSL x )' l7.~)

with Land S describing the long and short waves, respectively. We

observe again through gauge equivalence, that for certain parameter
choice both the above equations become exactly solvable and connected
with the NLS equation. In particular, for the choice k3 = 2k3 = -2a,
2 -2i8
k4 = -a by U(1) - GT described above (7.3) is reduced for q = Qe
with 8
x ~A, 8 t = as( IQI2) to NLSE

(7.3 ' )

where 5 -s(i+a), yielding easily the explicit Lax pair through Ui=
h- 1U.h - h- 1h i , h = exp(i8a 3 ) and soliton solutions. For example,
1
when 5 > 0 the exact 1-soliton solution of (7.3) may be given by

2tanh2vz
______4~V__ sech22vz, Q 2v
A sech2vz expi( wt+kz
vl1+ al Is 2vP + exl
where z = x - vt. Note that for a = -1 one gets s = 0 reducing (7.3)
to a linear equation. It is worth mentioning that for a more restricted
set of parameters the solvability of (7.3) has also been established[23]

100
through much involved prolongation theory. A similar GT with q = Se
2iG ,
where Gx = L - C g /4 and Gt = -c l L + E(a!S !2 5Q 2 2
+ 2 L ) + cg/8 transforms
(7.4) directly into the NLS equation of the form iqt+ q + y!q!2 q =O
xX_l
with Y = E(y-2a) and the choice 8 = 2(c g -c l ), 4S=2n=m=4E and E~=4.
However, if E~ - 4 = p tOone should include in (7.4) a higher order
2 . . .
nonlinear term like pL S to restore its gauge equivalence with the
NLSE.

Following the same line of argument one can show that the Langmuir
type equation (LTE) with some additional nonlinearity

[(n 2 +in )Q + 2inQ 1

x x
o

n - n
t x

is also reducible to NLS for the choice

Gx

Now we may depict the result of gauge unification scheme in Fig. 3.

CLL
DLL gEG gEH

G 11

CLL-NS
XNS
XLl - - - - - - - --(::-.::::=~-----{) JE
gEG gEH

ALLE(.t:.>O)
0-.. ..... LTE
G=SU(2)--- g~H
- - - -- - - --~+E=-------OLS 1
ALLE(.t:.~9L--~--------~S! HNLS-
0- LS 2
G-SU(1,1)

Fig. 3 Extended qauge equivalence scheme of different LLE and NLS type
equations. This establishes the relationship between different members
of the integrable family as well as real world candidates thus answer-
ing the questions raised in Fig. 1. Here H = U(l) and G = SU(2) (or
SU(l,l) corresponds to attractive (+) (or repulsive (-)) case

101
8. EXPLICIT AUTO BT THROUGH GT

As mentioned in Sec. 2, another interesting application of GT is the

auto BT for integrable systems, which is otherwise a tricky problem
[29] demanding 'guess' work. Here the parameter dependent gauge element
G(q,q' ,\) satisfies the equation

Gx , (q,q' ,A) = Ui (q' ,\)G(q,q' ,\)-G(q,q' ,\)U i (q,\), (8.1)

1
where xi = (x,t), i = 0,1. For finding out explicit Backlund Transform
(BT) for some concrete systems like KdV, sG, NLS, DNLS, LLE, modified
ab N ab , n
DNLS, we may express the elements G = L (an )/(1\) and from (8.1)
n=O n
choose N by matching coefficients of equal powers in (i\) to get con-
ab
sistent closed set of equations for a The value of N and explicit
solutions would depend on particular form of U1 [7].

:~:
11 22
Y
A::Sf::s:::Sw: :::: U1 =[!\ -{\] , which results in the choice
2' 2' 12 21 h
a = a o + 1\, a = 00 - 1\, a ~ 80 = a = Yo' were ao'So' Yo,oo
c
are independent of \. This leads to the relation a o v = u x - 2(u+v),
1
ao = 2(r' (u+v)-r(u-v)) with u = q'+q, v = q'-q. For deducing now
th~ BT for particular systems we have to imply different reductions
of AKNS, e.g., for KdV r = -1, q = -w x ' for sG r = -q = ~x, for NLS
r = ±q*, etc. leading to explicit BT's for these systems [7].

8.2 BT for Kaup Newell Problem

11 2 22
We have U1 given by (5.2) dictating the choice a = 1 + a2\ , a

-1 +02\2, a 12 ~1\' a 21 = Yl\ leading to the relation

i(q~02-qx02) = -2(q'+q) + (q'r'-qr)(q'02+qa2)'

which yields easily for r = ±q* the BT for DNS (5.2). Since the mixed
DNLS (5.5) is shown in Sec. 6 to be GE to DNS, the BT for XNS is also
easily obtained.

8.3 BT for WKI Problem: Modified DNS

Earlier attempt [25] for finding BT of modified DNS was not successful.
We, however, notice that since LLE and NLS are gauge equivalent, one
may construct the BT of the former system through the latter, which
-1
takes the form [7] S' = B 1SB 1 where S', S are different solutions
-1 -1
of LLE and Bl = ~o Go a3~0' where Go is the BT-gauge of NLS and ~o
is its Jost function at \ \0. Using now the fact [30] that a change
of variable iq = S-/S3, X = - Is3 dx transforms LLE to (5.9) we may
derive BT for the latter system [7].

102
9. CONCLUSION

We have demonstrated that through gauge equivalence method it is possible

to 'unify' a large number of integrable systems, pinpointing a few

basic ones. Using this machinery one may also obtain full information
of a system, for example, know its Lax pair, soliton solutions, and
Jost functions without doing any tedious 1ST investigation, but from
its gauge equivalent, well investigated counterpart. Application of GE
also allows us to generate new integrable systems along with their Lax
pair and sometimes solve exactly existing realistic systems. It also
opens up an elegant way of finding explicit BT of a large class of
systems. One of our interesting observations is that some realistic
nonlinear equations may turn into exactly integrable systems and some-
times they do so when some still higher nonlinear terms are added,
whereas the usual practice is to throwaway such terms in real models
to give the equations some 'elegancy'. Our investigation shows however
that in certain systems one may be able to gauge transform them to
some known integrable models without neglecting any of their nonlinear
terms (e.g., HNLS, LS1, LS2, Johnson equation), while in some other
cases higher order nonlinear terms are necessary to add for restoring
the integrability. Some models may even be linearized through such
procedure. Therefore, in handling realistic equations the possibility
of application of GE should be carefully explored.

In spite of some encouraging achievements, our ambitious programme

of gauge unification is far from being complete. The extension of
such equivalence to the corresponding quantum models is indeed a challen-
ging problem, which is being taken up by us at present.

ACKNOWLEDGEMENT

I would like to extend my sincere thanks to Prof. M. Lakshmanan for

giving me the opportunity to participate in this excellent Winter School
and also for suggesting me the possible simplification of the XLL
equation.

REFERENCES

[1] C. S. Gardner, J. M. Greene, M. D. Kruskal and R. M. Miura, Phys.

Rev. Lett. 19 (1967) 1095.
[2] V. I. Zakharov and A. B. Shabat, Zh Eksp. Theor. Fiz. 61 (1971)118.
[3] M. J. Ablowitz, D. J. Kaup, A. C. Newell and H. Segur, Stud. Appl.
Math. 53 (1974) 249.

103
[4a] A. Kundu, Lett. Math. Phys. 6 (1982) 479.
[4b] A. Kundu and o. Pashaev, J. Phys. C16 (1983) L585.
[5] A. Kundu, J. Math. Phys. 25 (1984) 3493.
[6] A. Kundu, J. Phys. A19 (1986) 1303.
[7] A. Kundu, J. Phys. A (1986) (in press).
[8] A. Kundu, Physica D (1986) (in press).
[9] M. Lakshmanan, Phys. Lett. 61A (1977) 53.
[10] K. Pohlmayer, Comm. Math. Phys. 46 (1976) 207.
[11] V. I. Zakharov and L. A. Takhtajan, Teor. Math. Fiz. 38 (1979)26.
[12] H. Eichenherr and J. Honnerkamp, J. Math. Phys. 22 (1981) 374;
J. Honnerkamp, J. Math. Phys. 22 (1981) 277.
[13] S. J. Orfanidis, Phys. Rev. D21 (1981) 1513.
[14] R. Sasaki and Th. W. Ruijgrok, Physica 113A (1982) 388.
[15] V. I. Zakharov and A. B. Shabat, Zh. Eksp. Teor. Fiz. 61 (1973)118.
[16] V. Makhankov, Phys. Lett. 81A (1981) 156.
[17] D. J. Kaup and A. C. Newell, J. Math. Phys. 19 (1978) 798.
[18] M. Wadati, K. Konno and Y. H. Ichikawa, J. Phys. Soc. Japan
46 (1979) 1965.
[19] H. H. Chen, Y. C. Lee and C. S. Liu, Phys. Scr. 20 (1979) 490.
[20] M. Wadati, K. Konno and Y. H. Ichikawa, J. Phys. Soc. Japan
47 (1979) 1698.
[21] V. S. Gerdzikov and M. I. Ivanov, Preprint E2-82-959 JINR, Dubna
(1982) .
[22] R. S. Johnson, Proc. Roy. Soc. Lond. A357 (1977) 131.
[23] A. C. Newell, SIAM J. Appl. Math. 35 (1978) 650.
[24] D. J. Benney, Stud. Appl. Math. 56 (1977) 81.
[25] M. Boiti, F. Pempinelli and G. Z. Tu, Prog. Theor. Phys. 69
(1983) 48.
[26] L. A. Takhtajan, Phys. Lett. 64A (1977) 235.
[27] A. V. Mikhailov and A. B. Shabat, Phys. Lett. 116A (1986) 191.
[28] R. Dodd and A. Fordy, Dublin Inst. preprint D/AS-STP 82-08(1982).
[29] R. M. Miura, ed., Backlund Transformations, Lecture Notes in
Mathemathics, No. 515 (Springer, Berlin, 1976).
[30] M. Wadati and K. Sogo, J. Phys. Soc. Japan 52 (1983) 394.
[31] N. N. Bogolyubov, A. K. Prikarpatskii, A. M. Kurbatov and V. G.
Samolienko, TMF 65 (1985) 271.

104
Prolongation Structure in One and Two Dimensions
A. Roy Chowdhury
High Energy Physics Division, Department of Physics,
Jadavpur University, Calcutta 700032, India

A computational basis for the prolongation theory applicable to soliton

equations is given with several examples included as illustrations.

1• INTRODUCTION

One of the most important developments in theoretical physics during

the past twenty years is the understanding of the complete integrability
property of a class of nonlinear partial differential equations(nlpde)[l].
The initial attempts for the systematic analysis of such equations
started with the famous papers of Gardner, Greene,Kruskal and Miura [2],
and also with that of Lax [3]. The observation of Lax, that these
integrable equations are representable as the consistency condition
of two linear equations was of utmost importance for the fast development
of the theory of solitons. However for a long time there was no concrete
way to arrive at a Lax pair for a given nlpde, though given the Lax
pair the problem can be analysed exhaustively. Then the now famous
work of AKNS (Ablowitz, Kaup Newell, Segur) [4] indicated a reverse
way of getting a class of integrable evolution equations starting
from a given Lax pair. The first detailed study of inverse scattering
was also taken up in this paper. One should of course mention the
pioneering work of Zakharov and Shabat [5] also. But the problem of
getting a Lax pair from a given nlpde was still not solved.

The basic problem was attacked and solved in the ingenious article
of Whalquist and Estabrook [6]. Later a simplified version was also
suggested by Corones [7], who showed how a pseudopotential can be asso-
ciated with a given nlpde. Later, different applications of the method
have been given by Gibbon [8], Pordy [9], Dodd [10], Roy Chowdhury
[11] and others [12].

2. RUDIMENTS OF THE THEORY OF DIFFERENTIAL FORMS

The whole formalism of Whalquist and Estabrook is based on the Cartan's

calculus of differential forms. Here we describe tersely the basic

105
rules for the operations on forms. Since we will be working mainly
in two spatial dimensions, we will use (x,y,t) as the set of independent
variables and u(x,y,t) as the dependent nonlinear field variable.

We start with the definition of differential forms:

(i) A differential dx, dt, du or an expression fdx, gdt, etc., is

called one form.
(ii) An expression of the form dXAdt, dUAdx, dyAdu is called two
forms, where A is the antisymmetric wedge product.
( iii) In fact one can also define the 1-form, 2-form, 3-form as the
integrands in respectively of a single, double or triple integral.
That is fdx of ffdx, gdxdy of ffgdxdy and hdxdydz of llfhdxdydz.
(iv) The symbol "d" stands for exterior derivative and acts in the
following manner: d(fdx) = dfAdx, d(gdxdt) = dgAdxAdt.
(v) It possesses the basic antisymmetric property: dXAdt = -dtAdx,
dXAdx = O.
i
(vi) In general for any coordinate representation, df f'i dx ,
elf
f'i = ~.
1
(vii) d(dw) 0, Poincare Lemma.
(mii) For any vector v, and forms wand 0

v. (wAo) (v.W)AO + (-1 )pwA(v.o).

(ix) For forms wand 0 WAO = (-l)pq OAW, where p rank of w,
q = rank of o.

3. nlpde AS THE SECTION OF A SET OF FORMS

The starting point of the WE formulation is to write the nonlinear

differential equation as a collection of differential forms. The process
of obtaining a particular equation from a set of forms is called sec-
tioning. In short it means that if one chooses a particular coordinate
representation then the "coordinate free" writing of the differential
form will reproduce the particular equation. In order to use a minimum
amount of abstraction we illustrate this with examples. Let us consider
the KdV equation

O. ( 1)

The first step is to define a set of primary sets of variables by

considering the derivatives of q, upto n-th degree (n being one less
than the highest degree of derivative of q occurring in the equation),
as independent set. In this case we consider,

106
 (2 )

So (u,p,q) form the primary set. To recast these in the differential

form language we set,

al = UdXAdt - dqAdt, a2 pdXAdt - dUAdt (3 )

whence the equation (1) is

o. (4 )

So we have
a 3 = dqAdx - qudxAdt - dpAdt (5 )

and (1) is equivalent to {a 1 , a 2 , a 3 } if we choose a coordinate such

that

dq (6 )

The above statement is quite easy to prove. From (6) and (3), we get,

a1 = udXAdt - (qxdx + qtdt)Adt = (u - qx)dxAdt. So a 1 = 0 implies

u = qx.

Consider a more complicated situation for an equation of the form

[ 121

o. (7)

In this case we set

o.
(8 )

These forms can be wr i t ten as

61 = duAdt - pdxAdt, 62 = dpAdt - rdxAdt,

(9)

The verification can be done following the same procedure as above.

4• CLOSED IDEAL

The next stage of the prolongation analysis makes use of an important

aspect of the differential forms, which is the closure of the set of
forms under exterior differentiation. In mathematical language we
write

107
 n
E ( 10)
j=l

As an example, let us consider equations (3), (4) and compute da i .

Using d(fdx) = dfAdx, d(dg) = 0, we get da 1 = dUAdxAdt = dx~ a 2 ,
1
da 2 = dpAdxAdt dX Aa 3 and da 3 ua1Adx + q a2Adx. So the fj's are
either dx or udx or qdx. On the other hand for equations (9) we obtain

0,

5. THE PROLONGATION ANALYSIS

After the closure property of the set of basis forms has been established
the final stage is set for the computation of the prolongation structure.
Here we actually search for a set of one forms wk '
( 13)

where Fk and Gk depend on the premitive set (u,p,q, ... ), independent

variables (x,t) and also on some new dependent variables Yk (which
are called the prolongation variables), in such a way that

m
E ( 14)
i

where n is the number of prolongation variables, m is the number of

basic defining forms a's and n~ is some set of one forms. Equation(14)is
actually an extension of the closure condition elaborated above. Ac-
tually they lead to an overdetermined set of partial differential
equations for F and G which are not always linear. Fortunately the
nonlinear part always has certain commutator-like structure and almost
always solvable. We will be using these features in the following
paragraphs by some explicit examples. Incidentally when Fk, Gk do
not depend on j ' s themselves, wk is called a potential. However if
F and G depend on the y's, they are called pseodopotentials. To illus-
trate the above ideas we consider again the case of KdV equation.

An important observation at this stage is that n i can be chosen

in various ways, on which depends the generality of the prolongation
forms wk' In the usual case we take

11·1 ( 15)

108
On the other hand, one can also set [13] n i = aidx + bidt + cidz +
didp + eidq. Equating to zero the coefficients of basic two forms
dXAdu, dUAdp, dpAdt, etc., in equation (14), with u 1 ' u 2 ' u 3 as in
(3), (4), (5) we arrive at

Fk 0; Fk 0; Fk + Gk 0 ( 16a)
,u ,p ,q ,p

aFk _ Fi aG k
uG k,q + p Gk,u - 12uG k,p + Gi o• ( 16b)
aYi aYi

In (16 ) we have used

d Wk d(dYk + Fkdx + Gkdt) = dFkAdx + dGkAdt

Equation (16) will determine the dependence of F and G on (u,p,q).

Differentiating repeatedly (16b) and utilising (16a), we get F 0,
qqq
Fp Fu = 0; Gqqqq = 0; Guuu = 0; Gpp =kO. Th~n it is not diffi-
cult to envisage the following structure of F and G ,

F k = Xk1 (y) + qX 2k (y) + q 2 X3k (y); Gk = -2(p+6q 2 )X .k

2 (y) +
23 k k k k k
3{u -9q -2qp)X 3 (y) + 8X 4 (y) + 8qX 5 (y) + 4q X6 (y)+4uX 7 (y) ,(18)
k
where Xi's are dependent only on the prolongation variables. If we
plugg in these Fk and Gk,s in equation (16b) and equate coefficients
of q2, q3, uq, pq, etc., we then get the following incomplete Lie
algebra

( 19)

As a next example, we consider the equation of a relativistic string

in a curved space-time. The governing equations are [14]

Xtt = e 2rp .
Sl.n 2 X, (20 )

where K is the curvature of the embedding space. We can also write

(2) as

109
 B (2' )

when a special value of K is chosen.

The basis variables are now defined to be ~2t = z, ~2x = p, ~'t q,

~'x = r. Then it is easy to observe that the basic two forms are

dqAdx + dpAdt - BdtAdx. (22)

The closure of the ideal generated by ri's can again be tested and we
search for w in the form

Following the procedure laid down previously we get

F o = F
r
P

[F,G] -F Z - F q + G p + G r - AF2 - BF q . (23 )

~2 ~, ~2 ~,

We choose Fang G as

(24)

Equation (23) then implies

(25 )

along with
[F 3 ,G 3 ] = -AX,(y) - BX 2 (y), (26)

~ , -~ 2 - ( ~ , +~ 2 ) ~2
G3 =e Y3(y) + e Y4 (y) + e Y5 (y) (27 )

110
which immediately leads to the incomplete Lie algebra

[X l 'Y3 1 X3 , [X 2 'Y3 1 -X 3 , [Xl' Y 11 = X 4 ;

[X 2 , Y4 1 X4 , [X 2 ' YS 1 0, [X l , X 3 1 Y3 ;

[X l , X 4 1 Y4 , [X 2 , X S 1 = 0, [X l , X S 1 = - Ys ;

[X 3 , Y3 1
1
2(X l -X 2 ), [X 4 , Y4 1 = 2(X l
1 + X 2 );

1 [X l , X 2 1 = 0, [X 3 , Y4 1 + [X 4 , Y3 1 0",
[X S ' YS 1 - 2 X l'

(28 )

Lastly let us consider a new integrable system, which is a generalisa-

tion of Liouville equation [15] written as

a+B-2q>
q>xt - e ,

2
a tt atq>t - (at) - eB-q>B x'

Bxx Bxq>x - (B ) 2 - ea-q> at"

x
(29)

If we set at = q, Bx = r, q>t s, q>x = p, the basic two forms become

(30 )
k k
In this case F , G are seen to be

(31 )

along with the Lie algebra

111
 [X"X s ] 0 , [X 3 ,X S ] = -2X 3 , [X, ,X 6 ] = -X 6 ;

[X 3 'X 6 ] 0, [X"X 7 ] = 2X 7 , [X 3 'X 7 ] 0;

[X, ,XS] -2X S ' [X 3 ,X S ] = X6 ' [X 2 ,X S ] = X2 ;

[X 4 ,XS] 2X 4 , [X 2 'X 6 ] = X,- XS ' [X, ,X 6 ] XS;

[X 2 'X 7 ] 0, [X 4 'X 7 ] -X 2 , [X 2 ,X S ] - X4 ' [X 4 ,X S ] o. (32 )

6. CLOSURE OF THE ALGEBRA

In all the examples cited above it is pertinent to observe that the Lie
algebra generated is never closed by itself. Of course there are some
examples where an exception can be seen ['6]. In the former case
there is no rule to attain this closure. However one may follow either
one of the following two broad strategies:

(,) First, to find a scaling, Galiean, Lorentz or conformal type

of symmetry of the original nlpde. Then impose these on wk' to obtain
on automorphism of F and G. This will have some nontrivial implications
regarding the unknown part of the Lie algebra. As an example, let
us consider the case of the extended Liouville equation (29). It remains
invariant under the transformation

x' ~ l;:x , t ' ~l;:

-, t, cp' ~ cp;

a'~ a - 3 logl;:, SI ~ S + 3 logl;:. (33 )

k' -, k k' k
Then we impose these on wk and get F ~ l;: F; G ~ l;:G , which lead
to the automorphism

X', X, , , X'2
2
l;: X2 ' X' l;:
-, X3 ' X' -+ l;:X 4 ,
~ ~
3 ~
4

Xs ~ XS ' X6 ~ l;:
-2
X6 ' X 7 ~ l;:X 7 , Xa -+ l;:
-, XS· (34 )

Now equation (32) does not give any information about [X 2 'X 4 ]. If
we study the transformation of [X 2 'X 4 ] under (34) we get [X 'X
3
-+ 2 4]
l;: [X 2 'X 4 ]· But there is no generator with such a scaling property in
(34). So its immediate implication is [X 2 'X 4 ] = O. On the other hand,
if we consider [X 3 ,X 4 ], under (34) we get [X 3'X 4] ~ [X 3 'X 4 ] and generators

112
with such scaling behaviour (with no factor of~), are Xl and XS. So
we infer

(3S)

and a and S may be fixed by Jacobi identity.

(2) The other route is the adhoc procedure adopted by WE who closed
the algebra by hand. The main algebraic reason for the effectiveness
of WE procedure was analysed by Shadwick [18]. Basically the procedure
of closing off is to obtain the various forms of embedding of SL(2,R),
SL(3,R) or SU(3), etc., in the incomplete Lie algebra. Such a procedure
is described in detail in references [9] and [10].

Once the Lie algebra is closed, then one can successfully obtain
the linearization, Lax pairs and Backlund transformations. For details
see [6-19].

7. EQUATIONS IN TWO SPACE VARIABLES

After the success of prolongation theory in (1+1) dimensions, extension

was made for (2+1) dimensional systems such as the K-P equation, Davey-
Stewartson equation by Morris [19], though the 1ST equations for these
were known a priori from the research of Dryuma [20] and Zakharov and
Shabat [21]. Here we indicate the procedure by considering a novel
application. We can summarize the basic rules for higher dimension
as follows:

(1) In n-dimension the nlpde's are equivalent to n-forms a i ,

A
a

(2) They satisfy ,closure as before daAC I I

(3) Then we introduce the prolongation variables y's and (n-l) forms

1 , 2 , 3 , ••• , dim yj, s~J are (n-2) forms to be

determined, s~
J

and wi is the connection one form defined as wi

113
(4) We again impose the closure for the extended set I'

An important and novel equation is the Benjamin-Ono equation [21],

f ux'x'
XI_X
dx' o (36 )

which can be written as a 3-d system following reference [22] as

u t + uU x + u xy = o. (37)

The basic three forms are [24]

drAdyAdt + updxAdyAdt + duAdxAdy.

(38 )
We then set

(39 )

F, G, H depend on (u,p,r,x,y,t'Yi) and

(40 )

0; O.

Then n j are written as

We then demand dn j =Ea.f~

1 1
+ (Adx + ~dy + Vdt)A n j which will then lead
to equations for L, M and N. Details of such a computation can be
found in refs. [25]-[27].

8. THE CONCEPT OF CONSTANT COEFFICIENTS IDEAL

A few years back an excellent method was suggested by K. Harrison [25]

for circumventing the difficulty of fixing the arbitrariness in the
form of F and G. He suggested that if it is possible to choose a special

114
class of two forms so that the nlpdes can be converted into an ideal
of differential forms with constant coefficients (C.C. Ideal) then
the process of obtaining the form of F and G can be streamlined and
one can arrive at the incomplete Lie algebra quickly. Let us refer
back to equation (22) and choose a new set of forms equivalent to the
nlpde's,

e: 1 = dt - dx, zdx + pdt, e: 3 = qdx + rdt,

(40a)
Then we can observe that

0, de: 2

de: 5 = - 2 [ e: 5 1\ e: 2 + e: 5 1\ e: 3 1, e: 4 1\ e: 5 = e: 4 1\ e: 6 = e: 5 1\ e: 6 = O. (41 )

It is interesting to note that these closure conditions do not involve

any variable coefficients so that (20) is equivalent to the C.C. ideal
generated by (41). The one form w is now written as

k (42 )
w

Bi being numerical matrices so that dw o leads to

o (43 )

which immediately implies

(44 )
The same procedure can also be illustrated with the basis of sine-Gordon
equation. It is written as, ~xt = sin~, Set r = ~x' rt = sin~.

115
The usual two forms are CY. = dcp 1\ dt - rdxl\dt and 13 = drl\dx - sincp dXl\dt.
But a set of C.C. ideals can be constructed by choosing the basis given
by

(45 )

k
is closed but with constant coefficients. We again write w
k i k k
-dy + (B ~i)Y' Then dw = a implies

a (46)

leading to a Lax pair for sine-Gordon equation.

9. USE OF PROLQNGATION STRUCTURE FOR

OBTAINING BACKLUND TRANSFORMATION

Lastly we only mention another important use of prolongation theory. One

can use the forms wk to deduce a B.T. [26] of the particular equation.
The basic principle is to assume that the new field variable depends
on the old one and also on the primitive variables along with the pro-
longation variables Yk' That is, u' = u' (cp, u, p, ... , x, t, Yk)'
For KdV, we write for the new set of forms as

dz'l\dt - p'dxl\dt, CY. 3 = -du'l\dx

and demanding that these be in the ring of the prolonged ideal, we
2
can obtain the BT, u' = -u - y + \. Detailed discussions can be
found in refs. [26]-[29] for other systems.

10, • CONCLUSION

In our above exposition we have tried to give a computational basis

for the prolongation theory that may be useful for a beginner. There
are many references for the geometric or differential geometric back-
ground for the prolongation structure, which we have not touched at
all. Finally, it is pleasure to thank the organisers, Prof. M. Lakshmanan
and Prof. P. K. Kaw for giving me this opportunity to deliver this talk
at the Winter School organized by SERC (DST, Government of India).

116
REFERENCES

[1] See for example, C. Rebbi and G. Soliani, Solitons and Particles
(World Scientific Press, Singapore, 1983).
[2] C. S. Gar~ner, J. M. Greene,M. D. Kruskal, R. M. Miura, Phys.
Rev. Lett. 19 (1967) 1095.
[3] P. D. Lax, Commn. Pure Appl. Math. 21 (1968) 467.
[4] M. J. Ablowitz, D. J. Kaup, A. C. Newell, and H. Seger, Stud. App.
Math. 53 (1974) 249.
[5] V. E. Zhakharov and A. B. Shabat, Sov. Phys. JETP, 34 (1972) 62.
[6] H. Whalquist and F. B. Estabrook, J. Math. Phys. 16 (1975) 1;
Phys. Rev. Lett. 37 (1973) 1386.
[7] J. Coronoes, Some Heuristic Comments on Soliton, Lecture Notes
in Physics, Vol. 810 (Springer, Berlin, 1979), p. 23.
[8] J. D. Gibbon, Lecture Notes in Maths. 775 (Springer, Berlin).
[9] R. K. Dodd and A. Fordy, Proc. Roy. Soc. A385 (1983) 389.
[10] R. K. Dodd and J. D. Gibbon, Proc. Roy. Soc. A359 (1978) 411.
[11] A. Roy Chowdhury and T. Roy, J. Math. Phys. 21 (1980) 189;
J. Phys. A12 (1979) 189; J. Math. Phys. 20 (1979) 1559.
[12] D. J. Kaup, Physica 1D (1980) 391; M. Lakshmanan, Physics Lett.
64A (1978) 354; J. Math. Phys. 20 (1979) 1667; App. Sci. Research
37 (1981) 127.
[13] A. Roy Chowdhury and S. Ahmad, Prog. Theo. Phys. 75 (1986) 1250.
[14] P. Molino, J. Math. Phys. 25 (1984) 2222.
[15] A. Roy Chowdhury and Shibani Sen, Prog. Theo. Phys. 76 (1986) 1.
[16] A. Roy Chowdhury and K. De Archan, J. Math. Phys.
[17] A. Roy Chowdhury and S. Ahmad, Phys. Rev. D32 (1985) 2780.
[18] W. F. Shadwick, J. Math. Phys. 21 (1980) 454.
[19] H. C. Morris, J. Math. Phys. 17 (1976) 1870.
[20] V. Dryuma, Sov. Phys. JETP 19 (1974) 387.
[21] V. E. Zakharov and A. B. Shabat, Func. Analysis App. 13 (1979)166.
[22] T. B. Benjamin and H. Ono, J. Fluid Mech. 25 (1966) 241; J. Phys.
Soc. Jpn. 39 (1975) 1082.
[23] B. Grammaticos, B. Dorizzi and A. Ramani, Phys. Rev. Lett.
53 (1984) 1.
[24] A. Roy Chowdhury and S. Ahmad, ICTP (Trieste, Italy), Preprint
No. IC/86/61.
[25] B. K. Harrison, J. Math. Phys. 24 (1983) 2178.
[26] R. K. Dodd and H. C. Morris, in Lecture Notes in Maths., Vol. 810
(Springer, Berlin), pp. 63, 95.
[27] A. Roy Chowdhury, Phys. Scr. 29 (1986) 289.
[28] C. Rogers and W. F. Shadwick, Backlund Transformations and Their
Applications (Academic Press, New York, 1982).
[29] H. C. Morris, J. Math. Phys. 18 (1977) 533; Int. J. Thecir. Phys.
16 (1977) 227; Chris Radford, J. Math. Phys. 27 (1986) 1266.

117
Integrable Equations in Multi-Dimensions (2+1)
are Bi-Hamiltonian Systems
A.S. Fokas 1 and P.M. Santini 1;2
1 Department of Mathematics and Computer Science and Institute for
Nonlinear Studies, Clarkson University, Potsdam, NY 13676, USA
2Permanent Address: Universita Degli Studi, Roma,
Istituto di Fisica "Guglielmo Marconi", Piazzale delle Scienze, 5,
1-1-00185 Roma, Italy

Recent developments by the authors in finding the recursion operators

and the bi-Hamiltonian formulation of a large class of nonlinear evo-
lution equations in (2+1)-dimensions is reviewed. The general theory
associated with factorizable recursion operators in multidimensions
is discussed. Both gradient and non-gradient master-symmetries are
simply derived and their general theory is developed, using the
Kadomtsev-Petviashvili equation as an example.

1. INTRODUCTION

Ablowitz, Kaup, Newell and Segur [1], following ideas of Lax [2] were
the first to solve in the concrete case of the Dirac problem the follow-
ing question: Given a linear eigenvalue problem find all nonlinear
equations that are related to it. They found that associated with a
given eigenvalue problem there exists a hierarchy of infinitely many
equations. This hierarchy is generated by a certain linear operator.
This operator is the squared eigenfunction operator of the underlying
linear eigenvalue problem. The operator generating the KdV hierarchy
(i.e., the squared eigenfunction operator of the SChrodinger eigenvalue
problem) was found by Lenard. For other eigenvalue problems see [3]-
[10].

Olver [11] established the group theoretical origin of the above

hierarchy: Finding the hierarchy associated with a given equation is
equivalent to finding the non-Lie point symmetries of the given equation.
He thus interpreted the squared eigenfunction operator as an operator
mapping symmetries onto symmetries; this lead to a simple mathematical
characterization of the recursion operator~. Olver was thus the first
to establish that certain integrable nonlinear equations possess infi-
nitely many symmetries. This motivates the following question: Is

118
there an algorithmic way for generating equations possessing infinitely
many symmetries? Fuchssteiner [12] discovered such a way: If an operator
~ has a certain mathematical property called hereditary then the equa-
tions u t = ~nux' n integer, possess infinitely many symmetries. From
the above discussion it follows that both linear eigenvalue problems
and hereditary operators yield hierarchies of equations possessing in-
finitely many symmetries. Actually Anderson and the author [13], follow-
ing ideas of Fuchssteiner, have shown that eigenvalue problems algorith-
mically imply hereditary operators.
Equations solvable by the Inverse Scattering Transform are Hamil-
tonian systems. Magri, in a pioneering paper [14], realized that integ-
rable Hamiltonian systems have additional structure: They are bi-Hamil-
tonian systems. Actually the underlying hereditary operator can be
factorized in terms of the two associated Hamiltonian operators. The
theory of factorizable hereditary operators has been further developed
by Fuchssteiner and the author [15] and by Gel'fand and Dorfman [16].
The understanding of the central role played by factorizable heredi-
tary operators for equations in 1+1, motivated a search for hereditary
operators for equations in 2+1. However, in this direction several
negative results have appeared in the literature. For example, Zakharov
and Konopelchenko [17], in an interesting paper proved that recursion
operators (of a certain type naturally motivated from the results in
1+1) did not exist in multidimensions. A similar result has been proved
for the Benjamin-Ono (BO) equation [18]. It should be noted that the
BO equation has more similarities [19] with the Kadomtsev-Petviashvili
(KP) equation than with the KdV equation. Fuchssteiner and the author
[18] after failing to find a recursion operator for the BO introduced
the concept of the master-symmetries T. Subsequently Oevel and
Fuchssteiner [20] found a master-symmetry for the KP equation. The T
theory for equations in 2+1 has been developed by Dorfman [21] and
Fuchssteiner [22]. However, the T is not related to the underlying
isospectral problem and also cannot be used to construct a second Hamil-
tonian operator. This is a serious drawback: several prominent investi-
gators, for example Gel'fand [23] have considered the existence of a
bi-Hamiltonian formulation as fundamental to integrability. Without
finding a recursion operator ~, one cannot find the second Hamiltonian
operator. Several investigators have noticed that master-symmetries
also exist for equations in 1+1. The theory for the master-symmetries
T in 1+1 was developed by Oevel [24] (see also [25]) and is more satis-
factory than the theory in 2+1: If one assumes that an equation is
invariant under scaling then there exists a one-to-one constructive
relationship between T and the recursion operator ~.

119
 Recently P. M. Santini and the author [26]-[28] have found the
recursion operator and the bi-Hamiltonian formulation of a large class
of equations in 2+1. They have also established the general theory
associated with factorizable recursion operators in multidimensions.
Furthermore, both gradient and non-gradient (the 2+1 analogue of T)
master-symmetries are simply derived and their general theory is deve-
loped.

2. MASTER SYMMETRIES

In this section we review certain aspects of non-gradient master-symme-

tries in 1+1 and gradient master-symmetries in 2+1.

Definition 2.1
A function , i s a master-symmetry of the equation qt K iff the map

a' [b] - b' [a] (2.1)

maps symmetries onto symmetries (prime denotes Frechet derivative).

The first example of a master-symmetry was given for the Benjamin-

Ono equation

(l-lf) (x) (2.2)

It was shown in [18] that if , ~ x(Hqxx + 2qqx) + q2 + ~ Hqx and on

is a symmetry then ~9+1 ~ [On"] is also a symmetry. It was further
shown in [18] that D , is a gradient function (,'D + D,'* = 0).

Master-symmetries are intimately related to time-dependent non-

Lie-point symmetries [25]. Indeed, the first non-Lie-point time-dependent
symmetry is a natural candidate for a master-symmetry: Consider the
'
evo l utLon ,
equatLon qt = K(l) an d l et K(2) , K(3) , . . . d enote L
't '
s tLme-
independent n6n-Lie-point symmetries. Let

, (2.3)

be a time-dependent non-Lie-point symmetry. Then

and , i s a candidate for a master-symmetry.

120
2.1 Master-symmetries for equations in 1+1

Lemma 2.1
Let

1 ,2. (2.4)

If ~ is hereditary, i.e., if ~I [~vlw - ~~I [vlw is symmetric with respect

to v, w, then

n
~m( E ~n-rS2~r-1)K1' (2.5)
r=l

m,n are non-negative integers.

Proof

See Theorem 2.1 of [281.

Corollary 2.1
Assume that LO is a scaling of both K and of the hereditary operator ~,
i.e. ,

aK , ~ I [ L 0 1 + [~, L 01 8~. (2.6)

Then

(i) (2.7)

i.e., ~LO is a master-symmetry for qt K.

(a + n8)~ n+m K n m
(ii) [~K, ~ LOlL' (2.8)

m
i.e., ~ LO is a master-symmetry of order m for qt = K.

n+1 n
(i iil (a + n8)U K + ~LO is a symmetry of qt = ~ K.

Proof

(i) Apply Theorem 2.1 with

8~.·

(ii) Similar to (i).

(iii) Use the definition of a symmetry.

In the above we derive T from ~. Now we obtain ~ from T.

121
Lemma 2.2

Let ~ be a hereditary operator such that ~8 =8~+, where 8 is a constant,

invertible, skew-symmetric operator. Then

(2.9)
where
S+

Proof
See [28].

Theorem 2.1

(i) If the hereditary operator ~ admits the scaling TO then~TO

is a master-symmetry for the hierarchy generated by ~.

(ii) Assume that the hereditary operator ~ admits the scaling TO

and that it also satisfies ~8 =8~+, where 8 is a constant, inver-
table, skew-symmetric operator which also admits the scaling TO.
Then

(2.10)

Proof

(i) If ~ admits a scaling and K is generated from ~ then K also admits

a scaling. Hence Corollary 2.1 implies (i) above.
(ii) Since ~ admits a scaling, ~+ also admits a scaling, hence S+ is
.
proport10nal to ",+
~ ,thus - - 1 1S
8S+8- . proport10na
. 1 to '"
~. Furt h er-
more, since 8
+ -1
0
admits the scaling TO' T 8 + 8TO+ = ~8, thus
TO + 8(T O) 8 equals a constant. Hence (2.9) implies (2.10).

EXAMPLES
1. ~ = 0 + q + qxo-1 is the hereditary operator associated with Burgers
equation. It admits the scaling q .... ~q, x ..... ~-lx, i.e., TO = q + xqx
2
Thus x(qxx + 2qqx) + q is a master-symmetry of Burgers equation.

2. ~ = 0 2 + 4q + 2qxo-1 admits the scaling q ..... ~, x ..... ~-2x, i.e.,

TO = q + 2xqx· Thus T = ~TO is a master-symmetry of the KdV.

3. If TO = q + 2xqx' then TO + O(T O)+o-l -3. Hence if T is the

master-symmetry of KdV,

is the recursion operator of the KdV.

122
2.2 Gradient Master-Symmetries for Equations in 2+1

A straightforward generalization of Theorem 2.1 to equations 2+1 fails:

(i) ~ could not be found; (ii) the known master-symmetries T were
gradient functions, hence T' + e(T' )+e- 1 = O. It will be shown in §3
that for equations in 2+1: (i) suitable generalizations of ~ , denoted
by ~12 can be found; (ii) there exist non-gradient master-symmetries
2
T12 (for example for the KP T12 ~12Q(Y1-Y2)' where 0 denotes the Dirac
delta function). Hence a generalization of Theorem 2.1 to equations
in 2+1 is given in §3.

One can still develop a theory for master-symmetries without using

the connection with the recursion operator~: see [21],[22].

3. SYMMETRIES FOR EQUATIONS IN 2+1

In this section we review the theory recently developed by Paolo Santini

and the author. We use the KP as an illustrative example and quote
the basic theorems when needed. We hope that this form of presentation
will aid the non-expert reader to become familiar with the notions and
methods developed in [26]-[28]. We advise the non-expert reader to
read [15] before reading this paper since many of the results presented
here are two dimensional generalizations of results given in [15].

3.1 Derivation of Recursion Operators

Given an isospectral eigenvalue problem there exists a simple algorith-

mic way of obtaining a recursion operator. This approach involves three
steps: compatibility, an integral representation of a certain differen-
tial operator, and an expansion in terms of delta functions. Let us
consider the eigenvalue equation

w + q(x,y)w + aW 0, a is a constant (3•1)

xx y

and for convenience of notation we suppress the t-dependence. Using

vector notation, (3.1) yields

w• [:1x
W
x
1

o JW, q
a
a-y' (3.2)

123
3.1a Compatibility

Associated with W UW we look for compatible flows Wt VW where

x

)
2C
, A, B, C, E polynomials in 0y.
E

Compatibility implies the operator equation

U = V - [U,V),
t x
or

CJ t
[C
Solving in the above equation for A, B, E in terms of C we obtain the
following operator equation:

qt = C + [q,C]+ + [q,C ]+ + [q,o-l[q,C]] + AOq - qA O' (3.3)

xxx x x

where
[ , ] is a commutator, [ , ]+ is an anticommutator, AO
is an operator such that

o and (0
-1
f)(x,y) = f x

""IX>
f(~,y)d~.

In what follows we take AO = 0 (the general case is considered in [27]).

3.1b An Integral Representation

The crucial step is to use an integral representation for the differen-

tial operator C:

(Cf) (x'Y1) (3.4)

Let

O. (3.5)
1

Equation (3.4) implies similar integral representations for all quanti-

ties appearing on the RHS of (3.3). For example

124
For

Thus

Similarly

+
where the operators q12 are defined by

(3.6)

Using the above integral representations in (3.3) we obtain

or

Let us introduce the operator ~12 via

(3.8)

Thus
(3.9)
D'I'12T12

3.1c Expansions in terms of delta functions

We expand T12 in .the form

(3.10)

It turns out that '1'12 admits a simple commutator relationship with

respect to h12 = h(Y1- Y2). Actually the following operator equation is
valid

4ah 12 , ( 3 • 11 )

125
Hence equation (3.7) yields

Thus
T(n) = 0, T(j-1) (3.12 )
12 12
x

Letting T~~) = we have the following proposition:

Proposition 3.1

The isospectral equation

0, q *q + aD;
y a constant (3.13 )

is associated with the equations

(3.14 )
where

~'2

(3.15 )

+
and '1'12' ~ 12 are related via D'I'12 ~12D. The operators q12 are defined
by

(3.16 )

(The notation qi is justified, since qi is indeed the adjoint of q1'

see §3.2)

EXAMPLE

1. Equation (3.14) with n o and 80 1/2 implies q1 = q1 .

t x
2. Equation (3.14) with n and 81 1/2 implies the KP equation

(3.17)

Remark 3.1

(i) The operators ~12 and '1"2 with Y2 = Y1 and a = 0 reduce to ~ and

~+ respectively, where ~ is the recursion operator of the KdV.

126
( i i) The starting symmetry (~'2D)., is given by q, +
x

Thus it reduces to q, •
x
the starting symmetry of the KdV, when Y2 = y,.

3.2 A New Directional Derivative and a New Bilinear Form

Recall that ~ generates symmetries and ~+ generates conserved covariants.

Similarly, it will turn out, that ~'2 and ~i2 generate extended symmet-
ries and extended conserved covariants respectively. To define these
extended notions we need to introduce a new bilinear form and a new
directional derivative:

(i) A new bilinear form

(3. '8)

where f'2 and g'2 are matrix valued functions of x, y" Y2 and obviously
the trace is dropped if f'2' g'2 are scalars. In association with the
above form we define Li2 to be the adjoint of L'2 iff

( 3 . '9)

We recall that the usual bilinear form and the usual adjoint are defined
by

(g,f) *f 2
R
dx dy trace gf, (g ,Lf) , (3.20 )

where f, g are matrix valued functions of x, y.

EXAMPLE

,. The adjoint of q, is given by qi

2. ( 3.2' )

3. ~* = 'l"2 •
'2

Note that the fastest way to compute the adjoint of an operator L'2
is to evaluate the adjoint as usually and then interchange , +-+ 2.

Let I be a functional given by

f
R
3 dx dy, dY 2 6 '2 trace P'2'
(3.22 )

127
The extended gradient of this functional is defined by

(3.23 )

where subscript d denotes a suitable directional derivative. It is

easily seen that a function Y12 is an extended gradient function, i.e.,
it has a potential I, iff

(3. 24)

Also

(grad I, .) * I f [.] (3.25 )

and Y is a gradient function iff Yf

(ii) A new directional derivative

Recall the crucial integral representation

Allowing f also to depend on Y2 we obtain

The above mapping between an operator and its kernel induces a mapping
between derivatives: Let subscript d denote the new directional deri-
vative. Then

The integral representation for q1 also induces, via (3.18) an integral

representation for the adjoint of q1:

g23,g3'1 f 12

=f 3 dY1dY2dx G21 f 12 , where we have used 3' +-+ 1, and

R

128
Thus

Furthermore, the qi mapping induces a mapping between derivatives.

Thus

(3.27 )

The above derivatives with respect to q1 and qi imply the following

+
derivatives with respect to q12' q12:

(3.28 )

Furthermore, using the chain rule and (3.28), if an operator K12 depends
only on q~2' q~2 its directional derivative L12 [° 1 2] is well defined.
This derivative is linear, and satisfies the Le~bnitz rule. Also, using
(3.28) it follows that the directional derivative in the direction of
012 reduces to the usual total Frechet derivative:

K12 [F] (3.29 )

f

where the subscript f stands for a Frechet derivative and

a
,,"e: K12(Q1,+F .. ,q,)\ ' 1 ,2, (3.30 )
o 11 J e:=0

+
Operators which depend only on Q'2 are called admissible. Similarly,
a function K12 is called admissible if it can be written in the form
K12 = R 12 H12 , where R12 is an admissible operator and H12 is an appro-
priate function (for the KP, H12 = H(Y1'Y2)).

EXAMPLE

The function A12 012 ~

since the operator A12

is easy to compute its directional derivative:

129
 2Da 1 2 .

3.3 Isospectral Problems Yield Hereditary Operators

Using the same methods as in 1+1, it can be shown that if the extended
gradient (G A )12 of the eigenvalue A of an isospectral problem satisfies

'¥ 1 2 (G A ) 12 ].l (A ) (G A ) 12' (3.31 )

then ~12 ~ '¥;2 is a hereditary operator. (One must again assume complete-
ness, a proof of which should follow a two dimensional version of the
method developed in [6].)

EXAMPLE

Consider the isospectral problem

V1 + (q1 - A)V 1 o. (3.32 )

xx

Taking the directional derivative of the above it follows that

Multiplying the above by V~, where V~ satisfies the adjoint of (3.32),

integrating with respect to dx dY1' and assuming fR 2 +

dx dY1 V 1V 1 =1,
we obtain

Using (3.26) to evaluate q1 [f 12 ] it follows that

d

( 3 • 33)

It is easy to show that ~;2 as defined by (3.7) satisfies

(3.34 )

Hence ~12 is a hereditary operator.

130
Remark 3.2

Konopelchenko and Dubrovsky [29) were the first to establish the impor-
tance of working with V(x'Y1)v+(x'Y2)' as opposed to V(x,y)v+(x,y).
They also found a linear equation satisfied by V 1V+
2 . However, they
failed to recognize that this equationcould actually yield the recursion
operator of the entire associated hierarchy of nonlinear equations.
Indeed, they used the above equation to obtain "local" recursion opera-
tors. Thus the question of studying the remarkably rich structure
of these recursion operators in particular its connection to symmetries,
conservation laws, and bi-Hamiltonian operators were not even posed.

3.4 Starting Symmetries

The theory of symmetries for equations in 1+1 is based on the existence

of "starting" symmetries KO, which via ¢ generate infinitely many
symmetries. For example, for the KdV KO = q For equations in 2+1
x
we find that the starting symmetries K~ 2 have the following important
properties: (i) Can be written in the form R~2H12' where R~2 is an
admissible operator and H12 is an appropriate function. (ii) The start-

°
ing operators R12 have simple commutator properties with respect to
h12 = h(Y1- Y2)· (iii) The Lie algebra of the starting operator R12
acting on functions H12 is closed. (iv) Using (ii) and the fact that
°
n-Q, °
¢12 also admits a simple gommutator relationship with h 12 , it can be
n ,,0 Q,
shown that 012¢12~12.1 =Q,:Ob n ,Q,¢12 R 12 · 012' where bn,Q, are appropriate
constants; hence 0 12 ¢72 R~2· 1 are admissible functions. It is thus
clear that in 1+1 one considers the Lie algebra of functions KO, while

°
in 2+1 one considers the Lie algebra of operators R 12 . This richer
algebraic structure of equations in 2+1 can be exploited in a variety
of ways. For example, different choices of H12 yield both time-indepen-
dent and time-dependent symmetries. Furthermore, all these symmetries
correspond to gradient functions.

We now discuss (i)-(iv) above for the concrete case of the KP: It
should be first noted that given an operator ¢12 there exists an algo-
;ithmic way of~finding its starti~g symmetries: One looks for operators
8 12 such that 5 12 H12 = °
but ¢12512H12 = K12H12 ~ 0.
~O
It can be shown
that if a starting symmetry is constructed in the above way and ¢12
is hereditary then ¢12 is a strong symmetry for this starting symmetry.

(i) For the KP there exist two starting symmetries:

(3.35 )

- -1
corresponding to 5 12 D and 5 12 = D(q12) D respectively.

131
 (ii) The following operator equations are valid:

o. (3.36 )

(iii) The Lie algebra of M12 , N12 is given by

~ ( 1) ~ (2)
[M 12 H12 ,M 12 H12 )d (3.37)
where

( 3.38 )

(3.39 )

Let us derive (3.37a):

Hence

Remark 3.3

The bracket (3.39) can also be traced back to the integral representation
of q1 (see [27)).

(iv) Equations (3.36) and the operator equation (see (3.11))

(3.40 )

(3.41 )

n 9 Q,
r b ~n-" M12 0 12 , bn,n
Q,=1 n,Q, 12 )(,

132
Let us indicate how the above equations can be derived: ~ntroducing an
operator ID, which commutes with all admissible operators K'2 and which
has the property that

it follows that

To derive equation (3.42) note that

n J!. n n J!. J!. n J!. n n-J!. J!.+'

E (-4a) ( )~ - M. 6'2 - 2a E (-4a) (n)~'2 D.6,2 .
J!.=O J!.' 2 J!.=O '"

(3.43 )
j j' A

The next step is to express ~'2D in terms of ~'2 M'2' where j, j'
are integers. This can be achieved as follows: It can be shown that
n+' n
~'2 D.' ~'2M.'. This equation implies

n
(2 N)j,,,n-j MA hj (3.44 )
E ~ ""2 , 2· , 2;
j=O

For example, multiplying ~'2D.' M'2.' by h'2' it follows that

(~'2-4a ID)h,2 D .' = (M'2-2a ID) .h, or ~'2D.h'2 = M'2.h'2. Similarly,

2 2
~'2D.' = ~'2M'2·' implies ~'2D.h'2 = ~'2M'2·h'2 + 2a M'2· h '2' etc.
Using (3.44) into (3.43) yields (3.42).

3.5 Basic Notions and Results

We consider exactly solvable evolution equations in the form qt = K(q),

q(x,y,t), on a normed space M of vector-valued functions on~; K is a sui-
table COO vector field on M. We assume that the space of smooth vector
fields on M is some space S of COO functions on the plane vanishing
rapidly as x, y .... ± 00. The above equation is a member of a hierarchy
generated by ~'2' hence more generally we shall study qt = K(n)(q).
Fundamental in our theory is to write these equations in the form

133
 K(n) (3.45 )
11 '

(in the matrix case, 1 is replaced by the identity matrix I), where
K;~)(q1,q2) belongs to a suitably extended space S, and S* denotes the
dual of S. In the extended spaces Sand S* we define the new directional
derivative (3.28) and the new bilinear form (3.18); the notions of the
adjoint and of a gradient are well defined with respect to (3.18) (see
(3.19), (3.23), (3.24)). In analogy with definition 2.1 we have:

Definition 3.1

(i) A function °12 £ S is called an extended symmetry of

(3.46 )

iff

o. (3.47 )

(ii) A function Y12£ S* is called an extended conserved gradient

(i.e., it is the extended gradient of a conserved functional I)
of (3.46) iff

0, (3.48 )

Functions which satisfy (3.48a) are called extended conserved

covariants.

(iii) An operator valued function ~12: S ~ s, is a recursion operator

for (3.46) (or it is a strong symmetry for K12 ) iff

(3.49 )

(iv) An operator valued function ~12: S ~ S, is a hereditary operator

(or Nijenhuis or regular) iff

( 3.50)

(v) An operator valued function 0'2: S* ~ S is a Hamiltonian operator

iff it is skew symmetric, i.e., °
12 = - 012' and it satisfies
(a'2,0,2 [012b,21c1) + cyclic permutation = O. (3.51)
d

134
 (vi) Equation (3.46) is a Hamiltonian system iff it can be written
in the form

(3.52 )

where 012 is a Hamiltonian operator and f12 is an extended

gradient function, i.e., f12 = f12 . Associated with (3.52)
d d
we define the following Poisson bracket

{I, H } (3.53 )

In the above, subscripts f and d denote total Frechet (see (3.29))

and directional (see (3.28)) derivatives respectively.

Remark 3.4

(i) Equation (3.47) can also be written as

Similarly, 4>12 [K]

f

(ii) Some of the above notions are well defined only if (o12K12)d
is well defined. However, for equations (3.45)

n
~
J!,=O
~o
Furthermore, by construction 4>12 and the starting operators K12
+ ~ (n)
depend on the basic operators q'2. Hence (012 K12 )d is well
defined.

In analogy with the basic results in 1+1:

Theorem 3.1

(i) If 4>12 is a recursion operator for (3.46) then 4>'2 maps extended
symmetries onto extended symmetries and 4>12 maps extended con-
served covariants onto extended conserved covariants.
(ii) If (3.46) is a Hamiltonian system then 0'2 = 0'2 Y'2·

( iii) I f 4>12 is a hereditary operator and a recursion operator for

~O

K'2.' then 4>'2 is a recursion operator for q1

t
= fR n ~O
dY2o'24>12K,Z'·

135
 (iv) If ~12 * e~~)(e~~))-1, where e~~) + ve~~) is a Hamiltonian
operator for all values of the constant v and e~~) is invertible,

then ~12 is hereditary.

(v) If ~12 as in (iv) and Y~2 * (e(1))-1~O
12 12·
1 is an extended gradient

function then all (~i2)mY~2 are extended gradient functions.

EXAMPLE

The hereditary operator ~12 of the KP equation is factorizable in terms

of the Hamiltonian cperators D and ~12D. Hence each member of the KP
hierarchy is a bi-Hamiltonian system, with respect to the following two
Poisson brackets

{I,H} i = 1,2

D,

3.6 Extended Symmetries

Lemma 3.1

(i) Let ~12 be hereditary, then

where

(3.55 )

m, n are non-negative integers.

(ii) o~~) is a time-dependent extended symmetry of order r of equation

(3.46) iff

o. (3.56 )

136
Proof

See [28].

We propose the following constructive approach to extended symmetries:

Given an isospectral problem construct a recursion operator ~'2. This
operator must be hereditarYA(seeA§3.3). Then construct its starting
symmetries operat~rs, say M'2' N'2. The operator ~'2 is a s~rong

symmetry of M'2' N'2 (see [27]). Compute the commutators of M'2~ON'2'

~'2 with h'2· Use the commutator relationships to derive o'2~'2K'2.'
n n-R,AO R,
R,:obn,R,~'~ K'2· o '2' K'2 is M'2 or N'2· Finally, compute the Lie
algebra of M'2' N'2. This Lie algebra together with (3.54)-(3.56) yield
infinitely m~ny time-independent and time-dependent extended symmetries.

EXAMPLE
,. m
~'2M'2."
A m
~'2N'2·' are extended symmetries of the KP hierarchy

q, = J dY2o'2~72 M'2·' (recall KP corresponds to n ,). M'2'

A t R
N'2 are defined in (3.35).

For

m+n-R,+' R,
~'2 N'2[O'2,H,2]d'
(3.57)
where we have used (3.54) (~'t is hereditary and it is also a strong
symmetry for M'2H'2' thus S~~ = 0), and (3.37c). Taking H'2 = ,

and using

n m A m
equation (3.57) implies [O'2~'2M'2·" ~'2M'2·']d = 0, i.e., ~~2~12·'
is an extended symmetry of the KP hierarchy. Similarly for ~'2N'2·"
since

2. ~~2M'2." ~~2N'2.' are extended symmetries of the hierarchy

137
3. The KP hierarchy admits two hierarchies of t-dependent symmetries
of order r given by (3.56) where

~(m) H(r) H (r) • ( )r

12 . 12 ' 12 T Yl + Y2

2j-l 2j-1 2s +1)

E (2j-l ) A(m+(2j-l )n+j-l - t=l 2s t +l) (r- t g l t
12 Ev(r,2j-l,s)M 12 .H 12

the summation E is over sl,s2, ... ,Sj' from zero to P n , P n (n-l )/2
if n is odd, (n-2)/2 if n is even,
and

~(m) H(r)
12 . 12 '

2j 2j
A(m+2jn+j -t~12st+l) (r-t~l 2s t +l)
Ev(r,2j,s)M 12 .H 12

E(2j-l)
12

wi th j > 1, b
n,t

v( r, j, s) * -J'-'-
(-2) j ( ne
.
j
ll= 1
(r - Ej 2s+1)~( n
t =1
j b
~ = 1 n , 2s t + 1
) r!
j
(r - E 2s +1)!
t=l t

For

Equation (3.56) implies that constructing a symmetry of order r is equi-

valent to finding a function E~~) with the property that its (r+l)st
commutator with 012K12 is zero. This can be easily achieved by using
suitable H12 's. For example, let H12 = Yl+ Y2' then (3.57) implies:

- ~ b m+n-t+l~ 20
t=O n,t ~12 12 1,t'

since

t
[012'Yl+ Y2 1r o if t '" 1 or 1 i f t 1.

138
 !I, m ~
Thus, using the fact that [012,01,!1,]1 = a it follows that ~12M12(Yl+Y2)
generates first order time-dependent symmetries

m
Similarly, to generate r-order time-dependent symmetries use ~12M12'

(Yl+ Y2)r, since the commutator of (Yl+ Y2)r with 0~2 produces (Yl+ Y2)r-!I,
and hence the r-th commutator of (Yl+ Y2)r with a~2 produces 1 which
commutes with 0;;):

where 0(r-s) denotes the Heaviside function with 0(0) = 1.

4. The hierarchy ql
t
= J,R dY 3 °
12 ~~ 2 N12' 1 admits two hierarchies of

t-dependent symmetries of order r given by (3.56) where

~(m) H(r)
12 . 12
j
(m+jn- !l,J1 2s!l, +1) (r
Ev (r,j,s)N 12 .H 12

and by

j
~(m+jn - !l,J:l 2s + 1)
Ev (r,j,s)M . !I,

where the summation E is over sl,s2, ... ,Sj from zero to P n ,

> 1, P n = (n-l )/2 if n is odd and (n-2)/2 if n is even. Also,

!I, n
b (-4a) (!I,)'
n,!I,

The above extended symmetries, under the reduction Y2 = Y 1 yield

sym~etries. This follows from the following theorem (see [27]).

Theorem 3.2

Assume that the admissible operators

139
where S, l3 are constants and 5 12 is such that 5 12 [.)H 12 o. Then
d

(i) If 012 is an extended symmetry of (3.45), 011 is a symmetry

of (3.45).

(ii) If 012 is an extended symmetry of (3.45) then 012 = o(ql,ql )=0

is an auto-Backlund transformation of (3.45), where ql and q2
are viewed as two different solutions of (3.45).

( iii) If Y12 is an extended conserved covariant of (3.45), Yll is

a conserved covariant of (3.45).

(iv) If Y 12 is an extended gradient function then Yll is a gradient

function.

EXAMPLE

Consider the extended symmetry of the KP

Clearly (M 12 .1 )11 2qlx which is a symmetry of the KP. Also, M12 ·1

= 0 is a well-known auto-Backlund transformation of the KP.

Remark 3.5

(i) It is quite interesting that both symmetries and Backlund trans-

formations of an equation in 2+1 come from the same basic entity,

the extended symmetry. Indeed, when a = 0 the recursion operator

~12 for the KP equation reduces to an operator that Calogero and

Oegasperis have introduced [30) and which generates the auto-
Backlund transformations of the KdV equation.

(ii) Using the interpretation that q±b ~ qb ± bq, q, b matrices, the

recursion operator of the KP becomes the operator generating auto-
Backlund transformations for the equations associated with the
N xN matrix 5chrodinger problem in one dimension (studied by
Calogero and Oegasperis [30)). This important connection is ex-
plained from the fact that certain 2+1 dimensional systems can
be viewed as reductions of certain evolution equations non-local
in y. These equations are directly connected to matrix evolution
equations [28), [31 ).

140
3.7 Extended Conserved Gradients

Lemma 3.2

Assume that 6 12 is a Hamiltonian operator. Then,

[612f12,612g12]d = 6 12 grad 12 < f12,612g12) +

612{<f12d- fhd)[612g121 - (g12 d - ghd)[612f121}. (3.58)

Proof

See [28].

One way of proving that ~i2 generates gradient functions is to use

Theorem 3.1, (v). However, this requires that ~i2 is factorizable
in terms of Hamiltonian operators. Alternatively, we propose the follow-
ing constructive approach, which only uses one Hamiltonia~ ope:ator:
Construct the Lie algebra of the starting operators, s~y M12 , N12 . Then
m -1 0
use this algebra and (3.58) t~ prove that all ~i2 612K12'AH12 ar~
"
gra dA1ents, prov1"d e d t h at 612K12
-1 0 H12 "1S a gra d"1ent~1~Oere
h 0 "1S M12
K12
or N12 . Finally, use Theorem 3.2 to show that (612K12H12)11 are gradients.

EXAMPLE

Consider the operators M12 , N12 associated with the KP. Then

(3.59)

(3.60 )

For
_l A

It is easy to verify tha: 1D ~12H12 is an extended gradient. Then (3.58)

and (3.37c) imply, that 0 ~12N12H12 is an extended gradient. Equation
(3.37b) implies

Since D-l~12N12H12' D-1M12H12 are exte~ded gradients, the above equation

with n = 0 and (~.58) imply that D-l~12M12H12 is an extended gradient.
Similarly D-l~~2M12 is an extended gradient. Equation (3.60) follows in
a similar manner using (3.37c).

141
Remark 3.6

It was shown in §3.6 that time-dependent symmetries of order r are genera-

ted via ~~2~12H12' ~~2~12H12 with H12 = (Y1+ Y2)r. The above results
show that D-1~~2~12H12' D-1~~2N12H12 are gradient functions for arbitrary
H12 . Hence the time-dependent symmetries correspond to gradient func-
tions. However, the time-dependent symmetries are closely related to
master-symmetr ies T (see §2). Hence the master-symmetries T correspond
to gradient functions.

3.8 Non-gradient Master-symmetries

Lemma 3.3

Assume that the hereditary operator ~12 satisfies ~12812 = 812~i2'

where 8 12 is a Hamiltonian operator (if ~12 is factorizable, then this
equation follows). Assume for simplicity that 8 12 O. Then
d

(3.61 )

where

(3.62 )

Proof

See [28].

The results of Lemmas 3.1-3.3 can be used to obtain non-gradient

master-symmetries T 12 . Such master-symmetries are explicitly related
to recursion operators ~12. Indeed given ~12 one computes T12 and
given T12 one .computes ~12. These formulae are the two dimensional
analogues of the formulae given in §2.1. The basic idea is to find
-1
a T12 such that T12 + 8 12 T i2 8 12 = 0 and Si2 = C.1, C constant.
d d

EXAMPLE
2
A master-symmetry of the KP hierarchy is given by ~12012. Indeed

B constant (3.63 )

142
 (3.64)

(3.65 )

For

~12 and M12 are given by (3.8), (3.35a), 8 12 = D. Let T12 = 8 12 .

-1
Then T12 + 8 12 T i2 8 12 o and Si2 = 4 (see (3.62». Thus, equation
(3.61) w~th m = 2 ~mplies (3.63) •
(1) A (2)
To derive equation (3.64) use Lemma 3.1 with K12 M!2· 1 , K12
8 12 , m = 2, sl~) = 0 (since ~12 is a strong symmetry of M12 .1) and
sl~) ~12f[8121 = 4. Thus

n+2 1 n+1 A

However, [M 12 ·1, 8 12 d is proportional to M12 ·1, hence the

A
~12 ~12

above implies (3.64).

To derive equation (3.65), use Lemmas 3.1-3.3 to obtain the follow-

ing general result: If~2 is a hereditary operator such that it is
a stong symmetry for M12 and ~12812 = 812~i2' where 8 12 is a constant
invertible Hamiltonian operator then

where

Using T12 8 12 in the abc"e and noting that S 12 = Si 2

812Ti2 8~~ 0, ~~;m[M12' 8 12 1d is proportional to

d
we obtain (3.65).

ACKNOWLEDGEMENTS

This work was partially supported by the Office of Naval Research under
Grant Number N00014-76-C-0867 and the National Science Foundation under
Grant Number MCS-8202117. One of the authors (A.S.F.) would like to
thank Professor Lakshmanan and his colleagues for their kind hospitality.

143
REFERENCES

[1] M. J. Ablowitz, D. J. Kaup, A. C. Newell and H. Segur, Stud. Appl.

Math. 53 (1974) 249; Phys. Rev. Lett. 30 (198~ 1262; Phys. Rev.
Lett. 31 (1973) 125.
[2] P. D. Lax, Comm. Pure Appl. Math. 21 (1968) 467.
[3] W. Symes, J. Math. Phys. 20 (1979) 721.
[4] A. C. Newell, Proc. R. Soc. London Ser. A365 (1979) 283.
[5] H. Flaschka and A. C. Newell, Lecture Notes in Physics, Vol. 38
(Springer, Berlin, 1975), p. 355.
[6] V. S. Gerdjikov, Lett. Math. Phys. 6 (1982) 315.
[7] M. Boiti, F. Pempinelli, G. Z. Tu, 11 Nuovo Cimento 79B (1984)231.
[8] M. Leo, R. Leo, G. Soliani, L. Solombrino, Lett. Al Nuovo Cimento
38 (1983) 45; M. Boiti, P. J. Caudrey, F. Pempinelli, 11 Nuovo
Cimento 83B (1984) 71.
[9[ F. W. Nijhoff, J. Van der Linden, G. R. Quispel, H. Capel, and
J. Velthnizen, Physica 116A (1983) 1.
[10] B. G. Konopelchenko, Phy~. Lett. 75A (1980) 447; 79A (1980) 39;
95B (1980) 83; 100B (1981) 254; 108B (1987) 26.
[11] P. J. Olver and P. Rosenau, On the 'Non-Classical' Method for
Group-Invariant Solutions of Differential Equations, Univ. of
Minnesota Math. Re. No. 85-125, 1985.
[12] B. Fuchssteiner, Nonlinear Anal. TMA 3 (1979) 849.
[13] A. S. Fokas and R. L. Anderson, J. Math. Phys. 23 (1982) 1066.
[14] F. Magri, J. Math. Phys. 19 (1979) 1156; Nonlinear Evolution
Equations and Dynamical Systems, ed.M. Boiti, F. Pempinelli and
G. Soliani, Lecture Notes in Physics, Vol. 120 (Springer,
New Yor~, 1980), p. 233.
[15] A. S. Fokas and B. Fuchssteiner, Lett. Nuovo Cimento 28 (1980) 299;
B. Fuchssteiner and A. S. Fokas, Physica 4D (1981) 47.
[16] I. M. Gel'fand and I. Ya. Dorfman, Funct. Anal. Appl. 13(1979)13;
14 (1980) 71.
[17] V. E. Zakharov and B. G. Konopelchenko, Comm. Math. Phys. 94
(1984) 483.
[18] A. S. Fokas and B. Fuchssteiner, Phys. Lett. A86 (1981) 341.
[19] A. S. Fokas and M. J. Ablowitz, Stud. Appl. Math. 68 (1983) 1.
[20] W. Oevel and B. Fuchssteiner, Phys. Lett. A88 (1982) 323.
[21] I. Y. Dorfman, Deformation of Bamiltonian Structures and Integrable
Systems (preprint).
[22] B. Fuchssteiner, Prog . Theoret. Phy. 70 (1983) 150; I. Ya Dorfman,
Deformations of Hamiltonian Structures and Integrable Systems
(preprints).
[23] I. Gel'fand (private communication).
[24] W. Oevel, A . Geometrical Approach to Integrable Systems Admitting
Scaling Symmetries (preprint, 1986).
[25] H. H. Chen, Y. C. Lee, J. E. Lin, Physica 9D (1983) 439; Phys. Lett.
91A (1982) 381.
[26] A. S. Fokas and P. M. Santini, Stud. Appl. Math. 75 (1986) 179.
[27] P. M. S~ntini and A. S. Fokas, Recursion Operators and Bi-Hamil-
tonian Structures in Multidimensions I , INS #65, preprint 1986.
[28] A. S. Fokas and P. M. Santini, Recursion Operators and Bi-Hamil-
tonian Structures in Multidimensions II, INS #67, preprint 1986.
f?Ol B. G. Konopelchenko and V. G. Dubrovsky, Physica 16D (1985) 79.
[30] F. Calogero and A. Degasperis, Nuovo Cimento 39B (1977) 1.
[31] P. Caudrey, Discrete and Periodic Spectral Transforms Related
to the Kadomtsev-Petviashvili Equation, preprint UMIST (1985).

144
Painleve Analysis and Integrability Aspects
of Nonlinear Evolution Equations
M. Lakshmanan 1 and KM. Tamizhmani 2
1 Department of Physics, Bharathidasan University,
Tiruchirapalli 620024, India
2Department of Mathematics, Pondicherry University,
Pondicherry 605001, India

A brief review of the singularity structure aspects of the solutions

of nonlinear ordinary differential equations and their generalization
to partial differential equations leading to the Painleve (P) property
is given. It is pointed out that the Painleve analysis leads naturally
to Lax pairs, B~cklund transformations, linearizations and Hirota's
bilinearization of nonlinear evolution equations. Specifically we
treat the Burgers', Liouville, Korteweg-de Vries, coupled nonlinear
Schrodinger and Kadomtsev-Petviashvili equations as examples.

1• INTRODUCTION

During the past two decades or so, there has been considerable progress
in the understanding of classes of nonlinear evolution equations leading
to many fascinating new concepts such as solitons, Backlund transfor-
mations, generalized symmetries, etc. [1-3] (as exemplified by the
various articles in this book). However, in this process an important
question arises as to how to search, identify, characterize and classify
the integrable nonlinear equations systematically and then understand
the solution characteristics. In the last century, this question was
analysed for ordinary differential equations by various authors [4-6]
through the singularity structure analysis of the solutions in the
complex plane. With the recent developments in soliton equations this
analysis has again received much attention [3,7-19], and now many of
the integrable dynamical systems are associated with the so-called
Painleve property, in that they are free from movable critical points/
manifolds. The development of the singularity structure analysis can
in fact be traced down to the following four main aspects:

1. The classification of first order and second order nonlinear ordinary

differential equations (odes) which are free from movable critical

145
 points achieved through the works of Fuchs, Painleve and his co-
workers [5] in the last century.
2. S. Kovalevskaya's (1886) investigation for finding the integrable cases
of rigid body motion around a fixed point under the influence of
gravity [6] through the singularity structure analysis.
3. Ablowitz, Ramani and Segur's (ARS) [3] conjecture in 1978 that
every ode obtained by an exact reduction of a soliton system which
is solvable by the inverse scattering method is of Painleve (P)
type (see below), a fact verified from the invariance point of
view by Lakshmanan and Kaliappan [7].
4. Weiss, Tabor, and Carnevale's (WTC) [8] generalized version of
the Painleve property directly applicable to partial differential
equations (pdes). Here the solutions of the pdes are required-
to be single-valued around movable singular manifolds in order
that they be integrable.

In Sec. 2, we begin with a brief discussion of odes and their

singularities, and the application of Painleve analysis to them. We
review the generalized WTC procedure for pdes and apply it to a class
of nonlinear evolution equations in (1+1) dimensions in Sec. 3 and
obtain the basic solitonic properties. Furthermore, in Sec. 4, we
extend the application of P-analysis to the (2+1) dimensional Kadomtsev-
Petviashvili (K-P) equation and derive the associated Backlund trans-
formation and Lax-pair straight forwardly through the P-analysis.

2. PAINLEVE PROPERTY AND ORDINARY DIFFERENTIAL

EQUATIONS: INTEGRABILITY

The singularities of an ode can be classified as (i) fixed and (ii)

movable [4,5]. While the location of the former is fixed by the nature
of the coefficients of the ode, the latter is a function of the integ-
ration constant or initial condition.

dw -1
Consider for example the linear first order ode dz + z w O.
It has the solution, w = C exp(1/z), C, arbitrary. So z = 0 is the
fixed (essential) singular point. More generally, for an n-th order
linear ode

n n-1
d w + P 1 (z) ~ + . . . + Pn(z)w 0, (1 )
dz n dz n - 1

where Pi(z), i = 1,2, ... ,n, are all analytic at z admits n

146
linearly independent solutions in the neighbourhood of zo' so that
the general solution may be written as

n
w(z) r ciwi(z), (2 )
i=1

where ci's are integration constants. Here the singularities of the

solution must be located at the singularities of the coefficients Pi(z)
which are all fixed and apriori known and that they do not depend
on the constants of integration c i ' i = 1,2, ... ,n, at all.

However, in the case of nonlinear odes the singularities have

the additional property that they can depend on the integration cons-
tants and so, they can 'move'. Examples: ~~ + w2 = 0 has the solution
w = (z - z )-1, where z is a constant of integration.
o
Thus, at z = z
0'
w has a singularity, a pole of order one; it is movable because its
location depends on z . Similarly, ~w + w3 = 0, w = ~2(z - z )-1/2;
dw 2 0 1 dw z 1i' L. 0 1
dz + w log w = 0, w = exp(z=z-); dz + w exp(w) = 0, w = log (z=z-)'
which admit movable algebraic 8ranch point, essential singularityOand
logarithmic singularity respectively. In fact, Fuchs [4] had shown
that the only first order equation which admits no movable critical
points is the generalized Riccati equation

dw (3 )
dz

Following the work of Fuchs for first order odes, Painleve, Gambier,
Garnier and others classified 50 canonical nonlinear second order
odes which do not exhibit movable critical singularities and which
can be integrated in terms of elementary functions, including elliptic
functions, and Painleve transcendentals [5]. Also the nature of entire
bounded functions and the constancy of integrals of motion, coupled
with the works of Euler (1780), Lagrange (1788) and Fuchs (1884) promp-
ted S. Kovalevskaya [6] to treat the dynamical problem of the motion
of a spinning top fixed to a point in terms of meromorphic functions.
She concluded that only for three parametric choices (including those
found by Euler and Lagrange earlier) the system is free from movable
critical points and hence integrable, a result which even today stands
undisputed.

The classification of integrable dynamical systems through the

singularity structure analysis was not utilized much for more than
eight decades or so. The ARS conjecture [3] on the connection between
the integrable soliton systems and the Painleve equations (noted in
Sec. i) revived the interest again in the singularity structure analy-

147
sis for dynamical systems recently. Such a revival has also led to
the concept of strong and weak Painleve properties [18] for dynamical
systems, which are then associated with the integrability aspect.
Here the strong P-property stands for solutions which are meromorphic
around a movable singular point while the weak P-property [18] allows
for a relaxation of the meromorphicity condition on the solution so
that under necessary circumstances determined solely by the nonlinearity
of the equation movable algebraic branch points are also allowed for
algebraic integrals of motion [17] to exist. Some of the typical
integrable dynamical systems which possess the P-property are (i)
Henon-Heiles system [18], (ii) coupled polynomial nonlinear oscillators
[19], and (iii) Toda-lattice, for specific parametric choices.

An n-th order ode of the form

d
F ( z,w, d~' ... ,
dn- 1
n-1
W) (4 )
dz

n-1
where F is rational in (:zn- ~ , ... , w) and locally analytic in z,

can be analysed algorithmically for its singularity structure as follows.

Let us look for a Laurent series solution of (4) in the neighbourhood
of a movable singular point in the form

q.
w (z - z ) J ""I: a. (z - z )m j 1,2, .••
o m=O J,m 0'
(5 )

and determine the allowed values of q., a. and the powers at which
J J,m
(n-1) arbitrary constants enter into the series (5). In order to
avoid inconsistencies, it may be necessary to introduce further logarith-
mic terms in the above series. Finally, after verifying the existence
of the above solution, one classifies the conditions under which the
solution is free from movable critical singularities to within trans-
formations, which are then the possible cases for integrability. The
latter may be ,proved often by finding the appropriate integrals of
motion by other methods.

,
3. THE PAINLEVE ANALYSIS FOR PARTIAL
DIFFERENTIAL EQUATIONS: INTEGRABILITY

The natural extension of the P-property discussed iO Sec. 2 to pdes was

suggested by Weiss, Tabor and Carnevale (WTC) [8], who required that
the solutions be single-valued around movable singularity manifolds.

148
The major difference between the P-analysis of odes and pdes is that
now the singularities of the latter are in general not isolated, as
the solutions are functions of several complex variables (zl,z2, ... ,zn)'
but rather lie on manifolds determined by the condition

o. (6 )

Thus if u = u(zl,z2, ... ,zn) is a solution of the pde

u t + K(u) 0, (7 )

then we require that in the neighbourhood of the manifold

u ,na.
T
~ j
uJ'cp , (8 )
j=O
where u
o *
0, u j = u j (zl,z2, ... ,zn) and cp = CP(zl,z2' ... , zn) are
analytic functions of (z.) in a neighbourhood of the manifold (6)
J
and that a. is a negative integer. By Cauchy-Kovalevskaya's theorem
such an expansion of the general solution must have sufficient number
of arbitrary functions equal to that of the order of the pde. Impli-
mentation of this procedure is direct and follows algorithmically in
a manner similar to that of the odes.

There are essentially four steps involved in the P-analysis of pdes:

(i) Determination of the leading-order behaviours; (ii) Identification
of the powers at which arbitrary functions can enter into the Laurent
series called resonances; (iii) verifying that at the resonance values
sufficient number of arbitrary functions exist without the introduction
of movable critical manifolds; (iv) Establishing connections with
the soliton and other integrability properties. The remarkable feature
of the P-analysis, particularly for soliton equations, is that a natural
connection exists between the P-property and the linearization property,
Lax pairs, Backlund transformations, integrability, etc. [8-151. In
the following subsections we will explain each of the stages succinctly.

3.1 Leading order analysis

The analysis starts with the determination of the all possible value(s)
of a. and Uo in the expansion (8), and under what conditions a. is a
negative integer so that no movable critical manifolds enter at this
stage. For each value of a., the homogeneous terms with the highest
degree may balance each other. These terms are called leading terms
(or dominant terms). The value of U o can be determined by equating

149
the coefficients of the dominant terms to zero and solving the resul-
ting algebraic equation for u o .

3.2 Resonance analysis

Next, one has to find the "resonance" values, j, that is the power(s)
at which the coefficient(s) u. of the term ~j+a in the expansion (8)
J
is arbitrary. To find these, we substitute (8) into Eq. (7), and
obtain appropriate recursion relations for u. and extract the coefficient
j+a-N J
Q(j'u. of the term ~ , where N is the order of the pde. Then
J
Q(j) o is called the resonance equation, for which -1 is always
a root, which corresponds to the arbitrary nature of~. In order
to avaoid any movable critical singular manifolds, we require that these
remaining roots are non-negative integers.

3.3 Arbitrary functions

Let js be the highest of the allowed resonance values. Then we subs-

titue
js j+a
u E u J' ~ (9 )
j=O

into Eq. (7) and collect the coefficients of ~j+a-N to obtain

Q(j)u. + R. 0, ( 10)
J J

where R j is a polynomial in the partial derivatives of ~ and uk's

(k = O,l, ... ,j-l). Since Q(j) = 0, for any resonance value j, R.
J
should identically vanish. In this case u. is arbitrary. In case
J
if it is not so, we have to introduce logarithmic terms of the form
a. + b. log~ in the series. But due to this addition, logarithmic
J J
singularities will appear in the solution manifold. Thus, R. = 0 is
J
the condition to ensure that the solution is free from movable critical
manifolds at a particular resonance value j. In this way, we can check
that the general solution is free from movable critical manifolds.

3.4 Backlund transformation (BT), Lax pairs, etc.

The Backlund transformation of Eq. (7) can be obtained by truncating

the expansion (8) at the 'constant' level term, by setting

-n -n+l ( 11 )
u Uo ~ + u1 ~ + .

Then one can find an over-determined system of equations for ~ and u.,
J
j = O,l, ... ,n, where un will satisfy Eq. (7). Upon solving the over-

150
determined system, the other soliton properties of Eq. (7) such as
linearization, Lax pairs, etc., can also be obtained, in general [8-14].

4. EXAMPLES

In this section, we briefly illustrate the theory discussed above with

some typical examples [8-14].

4.1 Burgers' equation

The Burgers' equation is of the form [2]

au ( 12)
xx

Using Eq. (8) in Eq. (12), we can easily find from the leading order
analysis that

a. ( 13)

The recursion relations for u j are found to be

j
u]'-2,t+ (j-2)u]'_1 q>t+ E u, [u 1 + (m-l)q>u]
m=O ]-m m- ,x x m

a[u, 2 + 2(j-2)u, 1 m + (j-2)u, 1m + (]'-1)(]'-2)u,m 2 ].

]- ,xx ]- ,x "'x ]- "'xx ] "'x
( 14 )

The resonance analysis shows that the resonances occur at j -1,2.

In fact, j = -1 corresponds to the arbitrariness of the manifold
q>(x,t) = O.

From Eq. (14), we find that

0 u -2aq>x' ( lSa)
0

j q>t + u 1 q>x aq>xx' ( lSb)

a
j 2 ilx(q>t + u l q>x - a q>xx) = O. ( lSc)

By (lSb), (lSc) is identically satisfied and hence u 2 is arbitrary.

The coefficients u" j > 2 can then be obtained uniquely in terms of
]
uO' u l ' and u 2 . Therefore, the general solution of (12) contains
the required two arbitrary functions and that (12) possesses the
Painleve property [8] in the sense of WTC.

151
 If now one sets the arbitrary function u 2 0, then all u. 0,
J
j > 2, provided

oU lxx ' ( 16)

which is just the Burgers' equation. Thus, we obtain the Backlund

transformation for Burgers' equation,

u ( 17)

where both u and u l satisfy the Burgers' equation and ~ obeys (lSb).
When we consider the vacuum solution u l = 0 in Eq. (17), the well-
known Cole-Hopf transformation results, which is the linearizing trans-
formation for the Burgers' equation.

4.2 Liouville equation

We next consider the Liouville equation [11] in the form

( 18)

Under the transformation

u log V, ( 19)

(18) becomes

O. (20 )

On expanding

-2 j (21 )
V ~ l: V.~ ,
j=O J

we find that

V (22a)
o

and

O. (22b)

152
After making use of Eqs. (22a) in·, (22b), we observe that V2 is arbit-
rary and thus Eq. (18) passes the Painleve test [11]. Furthermore,
a BT is obtained with

v (23 )

provided V2 satisfies Eq. (20) and the linearized wave equation is

true:
o. (24 )

The solution of (24) is given by

h(x) + g(t), (25 )

where h(x) and g(t) are arbitrary functions. Consequently, we obtain

the general solution

u (26 )

4.3 Korteweg-de Vries equation (K-dV)

The K-dV equation

o (27)

is easily shown to possess the Painleve property. One can find the
leading order and the resonance values as ~ = -2 and j = -1,4,6 res-
pectively. From the recursion relations it is easy to check that

j o -12C1q>~ (28a)

j u
1
= ( 28b)

j 2 o (28c)

j 3 (28d)

j 4 o (28e)

j 5

153
 Now it is clear that by the condition (28d), the determining Eq. (28e)
is always satisfied. Similarly, we can verify that at j = 6 the com-
patibility condition is satisfied identically and so, u 6 is arbitrary.
Thus, we conclude that the K-dV equation possesses the Painleve property
(8 ).

If the arbitrary functions u 4 and u 6 are chosen identically to

be zero, and if we require that u 3 = 0, then we can verify that u j 0,
j ~ 3, provided u 2 satisfies the K-dV equation. Thus, we get the BT
for the K-dV equation in the form

u (29 )

with

0, (30a)

o , (30b)

and

O. (30c)

Eliminating u 2 from (30a) and (30b), we obtain after an integration

<P ...
----'- + 0 {<pi x} = A, ( 3' )
<Px

where

{<p i x} (32 )

is the 'Schwarzian derivative' of q> and A is a constant parameter.

With the substitution <p v,/v 2 , Eq. (3') can be rewritten in the form

e-+0
'¥ t
x
[
{<pi
2v 2xx
x} - --v--- +
2
A, (33 )

where ex =
v 2 v'x - v,v 2x ' '¥t = v 2 V,t - v,v 2t · If (v"v 2 ) satisfy a
linear system (say)

v av (34 )
xx

v
t
= bv
x
+ cv (35 )

154
then we can check that e 0 and 'I't = be Using these values
lfx x
in Eq. (33 ) we get a = - 6(u 2 + A) , b u 2 /3 +(2/3lA and c = u 2x /6.
With these values, Eqs. (34 ) and (35 ) are just the Lax pair of the
K-dV equation.

4.4 Coupled nonlinear Schrodinger equations

Our next example is a system of two coupled nonlinear Schrodinger (NLS)

equations defined by

2 2
c1X1xx + 2alx11 Xl + 2Slx21 Xl' (36a)

c 2 X2xx + 2ylX 2 12 X2 + 2Slxl12 X2 · (36b)

The system (36) is known to be integrable [16] for the following

specific parametric restrictions

a Y ( 37)

a = -S = Y (38 )

Now we will show that for exactly the same parametric choices, (37)
and (38), the system (36) passes the Painleve test [12].

We rewrite the Eqs. (36 ) as

-Qt c 1 P xx + 2a(p2 + Q2)p + 2S(R 2 + S2)p (39a)

2 + S2)Q
Pt c 1Qxx + 2a(p2 + Q2)Q + 2S(R (39b)

-St C 2 Rxx + 2S(p2 + Q2)R + 2y(R 2 + S2)R (39c)

Rt = c 2 S xx + 2S(p2 + Q2)S + 2y(R 2 + S2)S, (39d)

where Xl = P + iQ and X2 = R + is, P, Q, R, and S are reals. By

expanding P, Q, R and S about the singularity manifold cp as
00
-1 -1
P ep ~ P . ep j , Q cp ~ Q j cpj , (40a)
j=O J j=O

00 00
-1 j -1 S . epj (40b)
R ep ~ Rjep , S ep ~
j=O j=O J

one can easily find the following two possibilities:

155
Case 1 (41 )

and the associated resonances are

j = -1, 0, 0, 1, 2, 3, 3, 4. (42 )

Case 2 o (43 )

and
j - 1, 0, 0, 0 ,3, 3, 3, 4. (44)

We can verify that for case 1 the associated series solution will
have a lesser number of arbitrary functions only since none of the
functions P l , Ql' Rl and 51 is arbitrary and so it does not correspond
to the general solution. For case 2, the resonance values (44) require
that three of the four functions Po' QO' RO and So are arbitrary.
From the leading order analysis we have

2 2
a(P o + Q2) + 6(R~ + 52) - c l q>x' (45 )
0 0
(twice)

6(p2 + Q2) + Y(R~ + 52) 2

- c 2 q>x (46)
0 0 0
(twice)

and hence two, say Po and QO' of the four functions (PO,QO,RO'SO)
are arbitrary. Also, from Eq. (43) we can show that RO (or SO) is
arbitrary only for the conditions (37) and (38). Proceeding further
we can establish the required arbitrariness at the other resonance
values. Thus, Eqs. (36) possess the Painleve property for the para-
metric choices of Zakharov and Schulman only.

We will discuss the connection between the corresponding BT and

soliton solutions in Sec. 5. We remark that this analysis can be
extended straightforwardly to the case of a set of N coupled nonlinear
Schrodinger equations [12) as well.

4.5 Kadomtsev-Petviashvili equation

The Kadomtsev-Petviashvili (KP) equation is a two-dimensional genera-

lization of the well-known Korteweg-de Vries equation and describes
unidirectional weak (quadratic) nonlinear disturbances perturbed by
weak transverse balancing fourth order dispersion. The system admits
infinitely many non-trivial symmetries and constants of motion and

156
a Lax pair exists. In this section, we will demonstrate the existence
of the Painleve property for the KP equation and obtain its Lax pair
therefrom.

The KP equation is of the form [10]

1-(u + 6uu + u ) + 30 2 u ± 1. (47)

ax t x xxx yy

Now, the Laurent series solution takes the form u = ~ u,~j+a

'-0 J
around the singularity manifold ~(x,y,t) = 0, whereJu= -2 and U o =
-2~~. Substituting the above series solution into Eq. (47), we find
that the resonance values are j = -1,4,5 and 6. Then equating the
lowest order coefficients to zero, we obtain,

2
j 0 Uo -2~x (48a)

j u1 12~xx (48b)

j 2 3 2 2 30'2~ (48c)
~x~t - ~xx + 4~x~xxx + 6u2~x + 0
Y

2
j 3 ~xt + 6u2~xx 6u3~x + 302~ yy + ~xxxx 0 ( 48d)

j 4 o (48e)

j 5 6 2 + + 30 2 ~yy ) xx o (48f)
u3~x ~xxxx

j 6

2
30 ~
yy u 3 o. (48g)

From the above sets of equations, one can verify that the functions
u4 ' Us and u 6 are arbitrary [9,10] and so the P-property holds.

If we choose the arbitrary functions u 4 ' Us and u 6 identically

to be zero, it is easily seen that u j = 0, j >7 provided u 3 = 0,
in which case u 2 satisfies the KP equation

157
 o. (49 )

Accordingly, we have the Backlund transformation

u (50 )

where ~ satisfies Eqs. (48).

Eliminating u 2 from Eqs. (48c) and (48d), it is found that

2
a ~t
30 1-(2)
- ( - +{~ ; x} +
3 2 ~ + 2
o. ( 5' )
ax ~x 2" 0 2)
ay ~x
~x

Integrating (5') with respect to x and inserting ~ v,/v 2 ' we require

that (v, ,v 2 ) satisfy

ov y f,v xx + f 2 , (52 )

where f" f 2 , f 3 , f4 and f5 are functions of (x,t,y,u) to be determined.

Finding the fi's, we obtain finally the Lax-pair in the form

ov
Y
+ v
xx
+ uv o (53a)

x
v t + 4v xxx + 6uv x + 3(u x - 0 f u y dx') o (53b)

the compatibility of which is the KP eq. (47).

5. INTERCONNECTION BETWEEN THE PAINLEVE PROPERTY

AND HIROTA'S METHOD

Among the direct methods, Hirota's bilinearization technique is the

most convenient one to construct soliton solutions of npdes. Essen-
tially, the method advocates transforming the given equation into
homogeneous bilinear forms and then by 'self-truncating' the series
to obtain a set of bilinear equations. In this process, the reasons
for both the choice of the initial transformation and the self-trun-
cation of the series is rather obscure. From the P-analysis, it
is clear that the truncation of the Laurent series at the constant
level term results in the BT. For the vacuum solutions using the BT
we can obtain in most cases the required Hirota's dependent variable
transformation to transform the given equation into the bilinear

158
operator form [13-15]. As an example we consider [12], the coupled
nonlinear Schrodinger eq. (39). The BT for Eq. (39)

P Q (54a)

-1
R S = So 'P + S1 (54b)

in the vacuum case leads to the following set of equations:

[QOt'P-Qo'P t + c 1 (POxx'P+ PO'Pxx - 2P Ox'P x )]'P

(55a)

[PO'Pt-POt'P +c 1 (QOxx'P+QO'Pxx-2QOx'Px)]'P
(55b)

[SOt'P-SO'Pt+c1(ROxx'P-2RO'Pxt - 2R Ox'Px)]'P
(55c)

[RO'Pt-ROt'P+C1(SOxx'P+SO'Pxx-2S0x'Px)]'P,
(55d)

where

Eqs. (55) can be reexpressed in terms of the Hirota's bilinear opera-

tors and after decoupling we obtain the form

).I) 'P. 'P 2 ar (56a)

2 (56b)
DtQO'P + c1DxPO'P -).I PO'P 0

2 0 (56c)
-DtPO'P + C1 Dx QO'P ).IQO 'P

2
DtSO'P + c1D~RO'P - ).IR O'P 0 (56d)

2 - ).ISO 'P 0, (56e)

-DtRO'P + C1Dx S O'P

where ).I is a constant to be determined and

a a n a a m
(at - at,) (ax - ax') f(x,t)g(x' ,t')
x'=x
t'=t

159
In particular, from (56a), we get

a2
p2 + Q2 + R2 + S2
0 0
l( c log cp _ ..l:!) ( 57)
0 0 1 2 '
ax 2
CI.

and for the choice (45 ) we obtain

p2 + Q2 1 a2
(R 2 + S2) -(c - log cp ~). (58 )
0 0 0 0 CI. 1 ax2

Now expanding the functions cp, PO' QO' RO and So as power series
[12] and using them in (57) we can construct the N soliton solutions.
Similar analysis can be applied to the other systems as well.

6. CONCLUSIONS

In this review, we have briefly pointed out how the Painleve analysis
lS a useful and systematic procedure for investigating the integra-
bility properties of nonlinear partial differential equations. For
a class of physically important equations we have demonstrated that
this technique can in a straightforward manner lead to linearizations,
Lax pairs and Hirota's bilinearization, showing the deep interconnec-
tions between the Painleve analysis and integrability.

REFERENCES

[ 1] M. J. Ablowitz and H. Segur, Solitons and the Inverse Scattering

Transform (SIAM, Philadelphia, 1981).
[ 2] P. J. OlveL, fippLlca~lons ot Lie Groups to Differential
Equations (Springer, New York, 1986).
[3] M. J. Ablowitz, A. Ramani and H. Segur, Lett. Nuovo Cim. 23,
(1978) 333; J. Math. Phys. 21 (1980) 715.
[ 4] E. Hille, Ordinary Differential Equations in the Complex Domain
(Wiley, New York, 1976).
[ 5] E. L. Ince, Ordinary Differential Equations (Dover, New York,
1956) .
[6 ] S. Kovalevskaya, Acta Math. 12 (1889) 177; R. Cooke, The Mathe-
matics of Sonya Kovalevskaya (Springer, New York, 1984).
[7] M. Lakshmanan and P. Kaliappan, J. Math. Dhys. 24 (1983) 795.
[ 8] J. Weiss, M. Tabor and G. Carnevale, J. Math. Phys. 24 (1983)
522.
[9] J. Weiss, J. Math. Phys. 26 (1985) 2174.
[10] W. Oevel and W-H. Steeb, Phys. Lett. 103A (1984) 239.
[11] K. M. Tamizhmani and M. Lakshmanan, J. Math. Phys. 27 (1986)
2257.
[12] R. Sahadevan, K. M. Tamizhmani and M. Lakshmanan, J. Phys. A19
(1986) 1783.
[13] K. M. Tamizhmani, "Geometrical, Group Theoretical and Singularity
Structure Analysis Aspects of Certain Nonlinear Partial
Differential Equations," Ph.D. Thesis, University of Madras (1986).

160
[ 14 ] J. D. Gibbon, P. Radmore, M. Tabor and D. Wood, Stud. In Appl.
Math. 72 ( 1985) 39.
[ 15] M. Tabor and J. D. Gibbon, Physica 18D ( 1986 ) 180.
[ 16 ] v. E. Zakharov and E. I. Schulman, Physica 4D (1982) 270.
[ 17 ] H. Yoshida, Celest. Mech. 31 ( 1983 ) 363; 381.
[ 18 ] A. Ramani, B. Dorizzi and B. Grammaticos, Phys. Rev. Lett.
49 ( 1 982) 1539 ; J. Math. Phys. 24 (1983) 2252.
[ 19 ] M. Lakshmanan and R. Sahadevan, Phys. Rev. A31 (1984) 861 ;
R. Sahadevan and M. Lakshmanan, Phys. Rev. A33 ( 1985) 3563.

161
Generalised Burgers Equations
and Connection Problems
for Euler-Painleve Transcendents
P.L. Sachdev
Department of Applied Mathematics, Indian Institute of Science,
Bangalore 560012, India

Some recent results on a class of generalised Burgers' equations (GBE)

are reviewed. Characteristically they reduce to nonlinear Euler-
Painleve equations, on similarity reduction, possessing single hump type
solution. GBE's with variable viscosity coefficients are briefly dis-
cussed. It is pointed out that the Korteweg-de Vries type equations and
GBE's behave somewhat analogously in their self-similar forms.

1. INTRODUCTION

There has been a pervading and persistent interest for a couple of

decades in the study of the model nonlinear equation, called the Korteweg-
de Vries equation,

o. ( 1)

This describes a certain balance between the (simplest) dispersion and

nonlinear convections. Equation (1) and its kindred class have three
main unifying features: (1) they exhibit clean soliton interaction;
(2) they can be exactly linearised through inverse scattering theory
to linear integral equations of Gelfand-Levitan type; (3) their self-
similar form, as it is or through some simple transformations, belongs
to the class of nonlinear ODE's whose only movable singularities are
poles; this characteristic of the ODE's (or POE's) is called Painleve
property, after Painleve who first studied second order nonlinear ODE's
from their singularity structure point of view.

There is another distinguished model equation,

(2 )

called the Burgers equation, which has a certain kinship with (1).
Ithad preceded the latter in its inception and investigation. Equation

162
(2) represents a balance between linear diffusion and nonlinear convec-
tion. Equations (1) and (2) have the same convective term but different
higher order terms. Each of these equations has influenced the study
of the other. Equation (2) is simpler and can, in fact, be exactly
linearised to the heat equation by the Hopf-Cole transformation [1].
In contrast, Eq. (1) has no such simple transformation and attempts
to linearise it by a Hopf-Cole type transformation have led to its further
'non-linearization' [2]. Equation (2) does not display soliton behaviour;
indeed it has no solitary wave solution. Actually, Eq. (2) is a severe
idealisation of reality. The equations which actually occur in applica-
tions are invariably more complicated. We quote two such equations:

ut + u8u + Au a 0
x 2" u xx' (3 )

u t + u a u + ju 0
x 2t 2" u xx' j 0,1 ,2. (4 )

Equation (3 ) has a lower order damping term and describes stress waves
in a nonlinear Maxwell rod, while Eq. (4 ) has a spherical or cylindrical
term besides those in (2). Both these equations have more general non-
linear convective terms than (2). We shall mention other generalised
Burgers equations (GBE's) in the sequel.

It does not seem possible to linearise GBE's of the type (3) and
(4) through Hopf-Cole like transformations. The question arises whether
we can unify this class of equations in some manner. Since soliton
behaviour and exact linearisation do not seem possible, the one option
open is their characterisation through the ODE's obtained by similarity
transformations. This turns out to be possible. We find that there
is a class of nonlinear ODE's, which we refer to as Euler-Painleve equa-
tions, which describes the self-similar single hump (and sometime other)
type solutions of several GBE's. A very special case of this equation
was studied by Euler and Painleve [3], hence the name Euler-Painleve
transcendents.

In the presen,t paper, we summar i se our work on GBE' sand Euler-

Painleve transcendents [4,5]. We first study the Burgers equation and
its self-similar single hump solution. The self-similar solutions of
(3) are then studied with reference to a connection problem with respect
to the 'reduced' nonlinear ODE's on the whole real line. This is followed
by a similar study for (4). Some current work on GBE's with variable
viscosity coefficients is then briefly discussed. Finally, we refer
to a large number of DE's, which have been categorised by Kamke [3]
according to their formal appearance and appended with geometrical or

163
physical significance. These are special cases of the Euler-Painleve
equations if the coefficients in the form of the latter are made to
vary with the independent variable. A reference to the connection pro-
blem for the K-dV equation as represented by the second Painleve trans-
cendent is made again to show how K-dV type of equations and GBE's
behave somewhat analogously in their self-similar forms. However, we
caution that Euler-Painleve equations are free from singularities on
the finite real line, unlike Painleve equations, which, in general,
have an infinite number of poles.

2. SINGLE HUMP SOLUTION OF BURGERS EQUATION

We study in the following a specific self-similar solution which has

a single hump form and vanishes at x = ± 00. For the Burgers equation
(2), this solution can be written out as

u (6/t) 1/2 f(n), n = x(26t)-1/2 (5 )

where
(2Tf)1/2 2 1/2 2 1
f (n) [ R exp(n ) + (Tf/2) exp(n) .erfcnl-
e - 1

, say, (6 )

where the function f(n) satisfies the DE

f" + 2nf' + 2f - 2 3 /2 ff' o . (7 )

The 'inverse' function HB(n) plays an important role for GBE's, as we

discusss subsequently. In the present case, it satisfies the DE,

o. (8 )

We note a few properties of this single hump solution, which, in a sense,

is closest to the solitary wave. We emphasise that this solution is
not symmetric, in contrast to the solitary wave. The solution (5) arises
out of singular initial conditions u(x,o) = Co + A6(x), where co=const.,
6(x) is the delta function--singular initial data are common to most
self-similar solutions. By integrating (2) from x = -00 to x = +00 ,

one may easily conclude that the Reynolds number

1
R f u dx (9 )
"6

164
is constant. Here, we use the conditions that u and its x-derivatives
vanish at x = too. The solution (5) vanishes exponentially as x~ +00
and algebraically as x~ -00. One may further verify that in the limit
6 ~ 0, t,hat is, R ~ 00, the solution reduces to

u x/t, ( 10)

the solution of the inviscid Burgers equation with 6 = O. For large

but finite R, the single hump consists of the inviscid solution for
~</2, followed by a thin shock ending in a diffusive tail. For R~ 0,
it easily follows from (6) that the solution (5) is essentially diffu-
sive; the nonlinear convective term plays no essential role. Hopf [121
pointed out that the solution (5) represents a 'stationary' solution,
to which an infinite number of solutions, with essentially similar be-
haviour at x ~ ± 00 in their initial conditions, converge as t ~ 00. Such
self-similar solutions in Soviet literature are now referred to as inter-
mediate asymptotics [11.

3. SINGLE HUMP SOLUTIONS FOR GBE WITH DAMPING--

THE CONNECTION PROBLEM

Now we seek self-similar single hump solutions of (3) [41. We find

that the solution has the form

t 1 / ( 1 -0: ) n), ( 11 )
u f(

n x{26t)-1/2,

provided

B {o: - 1 )/2. ( 12)

Equation (3) then reduces to the ODE

f" + 2nf' - _4_f - 4(26)-1/2 f}{O:-l) f' - 4Ho: O. ( 13)

'1 -0:

The solutions (11) decay (explicitly with time) if 0: > 1 and grow if
0:< 1. We employ the 'inverse' transformation (cf. (6))

H 61 / 2 f ( 1- 0:) /2 ( 14)

and obtain the form

o ( 15)

165
of (13), free from fractional powers of f. Here,

1 3-a
)./)(1 - a). ( 16)
"2a-1' ).1

Equation (15) generalises (8) in two ways: the coefficient of H' becomes
-2(1 + ( 1 ) instead of -2, and a constant -2).1 gets added on to (8).
These 'minor' changes represent a fairly general GBE (3). Before we
pose a connection problem for (13), we note two special solutions of
(13) and (15). First, (13) has a constant solution,

f f m, say. (17 )

It is easy to check that fm is the maximum value of f that the maxima

of the single hump solutions can attain. The other exact solution is

11 > 0,

11 < 0, (18 )

where

A+ (2/6) 1/2 (a-1) [ (1 + )./) ( l+a) ) 1/2 + 1 1 ,

a+1

A (19 )

The corresponding solutions for H(11) can be written with the help of
(14). Equation (15) has the Taylor series solution

H (20 )

for the decaying case a >1, where

(21 )

Here a o and a 1 are two arbitrary constants. It is easy to check that

the solution HB in (6), when eXP(11 2 ) and erfc 11 are suitab;y expanded,
is a special case of (20) with a = 3, ). = 0, and a 1 = _22. For this
choice of a 1 , we obtain a single parameter family of solutions of Burgers
equation, the single parameter a o representing the amplitude at 'in-
finity' of the various curves, or their Reynolds number. In general,
it is difficult to identify the ranges of parameters a o and a 1 in (20),

166
for which the series converges for -00 < n < 00. For this purpose we
go back to the self-similar function f governed by (13), and pose for
the latter a connection problem. Since we require the solutions to
vanish at n = ± 00, we linearise (13) for its asymptotic behaviour:

4
f" + 2n f' - 1 -a f o. (22 )

This equation has the solution

2 2 2a 1
f A exp (-n ) Hv (n ) - A exp (-n ) ( 2n ) as n t 00, (23a)
and
-2Cl 1 -1
f - 0 (n ) as n .. -00 , (23b)

where H is the Hermite function of order v and A is the amplitude

v
parameter. Thus, the asymptotic behaviour is similar to that for Burgers
equation--exponential and algebraic for n - 00 and n --00, respectively.

Equation (13) was solved numerically starting with the asymptotic

behaviour (23a) for large positive n. The main result of this study
was that in the decaying range a >1, the self-similar form of single
hump type exists if 1 < a 2 3; the amplitude parameter A for each admis-
sible value of (a,S) could at most reach a maximum value of A = Amax
For this value of A, the solution starting with (23a) would rise to
approach the special constant solution f = fm (see Eq. (17)). For
A > Amax' there is no single hump solution. The solution grows to become
unbounded at n = -00. Sachdev et al. [4] carried out a thorough investi-
gation of the intermediate asymptotic nature of these self-similar solu-
tions by integrating the PDE (3) by pseudo-spectral and implicit schemes.
Here, we emphasize that the self-similar form of (3) is governed by
the equation (15) for H, the inverse function. Equation (15) is a special
case of Euler-Painleve equation which we introduce in Sec. 6.

4. SINGLE HUMP SOLUTIONS OF NON-PLANAR GBE'S

Now we consider the self-similar solutions of (4) in the form

u t- 1 / 2a f(n), n = x(2tSt)-1/2, (24 )

so that f(n) is governed by the ODE

f' , - 2 3 /2 o-1/2 f a f , + 2n f ' + 2 ( 1 - aj) f 0, (25 )

a

167
where a is a parameter. The 'inverse' transformation

H (26 )

changes (25) to

HH" - a+1 H,2 + 2n HH ' - 2{1 - aj)H 2 - 2 3 /2 H' O. (27 )

a

Equation (27) m3Y be compared with (8) corresponding to the standard

Burgers equation (2): only the numerical coefficients in (27) differ
from those in (8) and the former reduces to the latter when j = 0 and
a = 1. Equation (27) belongs to the Euler-Painleve class (see Sec. 6).
Equation (27) has a series solution (20) with two arbitrary parameters,
and the recurrence relation for the coefficients can be written out
[5]. To identify the single hump solutions for (25) and the inverted
hump solutions for (27), we again pose the connection problem for (25).
The linearised form of the latter,

f I' + 2n f ' + 2 ( 1 - aj) f 0, (28 )

a

has the solution

2
f (n) A e- n Hv (n) , n > 0

B 1\1/2
f (n) - Inl j - 1 / a , n« 0, (29)
I{-v)
provided v = l/a - (j+1), H (n) is the Hermite func-
aj < 1. Here
v
tion of order v, and A and B are the amplitude parameters. Thus, the
linear solution decays exponentially as n ~ooand algebraically as n~- 00.
The connection problem may thus be stated as follows:

fll - 2 3 /2 8- 1 / 2 faf' + 2n f ' + 2{1-aj) f = 0, (30 )

a
2
f A exp{-n ) Hv{n), n t 00 ( 31 )

f ~ o (n .j.- 00), ( 32)

and

If I < 00, -00 < n < 00 • ( 33)

Before we give the numerical results for the above problem, we dis-
cuss a class of special exact solutions which provides some clues to
the general case. If a = 1/{j+1), Eq. (30) reduces to

f + nf I + If' I ( 34)
2
168
Integrating (34) and using vanishing conditions (31) at n 00, we get

n f + 21 f' (35 )

The transformation G = f- a changes (35) to

G' - 2anG a~~ (2/0) 1/2. (36 )

This equation can be integrated. We, thus, have

2
-l/a
f (n) exp(-n ) { c _ ~2 (2a/6) 1/2 dt} , ( 37)
a+1

where c = f-a(O). This solution generalises the solution (6) for Burgers
equation to other geometries, provided the parameter a equals 1/(j+1);
thus for j = 1, a = 1/2 and j = 2, a = 1/3, we have explicit single
hump solutions of (4). We can infer more about the solutions of (4)
by deriving relations involving some integrals. Writing F = fa, Eq.(25)
can be transformed as

1 FF' - a-1 F,2 + (1 - a]')F 2 + nFF' _ (2/6) 1/2 F2F' O. (38 )

2 ~

Integrating (38) from n = -00 to n 00, and using vanishing conditions

there for F and F', we get

1-2a (39 )
(2aj - 1)
a

Equation (39) implies the following:

(i ) j O. The ratio

fF2 dn
r -
(1 - 2a)
> 0 if a
1 (40 )
a > 2·
-Z F,2 dn

1
Therefore, the single hump solutions in this case exist only if a >2.

1
( i i) j 1. In this case (39) holds i f a 2. This corresponds to
the exact solution (37).

(iii) j = 2. In this case, the ratio r of the integrals in (40) is

(1 - 2a)/a(4a - 1). This is positive if 1/4 < a<1/2. This is the
range for which the single hump solutions may exist.

169
 The numerical study of the connection problem (30)-(33) indicated
the existence of solutions for a < 1/(j+1), j = 0,2, which go to zero
at a finite point n = nO' say, instead of n = -co. Indeed, one can
easily derive the relation

co
a( j+1 ) - 1) 1
a
I f dn - "2 f I ( nO) (41)
no
for this case. Since, evidently, f I ( nO)

Table 1 Single hump, monotonic and diverging solutions of Eq. (30)

Behaviour at left boundary j 0 j j 2

Solutions vanishing at a = a 1/2 1/3 <a < 1/2

n = -co
Solutions vanishing at 1/2 <a < 1 1/4 < a < 1 /3
n = nO
Solutions monotonically a = a 1/2
approaching a constant
at n= -co

Solutions diverging to a > 1 a > 1/2

infinity at n = -00

The details of the numerical study of the problem (30)-(33)may be seen

in Sachdev and Nair [5J. Apart from the results summarised in Table 1,
we note that, unlike the single hump solutions of GBE (3) with damping,
the nature o~ the solution depends on the parameters a and j. The
solutions exist for the permissible values of a and j for all values
of the amplitude parameter. The particular value a l / j seems to
bifurcate the generality of solutions. For this value of a, f = fc'
a constant, is a solution of (30) so that, for a given value of A, the
solution starting at n = -00, tends to a constant nonzero value at n =-co
instead of vanishing there or at n = nO. For example, for j = 1, A = 1,
fc 0.41187, and for j = 2, A 1, fc = 0.10197. The numerical solu-
tion of the connection problem agrees very closely with the series solu-

170
tion for H, governed by (27), when the values H(O) and H' (0) are suitably
identified from the numerical solution. The intermediate asymptotic
nature of the self-similar solutions was verified by solving the PDE[4]
numerically for the relevant values of the parameters a and j and 'appro-
priate' initial conditions vanishing at n= ± 00.

5. GENERALISED BURGERS EQUATION WITH

VARIABLE VISCOSITY

We consider the GBE

6 (42 )
2" g(t) uU x '

where g(t) is a smooth positive function, representing the dependence

of viscosity on time, Scott [6] considered a special case of (42) with
a = 1; the role of t and x, however, was interchanged in his study.
He considered a piston problem for this equation and proved the inter-
mediate asymptotic nature of the self-similar solution u = nIt/x) for
cylindrical and sub-cylindrical cases. The latter case was defined
such that ~
x
~ 0 as x ~ 00. Conversely, if. 9l!l
x '
~the GBE is
00

referred to as super-cylindrical. Here, we consider the cylindrical

and sub-cylindrical form

.§.( 1 + tIn u (43)

2 xx'

where n is a parameter and a >0. We shall find that self-similar solu-

tions either decaying or oscillating at x = ±ooexist for (43) . only when
-1 < n ::. 1, i. e., when (43) is, ei ther cylindrical or sub-cylindrical.
We easily check that the self-similar form of the solutions of (43) is

u = (44 )

so that it becomes

f" - 2f a f' + (n+1 )n f ' --a-

(n-1) f O. (45 )

The inverse function

H (46)

is, therefore, governed by

HH" a+1 H,2 + (n+1)n HH' - 2H' + (n-1 )H 2

- -a- O. (47 )

171
For single hump solutions of (45), we enquire when f has a maximum:
f' = 0, f" = nS 1 f < O. A necessary condition for single hump solu-
tions for the case S > 0 is that n < 1. Equation (47), as we shall
see, again belongs to the class of Euler-Painleve equations. Before
we pose a connection problem for (45), we note that it has a single
1-n .
parameter family of exact solutions. Assuming S l+n' we wrIte (45)
in the form

f" - 2f S f' + 2 (nf' + f) O. (48 )

S+l

Integrating (48) and assuming that f and f' tend to zero (such that
nf 4 0) as n 4 ± 00, we have

2
(49 )
S+l

Integrating (49), we get

a 2 erf(/a72 n )}]-l/S,
f [exp(2 n {A - ;2a (50 )

where a = 2S/(S+1) and A is the constant of integration equal to f(O)-S.

For decaying solutions, we require that a > 0 so that n < 1. We also
note that the special case S = 1, n = 1, which was treated by Scott
[6] can be solved in a parametric form. The linearised form of (45) is

n-1 f
fll + (n+1)n f' - -S- ( 51 )
O.

Its solution is simply the confluent hypergeometric function

1-n 1 n+1 2 .
~(2S(1+n) , 2; z), where z = - --2- n. The asymptotIc form of ~ im-
mediately suggests that the solutions vanish at n = ± 00 provided
1 1-n
2j3 1+n > 0, i.e., -1 < n<1. The case n = 1 leads to the erfc(n) solu-
tion of (51). The series solution for (47) exists in the permissible
range of n and describes either the single hump solution or shock-like
solution or the solutions with oscillatory tail and/or front. The re-
sults of the connection problem for (45) and (47) and other details
will be published elsewhere [7].

6. EULER-PAINLEVg TRANSCENDENTS

In the compendium of nonlinear differential equations, compiled by Kamke

[3], the equation

yy' , + ay' 2 + f(x)yy' + g(x)y 2 o (52 )

172
is attributed to Euler and Painleve; the motivation for the study of
this equation is not clear. This equation is, in fact, exactly linearised
by the transformation

y
v l/( 1+a)
(53 )

to
v" + fv' + (a+1 )gv o. (54 )
We have generalised Eq. (52) to the form

yy' + a(x)y' 2 + f(x)yy' + g(x)y 2 + b(x)y' + c(x) = 0, (55 )

where the constant a in (52) has been made to vary with x and a linear
term in y' and a function c(x) have been added to (52). Our claim is
that Eq. (55) characterises the GBE's in their self-similar form in
the same manner as the Painleve type equations describe Korteweg-de
Vries and other model dispersive equations. Indeed, the three equations
corresponding to the GBE's (3), (4) and (42), namely (15), (27) and
(47) easily follow from (55) if special choices of the functions a(x),
f(x), g(x), b(x) and c(x) are made. We must, however, emphasize the
difference between the nature of Eq. (55) and the Painleve type of equa-
tions. For the physically realistic cases of Eq. (55) which we have
studied, the solutions are either single hump type or shock-like or
have oscillatory tails, and there are no singularities of any kind in
the finite part of the real line. In contrast, the Painleve equations
are typified by the property that their only movable singularities are
poles. That Euler-Painleve equations should be 'nicer' than Painleve
equations follows also from the nature of the BGE's and K-dV type of
equations; roughly speaking, diffusion is 'smoother' than dispersion.
We must also point out that Eq. (55) for general (smooth) coefficients
would need more extensive investigation than we have carried out for
the special cases, arising directly from GBE's.

It is remarkable that 65 nonlinear DE's in the compendia of Kamke

[3] and Murphy [8] are special cases of (55) directly or by simple
transformations. These equations are either autonomous or are lineari-
sable by a power law or logarithmic transformation. Alternatively,
they are reducible to first order equations of Riccati or Bernoulli
type. Most of these equations are either integrable explicitly or admit
at least one quadrature allowing their treatment in the phase plane.
Here we give a listing of these equations and refer the reader to Kamke
and Murphy for their physical importance and solution. In Kamke's book,
these are listed in Sec. 6, p. 542 as 104-111, 117, 122, 124-127, 129,

173
131, 133-4, 136-9, 150-2, 155-8, 164, 166, 168-70, 173-9. In Murphy's
book, these are enumerated as 133, 140, 142, 150, 190, 195, 199, 201,
203-4, 219-22, 227-31, 233-4 in Part 2, Chapter 4. It would thus appear
that Eq. (55) has a quite ubiquitous nature and deserves close study
and investigation.

7. CONCLUSIONS

We have shown with the help of 3 GBE's--(3), (4) and (42) that Eq.(55),
which we have referred to as Euler-Painleve equation, seems to charac-
terise this class of equations. We have drawn an analogy with Painleve
type of equations which unify the study of K-dV equation and its kindred
class. We have summarised the results of the connection problems for
some GBE's. Similar study for Painleve second equation has been carried
out by Miles [9] and Rosales [10]. It is conceivable that not all GBE's
will be typified by (55). Indeed, self-similar form of solutions does
not always represent intermediate asymptotics. Besides, sometimesit
is the self-similar form of the linearised POE which may represent inter-
mediate asymptotics to which a large class of solutions arising from
a certain set of initial conditions approach [11]. This is the case,
for example, for the super-cylindrical equation, as shown by Scott [6].
In this case, it is not the nonlinear Euler-Painleve equation (55) which
describes the asymptotic behaviour of a certain class of initial value
problems but the corresponding self-similar solutions of the heat equa-
tion with a variable coefficient. To conclude, we would have to investi-
gate other GBE's and study (55) more deeply to come to a firm under-
standing of the role of Euler-Painleve equations.

REFERENCES

[1] P. L. Sachdev, Nonlinear Diffusive Waves (Cambridge University

Press, London, 1987).
[2] G. B. Whitham, Linear and Nonlinear Waves (Wiley, New York, 1974).
[3] E. Kamke, Differential Gleichungen: Losungsmethoden und Losungen
(Akademische Verlagsgeselleschaft, Leipzig, 1943).
[4] P. L. Sachdev, K. R. C. Nair and V. G. Tikekar, Generalised Burgers
equations and Euler-Painleve transcendents-I, J. Math. Phys.
27 (1986) 1506.
[5] P. L. Sachdev and K. R. C. Nair, Generalised Burgers equations
and Euler-Painleve transcendents-II, J. Math. Phys. (1987),to
appear.
[6] J. F. Scott, The long time asymptotics of solutions to the genera-
lised Burgers equations, Proc. R. Soc. Lond. A373 (1981) 443.
[7] P. L. Sachdev and K. R. C. Nair, Generalised Burgers equations
and Euler-Painleve transcendents-III (1987), to be published.
[8] G. M. Murphy, Ordinary Differential Equations and Their Solutions
(Van Nostrand, New York, 1960).

174
 (9] J. W. Miles, On the second Painleve transcendent, Proc. R. Soc.
Lond. A361 (1978) 277.
(10] R. Rosales, The similarity solution for the Korteweg-de Vries
equation and the related Painleve transcendent, Proc. R. Soc.
Lond. A361 (1978) 265.
(11] G. I. Barenblatt, Similarity, Self-Similarity and Intermediate
Asymptotics (Consultants Bureau, New York, 1979).
(12] E. Hopf, The partial differential equation u t + uU x QU xx '
Communs. Pure Appl. Math. 3 (1950) 201.

175
Backlund Transformations and Soliton Wave Functions
J.A. Rao 1 and A.A. Rangwala 2
1 Departmentof Physics, Ramnarain Ruia College,
Bombay 400019, India
2Department of Physics, University of Bombay, Vidyanagari,
Bombay 400098, India

USing Darboux-Bargmann technique, we obtain (1) the Backlund trans-

formations for any nonlinear evolution equation (NLEE) solvable by
the inverse scattering method of Zakharov-Shabat--Ablowitz-Kaup-Newell-
Segur (ZS/AKNS) and (2) the ZS/AKNS wave functions corresponding to
the n-soliton solution of this NLEE.

1. BKcKLUND TRANSFORMATIONS AND SOLITON WAVE FUNCTIONS

The ZS!AKNS scattering problem [1,2] is defined by the eigenvalue

equation

q(x,t)v 2 , (1. la)

v - r(x,t)v 1 , (1. lb)

2x
v1
where v = ( ) is the two-component ZS!AKNS wave function and q and
v2
r are functions of x and t satisfying the NLEE of interest. The eigen-
value is k. Equations (1.1) give the space evolution of the wave func-
tion. Let us distinguish the quantities, wave function and potentials
q and r, referring to n solitons by primes and those referring to (n-l)
solitons by unprimed ones. Thus:

v'1 (
viI); q' r' (1. 2a)
v2
and

v 1 ,2 satisfy eqs. (1.1). v1,2 satisfy similar equations:

v'2x - ikv 2' r'vi· (1. 3)

We appeal to Darboux's method [3] to expand the n-soliton wave

function v' in terms of (n-l)-soliton solution v:

176
 v'1 v' CV l + DV 2 , ( 1 .4 )
2

where A, B, C, D are functions of x and t. From eqs. (l.l), (1.3) and

(1.4), we see that A, ... ,D satisfy the differential equations:

A -rB + q'C, (1. Sa)

x

B -qA - 2ikB + q'D, (1.5b)

x

C r'A + 2ikC - rD, ( 1 .Sc)

x

D r'B - qC. ( 1 • 5d)

x
We next invoke an idea due to Bargmann [4]. It is well known that
the ZS/AKNS equation corresponding to the Korteweg-de Vries (KdV)
equation is the Schrodinger equation [5]. In the context of the
Schrodinger equation, Bargmann has shown that for a potential capable
of giving n bound states, the solution of the Schrodinger equation
can be written in the form e ikx X(k,x}, where X(k,x} is an n-th degree
polynomial in k. On the other hand, we know that n-soliton solution
of an NLEE envisaged by the ZS/AKNS eqs. (l.l) can be looked upon as
a potential giving n bound states [5]. Thus the idea due to Bargmann
suggests that v and v' will differ by a linear function of k. We there-
fore write:

A B (1. 6a)

C D (1. 6b)

where ai, ... ,d i are functions of x and t through (q,r) and (q' ,r').
The differential equations obtained for a, ... ,d by using eqs. (1.6)
in eqs. ( 1 .5) are easy to solve and a partial solution is:

a, a.(t} ,

b, 0, bO ~[(i}q' - q],
21 a.

cl 0, Co - 2~[r' (~) r],

6 (t), -(i}a B (t) ,

0 +
dl dO ( 1 .7 )
a.

with a O satisfying the equation

a.
a Ox 2i(qr - q' r'} (1. 8)

177
and, additionally, two equations of constraint need to be satisfied:

~[ (.Q.) q' - qx 1 - [q + (~) q' 1a 0 + Sq', (1. 9a)

21 CL X

(1. 9b)

Here CL, S, 0 are constants of integration and in general depend on t.

Subsequently we shall see that to be able to describe the soliton solu-
tions they need to be taken as constants. From eqs. (1.7)-(1.8), we
can see that we can obtain A, ... ,D in eqs. (1.4) provided we can solve
eq. (1.8) for a O and be able to satisfy eqs. (1.9).

It is easy to obtain a O as a function of (q,r) and (q' ,r') and hence

to show that eqs. (1.9) in fact represent the space part of Backlund
transformations (BTs). To see this, we first notice that the choice
of overall factor, CL say, is at our disposal. We set CL = -i and define
€ = -(.Q.). We multiply eq. (1.9a) by r' and eq. (1.9b) by q, add the
CL
resulting equations and use eq. (1.8) to obtain

( 1. 10)

where

( 1. 11 )

Similarly multiplying eq. (1.9a) by rand eq. (1.9b) by q' and adding
the resulting equations and once again using eq. (1.8), we get

( 1. 12)

Adding eqs. (1.10) and (1.12), we see that the resulting equation is
integrable only if

2
€ or € ±1 • ( 1. 13)

In all the subsequent discussion, we shall assume eq. (1.13) to hold.

The sum of eqs. (1.10) and (1.12) when integrated gives a quadratic
equation for a O whose solution is

1 2 !.
± 2[4v' +(q'+€q) (r'+n) J2H(~). ( 1. 14)

This solves eq. (1.8) consistent with eqs. (1.9). Here~' and v' are
integration constants. They can in principle be functions of t and

178
may even be complex. Soon, however, we shall identify them with soli-
ton pole position which requires them not only to be constants but also
real. We believe that the present method can be extended to include
perturbed NLEEs and then this possibility of time dependence of pole
position parameters becomes necessary. The function H(s) is the Heavi-
side function

1, s > 0
H( s) (
-1 , s < o. (1. 15)

It can be shown that in the pure multisoliton case, the argument s is

given by

x + ~ Imw(k') - xO' ( 1. 16)

v'

where w(k) is the dispersion function and Xo is a constant. The Heavi-

side function is required to ensure the continuity of both the soliton
solutions and the wave functions [6].

Using a O from eq. (1.14) in eqs. (1.9), we get the space part of
the BTs:

2 l
-2ill' (q'+Eq)±(q'-Eq) [4v' +(q'+Eq) (r'+Er) ]2H(S), (l.17a)

2 !
r~ +Er x = 2ill' (r'+Er)±(r'-€r) [4v' +(q'+€q) (r'+€r) ]2H(s). (1.17b)

We see that the constants 11', v' appearing in the above equations are
in fact the parameters of the n-th soliton. Since soliton solution
represents a pole in the complex k-plane at k = k' = 11' + iv', we
require (11' ,v') to be both real and independent of time. If the NLEE
of interest is such that if (q,r) is a solution so is (-q,-r) then both
values of € ,±1, are permitted. Otherwise we would have either E=+l
or €=-1. The two possible signs accompanying the discriminant are
linked with two possible directions of incidence; positive (negative)
sign corresponds to the incidence from the left (right). These direc-
tions of incidence in their turn give rise to relevant analyticity
structure of the scattering amplitude or the transmission coefficient
in the upper half k-plane (right incidence) or in the lower half k-plane
(left incidence).

It is easy to show that eqs. (1.17) reproduce the BTs in the stan-
dard cases of KdV, sine-Gordon (sG) and nonlinear Schrodinger equation
(NLSE) [6,7] corresponding to the three classes r = constant, r=-q and

179
r=-q* respectively. Our present treatment gives a unified treatment
of the BTs for all NLEEs encompassed by the ZS/AKNS scheme.

Using a O from eq. (1.14) in eqs. (1.7) and substituting the results
into eqs. (1.6), we get
1 2 1
A -i(k-Il')± 2[4\1' +(q'+e:q)(r'+e:r)2 HU;), ( 1 . 18a)

B ~(q'+e:q) , ( 1 . 18b)

c ~(r' + e:r), ( 1 . 18c)

1 2 1
D d i ( k -11 ' ) ± 2 [ 4 \I' + ( q , + e: q )( r ' + e: r)2 H U; )], ( 1 . 18d)

which solve the differential equations in eqs. (1.5) and yield, when
used in eqs. (1.4), the n-soliton wave function v' in terms of (n-l)-
soliton solution v and the n-th and (n-l)-th soliton solutions (q' ,r')
and (q,r) respectively. These results agree for the three cases (1)
r = -1, (2) r = -q, (3) r = -q* discussed in our previous work [8].
The time evolution of ZS/AKNS wave functions is given by [2]

Vlt =:.4 (k;q,r)v l + (B (k;q,r)v 2 , ( 1 . 19a)

V2t =~ (k;q,r)v l -~ (k;q,r)v 2 · (1. 19b)

In the ZS/AKNS scheme one stipulates that k = O. The functions.A ,(B,~

t
depend on particular NLEE of interest. Use of eqs. (1.4) leads to

At A(.A' -cA) B~ + dB' , (1.20a)

Bt B(cA' +J) - A(B + D(B' , ( 1 .20b)

Ct -c (oA ' +04) + A~' D..e, (1.20c)

Dt -D(,4 '-.,4) + B.&' - c<il, ( 1 . 20d)

where .A' = .A '(k;q' ,r'), etc. From eqs. (1.18), we see that eqs.
(1.20b,c) give the time part of the BTs.

2. RICCATI EQUATIONS

It is possible to extend the previous analysis further. To this end

let us denote A, ... ,D given by eqs. (1.18) when evaluated at k=k' =

Il'+i\l' by A, ... ,D. Thus, we have

180
 1 2 1
A v' ± "2[4v' +(q'+e:q)(r'+n)]2H(F,;), (2.1a)

B ~(q' + e:q) , (2.1b)

C 1 ( , + e:r) , (2.1c)
"2 r

1 2 1
0 e:{-v'± "2[4v' +(q'+e:q)(r'+e:r)12H(~)}, (2.1d)

and we note that

A.D B.C. (2.2)

Also, we denote the wave functions in eqs. (1.4) evaluated at k=k' by

v 1 ,v 2 and vi,v 2 so that

(2.3)

If we define

r r' v-'1-'
1 v2, (2.4)

then it is easy to see from eqs. (1.1) and (1.19) that r satisfies
the Riccati equations:

q 2ik'r - rr2, (2. Sa)

(2.Sb)

with r' satisfying similar equations. If we now require that v'=O=v'

1 2
in eqs. (2.3), then we obtain

r = - BIA D/C. (2.6)

The consistency of last equality coming from eq. (2.2). It also follows
from eqs. (2.3) that cvi - A v 2
= (13 C - A D)v 2 • The right-hand side
of this equation vanishes in view of eq. (2.2) giving

r' A/C BID. (2.7)

rand r' are particular solutions of the Riccati eqs. (2.5) and their
primed counterparts.

Two useful relations between rand r' can be easily established.

Using eqs. (2.1) in eq. (2.6) and in eq.
(2.7l, i t is seen that e:r and
r' both satisfy the quadratic equation (r'+e:r)y 2 -4v'Y - (q'+e:q) = O.
It therefore follows that

181
 e:r + r' 4\1'/(r'+e:r), (2.8 )
and
( e:rlr' -(q'+e:q)/(r'+e:r). (2.9)

They can alternatively be written as

q'+e:q -4\1' (e:rlr' /(e:r+r'), (2.10)

and
r'+e:r 4\1'/(e:r+r'). (2.11 )

These equations are generalizations of those given by Konno,Sanuki and Wadati

[7] and were employed by us in Ref. 8. In the present general case,
it is unfortunately not possible to go beyond this stage. In special
situations, however, we can make further predictions. As an illustration,
consider the situation where we stipulate a relationship between q and
r like

(a) r = r' -1 (with e:=+1); (b) r = -q, r'=-q'; (c) r=-q*,r'=-q'*.

(2.12 )
These then imply relationships between e:r and r'. For instance,
corresponding to choice (a) in eqs. (2.12), we get from eq. (2.8) that

r' = - r - 2\1'; (2.13a)

for the choice (b), we obtain from eq. (2.9) that

r' l/e:r ; (2.13b)

and for the choice (c), we obtain from eq. (2.9) that

r' 1/e:r*. (2.13c)

Equations (2.13) form important ingredients in the methods of Konno-Sanuki-

Wadati [7] and Chen [6]. The above process is reversible. If we sti-
pulate relationships in eqs. (2.13),we obtain from eqs. (2.9) and (2.10)
relationships between q and r given in eqs. (2.12).

We now briefly outline the method of obtaining multi-soliton solu-

tions and cor,responding wave functions in the cases where r is stipulated
as some known function of q as, for instance, in eqs. (2.12). The
procedure is recursive and algebraic except for the starting zero-soliton
case where it requires solution of very simple differential equations.
We illustrate the procedure for the case of NLSE. Discussion for other
cases will be found in Ref. 8. For the NLSE case, on using eq.(2.12c)
in eq. (2.10), we get

q' - e:[q + 4\1' ---2]'

r (2.14 )
1+1 r I
182
whi.ch relates the n-soliton solution, q', to the (n-l)-soliton quantities,
q, v1 and v2 . This leads to the following procedure. We begin with
the zero-soliton solution q(O) of NLSE which is q(O) = O. This is sub-
stituted in eq. (1.1) to obtain the zero-soliton wave function v(O).
The time dependence of v (0) is obtained from eqs. (1.19) with.A, (jJ , ~
appropriate to the NLSE case. After this the procedure is purely
algebraic: We use these known q(O) and v(O) in eq. (2.14) to obtain
q(l). These q(l), q(O) and v(O) are now used in eqs. (1.4) together
with eqs. (1.18) to get v(l). This procedure can obviously be conti-
nued and furnishes in a simple manner all the higher order soliton
solutions and wave functions.

References

[1) V. B. Zakharov and A. B. Shabat, Sov. Phys. JETP 34 (1972) 62.

[2) M. J. Ablowitz, D. J. Kaup, A. C. Newell and H. Segur, Phys. Rev.
Lett. 31 (1973) 125;
M. J. Ablowitz, D. J. Kaup, A. C. Newell and H. Segur, Stud.
Appl. Math. 53 (1974) 249.
[3) G. Darboux, Lecons sur la Theorie Generale des Surfaces et les
Applications Geometriques du Calcul Infinitesimal, Vol. 2, 2d ed
(Paris: Gauthier-Villars, 1915), p. 210.
[4) V. Bargmann, Rev. Mod. Phys. 21 (1949) 488.
[5) See, for instance, G. L. Lamb, Jr., Elements of Soliton Theory
(New York: John Wiley, 1980) and R. K. Dodd, J. C. Eilbeck, J. D.
Gibbon and H. C. Morris, Solitons and Nonlinear Wave Equations
(New York: Academic Press, 1982).
[6) H. H. Chen, Phys. Rev. Lett. 33 (1974) 925.
[7) K. Konno and M. Wadati, Prog. Theor. Phys. 53 (1975) 1652;
M. Wadati, H. Sanuki and K. Konno, Prog. Theor. Phys. 53(1975)419.
[8) A. A. Rangwala and J. A. Rao, Phys. Lett. 112A (1985) 188.

183
Comparison of Some Numerical Schemes
for the K-dV Equation
A. Hasan and M.S. Kalra
Nuclear Engineering Program, Indian Institute of Technology,
Kanpur 208016, India

A numerical solution for the KdV equation has been obtained uSing a
finite element Galerkin scheme based on cubic splines in the space
variable. The results are compared with the finite difference schemes,
and the finite element Galerkin and Petrov-Galerkin schemes reported
in literature. It is found that the use of smoother trial and test
functions leads to much less L2 and L co errors. It is also seen that
quintic boundary polynomials used earlier are not necessary for obtain-
ing an accurate solution.

1• INTRODUCTION

It is well known that the initial and boundary value problems associated
with nonlinear partial differential equations are very difficult to
handle in a general way. The nonlinear evolution equations have recei-
ved particular attention over the past two decades or so. This is due
to the fact that they arise in a natural way in a large number of physi-
cal problems and in many cases possess special types of solutions which
may be of great practical use.

Our interest in the present work is in the numerical study of one

such evolution equation known as Korteweg-deVries (KdV) equation

o. ( 1)

This equation and its generalizations playa major role in the study of
nonlinear dispersive waves. Examples range from water waves and lattice
waves to plasma waves [1].

The numerical solution of (1) has been the subject of many papers
over the last few years. Zabusky and Kruskal were the first to study
the KdV equation numerically through a leap-frog finite difference
scheme [2]. Greig and Morris [3] proposed a Hopscotch finite difference
method and compared it with the original scheme of Zabusky and Kruskal.

184
Fornberg and Whitham [4] used spectral methods for x-variable and leap-
frog in t. The finite element Galerkin schemes and its modifications
have been used to solve the KdV equation by Alexander and Morris [5],
Sanz-Serna and Christie [6], and Mitchell and Schoombie [7], among others.

2. DETAILS OF THE PRESENT SCHEME AND

COMPUTATIONAL RESULTS

In this paper we present some numerical results obtained for the KdV
equation through a finite element Galerkin scheme. For this purpose
we approximate u(x,t) in (1) as follows:
N-2
u(x,t) U(x,t) L Ui (t) (jli (x), (2 )
i=2
where
( x-x o
(jli(x) (jl - h - - iJ , h (x N - xo)/N, (3)

(jl(x) (2+x)3, -2 < x < -1,

1+3(1+x)+3(1+x)2_ 3 (1+x)3, -1 < x < 0,

-
2 3
1+3(1-x)+3(1-x) -3(1-x) , 0 < x < 1,
- -
(2-x)3, < x < 2. (4 )
-
The function (jl(x) above is the basic or cardinal cubic spline function
[a]. Here we have considered the range of interest of the x-variable
from Xo
to x N · Since (jl1(x) as well as (jlN_1(x) extend beyond this
range, they have been omitted.

Now if we choose the test functions in the standard Galerkin scheme

[9] to be the same as the trial functions (jli(x), the problem of approxi-
mately solving (1) reduces to obtaining a solution for the following
system of ordinary differential equations:

dU i
Aik dt + B. 'kU,U, +
1 J. 1 J
c.1k U.1 0, (5 )

where summation over repeated indices is implied and

2
d (jl l' rI •• k
( _'"'1'_)
--2-' dx .
dx

185
Here we have used the standard notation for the L2 inner product,viz.,

(f , g) f f ( x) g ( x) dx,

and also used integration by parts in writing C ik .

For a given initial value uo(x), the initial values of Ui(t) are
obtained from the following:

2,3, ... ,N-2. (6 )

Equations (5) are solved by using IMSL subroutine DREBS.

In order to compare the results we define the following errors

in the computed solution, U(x,T):

L IU(x,T) - u(x,T)I,

xN ] 1/2
[
f (dx U ( x , T) - u ( x , T) ) 2 (7 )
Xo

where T is the time for which the solution is evolved and u(x,T) is
the exact solution. For the comparison of different schemes, we use
the same initial data and other parameters as used in the papers cited
in the Introduction.

Table gives the Loo and L2 errors in the present scheme and compares
them with four other methods we have referred to previously.

3. CONCLUSION

It is seen from Table 1 that the L2 error in the Galerkin scheme used
here is almost an order of magnitude less than that in the other schemes.
L error 1s also found to be somewhat less. This can be attributed
00

to the use of smoother trial and test functions. We have used the
smoothest functions of degree 3. Here we may point out that Alexander
and Morris [5] used quintic boundary polynomials especially constructed
to maintain C2 -continuity in addition to cubic splines as trial and test
functions. The use of these boundary functions does not appear to
be strictly necessary in view of the fact that the solution goes to
zero at the boundaries. In fact Table 1 shows that, if anything,

186
 Table 1

Comparison of Different Numerical Schemes for the KdV Equation

Error x 10 3

Zabusky-Kruskal Greig-Morris Alexander-Morris Sanz-Serna-Christic Present Nethod

(q = 0)* ~Petrov-Galerkinl

T
L ~
L2 L~
L2 L~ L~ L2 L~ L2

.0.5 t 63.5 122.7 67.4 122.4 57.0 51.9 102.5 37.0 16.1

1.0 161. 4 296.2 141.6 226.1 100.4 150.6 52.7 24.7

1.5 50.7 30 .. 4

2.0 53.6 31.6

*q is a parameter used by Alexander-Morris.

tAlexander-Morris L~ error is for T = 0.3956.

lesser Loo error results if we omit these higher order boundary functions.
Finally we note that the computation time Eor the above scheme was
37 to 120 seconds Eor T = 0.5 to 2.0 seconds respectively.

REFERENCES

[I] R. K. Dodd et al., Solitons and Nonlinear Wave Equations (Academic,

New York, 1982).
[2] N. J. Zabusky and M. D. Kruskal, Phys. Rev. Lett. IS (1965) 240.
[3] I. S. Greig and J. L. Morris, J. Compo Phys. 20 (1976) 64.
[4] B. Fornberg and G. B. Whitham, Phil. Trans. Roy. Soc. 289(1978)373.
[5] M. E. Alexander and J. L. Morris, J. Camp. Phys. 30 (1979) 428.
[6] J. M. Sanz-Serna and I. Christie, J. Compo Phys. 39 (1981) 94.
[7] A. R.Mitchell and S. W. Schoombie, in Numerical Methods in
Coupled Systems, ed. R. W. Lewis (Wiley, New York, 1984).
[8] J. T. King, Introduction to Numerical Computations (McGraw-Hill,
New York, 1984).
[9] C. A. J. Fletcher, Computational Galerkin Methods (Springer,
Be r lin, 1 984 ) .

187
K-dV Like Equations with Domain Wall Solutions
and Their Hamiltonians
Bishwajyoti Dey
Institute of Physics, Bhubaneswar 751005, India

We consider K-dV like equations with higher order nonlinearity and show
that these have domain wall (kink) solutions for particular values
of the coefficient of the nonlinear terms. The solutions are compared
with the domain wall solutions of relativistic field theories. The
exact Hamiltonian densities are also evaluated for these equations,
using Dirac's constrained Hamiltonian formalism. The conservation of
the Hamiltonians is explained in terms of the contribution of the corres-
ponding fields from spatial infinities.

We consider certain nonlinear partial differential equations which

are Korteweg-de Vries (K-dV) like equations with higher order nonlinea-
rity. These equations are [1,2]

a,o >0; n 1,2,3, ... (1 )

b,o > 0, (2 )

which can be derived respectively from the Lagrangian density

ab
a
1.. l8 8
2 x t
+ (3)
(n+1) (n+2) (2n+1) (2n+2)

and

21 8 8
x t
+ l-
12
b8 4 + l08 2
x 2 xx
(4 )

where u 8x ' and the subscripts denote partial derivatives.

In order to look for travelling wave solutions we make the simple

transformation

x - ct (5 )

188
where c is the velocity of the solitary waves. Integrating equations
(1) and (2) we get respectively the domain wall solutions of these
equations as [1,2]

c(n+l}(n+2}
{ }1/2 1
u(x,t} [ 1 ± tanh(c/6}2 .!2.I]l/n for n = 1 ,3,5, ...
2a 2 '
(6a)
c(n+l} (n+2) }1/2 1
u(x,t} ±{ [ 1 ± tanh(c/6}2 .!2.I]l/n for n=2, 4,6, ...
2a 2
(6b)
for
a (2n+l )
b (7 )
c(n+l} (n+2}2

and
1 1
u(x,t} ±(3c/b)2 tanh[(c/26)2(s+k)]. (8 )

It can be noted that the solutions (equations (6)) of equation (1)

resembles the solution of relativistic field theories with potential
[ 3]

(9 )

Similarly equation (8) (solutions of equation (2)) resemble the kink/

4
antikink solutions of A~ relativistic field theory.

A conservation law associated with a K-dV like equation is expressed

by an equation of the form

o I ( 10)

where T the conserved density and -X, the flux of T, are functions
of u(x,t). The K-dV equation (b = 0 and n = 1 case of equation (1))
has infinite number of conservation laws associated with it. However
for equation (1) we could write only first two conservation laws {see
[1] for n = 1 and 2 cases} and we are currently trying to find other
conservation laws (if any) which are not obvious. On the other hand,
equation (2) is a more interesting case, as one can write many conser-
vation laws for this equation [1]. The third conservation law associated
with K-dV type equations, usually describes the conservation of Hamil-
tonian. For example the Hamiltonian for equation (2) is conserved
by the third conservation law associated with this equation. However
for equation (1) we could not write the third conservation law. This

189
led us to investigate the Hamiltonian nature of this type of K-dV like
equations. It may be noted that the Hamiltonian density for K-dV like
equations is not obvious, as these belong to degenerate Lagrangian
system [4]. There are constraints in the system and one has to use
Dirac's theory of constraints [5], for evaluating the correct Hamiltonian
density.

Now to avoid higher derivatives in the Lagrangian we introduce a

new field ~(x,t) and write the Lagrangian for equations (1) and (2)
respectively as

a ab
en + 2 + e 2n + 2
x
(n+1)(n+2) x (2n+1) (2n+2)

( 11 )

and

( 12)

where as before u(x,t) ex .

Considering independent variations with respect to e(x,t) we get from
equations (11) and (12) respectively

o ( 13)

and
o ( 14)

while independent variations with respect to ~(x,t) give for both

equations (11) and (12)

o. ( 15)

Thus equations (13) and (14) together with u = ex and equation (15)
gives equatio~s (1) and (2) respectively. The canonical momenta are

o (16a)

1T
e = ~
2
ex ( 16b)

Thus these are degenerate Lagrangian systemsas the canonical momenta

for the field ~ is zero. So we use Dirac's theory [5] of constraints
for evaluating the exact Hamiltonian density for these systems. The
primary constraints

190
 "" 0 (17a)

lie - 1e
2 x
"" 0 ( 17b)

(where "" denotes weak inequality) satisfy the relations

0, (18a)

{C 1 (x), C 2 (x')} 0, ( 18b)

and

(18c)

where 6(x) denotes Dirac delta function. The symbol { , } denotes

the Poisson bracket. The fields have their usual relationship

6 (x-x') (19a)

and
6 (x-x'). ( 19b)

Equations (18) show that the constraints are second class. The total
Hamiltonian is defined as
00

f (~o + 1(1 ) dx J (20 )

where the free part of the Hamiltonian density is given by

'Ito 11
e e t -f.. (21 )

and

-;It 1 (22 )

where the Lagrange multipliers A and

have to be determined from a
the condition that the constraints are maintained in time, i.e.,

o. (23 )

This condition requires an extra constraint condition, which we denote

by a secondary constraint X, and~l is thus modified as

"1 (24 )

191
where ~ is another Lagrange multiplier. The Lagrange multipliers
are determined as [2],

ab
a
(2n+1 )
J
- 68 xxx (25 )

for Lagrangian in equation (11) and

a = (_ 1 b 8 3 + 68 ) (26 )
"3 x xxx

for Lagrangian in equation (12), and the multipliers

8 (27 )
xx
and

a (28 )
xx

The total Hamiltonian density is now obtained from equation (20) using
equations (17), (21), (24)-(28) which when evaluated for the Lagrangions
in equations (11) and (12) gives respectively, upto a surface term,

na nab
__________~On+2 +
a;J/T
2(n+1) (n+2) x (2n+1) (2n+2)

2 + a8 n 8-
-n ljJ (68 5x + na8 xn-1 8 xx x 3x

(29 )

where 8 3x denote 8 xxx ' etc., and

1 b84
TI x

(30 )

To check that the Hamiltonian densities obtained are correct ones,

we obtain the field equations (1) and (2) using the Hamiltonian equations
of motion

~(x,t) ( 31 )

192
 Now to examine whether the Hamiltonian is a constant of motion it
is sufficient to consider only the free (~) part of the total Hamilto-
nian density [5]. Thus the Hamiltonian H when expressed in:'rms of
the original field u(x,t) gives for equation (1)

a
n+2 ab u 2n + 2] (32 )
H u
(n+1) (n+2) (2n+1) (2n+2)

and for equation (2)

H (33 )

It is known that for a system, represented by K-dV type equation

ou + F"(u)u (34 )
xxx x

which can be derived from the Lagrangian density

(35 )

the Hamiltonian is given by

H (36 )

The Hamiltoniam (32) and (33) for our systems (equations (1) and (2))
agree with equation (36). However it is to be noted that for describing
the correct dynamics of the systems one has to use the total Hamiltonian
density given by equations (29) and (30).

Now, the Hamiltonian H is a constant of motion if

dH o. ( 37)
dt

For the Hamiltonian in equation (32) we get

dH
dt -f dx [""(i1+T)
1 n n
au n+1(o u 3x + a(l+bu )u u x ) + (2n+1)
ab u 2n+1("uU 3x

+ a(l+bu n )u n u ) - o2u u - aonu 3 u n - 1 - aou u

x x 4x x x xx

- 2nabou 2n - 1 u 3 - abou 2n u u ].
x x xx

193
Integrating by parts and using the fact that derivatives of u(x,t) are
zero at spatial infinities (see equations (6)) we get
2 2 00
dH a 2n+2 a b 3n+2 4n+2
dt [ u + u ]
= 2(n+1)2 u + (n+1) (2n+1)
_00
(38 )

which is not zero as the contribution from spatial infinities do not

cancel, since the field configuration corresponding to equation (1)
is asymmetric (equation (6)). Thus the Hamiltonian corresponding to
equation (1) is not a constant of motion. This explains why we could
not get more than first two conservation laws for this system. However,
for the Hamiltonian (equation (33)) of equation (2) we get

dH
dt I dX(~ b 2u 5ux - ~ bcu 3 u 3x + 2bCUU~ + bcu 2 u x u xx

Integrating by parts and using the fact that derivatives of u(x,t) are
zero at spatial infinities we get (see equation (8))

dH
(39 )
dt

which is zero, as the field configuration corresponding to equation

(2) is antisymmetric. Thus the Hamiltonian corresponding to equation
(2) is conserved. It should be noted that, this Hamiltonian is also
conserved by virtue of the third (1) among the many conservation laws
satisfied by equation (2).

In case of relativistic field theories however such problem does

not arise, where it can be shown by a simple calculation, that, if
the Lagrangian does not contain explicit time dependence (as in equations
(3) and (4)) then the Hamiltonian is conserved, even if the field con-
figuration is nonzero at spatial infinities (but its derivatives are
zero) .

REFERENCES

(1) B. Dey, J. Phys. A: Math. Gen. 19 (1986) L9.

(2) B. Dey, to be published.
(3) B. Dey, J. Phys. C: Solid State Physics 19 (1986) 3365.
(4) Y. Nutku, J. Math. Phys. 25 (1984) 2007.
(5) E. C. G. Sudarshan and N. Mukunda in Classical Dynamics: A Modern
Perspective (Wiley Interscience, New York, 1974).

194
 Part III

Lattice Solitons
Lattice Solitons and Nonlinear Diatomic Models
P.C. Dash
Department of Physics, College of Basic Science and
Humanities Orissa University of Agriculture and Technology,
Bhubaneswar PIN 751003, India

The Fermi-Pasta-U1am problem together with the explanation by Zabusky

and Kruska1 can be rightly considered as the origin of lattice soli-
tons. This problem is reviewed in some detail along with a nice integ-
rable nonlinear lattice, the Toda lattice. The recurrence phenomenon
in case of KdV system and FPU discrete limit is also discussed. Three
diatomic nonlinear lattice models as well as their solutions are con-
sidered. These are the simplest cubic nonlinear model in continuum
limit, diatomic Toda system and continuum model with nonlinear onsite
potential at one of the mass points and harmonic potential at the other,
connected by harmonic springs.

1. ORIGIN OF LATTICE SOLITON

Though solitary wave was first discovered in 1834, the present upsurge
of interest in solitons is mainly due to the attempts made to explain
a nonlinear lattice problem, the Fermi-Pasta-Ulam recurrence found from
computer experiments in 1955. Its explanation by Zabusky and Kruskal
signalled the birth of lattice solitons. I intend to discuss in some
detail this recurrence phenomenon not only because it supplied a major
impetus to the development of soliton-physics but also it combines past
excitement with present vigour.

2. FERMI-PASTA-ULAM PUZZLE

The equation of motion for a chain of mass points interacting through

a potential ~(r) can be written as

(1 )

where the force.f( r) = -~, (r) and r n = y n+ 1 - Yn and dot represents der i-
vative with respect to time and prime stands for spatial differentia-
tion with respect to the argument. Linear lattice: ~ (r) = ~y r2

196
so that

{2 }

Solutions of eq. {2} with fixed boundary conditions {for n 0,

n = N+1, Yn = O} can be represented by

{3a}

with

2/YTm 1,2, ••. ,N. { 3b}

Now one can introduce normal coordinates

{ 4a}

with

w
r 21YTm {4b}

such that energy

E { 4c}

The natural motion of the harmonic lattice can be expressed as a

superposition of these normal modes. The energy of each normal mode
remains always constant. It was widely believed that a mild nonlinearity
should bring the system to a state of statistical equilibrium. To veri-
fy this widely believed conjecture Fermi, Pasta and Ulam [1] considered
a chain of masses and varied their number from 16 to 64. The nonlinear
potentials considered were

<P{r} 1 2 1 3
"2 yr + "3 yar {5a}

1 2 1 4
<P {r}
"2 yr; + "4 Byr { 5b}
1 2
"2 Y1 r Ir I < d,
<P{r}
[, 2
"2 Y2 r + or Ir I > d. { 5c}

Their aim was to obtain the energy distribution among the normal
modes of the systems in the presence of weak nonlinearity and to deter-
mine the time of relaxation to equilibrium. The result they found
from computer experiments was least expected: if the initial data assig-

197
ned all the energy to the lowest mode, only a few modes were excited
as time went on and almost all energy was eventually given back to
the lowest mode (Recurrence phenomenon). The aftermath of this finding
is really an exciting chapter in the history of soliton-physics. I
shall briefly discuss the two serious consequences of FPU puzzle:
(a) How to explain the recurrence?
(b) what happens to statistical equilibrium?

After FPU, extensive studies were carried out with the intention
of verifying and explaining the recurrence phenomenon. The conclu-
sions are as follows [2]: Recurrence time

(6a)

with fixed boundary condition Yo o and

B 71n ),
sin(N y! o (6b)
Yn!t=O n t=O

and tL is the linear period defined by

1
2N/(y/m)'. (6c)

3. EXPLANATION OF RECURRENCE

Zabusky in 1967 first showed that the continuum limit of FPU lattice
was the Korteweg_de Vries (KdV) equation. This was a major breakthrough
and signalled the birth of lattice solitons. It also became very use-
ful for providing an explanation to FPU recurrence. Subsequently,
the continuum approximation to lattice problems are used in many con-
texts because (i) continuum approximation is easier for analytical
as well as numerical study than its discrete counterpart; (ii) results
can be conveniently related to the discrete version in many cases;
continuum approximation is physically acceptable when wavelength is
very large compared to spacing of particles in a lattice. To illustrate
a continuum case I choose the following example which is not only the
first historical model but contains alII the essential features of
any nonlinear lattice problem in long wavelength limit.
Let

~ (r) (7 )

198
then

(8 )

In continuum approximation wavelength being large compared to spacing

of particles, wave is very smooth so that one can make the following
Taylor expansion:

= y+h f l + -h
a 1 2 :\2v
.£.......L + •.• , (9 )
- aX 2 ax2

where h is natural length of a spacing in lattice and y(x)

2
nh. Substituting eq. (8) in (7) and keeping terms upto h

( 10)

where c h/YTm.

For h '" 0, linear wave equation (lOa)

Keeping terms - h,

(lOb)

This is hyperbolic equation, whose solutions become discontinuous

-1
after a time -(ahyoc) where Yo is the maximum amplitude at t = 0
(obtained numerically), that is, the solutions break down. Zabusky's
conjecture is that this is not the solution in discrete case and hence
2
not physical. So, considering terms h , we have

(10c)

with E = 2ah.

Let us consider waves travelling to the right:

1
u - ~
a~'
~ x - ct, T E*t and E* "2 EC

a a a a - a
so that
ax - ~' at -
E*
at c ~.

199
Noweq. (10c) takes the following form,

0, ( 11 )

with 11 h/24a.

This is the Kortweg de Vries equation as obtained by Zabusky. It

will not be out of place to mention that even when the potential is
of exponential type, the continuum limit of the discrete case is also
the KdV equation (Toda).

Recurrence phenomenon is now known to be due to the motion of soli-

tons which carry energy but the first explanation in this line was
advanced by Zabusky and Kruskal [4] using the continuum version of
the lattice equation of motion (11). FPU considered the initial condi-
tion

B sin 1Tn (12a)

Ynlt=o N

O. (12b)

From (4a) the corresponding normal coordinate

N
2 B sl'n 1Tn ,1Tr B~ ( 13)
N+1 1: N sln N+ 1n = u1r·
n=l

From (12b) there is no initial kinetic energy, so total energy from

(4c)

~Bw~ (lowest mode is only excited).

Equation (12a) represents a sine-wave which remains confined in the

lattice. This can be thought of as a stationary wave which can be
approximated,by the superposition of two progressive waves of half
amplitude:

( 14)

Writing ~ = nh = x, T ~ Ect, u = ay/a~ and considering a cyclic

lattice of 2N particles one can find in the continuum limit

( 15)

200
with boundary condition

u(x) u(x+2) (16 )

where A Bn/2.
o

4. RECURRENCE IN KdV CONTINUUM

KdV equation (11) has a solution (soliton)

2
u Uoo + a sech S(s - c,) (17 )

where u and a are constants (a > 0) ,

a
c u 00 + "3 and S = ,.&7TT11·

Further with KdV equation (11) is associated an eigenvalue equation

lcp
611-- + (A- U)cp o ( 18)
a(
where U = -u, such that if u develops in time as in KdV, then A is
independent of time. Equation (18) is analogous to a Schrodinger eigen-
value equation with 0 2 , the Planck's constant being replaced by 1211.
As A is independent of time we can use ul T=O = 0 and find out eigen-
values which will remain same for all time. For a single soliton

a
A = u - "2 and c = constant - ~ A. (19a, b)

For different eigenvalues, different 'a' values can be obtained and

so, each eigenvalue of eq. (18) is associated with a soliton. Now
from eq. (15) expanding near the bottom of the potential-well

U A
0
+ 1 A n2 2
"2 o. s (20a)

and
1 2 (20b)
An = - A0 + (n+2") .; 1 211A 0 n

where n = 0,1,2, ...

Using eq. (19b) the velocities of the solitons associated with An form
.an arithmatic series with common difference

( 21 )

201
These are the solitons which move independently because actually these
solitons travel, interact as they collide and pass through one another
(here we neglect acceleration during interaction). When solitons move
in the periodic region (length = 2), the same configuration of solitons
as the initial one will come back again after a time interval

2
'R 0.364/v\IA O (22)
f1c

when A
0
1, III 0.0222

40/n . (23a)

Numerical experiment with the same values of Il and Ao gives

30.4
(23b)
n

The discrepancy in between (23a) and (23b) may be ascribed to change

in velocities during interaction.

4.1 FPU Recurrence

From numerical experiments Zabusky obtained the recurrence time for
FPU lattices as

(24a)

(24b)

using expressions ,= ~ €ct, € = 2ah, c = h/y/m, Ao = Bn/2 and

Il = h/24a in eq. (22), theoretical estimate for FPU discrete case gives

(25 )

The discrepancy between (25) and (24a) is still less.

This gives a very good account of the recurrence phenomena in both

discrete and continuum limit.

4.2 Statistical Equilibrium

Ford et ale and Saito et ale [2] advanced reasonable explanation to

the problem of statistical equilibrium, energy sharing, ergodicity
and the equipartition. It is now clear that there exists a critical

202
value a c of the coupling constant a which determines ergodicity. For
a >a c ' the system will be stochastic and energy sharing would take
place. If a <a c ' the recurrence phenomena may occur and one may find
nonlinear normal modes which is a consequence of Kolmogorov, Arnold
and Moser theorem. Here again as suggested by Ford a resonance relation
exists: r nkw k = 0, n k being integers, all not equal to zero. When
this condition is satisfied for small nonlinearity case there will be
energy sharing, otherwise recurrence phenomena will occur. Further
if the initial state is far from a normal mode the resonant nonlinear
system exhibits rapid energy sharing and equipartition of energy is
readily established.

5. TODA LATTICE

Previous sections give an account of the origin of lattice solitons

but the most remarkable model for their study is the Toda monatomic chain.
With the nearest neighbour interaction the Toda chain happens to be
the only integrable nonlinear model [3]. Here the nearest neighbour
interaction potential is given by

4l(r) ba e -br + ar + const. (ab > 0) (26 )

and the equation of motion becomes

( 27)

where
-br
a(e n - 1) (28)

or equivalently,

(29 )

The exponential potential (26) includes the linear harmonic case when
b ~ a and strongest nonlinear case of a system of hard spheres if b~oo.

The equation of motion (27) or (29) admits exact M-pulse solutions

whose form for M = 1 is given by

(30)

or with

203
 (31 )

(m/b) log {1+exp[2(an+Bt+6)JL (32 )

1
where B = (ab/m)2 sinha. (B/a) gives the speed and eq. (30) is the
single pulse solution. The M-pulse solution can be obtained in closed
analytic form and is represented assymptotically (for t ~ + 00) by

22+
(m/ab)B· sech (a.n+B .t+6.) (33 )
J J J J

with j 1 , 2 , ••• , N and constants 6. are related by

J

H-: 2: 6~. (34 )

j J j J

This last expression represents conservation of momentum.

These pulses move almost independently in the lattice. They emerge

after collisions with same shapes and velocities. So they behave like
particles and are called solitons or lattice solitons. Further it
can be seen from eq. (32) that when the soliton moves in the lattice
with a constant velocity it causes a contraction of the lattice.

In addition to this M-soliton solution the Toda lattice admits a

nonlinear periodic solution known as Cnoidal solution:

2 2 n E
(m/ab) (2kv) [dn {2('I - vt)K - k } J I ( 35)

where K and E are elliptic integrals of the first and second kinds, and
dn is an elliptic function (Jacobian).

If the modulus k is small, k ~ 0, a cnoidal wave reduces to a sinu-

soidal wave,

(36 )

w 2(ab/m) 1/2 sin(n/A) . (37 )

The cnoidal wave (eq. 35) can be written as

00
2 2
exp(-br n ) - (m/ab) [ 2: B sech {a(n-A~)-Bt} - 2BvJ (38 )
~=-oo

with a nK/K I ,
B nKv/K I •

204
Equation (38) represents an infinite sequence of solitons at equal
intervals A and shifted downwards.

As a mathematical model for lattice-soliton Toda chain no doubt

occupies a unique place but its applications in different fields (like
wave propagation in nerve systems, ladder circuit, chemical reactions
in atoms and molecules, and ecological systems) make it very important
and interesting from physical point of view.

6. NONLINEAR DIATOMIC MODELS

Lattice solitons are extensively studied using monatomic models in both

discrete and continuum limits. One of the applications of these studies
is to explain certain important characteristics of structural phase
transitions. As many of the solids undergoing displacive type phase
transitions have a diatomic structure along (100) symmetry direction,
the study of diatomic nonlinear models attracts much attention. Besides,
some of these nonlinear diatomic lattices are very helpful in explaining
nonlinear flow of heat in solids.

In the literature mainly two types of models are available for non-
linear diatomic cases: one deals with nonlinear interactions between
nearest neighbours [4-6) and the other with nonlinear onsite potentials
connected by harmonic springs [7,8). Study of diatomic Toda chain [6)
happens to be the first attempt in arriving at an exact solution to a
discrete nonlinear diatomic model. Now it is becoming gradually clear
that diatomic Toda chain represents a non integrable system [9). How-
ever, very recently non integrable rational billiard systems are found
to be analytically tractable in terms of Fourier expansion [10) and
so the earlier study of diatomic Toda chain with the help of Fourier
series deserves some special mention. On the other hand, Buttner and
Biltz [7) reported exact solutions to a lattice with nearest and a
next nearest inte!action, having a nonlinear ~4 onsite potential. I
wish to consider here the following three nonlinear lattices which
will involve all the techniques and characteristics of the available
models.

6.1 A Continuum Diatomic Model of First Kind

A large number of studies on diatomic models include nonlinear inter-

action between nearest neighbours. These are discussed in continuum
or long wavelength limit using a procedure by which solutions can be

205
separated into optic and accoustic modes. The following nonlinear di-
atomic model is chosen to illustrate the main features of this type
of lattices. Here we consider a chain of alternate mass points m1 and
m2 connected by nonlinear springs, with potential energy,

~(r) (39)

and equation of motion for displacements Y1n and Y2n of atoms m1 and
m2 of the n-th cell can be written as

2
Y
m1 1n k 2 (Y2n - Y1n) + k 2 (Y2n-1 Y1n) + k 3 (Y2n - Y1n)

2
- k 3 (Y2n-1 - Y1n ) , (40 )

- 2
Y
m2 2n -k 2 (Y2n - Y1n + 1 ) k 2 (Y2n Y1n) + k 3 (Y2n - Y1n + 1 )

(41 )

To solve in the continuum limit the expansions used are

Ypn ± 2hy' + 2h 2 y" ±! h 3 y m + . . . (42)

pn pn 3 pn

(43 )

Consideration of a harmonic lattice using these expansions suggests

that a = 1, corresponds to accoustic mode and a = -m 1 /m 2 to optical
case. Now substitution of equations (42,43) in (41) yields (with
appropriate choice of variables), the following KdV equation for a=l:

o. (44)

Therefore in the accoustic region KdV type pulse solitons are obtained.
Now for a = -M 1 /M 2 as in [4] we obtain travelling wave solutions of
the form exp(±ie) whose amplitude satisfies the nonlinear Schrodinger
equation:

p p p
+ --21(~ + -----2
N ) IAI2A + 2w1 A o. (45 )
w D2 v - 1
q

It describes the motion of the amplitude as nonlinear wave modulation.

206
6.2 Diatomic Toda Lattice

Using the notations of Section 5 and putting constants a b ,,

equations of motion for diatomic Toda chain becomes [6]

(46 )

We look for periodic solutions in the following form because this form
exists in the harmonic limit as well as in the equal mass limit.

L a. exp[i2TIj(vt + 2n)] (48)

j=-oo J "

L b. exp[i2TIj(vt + 2~-' )]. (49 )

j=-oo J

Substituting expressions (48,49) in (47) and then integrating over

a time period we get

(50)

where

M i2TIv(m,+m 2 ) L
j=-oo
ja .a, . - i2TIV L
J -J j=-oo
ja.b, .e('-j)
J -J
(5' )
N i2TIv(m,+m 2 ) . L jb .b, . -i2TIV L jb .a, . e(j-'), (52 )
J -J j=-oo J -J
J=-oo

with e(j) m, exp(-i2TIj/,,) + m2 exp(i2TIj/,,). (53 )

Equation (50) can be solved for getting the coefficients of the Fourier
expansion, so that
2 00 2n j 2
4m,m 2 sin ~ L ~ sin2TIj(vt + ~)
a - 2 TIV " j=' '_q2j "
o (54 )

00 i
L ~ sin2TIj(vt + 2n-')
j=' ,_q J "
• ( 55)

207
6.3 Henry and Oitma Diatomic Model

Let us consider a diatomic chain of harmonically coupled nearest neigh-

bour atoms m, and m2 including a nonlinear ~4 potential on mass M,.
The Hamiltonian is given by

2' 2
H ( u i - v i _,) + 2 y ( u i -v i) + V( u i ) 1 ,
(56 )

where u i ' vi are displacements of the two types of atom in the i-th
unit cell. V(u i ) is the nonlinear onsite potential at lattice points
with mass m1 • With the following prescription, equations of motion
can be obtained in continuum limit in the displacive regime:

ui -+ u(x,t) -+ u

a
vi -+ v(x + 2' t)

a
v i _, -+ v(x - 2' t)

Jdx (57 )
1: -+
a
i

(58 )

(59)

(a) If V = 0, the equations reduce to continuum equations of motion

for a harmonic diatomic chain.

(b) If V * 0, but instead is given by a ~4 nonlinear single site

potential then the following three types of excitations are obtained
as exact solution to the field eqs. (58) and (59).

(i) Linearize'd periodic solutions

Here
V(u) (A, B > 0) (60 )

with potential minima at ±u o (= ±A/B). These linearized solutions

are low energy phonons and represent oscillation of m1 atoms in one
of the double well potential (±u o ) and oscillation of m2 atoms about
±u o from their equilibrium position.

208
 u (61 )

v (62 )

with the dispersion relation

W~(ir

(63)
(ii) Solitary wave solutions (large amplitude solutions)

Equations (61) and (62) represent low amplitude solutions. Besides,

the field equations (58) and (59) represent large amplitude solutions
also. v = u = iu o is the simplest large amplitude solutions with lowest
energy. It is the ground state and is taken as the reference level.
Other solutions are regarded as excitations above this level:

+00
dx
E f a
-00

Seeking, now, solutions in the form u(x,t) f(s), v(x,t) g(s) with

::ll ::i } [
s = x-ct equations (58) and (59) becomes

~2 l
+ 2yf +
av
at
[
+ 2yf

For c c ,where c lya 2 /4m2

o· 0
f
0
2
1
df
g(s) fo (s) , s = [c o (m 1 - m )]2 f I
2
[c -2V( f) ]2

For a ~4 potential depending upon the values of the integration

constant c equal displacement field solutions can be obtained in the
form of a tanh-kink or a sech-pulse.

(iii) Nonlinear periodic solutions

For c = co' the displacement fields are not equal and solutions may
become extended instead of localized. The exact solutions can be written
as

209
 f fo sin(ks), k

g go sin(ks) + gl sin(3ks)

with

f
o

2 2
(2y ~ )f /(2y
4 0

1/2 1
2y } w ck [(2y)/(9m 2 )]2
c

The linearized periodic solutions are low energy phonons whereas

these nonlinear periodic solutions are high energy phonons. For dis-
crete diatomic case, B~ttner and Biltz [7] found "periodon" solutions
and the above nonlinear periodic solutions are shown to be the long
wavelength (continuum) limit of these solutions.

7. CONCLUSION

In conclusion, it may be stressed that the history and development

of the soliton concept is intimately connected with the studies on
lattice solitons. Now it is known that further researches will help
understand among other things, some unexplained facts in the field
of ferro-electrics, conformational change in biological molecules and
chopping phenomenon. However, some challenging areas of research in-
clude mono and diatomic nonlinear lattices in higher dimensions [11],
scattering and destruction of solitons by impurities (thermal conduction
[12]), onset of chaotic behaviour (that is, how solitons and chaos
compete or compromise) [13], effect of perturbation on solitons and
quantization (14].

REFERENCES

[1] E. Fermi, J. Pasta and S. Ulam, Collected Papers of Enrico Fermi,

Vol. II (University of Chicago Press, 1965), p. 978.
[2] A good review of the recurrence phenomena can be found in the
review article by M. Toda, Phys. Rept. laC (1975) 1; Theory of
Nonlinear Lattices, Springer Series in Solid State Sciences,
Vol. 20 ( Springer, Berlin, 1981).

210
 [3] K. Sawada and T. Kotera, Proc. Theo. Phys. Suppl. 59 (1976) 101.
[4] P. C. Dash and K. Patnaik, Proc. Nucl. Phys. and S. S. P. Symp.
(India) 21C (1978) 483; Prog. Theor. Physics 65 (1981) 1526, and
references therein.
[5] H. Buttner, and H. Biltz, in Solitons and Condensed Matter Physics,
ed. A. R. Bishop and T. Schneider (Springer, Berlin, 1978),p.162.
[6] P. C. Dash and K. Patnaik, Phys. Rev. A23 (1981) 959.
[7] H. Buttner and H. Biltz, in Recent Developments in Condensed
Matter Physics, ed. J. T. Devreese (Plenum, New York, 1981),
Vol. 1, p. 49.
[8] B. I. Henry and J. Oitman, Aust. J. Phys. 36 (1983) 339.
[9] B. Kostat, Adv. Math. 34 (1979) 195.
[10] P. J. Richens and M. V. Berry, Physica 2D (1981) 495.
[11] H. Buttner, H. Frosch, C. Behnke and H. Biltz, Springer Series in
Solid-State Sciences, 47 (1983) 281, and references therein.
[12] G. Casati, J. Ford, F. Vivaldi and W. M. Visscher, Phys. Rev. Lett.
52 (1984) 1861 and ibid 53 (1984) 1120; F. Mokross and H. Buttner,
J. Phys. C16 (1983) 4539.
[13] Till now there are no serious attempts in lattice models for study-
ing the coexistence of chaos and solitons. However in differential
equations some studies are made in this direction. A good and
brief review is given by A. Bishop in Springer Series in Solid-
State Sciences 47 (1983) 197 and Y. Imry, ibid 47 (1983) 170.
[14] N. Theodorakopoulos and F. G. Mertens, Phys. Rev. B28 (1983) 3512.

211
Recent Results in Toda Lattice
Z. Popowicz
Institute of Theoretical Physics, University of Wroclaw,
ul. Cybulski ego 36, PL-50-205 Wroclaw, Poland

It is shown how different generalizations of Toda lattice problems occur

in the context of eighteenth and nineteenth century mathematics. A
possible classification and applications of different generalitations
of the Toda 1 a t tice are gi ven, with special a t ten tion to gauge theory.

The topic of this paper will be the following system of equations

( 1)

where ~n = ~n(x,y) and nEZ, and its different generalizations which

appear in the physics and mathematics. For our purpose, it will be
convenient to define the system (1) in a different manner as

r r
2e n + e n-l (2 )

±k h
nm m
e (3 )

where rn ~n+l
- ~n and h = hn(x,y) ,
n

[-;
for n = m
for n = m±l (4 )

for rest

and for x = y as

9.....M
dx n
(5 )

(6 )

where Mn

Equations (1-6) are known to the physicist as the Toda lattice.

Toda defined equation (2) for x = y in 1967 [1,2] considering a one-

212
dimensional lattice, which consisted of N particles of unit mass inter-
acting through the potential
-r
4> ( r ) e + r. (7 )

Toda considered such a potential because it admitted analytic solutions

and it generalized the famous computer experiments of Fermi, Pasta
and Ulam [3]. Fermi, Pasta and Ulam considered a finite number of pen-
dula arranged in a line interacting with their nearest neighbour via
anharmonic forces. The system was started by displacing the end pen-
dulum with the other at rest. Soon all were moving but after a finite
time, the initial situation recurred. This meant that the system was
not ergodic. The same situation has been obtained for Toda lattice [4].

After introducing the Toda lattice its importance was quickly and
widely recognized by physicists. It appears that the Toda lattice
and its generalizations can describe different physical phenomena and
are contained among the soliton equations. These facts have given a
strong impetus for a deep investigation of these equations.

Surprisingly, recently [5-8] it has been noted that the Toda lattice
in the forms (1)-(6) has been "known" for the mathematicians of the
eighteenth and nineteenth centuries. Here one should specify the meaning
"known" and distinguish those mathematicians who used that system cons-
ciously [9] and those who were close to define it [10,11]. I do not
define here the meaning of "known". It is the problem for the histo-
rians and philosophers, similar to the question "Did ancient Greek
know the differential calculus?" On the other hand, I would like to
apply the Toda lattice to the eighteenth and nineteenth century mathe-
matics using the ideas of Euler and Sylwester.

First let us note that Euler solved the following linear differential
equation

g a
o
9 + b
0
g, (8 )

where g g(x), a o ' b o are arbitrary functions of x, differentiable

=
infinite times, and

~ -~
- dx'
..g = [ g• ]2 , ., (9 )

without using the series expansion of g. His method known now as the
Euler method is based on the following trick. Let us write down
equation (8) as

g/g = b + a (~/g)-1 ( 10)

o 0

213
Using equation (8) we can compute

( 11)

and similarly for

b + a ( 12)
n n

where the functions an' b n could be computed explicitly from (8),

differentiating n times. Substituting (11 )-(12) into (10) we obtain

a a
(~ In g)-l b + n
0 ( 13)
dx o b 1+ b n + 1+

In this way, Euler was able to represent g as a continuous fraction

[12], which is denoted by us by the use of the special notation in (13).
On the other hand, the investigation of the convergence of the continuous
fraction is equivalent to the investigation of the corresponding series.
It may be that this was the reason why Euler stopped these investigations
at this stage. Following the line proposed by Common and Roberts [8],
let us apply the Euler method to the Riccati equation

( 14)

where Eo' Fo are given functions of t. In [8], the authors proved the
following theorem.

Theorem If gk-1 satisfies the Riccati equation for some Ek _ 1 ,F k _ 1 ,

then gk ' defined by

( 15)

where Uo is an arbitrary constant and k 1,2, ... ,satisfies the Riccati

equation for E k , Fk if

Mk Ek - 1 + Uo F k _ 1 + u0'
2 ( 16)

. -1
Nk MkMk - (F k - 1 + Uo )' (17 )

Ek Nk + NkU o + Mk , ( 18)

Fk -N k Uo' ( 19 )

214
Note that the initial functions M1 and N1 are determined from the coeffi-
cients of the original Riccati equation (14) by using (16) and (17)
with k = 1. If we eliminate the functions Ek , Fk and the coefficient
Uo from (16-17), we obtain Mk , Nk satisfying the Toda equations (5-6),
where now n = 2,3, .... In this way, we have obtained the solution of
the Riccati equation (14) in the form of continuous fraction (15) which
is constructed from the solutions of the Toda lattice. It is surprising
that the "old is so new."

Quite a different application of the Toda lattice, with reference

to the nineteenth century mathematics, can be constructed using a work
of Sylvester (11). Note that Sylvester is also well known for coining
many new words or jargons in mathematics, for example, the persymmetric
determinant. He defined it as the determinant which possesses the
same elements on the perpendicular line to the main diagonal. For
example, the persymmetric determinant of third order is defined by

(20 )

Assuming that

( 21 )

Sylvester proved that

(22 )

and in the general case that

Dn+1 (g)D n - 1 (g). (23 )

Putting

DN+1 (g) 1, (24 )

for some N we obtain a complicated nonlinear differential equation for

the function g. Probably, Sylvester would have fallen head over heels
to know that equations (23) and (24) are equivalent with the finite
one-dimensional Toda lattice with the free endpoints in the form (3).
This connection can be easily seen due to the following correspondence.
Let us rewrite the equation (3) for minus sign and for x = y =z as

215
 -2h 1 +h 2
a2zz h1 e (25)
-2h 2 +h 1+h 3
a2zz h2 e (26 )

.
-2h N + h N_ 1
a2zz hN e (27 )

from which we obtain

h1 dt h1 dt
e D1 (e ) D1 (g) , (28)

h2 2h h1 h1
e a2zz h 1 ·e e a2zz e (29 )

(30 )

Using equation (23) we obtain

(31 )

and by recurrence

h
n
e (32 )

for n = 3,4, ... ,N. In order to terminate this recurrence according

to (27), we assume (24). Hence, we have established the above mentioned
correspondence.

Surprisingly the Toda lattice (2) where rn = ln fn can be seen in

the book of Darboux [9). Darboux found the singular solutions of this
lattice in the form

(n-a)(b-n)(x-y) -2 , (33 )

where a, b are arbitrary constants.

In this way, we showed that the different forms of the Toda lattice
can be recovered in terms of the eighteenth and nineteenth century
mathematics. Furthermore, from both physical and mathematical points
of view, it is possible to generalize the Toda lattice in other ways
also. We will try to introduce a possible classification of such
generalized Toda lattices in the following. We call the one-dimensional
system introduced by Toda as the standard Toda lattice, which has been

216
thoroughly investigated. We have three possibilities: finite, infinite
and periodic Toda lattices. Out of these infinite and periodic Toda
lattices are completely integrable Hamiltonian systems [13,14] and can
be solved by the inverse scattering transformation [15] or by the Hirota
method [16] or by the Backlund transformation [4]. The finite Toda
lattice has been considered by Kostant [17] and by Olshanetsky and
Perelomov [18].

One can also generalize the standard Toda lattice in one-dimensional

space to the two-dimensional space-time [19,20]. This model corresponds
to a non-trivial relativistic invariant model in field theory, the so-
called nonlinear Klein-Gordon equation and can be solved by the inverse
scattering method [19] or by the Backlund transformation. Barbashov
and Nesterenko in 1981 [21] showed that the relativistic string model
in a space-time of constant curvature (de Sitter universe) is described
by the system of equations

CPxt
eCP cos1/! - e-1/! , (34 )

eCP sin 1/! (35 )

1/!xt

Interestingly, this system is equivalent with the N = 4 periodic two-

dimensional Toda lattice. Indeed as was shown by Fordy and Gibbons [20]
that in this case the Toda lattice can be reduced to (34-35) if we make
the substitution

1
e '2 cP , 4> = (36 )

after rescaling the variables x and t.

Veno and Takasaki generalized [22] the standard Toda lattice to the
multidimensional case using the idea of Kadomtsev-Petviashvilli hierar-
chies [23]. Its algebraic structure and bilinearization in terms of
the T function and some special solutions were investigated in detail
in [24].

The fourth generalization is a purely theoretical approach where

one generalizes the Toda lattice by including supersymmetry. Here, one
can distinguish between the non-extended [25] or extended [26] super-
symmetric Toda lattice.

In the next approach, one can utilize the connection, discovered

by Bogoyavlensky [27], between the Toda lattice and simple Lie algebras.

217
It appears that for every simple Lie algebra one can associate a Toda
lattice which bears the name of the Lie algebra. For example, the SU(N)
one-dimensional finite, nonperiodic Toda lattice is exactly the standard
finite, nonperiodic Toda lattice with the free endpoints. The standard
periodic Toda lattice corresponds to contragradient Lie algebras. It
appears that these equations have important applications in the gauge
field theory, namely, in the construction of the spherically symmetrical
instantons or monopoles [28-31]. Let me briefly present how one can
recover the Toda lattice from the self-dual equation for the Yang-Mills
field theory. The instantons are defined as the finite action self-
dual of the Yang-Mills field theory [32]. The monopoles are defined
as the static solutions with finite energy of the self-dual Yang-Mills-
Higgs field assuming the so-called Bogomolny-Prasad-Sommerfeld limit
[32]. Here the self-dual condition means that

FjJV'
(37 )

where

F
jJV
ajJ Av av AjJ + [A ,A ],
jJ v
(38 )

and A is a Lie algebra valued matrix function.

jJ

The self-dual conditions (37) are usually written down in the com-
plexified space-time. This is achieved by introducing new coordinates

y z n- (x 2 -ix 3 ) (39 )

y z n- (x 2 +ix 3 ) (40 )

in terms of which (39) takes the form

F F-- 0, (41 )
yz yz

F + F o. (42)
yy zz,

Notice that due to the complexification of the space-time the gauge

group is also complexified. Now let us consider the special ansatz
[33] for Au' A_ assuming that u = y = z, u = y = z.
u

A
u
E +(y
oEM 0
exp[-E k B~BE
B 0 -0
+ au ~ 0 H),
0
(43 )

A E +(y exp[-E k B~BE+ + a ~ H ), (44 )

u oEM 0 BOO U 0 0

218
where Ya' Ya,$a' ~a are functions of z, z and E±a' Ma are the genera-
tors of the gauge group which we choose in the Cartan-Chevalle basis
[34]. Hence they satisfy the following commutation relations

(45 )

(46 )

Here M+ denotes the set of the simple roots of a given Lie algebra
of a gauge group and kaa is the Cartan matrix. For the SU(N) gauge
group it has the form (4) with finite n. Substituting (43-46) after
equating the coefficients corresponding to the same generators we obtain

exp(-~ kaa($a+~a)' (47 )

exp(-~ kaa($a+~a». (48 )

Equation (48) is the Toda lattice in the form (3) with the free endpoints.
for the SU(N) gauge group. For different groups, we have different
Cartan matrices and hence different Toda lattices. These equations
can be solved by pure algebraic method [31,33,35] or for the classical
semisimple gauge group by the non-auto Liouville-Backlund transformat-
ion [36]. This name follows from the fact that for the SU(2) gauge
group, equation (48) reduces to the Liouville equation which possesses
the non-auto Backlund transformation which transforms the solutions
of the two-dimensional Laplace equation onto the Liouville equation.
The Liouville-Backlund transformation transforms N-1 solutions of the
two-dimensional Laplace equations onto the SU(N) Toda lattice.

On the other hand, equations (41-42) can be solved in a different

way. First, let us notice that equation (41) tells that the potentials
Ay' Ay' Az ' A z are pure gauge. This means that

A D- 1 ay D A- 5- 1 a-5
y , (49 )
y y

D- 1 az D --1 -
A
z
A-
z D azD, (50 )

where D, 5 are arbitrary matrices. Introducing the new matrix R

5- 1 D equation (42) reduces to

o. (51 )

219
In the special case when z = z one can show [32] that equation (51)
describes the monopole solutions for the Yang-Mills-Higgs field theory.
Surprisingly if we now consider the following equation

(52 )

where Rn is the matrix-valued function of y, y belonging to SL(N,C),

then it can be considered as the lattice approximation of the equation
(51) for z = z, what can be easily seen [37] using the Taylor expansion
of
(53 )

In the special case when

<jln
R e .1 (54)
n
we obtain the equation (1) the Toda lattice. This generalization of
the Toda lattice is known as the nonabelian Toda lattice. Perk and
Capel [38] were the first to introduce this concept to physics. They
showed that the correlation between x and y components of the spin
in the inhomogenous X-Y model can be described by the one-dimensional
nonabelian Toda lattice. The present author has shown that this one-
dimensional nonabelian Toda lattice can be considered as the lattice
approximation of the chiral models [39]. This nonabelian infinite
Toda lattice possesses the inverse scattering transformation as well
as a Backlund transformation [37,39,40]. The multisoliton solutions
for the periodic nonabelian Toda lattice can be obtained from the inverse
scattering transformation [41] by the use of the so-called "soliton
correlation matrix" [42].

In order to show the basic difference between abelian and nonabelian

solutions of Toda lattice let us present a straightforward method [43]
of constructing the solutions of Toda lattice in both cases. Let us
define the Backlund transformation and the inverse scattering transfor-
mation [37] respectively as

R- 1 a R - R,-l a R' a {R,-l R - R,-l R } (55)

n y n n-1 y n-1 n n n-1 n-1 '

a- R'R,-l - a-R R- 1
y n n y n n
1a {R'R
n n+1
- R' R- 1 }
n-1 n (56)

where a is an arbitrary parameter, Rn and R~ are respectively the

known and unknown solutions of equation (52),

220
 -1 ( 57)
a y ljJ(n) IjJ (n+ 1 ) R (n)a R(n)ljJ(n),
y

a- ljJ(n) R- 1 R IjJ (n-1 ) , (58)

y n n-1

-1
- ljJ(n)
a yy Rn + 1 Rn ljJ(n) - a-(R- 1 a R ) ljJ(n) - R- 1 a R a 1jJ( n), (59)
y n y n n y n y

where equation (59) is obtained from (57) and (58) by differentiating

(57) by a y and using (58). These formulae are obtained by the genera-
lization of the corresponding formulae in the abelian Toda lattice.
Notice that in (57) and (58) the absence of the so-called spectral
parameters. We can introduce these by scaling the matrix function

(60 )

where ]1 and ]1' are quite arbitrary parameters. We are now ready to
prove the following.

Theorem: I f Rand ljJ(n) satisfy (52) and (57-59) respectively, then

n

-1
R' IjJ (n+ 1 ) , ( 61 )
n

where a is an arbitrary parameter, satisfies (52). The proof is ele-

mentary; it is enough to substitute (61) into (52) and use (57-58).
In order to generate new solutions from (57-59), we should specify
the seed solution Rn and then solve (57-59). First let us consider
the abelian case for which we choose

R const. 1, (62)
n

then equations (57-59) read as

ay 1jJ( n) 1jJ( n+ 1 ) , (63)

a--y 1jJ( n) 1jJ( n-1 ) , (64 )

ayy- 1jJ( n) IjJ (n). (65)

The system (63-65) is well known in physics because it is the special

case of the telegraph equation. Kametake [44) studied this system
in the context of the soliton solutions of the Toda lattice and the
present author has shown [45) that the Kametake method is equivalent
to the Bicklund transformation. Kametake showed that the base of the
solutions of (63-65) can be chosen as

221
 An exp ( AY + i y), (66)

where A *0 is an arbitrary constant. The one soliton solution is

obtained from (61) where

I/J(n) (67)

and are reduced to the well-known one (anti) soliton solutions for
the one-dimensional infinite Toda lattice for y = y. For the non-
abelian case we made the following choice of Rn for the SL(2,C) group

i
R detR 1, s E R, (68 )
n n
[ -nS 1-inS

which satisfies the nonabelian Toda lattice. Indeed

[
l-iS ~ 1 (69 )
s 1+lS

Substituting (68) and (69) into (57-59) one can quickly realize that
these equations constitute a system of linear partial differential
equations with constant coefficients. We can solve that system using
its characteristic equation. In our case this equation is

det [l-iS-]J ~ 1 ( ~ - ]J)

2
0, (70 )
s 1+1S-]J

and possesses a two-fold root only. According to the general theory

of linear partial differential equations the solution can be represented
as

o 1 - 2 Ai y+ Ai
.1. Y
I/J . . (n) (A . . (n)y + A . . (n)y+ A . . (n)~ ( 71 )
1, J 1,J 1,J 1,J

where A1. are arbitrary constants and A~ ., k = 0,1,2, i,j = 1,2 are
1, J
coefficients which can be determined from the assumption that I/J(n)
satisfies (57-58), from the fact that R~E SL(2,C) and from the boundary
conditions. Let us compare the solutions (71) with (66). We see that
they coincide if s = O. For s *0 we obtain typical nonabelian solu-
tions. Similarly to the abelian case we can construct the linear com-
binations of (71). The matrix R~ obtained in this way I named in [43]
as the nonabelian one (anti) solitons.

222
 Let us return once more to the equations (1-6). From the soliton
theory point of view this equation can be considered as the model of
solitons in one discrete and in one or two continuous variables. One
can ask whether it is possible to construct solitons in two discrete
variables. Recently, Ablowitz et al. (46) considered the discretiza-
tion of the so-called "a" (DBAR) problem. They considered the finite-
differential analogue of the Schrodinger equation in the form

~(n-1,m) + B(n,m)~(n,m) + A(n,m)~(n+1,m) = A~(n,m+1), (72)

where (n,m) Ez2 and the "potentials" B, A-1 vanish sufficiently fast
as nand/or m goes to infinity. Due to it they obtained the Toda lattice
equation in two discrete variables and in one continuous as the compa-
tibility conditions between (72) and

d A-1 ~(n,m) + gO(n,m)~(n+1 ,m), (73)

dt ~(n,m)

IT A(n-j,m+j)/A(n-1-j,m+j), (74)
i=O

which reads as

d B(n+1,m)go(n,m+1) - B(n,m)go(n,m), (75)

dtA(n,m)

d (76 )
dtB(n,m)

This is the next possible way of generalization of the Toda lattice.

From (75-76) we recover the Toda lattice (5-6) by assuming that A and
B do not depend on m. Recently, Ragnisco and the present author used
in (47) the concept of Darboux transformation and obtained some new
formulae for the solutions of the equation (75-76). Let us briefly
demonstrate how it works. The concept of Darboux transformation can
be applied for the explicit integration of linear evolution equations
with scalar or matrix-valued coefficients. We define the Darboux trans-
formation for (75-76) by

f(n,m)
~(n,m) ~(n,m) - f(n+1,m) ~(n+1,m) ( 77)

f(n,m+1) f(n-1,m)
B(n,m) B(n,m) - f(n+1,m+1) + f(n,m) (78)

f ( n , m+ 1 )f ( n+ 2, m)
A(n,m) f(n+1 ,m) f(n+1 ,m+1) A(n+1 ,m), (79 )

where f(n,m) is a particular solution of equations (72-73) with A = A1 .

223
To see that ~, A and B satisfy the equations (72-73) let us substitute
the formula (77-79) to this equation and use the fact that f, A, B,
satisfy (72-73) also. Notice that in the special case when we assume
that A, B, ~ do not depend on m we recover the Darboux transformation
for the Toda lattice (5-6). This Darboux transformation is different
from the transformation in the paper of Matveev [48] because he used
a different representation for the so-called L-A pair. Notice that
the concept of Darboux transformation is similar to the concept of
Backlund transformation. Indeed, from the knowledge of the Darboux
transformation one can recover the Backlund transformation for the
Toda lattice (5-6) assuming additionally that

f(n) (80 )
f (n+1 )

where a is an arbitrary constant and ~'(n) and ~(n) are old and new
solutions of the Toda lattice. To see this, it is enough to substitute
(80) with (77) into (72-73) assuming that m dependence does not appear
and equate the coefficient standing on the same ~(n).

Concluding this brief review let us recapitulate what we considered

here. Firstly, we considered the applications of the concept of the
Toda lattice in the context of the eighteenth and nineteenth century
mathematics. Secondly, we tried to classify the different generalizations
of the Toda lattice which appear in mathematics and physics. We paid
special attention to the application of Toda lattice to the gauge theory.
However, our classification does not exhaust this subject and the scope
of this paper does not allow us to consider other interesting generali-
zations of the Toda lattice. Interested readers may refer to the papers
[49-53] where this problem is considered.

References

[1] M. Toda, J. Phys. Soc. Japan 22 (1967) 431.

[2] M. Toda, Phys. Rep. 18C (1975) 1.
[3] E. Fermi, J. Pasta, S. Ulam, Collected Papers of Enrico Fermi,
Vol. II (University of Chicago Press, 1965), p. 978.
[4] M. Wadati, Prog. Theor. Phys. Supp. 59 (1976) 36.
[5] Y. Kametaka, Proc. Japan Acad. GOA (1984) 145.
[6] Y. Kametaka, RIMS Kokyuroko 554 (1985) 26.
[7] H. Au-Yang and J. H. I-I,Perk,Physica18D (1986) 365.

[8 ] A. Common & D. Roberts, "Solutions of the Riccati equation and

their relations to the Toda lattice," (1986) Preprint.
[9] G. Darboux, Lecons sur La theorie generale des surfaces et les
applications geometrique du calcul infinitesimal II (Chelsea,1972).

224
[10] E. Ince, Ordinary Differential Euqations (New York: Dover, 1956).
[11] J. Sylvester, Compt. Rend. Acad. Sci. 54, No. 129 (1862) 170.
[12] H. Wall, Analytic Theory of Continued Fraction (New York: Van
Nostrand, 1948).
[13] S. Manakov, Soviet Phys. JETP 40 (1974) 269.
[14] H. Flaschka, Prog. Theor. Phys. 51 (1974) 703.
[15] V. Zakharov, S. Manakov, P. Novikov, and A. pytaevsky, The Theory
of Solitons (Moscow: Nauka, 1980).
[16] R. Hirota, J. Phys. Soc. Japan 35 (1973) 286.
[17] B. Kostant, Adv. Math. 34 (1979) 159.
[18] M. Olshanetsky and A. Perelomov, Theor. Math. Phys. 45 (1980) 3.
[19] A. Mikhailov, JETP Letters 30 (1979) 443.
[20] A. Fordy and J. Gibbons, Commun. Math. Phys. 77 (1980) 21.
[21] B. Barbashov and V. Nesterenko, Commun. Math. Phys. 78 (1981) 499.
[22] K. Ueno, K. Takasaki, Proc. Japan Acad. 59 (1983) 168, 216.
[23] M. Jimbo, and T. Miwa, Publ. RIMS. Kyoto 19 (1983) 943
[24] K. Ueno, and K. Takasaki, "Toda Lattice Hierarchy" RIMS 425
Kyoto, Japan (1983).
[25] M. Olshanetsky, Commun. Math. Phys. 88 (1983) 63.
[26] Z. Popowicz, J. Phys. A. Math. Gen. 19 (1986) 1495.
[27] o. Bogoyavlensky, Commun. Math. Phys. 51 (1979) 201.
[28] A. Leznov and M. Saveliew, Lett. Math. Phys. 3 (1979) 489.
[29] A. Leznov and M. Saveliew, Phys. Elem. Part A: Nuclei. 11 (1980)40.
830] A. Leznov, Theor. Math. Phys. 42 (1980) 343.
[31] A. Leznov, and M. Saveliew, The Group's Method of Integrations
of the Nonlinear Dynamical Systems (Moscow: Nauka, 1985).
[32] M. Prasad, Physica 2D (1980) 167.
[33] R. Farwell and M. Minami, J. Phys. A. Math. Gen. 15 (1982) 25.
[34] J. Humphreys, Introduction to Lie Algebras and Representation
Theory (New York: Springer, 1972).
[35] N. Ganoulis, P. Goddard and D. Olive, Nucl. Phys. B205 FS5
(1982) 60l.
[36] Z. Popowicz, J. Math. Phys. 25 (1984) 2212.
[37] Z. Popowicz, Z. Phys. Part and Fields 19 (1983) 79.
[38] J. Perk and M. Capel, Physica 92 (1978) 1563.
[39] Z. Popowicz, Phys. Lett. 81A (1981) 235.
[40] D. Levi, L. Pinolli and P. Santini, J. Phys. A. Math. Gen.
14 (1981) 1567.
[41] A. Mikhailov, Physica 3D (1981) 73.
[42] J. Harnad, Y. Saint-Aubin, and S. Shnider, Commun. Math. Phys.
93 (1984) 33.
[43] Z. Popowicz, "The nonabelian solitons for the SL(2, C) nonabelian
Toda lattice" to appear in Inverse Problems.
[44] Y. Kametaka, Proc. Japan Acad. 60 (1984) 79.
[45] Z. Popowicz, Phys. Lett. 116A (1986) 255.
[46] M. Ablowitz,s.Chitlaru Briggs, O. Ragnisco and P. Santini, "An
example of a problem in a finite difference context: Direct and
inverse problem for the discrete analogue of the equation
~ + u~ = o~ , to appear in J. Math. Phys.
xx y
[47] Z. Popowici, and o. Ragnisco, "Darboux transformation and one-
soliton solution for the two-dimensional Toda lattice" preprint
Universita di Roma "La Sapienza" (1987) (submitted).
[48] V. Matveev, and M. Salle, Lett. Math. Phys. 3 (1979) 425.
[49] A. Nakamura, J. Phys. Soc. Japan 52 (1979) 425.
[50] N. Saitoh, E. Takizawa and S. Takeno, J. Phys. Soc. Japan 54
(1985) 3701, 4525.
[51] N. Saitoh, J. Phys. Soc. Japan 54 (1985) 3261.
[52] M. Leo, R. Leo, G. Solani and L. Solombrino, Lett. Math. Phys.
8 (1984) 267.
[53] Li Yi-Shen, Cheng Yi, "Infinite dimensional Lie Algebra and a
the9ry on the symmetries and constants of motion in 1+1 dimensional
systems" preprint.

225
Construction of Exact Invariants for
One- and Two-Dimensional Classical Systems
R.S. Kaushal
Department of Physics, Ramjas College, University of Delhi,
Delhi 110007, India

A general method to construct the invariants of both time independent

and time dependent one- and two-dimensional classical system is outlined
by complexifying the two dimensions.

1• INTRODUCTION

If all possible constants of motion (briefly called invariants) of a

dynamical system are known or can be constructed, then it becomes
easier to study or to predict the behaviour of the system rather comp-
letely, and the system is said to be integrable [1]. Since there
exists scarcity of integrable systems in contrast to nonintegrable
systems, in recent years, there has been considerable interest in ex-
ploring the methods of construction of these invariants for both time-
dependent (TO) [2-12] and time independent (TID) systems [13-16,18].

These invariants play an important role in various disciplines,

particularly in plasma physics, hydrodynamics, astrophysics, biophysics
and in quantum field theory. In fact, nonlinear equations involving
higher degrees of freedom can be reduced to quadrature from the know-
ledge of these invariants. Besides their mathematical usefulness in
solving a problem, the physical interpretations of these invariants
have not yet been found for most of the systems. The most-studied
example [2-7] is that of one-dimensional TO harmonic oscillator. For
this system, the Hamiltonian

H "21 p 2 + "21 w2(t) q 2 (1 )

is not the constant of motion, and the invariant is found to be

1 2 2
I "2[~ + (pp - pq) ], (2 )
p

where p satisfies an auxiliary equation

2 -3
p + W (t)p p (3)

226
This is perhaps the only example for which some plausible physical
interpretations of I have been sought. Eliezer and Gray [6] argue
that I is equivalent to the angular momentum in a projected two-
dimensional plane in which the radial equation is given by the auxi-
liary Eq. (3). Takayama [6] relates I and subsequently the auxiliary
Eq. (3) to the betatron oscillations about the equilibrium orbits.
In the present talk, my main emphasis will be on the construction aspects
of these invariants for one- and two-dimensional classical systems.

2. ONE-DIMENSIONAL TD SYSTEMS

After the work of Lewis [2] and Lewis and Riesenfeld [2] several other
methods have been developed for the construction of invariants for one-
dimensional TD systems. Although most of these methods deal with the
invariants upto second order in momenta (this is perhaps to maintain
an analogy with the Hamiltonian of the system), the existence of higher
order invariants cannot be ruled out.

2.1 Second Order Invariants

Various approaches used for the construction of second order invariants
for the system

L "21 •2
q - V(q,t) (4 )

can broadly be classified as

(a) Ermakov approach as used by Ray and Reid [7];

(b) Lutzky's approach using Noether's theorem [5];
(c) Transformation group approach of Burgan et ale [3]; and
(d) Dynamical Lie algebraic approach [4].

While the approach (a) is more heuristic than the other three, it assumes
the form of auxiliary equation in advance. The approaches (b) and
(c) are rather involved and also used for solving a class of nonlinear
differential equations [5,17]. The approach (d), based on the closure
property of Lie algebra of phase-space function~ has relatively more
transparent connection with the corresponding quantum systems. More
general cases in which V depends on momentum and also being non-sepa-
rable can easily be dealt within this approach.

227
2.2 Higher Order Invariants

Recently, we have carried out [9) the construction of invariants of

third and fourth order in momen~a for the system (4), by assuming the
form of I as

I (5 )

where b. 's are the functions of q and t. The fact that I satisfies
dI 1
dt = 0, leads to a set of coupled partial differential equations

o 0,

ab 2 ab 3
3 + b 4 aV (6 )
aq at aq

abO ab 1
aq + b 2 av b 1 av
at aq aq

By method of elimination of the coefficients in (6) nonlinear differen-

tial equations in V (henceforth called 'potential' equations) can be
obtained separately for third and fourth order invariants. Further
assuming the separability of V in q and t variables, these potential
equations are reduced to simpler forms. The system,

(7 )

is found to be integrable and admits [9) a third order invariant of

the form

I (8 )

Other integrable systems can be found by solving these potential

equations.

3. TWO-DIMENSIONAL SYSTEMS

Several methods have been developed [10-16) and applied to the study
of two-dimensional systems. While the third and fourth order invariants
in momenta are constructed [13) for TID systems, construction of
invariants for TO systems beyond the second order seems to be not pur-
sued that vigorously. Here we outline a method which can uniformly

228
be used for the study of both TID and TD systems by complexifying [16]
the two-dimensions. The method has not only led to some simplifications
in the construction of invariants but also provided with several new
integrable systems in two-dimensions.

3.1 Time Independent Systems

We consider a dynamical system described by the Lagrangian,

1
L "2 z !.z - V( z ,z) (9 )

with z z -2 aV
az' z = -2 aaVz
We assume the second constant of motion (the first being the Hamiltonian
itself for the TID case) upto fourth order in momenta in the form

I a O + ai~i + 2T aij~i~j + 3T aijk~i~j~k + 4T aijkl~i~j~k~l'

( 10)

where i,j,k,l = 1,2 ; ~1 = z and ~2 = z and the coefficients a O ' ai'

a .. , a. 'k' and a. 'kl are functions of z and ~ only. These coefficients
1J 1J 1J
are symmetric with respect to any interchange of their indices. The
dI
fact that dt = 0, implies an expression in z, -
z which must vanish iden-
tically, viz.,

~ + a i,jSiSj
a O,isi ~ ~ + a1'~1'
S
+ 1
-2 ~ ~ ~
a ij,kSiSjSk

O. ( 11 )

Now, after accounting for the proper symmetrisation of the coefficients,

we obtain the following conditions on a ijkl , a ijk , a ij , a i and a O :

aijkl,m + ajklm,i + aklmi,j + almij,k + amijk,l 0, (12a)

aijk,l + ajkl,i + akli,j + alij,k 0, ( 12b)

0, (12c)

229
 a. + a. + aijk~k 0, ( 12d)
1, j J,i

aO,i + a IJ
.. ~j 0, (12e)

a·C O. ( 12 f)
1 1

From these equations it can be seen that the coupling between the
coefficients corresponding to even and odd powers of momenta in (10)
does not arise. As a result, the number of equations to be solved
simultaneously for the third and fourth order invariants is nine and
twelve, respectively. To solve these equations we adopt the following
procedure:
(a) Equations which are free from potential terms in the set (12) are
solved explicitly for the coefficients involved and with some
arbitrary constants;
(b) Equations involving the potential terms are reduced to a couple
of 'potential' equations by method of elimination;
(c) Most of the arbitrary constants which arise in the step (a), can
be fixed either by solving or by rationalising these potential
equations.
(d) Using these values of the arbitrary constants, various coefficients
in (10) can be determined which in turn provide the explicit struc-
ture of I.

The method is successfully applied [15] to a large class of poten-

tials (including the Toda, Holt and Fokas potentials and also a class
of Toda-type potentials) which admit the third order invariants. The
potential equation corresponding to the fourth order invariants requires
further investigation.

3.2 Time-Dependent Systems

If we restrict upto the third order invariants (a ijkl = 0), and the
potential V and the coefficients a O' ai' a ij , a ijk in Eqs. (9) and
(10) are allowed to depend explicitly on t, then the expression (11)
takes the form,

aa O aa i 1 • 1
(a-t + a.~.)
1 1
+ (a O,1. +"""'---t
aT:
+ -2 a IJ .. ~ J.)f,;.1 +
.. f,; J. + -2 a Jl

o. ( 13)

230
This expression, after accounting for the proper symmetrisation of
the coefficients as before, will now yield [12) the following restric-
tions on a O' ai' a ij and a ijk :

aijk,l + ajkl,i + akli,j + alij,k 0, ( 14a)

aa, 'k
~
aij,k + ajk,i + aki,j + at

0; o. (140 )

In contrary to the TID case, here the coupling between the coefficients
corresponding to the even and odd powers of momenta in (10) does exist.
Subsequently, the number of equations to be satisfied simultaneously
for a given order of invariant is much more as compared to the corres-
ponding TID case. This complicates the construction of invariants
for TD systems. Note that for the second order case (a ijk = 0), the
set (14) reduces to ten equations whereas for the third order it provides
fifteen equations. To solve these equations we follow the same procedure
as described before. However, the aribtrary constants which arise in
the step (a) now turn out to be the arbitrary functions of t and the
same can be fixed in terms of potential parameters.

Using the complexification method in the context of TD systems a

large number of integrable systems which admit the second order invariants
are found [11). Recently, Eqs. (14) are used [12) to obtain a third
order invariant for a TD Henon-Heiles system,

( 15)

The invariant turns out to be

~here c 1, c 4 and BO are arbitrary real constants and a satisfies,

a = -3a .

A part of this work has been carried out in collaboration with

Prof. K. C. Tripathy and Dr S. C. Mishra.

231
REFERENCES

[1) E. T. Whittaker, A Treatise on the Analytical Dynamics of Particles

and Rigid Bodies (Cambridge Univ. Press, London, 1965).
[2) H. R. Lewis Jr., J. Math. Phys. 9 (1968) 1976; H. R. Lewis and
W. B. Riesenfeld, J. Math. Phys. 10 (1969) 1458.
[3) J. R. Burgan et al., Phys. Lett. 74A (1979) 11.
[4) H. J. Korsch, Phys. Lett. 74A (1979) 294; R. S. Kaushal and H.J.
Korsch, J. Math. Phys. 22 (1981) 1904.
[5) J. R. Ray and J. L. Reid, J. Math. Phys. 20 (1979) 2054; Phys.
Lett. 74A (1979) 23.
[6) C. J. Eliezer and A. Gray, SIAM J. Appl. Math. 30 (1976) 463;
K. Takayama, Phys. Lett. 88A (1982) 55.
[7) J. R. Ray and J. L. Reid, J. Math. Phys. 23 (1982) 503.
[8) P. G. L. Leach, J. Math. Phys. 22 (1981) 465; Phys. Lett.
98A (1983) 89.
[9) R. S. Kaushal, Pramana J. Phys. 24 (1985) 663.
[10) G. N. Katzin and J. L. Levine, J. Math. Phys. 24 (1983) 1761;
B. Grammaticos and B. Dorizzi, J. Math. Phys. 25 (1984) 2194.
[11) R. S. Kaushal, S. C. Mishra and K. C. Tripathy, Phys. Lett. 102A
(1984) 7; S. C. Mishra, R. S. Kaushal and K. C. Tripathy, J. Math.
Phys. 25 (1984) 2217.
[12) R. S. Kaushal, "Third Order Invariants for Two-Dimensional Time-
Dependent Classical Systems: Henon-Heiles System," to be published.
[13) O. R. Holt, J. Math. Phys. 23 (1982) 1037; L. S. Hall, Physica
08 (1983) 90; T. Sen, Phys. Lett. lllA (1985) 97; B. Dorizzi
et al., J. Math. Phys. 25 (1984) 2200.
[14) M. J. Ablowitz, A. Ramani and H. Segur, J. Math. Phys. 21 (1980)
715; 1006; A. S. Fokas and P. A. Lagerstrom, J. Math. Ana. Appln.
74 (1980) 325.
[15) R. S. Kaushal, S. C. Mishra and K. C. Tripathy, J. Math. Phys.
26 (1985) 420; R. S. Kaushal and S. C. Mishra, Pramana J. Phys.
26 (1986) 109.
[16) S. C. Mishra, Ph.D. Thesis, Delhi University (1985).
[17) See e.g.,P.B.Burt and J.L.Reid,J.Math.Ana.Appln. 55 (1976) 43.
[18) R. Sahadevan and M. Lakshmanan, Phys. Rev. 33A (1986) 3563;
R. Sahadevan, Ph.D. Thesis, University of Madras (1986).

232
Nonlinear Chains and Kink-Impurity Interactions
Bishwajyoti Dey
Institute of Physics, Bhubaneswar 751005, India

One-dimensional monatomic and diatomic chains with harmonic coupling

between neighbouring sites and an on-site anharmonic potential V(~) =
c~2n+2 + B~n+2 + A~2 + 0 are examined in the displacive limit, which
serves as a model for a structural phase transition. It is shown that
these systems admit kinks, nonlinear phonons and periodon solutions.
The eigenfunctions of the small oscillations about the kink solutions
are also obtained. This complete set of eigenfunctions has been used
to investigate the influence of impurities on the amplitude kink for
n = 1 case of the potential. The kink solution change near the impurity
is evaluated with the help of linear perturbation theory.

In recent years one-dimensional field theoretic models exhibiting soli-

tary-wave solutions (kink, solitons, non-linear phonons) have played
an important role in condensed matter physics, as these provide a non-
perturbative approach to strongly anharmonic systems. These models
have been used to describe many systems, especially those that undergo
a structural phase transition. For example, the domain wall (kink)
solutions of A~4 field theoretic models are identified with the central
peak phenomena observed in ferroelectric crystals [1].

Here we consider a model of a one-dimensional lattice with harmonic

coupling between neighbouring lattice sites and an on-site potential

( 1)

with adjustable non-linearity for n = 1,2,3, .... This potential has

triply degenerate minima for even values of n and doubly degenerate
minima for odd values of n, and for

4AC with A, C > 0 and B < O. (2 )

This model describes a first-order structural phase transition at zero

temperature. We consider the model Hamiltonian for the monatomic chain
[2 ],

233
 (2. •2 2
H l: + V (cp i) ) + 2. l: Cij(CPi - cP j) I (3 )
2 mcp i 2
i i, j

where the on-site potential V (cp ) is given by equation ( 1 ) • The conti-

nuum limit is

H (4 )

and the equation of motion is

a (5 )

where a is lattice constant and CO' the velocity of sound. This

equation can be readily integrated to give the kink and anti-kink solu-
tions as

2- l/n
CPa
(1 ± tanh ¥) lin for n 1 ,3,5, ... (6a)

±2- l/n CPO(l ± tanh ~~) lin; for n = 2,4,6, ... (6b)

2 2
where CPa = (2A/B) lin; ~O = m(C O - v 2 )/2A.
The solutions are stable and have finite energy.

For the diatomic chain we take a model Hamiltonian

where u i and vi are the displacements of masses Ml and M2 respectively,

y is the harmonic force constant between neighbouring atoms and V(u)
is the on-site potential represented by equation (1). The coupled
equations of motion are given by

(8a)

(8b)

The continuum limit of the Hamiltonian is

H j ~x[~Mlu2 + ~M2v2 + y(u-v)2 + ~ ya2u'v' + V(u)] (9 )

-00

and gives the coupled fields equations

234
 dV

]
+ 2yf + df

( 10)
'
+ 2yf

where u(x,t) = f(5); v(x,t) g(5) and 5 x - Ct.

For a particular velocity

( 11 )

we get kink and antikink solutions (M 2 > M1 and D = 0) as

uk(x,t) vk(x,t) 2- 1 / n U o [1 ± tanh (~t )]1 /n for n 1 ,3,5, ...

k
( 12a)
±2- 1/ n U o[1 ± tanh(~~ )]l/n, for n 2,4,6, ...
k
( 12b)

These solutions are also stable and have finite energy.

For C * Co (equation (11)) we can obtain the nonlinear periodic

solutions. For n = 1 case the solutions of equation (10) are [2],

f(5) ';(1 + fO sink5) ( 1 3a)

and

g(5) ( 1 3b)

where the amplitudes f O' gl and g2 are determined in terms of the

parameters and k = (2Y/9M 2 C2 ) 1/2 The periodon solutions which solve
the discrete equations of motion (equations (8)), can be expressed
as (for n = 1 case of potential)

(14a)

( 1 4b)

One can also obtain kink and antikink solutions for a model Hamiltonian
generalised by including a single well on-site potential on the M2
species of atoms [2].

235
 The eigenfunctions and eigenvalues of the small oscillations about
the static kink solutions (equations (6)) is obtained from the equation
[41

o , ( 15)

where

A[n 2 + 1 - j(n+1 )(2n+1)sech 2 ~ + (n+1 )(n-1)tanh ~t 1.

2 0 0

( 16)

The number of bound states is given by the relation [51

nb = 0,1,2, ... < .l[n+1

n
-1711. (17 )

Thus for n = 1, there are two bound states corresponding to nb = 0

and 1. There is only one bound state for all other values of n. The
eigenfunctions of the bound states are given in terms of the hyper-
geometric series F(a,b,lclp) as (K = nx/2s0)

( 18)
where
2(n+1)(n-1) ( 19)
n(3n+2)-2(m+i)

and

b (3n+2) _ (m + 1) (20 )
m 2n 2 .

The frequency of the bound states is given by

- 1 2
(m + 2) J - (n+1)2(n-1)2 ]
2 . (21 )
n
(
4 3n+2
~
(
- m
+ -i )J
The continuum states are

-K
( .lik _ 1. k 1 1 .k 1 .k e
F - 2 + 21 + 2 -y, - 21 + 21 K -K)'
e +e
(22)
236
 222
where k_ = k+ + 4 - 4/n and the frequency of the continuum states
can be expressed in terms of the wave vector k+ as

(23 )

For n = 1 case only the hypergeometric case can be terminated and the
continuum states can be expressed in terms of hypergeometric functions.
We use this complete set of states for n = 1 case of the potential
to calculate the kink impurity interaction, using the linear perturba-
tion theory.

The model Hamiltonian for kink-impurity interaction is taken as[6],

00

H f~x[~m(tt)2 + ~mc~(~)2 + V(~) + ~V(x)~2], (24 )

-00

where

V(x) a.O(x-x ) (25 )

p

describes the short range kink-impurity interaction potential. The

equation for ~(x,t) is

-V(x)~ , (26 )

where the damping constant y describes phenomenologically the sto-

chastic character of kink motion between their collisions. The solution
of equation (26) is given by

~(x,t) ~K(x) + u(x,t) (27)

where ~K(x) denotes the static kink solution (equation (6a)) and u(x,t)
the fluctuation in ~K(x) due to its interaction with impurity. Subs-
tituting equation (27) in equation (26) we get the equation for u(x,t)
as
2 2
mau mC2au+A(2-3sech2x )u+ vau (28 )
at2 - 0 ax2 2t;O I at
Using linear perturbation theory, the fluctuation u(x,t) can be written
in terms of the basis functions {¢} (equations (18) and (22), for
n = 1) as [6],
00

(29 )

237
Substituting equation (29) in equation (28) we obtain coupled equations
for the coefficients {s} , the solutions of which determine u(x,t)
completely and hence ~(x,t) in equation (27). The coefficients {S}
is determined to be

0,1, k (30 )

where

2 2
mw]. + A .. - Y 14m, for j 0, 1 (31 a)
]]

2 2
mW k - y 14m, for j k ( 31 b)

A .. fdx[V(x)III'1!III·) , ( 31 c)
]] -co ] ]

(31 d)

x -x x -x xo-x P
(1 - tanh ~) sech 0 p tanh
2E;0 2E;0 2E;0

Ci.m
A (It')
4"'0
i -----------------------------------------, (31e)
x -x
tanh2 0 p
~

(31f )

Xo and xp denote the initial position of the kink and position of

impurity respectively and w~ denote the frequency of the eigenstates
]
(equations (21) and (23), for n = 1). This solution u(x,t) can be
used for calculating the kink-impurity binding (pinning) energy [6).

REFERENCES

[1) J. A. Krumhansl and J. R. Schrieffer, Phys. Rev. Bll (1975) 3535.

[2) B. Dey, J. Phys. C: Solid State Physics 19 (1986) 3365.
[3) B. I. Henry and J. Oitmaa, Aust. J. Phys. 36 (1983) 339.
[4] B. Dey, Proc. SSP Symp. (India) 29C (1986) 137 and unpublished
( 1986) .
[5] P. M. Morse and H. Feshbach in Methods of Theoretical Physics
(McGraw-Hill, New York, 1953), Vol. 2, p. 1651.
[6) V. L. Aksenov, A. Yu. Didyk and R. Zakula, Preprint E17-84-483,
JINR, DUBNA (1984).

238
 Part IV

Statistical Mechanics and

Quantum Aspects
Quantum Solitons: An Overview *
R. Rajaraman
Centre for Theoretical Studies, Indian Institute of Science,
Bangalore 560012, India

The quantisation of the classical solutions of nonlinear field equations

by semiclassical methods is explained.

1. INTRODUCTION

It is well known that nonlinear relativistic field equation permit

localised classical solutions, called solitary waves and solitons. Here
we discuss the relevance of such classical solutions to the corresponding
quantum field theories and show how these solutions can be associated
with quantum, extended, non-pertubrative-particle states. To obtain
the quantum particle-states we quantise the classical static solutions.
There is a variety of techniques for executing such quantisation. The
idea of associating these classical soliton-solutions with particle
states in the corresponding quantum theory comes from the fact that
these classical solutions resemble extended particles, as these are
localised (finite size) and have finite energy. However, this connection
between the classical solution and the quantum particle is not trivial.
For example, it is not correct to identify the extended classical solu-
tion with say, something like the wave function of the quantum particle.
It should be noted that though the classical soliton solutions resemble
particles, the concept of particles exists only in quantum field theory
and not in classical field theory. In classical field theory, the fields
are continuous c-number functions of space-time and these functions
specify the s~ates of the classical system. The dynamics of these fields
is governed by nonlinear partial differential field equations. In quantum
field theory however, the fields are operator functions of space-time
acting on vectors in a Hilbert space. The states of the quantum system
are specified by vectors in that Hilbert space and not by the fields.
In particular, 'particles' are special state-vectors of the quantum
system. They are simultaneous eigenstates of the Hamiltonian and momen-

*Notes taken by Bishwajyoti Dey.

240
tum operators, with eigenvalues forming a discrete hyperboloid in the
E-P space. Additional requirements are that the form-factors in these
states should be localised to reflect their particle-nature.

Thus we see that the classical solutions of field equations on the

one hand, and the concept of a particle in the corresponding quantum
field theories on the other are quite different entities. However the
two entities are related. Certain properties of the quantum states,
e.g., energy or form factors can be expanded in semiclassical series
and the leading terms in these series are found to be related to the
corresponding classical solutions. These relations between the two
entities can be easily established by generalising to field theory
certain concepts of quantum mechanics such as the correspondence prin-
ciple and the semiclassical expansion (which relate quantum levels to
classical orbits). To develop this generalisation, we recall the analogy
between objects in quantum mechanics and in quantum field theory.

Consider a non-relativistic, unit mass particle in one dimension

under the influence of a potential V(x). Classically, the particle is
described by giving x as a function of time, and this is obtained by
solving Newton's equation

dV (1 )
- dx

In quantum mechanics, the particle is described, not by giving the value

of x, but of the wave function ~(x,t). x now becomes an operator which
operates on the sJ;laceof all wave functions (Hilbert space). Similarly
momentum p X is also an operator which satisfies

[p,x) -Hi (2 )

where [,) denotes the commutator. The Hamiltonian which is also an

operator is given in terms of the operators x and p by

2
H ~ + Vex). (3 )

The energy eigenstates ~n(x,t) obey the time independent Schrodinger

equation

(4 )

Now consider the theory of a scalar field governed by the Lagrangian

241
The analogue of x(t) in field theory is the field ~(x,t) which is
classically obtained by solving the field equation. To quantise the
system we consider the commutator

[~(x,t), ~(y,t)l H'i.s (x - y), ( 5)

where ~(x,t) is momentum conjugate to the field ~(x,t) and they both
are operators. Corresponding to wave function ~(x,t) in quantum
mechanics here we have ~[~(x,t)l whose square gives the probability
of finding the field in the configuration ~(x,t). The energy eigen-
states are again determined by
(6 )

where the Hamiltonian operator H is given by

.2 (V) 2
H fdx(1- + -~ + u[~l). (7 )

The lowest eigenstate I~o> is called the 'vacuum' of the quantum field
theory and by a suitable choice one can make Eo 0 (and the momentum
= 0) for the vacuum state. The Hamiltonian and momentum operators
commute with each other, and a particle in quantum field theory corres-
ponds to a discrete hyperboloid in the spectrum, i.e., a family of
simultaneous eigenvectors of the Hamiltonian and momentum operators,
with eigenvalues E and P obeying

(8 )

with fixed m. We are interested in the relationship of these particle-

states in quantum field theory to localised solutions of the classical
field equation

o~ + aU o. (9 )
a~

Now, recall the relation between quantum energy eigenstates deter-

mined by Eq. l4) and the static solution of Eq. (1). Let us consider
a potential V(x) which has an absolute minimum at x = a, and a local
minimum at x = c. Then clearly there is a classical static solution,

x(t) a.

It may be called the 'classical ground state' as it represents, the

lowest-energy classical solution. Its energy is

V(a) . ( 10)

242
A particle placed at the point x = a will stay there classically. In
quantum theory such a state is not allowed. The uncertainty principle
will not permit the particle to have both zero momentum and a fixed
position. Consequently, even in the ground state, the particle will
fluctuate around x = a leading to a ground state energy

( 11 )

where ~o is the energy of zero-point motion. Furthermore, if the poten-

tial is harmonic (approximately) near x = a, one can make a Taylor
series expansion of V(x) near x = a

V(x)

Then, for those wave functions that satisfy

« w
2
«(x-a)
2
r = 3,4, ... (13 )

where < ••• > denotes expectation value, the effect of the anharmonic
terms of V(x) will be small. For Xr sufficiently small, the potential
will be dominantly that of a harmonic oscillator and the low-lyipg energy
eigenstates whose spread is localised in the vicinity of x = a will
have energies

VIa) + (n + ~)hW + 0(\). ( 14 )

In particular the ground state energy is given by

( 15)

Note that the first term in the expansion (15) is just the energy VIa)
of the classical solutions. Equation (15) represents the simplest
relation between some quantum states and corresponding classical solu-
tion. Apart from this, the classical solution also gives other fea-
tures of the ground state wave function. For instance the position
expectation value

a + ( 16)

The quantity on the r.h.s. is the classical solution whereas the l.h.s.
is a property of the quantum ground state.

Repeating the same procedure for the local minimum of the potential
at x = c, we have a classical solution

x(t) c. ( 17)

243
It has the classical energy

V(c) , ( 18a)

which is higher than V(a). This solution is interesting, as it is

analogous to the classical static soliton in field theory. The solitons
too are static solutions, but with higher energies than the classical
vacua in the corresponding field theories. As before we can make a
Taylor series expansion about x = c (local minimum) giving

1 2 2 A) 3 A4 4
( 18b)
V(x) V(c) + 2(w') (x-c) + 3T(x-c) + 4T(x-c) +.

and if A~ (r=3,4, ... ) are sufficiently small, then the approximate

harmonic oscillator states centred at x = c will have en€rgies

E
n
v ( c) + w' (n' + ~) + 0( A~) (18c)

and expectation values

<x> c + 0( A~) • ( l8d)

Thus again we have a set of energy eigenstates whose energy is related

to V(c), the classical energy of a static solution and whose <x> is
related to the classical solution itself.

However while deriving results in Eq.(18c,dl we ignored the effect

of the deeper potential well at x = a. Actually the wave packets built
in the potential well around x = c will tunnel into the well around
x = a and vice versa. But, if the Ar and A~ are all small, the two
minima x = a and x = c will be widely separated with a large potential
barrier between them. Therefore, the tunneling will be small. Thus
to any finite order in the weak coupling expansion, the set of levels
around x = c can be considered separately from the set around x = a.

In field theory on the other hand we have examples of both situations.

The ones where such tunneling takes place, are associated with "ins-
tantons" and the ones where tunneling does not take place, are associated
with "quantum solitons." The states built around most of the soliton
solutions can be shown not to decay at all through tunneling. This is
related to the non-zero topological index carried by these solutions,
which effectively places an infinite energy barrier between them and
the vacuum solutions.

Now in order to generalise these ideas to field theory let us first

consider a system with a large (but finite) number of degrees of freedom.

244
Let a bunch of unit-mass non-relativistic particles be collectively
described by a cartesian coordinate x = (x 1 , .. x N ) with a potential
V(x 1 ... x N ). Let x = a be a (local or absolute) minimum of V in the
N-dimensional space. Then x(t) = a will be a stable static classical
solution. Expanding V(x 1 , .. x N ) around xi ai'

V(x)

1
+ 3T ( x . -a. ) (x . -a . ) (x k -a k ) [ a
a3a v a ] _ + ... . (1 9 )
• 1 1 J ] xi Xj x k x-a

In the weak coupling limit we can construct a set of low lying states
around x = a by diagonalising the matrix of second derivatives
[a 2 v/ax.ax.] by changing variables to some normal modes ~l'. Then
1 J x=a
in terms of ~i' the problem reduces to a set of oscillators of fre-
quencies wi' with wf, the eigenvalues of the matrix. The lowest-energy
state constructed around x = a will have energy

N 1
E
o
V(a) + E 2hwi + higher order terms (20 )
i=1

and the higher excitations will be given by

N 1
V(a) + E (n. + 2)wi + higher order terms. ( 21)
i=1 1

These higher order terms are to be obtained by perturbation theory.

As can be seen these results are the generalisation of the one-degree
of freedom problem, with the set of eigenvalues (w.)2 of
1
[a~/ax.ax.]
1 J x=a
replacing the single number w 2

2. QUANTI SAT ION OF SOLITON SOLUTIONS

In field theory, instead of a finite number of degrees of freedom

x 1 ' ... ,x N ' we hav~ a continuous infinity of degrees of freedom ~(x),
i.e., the value of ~ at each space point x. We consider the example
of the sine-Gordon field ~(x,t) here. The action is

4
s !dt dx {~ (a~~a~ ~) + ~ (cos/~ ~ - 1). (22)

Changing variables to

x --> mx, t --> mt and I~ ~ --> ~ (23)

245
we get the Lagrangian in the standard form

L T[cp 1 - v[cp 1 (24 )

as

L = fdx[~(~)21 - fdx(~(cp,)2 + 1 - coscp) = T[cpl - V[cpl. (25)

The stable static solutions of the field equation

a2 cp(x,t) aV[pl
(26 )
acp( x, t)
dt 2

are once again the minima of the potential, satisfying

o cpU _ sincp ( 27)

These are given by

(i) cp (x) 0, ±2n, ±4n, . • • (28a)

( i i) cp (28b)

(iii) cp = CPanstisol(x), (28c)

+x
where CPsol(x) = 4 tan -1 (e-) (29 )
(antisol)

(the ± sign describes soliton and antisoliton solutions respectively).

In quantum field theory, as mentioned before, the field cp , n (the
cp
conjugate momentum) and the Hamiltonian are operators. The generali-
sation of the matrix of second derivative of V occurring in Eq. (19)
is the operator (-a 2 /ax 2 + coscp) evaluated at cp(x) = cpo(x) which is any
one of the solutions in Eq. (27). Its eigenvalues and eigenfunctions
are given by the differential equation

(30 )

where ni(x) are the orthonormal 'normal modes' of fluctuations around

cpo(x). Thus in quantum theory (weak coupling approximation), one can
construct a set of approximate harmonic-oscillator states, spread in
field space around cpo(x). The energies of these states would be

1
V[cpol + h E(n. + 2)w i + higher order terms (31 )
i 1

246
For the solution given by Eq. (28a) we have as the eigenvalue equation

_a 2 a2
(- + coscp) n i (x) (- - 2 + 1) ni (x)
dx 2 cp=O ax
2
The eigenvalues are wi k~1 +

(33 )

The set of approximate harmonic oscillator states built around cp= 0

have energies
1
E = V[cp =0 1 + L, (n i + 2M i + higher order terms (34)
ni 1

(we have used the units ~ = c 1). The vacuum state has energy

1
E E
vac
.l L (k~ + 1) 2 + higher order terms. (35 )
0 2 1' 1

The next excited state (one particle state) has energy

E1 E + (k 2 + 1) 2 + higher order terms. (36 )

0 1

E1 - Eo represents the energy of the single-quantum particle state.

Its mass, in terms of the original parameters in (22), is

mass m + O(A) ( 37)

2 1 221
(in terms of original variable (k +1)2 ~ (k +m )2, etc.). Similarly
for the soliton solution in Eq. (28b) the eigenvalue equation is

a2
{- ---2 + coscp(x) }ni(x) (38 )
ax sol

This equation is exactly solvable and has one discrete eigenvalue

o (39a)

followed by a continuum,

2
(qN + 1) (39b)

where ~N = 2nN + o(qN)

One can now associate a set of energy levels around cp(x) CPsol(x)
,with energies

1
V[mTsol 1 + L,(n 1, + -2 )w 1, + higher order terms. (40 )
1

247
Of these, the lowest state has energy

E V[~soll + ~ ~wi + higher order terms. ( 41 )

1

This is the basic soliton-particle state in quantum field theory. Higher

states are soliton-plus-meson states. Together they form the soliton
sector of states. Finally we make some comments:

i) All we are doing is quantising small fluctuations about any given

stable static classical solution, whether it be ~ = 0 or

~ = ~sol(antisol)·

ii) The method is 'semiclassical' but no less quantum mechanical than

standard perturbation quantum field theory, which amounts to
expanding about ~ = o.

iii) The states in the soliton sector carry a nonzero topological quantum
number

1
Q f ~ dx (42 )
"2 dx

The associated current is

(43 )

which is conserved, i.e.,

o (44 )

and gives rise to the quantum number

Q (45 )

iv) The quantisation of time-dependent solutions can be carried out

by extending the standard WKB method to field theory.

v) The occurrence of 'ultraviolet' divergences in quantum field theory

due to short-distance behaviour of products of field operators
is well known. It is removed by adding suitable 'counter terms'
to the Hamiltonian. When this is done, ultraviolet divergences
in the sum EW i over modes in expressions like (41), (40), (35),
cancelout. i

248
 Fuller details and references are given in the reviews listed below:

The method for quantising solitons was first developed systematically

by

R. F. Dashen, B. Hasslacher and A. Neveu, Phys. Rev. 010 (1974) 4130

and independently by
J. Goldstone and R. Jackiw, Phys. Rev. 011 (1975) 1486.
The extension of W. K. B. method to field theory to quantise time-
dependent soliton solutions (such as the sine-Gordon breather solutions)
was developed by
R. F. Dashen, B. Hasslacher and A. Neveu, Phys. Rev. 011 (1975) 3424.
A review of the above work and other developments is given in
R. Rajaraman, Phys. Reports 21C (1975) 227.

249
Soliton Statistical Mechanics:
Statistical Mechanics of the Quantum and
Classical Integrable Models
R.IC Bullough, D.J. Pilling, and J. Timonen *
Department of Mathematics, U.M.I.S.T., P.O. Box 88,
Manchester M601QD, United Kingdom

It is shown how the Bethe Ansatz (BA) analysis for the quantum statis-
tical mechanics of the Nonlinear Schr~dinger Model generalises to the
other quantum integrable models and to the classical statistical mecha-
nics of the classical integrable models. The bose-fermi equivalence
of these models plays a fundamental role even at classical level. Two
methods for calculating the quantum or classical free energies are
developed: one generalises the BA method the other uses functional
integral methods. The familia~ classical action-angle variables of
the integrable models developed for the real line ~ are used throughout,
but the crucial importance of periodic boundary conditions is recognized
and these are imposed. Connections with the quantum inverse method
for quantum integrable systems are established. The R-matrix and the
Yang-Baxter relation playa fundamental role in the theory. The lec-
tures draw together the quantum BA method, the quantum inverse method,
and the generalised BA and functional integral methods introduced more
recently.

1. INTRODUCTION TO THE CLASSICAL AND

QUANTUM INTEGRABLE MODELS

The simplest integrable model with real physical interest is probably

the 'repulsive' nonlinear Schrodinger (NLS) model with equation of
motion

(1. 1 )

~ is a complex scalar field. This classical system is a Hamiltonian

system with Hamiltonian

H [~ )
r
J [~ x~ x*
2 2
+ c~ ~* )dx ( 1 .2)

*On leave from Department of Physics, University of Jyvaskyla,

SF-40100 Jyvaskyla, Finland

250
the parameter c is a real-valued coupling constant, and, for the
'repulsive' NLS model, c > O. Evidently this is consistent with the
repulsive character of the nonlinear self-potential c~2~*2, which
is positive when c > O.

By imposing the Poisson bracket

{~(x,t), ~*(x',t)} ia (x - x') (1. 3)

one finds that Hamilton's equation of motion

( 1 .4)

is exactly (1.1). There is now an obvious extension to a quantum

t
theory in which ~, ~* become operators ~,~ with commutation relation

t
[~(x,t), ~ (x' ,t) 1 a (x - x') ( 1 .5 )

and (1.4) becomes Heisenberg's equation of motion. For a well defined

ground state of the quantum model ~, ~t must be placed in normal order:

H[~l (1. 6)

The 'hat' on H[~l indicates it is now an operator.

We shall see that H[~l, (1.6), describes N bosons on a line with

repulsive a-function interactions. Indeed there is evidently a number
~ t ~

operator N = f~'~ dx which commutes with H. Moreover, since (1.1)

is now

t 2
~xx - 2c~ ~ (1. 7)

(in normal order) in second quantisation, there is the equivalent

Schr6dinger problem in first quanti sat ion

{- a2
\'L --2 + 2c L a( Xi - x j) }'l' E'l' , ( 1 .8)
i=l aX i bj

and this displays the repulsive a-function interactions. The 'on

a line' means that the model is a one space-dimensional model defined
on (say) b::. x ::. d. The cases of most interest are b -+ -eo, d -+ +eo (i. e. ,
xE ~) with appropriate decaying boundary conditions, or periodic

251
 1 1
boundary conditions of per iod L (say) on - "2L 2. x 2. "2L. The quantum
model with periodic boundary conditions we have just described was
the model of a pioneering paper in quantum statistical mechanics by
Yang and Yang [1]. Their analysis was based on the quantum Bethe
Ansatz (BA) method which we very briefly describe later. There is an
ansatz for 'I' which solves (1.8) exactly [2]. However, the fact that
a quantum model with a Schrodinger problem such as (1.8) is exactly
solvable has much wider implications. The purpose of these lectures
is to show how the Yangs' result for the free energy of this quantum
model apparently generalises to all of the integrable models including
the classical integrable models.

The history of the classical integrable models is that they were

first solved with q>, defined for x E JR, vanishing 'fast enough' at
\x\ = 00. Typically, this is exponential decay at x = too [3]. However,
we shall find that it is periodic boundary conditions, rather than
these decaying boundary conditions, which playa fundamental role
in the statistical mechanics.

At first sight the 'attractive' classical NLS (c < 0) is more

interesting than the repulsive case (c > 0); for the former has soliton
solutions. For c = -1 the l-s01iton solution is

2n exp{-4i( e: 2 -n 2 )t + 2ie: x + i/)}

q>(x,t) ( 1 .9)
cosh [8ne: t - 2 n( x - x o )]

e:, n, /) and x are real free parameters and (for n > 0) decays
as e- 2n \x\ as Ix \ ~ 00. However when c < 0 the quantum NLS has no
lower bound to its ground state. Thus starting with the simpler cases
we should first of all be concerned with the repulsive case of the
quantum or classical NLS.

Still, in a sense which becomes plain once we transform to action-

angle variabl~s (§2), the two NLS models are non-relativistic forms
of two covariant models: the attractive NLS corresponds to the sine-
Gordon (s-G) model

q> xx - q>tt ( 1. 10)

the repulsive NLS to the sinh-Gordon (sinh-G) model

m2 sinhq> . ( 1. 11 )

252
The m's are "masses", so the units are such that ~ = c (= velocity
of "light") = 1. The classical s-G has the kink and antikink solutions

cp(x,t) = 4 tan- 1 exp{± m(x-Vt)'Y} ; ( 1. 12)

y ~ (1 - V2 )-1/2. It also has the "breather" (bound kink-antikink

pair) solutions

cp(x,t) = 4 tan- 1 [tann sin{Ym(cosnllt-Vx)} sech{ym(sinnllx-Vt)}].

( 1. 13)
These are the soliton solutions of the s-G model. Evidently the
boundary conditions (b.c.s.) are now cp ~ 0 (mod 2n) 'fast enough' as
Ixl ~ 00; and CPx ~ 0 fastenough. On the other hand the sinh-G, like
the repulsive NLS has no soliton solutions. We shall find in conse-
quence that we can develop the statistical mechanics of sinh-G wholly
in parallel with that of the repulsive NLS.

Both the s-G and sinh-G are Hamiltonian: for s-G

H[cpl Yo -1f [2 1 11 2 Yo2 + 21 CPx

2 + m2 (1 - coscp) ldx (1. 14)

and 10 (the real valued coupling constant) > O. Then sinh-G is found
by cp ~ i cp, Yo ~ - Yo. The bracket is

{11(x,t), cp(x' ,t)} o(x - x') (1. 15)

and (1.4) for s-G is

CPt =
y,
011, I1 t =
-1 [ 2.
Yo CPxx - m s1ncp
1 (1. 16)
1 1
which is (1.10). The canonical transformation cp.... :r~ cp,11 ~ Yo-2 11
2 -! ! 2 1 3
means Yo appears only as -m Yo 2 sin ( y~ cp) = -m [cp - '6 Yo cp + ••• 1 in
(1.16), so that yoscales the s-G nonlinearity just as c scales the
NLS nonlinearity. However, cp is a real field for s-G and sinh-G,
whilst it is complex for the two NLS models.

It is plain too that sinh-G is a continuation of s-G as Yo ~ -Yo

(and vice-versa) just as the NLS equations continue in c. We use
this fact later. The places Yo = 0, c = 0 are exceptional: Yo = 0
in sinh-G or s-G is the linear Klein-Gordon (K-G) model

m2 cp ; (1. 17)

c = 0 is the linear Schrodinger (LS) model

(1. 18)

253
Evidently both linear models are still Hamiltonian (set Yo ~ 0 in
(1.14)). Both are free-field models in an obvious sense.
There are other integrable models of physical interest: the Landau-
Lifshitz model is

1
H [ <p]
2"

00

1
( 1. 19)
2"

with J = diag(J 1 ,J 2 ,J 3 ), 0<J 1 <J 2 <J 3 S(x,t)

and the bracket is

(1. 20)

Hamilton's equations (1.4) are

S,t S x S + S x(J.S) . ( 1. 21 )
,xx

This model contains both the s-G and attractive NLS models as special
cases [4,5]; the repulsive cases can also be found. In the isotropic

case J 1 = J 2 = J 1 the model is the integrable Heisenberg ferromagnet

[6,7] with equation of motion S = S x S
,t ,xx·
As for the NLS models the classical s-G models (s-G and sinh-G)
have extensions to quantum ~heories: the commutation relations are

[n (x, t), <p ( x' , t) ] -i6(x - x') (1. 22)

and the equations of motion ((1.10) and (1.11)) and the Hamiltonians
((1.14)) must be placed in normal order. This cannot be done in practice
until we define operators of annihilation and creation type from <po
We develop such apparatus later.

One of the, more remarkable properties of quantum s-G, evidently

a massive boson problem, is its strict equivalence to the massive
Thirring model (MTM), a massive fermion problem. "Equivalence" means
equivalence of expectation values--especially the eigenspectrum [8].
The MTM is

f dX[-i<pto 3<Px + mo<pt °1<P + 2g o <p;<pi<p2<Pl]

<p
[:~] I 6(x - x') , ( 1. 23)

254
in normal order: 03' 01 are Pauli matrices. The free field has go
= 0, and 0 < g 0-
< 00. In connection with the quantum spin-t XYZ-model

H[cp] (1. 24)

which has spin commutation relations [S~, S~] = iEasy S~Omn and reduces
to the Landau-Lifshitz model in classical and continuum limit, one
defines a coupling constant ~ [9] which relates [10] to go as ~ =
cot- 1 (-tg ) [11]. The spin-tXYZ model (1.24) maps by Jordan-Wigner
o
transformation [14] and continuum limit to the MTM, and it then proves
that for the MTM-quantum s-G equivalence iyo = n -~. So go = 0 is
Yo = 4n in the quantum s-G model. Because s-G is covariant one finds
a "bare" Yo renormalises to

y" (1. 25)

o

Thus Yo = 4n is y~ = Sn, and it is at this value one finds there

are no quantised breather states of quantum s-G [5]. Thus go = 0
(free fermions) in the MTM corresponds to no quantised breathers of
quantum s-G. Similarly Yo = 0 (free bosons) in the s-G corresponds
to go = 00 in quantum MTM. This is the infinitely attractive massive
fermion problem and compares with c = 00 in the repulsive NLS, which
is the so-called 'impenetrable bose gas' [15].

Finally, for quantum sinh-G Yo < 0, and for Yo < 0 one would expect
any related fermion model to be repulsive. In particular for -4n~0< 0
and ~ = cot- 1 (-tg o ) we find 0> go> -00, repulsive, but on a different
branch of ~(n <~ <3n/2) compared with that for the attractive MTM
(n/2 < ~ < n). Korepin's repulsive MTM [13] with his different relation
[11] between g and ~ has 0 >g >-n which means 4n<yo< Sn(n/2>~>0).
o 1 0
If however ~ = cot-
(-tg ) then 4n < y < Sn means -<x> < go< 0, repulsive,
o 0
but on the branch of ~ connecting smoothly to the attractive case.
Without further investigation it therefore seems that Korepin's repulsive
MTM is not equi va'lent to quantum sinh-G case. We shall not explore
the MTM-s-G equivalence in the repulsive regime any further here. We
confine attention to quantum sinh-G which is still to be explored as
an MTM.
The renormalisation of Yo shown in (1.25) arises in connection
with the renormalisation of mass m in the (relativistic) quantum s-G
model [16]; it appeared first in the work of DHN [17] and necessarily
appears in equivalent form in the BA theory of the MTM [10]. Evidently
the corresponding result for quantum sinh-G is [lS,19]

255
 1''' = l' [1 + l' /871]-1, (1. 26)
o 0 0

and in this case the only free particle model seems to be where 1'0=0
(§5). Still we shall see that for quantum sinh-G with Yo > 0 there
is still a description in terms of bosons and a second, equivalent,
description in terms of fermions. Similarly, although quantum repul-
sive NLS is a boson model with commutation (1.5), it has an equivalent
fermion description. It was this fermion description which was used
in the work of [1] on the quantum statistical mechanics.

These bose-fermi equivalent descriptions are equivalent. Thus

in §5 we shall show how to follow out the work of [1] wholly in a
boson description. This equivalence apparently depends on the fact
that the classical integrable model are embedded in infinite dimensional
Lie algebras. The simplest is an embedding in an "untwisted loop"
algebra [20]

. abcLc m,n E ~ (1. 27)

1 ( m+n

in which (abc are structure constants of a finite dimensional Lie

algebra g, dimension dim g. All the classical models we have mentioned
have the Lie algebra g = s~(2, cr) which has dim g = 3 and satisfies

( 1. 28)

If A E ~ and La E g, then the algebra of AmL a is isomorphic to (1.27).

For dim g = 00 an algebra g~ (00) [21] or a larger algebra [22] replaces
(1.27). Embedded in g~(oo) is the integrable Kadomtsev-Petviashvili
equations in two space and one time dimensions (2+1 dimensions) [21,22].
In 2+1 dimensions much remains to be understood. But g~(oo) has both
"bose" and "fermi" equivalent representations and g~(n) carries this
down to 1+1 dimensions (one space and one time dimension). Perhaps
still more important for present purposes is that the algebras (1.27)
determine the,"integrability" of the models. The Kostant-Symes-Adler
theorem [23] shows that an arbitrary loop algebra (1.27) induces a
symplectic structure with infinite sets of quantities commuting under
a Poisson bracket; it also yields hierarchies of classical integrable
nonlinear evolution equations [23]. Evidently we are concerned with
the classical integrability of the classical models and a quantum
integrability of the quantum models. We look at the former from a
wholly elementary point of view in §2 next. Later (in §4) we introduce
the quantum R-matrix first recognized by the Leningrad school in their

256
work on the quantum inverse method [2,16,24,25]. The R-matrix allows
us to define quantum integrability and Wadati and Akutsu [26] in their
lectures show how it defines the integrability of their IRF models
in two-dimensional classical statistical mechanics (2+0 dimensions).
The loop algebras (1.27) extend to Kac-Moody-Lie algebras by inclusion
of the centre commuting with all the elements of the algebra [20].
The quantum integrable models are embedded in such algebras [20].
De Vega [27] has established some particular connections between these
algebras and corresponding algebras associated with the R-matrix(also
see [25]).

2• CLASSICAL INTEGRABILITY

We shall exploit both classical and quantum integrability of the integ-

rable models in the statistical mechanics of 1+1 dimensions we develop
in these lectures. By classical integrability we shall mean 'complete
integrability' in the sense of Liouvile-Arnold [28]. Recall that
for a classical Hamiltonian system with a finite number of degrees
of freedom N, that is N q-coordinates and N p-coordinates ({ p.,q.l =15 .. ),
~ . J LJ
the system is completely integrable when there are N independent first
integrals I k , k = 1, ... ,N (say) which are in involution, that is commute
under the bracket, {Ik,I~} = O. Then if the manifold of level lines
Ik = const. is compact and connected Arnold [28] proves the motion
lies on an N-dimensional torus.

The simplest torus is a 1-dimensional torus, and the simplest

integrable system is the 1-dimensional harmonic oscillator with H
~p2 + ~w2q2 Since H =
E, the energy, the 1-0 case is always integrable.
The energy E defines the action w- 1 E =P and there is a canonical so-
called angle variable Q = wt + c. With H = wP{ Qt = wand P t = 0 1

are Hamilton's equations. Evidently p = -(2wP)' s}n Q, ~ = (2W- 1 p)'

cos Q.
1
Note that the differential 1-forms dp
1 1 1
= -w'(2P)-' dP sin Q
1 1
- w'(2P)' cos Q ~Q, and dq = W-'(2P)-' dP cos Q - W-'(2P)' sin Q dQ.
Consequently the 2-form ID = dpl\dq = dPl\dQ and dID = O. This is just
the condition for a canonical transformation [28]. One easily checks
that {P,Q} = {~~ ~~ - ~~ ~~} = 1. This is the other way (invariance
of the Poisson brackets) of looking at a canonical transformation.
-1 1 2 1 2 2 -1
The check follows from P W [2 p + 2 w q ], Q = tan (p/wq).

It is easy to extend the theory to any number of oscillators and

any corresponding number of degrees of freedom--in particular, by
not being too searching about it, to a continuously infinite number

257
of degrees of freedom. This is just the situation for a field ~(x,t):

there is a q-type coordinate ~ (x,t) for each label x, and a p-type

coordinate rr(x,t) likewise; and x is not countable in any segment
-b< x< d in JR. Thus, for the linear K-G model (1.17) with xEJR
in particular, we expect an H

00

H fW (k)P(k) dk w(k) (2. 1)

""
and a Q ( k ), 0 < Q ( k) < 21T so tha t {p ( k) , Q ( k' )} .s (k-k').

In order to show this, note that under the complex Fourier trans-
form qfx,t)-+cp(k,t) (1.17) becomes

- 2-
~tt + (w(k)) ~ o • (2.2)

On the other hand, scaling ~, H -+ j[~cpt~t + ~(W(k))2~~*]dk, where

co -co 00

cp*(k) = cp(-k), while f(drrAd~)dx f(dcptAdcp*)dk and the bracket is

-00 co -00

{cpt(k),cp*(k')} = .s(k-k'). Then f(dcptA dcp*)dk i

co 1 -co -'P
f(dPAdQ)dk where a = (2w)-'(cpt + iwcp), a* = (2w)-'(~t - iwcp*);
_00 2
then P = lal ' Q = arg a. Thus (rr(x) ,~(x)) -+ (P(k) ,Q(k)) is canonical,

while {P(k),Q(k')} = .s(k-k'), O<P(k)<oo, O<Q(k)< 21T, and

f
00 00

H I I
w ( k ) a ( k ) 2 dk f w(k)P(k) dk (2.3)
-00

which is (2.1). Notice that (2.3) is characteristic of a linear problem:

the LS model (1.18) yields the same result with w(k) = k 2 . Notice
too that the Fourier transform (rr(x),~(x)) -+ (cpt(k),cp(k)) is canonical.

That the Fourier transform is canonical extends to the spectral

transform which is also canonical [3,5,29,30]. Recall that the spectral
(or scattering) transform is used to solve nonlinear models like sinh-G
[3]. In light-cone coordinates, ~xt = m2 sinh ~ has the spectral
transform [3,5]

(2.4)
-a/ax

This eig~n-condition, with eigen- or spectral-parameter~, can be

written tv = ~v. (The hat onI, indicates only that it is a (matrix)

258
differential operator). This eigen-condition is the basis of the
inverse scattering, or inverse spectral, method for solving sinh-G.
Combined with

i sinh CP]
- cosh cP
[V1]
v2 [:~] ,t
(2.5)

1 1 m2
or Av = V,t one finds [3] that L,t [A, L] is 2 CPxt 2 sinh cP'
together with the isospectral condition ~,t = 0.

The matrix operator L is self-adjoint and has only real ~igenvalues.

These are dense on the real ~-axis. The spectral transform Lv = ~v

maps cp(x) (at fixed t) to spectral data S = {a(k) ,b(k);k E:JR} and
Ia I 2 - I b I 2 = 1. From (2.5) one finds [5]

a(k,t) a(k,O)
arg b(k,t) w(k)t + o(k), (2.6)

in which w(k) m2 k- 1 is the dispersion relation of the linear K-G

in light-cone coordinates. Surprisingly, in view of the nonlinearity
of sinh-G, one can now prove [5], that H can be written in the form

co
H (2.7)

with this w(k). k)-1 lnl a(k)1 '

One checks [ 5] that i f P(k) = ('II

constants of the motion, and Q(k) = arg b(k,t) then

co
co
J{
-co
1
d(2CPx) A d(~CP) } dx lco(dP AdQ) dk (2.8)

and the transformation is canonical: P(k), Q(k) are action-angle

variables, the phase space is ° 2. P(k)< co, 02. Q(k) <2'11 for each k,
and {P(k),Q(k ' )} o(k - k'lo One includes the coupling constant
by P(k) = (Yo'llk)~1 lnla(k) I when Yo* 1.

A second constant of the motion for sinh-G in light cone coordi-

nates then proves to be

P 2 L
co
k P(k) dk , (2.9)

and covariant combination~ are H±P: these are interpreted as

259
 H

P (2.10)

in laboratory coordinates. We have set k = (m,2/B~-2~) P(~)d~ =

P'(k)dk, and m' = 2m. The mapping ~-+k is 2:1 since both -oo<~ <0
and map to -oo<k<oo. On the other hand one checks that ~ J(dIl(x)/\
1 -1
d~(x))dx = _ {d(2 Yo I 1 -00
~~(~))/\d(2~(~)} d~ when ~= x+t, n = x-to Then

4~~n = m,2 sinh~ = ~xx - ~tt (using m' = 2m). rurther one finds y;,
the coupling constant for the covariant problem is y; = 4 yo ' More
detail is given in [5].

Next, using

1
h(~) = 2(h' (~) + p' (~)), p(U = ~(h' (~) - p' (~)) (2.11 )

one
Dropping the primes we thus have the somewhat remarkable results that
00 00

H[p] fW(k)P(k)dk p[p] = fkP(k)dk (2.12 )

The notation H[p], etc. indicates that these quantities are expressed
in action-angle variables. Under the mapping ~ -+k, Q(~)-+ Q(k), and
{P(k), Q(k')} = 6(k-k') with O~P(k)<oo, O~Q(k)< 21T. Thus H[p],etc.
depend only on the action variables.

In the same way one proves for the repulsive NLS that

N[~ ] - f~*~ dx <+ N[p] =

f P(k) dk

00

p[~] - f i~~~ dx<+ p[ p] fkP(k) dk

00

H[~] (eqn. ( 1 .2) ) +-> H[p] fW(k)P(k)dk (2.13 )

and w(k) = k 2 . The last is the energy of a free non-relativistic

1
particle of mass m = 2' a result consistent with the form of the quantum
problem (1.7). These results for repulsive NLS indicate why that
model can be thought of as a non-relativistic limit of the covariant
s-G model.

260
 However, the more remarkable feature of (2.12) and (2.13) is that
these expressions are exactly those one would find respectively for
the linear K-G and linear LS models. There is no actual incompatibility
here since the results for the linear problems are found (essentially)
by Fourier transform, those for the nonlinear problems are found by
the spectral transform. They are both canonical transforms with dif-
ferent inverse transforms. Unfortunately the very real difference
this represents becomes obscured once we embark on the statistical
mechanics. For in the evaluation of the partition function Z for
the free energy one expects to average over the action (and angle)
variables. We therefore look at this problem next.

3. CLASSICAL STATISTICAL MECHANICS

For a classical 1 D- oscillator

Z = f exp -SH[q](2n)-1 dpdq f exp -SH[p](2n)

-1
dPdQ (3• 1)

the notation follows §2 so that

H[q] H[p] wP . (3.2)

Then indeed Z = (Sw)-l from the first expression and Z (Sw)-l from
the second. Implicit is the Jacobian a(p,q)/a(p,Q) 1 for a canonical
transformation. Then the two equal measures (2n)-l dpdq and (2n)-l dPdQ
appeal to the fact that t= 1, h = 2n so (2n)-1 is h- 1 .

In the case of a field ~(x,t) with running index x (§2) one evidently
needs a continuous infinity of integrals placed in product form in
the expressions for Z. This is a functional integral. The first
expression for Z in (3.1) applies to any classical system with one
degree of freedom and given H[q]. Generalising this to q""~:x,t),

p"" IT(x,t) we thus get the functional integral

Z f ID IT ID cp exp - SH [ cp ] (3.3)

and H[cp] would be given, for example, by (1.14) in the case of s-G
or its continuation in the case of sinh-G. We can then expect to
write down the equivalent of the second form for Z in (3.1), namely

Z f ID II exp -SH[p] (3.4)

261
where, now particularly for sinh-G, H[p) is given by (2.12). However,
although (3.4) is to follow from (3.3) by canonical transformation
we must still find the correct measure ID~ for the transformed integral.
This proves to be [18)

ID~ (3.5)

and the normalisations on each element dPndQn are natural. The precise
definitions of the P n and Qn are given below (4.2), but P n ~ P(k)
(actually P(k)dk), Qn ~ Q(k) as N ~ 00.

It is now easy to evaluate both (3.3) and (3.4) for the linear
K-G equation. Some details are given in [5): one point is that as
IDTIID~ is discretized to TIdTI d~ the natural measure factors (2 n )-1
n n n
must again appear with each dTInd~n' In this way the results for both
(3.3) and (3.4) for linear K-G coincide [5) in

-1
- -6 ln ZKG (3.6)

The parameter a is a lattice parameter which arises by writing the

functional integrals for Z, as their definitions demand, as products
of a finite number N+1 of integrals and taking the limit N ~ ex>. The
system is therefore placed on a lattice with a finite number of indepen~

dent degrees of freedom N and ultimately the limits N~oo and a ~ 0 are
to be taken. The lim a ~ 0 cannot be done in (3.6), but this is because
of the physics. The classical statistical mechanics has ultra-violet
divergences which can be removed (i.e. renormalised) only in the quantum
theories.

4. CLASSICAL STATISTICAL MECHANICS OF THE SINH-G MODEL

Plainly we are now faced with a flifficulty. If we use (3.4) for Z with
H[p) given by (2.12) we can only regain the classical K-G result (3.6)!
In a quantum calculation (§§4 and 5) we would recognize this as a free-
field result. On the other hand we can actually use (3.3) directly in
the classical cases even though H[~) describes a nonlinear system. The
method is the transfer integral method (TIM) and appeals to periodic
boundary conditions on ~[5,31). The calculation has been done for
sinh-G [18), and the result of the classical analysis is the low-T
asymptotic series

262
 Lim FL- 1 -1
mB
Ir+oo
478935069186 t12 ] (4 . 1)
223 + ... + F KG ;

FKG is the classical free energy of the free K-G given by (3.6), and
-1 -1 -1
T = temperature = B , t = (MB) and M = 8myo (Yo> 0), which would also
be the mass of the kink or antikink in the s-G model. This result shows
that the free energy of the classical sinh-G model is not FKG , and it
becomes this only when Yo = O. A corollary is that for linear K-G the
periodic b.c.s. of the TIM do not change the result (3.6) for F KG . In
§3 periodic b.c.s. were not imposed for F KG .

This linear result is misleading and does not extend to the non-
linear models. The error in simply using (2.12) in (3.4) is a failure
to define with sufficient care a proper thermodynamic limit. An exactly
equivalent problem arises in constructing an applicable form of the
quantum inverse method where [24], by defining the Hamiltonian operator
H = H -~ N on a finite interval L E m (~ is a chemical potential
~

and N is the number operator of §1 and (4.6) below) the energy separation
between the vacuum 10> and the minimum of H is rendered finite.
~
The method is once again to impose periodic b.c.s. of period L. We
sketch the argument later in this section and more details can be found in
ref.[2,24-26].There we also show how the quantum inverse method is
equivalent to the BA method. Lieb and Liniger [32] were perhaps the
first deliberately to solve such a problem by imposing periodic b.c.s.,
but periodic b.c.s. lie at the heart of the BA method anyway [2].
- -1
Notice that this procedure ensures a finite particle density n = NL .

We now see that we can expect to reach a proper classical or quantum

thermodynamic limit of finite density by imposing periodic b.c.s.
Conveniently the system is first placed on a lattice, spacing a, say,
so that N particles in a period L under periodic b.c.s. mean (N+1)a
= L. A proper thermodynamic limit is then lim NL- 1 = > O. One n
1 N 1-+ 00
such limit is evidently a- However for a fleld ~(x,t) we also
need a -+ O. We shall see how this combination is achieved shortly.
First we address the problem of periodic boundary conditions itself.

The action-angle variables of §2 were all found for x E m with

decaying boundary conditions at ± 00. Since L is now a priori infinite
the particle density is zero. Evidently we need action-angle variables
under periodic b.c.s. with period L < 00.

263
 For the integrable systems it is natural to start with an integrable
lattice with these b.c.s. Izergin and Korepin [33,34] have introduced
integrable lattices reducing to s-G and sinh-G as a ~ O. We have
found action-angle variables for these lattices [18,35,36] valid to
O(L- 1 ). We sketch some results for sinh-G.

With L (N+1)a we find

H[p] w(R )P +O(L- 1 ). (4.2)

n n
n=-iN

We find it convenient to carry both n = -iN and n = +iN in (11.2) and choose N
even: the period is still (N+1)a = L. The P n in (4.2) relate to the
P(k), as L ~a>, through P n = P(R )(2nL- 1 + O(L-1))~ P(R)dR as 2nnL- 1
n III!
= k ~ k as L ~a>. Thus as L~a>, (4.2) ~ j,w(k)P(k)dk providing the
n
modes Rn properly fill the k-space as L

The modes Rn are the allowed modes. The periodic b.c.s. restrict
the allowed modes to those satisfying the integral equation [18,35]

iN
\' 6. (R ,R )P (4.3)
L c n m m
m=-iN,m*n

The analysis determines the phase shifts 6.c and for classical sinh-G
and repulsive NLS these are [18,35]

(4.4)
kw(k') - k' w(k)

(with Yo > 0 and w(k)

-2c(k - k' )-1 (4.5)

respectively. For details we refer the reader to [35] where they

are found by using classical Floquet theory on the integrable sinh-G
and NLS lattices. Notice a does not appear, i.e. lim a ~ 0 is already
taken. However, for a > 0 w(k) in (4.4) actually involves a and
this will show itself in the free field contribution F KG to the free
energy as (3.6) has already shown.

To present the classical Floquet theory [5,18,35,36] here would

divert from our main theme. Instead we attempt to motivate the all
important result (4.3) by describing the corresponding approach in

264
the quantum inverse method. This allows us to bring together the
quantum inverse and BA methods and this way we can relate both of these
to our own work [5,18,36-38].

The quantum repulsive NLS model has played a fundamental role

in the development of the quantum inverse method [2,24,25]. The sketch
of the argument we give here is for that model and largely follows
Korepin [39]. The Hamiltonian of the model is the normally ordered
operator (1.6). The commutation relation is (1.5). Evidently

(4.6)

commute with H. The corresponding BA problem is (1.7) solved in [2,32]

and used in [1].

First we need a remark about the classical NLS under periodic

b.c.s. This is solved by introducing the monodromy matrix

A(A) B(A )]
T(A) [ A E e (4.7)
etA) D(A)

(say). For an integrable lattice ~ = ~n on distinct sites n spaced

with lattice spacing a the spectral operator L, such that Lv = ~v,
of §2 is replaced by L(nIA) such that vn+1 = L(nIA)vn and a Lax pair
description can be found [24,25,35,39]. Under periodic b.c.s. ~n+N=
~n so
-iN
T(A) II L(nIA) (4.8)
n=iN-1

which is ordered from n = ~N-1 to the right (and we again assume N

even). For present purposes it is sufficient to assume that the relevant
NLS lattice is integrable when a ~ O. Then to O(a 2 )

-iy'c ~~]
(4.9)
l+hAa

while

f
xn
-1
a ~(x) dx (4.10)
x n -1

so that

(4.11 )

265
 It is now plain that the bracket condition (4.11) on the ~ induces
n
bracket relations on the elements of T(A). The Leningrad school intro-
duces the classical r-matrix r(A,~)(a 4 x 4 matrix for NLS) such that

(4.12 )

As usual ® is Kronecker product: so T(A) ® T(~) is 4 x 4 and the

right side is the usual matrix commutator. The left side is simply
the corresponding matrix of Poisson brackets. For the repulsive NLS

[!
c o o
o 1 rrn . (4.13 )
1 o
o o

There are natural 8 x 8 matrices r12(A,~) =

r(A,~) ® I and r23(A,~)

= I ® r(A,~) (I = 12 is the 2 x 2 unit matrix). The 12 and 23 notation

describes two of the three ways of embedding a 2-particle configuration
space in a 3-particle space. Evidently there is r13(A,~) also and
we find

[ r 1 3 ( A, \)), r 23 ( ~, \))] + [r 1 2 ( A, ~), r 1 3 ( A, \))] + [r 1 2 ( A, ~) , r 23 ( ~, \)) ] =0 •

(4.14 )

(4.14) is a statement on the scattering of three particles-namely that

this breaks down to a product of two-particle scatterings. It is in
fact the semi-classical limit of the Yang-Banter condition also intro-
duced by Wadati and Akutsu [26] in their contribution.

The trace of the right side of (4.12) vanishes. So

o ll(A) = TrT(A) . (4.15 )

Consequently ll(A) is a generator of classical Hamiltonians H with

a continuous infinity of commuting constants {H,ll(A)} = 0 (the use
of ll(A) = TrT('A) here should cause no confusion with the use of II
for phase shift in what follows). The existence of the r-matrix satis-
fying (4.14) guarantees the existence of a Lax representation for
the lattice if it applies exactly for that lattice, it guarantees
a Lax representation for the field ~(x,t) derived from any lattice
as a ~ 0, and it also guarantees the integrability of both in the
spectral transform sense, and their complete integrability in the
Liouville-Arnold sense. Thus (4.14) is a classical integrability
condition.

266
 Korepin's integrable NLS lattice is chosen to preserve the same
r-matrix for both the lattice and its continuum limit. The lattice
has

(4.16 )

2 -1
with P n =
[1+~cW~Wna ]2, {W~' Wm} = -1a
.!. •
8 nm and {Wn,W m} O.
The transfer matrix L(nIA) satisfies

{ L ( n I A) ® L ( mIII ) } = 8 nm [ L ( n I A) ® L ( mIII ), r ( A , II )] , (4.17)

-1
r(A,Il) = c(A-ll) IT1I (i.e. (4.13», and (4.17) implies (4.12) for
T(A) defined by (4.8).

This summarises the r-matrix theory of the classical integrable

models epitomized by repulsive NLS. The generalisation to the quantum
case and the quantum inverse method is now relatively straightforward:
~~, ~n become operators ~~, ~n and
-1
8 (4.18 )
nm a

Then L(nIA) is the operator

-ilCa~~l + 0(a 2 ) . ( 4 • 19)

1+haA

Define the ordered operator

£(mIA) ... £(nIA) . (4.20 )

The periodic b.c.s. mean T(m+N,n+NIA) T(m,nIA) so focus on

(4.21 )

for which

(4.22 )

is an operator. It can be shown that

(4.23 )

267
The matrix R(A'~) is a 4 x 4 c-number matrix

(4.24 )

where 14 is the 4 x 4 unit matrix and ITTI is defined in (4.13).

R(A,~) is the R-matrix and (4.23) are the commutation relations for
the operator elements A(A), B(A), etc. of the monodromy matrix operator
T(A). Evidently T(A) is the quantum form of (4.7) and (4.23) that
of (4.12). The matrix R satisfies the Yang-Baxter relation

(R(~,\i)®Il(I®R(A,\i))(R(A,~) ®1) (4.25 )

A solution of this relation (namely (4.24)) determines the commutation

relations (4.23). From

(4.26 )

and invariance of the trace of 4 x 4 matrices

(4.27 )

Thus ~(A) is a generator of Hamiltonian operators Hand

o . (4.28 )

The existence of R guarantees an (operator) Lax representation. Thus

the Yang-Baxter relation (4.25) is a quantum integrability condition.

For the present purposes the following results, which can be

derived in the standard way(see for example ref.[2]),are what we need.
First of all for the repulsive NLS model there are statesl{Ao};N>,
J
j = 1,2, ... , N such that

(4.29a)

(4.29b)

(4.29c)

268
and these results correspond to (2.12). The operator B(v) is a raising
operator and B(V)j{A .}jN> =j~{A .}j N+l> , so B(v) creates the eigen-
J ]
states of (4.29). The eigenvalues in (4.29) are evidently the free
field eigenvalues. However, the second result we need is that under
the periodic b.c. only certain A' are allowed, namely those satisfying
J

-iN
An 2nn L- 1 + L- 1 ~ ~(Am-An)
m=,N-l,mtn

(4.30 )

-1
We can associate An and 2nnL simply to label An by n. However,
for c = 0, ~ = 0, so we can identify An with allowed wave vectors
kn . Then

-iN
\' ~(k,k ) (4.31 )
L n m
m=iN-l,mtn
with

~ (k,k' ) - 2 tan- 1 [c/(k-k')] (4.32 )

(We note in passing a problem in maintaining the limits -~N ~n ~ >,N-l

(v

if the k n are allowed wave vectors(compare with(4.39) below)).

The results (4.29) and (4.31) with (4.32) are exactly those found
by the BA method also [2]. The BA method starts from (1.8), makes
an ansatz for ~ which solves it exactly when (4.31) with (4.32) applies,
and thus arrives at (4.29) for the eigenvalues. Thus the results
(4.29) and (4.31) bring together these two otherwise different cal-
culational procedures.

One can show that the result corresponding to (4.31) for sinh-G is
(after writing exp A for A so that the new A is a rapidity)

(4.33 )

where,

(A, A' ) sinh(A-A'+i)l)

~ -i ln (4.34 )
sirih(A-A '-i).J)

269
m1 is a partially renormalised mass [16,40], and ~ is the coupling
constant (§1) related to Yo by ~ = n(l + (yo/8n)). However, since
sinh-G is covariant it is necessary to fill a Dirac sea with particles:
this dresses the "bare" particles and the result is finally (4.31)
again [18,40] with

11 (k, k' ) = -2 tan -1 [ (4.35 )

kw (k') -

where k = m sinh A, w(k) = m cosh A and y~ is given by (1.26). In

this expression m is the fully renormalised mass [16,40] and can be
identified with the physical m for sinh-G.

The form (4.31) is therefore rather general and apparently applies

whenever the classical H can be expressed in the form (4.2), that
is in the forms of (2.12) and (2.13). However, we must still specify
-1
the branches of the tan expressions for the various 11. We choose
these in the first place to be the continuous branch running between
l1(k,k') = -2n for k -+ to 11 (k,k') = 0 for k -+ +00 at fixed k'.
When k = k', 11 = -n. We then indicate this particular choice of branch
by writing l1 f for 11 However, it is evident that l1f is a form
of the 2-body S-matrix phase shift of the chosen quantum model (refer
to reference [ 2] for quantum NLS) . The true S-matrix phase shift should
vanish for k -+ +00 and for k -+ - 00. We therefore define a phase shift

(4.36 )

with this property: e (k)=l, e (k)=O, k(0.

k~O, Evidently lib is
singular at k = k'. However, for c -+ 0, yo-+ 0 then has the semi-
classical forms for NLS and sinh-G

-2c/(k-k') (4.37)

(4.38 )

respectively. These are precisely the expressions (4.5) and (4.4)

for l1c found in the classical Floquet theoretical analysis [18,35].

The classical condition (4.3) for the allowed modes Kn found by

classical Floquet theory, and the result (4.31) found by the quantum
inverse method as sketched, or by the BA method as given in e.g. [2],
now all become consistent if 11 in (4.3) is l1 f and

P
n
= 1 K
n
allowed Pn = 0, Kn not allowed. (4.39 )

270
It is clear from (2.3) in §2 that P ~ p(R )(2TIL- 1 +O(L- 1 )) ~ P(R)dR
n n
can be interpreted as a particle number in the mode Rn. Indeed compari-
son of (2.13), (4.6) and (4.29) confirms this view. Thus (4.39) is
a fermion description. Thacker [2] shows by BA that choosing the
smooth branch for 6 is equivalent to making a fermion description
despite the bose character of the model. We can then intuit that
6 b is a description in terms of bosons so that

P m 0, 1, 2, . . (4.40 )
n n

Then Pn~ p(R)dR is the natural extension from this description in

terms of bosons to a description in terms of classical Maxwell-Boltzmann
(MB) particles. Accordingly, by using appropriate phase shifts 6 f ,
6 b or 6 c ' (4.3) generalises the BA or quantum inverse method condition
(4.31) to descriptions in terms of fermions or bosons in the quantum
case and then extends it to the classical case as well.

5. QUANTUM STATISTICAL MECHANICS

A fundamental object to consider is the quantum partition function Z:

like the classical partition function (3.3) this will be a functional
integral. We choose

Z JIDII IDcp exp S[cp] (5.1 )

S [ cp] j U
dT II ( x , 'r! cp ( x , 1) , 1 dx - H [ cp ] ] . (5.2)
o
Note that H[cp],II and cp are classical quantities: {II,cp} = o(x-x'). This
form can be reached [5,41] by starting from the quantum mechanical
Feynman propagator expressed in Hamiltonian form and performing the Wick
rotation t ~-il followed by integration on 'time' 1 from 0 to B. Since
the trace is required for Z, cp(x,O) runs to cp(x,O) as 1 ~ B Thus (5.1)
is periodic in l·of period B. But cp(X,l) is also periodic in x, of
period L, because of the thermodynamic limit. Thus (5.1) is defined
on the space-time torus -!L~ x < !L; 0 < 1 ~ B(and 1 includes r-B ).

If one notes thatJ~dl in (5.2) is really 't- 1 J~hdl and lets 11 ~ 0,

one regains (3.3) [5] from (5.1) and (5.2). Thus for models like repul-
sive NLS and sinh-G which have no soliton solutions and have action-angle
variables P(k), Q(k), k E ~ but no others, the quantum form of (3.4)
will be

271
 z f ID]J exp S [ p ] (5.3)

with the measure ID]J given by (3.5):

8
S[p] f d, [i f P(k)Q(k), ,dk - H[p] ] (5.4)
o
and H[p] is given by forms like (2.12). We should stress again that
all of these quantities are classical quantities. Moreover P(k), Q(k)
and H[p] are all real.

Now we can see that F _8- 1 ln Z is a real free energy when (and
more-or-less only when)

(5.5)

in which mn is an integer. Since n = 1, h = 2TI as noted (§3), so (5.5)

is Bohr semi-classical quantisation. If we set mn = 0,1,2, ..• (bosons)
or mn 0,1 (fermions), then P n = 0,1,2, (bosons) or P n = 0,1 (fer~

mions). This is consistent with the results of the last section. Then
from (5.3) and (5.4)

Z f ID]J exp - 8H[p] , (5.6)

together with the quantisations P n 0,1,2, .•. -or P n 0,1 and the
allowed modes

k (5.7)
n
m*n
in the two cases. Equation (4.3) for classical MB particles now fits
smoothly into this: (4.3) is (5.7) with ~c replacing ~b or ~f and
Pm is the classical action variable. Notice that (5.6) for the quantum
cases is formally equivalent to the classical (3.4). Note too that
in (5.7) it is now convenient to let m,n run from -iN to fiN rather
than to iN-l ~refer by (4.2».

The functional integral (5.6) must keep the same value evaluated
in terms of bosons or in terms of fermions. Before demonstrating
this by an actual calculation, we first evaluate Z by the very different
route pioneered by Yang and Yang [1]. Here a fermion description
was used, so we shall differ from that work by making the equivalent
analysis in terms of bosons. This calculation is reported [18,42]
and Wadati [43] has done a similar calculation, but beyond this nothing

272
seems to have been done in boson description in either the BA or quan-
tum inverse methods. In particular the status of (5.7) is that it
derives from the commutation relations (4.23) and the Yang-Baxter
relations but then emerges in fermion description. We assume it
generally true and proceed. The classical form (4.3) is proved rigorously
in the classical case [5,18,35,36,44,45]. We handle both quantum
repulsive NLS and quantum sinh-G together.

Since P is a particle number in R , P L- 1 is a number density

n -1 n_ 1 n -1 -1
(of bosons) in R. Then P L ~ P(R)L (2nL + O(L ))~ p(k)dk
n n -1 n
(say), as L ~ 00; and p(k) = P(k)L . From (2.13), and guided by
(4.29) for NLS or from (2.12) for sinh-G, we can then define energy,
momentum, and number densities through

lim EL- 1
L~oo
lim PL- 1
L-+co
J kp(k)dk

lim NL -1
L-+co
_ n -I p(k)dk . (5.8)

2 2!.
Evidently w(k) = k 2 for NLS, while w(k) = (m +k )' for sinh-G upto
ultraviolet divergence corrections (§3). Notige that the independent
variable is Rn which is assumed properly to fill the k-space with
the vectors R as L ~ We can therefore set k = h(R) and can define
a density of allowed states through dh/dR - 2nf(R). Then, dropping
the tilda again as in (5.8), (5.7) b~comes

dll (k, k ' )

2nf(k) 1 + f b dk p(k' )dk' (5.9)

We then reach a second relation between f(k) and p(k) by defining

a free energy F and minimising this: F = E - 8- 1 s and we need an
entropy S.

Following Yang and Yang [1], but referring to bosons, the number
of possible states in dk consistent with p and f is

(L(p+f)dk) !/(Lpdk)! (Lfdk)! . (5.10)

So after using Stirling's formula one finds an entropy per unit length

f [(f+p)ln(f+p) - flnf - pln p] dk . ( 5 . 11 )

273
We now minimise FL- l = (E - S-l S )L- l : SL- l is given by (5.11) and
EL- l by (5.8). A condition for minimum FL- l is that the functional
derivative C(FL- l )/cp = O. After a little manipulation one finds

C(SL- l )/Cp= In{(f+p)p-l} - inf~k llb(k,k')ln{(f(k')+p(k'»/f(k'lkik'

(5.12 )
since f relates to p through (5.9). Consequently the condition C(FL- l )/
cp = 0 becomes

f
OOd
ln {( f+ p) p-l } - Bw(k) - 21i dkllb(k,k') In{ (f+p)f-l} dk' O.
-00
(5.13 )
It is therefore convenient to set

(f+p)p-l = fp-l + 1 _ expSe:(k) (5.14 )

in which e:(k) are energies. These energies are allowed energies if,
and only if, they satisfy (5.13), and this integral equation becomes

f
00

e: (k) 1 dll b ( k , k') 1 n (1 _ e -B e: (k ' ) ) dk '

w (k) + 2nS (5.15a)
dk

while e:(k) = e:(-k).From this result, and by using (5.9), we go on to

show that the free energy FL- l per unit length is given by

f
00

lim FL- l = (2nB)-1 dk ln (1 - e - Be: (k» I (5.15b)

L-+oo
-00

where the e:(k) are determined by (5.15a). The free-field results

are (c=O for NLS, Yo = 0 for sinh-G) then just (5.15b) with e:(k)=W(k).

It is now plain that these results (5.15) in boson form have the
classical limit

f ~k
00

e:(k) = w(k) + (2nB)-1 llc(k,k') In(Be:(k'))dk' (5.16a)

-00

f
00

Lim FL- l (2 n B) -1 dk ln ( S e: (k» . (5.16b)

L -+00
-00

Yang and Yang (1) prove that the fermion form of (5.15) (equations
-1
(5.19) below) iterates to yield an iterated expansion for FL for
the quantum NLS model. There is a problem in iterating the classical
form (5.16) for the NLS model (45), but no such problem arises for

274
the classical sinh-G model. Iteration of (5.16) in this case yields
a sequence of integrals all of which can be done and the result is
exactly the TIM result (4.1) [18]. Moreover, rather than take the
classical limit of (5.15), one can derive (5.16) for classical MB
particles. Hence the low temperature asymptotic result (4.1) confirms
the consistency of all the calculations.

We now convert (5.15) to its fermion form. Introduce a chemical

-1
potential ~ at (5.12) by minimising the negative pressure -p = FL
- ~ NL - 1 = (E - 8- 1 S - ~ N) L- 1; NL - 1 = n
is given by (5.8) so that
w(k) -+ w(k) - ~ in (5.15a) and ~ n adds to the right side of (5. 15b).
From (4.57), in which t:.f is smooth,

d
dk t:. f (k, k') - 211 C( k - k ') . (5.17l

If one defines E(k) through

(5.18)

one then finds

00

w(k) - ~- (2118)-1 f
-00
~t:.f(k,k'
dk
l1n(1+e- 8E (k' ))dk' (5.1 9a)

and

f
00

~n - (2118)-1 In(1 + e- 8E (k'))dk' . (5.1 9b)

-00

Note that it is now t:. f which appears under the integral sign. For the
NLS t:.(k,k') = t:.(k-k') and dt:./dk = -dt:./dk'. With this choice equations
(5.19) are exactly the result derived in [1] by fermion description.
It is now clear that the bose and fermi descriptions yield the same
values for FL- 1 .

However, for NLS in particular there are two interesting free

particle limits. Since

(5.20 )

and this -+ 0 for c -+ 00 and - 211c(k-k') for c -+ 0, C -+ 00 yields E(k)

= k 2 _ ~, a gas of free fermions (and this is the 'impenetrable bose
gas' [1,15]), while for c -+ 0, E(k) = w(k) - ~- 8- 1 In(1+e- 8E (k))
so E(k) = w(k) - ~, for a gas of free bosons. The free boson case
follows directly from (5.15a) of course (once ~ is included).

275
 In the case of sinh-G one sees from (4.3S) that, when Yo ~ 0,
one again obtains a gas of free bosons. However since sin(y~/8)~

0, 1, 0 as Yo ~ 0, 8n/1S, 8n/7 there seems to be no free fermion limit.

From (S.19) for both repulsive NLS and sinh-G one can examine
the excitations above the quantum ground state by defining k = kF
where E(±k F ) = 0 (note that E(-k) = E(k)). Then ~ = ~F is fixed
by this condition also and E(k) > 0, Ikl >k F , while E (k) < 0, Ik!<k F .
Thus for 8- 1 ~ 0

+kF
e:( k ) -w ( k) - ~F - (2 n 8) -1 f
-kp

-1
-w(k) - ~F + (2n) (S. 21 )

The excitation energies E - Eo _ El above the ground state energy

Eo correspond to lifting a particle to the state k above kF so creating
- - p
a hole at k h , I khl < k F : then [2] El = e: (k p ) - e:(k h )':'l Mo~~over,
the ground state energy Eo can be calculated from EoL = I -FW(k)P(k)dk
-1 -k F
+ constant by solving (S.22). For at 8 = 0, (S.14) in the form
fP- 1 = exp 8£ means that P(k) = 0, Ikl >k F and (S.9) then means that
P satisfies

dt.f(k,k' )
2np(k) 1 + dk P (k' )dk' (S. 22)

for Ikl < k F . We should note however that this integral equation is
for the bose particle density p(k): we discuss elsewhere how it relates
to the fermion density used by Yang and Yang [1]. Otherwise the finite
temperature results (S.19) evidently contain all of the quantum theory
-1
of the models at 8 = T = 0 as we should expect.

6. EQUIVALENT RESULTS FOR THE s-G MODEL

The TIM result (4.1) for the low temperature expansion of FL- 1 for
classical sinh-G has the continuation in Yo to -Yo which is (4.1)
with Yo ~ -Yo i.e. t ~ -to The series in (4.1) becomes negative in
every term, but F KG , independent of y , is unchanged. However in
addition (4.1) now gains terms in e-l~t, e- 2 / t , •.. [36] which apparently
come from the kinks and antikinks: the term in e- 1 / t is [36]

276
 e - 1/ t [1 _ 28 t - S9 t 2 - 897 t 3 - 7S00S t 4 - ]
128 1024 32768 ....
(6.1 )
It would be natural to associate the series for s-G as arising from
the classical breathers but sinh-G has no breathers and the series
is the continuation from the series in (4.1).

On the other hand, the whole of the classical MB analysis which

leads to (S.16) for sinh-G has been generalised to classical s-G.
The argument uses an H[p] which includes kink, anti-kink and K-G con-
tributions but not breather contributions [36].

However, it is known that in the quantum s-G [16] (or MTM [10])
filling the Dirac sea (using the modes Kn) dresses the system to quantum
breather-like terms plus kink-antikink terms (cf. e.g. [19,37,38,46,47]).
It has been shown that the classical limit of these quantum statistical
mechanical integral equations for quantum s-G yields the classical
continuation Yo~ -Yo of (4.1) [37,38,47]. The method of §S which
leads to (S.lS) and (S.19) for quantum sinh-G and (S.16) for classical
sinh-G we call [18,36] a 'generalised BA method'. The same metpod
yields the quantum statistical mechanics of the s-G model [48]. So
indeed does the functional integral method generalising (S.6) with
(S.7). We return to calculate the functional integral (S.6) for
quantum sinh-G in the next section.

7. EVALUATION OF THE FUNCTIONAL INTEGRALS

FOR CLASSICAL AND QUANTUM Z

The classical Z for sinh-G or repulsive NLS is (3.4): it is important

to notice that

iN
H[p]
~ W(Kn)P n (7 • 1)
-iN

in which W(K ) depends on Kn' while

n

K
n k
n ~ t>c(Kn,Km)P m (7.2)
m*n

The functional integral (3.4) can be evaluated by iterating (7.2)

in (7.1) and the result for F can be put in the form [18]

277
 f f
00 00

(21TS)-1 dk In(Sw(k)) - (21TS)-2 f dq[W(q)]-l dk

-00 -00

x ~c(k,q)d(lnw(k))/dk + (21TS)-3 f dP[w(p)]-l fdq

-00 -00

x ~c(q'P)~q [[W(q)]-l fdk ~c(k,q)d(lnW(k))/dkJ

-00

2
[fdk ~c(k,q)d(lnw(k))/dkJ
-00

+ .. (7.3)

This is exactly the (formal) iteration of the classical free energy

(5.16b) with energies €(k) given by the classical integral equation
(5. 16a) !

For sinh-G the same iteration (or the iteration of the system
(5.16)) yields the asymptotic series expression (4.1) for FL- l ; F KG
is given by the first integral in (7.3). Similar work for the classical
repulsive NLS is not possible beyond a formal iteration since (5.16a)
does not iterate then [45]. Otherwise the classical functional integral
and generalised BA methods come together.

It is now plain that in the correpponding quantum analysis we

use (5.6) with (5.7) and the quantisation P n = 0,1 (fermion form)
or P n = 0,1, 2, ... (boson form). One readily finds that iteration
of (5.7) through (5.6) with either of these quantisation conditions
yields (5.19) (fermions) or (5.15) (bosons). Thus the functional
integral methods derived from (5.3) and (5.4) in terms of action-angle
variables reach, in either of these fashions, exactly the same results
as the quantum forms of the generalised BA method described in §5.

It is worth stressing that the functional integral method follows

Feynman's original approach by using the classical action throughout:
but the action-angle variables are used after canonical transformation,
and a measure problem arises and is solved (for sinh-G, NLS [18] (and
also for s-G [36])): the semiclassical quantisations which follow
discretize the action variables and this is more-or-less necessary
for a real quantum free energy F. All of the quantum mechanics created
by normal ordering then rests in the periodicity conditions (5.7),

278
namely in the 2-body S-matrix phase shifts ~b or ~f. As noted in
the theory presented in §5, ~f comes essentially from the commutation
relations (4.23) of the quantum inverse method, therefore from the
R-matrix, and so from the Yang-Baxter relations. We shall show how
this can be done directly elsewhere [48]. The corresponding analysis
for bosons has not been worked out yet (apparently).

Now how different the final form of the functional integral is

from Feynman's original form [49]. The classical action is used,
but this is in Hamiltonian form, the trivial (Bohr) quantisation condi-
tions are imposed, so no time discretization is needed to evaluate
the quantum functional integrals and the periodicity conditions carry
the deeper quantum mechanics.

It is now evident that this functional integral method extends

to quantum and classical integrable models other than sinh-G and re-
pulsive NLS, namely to all of the integrable models. Bose-fermi equi-
valence plays a fundamental role even in the classical cases. We
cite the case of classical s-G [36] as a non-trivial example of this
extension: the results of the functional integral method coincide
with those found by the generalised BA method described in §6. Results
for the quantum s-G model seem to show a wholly similar measure of
agreement [48].

Corresponding results for other integrable models (e.g. Landau-

Lifshitz, §1) are to be reported [50].

REFERENCES

[1] C. N. Yang, C. P. Yang, J. Math. Phys. 10 (1969) 1115.

[2] H. B. Thacker, Rev. Mod. Phys. 53 (1981) 253.
[3] R. K. Bullough, P. J. Caudrey, in Solitons, ed. R. K. Bullough
and P. J. Caudrey (Springer, Heidelberg, 1980) and the other
chapters there.
[4] E. K. Sklyanin, LOMI preprint E-3-1979 (1979).
[5] R. K. Bullough, in Nonlinear Phenomena in Physics, ed. F. Claro
(Springer, Heidelberg, 1985), pp. 70-102; R. K. Bullough, D. J.
Pilling, J. Timonen, in Nonlinear Phenomena in Physics,
pp. 103-128.
[6] M. Lakshmanan, Phys. Lett. 61A (1977) 53.
[7] L. A. Takhtadzhyan, Phys. Lett. 64A (1977) 235.
[8] S. Coleman, Phys. Rev. D11 (1975) 2088.
[9] M. Takahashi, M. Suzuki, Prog. Theor. Phys. 48 (1972) 2187.
[10] H. Bergnoff, H. B. Thacker, Phys. Rev. D19 (1979) 3666.
[11] Korepin [12,13] in effect uses the different relation between

279
 ~ and g o that ~ = t(n+g).
0
In [12] he treats the attractive MTM

or, since n-~ = 81 Yo (see below),

4n >Yo> 0) and in [13] he treats a repulsive MTM with -n<go< 0

(0 < ~ < n/2 or 8n >Yo >4n).

[ 12 ] V. E. Korepin, TMP (USSR) 41 (1979) 169.
[ 13 ] V. E. Korepin, Comm. Math. Phys. 76 (1980) 165.
[ 14 ] A. Luther, Chap. 12 in Solitons Ref. [3], pp. 355-372.
[ 15 ] M. Jimbo, T. Miwa, Y. Mori, M. Sato, Physica lD (1980) 80 and
references.
[ 16] E. K. Sklyanin, L. A. Takhtadzhyan, L. D. Faddeev, Theor. Mat.
40 (1 979) 194.
[ 17] R. F. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. Dll (1975)3424.
[ 18 ] R. K. Bullough, D. J. Pilling, J. Timonen, J. Phys. A: Math. Gen.
19 (1986) L955.
[ 1 9] R. K. Bullough, D. J. Pilling, J. Timonen, in Physics of Many-
Particle Systems, ed. A. S. Davydov (Ukrainian Academy of Sciences
of the USSR, Kiev, 1986).
[20 ] P. Goddard, D. Olive, Int. J. Mod. Phys. Al, No.2 (1986) 303.
[21] M. Jimbo, T. Miwa, in Vertex Operators in Mathematical Physics,
ed. J. Lepowsky, S. Mandelstam and I. M. Singer (Springer,
Heidelberg, 1984), pp. 275-290 and other papers there.
[22] V. Kac, Infinite Dimensional Lie Algebras--An Introduction,
2d ed (Cambridge University Press, Cambridge, 1985).
[23] For example S. Olafsson, R. K. Bullough, to be published.
[24] P. P. Kulish, E. K. Sklyanin, "Quantum Spectral Transform Method-:
Recent Developments", in Proc. of the Tvarminne Symposium,
Finland, 1981, ed. J. Hietarinta and C. Montonen (Springer,
Heidelberg, 1982).
[ 25] L. D. Faddeev, in Proc. Ecole d'Ete de Physique Theorique, Les
Houches 1982, ed. R. Stora and J. B. Zuber (North-Holland,
Amsterdam, 1983).
[ 26] M. Wadati and Y. Akutsu, Exactly Solvable Models in Statistical
Mechanics, in this volume.
[27] H. J. de Vega, in Proc. Symp. on Topological and Geometrical
Methods in Field Theory (World Scientific, Singapore, 1986), in
press.
[28] V. I. Arnold, Mathematical Methods of Classical Mechanics(Springer,
Heidelberg, 1978).
[ 29] R. K. Dodd, R. K. Bullough, Physica Scr. 20 (1979) 514.
[30] L. D. Faddeev, Chap. 11 in Solitons Ref. [3], pp. 339-354.
[ 31] D. J. Scalapino, M. Sears and R. S. Ferrell, Phys. Rev. 86 (1972)
3409.
[ 32] E. H. Lieb and W. Liniger, Phys. Rev. 130 (1963) 1605.
[ 33] A. G. Izergin and V. E. Korepin, Lett. Math. Phys. 5 (1981) 199.
[ 34] A. G. Izergin and V. E. Korepin, in Problems in Quantum Field
Theory and Statistical Physics 3, ed. P. P. Kulish and V. N. Popov,
LO~I Vol. 120 "Nauka" Leningrad.
[ 35] Yi Cheng, Ph.D. Thesis, University of Manchester (1987).
[36] J. Timonen, M. Stirland, D. J. Pilling, Yi Cheng and R. K. Bullough,
Phys. Rev. Lett. 56 (1986) 2233.
[37] J. T. Timonen, R. K. Bullough and D. J. Pilling, Phys. Rev. B34
(1986) 6525.
[ 38 ] J. T. Timonen, ~. K. Bullough and D. J. Pilling, Classical Limit
of Bethe Ansatz Statistical Mechanics for the Massive Thirring
Model, to be published.
[39] V. E. Korepin, in Completely Solvable Systems in Field Theory,
Lecuure Notes at SERC-LMS Symposium, University of Durham (1986).
[40] D. J. Pilling, R. K. Bullough and J. Timonen, to be published.
[41] C. Itzykson and J. B. Zub'2r, Quantum Field Theory (McGraw-Hill,
New York, 1980)

280
[ 42) J. Timonen, D. J. Pilling and R. K. Bullough, in Coherence,
Cooperation and Fluctuations, ed. F. Haake, L. M. N~rducci and
D. F. Walls (CUP, Cambridge, 1986), pp. 18-34. Unfortunately,
in this paper a copying error replacing k = h(K) by K = h(k)
introduced sign errors in the equivalents of our equations
(5.9) and (5.22).
[ 43) M. Wadati, J. Phys. Soc. Japan 54 (1985) 3727.
[ 44) R. K. Bullough, D. J. Pilling and J. Tim~nen, in Magnetic
Excitations and Fluctuations, ed. S. W. Lavesey, U. B~lucani,
F. Borsa, V. Tognetti (Springer, Berlin, 1984), pp. 80-85.
[45) R. K. Bullough, D. J. Pilling and J. Timonen, in Dynamical Problems
in Soliton Systems, ed. S. Takeno (Springer, Heidelberg, 1985),
pp. 105-114.
[ 46) S. G. Chung, Y. C. Chang, Phys. Rev. Lett. 50 (1983) 791.
[471 N. N. Chen, M. D. Johnson and M. Fowler, Phys. Rev. Lett. 56
(1986) 907; 56 (1986) 1427 (Erratum).
[ 48) Yu- zhong Chen, D. J. Pilling, R. K. Bullough and J. Timonen,
to be published.
[49) R. P. Feynman and A. R. Hibbs, Quantum Mechanics and Path Integrals
(McGraw-Hill Book Co., New York, 1965).
[50) Yu-zhang Chen, D. J. Pilling, R. K. Bullough and J. Timonen,
to be published.
[ 51 ) M. Karowski, H. J. Thun, J. T. Truong and P. H. Weiss, Phys. Lett.
67B (1 977) 321.

281
Exactly Solvable Models In Statistical Mechanics
M. Wadatjl and Y. Akutsu 2
1 Institute of Physics, Collep;e of Arts and Sciences,
University of Tokyo, Mep;uro-lm, Tokyo 153, Japan
2Institute of Physics, Kanap;awa University, Rokkakubashi,
Kanap;awa-ku, Yokohama 221, Japan

Recent studies on exactly solvable models in statistical mechanics are

reviewed. A brief summary of the quantum inverse scattering method is
given to emphasize the soliton theoretic aspect of the theory. Intro-
ducing a class of lattice models called the IRF models, it is shown that
there exists an infinite number of exactly solvable models in 2-dimen-
sional statistical mechanics. Significances both in physics and mathe-
matics are discussed.

1. INTRODUCTION

Twenty years have passed since the discovery of soliton [1]. Advances
during the last two decades may be summarized as follows.

(1) Ubiquity of soliton equations. Canonical soliton equations, such

as the Korteweg-de Vries equation, the nonlinear Schrodinger equa-
tion and the sine-Gordon equation, appear in almost all branches
of physics.
(2) Systematic methods to solve soliton equations. The soliton equa-
tions can be solved by analytical methods such as the inverse scat-
tering method, the Backlund transformation and Hirota's method.
(3) Universality of soliton picture. Soliton system as a completely
integrable system includes solvable models in quantum field theory,
quantum spin systems and statistical mechanics.

In this lecture note we review our recent contribution to the theory

of exactly solvable models in statistical mechanics [2-11]. By doing
this as elementary as possible, we like to convey our finding that the
extension of the soliton theory implied in (3) is surprisingly wide and
deep. Before we proceed to the main theme, we briefly summarize the
quantum inverse scattering method [12]. This way of writing may be help-
ful for readers to understandtrefundamental strategy of the theory.

282
 An operator version of the auxiliary linear problem is

0, ( 1.1)

Here Q(x, A) and GL(x, A) are M x M matrix operators, and A is the spectral
parameter. We assume the boundary conditions:

(1. 2)

where I is an identity operator and TL(A) is called the transition matrix.

For later discussions, it is more transparent to formulate the quantum

inverse scattering method (QISM hereafter) on lattice. The system size
2L is divided into 2N intervals each of which has a length ~ = 2L/2N.
The local transition matrix Ln(A)
xn
I - f Q(x,A)dx (1. 3)

xn-~

describes the change of the Jost function matrix GL(X,A) over the n-th
interval. In terms of the operator Ln(A), the transition matrix TL(A)
is expressed as

(1.4)

For a quantum integrable system we can associate Ln(A) such that

R(A,lJ)-[L (A) ® L (lJ)] (1. 5)

n n

where symbol ® denotes the direct product of the matrices. The c-number
matrix R(A,lJ), of course, depends on the model. The relation (1.5) is
referred as the (local) Yang-Baxter relation. If Ln(A)'S with different
n commute, we further have

(1.6)

This is the (global) Yang-Baxter relation.

The QISM provides not only a powerful method to study completely

integrable systems, but also a unified viewpoint on the structure of
solvable models in (l+l)-dimensional quantum field theory and 2-dimen-

283
sional statistical mechanics. We list the consequences of the Yang-
Baxter relation.

(1) The relation (1.6) indicates the existence of an infinite number

of conserved quantities as seen from the relation

[TO.), T(jJ)) 0, T {).. ) (1. 7l

Further, off-diagonal elements of (1.6) yield an algebraic formu-

lation [12-14) of the Bethe ansatz method.

(2) When we regard the elements of L n (\) and R(\,jJ) as those of scat-
tering matrix (S-matrix), the relation (1.5) is nothing but the
factorization equation for the S-matrix [15).

(3) When we consider L n (\) as vertices of a vertex model in statistical

mechanics, (1.5) is a condition that the transfer matrices T(\)
Tr TL (\) with different A'S commute, which implies the mod~l is
exactly solvable.

It is to be remarked that we have already suggested a common feature

of exactly solvable models, that is, the commutability of the "transfer
matrices." We shall explain this in detail in Section 3.

2. IRF MODELS

To be concrete, we shall mainly consider a type of statistical mechanical

models in two dimensions, called IRF (interaction round a face) models
[16). Let us introduce an IRF model. Spins 0i (oi = 0, 1,
k-l) are located on the lattice points (sites) of a square lattice and
the Boltzmann weight is assigned on each face (or plaquette) depending
on the spin configurations round the face. By ((a,b,c,d), we denote
the energy ofa plaquette with spin configuration (a,b,c,d). The corres-
ponding Boltzmann weight is

w(a,b,c,d) (2 •1)

where kB is the Boltzmann constant and T is the temperature (Fig. 1).

Let N be the particle number. The partition function ZN' the free
energy per particle f and the spin density <0 1> (the average of a
certain spin, say 0 1 ) are defined by

284
 d c Fig. 1 Boltzmann weight w(a,b,c,d)

W( a,b,c ,d) =

a b

(2.2)
ZN E E II w(ai' OJ' ok' °1) ,
°1 oN (i,j,k,l)

f - kBT lim N- 1 log ZN' (2.3)

N->""

1
<°1> (2.4)
ZN

Here, the product is over all faces of the lattice and the sum is over
all values of all spins.

Exact solvability of a model in statistical mechanics means that

we can evaluate physical quantities such as the free energy and the
spin density without any approximation. It is known that the Ising
model, the next nearest neighbor Ising model, the 6-vertex model, the
8-vertex model and the 3-spin model are exactly solvable [16]. Note that
the IRF model is not special but v.ery general. Most of the exactly
solvable models are expressed in the form of IRF models. For instance,
the Ising model is defined by

e: (a,b,c,d)

- ~[(2a-1) ~2b-1 )+(2b-1) (2c-1 )+(2c-1) (2d-1 )+(2d-1) (2a-1) 1, (2.5)

a, b, c, d, = 0, 1,

and the 8-vertex model is defined by

e: (a, b, c, d)

= -J(2a-1)(2c-1) - J'(2b-1)(2d-1) - J 4 (2a-1)(2b-1)(2c-1)(2d-1),

a, b, c, d, = 0, 1. (2.6 )

285
In general, the model is called k-state model when spin variables can
take k values, a i = 0, 1, k-1. The Ising model and the 8-vertex
model are, as seen from (2.5) and (2.6), two-state models.

3. TRANSFER MATRICES

Exactly solvable models in statistical mechanics are used to be counted

on the fingers. This situation has been drastically changed by the
following two discoveries:
(1) Concept of the commuting transfer matrices,
(2) Evaluation of physical quantities via the corner transfer matrices.

The row-to-row transfer matrix V has elements (Fig. 2)

n
V II w(a j' a j+ 1 ' a ;+1 ' a;) , (3.1)
aa' j=l

where a {a 1 ' ... , an}' a' {a i' ••• I a~}, a n + 1

(J'
n

(J (J
n n+1

Fig. 2 Row-to-row transfer matrix Vaa'

Similarly, we define V' with w replaced by w'

n
V' II w'(OJ" a J·+ 1 , a J"+l' a J'.). (3.2)
aa'
j=l

Then, the elements of the matrix product VV' are

(VV') , V V'II ,
aa aa a a

n
II X ( a J" a J'!, a J" I a J' + 1 ' (3.3)
a,] j=l

where
X(a,b,c I a' ,b' ,c') w(a,a' ,b' ,b)w' (b,b' ,c' ,c). ( 3.4 )

286
We regard X(a,c I a' ,c') as the k x k matrix with element X(a,b,c I
a' ,b' ,c') in row b and column b' (Fig. 3). Then, (3.3) can be written
as

,
(VV') , Tr X(o" 0; 102' 02)X(02,02 103' 03) ... X(on,onl 0,' 0;).
00

(3.5 )

c c'

w'
b r - - - - - - j b'

w
Fig. 3 X(a,b,c I a' ,b' ,c') in (3.4)
a a'

Similarly, we define X' with wand w' interchanged in (3.4). Then we

have

(V'V)oo'

(3.6)

From (3.5) and (3.6), we see that V and V' commute if there exist
k x k matrices M(a,a') such that

X(a,a' Ib,b') = M(a,a' lX' (a,a' Ib,b') [M(b,b') ]-'. (3.71

Multiplying M(b,b') from the right and writing the element (c,d) of
Mla,a') as w"lc,a,d,a'), we obtain

E wlb,d,c,a)w'(a,c,f,g)w"(c,d,e,f)
c

E w"(a,b,c,g)w'(b,d,e,c)w(c,e,f,g). (3.8)
c

Equation (3.8) is a sufficient condition for the commuting transfer

matrices and will be referred as star-triangle equation (Fig. 4).

287
~ = ~
c c

Fig. 4 Schematic explanation of the star-triangle equation (3.8)

d-c

d c

d-a c-b

Fig. 5 Wu-Kadanoff-Wegner
transformation
a b

Commutability of the transfer matrices is a common feature of exactly

solvable models. By Wu-Kadanoff-Wegner transformation (Fig. 5), the
star-triangle equation for vertex models becomes

L w ( fl , a I Y, fl" ) w' (v, yl a, v" ) w" ( v" , fl" I v' , fl ' )

Yfl"V"

L w" ( v, fll v", fl" ) w' ( fl" , al Y, fl' ) w (v" , Yla, v' ) . (3.9)
Yfl"V"
The relation (3.9) is schematically shown in Fig. 6. Furthermore,
when we set an arrow on each bond of vertex and regard the vertex as
trajectories of two particles (Fig. 7), (3.9) is a condition imposed
on the 3-body scattering from initial state (v,fl,a) to final state
(fl' ,v' ,a):

L ffY S ,ya S" ~' "

Yfl"V" fl~' vv l1
v " fl'

S"fl V" s,ay sya

L (3.10)
Yfl"V" Vfl" fl"fl' \} "\} I

This relation is called factorization equation for the S-matrix.

288
 ))'
v V'
))"

y
1/

/I \1
IJ I I

IJ IJ
Q.

Fig.6 The star-triangle equation for vertex model

Fig. 7 2-body scattering S~~

1J

We have shown that (3.8), (3.9) and (3.10) are equivalent. Baxter
noticed the importance of (3.9) in the study of 8-vertex model [17]
and Yang [18] introduced (3.10) as a consistency condition of the Bethe
ansatz method [19]. After them, these equations in general are often
called Yang-Baxter relation as already used in Section 1.

We consider corner transfer matrices. The lattice is divided into

four quadrants as indicated in Fig. 8. The boundary spins are fixed
to their ground state values. For spin configurations (Fig. 9) in
the lower-right quadrant, we define corner transfer matrix A:

(3.11 )

Note that the spins a = {01' ... , am} and a' = {ai, ... , o~} are
not summed over. Similarly, we introduce corner transfer matrices
B, C and D in the upper-right, upper-left and lower-left quadrants.
The partition function ZN and the spin density <01> are expressed
as

289
 C B
0- 1m

0- '11 ()
II

b.
I
0'"1
6.
f).

Il Il
0 A
a-m 6.
Il A

-
Fig. 8 Fig. 9

Fig. 8 The lattice is divided into four quadrants. The center

spin is 0 1 •

Fig. 9 Corner transfer matrix Aoo '. Boundary spins denoted

by ~ are fixed to their ground state values.

Tr(A BCD), (3.12 )

lim Tr(01 ABC D)/Tr(A BCD). (3.13 )

m-+oo

Corner transfer matrices can be diagonalized in the thermodynamic

limit m -+ 00 using the star-triangle equation and the analytic and perio-
dic properties of the Boltzmann weights [16]. In contrast with the
row-to-row transfer matrix the eigenvalues of the corner transfer mat-
rices remain discrete. Recently, Thacker [20] found a link to the
field theory that corner transfer matrix may be interpreted as the
boost operato~ [21].

4. ROGERS-RAMANUJAN IDENTITY AND PARTITION THEORY

One of the fascinating aspects which we encounter in the study of IRF

models is a close relation to the theory of partitions in number theory.
There exists a direct correspondence of the conditions on the spins
and the constraints on the partition of natural numbers. Appearance

290
of the Rogers-Ramanujan type identities in the evaluation of the spin
density through the corner transfer matrix is a consequence of this
correspondence.

Celebrated Rogers-Ramanujan identities [22] are

co
G(q) 1 + 1: 2 n
n=1 (1-q)(1-q ) ... (1-q )

IT (4.1 a)
n=O ( 1_ q 5n+ 1 ) ( 1_ q 5n+4) ,

co
H(q) 1 + 1:
n=1 (1-q)(1-q 2 ) .•. (1-q n )

co
IT (4.1 b)
n=O

It is a famous story that in 1917 S. Ramanujan who had known (4.1)

incidentally found a paper by Rogers in the Proceedings of London
Mathematical Society. We shall give a short introduction to the theory
of partitions [23], which leads to the proofs of (4.1) and its ex-
tension.

A partition of a positive integer n is a finite nonincreasing

sequence of positive integers whose sum is n. The number of partitions
of n is denoted by P n . For instance, we have

3 + 1 + 1
2 +
3
P3 3.

It is useful to introduce a notation that makes explicit the number

of times that a particular integer occurs as a part:

(J (J2 (J (J
(11 2 3 3 4 4 . • . ). (4.2)

We often deal with problems that the number of partitions is enumerated

under restriction (restricted partition problems). We denote the number

291
of partitions of n under restriction R by Pn(R). It is known [23]
that
Pn(all partitions with distinct parts)

P n (all partitions with odd parts). (4.3)

For instance, we have

6 (1 2 3) = (1 5) (2 4) = (6),

which confirms (4.3) for n 6.

The partition problem is called equipartition if for two restric-

tions Rand R' it holds that

Pn(R) Pn(R' ) for all n. (4.4)

Define the generating function Z(q,R) by

00

Z(q,R) n
L q Pn(R) . (4.5)
n=O

In terms of the generating functions, equipartition condition (4.4)

is expressed as

Z(q,R) Z(q,R') . (4.6)

The following identity is useful for later discussions:

00

L Pn(R)qn II (1 _ qn) -1 . (4.7)

n=O nER

A simple example of (4.7) is

00

L P (all partitions with odd parts).qn

n
n=O
00

II (1 _ q2n-1)-1. (4.8)
n=l

We are in a position to prove Rogers-Ramanujan identity (4.1).

Define a function F(oo; q) by

(4.9)

292
where the sum r' is over 01' 02' ... under the constraints

R: °i = 0, 1 and ° < °i + °i + 1 ~ 1, i > 0. (4.10)

By a calculation, we see that F(O;qi and F(l:q) coincide with the

infinite series of (4.1) respectively. We can prove [23] that
the partition with the restriction R in (4JO) is equipartition to
that with restriction R' defined by

=
° for i 0, ±(2 - 00) mod 5

0· 0, 1, 2, ..• , otherwise. (4.11 )

~

Applying the identity (4.7) to (4.10) which is equivalent to (4.11),

we obtain the infinite products of (4.1). Thus, the Rogers-Ramanujan
identities (4.1) are expressed as

II (1 _ qn) -1 . (4.12 )
n*0,±(2-0 0 )mod 5

In 1961, B. Gordon [24] discovered a generalization of the Rogers-

Ramanujan identities. Define

(4.13 )

where the sum r' is over 01'02' •.• under restriction R given by

R: 0i = 0, 1, 2, ••• , (k-l)

and ° ~ 0i + 0i+l < k - 1, i > 0. (4.14 )

Equipartition holds for the restriction R and the following restric-

tion R'

R': 0i = ° ~or i = 0, ± (k - 00)mod (2k+l),

0, 1, 2, ..• , otherwise. (4.15 )

Then, we arrive at the identity

00

II (1 _ qn)-l. (4.16 )
n=l
n*0,±(k-o O)mod(2k+l)

293
This is called Gordon's generalization of the Rogers-Ramanujan identi-
ties. The case k = 2 corresponds to (4.12).

Ramanujan further discovered a list of identities including

P(x)
(4.17)
P( x3 ) ,

00

P(x) II (1 _ x 2n - 1 ) .
n=1

5. GORDON'S GENERALIZATION HIERARCHY

We go back to the IRF models. As shown in Section 3, if the Boltzmann

weights satisfy the star-triangle equation (3.8), the transfer matrices
commute and then the model is solvable. This fact offers an extremely
powerful method to construct exactly solvable models. Finding of
solvable models consists of two steps.

(1) Introduce a model with appropriate physical requirements (symmetries,

number of spin states, etc).
(2) Solve the star-triangle equation (STE, for short) for the model.
In general, k-state IRF model has k4 independent Boltzmann weights
and the number of the STE to be satisfied is k 6 . As k increases, the
system of equations is extremely overdetermined and does not have
interesting solutions. Therefore, the setting of physical requirements
is crucial in this game.

We introduce the spectral parameter

w w(u), w' w(u + v), w" w(v) . (5•1)

Then, the STE (3.8) reads as

L w(a,b,g,'f;u) w(f,g,d,e;u+v) w(g,b,c,d;v)

g

L w(g,c,d,e; u) w(a,b,c,g; u+v) w(f,a,g,e; v). (5.2)

g

We assume diagonal exchange symmetry among the Boltzmann weights

w(a,b,c,d; u) w(c,b,a,d; u)

w(a, d, c, b; u). (5.3)

294
 In 1980, Baxter [25] found that the hard hexagon model is exactly
solvable. In the hard hexagon model, spin variable can take either
of ° or 1, and the Boltzmann weight is non-zero only if there are no
neighbouring pairs of 1-spins. Then, the physical requirements on
the model are

0, 1, (5.4a)

for neighbouring spins 0i and OJ. (5.4b)

With the hard-core condition (5.4b), spin-1 particle looks "hexagon

shape molecule" on the lattice. The model has been successfully applied
4
to the adsorption problem of He monolayer on carbon surface [16].

Due to the hard-core condition (5.4b) and the symmetry (5.3),

the hard hexagon model has 5-independent Boltzmann weights:

W1 w(O, 0, 0, 0; u),

w(O, 0, 0, 1; u) w(O, 1, 0, 0; u),

w(O, 0, 1, 0; u) w(l, 0, 0, 0; u),

W4 w(O, 1, 0, 1; u),

w(l, 0, 1, 0; u). (5.5)

The STE to be satisfied by these Boltzmann weights are

W1Wi Wl + w3 w4w3 w2 wjw Z

w3 w jwl + w5 wi w 3 3 Z'
w1 w w

w1wi W3 + w3 w'w
,4 5
l1
3 Z'
w4 w w

3+
w3 w j w w5 wi ws S Z'
w2 w w

w3 wi w3 + w5 W 4s
W w4 wSw4, (5.6)

where the arguments of wi' wi and wi are u, u+v and v, respectively.

The solution of these functional equations is expressed in terms of
elliptic theta function. It is convenient to employ the definition

295
 sinu n (1 2n 4n 2n
2p cos2u + P )(1 - P ). ( 5.7)
n=1
2
We hereafter write H(u) for H(u;p). The solution is

H(3" - u)
H(A - u)
H(3,,) H(A)

H(u) H(4" - u)
1
[H(,,)H(n)]2
H(4,,)

H(n - u)
(5.8)
H(n)

where
n/5. (5.9)

Using an addition formula

H(a+x)H(a-x)H(b+y)H(b-y)

H(a+y)H(a-y)H(b+x)H(b-x) + H(a+b)H(a-b)H(x+y)H(x-y), ( 5.10)

we can prove that the solution (5.8) satisfies the STE (5.6).

The hard hexagon model has four regimes; I(the vacuum), II(the
triangular ordering), III(the vacuum) and IV(the square ordering). The
2
parameter u and p measure the anisotropy of coupling constants and
the deviation from critical temperature. In regime I, we use x and
w defined by

2
P - e -e: , x = w e 2nu/e: . ( 5 . 11 )

By using the analytical properties of corner transfer matrices and

the explicit forms of the Boltzmann weights, we can show [16] that
in regime I the spin density is given by

<°1> (5. 12a)

x6 , - xG(x)/H(x). (5.12b)

Here, F(oO;q), G(x) and H(x) have been introduced in (4.9), (4.1a)
and (4.1b). Using (4.12) and (4.17) in (5.12), we arrive at

296
 <°1> (S. 13)
P(x}

We can analyze the critical behavior exactly. For instance, critical

indices across the I-II regime boundary are a = 1/3 and B = 1/9.

The hard-hexagon model is a 2-state model where spin variable 0i

takes 0 or 1. We introduce a k-state model

0· 0,1, . .. , (k - 1) (S. 14a)

1

subject to the constraint

k - 1 for adjacent spins 0. and (S. 14b)

1

It should be remarked that the constraint (S.14) corresponds to the

restriction (4.14) which leads to the Gordon's generalization of the
Rogers-Ramanujan identities. We call the k-state model defined by
(S.14) GG(k} model. The number of independent Boltzmann weights is
k(k+l }(k+2}(k+3}/24. We succeeded in finding the solution of the STE
for the GG(3) [2], GG(4} [S,7] and GG(S} [8,11] models by solving
37, lS8 and S10 functional equations. We also presented a method to
construct the Boltzmann weights for the GG(k} model [11]. A beautiful
mathematical structure of the Boltzmann weights was found in the cons-
truction.

As an example, we write down the solution for the GG(3} model

which are obtained by solving 37 independent STE's.

H(S)..-u}H(3)..-u} H(2)..} H(u}H()..-u},

w(O,O,O,O;u}
H(S)..}H(3)"} H()"}H(3)..}2

w(l,O,O,O;u} w(O,O,l,O;u}

£ [H(2)"}]~ H(u}H(4)..-u}
1 H(3)..} H()..}H(4A}

w(2,0,0,0;u} w(0,0,2,0;u}

H(u}H(3A-u}
£2 1
[H(A}H(3)..} ]2H(2A}

297
 w(O,O,O,l;u) w(O,l,O,O;u)

H(A-U)H(4A-U)
H(A)H(4A)

w(O,O,O,2;u) w(O,2,O,O;u)

H(A-U)H(5A-U)
H(A)H(5A)

H(3A-U)H(4A-U)
w(l,O,l,O;u)
H(3A)H(4A)

H(2A-u)H(3A-U)
w(2,O,2,O;u)
H(2A)H(3A)

H(5A-U)H(4A-U)
w(O,l,O,l;u)
H(5A)H(4A)

H(GA-u)H(5A-U)
w(O,2,O,2;u)
H(6A)H(5A)

w(l,O,O,l;u) w(l,l,O,O;u) w(O,l,l,O;u)

€ H(u)H(A-u)
w(O,O,l,l;u)
2 [H(A)H(2A)]t H (3A)

w( 1 ,O,2,O;u) w(2,O,1,O;u)

H(u)H(6A-U)

w(O,2,O,1;u) w(O,1,O,2;u)

H(A-U)H(2A-u)
H(A) H(2A)

w(O,l,l,l;u) w(l,l,O,l;u)

H(u)H(5A-U)

w(l,O,l,l;u) w(l,l,l,O;u)

H(A-U)H(3A-U)
H(A) H(3A)

298
 w(l,l,l,l:u)
H(3A-u)H(5A-u)
(5.15 )
H(3A) H(5A)

where €i = ± 1 (i=1,2,3) and the parameter A (the crossing point) is

TT/7. (5.16 )

Independently, this solution has been found by Baxter and

Andrews [26]. For the general k, the Boltzmann weights are expressed
in polynomials of products of (k-1) elliptic theta functions and A is
given by [11]

1..= TT/(2k+1). ( 5.17)

Thus, there exists a series of exactly solvable models, that is, the
GG(k) model for k = 2 (hard hexagon model), 3, 4, 5, . . . . We call
it Gordon's Generalization (GG) hierarchy. The finding of the GG
hierarchy is a result of our belief that exactly solvable IRF models
are closely related to the theory of partitions.

6. GRAND HIERARCHY OF EXACTLY SOLVABLE MODELS

Besides the GG hierarchy explained in the previous section, we know

a hierarchy of solvable models which we call ABF hierarchy [27].
The ABF hierarchy is also closely related to the theory of parti-
tions. Originally, those models (restricted 8V 50S models) are defined
as the solid-on-solid (50S) models where the height variables {hi}
are assigned on the lattice sites. In terms of the spin variables
Vi}' the model is defined by the restrictions

0,1,2, ... , (k-1), (6.1 a)

k-2 <0. + O. < k-1 for adjacent spins 0; and OJ'. (6.1 b)
- ~ J ~

Recently, we discovered a new hierarchy (KAW hierarchy hereafter)

where the model i,s defined by [6]

O· 0,1,2, ... , (k-1), (6.2a)

~

k-2 <0. + o. < k for adjacent spins o. and o .. (6. 2b)

- ~ J - ~ J

Furthermore, we found a hierarchy (the special 52 hierarchy) where

the constraints are the same as (6.1) but the highest spin 0 = (k-1)
is replaced by a doublet [4].

299
 Those hierarchies are classified by the following constraints:
(1) number of states:

0,1,2, ... , (k-1), (6. 3a)

(2 ) hard-core conditions for adjacent spins and OJ:

°i

L < + 0· < L + f, (6.3b)

°i J

(3 ) additional constraints on adjacent spins 0. and 0·

1 J
for f > 3. (6. 3c)

We consider an (L,f) plane, L ~ 0, f ~ 1 (see Fig. 10). The ABF

and KAW hierarchies correspond to f = 1 and f = 2 line, respectively.
The L = 0 line is the GG hierarchy. Based on this observation, we
have predicted (6) that there exists an exactly solvable model at every
lattice point in the (L,f) plane. A set of those models which contains
00 x 00 models will be referred to as Grand Hierarchy.

The existence of the Grand Hierarchy was confirmed as follows (9).

We start from the eight vertex solid-on-solid model (8VSOS model,
for short) which is also an IRF model (Fig. 11). Each height variable
hi takes integer values under constraints

Ih.1 - h·1
J
for adjacent heights hi and h j . (6.4)

From the homogeneous model, we can construct the Z-invariant 8VSOS

model (the term HZ-invariant" by origin means that the partition func-
tion ZN remains invariant under the shift of lines forming the irregular
lattice (28).1. We construct the Z-invariant model on the square lattice

5
I
X

4
I
X

3
I
X Fig. 10 The (L,f) plane showing

2
I
X
the Grand Hierarchy. The
symbols (x), (0) and (e)
correspond to the GG,
ABF and KAW hierarchies,
o 2 3 4 5 L respectively.

300
 I I I
-+- ----- -1-- - - --- -I--
I I I

I
I
I
I
I
I
d c
I
I I
I I I
I I
I U U+(O-I) ..
I
I I
-+-
I
- - - - - Vj- Uj ----- -1--
I
~J

I I
I I
I I
I
I
I I
I I
I I
I

I
I
I I
U u+ A
-+- - - - - - --I-- i-- - - - - -r--
I I I

u-(O-I)'>" u- .>.. U

a b
Fig _ 11 Fig _ 12

Fig. 11 A Z-invariant 8VSOS model_ Height variables are assigned

on the sites of the lattice formed by the full lines_
To each dashed vertical (horizontal) line ~i (~;), spectral
parameter u. (v.) is assigned. The spectral parameter~ of
1 J
the IRF plaquette which contains the intersection of
~i and ~; is Vj - u i -

Fig. 12 Boltzmann weight wQQ(a,b,c,d;u) of the IRF model. The

symbol. denotes the summation over the "inner" height.
The weight does not depend on the edge heights denoted
by O.

in the following manner. On a plane we place M vertical lines{~j}

and N horizontal lines {~~} forming a square lattice. To each line
J
~i(~i) we assign a spectral parameter u i (vi)' An inhomogeneous 8VSOS
model is introduced on the dual square lattice. The spectral parameter
of an IRF-plaquette which contains an intersection point of ~i and ~;

is assigned to be v. - u .. We call the resultant 8VSOS model the Z-

J 1
invariant 8VSOS model. The homogeneous model is recovered by setting
Vj = u (all j) and u i o (all i).
We consider Q x Q periodic reduction for the Z-invariant 8VSOS
model, which means that we set (see Fig. 12)

UJA = i - mod Q,

(V.-U)/A j - 1 mod Q. (6.5)

J

301
We also set (r positive integer, r > 5)

IT/r, (6.6)

and regard Q as f in (6.3). From the edge-height independence (Fig.12),

resultant Q x Q blocks can be considered as plaquettes of an IRF model
in the (L,f) plane. Hence the partition function of the IRF model
is the same as the Q x Q periodic Z-inavriant 8VSOS model. Thus,the
solvability of the Grand Hierarchy is a consequence of the periodic
reduction of the Z-invariant 8VSOS model [9].

It is an extremely interesting fact that the special S2 hierarchy

[4] is not included in the Grand Hierarchy. It may be possible that
we extend the special S2 hierarchy into the Z-invariant model and
again by the periodic reduction we construct a new set of solvable
hierarchies.

7. SUMMARY AND FURTHER DEVELOPMENTS

(1) We have shown that an infinite number of solvable models in

statistical mechanics, at least ~ x ~ number of models, can be cons-
tructed by solving the star-triangle equation. Thus, the existence
of the Grand Hierarchy has been established [6,9]. Significance of
this achievement in statistical mechanics cannot be overemphasized.
Physical applications, for instance, to the interfacial problems are
very interesting.

(2) Let us relate our discussion to the conformal theory [29,30].

The conformal theory enables us to predict the possible values of cri-
tical index n in (l+l)-field theory or 2-dimensional statistical mecha-
nics. Since any analytic function of the coordinate z = x + iy is
the conformal transformation, conformal group in two dimensions is
infinite dimen,sional. Generators Ln' n = 0, ±1, ±2, •.• , of'the trans-
formations satisfy the Virasoro algebra [31]:

c 3
(n-m)L n + m + T2(n - n)6 n + m,0· (7.1)

From the representation theory of the Virasoro algebra [32] and the
physical requirements such as unitarity [30], the value of central
charge c is restricted to

302
 (1 ) c > 1

(2) c = 1 - 6
m 2, 3, 4, (7.2)
m(m+1)

Further, in the case of (2), the possible value of critical index n

is given by [30]

n 2(h+h),

2
[ (m+ 1 ) p - mq] - 1
h, h m 2, 3, 4,
4m(m+1 )

< p < m+ 1, < q < p. (7.3)

It is quite interesting that the critical index n calculated from

the exact values of a and B for the ABF hierarchy through a scaling
relation n = 4B/(2-a) agrees with the one predicted by (7.3) with p
= q [33]. In other words, the ABF hierarchy is a partial realization
of the conformal theory. The detailed analysis of the Grand Hierarchy
is under progress and at the moment we predict that there exists an
exactly solvable model in each universality class.

We note that while the conformal theory deals with the phenomena
only at the criticality the exactly solvable models can offer infor-
mation both on- and off-critical temperatures.

(3) As we have seen in Section 5, the Rogers-Ramanujan type identi-

ties appear in the spin density calculation by the corner transfer
matrix method. Conversely, we expect that there exists a solvable
model corresponding to any Rogers-Ramanujan type partition identity
[2,34]. We remark that elliptic functions (more generally, modular
functions) play an important role again in the exact theory of statis-
tical mechanics as we experienced in the soliton theory. It is interes-
ting that the modular functions are closely related to the partition
theory and the representation theory of Kac-Moody algebra [35].

(4) Very recently, we have found that the study of exactly solvable
models in statistical mechanics has deep connections with the von Neumann
algebra theory and the knot theory [10]. In 1984, v. Jones [36] dis-
covered new polynomial invariant (Jones polynomial) which is more power-
ful than the Alexander polynomial to classify knots and links. A
central role is played by the algebra A defined by
q,n

303
 2
e.
1

e.e. > 2, (7.4)

1 J

A certain trace is associated with the algebra A making it the von

q,n
Neumann algebra. Due to the existence condition of the trace, the value
of , in (7.4) is restricted to

(1) ,-1>4

(2 ) , -1 4 cos 2 (lT/k), k 3, 4, ... . (7.5)

On the other hand, the star-triangle equation for the ABF hierarchy
(restricted 8-vertex 80S model with the height variable h. = 1,2, ... ,
J
r-1) at criticality yields the Temperley-Lieb algebra {U.} [37]:
J

U~ 1/2u
J q j'

U.U.
1 J
U.U. ,
J 1
Ij-il > 2, (7.6)

with q given by

q (7.7)

We see from (7.4) and (7.6) that both algebras are the same when
1/2
we set U. q e j and
J

, -1 q. (7.8)

Moreover, it is known [33] that the (r-1 )-state 8V80S model exhibits
the generic (r-2)-fold multicriticality whose exponent realizes the
discrete series (7.3) of the Virasoro algebra through the correspondence

m r - 1. (7.9)

Combining the above results we arrive [10] at a novel relation between

the discrete series of the Virasoro algebra and the indices ,-1 for
sub factors of I I I factors [38]:

m+ 1 = r k. ( 7 . 10)

304
Note that the critical 8VSOS hierarchy intermediates the completely
different objects in mathematics.

More recently, we have found that new invariant polynomials for

knots and links can be constructed from solvable (vertex and IRF) models
describing critical statistical systems [39].

In conclusion we summarize this note by emphasizing that the soliton

theory including solvable models in statistical mechanics has much
more wide and deep extensions in physics and mathematics than we expected
a few years ago.

References

[ 1] N. J. Zabusky and M. D. Kruskal, Phys. Rev. Lett. 15 ( 1965) 240.

[ 2] A. Kuniba, Y. Akutsu and M. Wadati, J. Phys. Soc. Jpn. 55(1986)1092.
[ 3] Y. Akutsu, A. Kuniba and M. Wadati, J. Phys. Soc. Jpn. 55(1986)1466.
[ 4] Y. Akutsu, A. Kuniba and M. Wadati, J. Phys. Soc. Jpn. 55( 1986) 1880.
[ 5] A. Kuniba, Y. Akutsu and M. Wadati, J. Phys. Soc. Jpn. 55(1986)2166 ..
[ 6] A. Kuniba, Y. Akutsu and M. Wadati, J. Phys. Soc. Jpn. 55(1986)2605.
[ 7] A. Kuniba, Y. Akutsu and M. Wadati, Phys. Lett. 116A ( 1986) 382.
[ 8] A. Kuniba, Y. Akutsu and M. Wadati, Phys. Lett. 117A ( 1986) 358.
[9] Y. Akutsu, A. Kuniba and M. Wadati, J. Phys. Soc. Jpn. 55(1986)
2907.
[ 10 ] A. Kuniba, Y. Akutsu and M. Wadati, J. Phys. Soc. Jpn. 55(1986)3285.
[ 11 ] A. Kuniba, Y. Akutsu and M. Wadati, J. Phys. Soc. Jpn. 55(1986)3338.
[ 12 ] L. D. Faddeev, Sov. Sci. Rev. Math. Phys. Cl (1981) 107,
H. B. Thacker, Rev. Mod. Phys. 53 (1981) 253;
P. P. Kulish and E. K. Sklyanin, Lecture Notes in Physics
(Springer,Berlin, New York, 1982), Vol. 151, p. 61.
M. Wadati, in Dynamical Problems in Soliton Systems, ed S. Takeno
(Springer,Berlin, 1985), p. 68.
[ 1 3] L. A. Takhatajan and L. D. Faddeev, Russian Math. Survey 34(1979)11.
[ 14 ] H. J. de Vega, in Dynamical Problems in Soliton Systems, ed S.
Takeno (Springer,Berlin, 1985), p. 81.
[ 15] M. Karowski, H. J. Thun, T. T. Truong and P. H. Weisz, Phys. Lett.
67B (1977) 321,
A. B. Zamolodchikov and A. B. Zamolodchikov, Ann. of Phys. (N.Y.)
120 (1979) 253,
K. Sogo, M. Uchinami, A. Nakamura and M. Wadati, Prog. Theor. Phys.
66 (1 981) 1 284 ,
K. Sogo, M.. Uchinami, Y. Akutsu and M. Wadati, Prog. Theor. Phys.
68 (1982) 508.
[ 16] R. J. Baxter, Exactly Solved Models in Statistical Mechanics
(Academic Press, London, 1982).
[ 17] R. J. Baxter, Ann. of Phys. (N.Y.) 70 (1972) 193.
[ 18] C. N. Yang, Phys. Rev. Lett. 19 (1967) 1312.
[ 19] E. Lieb and W. Liniger, Phys. Rev. 130 (1963) 1605,
B. Sutherland, Phys. Rev. Lett. 20 (1968) 98.
[20] H. B. Thacker, Physica 180 (1986) 348.
[ 21] K. Sogo and M. Wadati, Prog. Theor. Phys. 69 (1983) 431.
[22] L. J. Rogers, Proc. London Math. Soc. 25 (1894) 318,
S. Ramanujan, Proc. Camb. Phil. Soc. 19 (1919) 214.
[23] G. E. Andrews, The Theory of Partitions (Addison,London, 1976).

305
[24] B. Gordon, Amer. J. Math. 83 (1961) 393.
[25] R. J. Baxter, J. Phys. A13 (1980) L61.
[26] R. J. Baxter and G. E. Andrews, J. Stat. Phys. 44 (1986) 249.
[27] G. E. Andrews, R. J. Baxter and P. J. Forrester, J. Stat. Phys.
35 (1984) 193.
[28] R. J. Baxter, Philos. Trans. R. Soc. London A289 (1978) 315;
A. B. Zamolodchikov, Sov. Sci. Rev. A2 (1980) 1;
R. J. Baxter, Proc. Roy. Soc. London A404 (1986) 1;
J. H. H. Perk and F. Y. Wu, Physica 138A (1986) 100.
[29] A. A. Belavin, A. B. Zamolodchikov and A. M. Polyakov, J. Stat.
Phys. 34 (1984) 763;
A. A. Belavin, A. M. Polyakov and A. B. Zamolodchikov, Nucl. Phys.
8241 (1984) 333.
[30] D. Friedan, Z. Qiu and S. Shenker, Phys. Rev. Lett. 52 (1984)
1575.
[31] M. Virasoro, Phys. Rev. Dl (1970) 2933.
[32] V. G. Kac, Lecture Notes in Physics (Springer,Berlin, 1979),
Vol. 94, p. 441;
B. L. Feigin and D. B. Fuchs, Funct. Anal. Appl. 16 (1982) 114.
[33] D. A. Huse, Phys. Rev. 830 (1984) 3908.
[34] C. A. Tracy, Preprint (1986).
[35] J. Lepowsky and R. L. Wilson, Proc. Nat. Acad. Sci. USA 78(1981 )7254.
[36] V. Jones, Bull. Amer. Math. Soc. 12 (1985) 103.
[37] H. N. V. Temperley and E. H. Lieb, Proc. Roy. Soc. London A322
(1971) 251.
[38] o. Bratteli and D. W. Robinson, Operator Algebras and Quantum
Statistical mechanics I, II (Springer,Berlin,New York, 1979,1981).
[39] Y. Akutsu and M. Wadati, "Knot Invariants and the Critical
Statistical Systems," J. Phys. Soc. Jpn. 56 (1987) 839.

306
 Part V

Applications:
Physics and Biology
Solitons and Some Other Special Solutions
in Field Theory
K. Babu Joseph
Department of Physics, Cochin University of Science and Technology,
Cochin 682022, India

Solitons in gauge theories such as monopoles and vortices are dis-

cussed. The applications of the Hirota bilinear operator and group
analysis methods for the SU(2) Higgs model are described.

1. INTRODUCTION

In field theory a soliton characterises a stable and extended solution

of a nonlinear field equation, having finite energy or action, whose
existence may sometimes be related to the topological properties of the
fields at spatial infinity. Though solitons arise in certain gauge as
well as scalar field theories we shall confine ourselves with gauge
theory solitons. Magnetic monopoles are the most widely studied soli-
tons of this category. In contrast to the singular point monopole dis-
covered by Dirac [1] in abelian gauge theory, the ' t Hooft-Polyakov
non-abelian monopole [2] is a finite energy static field configuration
carrying magnetic charge. It is a solution of the SU(2) Yang-Mil ls-
Higgs model, where the gauge group is SU(2), which is spontaneously
broken to U(l) by a triplet of Higgs scalar fields. Solitons carrying
both magnetic and electric charges, called dyons, have also been cons-
tructed [3]. However, monopole and dyon solutions can be expressed
in closed analytic form only in a special limit.

Vortices or strings form another class of gauge theory solitons. These

objects contain quantized units of magnetic flux. The abelian Higgs
model possesses a finite energy vortex solution [4]. A scale-dependent
vortex solution has been obtained recently [5], and it is the first
solution of its kind to be derived. It is important to note that a pure
gauge theory, abelian or non-abelian, has no finite energy solutions.

Finite action solutions of four-dimensional non-abelian gauge

theories in Euclidean space-time are called instantons [6]. Since
instantons are localized in time as well as space, they do not qualify
as particles. In the corresponding quantized theory instantons tunnel
between different gauge vacua.

308
 All of the foregoing types of soliton have associated topological
conservation laws. But in some models solitons arise from the con-
servation of an ordinary charge, and these are called non-topological
solitons [7]. Instantons and non-topological solitons will not be
treated in these lectures. The topics we discuss here include appli-
cation of the Hirota bilinear operator [8] and group analysis [9] methods
to the construction of soliton and some time-dependent solutions of
the SU(2} Higgs model. Among these are the celebrated Prasad-Sommerfield
(PS) monopole [10] and related solutions, which were discovered by mere
guess-work. No systematic derivation of these solutions has so far
been given in the literature. This explains the importance of the
approaches herein presented. There is a short section on vortices also.
A general familiarity with gauge theories at the classical level is
assumed on the part of the reader.

2. TOPOLOGICAL CRITERION FOR SOLITON SOLUTIONS

The existence and stability of a topological solution are guaranteed

by the conservation of a topological quantity called the winding number

or Pontryagin index. Two maps from a topological space to another are

said to be homotopic, if one can be continuously deformed into the other.
A homotopy class is an equivalence class of maps which are homotopic
to one another. Each such class is labelled by a winding number which
is conserved, implying that maps belonging to two distinct homotopy
classes are not deformable into each other.

To illustrate the winding number idea, let us consider the SU(2}

Higgs model mentioned in the introduction. It is described by the
Lagrangian

~ =_ ~ Fa F~va + ~ 0
4 ~v 2 ~cp
a - V( }
cp (2.1 )

where the Fa are .the three components of the SU(2} field strength tensor
~v

(2.2 )

D~ is the gauge covariant derivative operator, defined by

o a abc
~cp =0 cp + gE b A cp , (2.3 )
~ a c ~

and V(cp} is the Higgs potential,

309
 V(cp) A( 2
"4 cp (2.4)

with 2 a a
cp cp cp

For m2 2 2
> 0, the minimum of V(cp) corresponds to cp = m IA. By
convention, the classical vacuum is identified to be at spatial infinity,
and thus

as r ... 00. (2.5)

This condition defines a map of a sphere at spatial infinity to a sphere

of radius milA in the field space spanned by the Higgs fields cpa. We
2 2 2 2
denote this map by S ... S. Now, the homotopy classes of the map S ... S
form a group called the second homotopy group, IT 2 (S2), and IT 2 (S2) = Z,
the group of integers. Each element of IT 2 (S2) defines a homotopy
class and represents a winding number. Hence the winding numbers of
the present model are integers.

The total energy of any solution of the above model is

E (2.6)

For a static solution of the Yang-Mills fields the energy will be finite
2 2
if and only if V(cp) vanishes or cp ... m IA as r ... 00. Hence, a way is
open to classify the solitons according to their winding number. The
n=1 soliton is a magnetic monopole which generates (asymptotically) the
magnetic fields,
1\
1 ri
B. (2.7l
1 g 2"
r

where £ is a unit vector. The stability of a monopole sterns from the

topological conservation law, ~n = O. The conservation of magnetic
charge is a consequence of the Higgs field topology.

3. HIROTA'S BILINEAR OPERATOR METHOD:

CONSTRUCTION OF PS MONOPOLE SOLUTION

The limit, m2... 0, A ... 0, keeping m2 /A finite, is called the PS limit

in which an explicit, static monopole solution was obtained [10] by
ingenious guess-work. In this section we discuss an application of
Hirota's method [8] to a systematic derivation of the PS monopole and
other related solutions [11].

310
 The bilinear operator method of solving a nonlinear differential
equation consists in expressing the dependent variable h as the ratio
of two functions, h = g/f. When this ratio is substituted in the origi-
nal equation, an equation with two dependent variables, g and f, is
obtained. The derivatives of functions can always be expressed in
terms of Hirota's bilinear derivatives, defined by

(~x - ~x' )f(x,t)g(x' ,t) I . (3. 1)

x=x'

The nonlinear equation for g and f is split into two coupled equations,
and the functions g and f are expanded as power series in a parameter
E. The individual functions in the power series are evaluated by suc-
cessively integrating the differential equations that follow from equa-
ting the coefficients of equal powers of E on either side of each
of the split equations. Solutions can be obtained either by terminating
the series by some technique, or by actual summation.

Some useful identities that follow from (3.1) are listed below:

n
Ox g. 1 (3.2)

L(g/f) (3.3)
ax

2
~(g/f) (3.4)
ax

On(g.f) (3.5)
x

Turning to the monopole problem, the equations of motion derived

from (2.1) reduce to the following form, when a suitable ansatz [2]
which is spherically symmetric and static, with winding number 1, is
made use of:

(3. 6a)

H(2K 2 + ~H2 _ m2 r2), (3.6b)

g

where K and H are both functions of r, and a prime denotes differen-

tiation with respect to the argument. The energy integral is

311
 E
4n j dr {( K I ) 2 + (rH I - H) 2
g2 0 2r2

(3.7)

For finiteness of this integral, the necessary boundary conditions are

H ..... 0, K ..... 1, as r ..... 0 (3. 8a)

H ..... ~ r, K ..... 0, as r ..... 00 • (3.8b)

/>-
In the PS limit eqs. (3.6) become

(3.9a)

(3.9b)

One makes the dependent variables transformation:

K(r) A(r)/B(r), H(r) C(r)/B(r). (3.10)

Using (3.4), eqs. (3.9) are rewritten in the form

( 3.11 a)

(3.11b)

where D2 = 02
r'
a second order bilinear operator.

In the next step, (3.11b) is split into two coupled equations with
the help of a function G(r), which will be determined later:

(3.12a)

o. (3.12b)

One readily verifies that solutions to (3.12) are solutions of the

original set (3.11). Although other splitting patterns may also be
effective, the present one is advantageous in reducing the degree of
nonlinearity from three to two. (3.11a) now becomes

(3. 12c)

312
 The functions A, Band C are expanded as power series in a small
parameter £:

A(r) £A, (r) + £ 2 A2 (r) + (3. , 3a)

B(r) , + 2
£B, (r) + £ B2 (r) + (3. , 3b)

C(r) , + £C, (r) +

2
£ C2 (r) + (3. , 3c)

Inserting (3.'3) into (3.'2) and comparing the zeroth power of £ on

both sides, one finds that G(r) = 0 for consistency. Extending the
same procedure to higher powers of £ yields equations of the following
type:

o (3. '4a)

2A,(C,-B,). (3. '4b)

A consistent set of solutions of (3.'4) is the following:

acr 2 ( 3. , 5a)

C, cr+d, C2 = (2bc-N 2 )r 2 /2!,

2 2 2 2 2 3
C3 = -d(2bc-N )r /2! + [(c-2b)N + 2c(a +b )]r /3! ( 3. , 5c)

a 2)~ .

A solution of (3.9) by the above procedure is labelled by five

parameters a, b, c, d and £. However, by suitable choice of these
parameters, one can make summable series out of (3.'5). There are two
trivial solutions too. For, setting a = b = c = d = 0 gives

K 0, H 1. (3. '6)

Similarly, the choice a = b, c = 0 yields

K nr/( '+nr), H '/(1+n r ) (3.17)

313
where n = ac/(l+d), an arbitrary constant. Both of these solutions
represent point monopoles of infinite energy.

For a * b, c 0, and I dc I < 1,

A(r) car, ( 3. 18a)

B( r) ~ (b sinhSr + N coshSr), (3.18b)

where S Nc/(l+dc). Due to the large arbitrariness of d and c, one

can take S to be independent of a and b. After a straightforward cal-
culation, the series for C(r) becomes

C(r) = ~[(N - bSr)coshSr + (b - Nsr)sinhSr1. (3. 18c)

Transforming back to the original pair of dependent variables, one

finds

K(r) aSr e Sr /(a 2 e 2Sr - 1) ( 3. 19a)

H (r) (3.19b)

with a = a/(b -~), which is arbitrary.

This coincides with the
general point monopole solution obtained by Ju [121. Setting a= e Y
(say), one recovers the Protogenev solution [131. The most interesting
solution emerges with a 1, which is the PS monopole solution. It
is, however, not possible to sum the series for B(r) and C(r) for non-
zero c.

The Hirota method provides a unified scheme for deriving the PS

and some other solutions, which were originally obtained by trial and
error. Nevertheless, the procedure fails to yield closed expressions
in the general case away from the PS limit.

4. GROUP ANALYSIS OF SU(2) MONOPOLE EQUATIONS

The group analysis of partial differential equations (POE) is a standard

procedure [91 that exposes their invariances and often, leads to solu-
tions. For a second order PDE,

o (4•1)

314
a similarity transformation x-+x*, t-+t*, u-+u*, is one which leaves
the equation invariant. In infinitesimal form one writes

x* x + e: X(x,t,u) + O( l) (4. 2a)

t* t + e:T(x,t,u) + 0(e: 2 ) (4.2b)

u* u + e:U(x,t,u) + 0(e: 2 ), (4. 2c)

where X, T, U denote the infinitesimals of the transformation. From

these equations it follows that

2 2
u(x+e:X+O(e: ); t+e:T+O(e: )) u + e:U + 0(e: 2 ). (4.3)

Expanding, and equating the O(e:) terms on either side,

U. (4.4)

This is called the invariant surface condition and yields the charac-
teristic equations:

dx/X dt/T du/U (4.5)

which imply,

dx
f(x,t,u) (4. 6a)
dt

du (4.6b)
g(x,t,u).
dt

When (4.6) is independent of u, one deduces the solution,

x (4. 7a)

u (4.7b)

where k1 and k2 are arbitrary constants. One of these constants, say

k 1 , plays the role of a new independent variable, henceforth denoted
by p; then k2 becomes the dependent or similarity variable; thus,

u(x,t) F( p) • (4.8)

On inserting this into (4.1) there results an ODE for F(p).

315
 From a knowledge of the infinitesimals X, T, U, one can construct
a set of infinitesimal operators which generate the infinitesimal simi-
larity transformations and the Lie algebra corresponding to them. It
characterizes the group of invariances G of the system of equations,
and the procedure herein described, is called group analysis. Its
usefulness in field theory is clear from the possibility of generating
time-dependent solutions of the SU(2} Higgs model [14]. The time-
dependent equations of this model in the PS limit are

r2(K Ktt } K(K 2 + H2_ 1 } (4. 9a)

rr

r2(H - Htt } 2HK2. (4.9b)

rr

One defines a generic dependent variable u a (a = 1,2) such that

u 1 = K and u 2 = H, and considers a one-parameter family of infinitesimal
transformations in r, t, and u a with associated infinitesimals R, T
and ua , respectively.

To ensure the invariance of the system (4.9), the second derivatives

of u a must transform according to

a* (4.1 Oa)
ur*r*

a* (4.1 Ob)
ut*t*

where [U a ] denotes a second extension [9].

xx

When the transformed system corresponding to (4.9) is written in

view of (4.10), and coefficients of terms of order E are equated to
zero, one finds

r2[U~r] - r2[U~t] + 2rR(K rr - Ktt } + u 1 (1-3K 2 -H 2 }-2U 2 KH) = 0

(4.11 a)

O. ( 4.11 b)

From these equations one picks up the coefficients of different orders

of derivatives of K and H, and sets them equal to zero. This leads
to a large number of determining equations which are consistently solved
to yield the infinitesimals:

R 2;\rt + ltr, (4. 12a)

(4.12b)

0, (4. 12c)

316
where A, K, and a are constants. The occurrence of three independent
parameters in these equations, permits one to define the three genera-
tors Ga as follows:

2
G1 2rt a + (r + t 2 )L (4. 13a)
ar at

G2 a
rL + tat (4.13b)
ar

G3 a , (4. 13c)
at

which satisfy the Lie algebra:

(4.14 )

The group analysis approach yields a set of time-dependent solu-

tions. To this end, one considers a subgroup G 1 c G as well as
itself, defines a similarity variable in each case, and solves the
corresponding similarity-reduced equation.

Under the full group G , with A * 0, K *0 and a

similarity variable is

p (4.15)

The corresponding similarity-reduced equations have the same form as

that of the static 8U(2) Higgs model in the P8 limit, (3.9), with r
being replaced by p. This directly gives the solution

K(p) Cp/sinh(Cp) , (4. 16a)

H( p) Cplcoth(Cp), (4. 16b)

which coincides with that reported in ref. (15). One, however, obtains
a new solution also:

K( p) p/(A + p), (4. 17a)

H( p) A/(A + p), (4.17b)

where A is a nonzero arbitrary constant. Both K(p) and H(p) are singular
on the surface A + P = o.

When a subgroup G1c G, defined by setting K a = 0, is considered,

the similarity variable is

317
 TI r/(t 2 - r 2 ), (4.18 )

and the reduced system is again of the PS form, (3.9), with p ~ TI .

In addition to a PS-like solution, which is reported in ref. (15),
one obtains a new solution:

Til (A + TI), (4. 19a)

AI (A + TI), (4. 19b)

with A '" O.

5. VORTICES

We shall discuss here some aspects of a scale-dependent vortex solution

recently obtained in the SU(2) Higgs system [5]. Using a suitable
axially symmetric ansatz, that satisfies the boundary condition,

K (r) ~ 1, H(r) ~ 0 as r ~ 0 (5.1)

the equations of motion in the PS limit are obtained:

rH (5. 2a)
r

rK (5.2b)
r

With the substitution

K(r) 1 - H(r), (5.3)

which is in agreement with (5.1), (5.2) are converted into a single

nonlinear equation

r2H rH 2H2(H - 1 ) . (5.4)

rr r

So far no analytic solutions of this equation have been found. The

authors of ref. (5 ) have resorted to phase plane analysis, and found
that there exists a regular solution (5.4) with the behaviour

H(O) 0, H(oo) 1. (5.5)

318
 Since (5.4) is invariant under a scale transformation, r~/a r,
the solution as well as its energy, will be scale-dependent. Thanks
to the axial symmetry, this soliton is rightly designated as a vortex
or string.

The energy per unit length of a vortex is called its tension. In

the present case, the tension is scale-dependent and is given by the
integral

2 (5.6)
E -2lTJ r dr,
0
or

E ':!!"-J dr[3r2(H )2 - 4rHH + 4H2 - 4H 3 + 3H 4 ] , (5.71

2 3 r r
e 0 r

which has been evaluated only numerically [5] for different a values.
The scale-dependent vortex, as this solution may be called, is stable
because, even though a may be continuously varied, the value a = 0
is forbidden. The reason for this is that at this point, the boundary
behaviour at r (5.5) is altered.

The magnetic flux contained in the scale-dependent vortex, is cal-

culated from the formula,

J 2lTr dr F 312 , (5.8)

o
which gives

<I> = -2lT/e, (5.9)

as is the case with the Nielsen-Olesen string [4].

6. CONCLUSION

In these lectures we have discussed the concept of finite energy soli-

tons in gauge theories at the classical level and the application of
two well-known mathematical methods--Hirota's method and group analysis--
to the construction of some of these solutions in closed form. Much
work has been done in recent years to find multi-soliton solutions in
field theory. For instance, multivortex solutions where n vortices
are superimposed in an axially symmetric configuration have been dis-
covered [16]. However, there exists no formula for a general multi-
vortex solution. On the other hand, static multi-monopole solutions in
the PS limit have been obtained by methods of algebraic geometry [17]

319
and using Backlund transformations [18]. The soliton concept has per-
meated into supergravity and Kaluza-Klein theories and has even been
suggested as a model of baryons.

References

h!] P. A. M. Dirac, Proc. Roy. Soc. A133 (1931) 60; Phys. Rev.
74 (1948) 817.
[2] G. 't Hooft, Nucl. Phys. B79 (1974) 276; A. M. Polyakov, JETP
Lett. 20 (1974) 194.
[3] B. Julia and A. Zee, Phys. Rev. 011 (1975) 2227.
[4] H. B. Nielsen and P. Olesen, Nucl. Phys. B61 (1973) 45.
[5] Prakash Mathews, C. M. Ajithkumar and K. Babu Joseph, Phys. Lett.
171B (1986) 64.
[6] A. A. Belavin, A. M. Polyakov, A. S. Schwartz and Yu. S. Tyupkin,
Phys. Lett. 59B (1975) 85.
[7] R. Friedberg, T. D. Lee and A. Sirlin, Phys. Rev. 013 (1976)
2739.
[8] R. Hirota in Solitons edR.K. Bullough and P. J. Caudrey (Heidelberg:
Springer-Verlag, 1980).
[9] G. W. Bluman o.nd J. D. Cole, Similarity Methods for Differential
Equations (New York: Springer-Verlag, 1974).
[10] M. K. Prasad and C. M. Sommerfield, Phys. Rev. Lett. 35 (1975)760.
[11] C. M. Ajithkumar, Ph.D. Thesis, Cochin University of Science
and Technology (1986) (unpublished).
[12] 1. Ju, Phys. Rev. 017 (1978) 1637.
[13] A. P. Protogenov, Phys. Lett. 67B (1977) 62.
[14] K. Babu Joseph and B. V. Baby, J. Math. Phys. 26 (1985) 2746.
[15] W. Mecklenburg and D. P. O'Brein, Phys. Rev. 018 (1978) 1327.
[16] H. J. de Vega and F. A. Schaposnik, Phys. Rev. 014 (1976) 1100.
[17] R. S. Ward, Comm. Math. Phys. 79 (1981) 317.
[18] P. Forgacs, Z. Horvath and L. Palla, Phys. Lett. 99B (1981) 232;
Phys. Lett. 102B (1981) 131.

320
Solitary Waves of the "2-Dimensional Ferromagnet" *
R. Rajaraman
Centre for Theoretical Studies, Indian Institute of Science,
Bangalore 560012, India

Finite energy solitary wave solutions obtained by Polyakov and Belavin

for the isotropic 2-dimensional ferromagnet are briefly reviewed. The
ideas of Bogomolnyi connecting the winding number and potential energy
of the system play an important role in the discussion.

I will derive here some exact solitary waves solutions of the so-called
nonlinear a-model in particle-physics parlance. The system consists
of three coupled fields ~a(X,t), a = 1,2,3, in 2+1 (two-space, one-
time) dimensions. This system can also be viewed, as I hope to show
you, as a simple model of an isotropic 2-dimensional ferromagnet. These
solutions were obtained in the late seventies by Polyakov and Belavin
[1] using very elegant methods. The ideas of Bogomolnyi [2] will also
play an important role in our discussion. These solutions have been
known now for several years and have also been covered in reviews [3].
Nevertheless I decided to discuss them in this school for the following
reason. Most participants here are from either the applied mathematics
or plasma physics wings of the soliton community. Most of you are very
familiar, in great detail, with 1+1 dimensional systems, such as the
KdV, the sine-Gordon, the nonlinear SChrodinger equation, and with their
remarkable exact soliton solutions. But if we relax the requirement
of exact solitons (i.e., preservation of shape and velocity even after
collisons), and settle for just solitary waves, then there are many
interesting solut~ons in higher dimensions as well. In particular,
particle physicists have found several such solutions in 2, 3 and 4
dimensions, by using some elegant analytic methods. These solutions
also possess an interesting topological index. The solitary waves of
the 2-dimensional ferromagnet, which I shall discuss here, are perhaps
the simplest examples of this type. Other examples are the 't Hooft-
Polyakov monopole, the Skyrmion, and the Yang Mills instanton[3].

* Notes . 1.
taken and prepared by M. Dan1e

321
 The equation we are going to solve is

0, ( 1)

where

a a
cp cp (xl ,x 2 ,t), a = 1,2,3.

It is understood that the repeated indices are summed over. Thus, b

is summed and a is free. This equation results from the following simple
Lagrangian,

L E
a
f (2 )

subject, however, to the constraint

E cpa(x,t) cpa(x,t) at each x, t.

a

This is the Lagrangian of a massless Klein Gordon field in 2-dimensionp,

under the condition ~ cpa cpa 1, that is, the field is not allowed
to vary freely but varies only subject to the above condition. Let
us show that the field equation (1) is the Euler-Lagrangian equation
for Lagrangian (2), in the presence of the constraint Ecpacpa = 1. The
a
constraint can be incorporated using the method of Lagrangian multipliers.
Let us write a new Lagrangian by adding the Lagrange multiplier term
to the old one,

r. (3 )

The Euler-Lagrange equation for the Lagrangian (3) is

o. (4 )

This may appear like a linear equation, but remember that we must adjust
the Lagrangian multiplier A(X,t) in such a way that the constraint is
satisfied. Rewriting (4), we have

(5 )

Multiply both sides by cpa, and sum over a.

cp aD cp a (6 )

322
Using equation (6) in (5), we get

<p
a
- (<p to b
<p)<p a 0, (7)

which is just the original equation (1).

So far nothing has been said about ferromagnets. Now one can see
the following connection with the ferromagnets. Consider first a one-
dimensional ferromagnet, that is, a lattice with a magnetic moment
<P~l at each point. Here i lebels the lattice site and a are the compo-
nents 0 f h . moment. A1 so, we h ave t h e cons t
t e magnetlC '
ralnt aa
<Pi<Pi 1,
since the individual magnets can only rotate, and not change their
magnitudes.
a
<po
l
<$ <$
i i+l

The dynamics of these magnets depends on what they energtically favour.

Energy is least when they are parallel and most when anti-parallel.
We assume that each magnet interacts only with its nearest neighbour
and that this interaction is characterized by the following potential
energy.

v (8 )

Rewriting,

v (9 )

When the lattice size is small we can take the continuum limit, by
taking <P~ ~<pa(x),
l
E. = J dx. Then the potential energy for this
one-dimensional ferroffiagnet becomes, after absorbing constants,

v ( 10)

In two dimensions the same arguments can be carried out and one will
get

v ( 11 )

This is the potential energy appearing in the Lagrangian of the system

(2). Adding to this the kinetic energy we will get the full Lagrangian
(2). The corresponding equation of motion is given in eq. (1).

323
 Equation (1) is our starting point and the task ahead is to solve
this equation. To make life simpler we will look for time-independent
solutions: ~a(xl,x2). This will be sufficient for obtaining a class
of solitary waves. These time-independent solutions will satisfy the
following equations.

o. ( 12)

a
Once the stationary solution (x 1 ,x 2 ) is found the solution in the
~

moving frame can be constructed because the equation of motion is co-

variant under (2+1)-dimensional Lorentz transformations. Hence, by
Lorentz transforming, we get the function

~
a (Xl - ut
--------2---,-'
(1 - u )2

which is also a time-dependent solution of (1).

Note: However that such transformations will generate some but not
all time-dependent solutions.

Equation (12) was solved by Polakov and Belavin using the following
clever trick. The lowest energy trivial solution for the equation is
called vacuum solution which is of the form

a
u ( 13)

where
a a
u u 1.

Thus, we have an infinite number of trivial vacuum solutions, one corres-

ponding to each possible direction the unit vector u a could take.
Next consider non-trivial x-dependent solitary wave solutions. The
energy of these solutions (entirely potential, since they are static
solutions) is

1 2 J..(V~a)2
V
"2 J d x 2

'" 211 a
1 J de {J..(~)2 + _1_( a~a) 2}
"2 J0 dr r
( 14)
2 or 2~
0 2r

In order that this energy be finite, the energy density (the integrand
in (14))should go to zero as r~"'. This yields the boundary conditions

~ 0 and ~ ~ 0 as r ~ "'. ( 15 )
ae

324
In other words, as r ~ 00 the field ~a(x) must approach some fixed
value, independent of direction. All points on the circle at infinity
must have the same value of ~a. 8uch a function ~a(x), since it iden-
tifies (that is, it has the same value at) the boundary at infinity,
compactifies physical space R2 into a spherical surface 8 2 , In the
interior of this space, ~a(x) could be any continuous function of x,
with unit modulus (~a~a 1). The allowed values of ~a at each given
x, also forms a sphere 8 2 since the modulus of ~a has to be unity.
Thus, any finite energy configuration ~a(x) represents a mapping of
one sphere 8 2 into another sphere 8 2 ,

We know from mathematics that all such mappings can be divided

into homotopy classes, where mappings within each class can be conti-
nuously deformed into one another. Each class is characterised by
a "winding number." The expression for the winding number n for these
mappings is given by

( 16 )

A useful relation connecting the winding number and potential energy

of the system was written down by Bogomolnyi. We have

2
f d x ( a cp +£
II - ll\)
!p x a cp)
\)
(a !pi £
II lla
!p x a alP > 0, ( 17)

that is,
2
fd x( a cp. a !p) + £ £ (!pxa!p) (!pxa !p > + 2 f£lI\)a ll !P. (!pxa\)!pl.. (18)
II II ll\) lla \) a ~ ~

It is easy to check that the first two terms are equal. Then using
eq. (16), one can write

( 19)

Given (1), this is equivalent to the following:

4V>+1611n,
that is,

v > h\n\. (20)

Equation (20) is the Bogomolnyi inequality. It says that the energy

of any configuration is greater than or equal to 411 times its winding
number. Now, recall the important fact that any classical static solu-
tion extremises the energy functional. In particular, if the equality

325
in eq. (20) is satisfied, then the energy is minimised in any given
topological sector. Hence, a field ~a(x) for which the equality in
eqs. (17-20) holds, is automatically going to solve the field equation
(12). This in turn requires, because of (17) that

a .a~
(21 )

Thus, instead of solving the second order differential equation

(12), we need only to solve the first order equation (21). Any solu-
tion of the latter will also be a solution of the former. This can
be verified by applying a
. on equation (21).
make the following stereographic projection.
To solve eq. (21), we

(22 )

Let w where

In terms of this, the Bogomolnyi equation (21) now becomes

(23a)

(23b)

Using the stereographic projection (22), eq. (23) takes the form

(24 )

On making use of (23), eq. (24) becomes

(25 )

Equation (25) is the Cauchy-Riemann equation for which any analytic

function w(z) (z = xl + ix 2 ) or w(z*) is a solution. Thus our original
equation is solved. In terms of the variables w = 2(~1+i~2)/(1-~3)'
and z = xl + iX 2 any analytic function w(z) or w(z*) is an exact
solution of (21), and also of the original field equation (12). The
solution can be rewritten in terms of the original field variables
~a, by inverting the stereographic projection (22). These solutions
will, for reasons given earlier, have finite energy, equal to 4 times

326
its winding number n. In terms of the variable w (t), eq. (16) can
be written as

V
n
4n (26)

n
As an example, consider w = [(Z-ZO)/A] 0 , where Zo and A are cons-
tants and no is a positive integer. This is an analytic function and
hence a solution of the original field equation (12). Upon inserting
this function into (26), one can check that

2n
A 0
n
2n
(A 0 +

n
v ( 27)
o 4TI

The winding number is n . This is also evident from the fact that
o n
in the function w = [(Z-ZO)/A] 0, the Z plane is clearly mapped no
times into the w-plane. The constants Zo and A represent respectively
the location and the size of the soliton. Notice that thanks to the
translational and scale covariance of the field equation (12), the
energy (27) does not depend either on Zo or on A .

We had taken no to be a positive integer. Choosing no to be negative

yields an equally good solution, with the same energy but opposite
winding. That there is a pole at Z = Zo when no is a negative integer
is no cause for concern. The divergence of w as Z ~ Zo only means
that (j)3 ~ 1.

In this fashion, any meromorphic function of Z or z* is a solitary

wave solution of the system (12).

REFERENCES

[1] A. A. Belavin and A. M. Polyakov, JErP Letters, 22 (1975) 245.

[2] E. B. Bogomolnyi, Soviet J. of Nuc. Phys. 24 (1976) 449.
[3] R. Rajaraman, Solitons and Instantons (Amsterdam: North-Holland
Co., 1982).

327
Soliton Propagation In Optical Fibres
A.Kumar
Department of Physics, Indian Institute of Technology,
Hauz Khas, New Delhi 110016, India

Pulse propagation in optical fibres, including the effects of nonlinear

change in the refractive index, group velocity dispersion and losses,
is reviewed. An analysis of the effect of fifth-order nonlinearity
on the soliton propagation in the model based on the damped nonlinear
Schr~dinger equation is made. The experimental observation of bright
solitons and soliton laser are discussed in short.

INTRODUCTION

Nonlinear pulse propagation in optical fibres has been a subject of

intensive research [1-40] primarily because of the possibility of trans-
mitting undistorted pulses of high peak powers which can be used in
various fields such as communication, power transmission, medicine and
industrial processing [3,4], etc. The main activity in this connection
is centred around a suggestion given by Hasegawa and Tappert [1,2] that
the nonlinear response of the fibre, which tends to self-confine the
pulse, can be exploited to get rid of the pulse distortions, caused
by dispersion, frequency chirp, etc., which put a critical limitation
in realizing the full bandwidth capability of the linear communication
systems using fibres. The distortionless pulse propagation based on
this mechanism is usually referred to as soliton propagation. According
to this mechanism when the frequency shift due to the nonlinear change
in the refractive index of the fibre is balanced with that due to group
dispersion in the negative dispersion region of the fibre the optical
pulse should form an envelope soliton and a stationary pulse transmission
should take place. However, due to the losses in fibre the soliton pulse
continuously loses its energy, as a result of which its width increases
with the distance of propagation and it tends to vanish [13,18-21,27].
Therefore, for long distance propagation periodic amplification and
reshaping of the pulse should be done [14-16,22].

The prediction of Hasegawa and Tappert was successfully verified

by Mollenauer, Stolen and Gordon [28]. After this demonstration of the

328
feasibility of the soliton concept a large number of research papers
(both theoretical and experimental) dealing with further possible appli-
cations of optical solitons appeared in literature. All these investi-
gations led to the creation of what is called a soliton laser, which
is a valuable achievement.

The organisation of the lecture is as follows. Section I deals

with the basic nonlinear evolution equation for the complex envelope
amplitude and the soliton solutions to them in the lossless case. In
Section II we see the effect of fibre losses on soliton propagation
while in Section III we study the amplification and reshaping of solitons
and their effect on the stability of the pulse. Section IV describes
the effect of fifth-order nonlinearity on the propagation of a gaussian
input pulse taking into account the frequency chirp. Section V deals
with the modulational instability in optical fibres and the possibility
of generating a train of solitons using induced modulational instability.
In the last two sections, i.e., in the sixth and seventh we, in short,
talk about the experimental observations of bright soliton pulses and
the soliton laser.

I. BASIC NONLINEAR EQUATION: BRIGHT AND DARK SOLITONS

Consider the propagation of a pulse envelope ~(t,x) (where x is the

longitudinal coordinate along the axis of the fibre) through an optical
fibre including the effects of group velocity dispersion, nonlinear
change in the refractive index of the fibre and the fibre losses. It
is well known [1,13,18,28,37] that the evolution of the envelope is
governed by the following differential equation:

a a 1 a 2 rn & n (I~I 2 )~ o, (1. 1 )

i{K 1 ~ + y~ + ~) + -2K2 ~ -
at ax at 2 no NL

where y is the exponent of amplitude decay in linear regime, K1 =aK/aw

K2 = a 2 K/aw 2 and n NL { 1~12) represents the intensity dependent nonlinear
part of the refractive index:

n{w, lEI)
2 (1. 2)

In the case of cubic nonlinearity n NL n21EI2 and self-focusing medium

n 2 > 0, using the transformations

329
 S -1 -2
T (t-K 1X), ~ IK2 IT x

1/2 1 n2 Ko
1\! = T(a!lK 2 I) cp, a 2" n ( 1 . 2a)
0

equation (1.1) can be reduced to the following dimensionless form

o , ( 1 .3)

where r = YT 2 / IK2 I. Note that in the above transformations T is an

arbitrary time scale which allows a pulse of standard duration in the
dimensionless retarded time variable s to correspond to a pulse of any
desired duration in t [28].

In the absence of losses y = 0 and Eq. (1.3) reduces to well-known

nonlinear Schrodinger equation (NLS). This equation can be solved for
any input structure 1\! (~,S) at ~ = 0 using Zakharov-Shabat scheme based
on inverse scattering method. The general response of 1\!(~,S) can be
described by N solitons and continuous modes which vanish at ~ ~ 00.
The number of solitons N is determined by the area

+00
A f I1\!( 0, S) IdS.

The soliton solutions obtained by Hasegawa and Tappert [1,2] for

optical fibres are one soliton solutions to Eq. (1.3) which are usually
called bright or dark solitons depending on whether they exist in the
anomalous dispersion region or norma~ dispersion region respectively.

The bright soliton solutions to Eq. (1.3) in the anomalous dispersion

region when K2 < 0 are given by

( 1 .4)

where the soliton amplitude EBS is related to the pulse half-width So

and the wave number shift AB by

( 1 .5)

In the normal dispersion region when K2 > 0, Eq. (1.3) admits soliton
solutions which represent envelope shocks and are called the dark soli-
tons. They are given by

( 1.6)

330
where EDS ' SO' and AD are related through

(1. 7)

Note that for dark solitons the nonlinear wave number shift is twice
as large as the wave number shift in the case of bright solitons for
the same amplitude.

Further, note that

$U,;, S) ~(S - 1)
u
exp[i(~ -
u
11-)]
2 2
(1.8)
u

is also a solution which indicates that a frequency modulation at ~=O

can produce a soliton which propagates with the speed u different from
that without frequency modulation.

Let us now find out the expression for the peak power of the above
mentioned solitons. Using Eq. (1.2a) and Eq. (1.5) the relationship
between the peak electric field ~o and the pulse width 2to (for which
the pulse height drops to sech = 0.65) can be written as [27]

~o (1. 9)

where Wo = 2 TIC/A is the carrier angular frequency. The peak power

Po is related to ~o by

P
o (1. 10)

where S is the cross-sectional area of the fibre, v = c/n and the di-
g -12
electric constant of the fibre £ = £ n 2 where £0 = 8.854 x 10 F/m.
-22 2 0
Assuming n 2 = 1.2 x 10 (m/V) and n = 1.5 we get

A2 (a 2. n / aA 2 )
P 3.32 x,10 7 -
S(]Jm 2) (W) • (1. 11 )
o (W t )2
o 0

This is the required expression for peak power. From this expression
it follows that near the zero group dispersion the balancing power to
produce a soliton can be minimum. However, at a 2 n/aA2 = 0 the soliton
nature of the pulse is lost. Hence for realistic power one should choose
A in such a way that the higher order dispersions could be treated as
perturbations [27].

331
 2
For A = 1.3 ~m and S 20 ~m one gets that

1 . 6 W(P sec) 2 , (1. 12)

which shows that for 1 ps pulse one requires a minimum power of 1.6W.
However, this minimum power can be changed by the choice of A and some
other factors.

At the end of this section we want to note that in the derivation

of Eq. (1.1) the radial dependence of the field was treated by employing
the radial eigenfunction of a linear dispersive fibre [1,37] which in
general is not acceptable. Christodoulides and Joseph [29,30] have
studied the propagation of bright solitons through optical fibres treating
the radial dependence of the field in an exact way. For the weakly
dispersive fibres they reproduce the old results. However, for highly
dispersive fibres they conclude that higher order effects in the pertur-
bation scheme for radial field dependence become important and may play
a vital role in the formation of solitons. Using the same procedure
they have obtained a new class of dark soliton solutions which exist
in the normal dispersion region of the fibre.

II. EFFECT OF LOSSES

For y * 0, Eq. (1.3) cannot be solved exactly. However, if r~« 1 the

second term in Eq. (1.3) can be treated as perturbation and the one
soliton solution can be written in a simple form [8,18,19]

I/J(~,S) A sechAS exp(io), (2 •1)

where
1/J2
o
A = I/J exp[-2r~], a -ar[1 - exp(-4rS)]. (2.2)
o

Thus the pulse amplitude decreases with the distance of propagation

while the pulse width increases as So~ So exp[2 r~]. It puts some limi-
tations on a large distance propagation of solitons in lossy fibres.
In fact the condition for the perturbation to be valid is

yx « 1. (2.3)

The characteristic distance determined by Eq. (2.3) is known as the

period of soliton in lossy fibres [41].

332
 Numerical simulation results [18] of Blow and Doran confirmed the
perturbative results for small propagation distances. However, the
results obtained for the evolution of higher order solitons in lossy
fabric indicated that the complicated amplitude oscillations charac-
teristic of higher order solitons gradually disappeared and the output
shape at large distances is a single peak with a pulse width much less
than in the corresponding linear case. These results demonstrate that
for realistic optical fibre communication systems large amplitude soli-
tons can be used for a significant reduction in pulse spreading over
the linear regime, which is important for systems using long repeater
spacing.

III. AMPLIFICATION AND RESHAPING OF SOLITONS

From the above discussions it is clear that in real fibres with losses
it is not possible to have distortionless propagation for longer dis-
tances. For long distance communication systems periodic amplification
is needed to retain the original pulse shape. If the soliton is periodi-
cally amplified at a distance determined by group dispersion and losses
it should propagate without distortion for a large distance. To study
this problem let us consider the following.

It has been shown [14-16] that if we include the acts of amplifica-

tion the envelope amplitude ~(~,S) satisfies the following differential
equation

N
i a,10
~
+ 1
-
a2".
---~ + ir~ - i L: a ~ ~ ( s , ~ ~ - 0) a( ~ - ~ ~ ) + I~ 12~ = 0, (3 . 1 )
a~ 2 as2 ~=1

where ~~ = (~-1)~~ ,~ = 1,2,3, ... ,N are the positions of the amplifiers,

a(x) is a-function of x and a~ is the exponential gain of the ~-th

amplifier. Kodama and Hasegawa [14,15] showed that for r « 1 the soliton
can propagate for. a distance larger than 600.~~ provided that the ampli-
fier gain is chosen constant, a, such that it exactly compensates for
the loss rate

a exp[r.~~] - l. (3.2)

However, if a~is variable one has to study the effect of random

kicks, experienced by the soliton during amplification, on the stability
and the detoriation of the soliton quality of the pulse.

333
 Since due to the act of amplification only the amplitude of the
pulse is increased to (l+a~) while the pulse width is not affected the
pulse deviates from the ideal soliton structure. The amplified pulse
produces a new soliton with the amplitude (l+a~)~o where ~o is the ori-
ginal amplitude. The difference between the energy of the amplified
pulse and the newly formed soliton is radiated out as linear dispersive
wave. Kodama and Hasegawa [16] showed that the total energy E(~):

(3.3)
- co

F(~ )
(
~=1
~ (l+a~)J exp(-r~); ~n < ~ <~n+1 (3.4)

diffuses as

[exp(2D~) - l]E o (3.5)

where D is defined by
- 2
«(F n + 1 - Fn/) «(a n + 1 - a) ) ;.
lim lim (3.6 )
n
ll~ -+ 0 ll~-+O

under the assumption that the average value of a~ is given by a defined

by equation (3.2).

Now, right after the n-th amplification ~ ~n + 0 and the soliton

is expected to have the form

(3.7)

where en and an are the centre and the phase of the soliton and

[1 + 2a n + O(a 2 )] exp[-2rll~]~ 1. (3.8)

n n-

The deviation in ~n in the order a~ is due to the interaction between

the soliton and the radiation generated by the previous amplification
and loss in th~ absence of which we would have got ~n = (1 + 2an)~n_1.
From this equation we get that after the n-th amplification the energy
of the soliton can be expressed as

(3.9)

where (1-0) > 0 is the measure of the energy lost by the radiation of
the dispersive wave. The ratio Rn of the soliton energy after n-th
amplification and the total energy is given by

334
 E sol
n
1 + 20. n + 00. n2
-E-- ----------~2~----Rn-1 . (3.10 )
n (1 + an)

It follows from this equation that the deviation of the soliton energy
from the total energy is small after each amplification since 0. ~ r6~

is usually a small quantity. Thus the main problem in the study of

soliton transmission with amplification is to find out the quantity 0
since it determines whether there will be transfer of energy to the
soliton from the dispersive waves (0 <0 < 1) or from the soliton to
the dispersive waves (0 < 0) [16]. Besides that 0 1 indicates that
no radiation appears and all the energy goes to the soliton while 0=0
indicates that there is no interaction between the soliton and the dis-
persive waves which means that the dispersive waves escape from the
soliton quickly. For this purpose Kodama and Hasegawa numerically
integrated equation (3.1) for various values of 6~ and the parameter
20 which represents the random error in the amplifier gain. Their re-
sults show that even for large errors in the amplifier gain an optical
soliton can be stably transmitted through a glass fibre with low loss
for thousands of kilometres.

Concluding the section we note that another adiabatic amplification

scheme based on stimulated Raman gain has also been proposed by Hasegawa
[22] which has the advantage that here the repeater spacing is decided
only by fibre loss rate. In the first scheme discussed here the amplifi-
cation has to be done every time before the dispersive wave leaves the
soliton which means that the repeater spacing is decided by a distance
controlled by group dispersion. The latter scheme utilizes the stimu-
lated Raman process of the fibre itself in which the gain can be made
small to make the amplification adiabatic (so that the change in the
Raman gain due to fibre loss and pump depletion can be avoided). As
a result the area of the soliton remains constant and the transfer of
soliton energy to the dispersive waves is minimum.

IV. EFFECT OF FIFTH-ORDER NONLINEARITY

Let the index of refraction of the fibre be given by [24,25]

n (4.1)

where n2iEi2 and n4iEi4 are the intensity dependent nonlinear parts of
the refractive index. In this case the propagation of the pulse en-

335
velope ~(X'T)' where x is the longitudinal coordinate along the axis
of the fibre, is governed by the following differential equation [24]

(4.2)

where t is time, T = t - k~x is a reduced time coordinate, k~

(ak/aw) is the inverse of the group velocity at the group carrier

w=wo 1 2 2
frequency w , a = -2(a K/aw) , y characterizes the losses in fibre
o w=w
and Ko -won2/2c and Ao = wo g4 /2c express the magnitude of non-
-yx
linearity in terms of n 2 and n 4 . For ~ = ~(x,T)e Eq. (4.2) takes
the form

i~ (4.2a)
ax

The pulse propagation through a fibre characterized by Eq. (4.2a) can

be formulated as a variational problem in terms of the Lagrangian

L (4.3)

K(x) = Ko exp[-2yx], A(x) = Ao exp[-4yx] , (4.4)

where the asterisk denotes a complex conjugate. Assume that ~ initially

had a gaussian form, i.e., that

(4.5)

where Ao and a o are the initial pulse amplitude and ~ulse width respec-
tively. The further evolution of ~o can be analyzed in terms of the
trial function

2
A(x) exp[- T + ib(x)T 2 ] (4.6)
2a 2 (x)

where the amplitude A(x), a(x) and the frequency chirp 2b(x)T all vary
with the distance of propagation. Using the variational principle and
performing the integration over T we get four Euler differential equa-
tions for A(x), A*(x), a(x) and b(x) which after some algebra can be
reduced to a single second order ordinary differential equation [24]
for the normalized pulse width parameter y(x) = a(x)/a o

2~e -4yx \Ie -2yx

.?l:!. (4.7)
3 3 2
y y y

where

336
 2(/
-4- v
/2aK E
o 0 E a(x) 2
IA(x)l. (4.8)
o
a
o

Kumar et al. integrated this equation numerically for ~ = 1/8 and

! -
y= 0.5 dB/Km for various values of the parameters f = (v/ 2~) 2 and A
The initial value for y(x) was taken to be unity. The results of numeri-
cal integration are depicted in Figs. 1 and 2 and show that the pulse
width initially oscillates and then starts increasing monotonically with
the distance of propagation. While oscillating it passes through several
minima whose p03itions vary with the relative strengths of dispersion
and nonlinearity, which is in agreement with the character of solution
obtained for damped NLS equation [13]. In the case of damped NLS equation
for f = 2 the distance travelled by the pulse with consecutive compres-
sion and decompression before it disperses finally is the largest.
In this case the final minimum of the pulse width occurs at a distance
of about 8.6 km which in fact gives the spacing between the amplifiers
to be used for long distance propagation. However if we increase f for
nonzero X the position of the final minimum shifts towards larger dis-
tances. The best result is obtained for X = -0.035 and f = 2.5 (curve

16
>-
....:J:e
3 12
w
Vl
~

~
o
~ 8
~

<I:
~
a::
o
z
4

1S 20 2S 30
DISTANCE (Kmsl

Fig. l Results of the numerical solution of Eq. (4.7) for

different values of f and X. Solid lines: 1, f = 0.0,
X = 0.0; 2A, f = loS, X = 0.0; 3A, f = 2.0, X = 0.0;
4A, f = 2.5, X = 0.0. Dashed lines: 2B, f = 1.5,
X -0.025; 3B, f = 2.0, X = -0.025, 4B, f = 2.5,
X -0.025; 3B, f = 2.0, X -0.025, 4B, f = 2.5,
X -0.025; 4C, f = 2.5, X = -0.035.
337
 >-
12
....0
I
2A
~
UJ
Ul
...J
::;) 8
Q.

0
UJ
N
::::;
«
:£ /

a:: /
/
/
3A
0 /
z , /
1A
,
~
,,

20 25 30
DISTANCE (Kmsl

Fig. 2 Results of the numerical solution of Eq. (4.7) for

different values of f and X. Solid lines: lA, f = 2.6,
X = 0.0; 2A, f = 3.0, X = 0.0; 3A, f = 3.5, X = 0.0.
Dashed lines: lB, f 2.6, X = -0.025; 2B, f = 3.0,
X = -0.025; 3B, f = 3.4, X = -0.025.

4c of Fig. 1) when the pulse width passes through a minimum (before

final dispersion takes place) at a distance of approximately 23 km.
At this point the pulse width is about nine-tenth of the initial pulse
2
width. For a glass fibre with effective core area of 20 jJ m the minimum
laser peak power, at A = 1.55 jJ m, required to materialise the above si tua-
tion for a gaussian pulse of 1 psec width is of the order of 5W. Here
the values of n 2 and n 4 used are n 2 '" 1.2xl0 -22 (m/v) 2 and n 4
4.4 x 10- 37 (m/v)4

Finally we note that since in our case the final minimum of the pulse
occurs at a distance of about 23 km the frequency at which the ampli-
fication of the pulse should be done for long distance propagation is
reduced by a factor of 3 (compared with the model based on damped NLS
equation). This is a considerable advantage.

v. INDUCED MODULATIONAL INSTABILITY AND

GENERATION OF SOLITONS

It was pointed out by Anderson and Lisak [211 that modulational insta-
bility in optical fibres makes a cw optical signal unstable if the
carrier wavelength is more than 1.3 m, i.e., in the negative group dis-
persion region of the fibre leading to some difficulty for coherent signal

338
transmission using amplitude or phase modulation (or both employed
together) .

For cubic nonlinearity the envelope amplitude satisfies the equation

(see Eq. (4. 2 ) )

i(f* +y\)J) (5. , )

The steady state solution to it can be written as

x
\)Jo exp[-i J ~K(x)dx - yx], \)J~ = \)Jo (5.2)
o

with ~K(x) = Ko \)J~ exp[-2yx]. In the case of small modulation \)J, (X,T)
representing \)J(X,T) as

x
[\)Jo + \)J, (X,T)] exp[-i J ~K(x)dx - yxl, 1\)J,I«\)Jo' (5.3)
o

putting \)J, = u + iv and seeking the solution to equations satisfied by u

and v (obtained after the substitution of (5.3) into (5.')) in the form
u = U o cos[wx -~T] and v = Vo sin[wx - ~T] we get that the local growth
rate is given by [2']
2
2 ~ (x)
I w(x) a~ [_c__ ,]'/2 (5.4)
m ~2

where

(5.5)

wi th ~2 = 2 K I \)J I 2 la Note that ~ is /2 times the maximum

co 0 0 co
growth rate frequency in the lossless case y = O. The total growth
rate is determined by exp[K(x)], where

x
, ) , 12]
K(x) J I w(x' )dx'
m
o
(5.6)

Hence the maximum growth rate is given by exp[maxK(x)] where

a~2
co
max(K(x) ) (, - TI/ 4 ) 2""Y (5.7)

and occurs at ~ = ~2 12.

Thus if the product of the maximum growth
co
rate and the damping length is larger than' the modulational instability
becomes a limiting factor for coherent signal transmission because of
the instability of the cw signal.

339
 However, Hasegawa [20] has shown that if one uses an externally
applied amplitude modulation which could induce modulational instability
at the wavelength of the given modulation one would be able to produce
a series of solitons with the pulse width in the range of 1-50 psec and
repetition period ~ 250 psec.

For this purpose Hasegawa solved Eq. (1.3) numerically for different
wavelengths A M of the initial modulation and the depths of modulation
AM in a periodic boundary condition with a given period s = 48 which
corresponds to 130 psec for the parameters used by him. In particular
Eq. (1.3) was solved with the initial condition

1jJ( 0, S) (5.9)

with a fixed value of r = 5.18 x 10 -2 corresponding to 0.3 dB/km. His

calculations show that initially peak structures are formed which deform
at a longer distance of propagation splitting into two or more peaks
and a sequential pulse with a width So ~ 1 is produced at a repetition
frequency given by the initial modulation 'M independently of AM' How-
ever, the distance of propagation needed to form the peaked structures
varies as a function of AM and 'M (Table I of reference 20). This dis-
tance tends to increase as 'M is increased and as AM is reduced. It
suggests that to produce pulses at a shorter distance it is better to
use shorter modulation period and deeper depth of modulation.

VI. EXPERIMENTAL OBSERVATION OF BRIGHT SOLITONS

For the experimental verification of the existence of bright solitons

one needs a fibre with low losses in the negative group dispersion region.

The development of monomode silica glass fibres having low losses (0.2

dB/km) in the region of negative group dispersion [5,28]and mode-locked

colour centre laser[39] tunable over that region made possible the experi-
mental observ~tion of bright solitons. Mollenauer, Stolen and Gordon
[28] successfully demonstrated the distortionless propagation of a 7
psec pulse with peak power of ~ 1.24 W at A = 1.55
for a distance
~m

of about 700 meters using a monomode fibre of about 100 ~m 2 cross sec-
tion. For higher powers they observed substantial compression and well
resolved splitting of psec pulses. These observations were in close
agreement with the predictions of Hasegawa and Tappert. So far as the
experimental observation of dark solitons is concerned we note that noth-
ing has been reported so far.

340
VII THE SOLITON LASER

Using the basic properties of nonlinear pulse propagation through optical

fibres and the mode-locked colour-centre laser Mollenauer and Stolen
[17] created a device which can be called a soliton laser. It is basi-
cally a synchronously pumped mode-locked colour-centre laser with a
certain length (which can be altered according to the required value
of the pulse width) of a monomode polarisation preserving (highly bire-
frengent) fibre incorporated into its feedback loop. The operation of
the laser is based on N = 2 soliton of NLS equation and the shape of
the pulse obtained from this laser is sech2 in intensity. Besides the
ultra narrow pulses the soliton laser allows for longer precisely con-
trolled pulses which are important for several experiments on pulse pro-
pagation in optical fibres.

The pulse narrowing and solitons are obtained using the combined
effect of nonlinearity in the refractive index and negative group dis-
persion of the fibre as discussed earlier. The operation of the soli-
ton laser can in short be explained as follows. Broad pulses initially
obtained from the mode-locked colour-centre laser are narrowed by passage
through the fibre. The narrowed pulses are reinjected into the cavity
in such a way that they are coincident and in phase with those already
existing in the cavity. As a result the laser itself starts producing
narrower pulses. This process is continued until one obtains soliton
pulses.

Concluding we wish to note that the polarisation preserving ability

of the fibre is crucial for the operation of the laser for otherwise
feedback into the mode-locked colour-centre laser, which is highly polari-
zation sensitive, would fluctuate wildly with fibre length, wavelength,
etc. [17].

REFERENCES

[ 1] A. Hasegawa and F. Tappert, Appl. Phys. Letts., 23 (1973) 142.

[ 2] A. Hasegawa and F. Tappert, Appl. Phys. Letts., 23 (1973) 171.
[3] B. Bendo, P. Gianino, N. Tzoar and M. Jain, J. Opt. Soc. Am.,
70 (1980) 539.
[ 4] M. Jain and N. Tzoar, J. Appl. Phys., 49 (1978) 4649.
[5] T. Miya, Y. Terunuma, T. Hosaka and T. Miyashita, Elec. Letts.,
15 (1979) 106.
[6] S. Shimada, T. Matsumaoto, T. Ito, S. Matsushita, K. Washio and
K. Minemura, Elec. Letts., 15 (1979) 484.
[7] L. G. Cohen, C. Lin and W. G. French, Elec. Letts., 15 (1979) 334.

341
 [8) R. Olshansky and G. W. Scherer, in Rec. Optical Communication
Conf. (Amsterdam, The Netherlands) paper 12.5, Sept 17 (1979).
[9) R. H. Stolen, Proc. IEEE, 68 (1980) 1232.
[10) R. H. Stolen, in Optical Fibre Telecommunications, ed S. E. Miller
and A. G. Chynoweth (Academic Press, New York, 1979).
[11) C. Gloge, Appl. Opt., 10 (1971) 2442.
[12) C. V. Shank, Science, 219 (1983) 1027.
[13) D. Anderson, lEE Proc., 132 (1985) 122.
[14) A. Hasegawa and Y. Kodama, Opt. Letts., 7 (1982) 285.
[15] Y. Kodama and A. Hasegawa, Opt. Letts., 7 (1982) 339.
[16] Y. Kodama and A. Hasegawa, Opt. Letts., 8 (1983) 342.
[17] L. F. Mollenauer and R. H. Stolen, Opt. Letts., 9 (1984) 13.
[18] K. J. Blow and N. J. Doran, Opt. Commun., 42 (1982) 403.
[19) D. Anderson, Opt. Commun., 48 (1983) 107.
[20] A. Hasegawa, Opt. Letts., 9 (1984) 288.
[21] D. Anderson and M. Lisak, Opt. Letts., 9 (1984) 468.
[22] A. Hasegawa, Opt. Letts., 8 (1983) 650.
[23] R. H. Stolen and E. P. Ippen, Appl. Phys. Letts.,22 (1973) 276.
[24] Ajit Kumar, S. N. Sarkar and A. K. Ghatak, Opt. Letts.,11(1986)321.
[25] Ajit Kumar and M. S. Sodha, Elec. Letts., 23 (1987} 275.
[26] L. F. Mollenauer, R. H. Stolen and M. N. Islam, Opt. Letts.,
10 (1985) 229.
[27] A. Hasegawa and Y. Kodama, Proc. IEEE, 69 (1981) 1145.
[28] L. F. Mollenauer, R. H. Stolen and J. P. Gordon, Phys. Rev. Letts.,
45 (1980) 1095.
[29] D. N. Christidoulides and R. I. Joseph, Opt. Letts.,9 (1984) 229.
[30] D. N. Christodoulides and R. I. Joseph, Opt. Letts.,9 (1984) 408.
[31] M. Jain and N. Tzoar, Opt. Letts., 3 (1978) 202.
[32] C. V. Shank, R. L. Fork, R. Yen and R. H. Stolen, Appl. Phys.
Letts., 40 (1982) 761.
[33] B. P. Nelson, D. Cotter, K. J. Blow and N. J. Doran, Opt. Commun.,
48 (1983) 292.
[34] D. Grischowsky and A. C. Balant, Appl. Phys. Letts.,42 (1982) 1.
[35] W. J. Tomlinson, R. H. Stolen, and C. V. Shank, J. Opt. Soc. Am.,
B1 (1984) 139.
[36] D. R. Nicholson and M. V. Goldman, Phys. Fluids. 19 (1976) 1621.
[37] V. I. Karpman and M. D. Kruskal, Soviet Phys. JETP 28 (1969) 277.
[38] G. R. Lamb, Elements of Soliton Theory (Wiley Interscience, New
York, 1980).
[39] L. F. Mollenauer and D. M. Bloom, Opt. Letts., 4 (1979) 247.
[40) J. Satsuma and N. Yajima, Prog. Theor. Phys., Suppl. No. 55 (1974)
284.

342
Davydov's Soliton
A.C. Scott
Department of Electrical and Computer Engineering,
The University of Arizona, Tucson, AZ85721, USA

Davydov's theory for storage and transport of biological energy in protein

is described and related to recent infrared absorption measurements in
crystalline acetanilide. Some aspects of the quantum theory are consi-
dered in detail.

1. INTRODUCTION

In living organisms a fundamental mechanism for the transfer of energy

into function proteins or enzymes is the hydrolysis of adenosine tri-
phosphate (ATP) into adenosine diphosphate (ADP) according to the
reaction

( 1.1)

Under normal physiological conditions about 10 kcal/mol or 0.422 eV

of free energy is released by this reaction [1], leading to several
interesting questions: How is this free energy transferred into protein?
How is it stored there? How does it move inside a protein? How is it
transformed into useful work?

To answer questions of this sort a theory was proposed by Davydov

[2] which focused attention on the self-trapping of molecular vibra-
tional energy in the amide-I (or co stretch) vibration of the peptide
unit (CONH), a basic structural element of all proteins. According to
this theory, it was proposed that the localization of amide-I vibrational
energy would alter the surrounding structure (primarily the hydrogen
bonding) and that this local alteration would, in turn, lower the amide-I
energy enough to prevent its dispersion.

At about the same time as the original paper by Davydov, Careri [3]
published some unexpected spectral measurements in the amide-I region
of crystalline acetanilide (CH 3CONHC 6 HS )' or ACN. As the temperature
was lowered from room temperature, he observed an anomalous amide-I

343
band (at 1650 cm- 1 ) growing up on the red side of the normal amide-I
band (at 1665 cm- 1 ). This 1650 cm- 1 band was called anomalous because
it could not be explained with accepted concepts of molecular spectros-
copy (e.g.,Fermi resonance, Davydov splitting, etc.). At first Careri
suspected some unusual one-dimensional phase transformation might provide
an explanation, but no such evidence was found after several years of
experimental work. Recently a self-trapping theory was proposed which
is closely related to that of Davydov and explains the salient experi-
mental facts [4-6].

The present situation, therefore, is that the 1650 cm- 1 band in

ACN seems to provide direct experimental evidence for a self-trapped
state of molecular vibrational energy. The "red shift" of 15 cm- 1 from
the normal band can be considered as the binding energy of a Davydov-
like soliton, and this interpretation leads to quantitative predictions
of biological significance.

This paper is organized into three phases. The first is a review

of Davydov's soliton theory and the experimental observations in crys-
talline acetanilide. The second phase is a detailed comparison of various
attempts to provide a quantum mechanical explanation for self-trapping
of molecular vibrations. Finally some questions of biological signifi-
cance are briefly considered.

2. DAVYDOV'S SOLITON THEORY

This section is intended to provide a brief summary of Davydov's soliton

theory for the convenience of the reader. Such a summary is helpful

to appreciate the differences between the theory of self-trapping pro-
posed for proteins and the theory proposed recently to explain experi-
mental measurements on crystalline acetanilide. It is also necessary
in order to see how the quantum theory developed by Davydov as a basis
for self-trapping is related to other quantum analyses. Several surveys
of this work are available for further reference [7,8].

Careful inspection of the a-helix structure of protein reveals three

channels situated approximately in the longitudinal direction with the
sequence

etc. H-N-C=O---H-NC=O---H-N-C=O---H-N-C=O etc.

where the dashed lines represent hydrogen bonds. For a detailed analysis
it is necessary to consider the interaction of all three channels, but
one is sufficient to layout the basic ideas.

344
 A single channel is governed by the energy operator

(2.1)

Taking the components of A in order, A is an energy operator for the

co
co stretch (amide-I) vibration including the effects of nearest neighbor
dipole-dipole interactions. Thus

(2.2)

where Eo is the fundamental energy of the amide-I vibration, -J is the

nearest neighbor dipole-dipole interaction energy, and ~~(~n) are boson
creation (annihilation) operators for amide-I quanta on the n-th molecule.

Aph is the energy operator for longitudinal (acoustic) sound waves.

Thus

(2.3)

where M is the mass of a molecule, W is the spring constant of a hydrogen

bond, ~n is a longitudinal momentum operator for the n-th molecule, and
1\
un is the corresponding longitudinal position operator.

(2.4)

where Xa is the derivative of amide-I vibrational energy with respect

to the length (R) of the adjacent hydrogen bond. Thus

(2.5)

Davydov minimizes the average value of A with respect to the wave

function

11jJ> r an(t)exp[G(t)l~~ 10>, (2.6 )

n

where

(2.7)

A straightforward calculation shows that

(2.8 )

345
and
(2.9)

The wave function in (2.6) will be called Davydov's ansatz throughout

this chapter. One of the aims here is to study the range of validity
of this ansatz.

Assuming that Davydov's ansatz approximates the true wavefunction,

(2.8) and (2.9) show that Bn and TIn are the average values of the posi-
tion and momentum operators, respectively. Furthermore, an is the pro-
bability amplitude for finding a quantum of amide-I vibrational energy
on the n-th molecule. The normalization condition <~[~> = 1 implies
that

1. (2.10)

Thus Davydov's ansatz describes the dynamics of a single quantum of

amide-I vibrational energy.

Minimization of <~[A[~> with respect to an' Bn and TIn leads to

the differential-difference equations

0, (2.11 a)

(2.11 b)

Extensive numerical and theoretical analysis of (2.11) yields the follow-

ing results [9-12J: (i) it is reasonable to expect soliton formation
at the level of energy released by ATP hydrolysis (1.1), and (ii) such
a soliton travels rather slowly with respect to the speed of longitudi-
nal sound waves. This suggests neglecting the kinetic energy of longi-
tudinal sound by assuming Bn 0, whereupon

Xa
- -W [a n [2 (2.12 )

and, in this "adiabatic approximation," (2.11) becomes

0, (2.13 )

where
(2.14 )

346
 Davydov has emphasized that a solitary wave solution of (2.11) cannot
be created directly by absorption of a photon because of an unfavorable
Franck-Condon factor. This is because the necessary intermolecular
displacement in (2.11b) cannot occur in a time that is short enough
for photon absorption. The Franck-Condon factor will be discussed in
detail in the following section.

3. SELF-TRAPPING IN CRYSTALLINE ACETANILIDE

Just as in the a-helix, careful inspection of the crystal structure

of acetanilide reveals channels situated in the b-direction with the
sequence

etc. H-N-C=O---H-N-C=O---H-N-C=O---H-N-C=O etc.

Recent infrared absorption measurements on microcrystals of ACN

show an unexpected band at 1650 cm- 1 which rises with decreasing tem-
perature to become the dominant spectral feature below 100 K. When
this band was discovered, Careri suspected it to be caused by a subtle
phase change along b-direction of the crystal [3], but careful studies
over a period of several years failed to reveal any such evidence. The
lack of a viable alternative eventually led to the suggestion that the
-1
1650 cm band might be caused by direct absorption of an infrared photon
into a self-trapped state similar to that proposed by Davydov. The
qualifier "similar" is important because, as was noted above, the Franck-
Condon factor is unfavorable for direct photon absorption by a self-
trapped solution of (2.11).

The corresponding theory proceeds, as in the previous section, by

defining the energy operator

(3.1 )

where @CO is again given by (2.2) but with [6]:

-1
J 3.96 cm (3.2)

In the present analysis, however, self-trapping is assumed to be caused

by interaction with an optical phonon rather than an acoustic phonon .
. Thus

(3.3)

347
and

(3.4)

Minimization of <~IAI~> with respect to the parameters of the

Davydov ansatz wavefunction (2.6), where

"
(J
1
- -h
n
(;
E[q (t)p
n
"
- P (t)q],
n
(3.5)
n

leads to the dynamic equations

0, (3. 6a)

(3.6b)

As before

(3.7)

The adiabatic approximation (qn 0) reduces (3.6) to

0, (3.8)

where

(3.9)

A detailed numerical study of a system of equations similar to (3.8)

but representing one hundred molecules of ACN in two coupled channels
has recently been carried out by Eilbeck et al. [6].showsThis wor~
1
that the red shift from the normal amide-I band at 1665 cm- to the
1650 cm- 1 band is best fit by choosing

-1
Yo 44.7 em (3.10)

We turn next to an estimate of the Franck-Condon factor for direct

photon absorption by a self-trapped state of (3.6). Before absorption
lanl2 = 0, and after absorption lanl 2 = 0 over a localized region such
that (2.10) is satisfied. Thus the ground state wavefunction of (3.6b)
must shift from

4> (~)~
1TW
2 w )
exp (-qn 2w ( 3 • 11 )
o

348
before absorption to

~
o
w!
(nw)2 exp [ -(qn + Yo la n 12)2 ~)
2w ( 3.12)

after absorption, where

1
W = (w/m)' ( 3.13)

is the frequency of the optical mode that is mediating the self-

trapping. The transition probability for soliton absorption is there-
fore reduced by the Franck-Condon factor

[f~ o ~ 0 * dq n )2 ~ exp(-Y o /2w), ( 3 . 14)

which is close to unity for

y «w. ( 3.15)
o

The frequency (w) of the optical mode can be determined from the
-1
temperature dependence of the 1650 cm line. Such temperature depen-
dence is expected, because the probability of (3.6b) being in its ground
state, and therefore able to participate in self-trapping, is [1 -
exp(-w/kT»). Thus as the temperature is raised, the low temperature
factor given in (3.14) should be reduced by the additional factor
[1 - exp(-w/kT»)2. A least square fit to intensity data is obtained
for w = 131 cm- 1 . Together with (3.10) this implies exp(-Yo/2hw) = 0.84.

Further evidence tending to favor a self-trapping explanation for

the 1650 cm- 1 band is the recent observation of the overtone series
shown in Table 1 [13]. Since the overtones N > 2 are self-trapped states
involving more than one quantum of the amide-I vibration, it is interest-
ing to consider states that avoid the constraint of (2.10).

Table 1 Overtone Series for the

ACN Soliton

N v(N)

-1
1 1650 cm
2 3250
3 4803
4 6304

349
4. THE QUANTUM THEORY OF SELF-TRAPPING

In this section we approach the problem from a classical perspective.

Starting with the classical amide-I coordinates, P n and Qn' for which
the Hamiltonian is r n [p2n + Q2)
n'
it is convenient to define the complex
mode amplitudes

1
wo2 (P n + iQn)· (4. 1)

In terms of these complex mode amplitudes (including dipole-dipole

interactions)

(4.2)

where

(4.3)

is the classical oscillation frequency of an amide-I vibration. (From

here on we will assume n= 1 and measure energy and frequency in the
same units.)

With a classical interaction energy

(4.4)

where qn is the coordinate of some low-frequency phonon with adiabatic

energy

1 2
2" w r qn' (4.5)
n

one arrives at the total classical Hamiltonian

H (4.6 )

Minimizing (4.6) with respect to the qn requires

(4.7)

whereupon (4.6) can be reduced to

H (4.8)

350
where

2
Y - X /w. (4.9)

The corresponding dynamical equation for An is

(i d
dt
E )A
o n
o. (4.10)

In additional to the energy, H, another constant of the motion along

solutions of (4.10) is the number

N ( 4 • 11)

To this point the discussion of the present section has been entirely
classical. We now consider quantization in four special cases: (i)
J «y, (ii) y « J, (iii) semiclassical quantization, and (iv) the
Davydov ansatz. In each case it will be of particular interest to cal-
culate an overtone series corresponding to that presented in Table
for crystalline acetanilide.

4.1 The Case J « y

In this case we neglect the dipole-dipole interaction terms in (4.8)

and (4.10), and write the energy

H (4.12 )

where

( 4.13)

Under quantization, the terms in (4.12) become operators

h
n
-+ 11 n (4.14 )

through replacement of the complex mode amplitudes by creation and anni-

hilation operators for bosons. Thus

(4. 15a)

A* -+ jjt. (4.15b)
n n

Since the ordering of these operators is not determined by (4.13),

we take the averages

351
 -+ ~(~t~ + ~~t), (4.16 )

-+ ~(~t~t~~ + ~t~~t~ + ~t~~~t + ~~~t~t + ~~t~t~). (4.17)

6

where the subscripts have been dropped for typographical convenience.

Noting that ~t and ~ have the properties ~tIN> = IN+1 IN+1> and ~IN>
=IN IN-1> (where IN> is an harmonic oscillator eigenstate), it is straight-
forward to show that

~ (E - ~y) (~t~ + ~) 1 ~t~~t~.

0 2 2 2"Y (4.18 )
Thus

filN> E(N) IN>, (4.19 )

where

1 1 N2
E(N) (E
0
2"Y) (N + ~)
2 2"Y (4.20 )

In summary, eigenvectors of the operators defined through (4.13),

(4.14), (4.16) and (4.17) are identical to those of an harmonic oscilla~
tor, but the corresponding eigenvalues are given (4.20).

The form of (4.20) is significant. It can be written

E(N) (4.21 )

L
where EC is the ground state (N 0) energy, E Nand

(4.22 )

This "nonlinear" contribution is directly measured from the overtone

series in Table 1.

4.2 The Case y « J

In this case the classical equation (4.10) reduces to the nonlinear

Schrodinger (NLS) equation of soliton theory. To see how this goes,
assume the repeat distance between molecules of d and replace the dis-
crete variable n by a continuous variable, x = n, which measures distance
in units of d. Then (4.10) takes the form

( 4.23 )

352
 Quantization of this equation was originally performed using the
Bethe ansatz method and recently it has been shown tha~ such solutions
can be efficiently constructed from a quantum version of inverse scatter-
ing theory [14,15].

Under quantization, the functions A and A* are replaced by annihila-

tion and creation operators for boson fields, ~ and ~t. At equal times
these have the commutation relations [$(x), $*y)] = [t(x),$t(y)] = 0
and [$(x),$t(y)] = 6(x - y). In terms of the previous discussion, it
is evident that $(x) is equivalent (under scaling) to ~ in the conti-
n
nuous limit n = x. In effecting this limit two procedures are customary:
(i) neglect consideration of the ground state energy which is unbounded
in the limit, and (ii) "normal order" all operator expressions, i.e.,
move all creation operators to the left.

Since ~~t = ~t~ + 1, normal ordering of (4.18) and neglect of the

ground state energy imply

(4.24)

Thus to put (4.23) in standard form for quantum analysis, let

A ~ exp[-i(E o - 2J - y)t] (4.25 )

and scale time as t ~ t/J. Then (4.23) becomes

0, (4.26 )

where a subscript notation is used for the partial derivatives. Under

quantization ~ ~ $ and (4.26) becomes the operator equation

0, (4.261)

with energy operator

(4.28 )

number operator

N (4.29)

and momentum operator

(4.30 )

353
 The quantum inverse scattering method provides exact wavefunctions,
I~> , that diagonalize ~, ~ and A as follows.

(4.31 )

where

N integer> 0, (4.32 )

~I~>= Npl~>, (4.33 )

where p is a real number, and

AI~> = [NP2 +
2
~(N _ N3
48J 2
j ~>
1 (4.34 )

Furthermore in the limit y/J .... 0

I~> .... fdx e ipx $1 0 >. (4.35 )

Equations (4.32) and (4.34) imply an overtone series

E(N) (4.36)

(4.37)

4.3 Semiclassical Quantization

In the parameter range y ~ J, it is possible to impose elementary quantum

conditions on stationary solutions of (4.10). Writing such a solution
in the form

E
A (~) ~ Ct exp[-i(~ + w)t) (4.38)
n y n J

reduces (4.10) to the standard form

WCtn + Ct n+1 + Ct + Ct n3 o. (4.39 )

n-1

Using a shooting method [12) it is possible to find a family of numeri-

cal solutions for (4.39) with the following properties: (i) = Ct_ n '
Ct n

(ii) for n ~ 0, Ct n >Ct n + 1 , and (iii) limn.... ooCt n = O. From such a solution

354
the conserved quantities Hand N defined in (4.8) and (4.11) are readily
calculated as

J 2
N E Ct (4.40 )
Y n
and
E Ct n Ct n _ 1 4
E Ct n
H(N) E n 1 N2 n
- J N 2"Y (4.41 )
0 2
ECt~
[ ECt~J
Semiclassical quantization is effected by noting that stationary
solutions are of the form

A (t)
n
AnO exp[-ie(t)], (4.42 )

where ~ = dH(N)/dN. Thus Nand e are conjugate variables and the

quantum condition

eft N de 21T(integer) (4.43)

together with the definition of N (4.11) imply

N integer ~ O. (4.44 )

Equation (4.41) has the form

E(N) EL + ENL, (4.45)

where
4
E Ct n
ENL 1 N2 n (4.46 )
2"Y
~ Ct12

In the limit J« y, Ct n « Ct o for Inl ~ 1 so (4.46) evidently reduces

to (4.22). In th~ limit y« J it is straightforward to show that (4.46)
reduces to (4.37). Thus (4.46) is expected to provide an accurate cal-
cUlation of ENL over the entire parameter range.

It is now possible to consider how data from the overtone series

for the 1650 cm- 1 band in acetanilide compare with these calculations.
From (3.2) and (3.10),

11.3. (4.47)

355
This lies in the range of (4.46) for which

(4.48 )

so the line at 1650 cm- 1 implies

-1
1672.3 cm (4.49 )

From the measured values of overtone frequency, v(N) in Table 1, the non-
linear contributions to the overtone spectrum can be calculated as

(4.50)

In Table 2 we compare these calculations with those computed from

(4.48) .

Table 2 Nonlinear Terms in

ACN Overtone Series

_E NL (cm- 1 ) 1 2 -1
N ZyN (cm )

22 22
2 95 89
3 214 201
4 385 357

4.4 Davydov's Ansatz

We are now in a position to evaluate Davydov's ansatz. In the context

of an adiabatic approximation, the wavefunction introduced in (2.6)
takes the form

II\!> = E a (t)
n
Gtn 10>, (4.51 )
n

where the an(t) are solutions of (2.13). This form of the Davydov
ansatz has the following properties:
(i) In the limit J « y, it reduces to the first eigenfunction,
11>, in (4.19).
(ii) In the limit y« J, it reduces to the asymptotic form of
Bethe's ansatz in (4.35).
(iii) Between these two limits, Davydov's ansatz gives energies that
agree with semiclassical calculations.

356
 Thus one concludes that Davydov's ansatz is a useful approximation
to the exact wavefunction over the entire parameter range 0 < y/J < 00

with the constraint (2.10) which implies N = 1.

5. BIOLOGICAL SIGNIFICANCE OF SELF-TRAPPING

Measurements on crystalline acetanilide (ACN) confirm Davydov's theory

of self-trapped states (solitons) in hydrogen bonded polypeptide chains.

Furthermore, Table 1 shows that the "N = 2" state in ACN can absorb almost

all (95%) of the free energy released in hydrolysis of adenosine tri-

phosphate (ATP). It is reasonable to suppose that a corresponding state

can form on the hydrogen bonded polypeptide chains of a-helix.

Over a decade ago McClare argued that the free energy released in

ATP hydrolysis should transfer resonantly into a protein in order to

avoid thermal degradation [16,17]. To store and transport this energy
he posited an "excimer" state in protein which would be closely related
to the amide-A band of a-helix at 3240 cm- 1 . McClare's excimer is
qualitatively similar to the "conformon" of Green and Ji [18] and the
basic properties of both are provided by a Davydov soliton in the "N=2"
state [2,7,19-22]. In the past such suggestions have been rejected or
ignored by the biochemical community because a localized region of free
energy within a protein was believed to be physically impossible. Since
this view is no longer tenable, the early proposals of Davydov,
McClare [16,17,22~~and Green and Ji [18] must be reevaluated. A recent
paper by Careri and Wyman [24], suggesting a soliton mechanism for
cyclic enzyme activity, provides a first step in this direction.

REFERENCES

[1] R. F. Fox, Biological Energy Transduction (Wiley, New York, 1982),

p. 216.
[2] A. S. Davydov, J. Theor. BioI. 38 (1973) 559.
[3] G. Careri, in Cooperation Phenomena, ed H. Haken and M. Wagner
(Springer, Berlin, 1973), p. 391.
[4] G. Careri, U. Bountempo, R. Carta, E. Gratton and A. C. Scott,
Phys. Rev. Lett. 51 (1983) 304.
[5] G. Careri, U. Buontempo, F. Galluzzi, A. C. Scott, E. Gratton and
E. Shyamsunder, Phys. Rev. B30 (1984) 4689.
[6] J. C. Eilbeck, P. S. Lomdahl and A. C. Scott, Phys. Rev. B30
(1984) 4703.
[7] A. S. Davydov, Phys. Ser. 20 (1979) 387.

357
 [8] A. S. Davydov, Soc. Phys. Usp. 25 (1982) 898 [Usp. Fiz. Nauk
138 (1982) 603].
[9] A. C. Scott, Phys. Rev. A26 (1982) 578.
[10] A. C. Scott, Phys. Rev. A27 (1983) 2767.
[11] A. C. Scott, Phys. Rev.A29 (1984) 279.
[12] L. MacNeil and A. C. Scott, Phys. Scr. 29 (1984) 284.
[13] A. C. Scott, E. Gratton, E. Shyamsunder and G. Careri, Phys. Rev.
B32 (1985) 5551.
[14] E. K. Skylanin and L. D. Faddeev, Sov. Phys. Dokl. 23 (1978) 902
[Dokl. Acad. Nauk SSSR 243 (1978) 1430].
[15] H. B. Thacker and D. Wilkinson, Phys. Rev. 019 (1979) 3660.
[16] C. w. F. McClare, Nature 240 (1972) 88.
[17] C. w. F. McClare, Ann. N. Y. Acad. Sci. 227 (1974) 74.
[18] D. E. Green and S. Ji, in The Molecule Basis of Electron Transport
(Academic Press, New York, 1972), p. 1.
[19] A. S. Davydov, Biofizika 19 (1974) 670.
[20] A. S. Davydov, J. Theor. Bioi. 66 (1977) 379.
[21] A. S. Davydov, Int. J. Quant. Chern. 16 (1979) 5.
[22] A. S. Davydov, Biology and Quantum Mechanics (Pergamon, New York,
1982) .
[23] c. w. F. McClare, Ann. N. Y. Acad. Sci. 227 (1974) 74.
[24] G. Careri and J. Wyman, Proc. Natl. Acad. Sci., USA 81 (1984)
4386.

358
Generalized Nonlinear Schrodinger Equations
in Quantum Fluid Dynamics
B.M. Deb and P.K. Cbattaraj
Theoretical Chemistry Group, Department of Chemistry,
Panjab University, Chandigarh 160014, India

This review article discusses several inte-resting generalized nonlinear

Schrodinger equations (GNLSE) occurring in quantum fluid dynamics for
both time-independent and time-dependent situations as well as the
possibility of soliton or solitary wave solutions for such equations.
It concludes with a brie f report on a new GNLSE derived by the authors
recently.

1. INTRODUCTION: THE NATURE OF QUANTUM FLUID DYNAMICS

Paradoxical though it may appear, generalized forms of nonlinear

Schrodinger equation (GNLSE) do occur in the quantum theory of many-
particle systems--and therefore in nuclear, atomic, molecular and solid
state physics--provided one adopts Madelung's, rather than Schrodinger's,
viewpoint. In Madelung's approach [1], the Schrodinger equation for
a single particle is recast in the form of two classical fluid dynamical
equations, a continuity equation and an Euler-type equation of motion
(EOM) [2]. This is the origin of quantum fluid dynamics (QFD) where
the system is described in terms of two real quantities, the charge
density and the current density (both obeying the above two fluid dyna-
mical equations), instead of the complex-valued wavefunction [3]. At
a first glance, this Madelung fluid does not have a particle interpreta-
tion, unlike the Hamilton-Jacobi fluid [4] whose dynamics are governed
by transformed Hamilton's EOM~. However, following Nelson [5], one
may ascribe a particle interpretation to the Madelung fluid by consi-
dering the stochastic nature of the particle trajectories. In the semi-
classical limit, the Bohm potential term [2] vanishes for the Madelung
fluid which then reduces to the Hamilton-Jacobi fluid. Apart from
'employing a "classical" language within the quantum context, the main
strength of the QFD viewpoint lies in its adoption of three-dimensional
(3D) single-particle densities [3] as the basic variables. Even for

359
a many-particle system, the continuity equation and the EOM can be
brought to 3D space in terms of an orbital partitioning, e.g., natural
orbital~ [6] or Kohn-Sham orbitals [7], of the charge and the current
density.

In this short review article, we discuss several interesting GNLSE's

in both time-independent and time-dependent situations, the possibility
of soliton or solitary wave solutions, along with a brief report on
a new GNLSE derived by us recently [8]. The GNLSE's derived in our
group have arisen in our attempts to deal with atomic and molecular
phenomena using a blend of density functional theory (DFT) and QFD [2,3].

2. TIME-INDEPENDENT GNLSE

For a many-electron system, the current density vanishes in the ground

state and therefore the QFD viewpoint reduces to the DFT viewpoint.
The latter states that the ground state energy E[p] is a unique func-
tionalof p(r), the charge density, and attains its minimum value
for the true p. Variational optimization of E[p] with respect to a
trial density, subject to the conservation of the total number of parti-
cles, leads to an Euler-Lagrange equation which is nothing but a time-
independent GNLSE for the direct calculation of electron density, by-
passing the wavefunction. One such equation, derived by Deb and Ghosh
[9], has the form (atomic units employed throughout this paper)

[- 21 V
2
+ veff(r;p)]~(r) ~~(r), (1 )

p(r) I ~(r) I 2 ' (2 )

where ~ is a Lagrange multiplier and veff is a one-body nonlocal,

nonlinear effective potential consisting of kinetic, Coulomb and ex-
change-correlation terms. The nonlinearity in v eff arises due to non-
integral powers of ~ as well as an integral. Similar equations have
subsequently been derived, though not solved, by Levy et al. [10] and
Hunter [11]. Equation (1) was numerically solved by Deb and Ghosh [9]
in a model potential framework for noble gas atoms (Ne, Ar, Kr, Xe).
For such systems, Eqs. (1) and (2) become one-dimensional in terms
of the radial variable r. The calculated radial densities, energies
and ~-values were quite satisfactory. However, an analytical solution
was not attempted.

360
3. TIME-DEPENDENT GNLSE: THE POSSIBILITY OF
SOLITON OR SOLITARY WAVE SOLUTIONS

The time-dependent GNLSE's are QFD EOM~. In order to examine the possi-
bility of their soliton or solitary wave solutions, it is useful to
fall back upon the familiar cubic NLSE (one spatial dimension),

. aiP +a
ld"t 0, (3 )

whose envelope soliton solution exists when as> 0 and iP-+ 0 as\x\-+oo .
This is the case of modulational instability of a train of modulated
waves in a nonlinear, dispersive medium [12,13]. Note that, in general,
a GNLSE may involve three spatial variables apart from time. For
spherically svmmetric systems, e.g., closed-shell atoms, such a GNLSE
reduces from three to one spatial dimension. If one identifies x in
Eq. (3) as the radial variable r, then Eq. (3) may be regarded as a
spherically symmetric version of Eq. (1), apart from time, with an
appropriate form for v eff and suitable boundary conditions. Therefore,
the solution of Eq. (3) can provide insight into its more general forms.
Due to the difficulties associated with the analytical or numerical
solution of higher dimensional GNLSE's, one frequently resorts to simpler
forms of reduced dimensionality whose analytical/numerical solution is
relatively easier to obtain. This is the approach adopted in Section 4.
For the present, we continue our discussion on both one- and three-
dimensional GNLSE's.

The following GNLSE [14] was helpful in the analysis of the Heisen-
berg spin chain with site-dependent exchange integral [15]:

aa 2 (fiP) + 0, (4 )
ax 2
L f\iP\ ~x (f\iP\)dx, (5 )

where f is a real function of x. The corresponding Madelung fluid

equations were obtained by Belic [16], by adopting the usual polar form
for iP , viz.,

p 1/2 e iX (6 )

in terms of the density p and the velocity potential X Equation

(4) has also been linked with the dynamics of a surface [16]. Earlier,
Nonnenmacher and Nonnenmacher [17] had explicitly shown that the usual
Madelung fluid (obtained from the Schrodinger equation) cannot have

361
a soliton solution unless one adds a nonlinear term to it. Equation
(3) is such a nonlinear equation and its corresponding Madelung fluid
will have the expected solitonic behaviour. Such a Madelung fluid
is equivalent to a nondissipative (ideal), isentropic and irrotational
fluid [17].

Nassar [18] has suggested that the QFD viewpoint, using both the
Madelung approach (i.e., transformation via the polar form of the wave-
function) and the stochastic mechanical approach [4,5,19] which incor-
porates a stochastic force in Newtonian mechanics, may also be helpful
in dealing with the quantum mechanical treatment of dissipative proces-
ses. Such processes occur, for example, in photochemistry, solid state
physics, statistical physics, fission and heavy ion physics. Accordingly,
a generalized nonlinear Schrodinger-Langevin equation (GNLSLE) of the
following form has been derived [20]:

a<!>(x,t) 1 a 241 (x, t)

i +"2 2
at ax
41(x,t)
-[~i in + b lnp(x,t) + V(x,t)] 41(x,t) 0, (7 )
41*(x,t)
2 (8 )
p (x,t) 141(x,t)1 '
where v is akin to a plasma collision frequency, b is a free parameter
and v(x,t) is an external potential which may contain a white-noise
random force. The first term in square brackets in Eq. (7) accounts
for dissipation. Nassar [21,22] has then employed the usual Madelung
transform, Eq. (6), to obtain the QFD equations corresponding to Eq.(7).
Note that Eq. (7) is not expected to have a solitary wave solution.

A different type of GNLSE (in three spatial dimensions) has been

suggested by Himi and Fukushima [23] via QFD, in order to deal more
effectively with quantum phenomena such as heavy ion collisions. The
two QFD equations result on applying a generalized scaling approxima-
tion in the time-dependent Hartree-Fock framework. This GNLSE is, in
effect, formally a time-dependent generalization of Eq. (1), viz.,

0, (9 )

where v eff is a nonlocal, nonlinear one-body potential given by

( 10)

with E as the total energy and v ext as the external potential.

362
4. A NEW GNLSE BASED ON AN AMALGAMATION OF QFD AND DFT

We now return to our earlier theme in Section 1 of using a blend of

QFD and DFT (called QFDFT). Since time occurs naturally in the QFD
equations and a nonzero current density (in addition to the charge
density) is obtained by solving them, through QFD DFT can go into
both time-dependent situations [7,24] and excited states, two types
of phenomena which have been outside the purview of traditional DFT.
We have considered [8] the physical problem of high-energy proton-neon
atom collisions and have derived a single equation (a GNLSE) for directly
calculating the net time-dependent electron density and current density.
The GNLSE has also been derived by us alternatively through stochastic
mechanics, via a set of Fokker-Planck equations. The GNLSE has the
following form, similar to Eq. (9), viz.,

.a¢(r,t) + ~ 2 o.
l at 2 0v..,m(r,t) + veff (r,
t ) ..,
m( r, t) ( 11 )

However, v eff is of a different form from (10) and can be explicitly

written as

a (N ) _ 9. _ _ _ + f ( R, N )j (1 2 )
--2- r N '
r IR-r I
p(r,t) I¢(r ,t) I2 ' ( 13)

where Ck and Cx are constants, a(N) depends on N, f(R,N) depends on

both Rand N, R being the internuclear distance and N being the number
of electrons. Here Q refers to a screened nuclear charge. The non-
linearity in Eq. (11) enters through the one-body nonlocal potential
v eff ' via nonintegral powers of ¢ as well as an integral occurring
in Q (cf. Section 1). In order to solve Eq. (11), for studying high-
energy proton-atom collisions, we first consider a restricted case of
spherically symmetric scattering system. For a spherically symmetric
system (one spatial dimension), the GNLSE (11) may be transformed to

i a1J;(x,t)+
at -2
{ _1_
3 a
ax + V eff
[]}
1J;,t 1j.(x,t) 0; a>O,O~x~oo
ax ax
( 14 )

where 1J; = r¢ and x = Jr. 1J; tends to zero as x goes to zero or

infinity. We have numerically solved [8] Eq. (14) by applying a Crank-
Nicolson-type finite difference scheme which is stable [25] and con-
vergent. Figure 1 depicts the initial radial density profile and the
corresponding time-evolved quantity after 61 time-steps (each time-
step = 0.01 atomic unit). Since the initial profile is not a solution

363
 12.5 ..-------"""2=-.-=5,,-----------. Fig. 1 Plot of radial den-
t=O.O a.~, sities against tr, in the
spherically symmetric ap-
proximation, of a target
neon atom colliding with
a proton. The first figure
at t = 0.0 corresponds to
the internuclear distance
R = 10.05 and clearly
shows the two atomic
shells (Hartree-Fock wave-
function [27] used). The
second figure corresponds
to R = 9.44 after 61 time-
steps, each time-step be-
ing 0.01. All values are
in atomic units.

of Eq. (14), it adjusts itself in course of time and from the thirtieth
time-step onwards the envelope of the profile remains unchanged. In
other words, these results indicate the possibilIty of solitary wave
solutions. This requires further analysis. We also hope to look into
the integrability properties of this type of equation, which might lead
to an analytical solution. Incidentally, a spherically symmetric GNLSE
also occurs in the study of higher dimensional Heisenberg spin chain
[26].

Th2 GNLSE (11) has also been solved numerically for a scattering
system of cylindrical symmetry (two spatial dimensions). An alternating
direction implicit-type finite difference technique has been used for
this purpose. As the proton approaches the neon atom and the interaction
is switched on, a nonzero current density is obtained which signifies
that contributions from excited states are coming into picture [8]. We
feel that, unlike the usual approaches in molecular reaction dynamics,
our QFDFT approach [8] has the potential to monitor a time-dependent
quantum process from start to finish, where the process occurs on a
pUlsating (time-dependent) potential surface given by veff(r,t).

Acknowledgement

BMD and PKC would like to thank the CSIR, New Delhi, for a research
grant and a Senior Research Fellowship respectively.

364
References

[ 1) E. Madelung, Z. Phys. 40 (1926) 332.

[ 2) S. K. Ghosh and B. M. Deb, Phys. Rep. 92 (1982) 1.
[ 3) A. S. Bamzai and B. M. Deb, Rev. Mod. Phys. 53 (1981) 96,593.
[ 4) F. Guerra, Phys. Rep. 77 (1981) 263.
[ 5) E. Nelson, Quantum Fluctuations (New Jersey: Princeton University,
1985) .
[ 6) S. K. Ghosh and B. M. Deb, Int. J. Quant. Chern. 22 (1982) 871.
[ 7) B. M. Deb and S. K. Ghosh, J. Chern. Phys. 77 (1982) 342.
[ 8) B. M. Deb and P. K. Chattaraj, to appear.
[ 9) B. M. Deb and S. K. Ghosh, Int. J. Quant. Chern. 23 (1983) 1.
[ 10 ) M. Levy, J. P. Perdew and V. Sahni, Phys. Rev. A30 (1984) 2745.
[ 11) G. Hunter, Int. J. Quant. Chern. 29 (1986) 197.
[ 12) A. Hasegawa, Plasma Instabilities and Nonlinear Effects (Berlin:
Springer-Verlag, 1975), p. 201.
[ 13 ) R. K. Dodd, J. C. Eilbeck, J. D. Gibbon and H. C. Morris, Solitons
and Nonlinear Wave Equations (London: Academic Press, 1982),p.495.
[ 14) M. Lakshmanan and R. K. Bullough, Phys. Lett. 80A (1980) 287.
[ 15) R. Balakrishnan, J. Phys. C15 (1982) L1305.
[ 16) M. R. Belie, J. Phys. A18 (1985) L409.
[ 17) T. F. Nonnenmacher and J. D. F. Nonnenmacher, Lett. Nuovo Cim.
37 (1983) 241.
[ 1 8) A. B. Nassar, Lett. Nuovo Cim. 41 (1984) 476.
[ 19) G. C. Ghir~rdi, C. Omero, A. Rimini and T. Weber, Rev. Nuovo
Cim. 1 (1978) 1.
[20 ) A. B. Nassar, Phys. Lett. 109A (1985) 1.
[21 ) A. B. Nassar, J. Phys. A18 (1985) L423.
[ 22) A. B. Nassar, J. Phys. AlB (1985) L509.
[ 23) M. Himi and K. Fukushima, Nucl. Phys. A431 (1984) 161.
(24) E. Runge and E. K. U. Gross, Phys. Rev. Lett. 52 (1984) 997.
[ 25) P. K. Chattaraj, S. Rao Koneru and B. M. Deb, J. Compo Phys.
(in press).
[ 26) M. Lakshmanan and M. Daniel, Phys. A107 (1981) 533.
(27) E. Clementi and C. Roetti, At. Data Nucl. Data Tables 14(1974)174.

365
Index of Contributors

Ablowitz, M.J. 44 Joseph, K. Babu 308 Rao, J.A. 176

Akutsu, Y. 282
Kalra, M.S. 184 Sachdev, P.L. 162
Bullough, R.K. 7,250 Kaushal, R.S. 226 Santini, P.M. 118
Kumar, A. 328 Scott, A.C. 343
Chattaraj, P.K 359 Kundu, A. 86 Sudarshan, E.C.G. 2
Chowdhury, A. Roy 105
Lakshmanan, M. 145 Tamizhmani, KM. 145
Dash, P.C. 196 Timonen, J. 250
Deb, B.M. 359 Papageorgiou, V. 66
Dey, Bishwajyoti 188,233 Pilling, D.J. 250 Wadati, M. 282
Popowicz, Z. 212
Fokas, A.S. 66,118
Rajaraman, R. 240,321
Hasan, A. 184 Rangwala, A.A. 176

367
 F. Claro (Ed.)

Nonlinear Phenomena
in Physics
Proceedings of the 1984 Latin American School of
Physics, Santiago, Chile, July 16-August 3, 1984
1985. 110 figures. IX, 441 pages. (Springer Proceed-
ings in Physics, Volume 3).
ISBN 3-540-15273-3

Contents: Mathematical Methods and General. -

Quantum Optics. - Fluids. - Astrophysics and
General Relativity. - High Energy Physics. - Index
of Contributors.

G. Eilenberger

Solitons
Mathematical Methods for Physicists
2nd corrected printing. 1983. 31 figures. VIII, 192
pages. (Springer Series in Solid-State Sciences,
Volume 19). ISBN 3-540-10223-X

Contents: Introduction. - The Korteweg-deVries

Equation (KdV-Equation). - The Inverse Scattering
Transformation (1ST) as Illustrated with the KdV. -
Inverse Scattering Theory for Other Evolution
Equations. - The Classical Sine-Gordon Equation
(SGE). - Statistical Mechanics of the Sine-Gordon
Springer-Verlag System. - Difference Equations: The Toda Lattice.
Berlin Heidelberg New York - Appendix: Mathematical Details. - References. -
London Paris Tokyo SUbject Index.
 B. G. Konopelchenko

Nonlinear
Integrable
Equations
Recursion Operators, Group-Theoretical
and Hamiltonian Structures of Soliton
Equations
1987. VIII, 361 pages. (Lecture Notes in
Physics, Volume 270). ISBN 3-540-17567-9

Contents: Introduction. - General Ideas of

the Recursion Operator Method and
Adjoint Representation. - Linear Matrix
Bundle. - Generalizations of the Linear
Bundle. - Backlund-Calogero Group and
General Integrable Equations Under
Reductions. - Quadratic Bundle with Z2
Grading. - Polynomial and Rational Bun-
dles. - General Differential Spectral
Problem. - Generalization and Reductions
of the Differential Spectral Problem and
Integrodifferential Spectral Problem. -
Towards to the General Theory of Recur-
Springer-Verlag
Berlin Heidelberg New York sion Structure of Nonlinear Evolution
London Paris Tokyo Equations. - References.