8 Springer Series in Solid-State Sciences

Edited by Peter Fulde

Springer Series in Solid-State Sciences
Editors: M. Cardona P. Fulde H.-J. Queisser

Volume 1 Principles of Magnetic Resonance 2nd Edition

By C. P. Slichter

Volume 2 Introduction to Solid-State Theory

By O. Madelung

Volume 3 Dynamical Scattering of X-Rays in Crystals

By Z. G. Pinsker

Volume 4 Inelastic Electron Tunneling Spectroscopy

Editor: T. Wolfram

Volume 5 Fundamentals of Crystal Growth I. Macroscopic Equilibrium

and Transport Concepts
By F. Rosenberger

Volume 6 Magnetic Flux Structures in Superconductors

By R. Hiibener

Volume 7 Green's Functions in Quantum Physics

By E. N. Economou

Volume 8 Solitons and Condensed Matter Physics

Editors: A. R. Bishop and T. Schneider

Volume 9 Photoferroelectrics
By V. M. Fridkin

Volume 10 Phonon Dispersion Curves in Insulators

By H. Bilz and W. Kress
Solitons and
Condensed Matter Physics
Proceedings of the Symposium on
Nonlinear (Soliton) Structure and Dynamics
in Condensed Matter
Oxford, England, June 27-29, 1978

Editors
A.R.Bishop T. Schneider

With 120 Figures

Springer-Verlag Berlin Heidelberg New York 1978

Alan R. Bishop, B.Sc. Ph.D.
Department of Physics, Queen Mary College, Mile End Road
London E1 4NS, England

Dr. Toni Schneider

IBM Ziirich Research Laboratory, CH-8803 Riischlikon, Switzerland

Series Editors:

Professor Dr. Manuel Cardona

Professor Dr. Peter Fulde
Professor Dr. Hans-Joachim Queisser
Max-Planck-Institut flir Festkorperforschung
Biisnauer Strasse 171, D-7000 Stuttgart 80, Fed. Rep. of Germany

ISBN-13: 978-3-642-81293-4 e- ISBN-13: 978-3-642-81291-0

DOl: 10 .1007 / 978-3-642-81291-0

This work is subject to copyright. All rights are reserved, wether the whole or part of the
material is concerned, specifically those of translation, reprinting, re-use of illustrations, broad-
casting, reproduction by photocopying machine or similar means, and storage in data banks.
Under § 54 of the German Copyright Law, where copies are made for other than private use,
a fee is payable to the publisher, the amount of the fee to be determined by agreement with
the publisher.
© by Springer-Verlag Berlin Heidelberg 1978
Softcover reprint of the hardcover 1st edition 1978
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.

2153/3130-543210
Preface

Nonl inear ideas of a "sol iton" variety have been a unifying influence on the
na tura 1 sci ences for many decades. HO~/ever, thei r uni versa 1 a pprec i at i on in
the physics community as a genuine paradigm is very much a current develop-
ment. All of us who have been associated with this recent wave of enthusiasm
were impressed with the variety of applications, their inevitability once the
mental contraint of linear normal modes is removed, and above all by the
common mathematical structures underpinning applications with quite different
(and often novel) physical manifestations. This has certainly been the situ-
ation in condensed matter, and when, during the Paris Lattice Dynamics Con-
ference (September 1977), one of us (T.S.) first suggested a condensed matter
soliton Meeting, the idea was strongly encouraged. It would provide an
opportunity to exhibit the common mathematical problems, illuminate the new
contexts, and thereby focus the "subject" of nonlinear physics at this
embryonic stage of its evolution.
The original conception was to achieve a balance of mathematicians and
phy~cis~ such that each would benefit from the other's expertise and out-
look. In contrast to many soliton Meetings, hO~/ever, a deliberate attempt was
made to emphasize physics contexts rather than mathematical details. The
beautiful examples and techniques of rigorous soliton theory and algebraic
topology developed so fully in the last few years have motivated the physi-
cists' appreciation: we hope that this survey of condensed matter appli-
cations will be accepted as a tribute to the mathematics, and also serve to
highlight some of the more pressing and universal mathematical problems now
facing the nonlinear physicist.
These Proceedings should give a sufficiently coherent flavour of current
activity to introduce nonlinear physics to the newcomer but they are cer-
tainly not exhaustive. Some omissions were deliberate (we considered non-
linear plasmas, optics, hydrodynamics, etc., as extremely important but
sufficiently established in their own right); others were imposed by time
limitations. We apologise to all those subjects and authors who were not
represented, but feel sure that they too will benefit from this exposure of
the greater nonlinear cause. Our main hope is that this contribution will
mark a small step towards the ultimate percolation of the soliton paradigm
throughout all levels of the physics community
The gratitude of the Organizing Committee:
S. Aubry, A.R. Bishop, R. Blinc, A. Bruce
R. Bullough, R.J. Elliott, J. Krumhansl, A. Luther
T. Schneider, R. Stinchcombe, H. Thomas and
S. Trull inger
goes to many individuals and organizations.

v
 Without our Sponsors:
European Research Office (U.S. Army)
IBM Zurich Research Laboratory
National Science Foundation (U.S.A.)
RCA Laboratories (Zurich)
this venture would have been impossible. We are especially grateful for
their encouragement and administrative cooperation which enabled us to meet
optimistic deadlines. Queen Mary College (London University) (through A.R.B.
and D.A.L. Jones) kindly agreed to administer the European grants and the
University of Southern California (through S.E. Trullinger) acted as inter-
mediary for the travel support from NSF.
We thank Springer-Verlag for undertaking such a rapid publication. This
has necessarily placed uncomfortable burdens on authors; we apologize for any
strong-arm tactics and trust they find the present outcome some compensation.
Special thanks to the indefatigable E. Stoll and to the IBM Publications
Department (Zurich): U. Bitterli, M. GrUneisen, P. Theus, K. Thoma, and R.
Trachsler, for their unfailing patience and accommodations.
The Clarendon Laboratory and St. John's College, Oxford, provided splendid
environments for a convivial and stimulating gathering. We are also deeply
indebted to R.J. Elliott and his staff (especially Miss B. Green) for their
hospitality and meticulous attention to local organization, both before and
during the Symposium.
So onwards to the next Sol iton t1eeting. Meanwhile, it only remains for
us to introduce the specific topics included in this one. Fortunately, we are
rel ieved of this duty by the inventive mind of J .A. Krumhansl - ~/e take great
pleasure in reproducing "JAK's Odd Ode to Solitons"", which accomplishes the
task with eminent clarity and distinction.

ZUrich, July 1978 A.R. Bishop T. Schneider

*Taken from Professor Krumhansl 's presentation at the Symposium Dinner. The
editors accept full responsibility for its punctuation and introduction!

VI
Solitons in Oxford

First from Bullough we have heard,

of concepts wild and weird.
How to find waves solitary,
you must be bright and wary,
backing round an inverse path
a la Kpuskal and his math.
Enter now the marching cordon,
flying banners of sine Gordon!

Then from Scott we hear the history,

how his namesake (hyphen Russell),
just to exercise his muscle,
rode his horse and set the course
we've followed hard these days.

Eilbeck, Gibbon, Kitchenside sing praise!

Now along comes Alan Newell,

and he's nobody's fool,
and tells us if we mess around with change
from systems thought conditioned well,
things will really go to hell.

In l-d only can you freeze that pulse of K. DeVries!

Thence on we switch to physics statistical

and how those kinks somewhat mystical
jump through hoops in Bishop's rendition
of Gibbs' function of partition.

Through the saddle we must jump with the solitary hump!

Then shades of Alice's Cheshire cat,

as Wallace tells us where its at.
Through the complex plane, it's gone,
wondrous, wondrous instanton.

But now with dreaming call a truce,

listen to the words of Bpuce.
Renormalize this group's odd missions
with pules well known for phase transitions.

Order, order, everywhere, but also lots of kinks!

VII
 And thenoe to Sohneider's show with StoZZ:
my how those IBM oomputers raZZ
with aZZ the power they oan muster
forming oZuster after oZuster.

l-d,2-d,3-d, 4, - now reaZZy is there need for more?

Thenoe, onward, onto wordZy faots,

we hear from Messrs Bak and Axe
if there is sanity to save,
charge density's your wave.

And that is it so far this week,

but more is to come the morroW:
SoZitons to be more truther
by the words of Awn Luther,
and perhaps sound neat and nice
in prose whioh roZZs from MiohaeZ Rioe.

More new words from Friend and Aubry

and finaZZy, then, that Ziquid waoky
distorted by Kazumi Maki.
Thenoe to wind-up with Steve TruZZinger -
(what rhymes with that?)

So farewell - and end!

These names were but a few of those

whose mystio, mathie, physio prose
has made these days true seem
Midsummer's soZitonio dream.

VIII
Contents

1. Introduction

Solitons in Mathematics: Brief History

By R.K. Bullough and R.K. Dodd .................................... 2
Solitons in Physics. By J.A. Krumhansl 22

II. Mathematical Aspects

Numerical Studies of Solitons. By J.C. Eilbeck 28

Poles of the Toda Lattice. By J. Gibbon 44

Perturbation Theory of the Double Sine-Gordon Equation
B~ P.W. Kitchenside, A.L. Mason, R.K. Bullough, and P.J. Caudrey 48
Soliton Perturbations and Nonlinear Focussing. By A.C. Newell 52

Novel Class of Nonlinear Evolution Equations Solvable by the Spectral

Transform Technique, Including the So-Called Cylindrical KdV Equation
By F. Calogero and A. Degasperis .................................. 68
The Complex Modified Korteweg-de Vries Equation, a Non-Integrable
Evolution Equation. By C.F.F. Karney, A. Sen, and F.Y.F. Chu ..... 71

III. Statistical Mechanics and Solid-State Physics

Soliton-Bound States in the Magnetic Gap. By A. Luther 78

Statistical Mechanics of Nonlinear Dispersive Systems

By A.R. Bishop .................................................... 85
Some Applications of Instantons in Statistical Mechanics
By D.J. Wall ace ................................................... 104
The Theory of Structural Phase Transitions: Cluster Walls and Phonons
By A.D. Bruce ..................................................... 116
Nonlinear Lattice Dynamics: Molecular Dynamics Studies
By T. Schneider and E. Stoll ...................................... 135
Computer Simulation of Structural Phase Transitions. By W.C. Kerr .,. 150

IX
Soliton-Like Features in a Two-Dimensional XY Model with.Ouartic
Anisotropy. By E. Stoll and T. Schneider ......................... 154
Behavior of a ¢4-Kink in the Presence of an Inhomogeneous Perturbation
By·N. Theodorakopoulos, S. Hanna, and R. Klein .................... 158
Solitary Wave Solutions in a Diatomic Lattice
By H. BUttner and H. Bilz .... 0......•..•.••. 0.••••••..• 0.......... 162
Lattice Models of High Velocity Dislocation Motion
By N. Fl ytzani s .......................... 0... 0.. 0• . . • . . . . . . . . • • • • • 166
Grain Boundaries as Solitary Waves
By R.J. Harrison, G.H. Bishop, Jr., S. Yip, dnd T. Kwok 183
The Relation of Solitons to Polaritons in Coupled Systems
By D.F. Nelson .................................................... 187
Solitons in CsNiF3: Their Experimental Evidence and Their Thermo-
dynamics. By M. Steiner and J. K. Kjems ........................... 191
Structure and Stability of Domain Walls - Phase Transition
By J. Lajzerowicz and J.J. Niez ................................... 195
Periodic Lattice Distortions and Charge Density Waves in One- and
Two-Dimensional Systems. By R.H. Friend .......................... 199
Solitons in Incommensurate Systems. By P. Bak 216
Fluctuations and Freezing in a One-Dimensional Liquid: H9 3_oAsF 6
By J.D. Axe ................................................. 0 •••• 0 234
Charge Density Wave Systems: The ¢-Particle Model. By M.Jo Rice.. ..... 246
The Soliton Lattice: Application to the w Phase. By B. Horovitz 254
The New Concept of Transitions by Breaking of Analyticity in a
Crystallographic Model. By S. Aubry .............................. 264
Textures in Superfluid 3He . By K. Maki 278
Creation of Spin Waves in 3HeB .
By P.W. Kitchenside, R.K. Bullough, and P.J. Caudrey 291
The Interaction of Spin Waves in Liquid He 3 in Several Dimensions
By J. D. Gi bbon ............. 0 • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 297
Josephson Transmission Line Oscillators. By A.C. Scott .............. 301
Dissipative Structures in Quasi-One-Dimensional Superconductors
By A. Baratoff ............. 0..........•• 0•....• 0•.•.•.•.•..... 0. o' 313
Solitary Phenomena in Finite Dissipative Discrete Systems
By E. Ben Jacob, and Y. Imry ...................................... 317

x
Stability of Nonuniform States in Systems Exhibiting Continuous
Bifurcation. By M. BUttiker and H. Thomas ........................ 321
The Sine-Gordon Chain: Mass Diffusion. By T. Schneider and E. Stoll 326
Solitary-Wave Propagation as a Model for Poling in PVF2
By A.J. Hopfinger, A.J. Lewanski, T.J. Sluckin, and P.L. Taylor 330
Theory of One-Dimensional Ionic and Sol itary Wave Conduction in
Potassium Hollandite. By J.B. SokDloff and A. Widom .............. 334

IV. Summary
Summary: Where Do Solitons Go From Here? By S.E. Trullinger 338

Index of Contributors 341

XI
I. Introduction
Solitons in Mathematics: Brief History

R.K. Bullough and R.K. Dodd

Department of Mathematics, U.M.I.S.T., P.O. Box 88
Manchester M60 lQD, Great Britain
and
Schoo'l of Mathematics, Trinity College, Dublin 2, Ireland

1. Introduction

This article presents a brief history of the inverse scattering method for
solving nonlinear evolution equations and the Hamiltonian structure associ-
ated with it. It is not a comprehensive survey of the different mathematics
now concerned with soliton theory. To attempt the latter would be inappro-
priate for a meeting concerned with nonlinear dynamics and structure in
condensed matter. In any case, soliton theory already ramifies into areas
of mathematics, algebraic geometry, theory of Jacobian varieties, on the edge
of the mathematical range of one of us (RKB).

The subject surely begins with the observation [lJ of the single soliton
solution of the Kortewes-de Vries (KdV) equation, in the month of August
1834, by JOHN SCOTT RUSSELL. Since then, and until quite recently, the
physics and mathematics of the subject have evolved together, each influen-
cing the other. At the present time there are signs of schism into distinct
and disparate disciplines.

RUSSELL in effect established by experiment [lJ that arbitrary initial

data for the KdV equation

(1)

broke up into solitons. The speed of the soliton depended on its amplitude.
No serious mathematical developments followed this observation until the
work of ZABUSKY and KRUSKAL [2,3J

In contrast, the sine-Gordon equation

u - u = sin u (2)
xx tt
arose in a strictly mathematical context - in differential geometry in the
theory of surfaces of constant curvature [4J. It is worth remarking that
LIOUVILLE gave the general solution of the nonlinear Klein-Gordon equation

uxt = exp m u (3)

(m = parameter) in 1853 [5J. Eq. (3) is a Klein-Gordon equation u = F(u)

'light cone' coordinates, but this is easily expressed in the Lorenf~ co-

2
variant form u - u = F(u), the form of (2). LIOUVILLE's method was
his own, but re2~nt1~t[6J we have used the general solution u = f(x) + g(t)
of u = 0 and a Backlund transformation to rederive LIOUVILLE's solution
in th~tform

exp mu (4)

The Backlund transformation (BT) with free parameter k

ul
X
=
x
U + 2 k sin l (u'+ u)
(5)
-1 2 1
u' :::; -u + 2 k
t t
sin (u'- u) 2
for the sine-Gordon

u sin u (6)
xt
in light cone coordinates was known to Backlund before 1882. We now know
that a BT is characteristic of the 'integrable systems'. This phrase is
used loosely here to mean those systems of nonlinear evolution equations
which have soliton solutions. We define 'integrability' more precisely
later in terms of 'integrable Hamiltonian systems'. We shall show elsewhere
[7J how these ideas relate to the classical integrability theorem of
FROBENIUS.

Integrability is concerned ultimately with integrability conditions

exemplified by (5). The BT (5) relates a solution u of the sine-Gordon
(SG) (6) to another solution u' (it is strictly speaking an auto BT [6J).
The integrability condition is equality of the second partial derivatives
u' = u' . This means that u satisfies a partial differential equation
wB~ch ~~ (6). By adding u' and u' it also means that u' satisfies
a partial differential equafton whichttn this case is also (6). BT's are
generalizations of contact transformations (canonical transformations) in
which the derivatives of u' are expressed in terms of u, its derivatives,
and u'. The structure has been extended by PIRANI [8J. The line of attack
is to replace 'transformation' by 'map' in which x,t,u,u ,u are indepen-
dent variables. Introduce the idea of COO (= infinite1yXdifferentiable)
functions u equivalent to v if u(x,t) = v(x,t) and all their derika-
tives at (x,t) are equal to order k. The k-jet of u at (x,t), j u,
is the equiva~ence class thus defined. The k-jet bundle J (x,t;u) is x the
set of all j u. This is a manifold coordinatized by, e.g., for k = 1,
just (x,t,u,~ ,u). In this language, the Backlund map is a mag ~ : Jl(x,
t,u,u ,u) x N(u~) + J 1 (x,t,u',u',u') in which N(u') is a C manifold
coordtnaEized by u'. Thid defi~itton can be generalized [8J. In essence
Jl is the natural manifold on which to calculate BT's like (5): we have
used this covertly in finding BT's for the Klein-Gordon equations [6J.

Not all equations with BT's are 'integrable' in the sense of having
soliton solution~: in particular LIOUVILLE's equation (3) does not because
u does not vanis~ as Ix! + 00 (it is plain that u = 0 is not a solution:
the soliton solutr~ns decay exponentially to zero at Ixl + 00 so there can
be none of these). '

3
 LUND [9J recently extended the early work on the SG in differential geo-
metry. We use his work here to exemplify the mathematics. Consider the
problem of embedding the n-dimensional surface V in the n+l dimensional
Euclidean space E n
i
E : x (i 1,2, ... , n+l)
\1
V : y (\1 1,2, ... , n),with metric
n
2 \1 v
ds = g\1V dy dy

The embedding is isometric (preserves scalar products) if V can be defined

n
through
i i 1
x x (y , ... , yn)
i \1 i v .
such that g = ax lay , ax lay . Vectors X ~n the tangent s~ace of
V , and the ~~rface normal X define a basis\1 in E, X = ax/ay X
TRe expression of the vectorsn~t layV, ax l/ayv in terms\1of this ba~isnt~'
. \1
' ) system 0 f Gauss-We~n%arten
t h e (1 ~near n+.
equat~ons. e '~ntegra b'~ l'~ty con-
Th
ditions a 2x layVayD = a 2x layDay (for all pairs yD,yV) lead to a non-
linear systeM, the Gauss-Cgdazzi equations.

For a two-dimensional surface in three-dimensional Euclidean space, LUND

[9J finds Gauss-Codazzi integrability conditions which ultimately take the
form
2
+ - (cot 8 \) = 0
(7)
1
2
sin 28 = 0

for the Gauss-Weingarten system

(-i~+*iP. ~)
-q ~~ - ~p
(Vl)
v2
(8)

(::::)" { -:) c:)

in which p = (cos28/2sin 2 8)A , q = 8 + icotSA , r = (V4~)cos28 - (2sin 2 S)-1
A , s = (1/4~)sin28. An eige~value ha~ been ineroduced by appealing to Lie
iavariance (essentially invariance under interchange of ~,n) of the pair
of coupled evolution equations (7) in the two fields 8,A, The conditions
(7) are sufficient and necessary for the integration of (8). The system (8)
is however a generalized ZAKHAROV-SHABAT [10J scattering solution problem
as we shall see shortly. Integration of this linear system is equivalent to
solving the nonlinear system (7). We thus have a natural geometrical basis
to an inverse scattering problem,

4
 However, the BT (5) already contains a similar scattering problem for
the SG. Set f = tan[(u+u')/4J. Then

f lx (9)

k- l fcosu - (2k)-1 (1

The Ricatti transformation f = v 2 /v l now yields

C \ (l_1 1

:~ c:)
n
2
vl,X)= k
1
u +-
2,x 2 x 2

(>') 2,t
1
2k (.':'" 'iun)
Slnu -cosu v2
(10)

This is the same as (8) for A = °

and u = 28 (corresponding to a surface
of constant intrinsic curvature) if i1; - k!2. POHLMEYER [llJ has obtained
a BT for the nonlinear sigma models in n fields ul, ... ,u n coupled only
by the constraint
n
2
I u.
1
1.
i=l
For n 4, the 04 invariant model, the system reduced to LUND's. For
n = 3, the 03 invariant case, the system is the SG. POHLMEYER [llJ has
shown that the results (10) can be extended to (8) by using his BT. The BT
may be much more fundamental than the idea that inverse scattering trans-
forms like (8) or (10) are geometrical in origin. However, BT's can be
derived for example, by the method of prolongation structures due to WAHLQUIST
and ESTABROOK [12J. It is plain that geometrical ideas underlie this theory
although the precise nature of these is not yet fully understood.

The BT can generate infinite sets of conserved quantities. We have demo-

nstrated this elsewhere [5,13J and remark only that by using the BT between
the SG and

u' (11)
xt
which is
u' k-lsin(u' - u)
x (12)
u~ u t + k sin u'

one obtains the conserved densities of the SG

T2 2 T4= u 2_ l u 2 T6 u 2_ ~ u 2 u
2 1
+ - u
6
Ux ' xx 4 x ' xxx 2 xx x 8 x
2 7 2 2 7 4 35 2 4 5
u
xxxx 2 uxxx u x + /; u xx +8- u xx u
x + 64

T10 = etc. (13)

5
 a
Note that Tr has the rank r in that any term in Tr,rr u n, has rank
L n a = r. The odd rank densities have been removed [13J R~cause they
are t¥ivial. Note tha~, because ofoothe assumed boundary conditions u, u x '
u
xx
~ 0, etc., d/dt i
co
T d x = - f X dx = X(+oo) - X(-oo) = 0 since a
x 100 Tn dx is a
density T satisfies a conservation law T + X = O. Thus
-00

constant of the motion I (say). The exis~encexof an infinit~OOset of con-

stants of the motion I n is necesary but not sufficient for the SG to be a
completely integrable (~nifinite dimensional) Hamiltonian system. We return
to this later.

A third equation, the nonlinear Schrodinger equation (NLS) [10J

i u + u + 2u lul 2 = 0 (14)
t xx
has played an important role in the history of the subject. The equation of
a 'simple wave' u + uu = 0 describes a nonlinear system without dis-
persion; the KdV d~scriSes a weakly dispersive, weakly nonlinear system; the
NLS describes a strongly dispersive, weakly nonlinear system. The equation
is of relatively recent origin [14J. Its role [10J in the development of
the inverse scattering method is described below.

2. Discovery of the Inverse Spectral Transform

In §l we mentioned three equations, the KdV, the SG and the NLS, important
to the development of the inverse scattering method. The critical step was
taken by KRUSKAL and co-workers [2,3J for the KdV. It seems worthwhile
sketching the argumentation adopted by these authors as we have heard it
[15J since it throws up a number of points. The argument is physically
based and the mathematical content has emerged later following the work of
LAX [16J. We indicated the nature of this in §3 following where we also
connect the theory with the inverse scattering methods associated with Eqs.
(10) and (8) introduced without further comment in §l.

KRUSKAL in the period 1955-65 was apparently concerned with the period-
icity exhibited by one-dimensional lattices - the FERMI-PASTA-ULAM problem
[17J. First results obtained with ZABUSKY appeared in 1965 [2J. The
two authors were concerned with a continuum limit of a lD-lattice which
leads to
(15)

The Riemannian invariants of the linear problem u = V2(y + y ) = 0(1),

v = V2(yx- Yt) = O(E) lead, for E« 1, to the one-wayXgoin~ wave [T-
4E(t-x) J
(16)

the simple wave. Characteristics of this equation are [18J dx/dT = u =

constant. These lines intersect and the system shocks (the equation can be
integrated by the hodograph transformation [18J). It is interestin thatR
(16) has an infinite set of polynominal conserved densities Tn = u /n: it
has some of the structure of the integrable systems like the KdV equation
(the I = fTndx are the involution - see below). The addition of the
n

6
 2
dispersive term [(6x) /12]y , a further term in the Taylor expansion
used to convert the discretex!~ftice model to a continuum~ balances dis-
persion against nonlinearity and leads to u + uu + ( x ~2()u = 0, the 2
KdV. [For the record 6x ~ 1/64, E ~ 1/20 and x8 2 = 6x /12 XtX~ (0.022) .]
Numerical solutions under periodic boundary conditions with a harmonic
initial condition lead to periodic behavior in time. It is now known that
more generally, under these conditions, the KdV is periodic in space and
'almost periodic' in time.

The simple wave shocks and with the small dispersive term added, the
resultant KdV equation also develops a (smooth) jump. KRUSKAL and co-
workers looked for conservation laws across the jump and found the polynomi-
nal conserved densities u [from the equation in the form u + (V2 u 2+ 8 2
u ) = 0] u 2 and 2u 3 - u 2. They subsequently found ten p6lynominal con-
s~~v~d densities Tr : theXrank of rrtU~ in this case is r = I(l + n/2)a .
T IO has 32 terms (VlO)u IO - 36u7uix-63D u4u~x ... (419904/12155) u~x [19].n

FPU also looked at a lattice with quartic rather than cubic anharmonicity.
This leads to Ytt = yxx(l + EYx2) and thence to the modified KdV

= O. (17)

This also has conservation laws. MIURA found the transformation u = v 2 +

/-6 v connecting these. The remarkable /-6 is actually innocuous: the
MIURAXtransformation [20] u = v 2 + v is a BT connecting
x
u
t
- 6 uu + U
xxx
=0
2x (18)
v - 6 v v + v
xxx
= o.
t x

2
The Ricatti transformation v = a(£n ~)/a~ linearizes v + v and
x
u = ~xx/~. The Galilean transformation

x' = x - Vt, t' ~ t, u ~ u + V (19)

leaves u - uu + U =0 invariant. Hence, u ~ u + A ~ N and the

Schroding~r eig~nvalt~Xproblem has emerged! Furthermore, u =x~ N - A
xx
into the KdV yields

(a) Ie
t
o
(20)
(b) a functional B[~,u] such that B[~,u] = ~t.

GARDNER et al. [3J used these results and the property u, u ,u ~ 0

as Ixl ~ 00 to solve the KdV by the route u(x,O) (initial dat~) ~xxscat­
tering data at t = 0 (by B[~,u] = ~ ) scattering data at time t ~ [via
the G<:>l'fand-Levitan-Marchenko linear i~tegral equation (see, e.g., [1:3])]
u(x,t), the solution of the KdV. The method is now called the 'inverse
spectral transform' (1ST) because (a) A = 0, i.e., the spectrum of the
Schrodinger operator is invariant undertthe KdV flow, and (b) the method
generalizes the Fourier transform for linear systems [21]. The remarkable
discovery by KRUSKAL and co-workers has been the source of all the develop-
ments since, with the single exception of the direct methods developed by

7
HIROTA [22] and CAUDREY [22]. The precise connection of these with the 1ST
is still to be established.

3. Operator Pair Formulation of Nonlinear Evolution Equations

LAX [16] stimulated an important development and added mathematical under-

standing to the 1ST. Given the relati¥ely prime differential operators
u + 1 = (-a 2 /ax2+ u), B = a2n+1/ax2n 1 + lower degree (B is skew sym-
metrig, B* = -B) such th~t there is a one-parameter family of unitary
operators 1 U satisfying U= BU; and t is unitary equivalent under n,
i.e., U- (t) B(t) U(t) istindependent o¥ t:.

(i) The eigenvalues A of L

ari integrals of the motion, i.e., A =
O. [The proof is triviaY: (O)u= L U-
(t) t (t)U(t), t (O)~(O) = A ~(O)~,t
Then t (t)~(t) = A ~(t) witR ~(t) = U(t)~YO), i.e., uU(t) is th~ evolution
operato¥.] u
(ii) f-
at
~-l(t) i: U
(t) ~(t) = 0 ~

_ ~-l ~ ~-l i: u + U + ~-l i:u B U O.

t u
B U =:>- ~-l B L
u
U + U + ~-l i: Ii ~ = 0 and

[B, L] (21)

(ii) Since L = u , the operator equation (21) is an evolution equation

u t = K ~u] in wfiichAtK[U] is a functional of u. By trying for. the skew
symmetr~c operator B = a 3 + ba + ab (a = a/ax), LAX found that w~th b =
-3u/4 (our scaling is different from that used by LAX [16]) he regained
from (21) the KdV equation. Further, the method generalizes to yield an in-
finite hierarchy of KdV equations of degrees 3 (the KdV) , 5, 7, etc.

It is remarkable that LAX's operators can be found by defining the square

root R of the operator. -i by R2 = -t and R = a + c + cIa 1+ c 2 a 2+ •••
with a = a/ax). Then -Lu= ~2_ u and R2u = -L is a 2 + 2c oa + (co + c 0 2+
2c l ) + (c l + 2c 2+ 2c Oc ) a- 1+ .•• , so that c ~ 0, c l = -V2 u, c 2= V4'~
Ev~dently,'fhe principal part of R (that exc2uding inverse powers of a~ is
a and [R,i] Lt is

(22)

-3/2
However, the principal part of R = (-Lu ) R is

a 3 - (3/4) au - (3/4) ua (23)

-5/2
and this is LAX's operator for the KdV. Other fractional powers R ,etc.,
generate LAX's hierarchy.

GELFAND and DIKII [23], and other workers (NOVIKOV, DRINFELD) referenced by
MANIN [24], in particular show how to solve the KdV for its multi-soliton
solutions on the real line < x < by reducing the problem to a system
00

8
of algebraic equations. Analogous but more difficult methods are used by
NOVIKOV [24,25] to solve the KdV under periodic boundary conditions. MANIN
[24] notices interesting new properties in the multi-soliton solutions on the
real line when the operators L, B are not relatively prime,i.e., are of
degrees d (the order of the highest derivative) which are not relatively
prime, 2 and 4 for example. He obta:tns remarkable 'geyseron' solutions in
this case [24]!

4. Symplectic Structure

GARDNER [26] introduced the idea of a bracket (a Poisson bracket) but worked
under periodic boundary conditions. For present purposes, define the func-
tional or Frechet derivative eF/eu of the functional F[u] = foo~u] dx 00

by
dF
dt eu at
r
~ au dx
-00
(24)

where t, usually the time, parametrizes u. Alternatively,

lim
E-+O
a
aE
r ,9T[u + (eu] dx = r~
-00 eu
eu dx. (25)

The bracket corresponding to that found by GARDNER is the unusual one

{F, G}
eF eG (26)
eu ax eu dx.

GARDNER in effect proves

or
o=
dI
dt
n
r--.E.
-00 eu
au
at
(27)

if u = a/ax (eI /eu). This evolution equation is a Hamiltonian flow with

Hamiltonian the mtW constant of the motion I of the KdV equation. Since
I = iooTmdx and T3 = V2(2u 3 - u 2), one fiWds that the KdV is a Hamiltonian
fTow. The constants of the motion x I are in involution, there are precisely
the right number of these, and the syWtem constitutes a completely integrable
infinite dimensional Hamiltonian system - the first example of this class.

To expand this a little more:

au oH3
6uu - u
at op x xxx

C:
(28)
~ .L 3 - 6uu + u
at ax ) x xxx
can be compared wi th the classical prescription for a Hamiltonian flow

(!) (-~ ~) (aH/a q )

aH/ap
= J grad H. (29)

In this, J is a linear map taking phase space to phase space. It is

9
isometric (i.e., preserves inner products) and, and this is the key point,
is skew. It connects the contravariant vector field with a co- (q,p)
variant one (grad H). Phase space is symplectic: it carries a closed skew
symmetric differential two-form w [27J. For (29)
w w.. opi A oqj (i ,j 1,2) (30)
~~l ol oql op2 _ op A oq.

opl and op2 (oql and oq2) are independent variations in p (q). The
matrix J is w.. which is skew. The wedge product (A) is (by definition [27J)
a skew symmetric1tensor op A oq = - oq A op.

The symplectic form w is invariant (coordinate free) and closed, dw O.

From Hamilton's principle if

o !(p dq - H dt) = 0, o f(p dq - H dt) o (31)

describe the same trajectories in phase space,

f(p dq - Hdt - P dq + H dt + dF) = 0

and
o (p dq) o(p dq) + 0 dF 0
so that
d(p oq) - d(p Oq) + d of 0
and
dp o q - dp oq op dq op dq
or
dp oq op dq dp oq - op dq (32)

i.e., w = w under the canonical transformation p,q -+ p,q , H -+ H. In this

somewhat heuristic demonstration, we have used that H(p,q) = H(p,q), that
is take the same values although their functional forms are different: hence
H dt = H dt above.

In (28) the discrete variables q,p in (29) are replaced by the running
set u,p = LX u(x')dx', for each x, and the symplectic form matrix w.. is
replaced by~the skew symmetric operator a/ax. This suggests the sympl~~tic
form
w = dx r (33)

and the Poisson bracket (26). This bracket satisfies Jacobi's identity

+ {{B, C}, A} = 0, (34)

a result which is conveniently proved via the identity

(~~) ]
oB o2B
{{A,B},C} =f [o2A
ou2 ax ou
o}
ax ax (~~) dx . (35)

To prove involution of the constants of the KdV motion I it is sufficient

to know [28J that there is a skew symmetric operator K n - 3 3+ 2ua + 2au
such that the vector field u _ a u for the nth KdV flow (namely u
t n t

10
K [uJ = [L, B J with B LAX's nth skew symmetric operator B) satisfies
an u = K oH /Ru. For th~n
n+l n

0:
61 01 6I
m n
{I ,I }
n m f ax ou
dx = f ou
d
m
u dx
or n K orm-l
or orm-l dIm_1
J Ou Ou
dx f K~
ou ou
dx = - fa n+l u
au
dx

01
n+l
orm-l 01
n+l
orm-l
f ~x ou ou
dx f Ou ax Ou
dx {In+l' 1m- I }, (36)

and continuation down to 1 1= u yields zero [28J.

We shall take as our definition of a completely integrable (or integrable)

system with 2n degrees of freedom such that (a) the system has n con-
stants of the motion I , and (b) these are in involution, i.e., {I ,I }= O.
Integrable systems can Re given a more fundamental description and w~ p~ovide
this elsewhere [7J. For the KdV the infinite set of constants I in
involution is not enough for complete integrability since preciseZp the right
number of constants n = 00 is required. However, the question of integrability
was settled by ZAKHAROV and FADEEV [29J who explicitly and independently
enunciated the fact that the KdV was an infinitely dimensional, completely
integrable Hamiltonian system proved the symplectic form (35) is closed under
the transformation which maps u(x,t) via the Schrodinger eigenvalue problem,
to the scattering data wet), ( and c, and integrated the equations. The
scattering data S = {wet), (~£, c} ~re constituted as follows: wet) is the
re f ect10n
l' coe f f ""
1c1entfor £.
Jos funct10ns [13J ~ ~ e
+ll{X ,x + _00 ; ~£ 1S
. t he
£th bound state eigenvalue (which proves to lie on the positive imaginary axis
in the (-plane) and c,e is the normalization of the eigenfunction. This
scattering data is sufficient to reproduce the potential u(x,t) via the Gel'
fand-Levitan-Marchenko equation. The Hamiltonian in terms of the scattering
data proves to be of action angle type:
m
H
8
'IT r k log (1 - Iw1 2 ) dk - E5
£=1
L (£
5

r 32 m
k 3 P(k) dk 5/2
8
5 L PiC
(37)
£=1

This Hamiltonian reproduces the time evolution of the scattering data:

(38)

These results are not peculiar to the KdV (and the TODA lattice [30J).
ZAKHAROV and SHABAT almost immediately [10J integrat~d the NLS equation
2 ~
1U + u + X ulul = 0 in the LAX form L i [L, AJ with
t xx t
2
L
[~ u'~J x --2
o ' l-p

11
 A= - P [01 OJ ~
1 a/
+ [ iu Ii ~1 +p) iu x *
2 J (39)
-1U
X
-lui l(l-p) •

The scattering problem Lv ~v with v = [v ,v 2 J yields an infinite set

of conserved densities. The NLS is also a Hami tonian flow and proves to be
1
completely integrable.

ABLOWITZ, KAUP, NEWELL and SEGUR [31J then provided a natural generaliza-
tion: namely find £, A such that
L v ~v, L
-iq J
-ia/ax

A v A
[~ (40)

Choose A,B,C,q,r
such that, under ~t = 0, L [A,LJ is the required
pair of evolution equations q = K [q,rJ, r t K2 [q,rJ (here, K. [q,rJ
means functional of q and rJ. T~e potentials q,r form a natural canon-
ica1 pair and the appropriate bracket is

{F, G} r
_00
(OF
oq
oG
or
_ of
op
OG)
oq
dx • (41)

An extension essentially due to CALOGERO [32J yields [33J

a s(x,~,t)
-t
+ h(L ,~, t)· aa a s(x,~,t)
at
+ 2 $"l
-r
1_
,L ,~, t
)
~
s(x,~,t) =0 (42a)
where
a
s =
(~) [~ -~ ]

i> e dx'q J~ dx'r J

e
1 [a/ax - 2r 2r (42b)
=
2i -2q dx'q
a
ax
+ 2q LX
.00
dx'r

as an integrable pair of evolution equations in x,t and any number of

'time-like' variables ~ The extended LAX-AKNS operator pair formulation
is
I1v (x,:t,t,~) A v(x,~,t,~)

11 ~l...- + ~ (~,~,t)· a
a~
at
-iq ] ~ v (43)
-ia/ax v

and 11 L [L,AJ with 11~ = O. We have reported a BT for this system [34J
and its canonical structure is established [33J. No physical application of
this extended formalism is known to us. An extension to bigger N x N

12
matrices has brought some mathematical solutions - the boomerons - with
remarkable properties. This ingenious extension is due to CALOGERO and
DEGASPERIS [32] who exploit Wronskian relations for example an N x N
Schrodinger problem (in which u becomes an N x N matrix) to obtain linear
equations of motion for the matrix reflection coefficients and to which there
corresponds a nonlinear evolution equation for u. NEWELL [35] has recently
extended the AKNS [31] scheme to N x N matrices and generalized the canon-
ical structure to this case.

As an early example of the use of the AKNS 2 x 2 inverse scattering

scheme consider the 'reduced MAXWELL-BLOCH' system of equations [36]

-s, r
(44)
x
Eu + ]Jr, u -Es.
x
It is easy to check that

A B [s + i ]Jr
21;

C r = 102 E - q. - q* (45)

with r -]JS [do not confuse the q,r potentials with the three depen-
dent va~iables r,s,u in (44)], reproduces (44) in the form = L [A,L]
under the condition 1;t = O. The N-soliton solution is [36] t

4
a2 9-n det IMI
ax 2
M
nm
2
E +E cosh [i (e
n
+ e )]
m
m n
1 2 2 -1
en 2
E
n
(x - 4 [E
n
+ 4]J] t + <5 )
n
(46)

Note that when ]J = 0 the system is the SG: equations (44) are Et~ -s,
s Eu, u -Es. Put s = -sino, u = -coso, E = ox; the Oxt= SlnG. In
x
dhs case
i [c~so sino]
A (47)
41; Slno -coso

with q and r in the scattering problem (40) given by q = -Gx/2,=~-r,

This is the form obtained from LUND's result (8) when II =0. The L,A pair
agree with the results obtained from the BT (5) quoted in (10). This pair
was given first by ABLOWITZ, KAUP, NEWELL and SEGUR [37].

Whilst the BT thus shows that the L,A pair for the SG is a natural pair
it is important to notice that this extension of the LAX pair formation by
AKNS is critically different from LAX's original formulation and from its
extensions by ZAKHAROV and SHABAT [10] and those reported by MANIN [24] .

13
For the sine-Gordon equation, a LAX pair of differential operators is a pair
of 4 x 4 first-order (= first degree) differential operators [38J. In the
2 x 2 formulation in the AKNS scheme is not a differential operator [see A
(47)J and it involves the eigenvalues of the scattering problem i v = s v.
Thus, if (i) [A,I,J = £, (ii) Lv = sV, (iii) s = 0, then L(Av - v ) = S
A t " t t
(Av - v t ) and (Av - v) is an eigenvalue belonging to s ,i.e, Av - v
A(S)V or A(S) = O. tIt is easy to see that sol in A acts as an t
integral operator by calculating [i,AJ from (47) explicitly. One finds

[: ~] ~x ]
-sincr [a lax
[L,AJ
2s
~x -d/aX

[~xt t>~t]
sino rO
i
2
II ~] i Lit ( 48)

The first step uses Lv SV to cancel the sol.

5. Theory of the SG

The rest of this article is concerned with developing the theory of the SG
in detail since the result is relevant, in particular,to the work of BISHOP
[39J and TRULLINGER [40J and the work of this meeting.

The SG is a Hamiltonian flow: the two potentials q,r in the AKNS gen-
eralization (40) are q = -ox/2 r = 0x/2 and q = -r = -r*. The pair
cannot constitute an independent pair of canonical variables. However, in
this case, one finds quite generally [33J that
1 oH
H(r,q) H(q) -
2 f oq
(q - r
Ix
) dx.

Then
oH 1 oH
qt or 2 ax oq
1 oH 1 o2H
r +- (q - r lx )' (49)
t 2 oq 2 ""fciL
Thus, if q = rl x ' i.e., r = JX q dx' initially, then it is true for all
time. This agaln indicates in aOOnatural way that the correct choice for the
Poisson bracket is GARDNER's choice (26) and that for the symplectic form is
(33) .

With this in mind, a symmetrized Hamiltonian for the SG is [13,33J

1 -1 Joo fX
H = "4 YO dx [cos 2y O P + cos {-2 q(x' ,t)dx'} -2J

with P = 1/2 Yoo , q = -1/2 °x From this

1 oH 1 1
- 2" °xt = qt = 6p
= 2
sin 2y o p =
2
sino

14
 1
2 YO
-1
at Pt =
oR
6q
1
- 2" YO
-1
rx sin {-2 r q(x")dx"}dx"

or 1 -1 1 -1
er siner (50)
2" YO tx 2" YO

The dimensionless number YO is a coupling constant: note that it does not

?lay a role in the equations of motion. It does playa role in the definition
of the momentum conjugate to q. Thus, it plays a role in the Poisson
brackets and is fundamental to the canonical quantization of the SG equation.

The scattering problem (40) for the SG is studied in terms of Jost func-
tions. Consider the Jost function solutions for v, <p and <P, such that

<r",
[~J e
-it;x
<!''''
[~J e
it;x x -+ _00

if,;x -it;x
ljJ '" [~] e ljJ '" [~J e x -+ +00 . (51)

The eigenvalue t; is real. Since there can be only two linearly"indepen-

dent functions belonging to t;

aCt;) ~ (x,t;) + bet;) ljJ(x,t;)

b(O ljJ (x,o + a(t;) ljJ(x,O. (52)

The quantities w = b(t;)/a(t;) and l/a(t;) are the reflexion coefficient and
transmission coefficient for functions

ljJ '" [~J e -it;x, as x..,. 00

Both wand a- l can be continued into the upper half s-plane: the zeros
of a(s) there define the bound states of the scattering problem (40). The
key result is that Aen/a) <PJ = a/at [(l/a) <PJ ~ a is a constant of the
motion. We demonstrate this explicitly to show what is going on: we have
for x..,. + 00 that
1
at a

[~ ]
b it;x e +it;X} (53)
e
at a
from which

AO f, a(t;, t) aCt; ,0)

b (t;, t) b(t;,O) e- 2A Ot (54)
~

and these results can be continued for 1m s > O. Notice that A is re-
placed by A + f(s,t)I, but [f 1,1J = 0 in agreement with the analysis
above (48). Further, only lim x..,. + 00 of A is needed. Since the ,
potential Ox/2 vanishes for the SG there, lim = is essentially th~ A A
linearized dispersion relation. Indeed from sin2e a"" 0(mod2rr), jxj-+oo, Z47),

15
AS = (i/4s) diag [I, -lJ, whilst the linearized dispersion relation of the
S (6) is w = k-1 ~ AO = i/4s [see (lO)J. The situation is quite general:
for example, for the aRB equations (44), B,C ~ 0 as x ~ 00 since r,s + 0
as x + 00. Then u + -1 and A = is [4 s 2- ~2J-1 = ~(s). In general ~(s)
= -1/2 iw (-2s) = 1/2 iw (2s) inOterms of the linear dispersion relations for
q and ~ of (40) [33]. It is easily checke~ that w (k) = w (k) = k[k2_~2J-1
for thetRMB equations. It is clear that ~(L ,~,t) isqthe nat~ral extension
of ~(L) and hence of the linearized dispersion relation in (42).

With a(s) a constant of the motion established bt (54), it is expressed

as the Wronskian a(s) = W( ,\)I) = Cjl1\)12 - Cjl2h '\. 'l'J.el. x for x ~ 00. In
W(Cjl,\)I), 'I'J. and Cjl2 are the components of Cjl ,i.e., ~ =[Cjl1,Cjl2J and similarly
for \)I. From Lv = sv, the AKNS scattering problem (40) in terms of the
potentials q,r, one finds that <I> = R-n ('I'J. exp isx) satisfies
2 a -1
2 is <I> Ix <I> Ix - q r + q ax (q <I> Ix) . (55)

As Is I + 00 , <I> + 1 so
00

<I> Ix I ~n
(2i s )-n (56)
n=l
(say). Then from (55)
-1
~n+l q
ax
(q .~n ) + I ~ . ~k' n = 1,2, ....
j+k=n J

!ffo - q r (57)

As x + 00, R-n a(s) '\. <I> and since a(s) '\. 1 as lsi ~

n=l
I s
-n (2i) -n
r ~ dx I s
n=l
-n
H •
n
(58)

Then R-n a(s) a constant of the motion yields H = (2i)-n Ioo~dx are con-
stants of the motion and the ~ are conserved d~nsities. -Fornthe SG q =
-r = -ox/2 and these densitiesnare T2, T4 , etc., in agreement with (13).

By expansion about the ordinary point s = 0 one finds a second infinite

set of conserved densities. The point s = 0 is chosen since it is a pole
of the linearized dispersion relation for the SG ~(s) = s/4. This choice
makes ~(s) take its proper form in the evolution equation described by
Hamilton's equations of motion. For the SG this is just the SG itself,
Eq.(6), in light cone coordinates and evolution equation form. This analysis
for the SG is given in Refs.[13J and [33J. However, there is a trivial route
to the first conserved density. Observe that T2= a 2/2 satisfies
1 -1 2 1 -1 x
4 Yo (ox)t + 2 Yo {coso - l}x = 0 (59)

because a = sino. But this SG equation is invariant under interchange of

x and tXt Lie invariance [41J) and therefore
1 -1
2 Yo {coso - l}t O. (60)

16
 The density V2y -l{coso - l} is the first of the sequence obtained by
expansion of ~n a(s9 about s = 0 [13,33]. We now have in V4YO-1a 2 a
momentum density and in V2Y O- 1 {coso - l} a Hamiltonian density. ~ wish
to express these in terms of the scattering data for the following reason:
the scattering problem Lv = sv in (40) is again a canonical trans.formation;
and the Hamiltonian in terms of scattering data again takes the simpler form
of action-angle type depending only on the momenta. This is. exactly the
situation found by ZAKHAROV and FADEEV for the KdV [29] [compare (37)]. The
expression of the Hamiltonian H and the momentum P in terms of scattering
data is achieved by expressing ~n a in terms of the scattering data. This
is done through CAUCHY'Stheorem [33]. The result, which is (65) below, shows
that the SG a = sino is a completely integrable infinite dimensional
Hamiltonian sy~fem. Then by (66) the SG (2) is also.

The symplectic form (33) in the case u = q, p = -Yo L~ q dx' transforms

into [13,33]

w Idp./\dq.
J J
r dt;[d P(O/\dQ(t;)J. (61)

One set of complex canonical coordinates is defined in terms of the scatter-

ing data [13,33] by

Pj = YO ~n 1;;. , q. = 2 ~n b. , j = 1,2, ••• , M.

J J J
-1 - (62)
P(I;) = (2rr I;y o ) ~n {a(l;)a(I;)}, Q(O = arg b(O.

The symmetrized Ham iltonian and momentum derived from the den'sities V2y 0- 1
(coso - 1) and V2Y O- 1o 2, respectively, can be put in the forms [compare
(50) and note we have no~ introduced a mass m in H]
2
H(p,q) = : foo {cos(-2 fX q dx') + cos (2y O p) -2} dx
YO -00 -00

1 foo 2 2 2 (63)
P(p,q) = - (q + Y P ) dx.
2yO 0 x

By expressing these conserved quantities in terms of the scattering data

[13,33] they can be expressed in the forms

H=
2.
m l.
2y o { -2
K
L (s.
J
-1
- 1;.
J
*-1
L
J
-1
) + 2i L 11. + iy o _oodl; I; -1 p(l;) } r
j=l j=l
K L
2i
p = + Y { -2
j=l
*
L (s.J - s.J ) - 2i L 11.J - i YO dE; i; Pco}
j=l
~:
(64)

0
in which 2K + L = M [compare (62)]. ,The result for P suggests new
momenta different from those quoted in (62) for which
K L
H=
L
j=l
h. +
J
L h.J +
j=l
K L
(65)
P L P.J + j=l L p.J +
j=l

17
where h.p. = 4m 2Y o- 2 , h.P.
= 64 m2Y o- 2 sin 2 8, h(~)p(~) = m2 and e. = args .•
These mo~e~ta with correa p6nding coordinates q.,q., etc, also closeJ(61). J
Define now still new energies and momenta by h~OJ= 1!2{h'(O + p~(O}, p(O =
V2{h'(~)- p'(~)} and define similar quantities for h.,p. and h.,p .•
Define a new mass m'= 2m and drop the primed notatio~. J Then J J
K 2 -2 2 ~2 1/2
H I [256m YO sin 8 j + ~ J
j=l
+

K L
P L p.
J
+ I p.
J
(66)
j=l j=l

The expression for H in this form was first achieved by TAKTADJAN and
FADEEV [38J who used the LAX pair of first-order 4 x 4 matrix operators
mentioned earlier. The route used here is via the 2 x 2 ZS-AKNS scattering
problem (40). We can follow it because in light cone coordinates the SG
(2) is the evolution equation u t L~ sin u dx' in which the right side is
a functional of u.

The expression (66) transforms the Hamiltonian of (63) to that for a

collection of free relativistic particles of masses m, 8m YO-l (the SG
kinks) and 16m YO-lsine (the SG breathers). The transformation is canonical
and the result exact. It is an easy matter to canonically quantize it since
the canonical coordinates q., q., q(~) can be found from the symplectic
form. One finds the canonicll plir YO -le., 4 arg b. for the internal co-
ordinates of the breather. The phase spac~ 0 < 8. ~ rr/2, 0 < 4 arg b.< 8rr
is compact with volume 4rr2y -1. The corresponding breather ;pectrum i~ (65)
is therefore discrete. One ~inds [13J a discrete set of breather masses for
(66) given by
16m . n yo
M
n
Sl.n 16 n 1, 2, ••• , N
YO

where N is the largest integer < 8 11YOl This beautiful result has been
reached by a number of authors by very different routes [42J.

For the purposes of this meeting on nonlinear structure and dynamics of

condensed matter, the SG Hamiltonian in the form (66) provides an excellent
model for the statistical mechanics of a soliton system. It is clear that
the solitons and breathers are new elementary excitations; the continuum
contribution is phonon-like. The complication in (66) is that although the
Hamiltonian is separable, it depends on the initial data. In particular,
the number of solitons L, of breathers K, and the occupation of phonon-
like state P(~) depends on the initial data. It is therefore necessar~ to
trace over all sets of available initial data, namely over the set of C
functions decaying exponentially at x = + "'. BISHOP [39J takes the somewhat
different view involving (x,t) space rather than momentum space where
interactions between solitons, breathers and 'phonons' appear as phase shifts
and emptied phonon states. It remains an interesting technical problem to
ally the two different points of view.

18
 Notice that, because the soliton and breather masses depend on YO·',
perturbation theory in YO must be singular perturbation theory above sine-
Gordon solitons and breatners (see [43J and [44J).

Finally, we note that we cannot hope to indicate all the ways in which the
inverse spectral transform method is now being extended. We simply refer the
reader to the paper [35J by NEWELL, recent work on the prolongation structure
method of WAHLQUIST and ESTABROOK by DODD [45J and others, the work quoted by
MANIN [24J and two remarkable papers by ZAKHAROV ~6 ,4~, especially.

References

l. J. S. Russell: "Report on Waves", British Association Reports (1844)

2. N. Zabusky, M.D. Kruskal : Phys. Rev. Lett. 12, 240 (1965)
3. C.S. Gardner, J.M. Greene, M.D. Kruskal, R.M. Miura: Phys. Rev. Lett.
l2., 1095 (1967)
4. G.L. Lamb: Rev. Mod. Phys. 43, 99 (1971), and references therein
5. J. Liouville: J. Mathematiques Pures et Appliquees (Paris) ~ (1) 71-72
(1853)
6. R.K. Dodd, R.K. Bullough: Proc. Roy. Soc. (London) A 352, 481 (1977)
7. R.K. Dodd, R.K. Bullough: "Integrability of nonlinear~olution equations:
prolongations and solitons", to accompany Proc. of the Chalmers Symposium
on Solitons and their Applications ':rI Science and Technology, Goteborg
June 1978 (to be published in Physica Scripta, 1979)
8. Details communicated by W. Shadwick, Warsaw Meeting, September (1977)
9. F. Lund: Phys. Rev. Lett. 38, 1175 (1977); Proc. of NATO Advanced Study
Institute on Nonlinear Problems in Physics and Mathematics (Istanbul,
August 1977), ed. by A.a. Barut (D. Reidel Publishing Co., Dordrecht,
Holland 1978)
10. V.E. Zakharov, A.B. Shabat: Zh. Eksp. Teor. Fis. (Soviet) ~, 118 (1971);
JETP (Soviet) 34, 62 (1972)
11. K. Pohlmeyer: Cornrn. Math. Phys. 46, 207 (1976); New Developments in
Quantum Field Theory and Statistical Mechs., ed. by M. Levy and P. Nitter
(Plenum Press, New York 1977) p.339
12. H.D. Wahlquist, F.B. Estabrook: J. Math. Phys. 16, 1 (1975); 17, 1293 (1976)
l3. R.K. Bullough, R.K. Dodd: In Synergetics. A Workshop, Proc. Int. Workshop
on Synergetics at Schloss Elman, Bavaria, May 1977, ed. by H. Haken
(Springer-Verlag, Berlin, Heidelberg, New York 1977) pp.92-ll9
14. V.L. Ginzburg, L.P. Pitaevskii: JETP 34, 1240 (1958); L.P. Pitaevskii:
JETP ~, 408 (1968); R.Y. Chiao, E. Garmire, C.H. Townes: Phys. Rev.
Lett. 13, 479 (1964); P.L. Kelley: Phys. Rev. Lett. 15, 1005 (1965); T.B.
Benjamin, J.E. Feir: J. Fluid. Mech. 27, 417 (1966);-o.J. Benney, A.C.
Newell: J. Math. Phys. 46, 133 (1967);-V.I. Bespalov, A.G. Litvak, V.I.
Tulanov: Nanka ,2nd All-Union Symposium on Nonlinear Optics, Collection
of Papers, (Russian) (Moscow, 1968)
15. M.D. Kruskal:In Proc. of Symposium on Nonlinear Evolution Equations
Solvable by the Inverse Spectral Transform, Accademia dei Lincei, Rome
June 1977, ed. by F. Calogero (Pittman,London1978); Proc. of NATO Ad-
vanced Study Institute on Nonlinear Problems in Physics and Mathematics
(Istanbul, August 1977), ed. by A.a. Barut (D. Reidel Publishing Co.,
Dordrecht, Holland 1978)

19
16. P.D. Lax, Comm. Pure Appl. Maths. 21, 467 (1968)
17. E. Fermi, J.R. Pasta, S.M. Ulam: Studies of NonZinear ProbZems, Vol. 1,
Los Alamos Rept. LA-1940, (May 1955); CoZZected Works of E. Fermi, Vol. 2,
(Univ.of Chicago Press, 1965) pp.978-88
18. R.K. Bullough: In Proc. of NATO Advanced Study Institute on NonZinear
ProbZems in Physics and Mathematics (Istanbul, August 1977), ed. by A.O.
Barut (D. Reidel Publishing Co., Dordrecht, Holland 1978)
19. R.M. Miura, C.S. Gardner, M.D. Kruskal: J. Math. Phys. 9, 1204 (1968)
20. R.M. Miura: J. Math. Phys. 9, 1202 (1968) -
21. F. Calogero: "Nonlinear evolution equations solvable by the inverse
spectral transform", invited lecture presented at Int. Conf. on Mathe-
matical Problems in Theoretical Physics, Rome University (June 6-15,
1977); A.C. Newell: "The Inverse Scattering Transform" in SoUtons,
ed. by R.K. Bullough, P.J. Caudrey, Topics in Current
Physics (Springer, Berlin, Heidelberg, New York 1979)
22. R. Hirota: In, for example ,"Direct Hethods in Soliton Theory" in SoUtons,
ed. by R.K. Bullough, P.J. Caudrey, Topics in Current Physics
(Springer,. Berlin, Heidelberg, New York 1979) ; P.J.
Caudrey: Proc. of NATO Advanced Study Institute on NonZinear ProbZems in
Physics and Mathematics (Istanbul, August 1977), ed. by A.O. Barut (D.
Reidel Publishing Co., Dordrecht, Holland 1978); R. Hirota: Phys. Rev.
Lett, 27, 1192 (1971); P.J. Caudrey, J.C. Eilbeck, J.D. Gibbon: J. Inst.
Maths.~pplics. 14, 375 (1974)
23. I.M. Gel'fand, L~Dikii: Uspeki mat. nank. 30, 67 (1975); Russian Maths.
Surveys 30, 77 (1975); Funkt, Anal. i Ego Prilog. 10, 18 (1976)
24. Yu, I. Manini: Itogi Nanki i Tekniki 11, 5 (1978) --
25. S.P. Novikov, Funkt. Anal. i Ego Prilozh. 8:3, 54 (1974); B.A. Dubrovin,
I.M. Kricheven, S.P. Novikov: Dokl. AN SSSR (1976)
26. C.S. Gardner: J. Math. Phys. 12, 1548 (1971)
27. H. Flanders: DifferentiaZ Forms with AppZications to the PhysicaZ Sciences
(Academic Press, New York 1963)
28. This argument and some earlier remarks were stimulated by access to notes
by Prof. David Simms on Lectures by H.P. McKean at Calgary, 1978. We have
not been able to check original sources and rely on our recollections
29. V.E. Zakharov, L.D. Fadeev: Funkt. Anal. i Ego Prilozh. 5, 18 (1971)
30. M. Toda: In Studies on a NonZinear Lattice, Ark. for Der-Fysiske Sem.,
Trondheim 2 (1974); "an a Nonlinear Lattice - the Toda Lattice" in SoUtons
ed. by R. K-:- Bullough, P. J. Caudrey. Topics in Current Phys ics
(Springer, Berlin, Heidelberg, New York 1979)
31. M.J. Ablowitz, D.J. Kaup, A.C. Newell, H. Segur: Phys. Rev. Lett. ~,
125 (1973); Studies in Appl. Maths. 53 No.4, 249 (1974)
32. F. Calogero, A. Degaspeds: In SoUtons, ed. by R.K. Bullough, P.J. Caudrey
Topics in Current Physics (Springer, Berlin, Heidelberg, New York 1979)
; Proc. NATO Advanced Study Institute
on NonZinear ProbZems in Physics and Mathematics (Istanbul, August 1977),
ed. by A.D. Barut (D. Reidel Publishing Co., Dordrecht, Holland 1978)
33. R.K. Dodd, R.K. Bullough: In The GeneraZized Marchenko Equation and the
CanonicaZ Structure of the AKNS-ZS Inverse Method (to be published 1978)
34. R.K. Dodd, R.K. Bullough: Phys. Lett. 62A, 70 (1977)
35. A.C. Newell: "The general structure ofintegrable evolution equations"
(Preprint 1977)
36. J.D. Gibbon, P.J. Caudrey, R.K. Bullough, J.C. Eilbeck: Lett. al Nuovo
Cimento~, 775 (1973)

20
37. M.J. Ahlowitz, D.J. Kaup, A.C. Newell, H. Segur: Phys. Rev. Lett. 30,
1262 (1973)
38. L.A. Taktadjan, L.D. Fadeev: Teor. Mat. Fis. 21, 160 (1974)
39. A. R. Bishop: these Proceedings -
40. S. Trullinger: In discussion at this Symposium
41. R.K. Dodd, R.K. Bullough: Proc. Roy. Soc. (London) A 351, 499 (1976)
42. R.F. Dashen, B. Hasslacher, A. Neveu: Phys. Rev. D 11, 3424 (1975);
V.E. Korepin, L.D. Fadeev: Teor. Mat. Fis. ~, 47 (1975); A. Luther:
these Proceedings and published work
43. A.C. Newell: these Proceedings
44. P.W. Kitchenside, A.L. Mason, R.K. Bullough, P.J. Caudrey: these Pro-
ceedings
45. R.K. Dodd: Proc. NATO Advanced Study Institute on Nonlinear Problems in
Physics and Mathematics (Istanbul, August 1977), ed. by A.a. Barut
(D. Reidel Publishing Co., Dordrecht, Holland 1978)
46. V.E. Zakharov, A.B. Shabat: Funkt. Ana1iz. i Ego Pri1ozh. 8, 43 (1974)
47. V.E. Zakharov: In Solitons, erl. by R.K. Bul1ough, P.J. Caudrey, Topics
in Current Physics (Springer, Berlin, Heidelberg, New York 1979)

21
Solitons in Physics

J.A. Krumhansl
Office of the Assistant Director for Mathematical and
Physical Sciences, and Engineering, National Science Foundation
Washington, DC 20550, USA

1. Introduction

Given Bullough's review of the mathematics of soliton solutions of nonlinear

field equations, we see how distinctly different those are from our tradi-
tional linear wave experiences. Therein lies the most serious barrier which
limits much of physics today in this regard: during this century, generations
of physicists have trained in and practiced linear thought, normal mode de-
composition or superposition, and nonlinearity introduced only by perturba-
tive corrections was used to varying degrees to correct this limited base.
But in no way does an extended wave turn into a soliton by perturbation
theory. The other side of the coin, however, is that in strongly nonlinear
problems, mathematicians had not generally provided us with a game plan for
systematically approaching nonlinearity, in the sense that mode analysis was
utilitarian. But that is just what is happening with regard to solitons. It
is being discovered that there is a non-negligible class of nonlinear field
equations which, if not bearing exact soliton solutions, clearly have both
spatially extended and spatially compact solutions, that these solutions are
functionally independent to a high degree (if not exactly so), and that they
are stable with respect to small perturbations. Now the physicist can do
something with these as an augmented set of ingredients in the analysis of
nature - in statistical mechanics, quantum liquids, structural phase transi-
tions, quantum field theory, epitaxy in surface physics, polymer science,
etc., to name some major areas whose utilization is growing.

This conference is about that theme; the mathematical situation is far

more advanced than its deployment in these areas, but more than that I think
there enters another consideration as we address the marriage of the soliton
lore to physics. Specifically, it isn't so much that physical situations
will often conform exactly to the sine-Gordon, nonlinear Schrodinger equation,
or the rest of the integrable arsenal now available, the important point is
that the dichotomy of nonlinear behaviors to be realized leads us to look for
new phenomena. In that sense, a definition:
A Soliton is a Paradigm.

This is meant exactly in the sense of T.S. Kuhn the philosopher of science,
i.e., a fundamentally new pattern of thought for regarding both old and new
problems. That, for the physicist and engineer, is what is responsible for
the excitement in this subject, and brings us together to see to what extent
the joining of this new thought pattern to experiment and theory of condensed

22
matter can be carried out. In the rest of this paper I point to examples
where progress is taking place. The literal term soZiton is overused; the
thought pattern is a different matter.

2. Excitation Modes in Classical Nonlinear Systems [1]

The soliton class of equations allows us to write a field solution as

ljJ(x, t)

This additive statement is utilitarian and totally surprising, indeed it

looks like superposition allover again. Of course, it is not; in point of
fact ljJOS depends implicitly on the form of ljJ 0 ' whether that is a one-
soliton, ~wo-so1iton etc. solution. But a1thougt kot a mathematically linear
superposition it is perfectly good for calling out two distinctly different
behaviors. Details are well developed in several instances; for example, in
sine-Gordon put in a single soliton, then:

-1
ljJ tan exp (la)
E; h - (v /c)2
o s

tanh (lb)
c c

(lc)

Indeed, in this case the separation is exact and a sum over _00 < k < 00 can
be used to make up ljJO . The sine-Gordon equation has many remarkable
features, but among ott~rs note the perfect transmission (save for phase
shift) of the plane waves through the soliton, a significant measure of
independence which the physicist finds useful. Next, note that ljJSOL + E
ljJT NS boosts ljJSOL(x,t) 7 ljJSO (x + E,t), another useful operation physi-
caHy (and to be expected from translational invariance). Finally, we note
that the eigenfrequencies of the oscillatory modes are all real, wk 2 w0 2
+ c 2 k 2 ; thus the soliton is stable against perturbations.

In summary: (a) Simultaneous presence of localized and extended fields,

(b) recipe for translating the soliton, and (c) stability against fluctua-
tions.

As it turns out this pattern of setting out the problem now has consid-
erable practical utility [1] as a general recipe for perturbations about a
fully nonlinear solution. That is:

(a) Write

(b) Substitute in the (fully nonlinear) equation of motion

23
 (c) Collect terms linear in 6~; the resulting equation is linear
and self adjoint, for which a Green's function may be determined.
The eigenfunctions and eigenvalues form a basis for stability
analysis, and response to applied forces, etc.

Application of this method is straightforward; for example, FOGEL, TRUL-

LINGER, BISHOP, and KRUMHANSL [2J have studied the motion of a soliton in the
presence of impurities and applied fields.

3. An Extension to Quantum Mechanics

Luther will discuss fully quantum theoretical methods for an electron gas
system and others equivalent by transformations to the quantum sine-Gordon
system; these are exact models.

However, the method of the previous section can be extended to include

quantum corrections to the classical field. Thus, JACKIW [3J, and others
have written $(x,t) ~ ~S L(xO,t) + 6$ where the tent indicates quantum
variables; that is, we pr2mote the coordinate xo of the soliton to a
quantum variable x O' and eventually find a systematic expansion of the
quantum Hamiltonian in x O' 6$ and their canonical momenta. In lowest order,
the soliton behaves as a massive free particle, and the 6$ as a quantized
oscillator field.

This forges one more link to workaday physics, i.e., quantum calculations.
Significantly, the classical soliton solution is the basis for the method,
and the perturbation about this develops into the quantum fluctuations from
motion of the soliton as a classical particle.

4. Consequences in Statistical Mechanics

BISHOP will review this topic in some detail but to point the way from the
discussion above, we again adopt the physical dichotomy of the two classes of
excitation. The central quantity, in_Baneral, for statistical thermodynamics
is the partition function Z ~ {J} e (or quantum mechanically an appro-
priate trace). It is phenomenologically tempting to hope that Z ~ ZSO ZOSC'
or that F ~ -k T In Z ~ F + F . In fact, KRUMHANSL and SCHRIEFFEk
[4J showed, andBchecked wi~RLan eg~~t functional integral calculation in l-d,
that this separate indeed holds to significant accuracy, although much further
work has been needed to understand the conditions of validity and determine
higher-order corrections.

Thus we have proceeded one step further in the deployment of the solitary
nonlinear excitations into physics: the counting of phase space in the Gibbs
ensemble lends significant weight to soli tonic excitations in many nonlinear
systems.

A very important point to note here concerns the different requirements

of quantum theory vs. statistical mechanics. To lead to a truly stationary
quantum state the system must be integrable, whereas this is not necessary in
order to appear in the partition function - where it is only necessary that

24
extremal paths of the Hamiltonian phase space be dense for a solitary mode.

4. Topology, Dynamics, and Structural Phase Transitions

The spatial variation of the nonlinear fields frequently has a geometric

character associated with it. For example, if the field variable in a sine-
Gordon equation is the rotation of a spin in a plane perpendicular to a
straight line of spins in an applied field, the single soliton is a 2IT
kink in the chain; if a uniaxial perpendicular anisotropy is introduced, IT
kinks can result.

The local potential (_A~2/2 + B~4/4) is a widely used potential (called

~-foup) to describe a field system having two possible minimum energy con-
figurations locally - and if carried uniformly through the region, lead to
long-range order. While ~-foup is not truly integrable, it has kinks which
are highly stable, and carry ~ from one ordering orientation to the other.
A consequence is the possibility that the field breaks up into locally-
ordered microdomains, with nonlinear solitary wave configurations forming the
domain walls. The ramifications of this insight are many - far too extensive
to even begin to summarize here. SCHNEIDER and STOLL will report on several
ongoing series of molecular dynamics studies which not only agree with known
statistical mechanics predictions for phase transitions, but they also show
graphically the presence of clusters; in addition BRUCE develops an analysis
which goes far towards incorporating these phenomenological ideas into an
analytical framework.

Moving on to 3He we find an even richer topological ingredient, as will

be evident in MAKI'S discussion. Again, solitons play an important role in
breaking the symmetry of the ordered phase, as well as introducing dynamic
effects on an entirely different time scale from the traditionally thought of
spin-wave excitations.

Finally, I mention the fertile studies of charge-density wave, mass-

density wave, and similar fields being studied in solids and reported here
(RICE, BAK, FRIEND, AXE, HOROVITZ). The main physics is: (a) the materials
are highly l-d and 2-d in character, thus enhancing the stability of
solitary structures, (b) structural phase changes can take place via propa-
gation of nonlinear kinks, (c) frequently the displacing ions (electron gas)
are loosely bound so that anharmonicity sets in easily.

Thus, in still further ways, we see how the localized and topological
character of solitary nonlinear excitations can signficantly broaden the
concepts we use in condensed matter.

5. Summary and Acknowledgements

By means of several general examples, I have tried to point out that the
studies of solitons in mathematical physics present us with a paradigm for
new perspectives on a wide class of phenomena in condensed matter. The
ideas are exciting, and although physicists may not find many true solitons
the hunt will reveal many common cousins.

25
 My appreciation for discussions goes to many here, but my particular
thanks go to my former colleagues: Alan Bishop, Steve Trullinger, and
Baruch Horovitz.

References

1. A.C. Scott, F.Y.F. Chu, D.W. McLaughlin: Proc. IEEE 61, 1443 (1973)
2. M.B. Fogel, S.E. Trullinger, A.R. Bishop, J.A. Krumhansl: Phys. Rev.
Lett. 36, 1411 (1976), 37, 3l4(E) (1976); Phys. Rev. B 15, 1578
(1977)- - -
3. R. Jackiw: Rev. Mod. Phys. 49, 681 (1977)
4. J.A. Krumhansl, J.R. Schrieffer: Phys. Rev. B 11, 3535 (1975)

26
II. Mathematical Aspects
Numerical Studies of Solitons

J. Chris Ei1beck
Department of Mathematics, Heriot-Watt University
Edinburgh, EH14 4AS, Scotland

1. Introduction
It is now well-known that the first quantitative study of soliton-like behav-
iour was made by SCOTT-RUSSELL [1,2] in 1834. Using what we would call today
an analogue computer, but refered to in Victorian parlance as a wave tank, he
observed the remarkable stability of the solitary wave in shallow water and
the breakup of an initial pulse into two solitons. The delightful description
of his first encounter with the 'Wave of Translation' has been reprinted in
the survey by SCOTT et a1 [3] : it seems likely that the canal involved was
the Grand Union Canal which by happy coincidence passes close to the Riccarton
Campus of Heriot-Watt University.
The work of KORTEWEG and DE VRIES [4] lead to an analytic treatment of the
solitary wave in 1895, but the full richness of the nonlinear phenomena in-
vol ved was ignored until i nteres twas revi ved by the numeri ca 1 experiments
of FERMI PASTA and ULAM [5] in an apparently unrelated field. In 1965 ZABUSKY
and KRUSKAL [6,7] derived the KdV equation as an anharmonic crystal model
and showed by a series of classical numerical experiments that solitary wave
solutions were extremely stable on collision. Independently, and three years
earlier, PERRING and SKRYME [8] had investigated a nonlinear Klein-Gordon
equation later to be christened the Sine-Gordon (SG) equation, and had shown,
first numerically and then analytically, that a two soliton (kink) solution
of this equation existed. However the world of high energy physics was at
that time unready for nonlinear field theories, and this work received little
attention. In contrast the work of Zabusky and Kruska1 fell on fertile ground
and quickly let to the explosion of theoretical work which is still growing
today.
Once the full richness of the analytic results became available, the num-
erical work took second place for some years. Indeed it seemed at one time
that almost all equations with solitary wave solutions would be shown to
exhibit exact multisoliton behaviour, if only the right scattering problem
could be found. It is now becoming more and more apparent that there are
many examples of inexact, quasi-soliton behaviour. In addition 'exact' soli-
ton equations extended to two or more space dimensions have localized solutions
which exhibit facinating features but which are not solitons in the classical
sense [9]. For these problems little or no analytic results are known and
numerical studies are essential for a quantitative and qualitative understand-
ing of the phenomena.
In a paper on numerical studies of solitons it is necessary to say a little
about numerical methods for nonlinear dispersive partial differential equations

28
especially since several recent papers are guilty of exhibiting numerical
results with no mention of the associated numerical analysis. Section 2 is
devoted to a lightning survey of some basic methods commonly used in the lit-
erature. In Section 3 studies of nonlinear Klein-Gordon type equations are
reported, together with a discussion of the aprropriate numerical techniques.
The rather different problems associated with 'long wave' equations such as
the KdV and the Regularized Long Wave equation are discussed in Section 4,
followed by a brief and rather random survey of other single and coupled
equations in one space dimension in Section 5. Finally in Section 6 some
recent work on equations in two and three space dimensions is reviewed.
2. Basic Numerical Methods for Nonlinear Dispersive Wave Equations
In the field of partial differential equations, numerical analysis is still
almost as much an art as a science. For any specific equation, the question
of the 'best' numerical method is an extre~ely complicated one. Even a part-
ial answer to this question depends on many factors: for example the final
accuracy required for a specific range of the independent variables, the
limitations of time and storage space, and machine word length. Equally imp-
ortant is the amount of time and effort available for the development of the
appropriate software. Most computational physicists or applied mathematicians
are content to derive one reasonably efficient method and few detailed com-
parisons are made with other methods. For this reason no outstanding claims
are made for the methods described below except in the cases where meaningful
tests have been carried out. Much further work is required, not just in theo-
retical analysis of the accuracy and efficiency of different methods, but in
practical field tests involving near-optimum codes.
In this brief section I can do no more than outline a very basic set of
numerical tools: a full description will be found in the books by RICHTMYER
and MORTON [10], MITCHELL [11 ], MITCHELL and WAIT [12] and AMES [13], and
in the review talks in the Conference on the State of the Art in Numerical
Analysis held in York in 1976 [14]. For definiteness I shall discuss methods
as applied to the general evolution equation
ut = L(u) (2.1 )
Where L(u) is some general nonlinear differential operator in the space vari-
able x, and finite boundaries are prescribed with appropriate boundary condi-
tions.
Two basic approaches are used to reduce the problem to one involving only
a finite number of parameters: the function approximation approach and the
finite difference approach.
The function approximation method, as its name implies, approximates the
exact solution u(x,t) by an approximate solution defined on a finite dimension
subspace (usually in the x variable only):
u(x,t) ", u(x,t) = ~ co (t) cpo (x) (2.2)
i =1 1 1

The cpo(x) are appropriately chosen basis functions. Common choices for these
are t~e trigometric functions, leading to a finite Fourier Transform or pseudo
spectral method and piecewise polynominal functions with a local ba~is, giving
the Finite Element Method. Assuming the cpo1 have been chosen to satlsfy the

29
boundary conditions, (2.1) becomes
N
r(x,t) = ut - L(u) = 1E c.(t)¢.(x)
1 1
- L(u) (2.3)

where the dots represent differentiation with respect to t,

The residual r is to be made small in some sense: the most common criteria
is the Galerkin approach
1

f r(x,t)¢j(x)dx = 0 j 1, •.. ,N (2.4)

leading to a system of ordinary differential equations.

An alternative approach is to set the residual to be zero at a given set
of points X1, ... ,X N

j = 1, ..• ,N (2.5)

This is the collocation method: although little used at the present in part-
ial differential equations, it is gaining popularity in the field of ordinary
differential equations [15 I.
Function approximation methods are relatively new, and for many applications
the more familiar finite difference methods are still the most popular. For
this approach we seek approximations u~ to the original function u(x,t) at a
set of points xm' tn on a rectangular grid in the x,t plane, where xm = hm
and t = kn. By expanding function values at grid points in a Taylor series,
n
approximations to the differential equation involving algebraic relations
between grid point values can be obtained. Assuming some suitable approxi-
mation Lh(u~) for L(u) has been derived, various types of scheme can be con-
structed according to the type of approximation for the time derivative.
Some common choices, together with the appropriate computational molecule
[131 are shown below.
n+l n
urn - urn = k Lh (u~) .-L. (2.6)

n - un- l = (2.7)
um k Lh (u~)
m T

+
n+l n-l
urn - urn = 2k Lh(u~) (2.8)

Eqns (2.6-8) are known as the simple explicit, simple implicit, and leap-
frog schemes respectively. Two further permutations are important, the Crank-
Nicholson scheme obtained by adding (2.6) and (2.7), and the Hopscotch scheme
[161 which applies (2.6) at odd values of (n+m) and (2.7) at even values:
this cunning combination can result in (2.7) becoming explicit, as the

30
following diagram shows.

For (2.7) to become explicit, the nonlinear part of Lh(u~) must be replaced
by a space average Lh(~(unm+ 1 + unm- 1))' The computational molecules shown
above are for the case where L is a second order operator and the simplest
difference schemes are used: for more complicated cases (e.g. the KdV equ-
ation) some modification will be necessary.
All methods must satisfy the basic stability requirement, that is
lu(xm,t n) - u~ I must remain bounded as n -> "". In addition we would like
this error term to remain as small as possible. Stability requirements often
lead to a limit on the time step k as a function of h, and accuracy require-
ments require both these to be small. The local truncation error is defined
to be (p.d.e - f.d. approx) and the method is said to be of order 1 in time
and j in space if it is O(k l + hj ).
Comparing these finite difference methods, for linear equations the Crank-
Nicholson scheme is probably the most efficient of those discussed here, but
in nonlinear cases it results in a set of nonlinear simultaneous equations
at each time step, requiring much more work. The leapfrog scheme (2.8) is
unstable for all choices of the timestep k for the linear heat equation, but
a similar version for hyperbolic second order equations is an extremely pract-
ical method, as we shall see in Section 3 below. Finally the Hopscotch scheme
[16 I is extremely simple, fast, and stable but the step length k in the para-
bolic case is constrained to k ~ h2 to preserve reasonable accuracy. Little
work has been done on comparisons of finite difference and function approxi-
mation methods for nonlinear problems.
With this Simple survey of numerical methods we can now discuss some num-
erical studies of solitons, starting with a review of work on nonlinear Klein-
Gordon equations.
3. Nonlinear Klein-Gordon Equations
The equations
uxx - utt = F(u) (3.1 )

for various choices of F, have played an important part in the study of soli-
tons. The most common are
F(u) sin u (Sine-Gordon) (3.2)
F(u) -u + u3 (CP~) (3.3)
F(u) u - u3 (CP:) (3.4)
F(u) sin u + Asin 2u (Double Sine-Gordon) (3.5)

31
 The ¢-four equations (3.3) and (3.4), so named from the Lagrangian from
which they are derived have kink and soliton solutions respectively, whereas
the Sine-Gordon equations have kink solutions (different asymtotic values as
x-+±oo).

As mentioned in the introduction, PERRING and SKYRME [8] were the first
to investigate the SG equation numerically. They considered two simple num-
erical methods. First was a simple leapfrog scheme, equivalent to

(3.6)

Numerical tests, and a linear stability analysis, showed that this scheme
was unstable for k = h, but it was found that reducing k to 0.95 h was suffi-
cient to remove this instability. The second scheme involved re-writing the
equation as a pair of first order equations, probably in the form
(3.7)
sin u
and introducing new variables n,~ t ± x so that (3.7) become
un = ~ v , \ = -~ sin u ( 3. 8)
the socalled characteristic form. Since the characteristics in this case are
straight lines, (3.8) can be solved by standard predictor-corrector techni-
ques for ordinary differential equations [13]. This method in [8] was prob-
ably based on the familiar trapezium rule approximation which gives a second
order accurate method. Although slightly more accurate than (3.6) it involves
more work since an iteration of the corrector step is required at each time
step.
Using both these schemes the kink-kink interaction was investigated and
an analytic formula was derived following the numerical results. The bound
state kink-antikink pair (meson/bion/breather) was also investigated and a
numerical study showed that a kink-bion interaction was also stable.
Following the well-known analytic discoveries concerning the solutions of
the SG equation, little progress was made on numerical methods until ABLOWITZ,
KRUSKAL and LADIK [17] showed that the leapfrog scheme (3.6) can be stabilized
n in the final
for k = h by using a space average I(u n . + un 1) instead uf um
m+1 m-
term of (3.6). This gives, for k = h
n+ 1 n- 1 n n n n (3.9)
urn = -urn + um+l + um_l - k2F(~(um+l + um_l ))

This modification reduces the number of multiplications by one, but more

importantly the term in unm is now abs~nt, and calculations can be performed
on a diagonal grid involving only even values of (n+m). This reduces the
running time by a factor of 2. In practice only two time levels need to be
stored and the final code is extremely simple and efficient. Numerical tests
carried out by the author have confirmed that at least for medium size step
lengths (h = 0.1) the scheme (3.9) is more efficient than both second and

32
fourth order predictor-corrector methods based on characteristics. The num-
erical tests have also shown no sign of the nonlinear instability of the kind
exhibited by NEWELL [18] in schemes for nonlinear diffusion equations and the
KdV equation, though this possibility is still under investigation.
Using scheme (3.9), ABLOWITZ et al [17] investigated the collisions of two
solitary wave solutions of both the ¢~ equation (3.3) and the Double Sine-
Gordon equation (3.5) with 0 ~ A ~ 10. For the DSG equation they showed that
when the two pulses were moving very slowly with respect to each other there
was a significant amount of radiation during the collision process, but at
higher relative velocities very little, if any, radiation was present. Inde-
pendent calculations by DUCKWORTH et al [19,20] confirm that at moderate
relative velocities the DSG kinks show near-exact soliton behaviour. It
seems likely that the inelasticity on collision is present at all relative
velocities, but is so small at high velocities that more accurate calculations
would be required to observe it. However the possibility of some sort of
inverted threshold effect cannot be ruled out. The work by DUCKWORTH et al
[19,20] also reveals a large amount of interesting phenomena in solutions of
the DSG equation which are discussed in detail in the references.
For the ¢~ equation ABLOWITZ et al [17] found evidence of inelastic beha-
viour at all relative velocities, a result obtained independently by
KUDRYAVTSEV [21]. Similar results were found for a square wave potential
version of F(u). If the relative velocities of the kinks is small enough,
their identity is lost on collision, and a relatively long lived oscillating
state is produced which behaves much like the Sine-Gordon bion.
It is interesting to note that analytic multisoliton-like solutions in
two space dimensions have been found for the DSG and the ¢4 equations [22].
These solutions collapse into a single kink in one space dimension, so they
are of little help in understanding the results discussed above.
Figure 1 shows some frames from a computer film in preparation showing
solutions of the Sine-Gordon and ¢~ equations. Although the sequence depicted,
a two bion collision, was obtained from the exact analytic solution of the
SG equation, it would probably have been more efficient to generate it num-
erically. The film also shows the collision of two kinks in both the SG and
¢~ models.

The bion solution of the SG equation is an exact solution formed by taking

a two-kink solution with complex conjugate ener0ies for each kink, in other
words a bound state of two kinks. Since kink collisions in the SG equation
are elastic, there is no way that the bion SG solution can be formed from two
free kinks. However in the ¢~ model, two kinks travelling at small relative
velocities can lose enough energy during collision to fall into the bound-
state potential well. These ¢~ bions are a single oscillating hump [9] and
it is possible that solutions with more structure as in the SG bions shown in
Fig.l remain to be discovered.

33
 O,OJ 0,50

1 , 00

~ Frames from a computer produced 16 mm film showing analytic and num-

erical solutions of the Sine-Gordon and other nonlinear Klein-Gordon equations.
This sequence shows the analytic two-bion solution of the SG equation.

34
4. 'Long Wave' Equations
The most familiar equation for long water waves is that due to KORTEIJEG and
DE VRIES (4) in 1895.
ut + ux + uu x + uxxx = G (4.1)
The linear term in Ux can be removed by a transformation to a frame moving
with unit velocity to give the more familiar form
(4.2)
This equation played a fundamental role in the understanding of soliton phen-
omena: it was first studied numerically by Zabusky and Kruskal [6,7). They
used a leapfrog finite difference scheme.

(4.3)

The space average term for u in the finite difference approximation for uU x
is a device to conserve the energy ~E(U~)2 to within terms of order O(k2).
The linear stability requirement for this scheme is that (23)

k/h 3 < 4+h}lu I (4.4)

o
which means that a very small time step must be used to preserve stability
(here uo is the maximum value of u in the range of interest).. The nonlinear
stabil ity properties of this scheme have been extensively studied by NEI~ELL
(18). A survey of some finite difference schemes for the KdV equation appeared
in the paper by VLIEGENT HART (24) in 1971 and in 1976 GREIG and MORRIS (23)
proposed a Hopscotch scheme for this equation. This scheme has a considerably
less stringent stability·requirement.
k/h 3 c .1 (4 5)
.... 12-h 2 u I .
o·

The Hopscotcn scheme requires less storage than the scheme (4.3) and can be
shown to possess only a small phase error in its Fourier modes. However the
small time step required even with (4.5) is apparent: a two soliton co'llision
described in (23) required 4800 time steps to achieve an accuracy of 1% in
the final amplitudes.
Very recently, FORNBERG and WHITHAM (25) have published details of numeri-
cal studies of the KdV equation and related equations, especially of the type

(4.6)

35
where K(x) can be chosen to give various dispersive effects. These authors
used a pseudo spectral method for the x variable together with a leapfrog
method in t. This approach is ideally suited to equations with linear inte-
grals terms such as (4.6). The scheme used in [251 has the stability require-
ment
k/h 3 ~ 2;2 ~ 0.1520 (4.7)

The accuracy of this method is claimed to be high, but unfortunately there

is no direct comparison with the Hopscotch scheme in [231. Ref.[251 also
considers the higher order KdV equations

ut + uP Ux + u = 0 (4.8)
xxx
and numerical results show that with p ~ 3 the soliton collision is inelastic.
i
However, results for (4.6) with K(x) = exp(-JxJn/2) show some evidence for
an exact two-soliton solution.
As an alternative model to the KdV equation, PEREGRINE [261, and BENJAMIN
et al [271 have proposed the socalled regularized long-wave equation
ut + Ux + uU x - uxxt = 0 (4.9)
Numerical schemes for this equation were studied by EILBECK and MCGUIRE [28,
291. They derived a three level finite difference scheme which was stable
for all practical values of k, and in fact works best if k = h. This means
that a considerably greater time step can be taken in the numerical study of
(4.9) as compared to the KdV equation (4.1). The scheme in [291 can be made
conservative by a space-averaged nonlinear term as in the scheme (4.3).
The numerical study in [291 showed that after a collision of two solitary
waves the pulses re-appeared with amplitudes within 0.3% of their original
amplitude. On the basis of this evidence, and on the evidence of a computer
produced film showing the two and three solitary wave collision, it was con-
cluded that the RLW equation (4.9) had exact multisoliton solutions. A sequ-
ence of frames showing the three 'soliton' solution is shown in Fig 2.
In addition to the two and three-solitary wave collision the film shows
the breakup of an arbitrary initial pulse, in this case chosen to be a square
wave of height 0.9 and length ~ 12.17. With these initial conditions in the
KdV equation, theory tells us that two solitary waves plus a large oscillat-
ing tail (radiation) would be observed. The numerical results for the RLH
equation show results in close agreement with the theoretical results for the
KdV equation, even though the amplitudes involved are large enough for sub-
stantial differences between the two equations to manifest themselves. At
present we have not even a qualitative theory for the evolution of one or
more solitary waves from arbitrary initial data for the RL~ equation, and
this remains an outstanding problem for the analyst.

36
Fig.2 Frames from a computer
produced film showing numerical
solutions of the regularized
Long-Wave equation (4.9). This
sequence shows the collision and
re-emergence of three 'solitons'.

37
 Unfortunately the pictorial evidence was not sufficiently accurate to pick
up the inelastic radiation occurring for the two solitary wave collision in
this model. It was left to a Russian team, ABDULLOEV et al [30] to point
out that when very large sol itary waves (-10) coll ide in the RLW equation,
a very small oscillating tail (-10- 3 ) appeared, which could not be adequately
explained as numerical error. These authors showed that for higher order RLW
equations corresponding to (4.8) the inelasticity was even more pronounced.
A more striking demonstration of this effect has been achieved by CALOGERO
and SANTARELLI [31] who made use of the fact that (4.9) has solitary wave
solutions travelling in both directions. I-Jhen two travelling in opposite
directions are allowed to collide the inelastic effect is much greater and
extra solitary waves appear.
Finally, BONA, PRITCHARD and SCOTT [32] have recently derived a scheme
which is fourth order in space and time for the RLW equation, and have demon-
strated rigorously the existence of the small oscillating tail for a range
of two solitary wave collisions.
Two important points arise from the study of the KdV and RLW equations.
Firstly, if testing for exact soliton behaviour, a very careful analysis and
an accurate numerical scheme are required. Secondly, in deriving an efficient
numerical scheme to model a physical situation, it may be necessary to return
to the original mathematical model and modify it in a way which will lead to
well-behaved numerical schemes. It was precisely for this reason that
PEREGRINE [26] first proposed the RLW equation.
Exactly why numerical schemes for the KdV equation should require such a
small time step is unclear, but one possible explanation is that for large
wave numbers the linear Fourier modes of the KdV equation have a large nega-
tive velocity, whereas the velocity of the corresponding modes of the RLW
equation remain small and positive [27].
5. Other Equations in One Space Dimension
Nonlinear Optics has always provided a fertile area for interaction between
experimental results, numerical studies and analytic work. The discovery of
the phenomena of Self-Induced Transparency (SIT) by MCCALL and HAHN [33]
followed from the numerical and analytic study of solitary wave solutions of
the appropriate equations for the envelope of an intense coherent light pulse
in a low density dielectric. Other numerical work in this field is reported
in the excellent review by LAMB [34]. Although the envelope and approximate
field equations have exact multisoliton solutions [35,36], numerical studies
of the exact field equations [37,38] based on fourth-order predictor corr-
ector methods have shown that exact soliton behaviour breaks down at high
densities.
The field of plasma physics has also provided many nonlinear wave equ~tions
with soliton-like properties. Some important studies on KdV-type equations
have been reported by TAPPERT [39,40]. A great deal of work on soliton phen-
omena in plasma physics has been performed in the USSR, and the recent review
by MAKHANKOV [9] is a valuable source of reference. In this work, numerical
studies of various modifications of the Boussinesq equation and of the non-
linear Schrodinger equation with self consistent potential are made, with
emphasis on soliton-like phenomena in plasma physiCS. In all cases the modi-
fications to the basic equations introduces some degree of inelasticity into
the soliton interaction. Details of the numerical scheme used in the case of

38
the modified Boussinesq equation are given in the paper by BOGOLUBSKY [41 I.
Some interesting work on nonlinear partial difference equations approxi-
mating the nonlinear Schrodinger equation has been described by ABLOWITZ and
LADIK [42 I. By construction these difference schemes can be solved by the
inverse scattering method and have multi-soliton solutions. The resulting
scheme is rather inefficient and slow, but studies of this kind may be useful
in "shedding light on the difficult questions of fully nonlinear analysis".
The wide-ranging review by MAKHANKOV [91 also deals with realistic field
theory models. In one space dimension, in addition to the SG and ¢4 equations
described in section 3, he also reports work on various coupled relativistic
field equations. However the most interesting results in this review occur
in three-space dimentional calculations, which will be discussed in the follow-
ing section.
6. Numerical Studies of Solitons in Higher Space Dimensions
Studies of nonlinear wave motion in more than one space dimension are more
restricted than the more familiar results in one dimension. In general, most
papers are devoted solely to a study of a single 'solitary wave' solution,
and the goal is to find such solutions which remain localized for all time.
Such solutions are commonly called 'solitons', especially in the particle
physics literature, although no collisional stability of the sort exhibited
by one dimensional 'classical' solitons are demonstrated. Some multisoliton
solutions of equations in several space dimensions have been derived, but
these are almost always plane wave solitons: one rather complicated exception
with localized multisoliton solutions is given by NEWTON [43 I.
MAXON and VIECELLI [44,451 have derived spherically and cylindrically sym-
metric versions of the KdV equation for small amplitude acoustic waves in a
collisionless plasma of warm electrons and cold ions. The equations are
-1
un + (vn) u + UUt; + ~Ut;t;t; = 0 (6.1)

where n is the time coordinate, t; the radial coordinate, and v = 1,2 for
speerical and cylindrical symmetry respectively. The authors examined the
evolution of a single solitary wave using a leapfrog finite difference scheme
based on that used by ZABUSKY and KRUSKAL [7]. It was found that the pulse
steepened and narrowed as time increased, a result confirmed recently by the
ana lyti c work of CUMBERBATCH [4·6 I.
Similar work on the collapse of plasmons in the dynamics of Langmuir turb-
~lenc~ in plasmas is described in the review by MAKHANKOV [9 I. The problem
1S qU1te analogous to the self-focusing effect of a laser pulse in matter and
plasmas, reviewed in the paper by AKHMANOV et al [471 and recently discussed,
with appropriate numerical schemes, by MATTAR [481, and IiATTAR and NEWSTEIN
[49 I.

Perhaps the most interesting results on soliton-like solutions in more than

one dimension are the discovery of long-lived sQherically symmetric pulsating
solutions of the ¢4 and SG equations by BOGOLUBSKY and MAKHANKOV [7,50-54 I.
These solutions, called pulsons, are seen when the radially symmetric equations
a~e numerically integrated with initial conditions corresponding to a spherkal
k1nk. For example for the ¢~ equation initial values of u(r,t) were taken
to be

39
 u(r,t) = tanh [(r-R o)/12], Ro» 1 (6.2)

together with unspecified boundary values for ut (presumably ut = 0, although

the numerical results were not sensitive to small changes in the initial con-
ditions). Initially this spherical kink collapsed towards the origin, and
close to the origin the solution oscillates wildly and is reflected, in much
the same way as a wave reaching the unsupported end of an elastic string or
bar. After reflection the wave returns to a state close to its initial one,
losing kinetic energy in the process, then the whole cycle is repeated.
During each cycle energy is gradually leaked away to infinity, but, depending
on the initial value of Ro' the pulson can maintain its identity for several
cycles until a final outburst of energy leads to its collarse.
Figure 3 shows the results of a similar calculation made by the present
author during preliminary work on a computer film on pulson solutions of non-
linear Klein-Gordon (NLKG) equations. Unfortunately no details of the numeri-
cal methods used by BOGOLUBSKY and MAKHANKOV have been made available. For
the calculations presented here and in the accompanying cine film extracts I
used a simple leapfrog scheme analogous to that used in one space dimension
for the NLKG equations (3.9), applied to the spherically symmetric NLKG equ-
ati on
-1
urr + 2r ur - Utt = F(u) (6.3)
At r = 0 this becomes, assuming ur(O) o [ 55 ]
3u rr - Utt = F(u) (6.4 )
A leapfrog scheme for (6.4) at the r = 0 gridpoint would give rise to large
oscillations in the case 6r = 6t, so at this point an implicit scheme based
on a time averaged approximation to the space derivative was used. A fixed
grid spacing of 6r = 0.025 was taken, but in view of the large gradients
during the reflection process it would be more satisfactory to introduce a
finer grid near the origin or to stretch the variables in this region to give
higher accuracy [56 ].
BOGOLUBSKY and MAKHANKOV studied pulson solutions to the SG equation and
the ¢~ equations (3.2,3.4) in addition to the ¢~ equation, and found similar
results. For these three equations, near-exact small amplitude pulson solut-
ions were found analytically. In addition BOGOLUBSKY [53] has studied 'heavy'
pulson solutions of the SG equation, in which the initial kink drops from 4n
to O. He showed that this pulson behaved like a '4n' Dulse for some oscilla-
tions,then 'cascaded' into a '2n' pulson for further oscillations before its
identity eventually disappeared. Finally in a recent letter BOGOLUBSKY et al
[54] have studied the head-on collision of two small-amolitude Dulsons in the
¢~ case, and have shown that these pulsons regain some individual identity
after the collision process, if relative velocities are large enough.
These initial results on multi-dimensional localized solutions are very
interesting, and no doubt many more results will follow from this line of
investigation. It will be interesting to see if any analogue of the 'reson-
ance' effect observed by ABLOWITZ et al [17] in the low energy collision of
the ¢4 kinks in one space dimension is also seen in higher dimensions. An-
other area of interest is the effect of the departure from spherically

40
 ,\
'\
/\
,
I \

,,
\ I
\ I \
'.
---------------,/ ,,------------ ----------" ' '.
,--------------------

IV

,\
'\

-'-\. /\
~
______________ ,1 \" ......... __________ ___________ ,1
, \

\ ... ________________ _

Fig.3 Frames from a computer cine film showing a pulson solution of the
spherically symmetric ¢~ equation. The solid curve is u(r,t i ) and the dashed
curve is the energy dens ity H(r, ti ) . Each frame represents a change of 5
units in t, and the horizontal axis range is 0 < r < 15, the vertical axis
-3.5 < u < 3.5 or -875 < H < 875. The pulson is initially centred at r = 8.0.

41
symmetry on the evolution of pulsons. SCOTT [57] has suggested that some
amount of angular momentum in nonspherical initial conditions may offset the
pulson collapse to some extent and ~rolong the pulson lifetime. The present
author is currently investigating the behaviour of nonspherically symmetric
pulsons in the NLKG models, using a two-s~ace dimensional version of the
scheme (3.9) with k = h/12 for stability.
Although these pulson solutions are of interest in Darticle physics [9],
it would be nice to kno~1 if these results have any applications in other areas
where the SG and ~~ equations are physical models. It is to ~e hoped that
one success of this present symposium will be to spotlight areas in solid-
state physics where these computational methods and computational studies can
be usefully applied.
Acknowledgements
I apologise to all the people whose work I have failed to recognise in these
pages. In particular I have said nothing about soliton-like behaviour in
discrete systems, a topic which will be discussed in other papers at this
symposium. Any other ommissions arise from ignorance rather than deliberate
intent. If all those working in this area send me pre?rints I will have no
excuse in future:
The work described here on the RLW and flaxwell-Bloch equations ~Ias joint
work with Drs G. R. ~!cGuire and P. J. Caudrey respectively, and I am most
grateful for many helpful conversations. In addition it is a ~leasure to
acknowledge help and encouragement from many of my colleagues referred to in
this paper. Finally, I would like to acknowledge travel grants from the Royal
Society of London, the National Science Foundation and the Carnegie Trust for
research visits to Italy and the U.S.A. during the last two years, and to
thank the Science Research Council for financial and computational support.
References
l. J. Scott Russell, Report of the 14th tleeting of the British Association,
311 (1845)
2. G. S. Emmerson, in'Scott Russell', London: flurray (1977)
3. A. C. Scott, F. Y. F. Chu, and D. W. flcLaughlin, Proce. IEEE 61 1443(1973)
4. D. J. Korteweg and G. DeVries, Philos. tlag. 39 422 (1895) -
5. E. Fermi, J. Pasta and S. Ulam, Los Alamos Report LA 1940, 143 (1955)
6. N. J. Zabusky, in Proc. Symp. on Non 1i near Parti a1 Differenti a1; W. F. Ames,
ed., Academic Press, Ne\~ York, 223 (1967)
7. N. J. Zabusky and fl. D. Kruskal, Phys. Rev. Lett. 15 240 (1965)
8. J. K. Perring and T. H. R. Skyrme, Nucl. Phys. 31 !So (1962)
9. V. G. Makhankov, Phys. Lett. C35 1 (1978) -
10. R. D. Richtmyer and K. W. Morton, Difference tlethods for Initial Value
Problems, 2nd Ed. Interscience, London (1967)
11. A. R. Mitchell, Computational tlethods in Partial Differential Equations,
J. Wiley, London (1969)
12. A. R. Mitchell and R, Wait, The Finite Element t·lethod in Partial Differ-
ential Equations, J. I-liley, London (1977)
13. W. F. Ames, Numerical Methods for Partial Differential Equations 2nd Ed.
Nelson, London (1977)
14. D. Jacobs, ed., The State of the Art in Numerical Analysis, Academic
Press, London (1977)
15. J. H. Ahlberg and T. Ito, Math. Comp 29 761 (1975)

42
16. A. R. Gourlay, Proc. Roy. Soc. Lond. A 323 219 (1971)
17. M. J. Ab1owitz, M. D. Kruska1 and J. F.ladik, 'Solitary Wave Collisions',
preprint, to be published in SIAM J. App1. Math.
18. A. C. Newell, SIAM J. App1. Math. 33 133 (1977)
19. S. Duckworth, R. K. Bu11ough, P. J~Caudrey and J. D. Gibbon, Phys. Lett.
52A 19 (1976)
20. rK. Bullough, in "Nonlinear Evolution Equations Solvable by the Spect-
ral Transform", ed. F. Calogero, Pitman, London (1978)
21. A. Kudryavtsev, JETP Lett. 22 82 (1975)
22. J. D. Gibbon, N. C. Freeman-and R. S. Johnson, Phys. Lett. 65A 380 (1978)
23. I. S. Greig and J. Ll. Morris, J. Compo Phys., 2064 (1976)-
24. A. C. Viliegenthart, J. Engrg. Math., 51 137 1T971)
25. B. Fornberg and G. B. Whitham, Phil. Trans. Roy. Soc. 289 373 (1978)
26. D. H. Peregrine, J. Fluid Mech. 25 321 (1966) -
27. T. B. Benjamin, J. L. Bona, and ~ J. Mahony, Phil. Trans. Roy. Soc. 272
47 (1972)
28. J. C. Ei1beck and G. R. McGuire, J. Compo Phys. 19 43 (1975)
29. J. C. Eilbeck and G. R. McGuire, J. ComD. Phys. 21 63 (1977)
30. Kh. O. Abdu11oev, I. L. Bogo1ubsky and V. G. Maknankov, Phys. Lett. 56A
427 (1976)
31. F. Ca1agero and A. R. Santarell i, private communication, (1978)
32. J. L. Bona, W. G. Pritchard and L R Scott, University of Essex Fluid
Mechanics Research Institute Report (1978)
33. S. L. McCall and E. L. Hahn, Phys. Rev. Lett. 18 908 (1967)
34. G. L. Lamb Jr., Rev. Mod. Phys. 43 99 (1971) --
35. G. L. Lamb Jr., Physica 66 298 (1973)
36. J. C. Ei1beck, J. D. Gibbon, P. J. Caudrey and R. K. Bu11ough, J. Phys.
A. 6 1337 (1973)
37. J. C. Eilbeck and R. K. Bullough, J. Phys. A. 5 820 (1972)
38. P. J. Caudrey and J. C. Ei1beck, Phys. Lett. 62A 65 (1977)
39. F. D. Tappert and C. N. Judi ce, Phys. Rev. Lett29 1308 (1972)
40. F. D. Tappert, Lect. app1. math. Am. tlath. Soc. 15 215 (1974)
41. I. L. Bogo1ubsky, Compo Phys. Comm. 13 149 (1977)
42. M. J. Ab10witz and J. F. Ladik, Studles in App1. Math. 55 213 (1976)
43.· R. G. Newton, J. Math. Phys. 19 1068 (1978)
44. S. Maxon and J. Viecelli, PhyS:- Rev. Lett. 324(1974)
45. S. Maxon and J. Viecelli, Phys. Fluids. 17 Tb14 (1974)
46. E. Cumberbatch, Phys. Fluids 21 375 (1978)
47. S. A. Akhmanor, A. P. SukhorukovandR. V. Khokh1ov, Soviet Phys. Uspekhi
10 609 (1968)
48. F. P. Mattar, 'Transient Propagation of Optical Beams inActive t1edia',
preprint, Department of Physics, University of Rochester (1978)
49. F. P. Mattar and M. C. Newstein, in "Cooperative Effects in Matter and
Radiation", eds. C. M. Bowden and D. vi. Howgate, Plenum Press, N. Y. (1977)
50. I. L. Bogo1ubsky and V. G. t1akhankov, JEPT Lett. 24 12 (1976)
51. I. L. Bogu1 ubsky, JETP Lett. 24 535 (1976) --
52. I. L. Bogu1ubsky and V. G. Makhankov, JETP Lett. 25 107 (1977)
53. I. L. Bogo1ubsky, Phys. Lett. 61A 205 (1977) --
54. I. L. Bogo1ubsky, V. G. Makhankov, and A. B. Shvachka, Phys. Lett. 63A
225 (1977)
55. G. D. Smith, "Numerical Solution of Partial Differential Equations",
Oxford U. P. (1969)
56. P. J. Roache, "Com[Jutationa1 Fluid Dynamics", Hermosa Publishers,
Albuquerque, New Mexico (1972)
57. A. C. Scott, private communication.

43
Poles of the Toda Lattice

John Gibbon*
Department of Mathematics, U.M.I.S.T., P.O. Box 88
Manchester M60 100, Great Britain

Introducti on
The Toda lattice is a simple nonlinear differential-difference equation which
was derived by TODA [1] as a model describing the propagation of waves along
them. If °
a chain of particles of equal mass with an exponential restoring force between
is the displacement from equilibrium of the nth particle then
the equationnof motion is
On = exp(On_l - On) - exp(On - 0n+l) (1)
Eq. (1) is also a model for a nonlinear electric filter [2]. The great in-
terest in (1) lies in the fact that it is a completely integrable Hamiltonian
system [3,4]. Here it is our intention to firstly show that there is a con-
nection between (1) and other interesting Hamiltonian N-body systems and
secondly to investlgate the equivalent p.d.e. 's obtained in the continuum
limit. In both cases there is a simple connection with the KdV/Boussinesq
equations.
N-Body Sys tems
In 1971 CALOGERO [5] showed that the quantum mechanical problem of a set of N
particles on a line interacting via inverse square potentials is a solvable
one. MOSER then showed [6] that the equivalent classical problem is also
completely integrable. This system can be written in the Hamiltonian formu-
lation
H= Hx.2 + L V( x. - xk ) (2)
j J kfj J
where V(x) = x- 2 for the CALOGERO-MOSER problem. The equivalent equation
of motion is
X. = 2 I (x. - xk ) - 3 ( 3)
J kfj J
It has been shown subsequently by various workers [7,8,9] that several other
N-body problems of type (2) are of great physical interest some of which are
integrable and some not: (i) Sutherland's problem: V(x) = 1/sin 2 x(6) (ii)
V(x) = x- 2 + ax 2 [7] (iii) V(x) =»(x) [8] where Y is the doubly periodic
Weierstrassian elliptic functlon {iv) V(x) = -logisinxi [9]. The first three
are integrable by an extension of the method used for (3). Case (iv) has been
considered by CALOGERO and PERELOMOV [9] although a method for its complete
integrability is not yet known.

Connection with the Toda Lattice and Evolution Eguations

The connection between N-body systems such as (3) and nonlinear evolution

* Address after 1st Oct. 1978: Dept. of Mathematical Physics, University

College, Belfield, Dublin 4, Eire.
44
equations was pointed out by AIRHAULT, MCKEAN and MOSER (AMM) [10]. Consider
the Boussinesq equation
(uxx +6U2+U) xx =U tt =0 (4)
AMM chose u to be in the form of an expansion about poles x.(t) such that
J
u = I(x - Xjr2 (5)
j
The choice of (5) seems a little odd but if we take x· = (2j + l)irr + ct
then (5) is just the Laurent expansion of a squared hyperbolic secant about
its poles and hence (5) in this particular case represents an expansion about
the poles of a soliton. Using (5) in (4) yields two equations of motion
x· = 2 ~ (x. - xk\-3
.• J (6 )
J k j J

x. 2 =1- I(X.-xj2 (7)

J J k k~j
Taken together (6) and (7) show that the velocities and positions of the
poles in the complex plane must satisfy
I (i k + i.)(x k - x.)-3 = 0 (8)
k~j J J
Eq. (6) is exactly the inverse square potential problem except that this pro-
blem is in the complex plane whilst the N-body problems previously mentioned
are on the real axis.
Where does the Toda Lattice fit into this set up? It is different to the
Hamiltonian system (2) in the sense that the equation of motion only contains
nearest neighbour interactions while (2) contains all possible pair interac-
tions. It is possible to show that, similar to the Boussinesq equation the
Toda lattice can be transformed into a complex pair-potential Hamiltonian
system. A more convenient form of (1) can be obtained by taking a new varia-
ble Qn = Sn- 1 - Sn yielding an equation of motion
1 + Sn = exp(S n+ 1 + Sn- 1 - 2S n ) (9)

A travelling wave solution of (9) is (equivalent to the one soliton solution)

S = log cosh ~(an - wt) w = ±2sinh(~a) (10)
n
If we appeal to the usual product expansion for cosh ~z then this indicates
that the equivalent expansion for Sn in analogy to (5) is
Sn = ~log(n - Xj) (11)
J
Using (11) in (9) yields (for an infinite lattice only) the equations of
motion
X. = 2I(x. - x r 3 Il [1 - (x J' - \t2] (12)
J £ J £ k~j~£

i .2 = Il [1 - (x. - x )-2] (13 )

J k~j J k
The combination of the two shows that in analogy to (8)
k~j(Xj + \)(x j - \ r 3 [1 - (Xj - \)-2] = 0 (14)

The system can be more tidily expressed by the Hamiltonian

45
 H = ~p.2 _ n [1 - (X.-X )-2] (15)
J J k~j J k
with canonical variables x· and x'. The Hamiltonian (15) and (14) are
obviously generalisations of the redults that AMM obtained for the Boussinesq
equation since the first term in the product expansion in (15) yields the
inverse square potential.
We have yet to demonstrate complete integrability for this system on the
real axis as MOSER has done for the inverse square potential. However since
it is a generalisation of the inverse square case it may prove in the future
to be tractable.
Lack of space precludes us from giving a fuller analysis of the motion but
some information about the motion can be obtained very quickly. Briefly if
we set q = n- 1 and recursively differentiate w.r.t. q and set q = 0 at
each step, we obtain a set of relations the first of which is Ix. = O. A
combination of the rest gives the result J
\ m_ m p _
LX. - I a t (m - 1,2, ........ ) (16 )
j J p=O mp
where the a's are constants. The first result simply means that we have a
closed system whose centre of mass is stationary and is confirmed by summing(12)
Equation (16) tells us that asymptotically the x. approach a limit which is
no faster than Xj - cjt. Two examples of the motion can be considered. First-
ly, if we choose the x. to be the poles of a soliton: x. = ct ± (2j + 1 )irr,
then the equation of motion and (14) are satisfied as we ~ou1d expect. Second-
ly, to show that not all motions are linear in time, a two particle solution
is 1
xl = a + [(t - 13)2 + U~ ; X2 = a - [( t - 13) 2 + !]2 (17 )
]
This means that xl + x2 = 0 and so (14) is satisfied.
The Continuous System
In order to go into the continuum we shall use the shift operator [1]
exp(±a/an)f(n) = fIn ± 1) and consider S to be a function of a continuous
variable. Further, it is convenient to in~roduce a small scaling parameter
E and a new variable x such that x = E~n. Equation (9) now turns out to be
1 + S = exp{[2COSh(E~dax) - l]S} (18)
Expanding (18) in powers of E and introducing a new variable u = S , we
find that xx
j+l
Utt = 2.I E
(2J+2)~
A
xx
(j)
(19 )
J=O
A(O) = u

A(2) u4x + 30uu 2x + 60u 3 unx = d~U

To order E2 therefore, the Toda 15ttice is ~Q~iva1ent to the Boussinesq
equation(wlth contributions from A() and A\r) . The contributions from the
other A J) (j>2) are higher nonlinear dispersive corrections to the Rous-
sinesq equation. It is these higher terms which contribute to the higher
power terms in the product expansion in the potential in the Hamiltonian (15).
A set of higher KdV equations can be obtained if we ignore leftward travelling
waves: . 1
EJ + (j )
ut + 2j~0 (2J + 2): Ax =0 (20)
46
After the KdV equation itself (A(l)) the next operator A(2) is an evo1u-
tionequation on its own (inconvenient numbers have been scaled out)
u + A (2) = 0 (21 )
t x
has been shown to be solvable by an inverse scattering operator and(~,ck1und
transformations and N-so1iton solutions exist [11,12]. The set of A form
a ladder of KdV equations which are different to those found by LAX. In [12]
it was shown that the scattering operator is of the third order:
L = a3 + 6ua B = 9a 5 + 90ua 3 + 90u xa2 + 60(u 2x + 3u 2 )a (22)
such that L' = [B,L] yields the evol\ltion equations (21). Simple recursion
relations also exist between the A\J) :
aA(j)_t2j+2)! 2 (j-n-1) (23)
au 2nx - 2J - 2n): • (2n + 2): A
which enables the complete ladder to be constructed [11]. Quite a lot of
other information can be found which we do not have room for here. However,
in conclusion, it is interesting that the Toda lattice generates at higher
orders of perturbation, integrable nonlinear evolution equations such as (21).
It seems to be these which account for the simplicity of its pole structure.
This is certainly not always the case for discrete integrable equations. For
instance, KAC and VAN MOERBEKE's equation [6]
un = He xp un+1 - e xp un- 1) (24)
.
is integrable in the same way as the Toda lattice and to first order in the
continuum yields the KdV equation. However no "sensible" or solvable equa-
tions are generated at higher orders of perturbation nor does it have a simple
pole structure.
References
1. M. Toda: Prog. Theoret. Phys. Supp1. ~, 174, 1970
2. R. Hirota and K. Suzuki: J. Phys. Soc. Japan 28, 1366, 1970
3. M. Henon: Phys. Rev. B: ~, 1921, 1974
4. H. F1aschka: Phys. Rev. B: ~, 1924, 1974
5. F. Calogero: J. Math. Phys. ~, 419, 1971
6. J. Moser: Advances in Mathematics ~, 197, 1975
7. M. Adler: Comm. Math. Physics ~, 195, 1977
8. F. Calogero: Lett. Nuovo Cimento 11,411, 1975
9. F. Calogero and A. Pere1omov: Comm. Math. Phys. ~, 109, 1978
10. H. Airhau1t, H. McKean and J. Moser: Comm. Pure App1. Math. 30, 95, 1977
11. P. Caudrey, R. Dodd and J. Gibbon: Proc. Royal Soc. Lond. A~, 407, 1976
12. R.K. Dodd and J.D. Gibbon: Proc. Royal Soc. Lond. A 358, 287, 1977

47
Perturbation Theory of the Double Sine-Gordon Equation

P.W. Kitchenside, A.L. Mason, R.K. Bullough, and P.J. Caudrey

Department of Mathematics, U.M.I.S.T., P.O. Box 88
Manchester M60 lQD, Great Britain

A general perturbation theory has recently been developed by KAUP and NEWELL
[lJ for dealing with almost completely integrable nonlinear partial differ-
ential equations, for which conventional inverse scattering transform (1ST)
techniques are inapplicable. One such equation is the double sine-Gordon
equation (DSGE).
A. V
+ 2 Sln 2u.
. (1)
uxx - U tt = Slnu

Eq. 1 arises in nonlinear optics in the theory of resonant pulse propagation

in degenerate media (DUCKWORTH [2J). A solution of particular interest here,
with boundary conditions u ~ 2~ when Ixl ~ 00, is given by

u = 4tan-l(~ COSh~ 14-n x) (2)

in its rest frame. This represents an unstable 2~-kink-antikink bound pair.

In the optical context such a solution is known as a 'O~' pulse.

For initial data corresponding to a compression of the spacing of the

kink and antikink, it oscillates (Fig. 1) much as the breather solution of
the sine-Gordon equation (A = 0): for data corresponding to an extension
it unbinds as shown in Fig. 2.

Fig.l A numerical integration

~) for the oscillatory
regime

48
 Fig.2 The break-up regime

The SGE is completely solvable by the 1ST and it is natural to regard the
DSGE as a perturbation, for small A. A convenient starting point is a two-
kink solution of the SGE, which may be written using standard notation, as

2 i2 (k l +k'2)X
1;1-1;2 Yl Y2 e
1 -

i ~i- 4~J
-1 1;1+1;2 21;1 21;2
u 4tan , ki = (3)
i2klK i2k 2X
Yl e Y2 e
+ --
21;1 21;2

in terms of the scattering data. Here Yl and Y2 are the residues of the
reflection coefficient bfa, (when this quantity may be analytically con-
tinued into the upper half plane), evaluated at the complex eigenvalues 1;1
and 1;2 respectively. For the SGE, the time evolution of ~he scattering
data is trivial. For (1), however, the appropriate equations are [1J

sin ~ 2 ~ 2J
f 2 • [¢l + 'f2 1;.
dX, (i=1,2) (4.1)
~

y.a.
~ ~
1;. - f oo sin ~
2 °azd [¢l2 + ¢2 2 J 1;. dX (4.2)
a. ~t 8a.2 ~
~ ~

dX (4.3)

b.
~

a.
~

where a. denotes differentiation of a with respect to 1; evaluated at

1; = 1; .. ~The expression (¢jZ + ¢zz) denotes the squared eigenfunction of
~ 1;

49
the SGE IST evaluated at S. Eqs. (4) are simplified considerably if we
discard the continuum components [and therefore (4.3)] on the assumption
that they are negligible. This seems intuitively plausible if we start
with the purely discrete (3) and A is taken to be sufficiently small.
U~ing.the breather parametrization sl = -sZ* and Yl = -Y Z* the identi-
fl.catl.ons

(5)

ensure that (3) is symmetrical about the orl.gl.n in the rest frame. Putting
~ = cos-1/i/4 and e = ~-~ = 0 in (3) give the required initial solution
(Z). The subsequent time evolution may not be determined from (4) neglect-
ing the continuum. This has been done by MASON [3], and provided the
breather solution remains valid the resulting equations are
-1
A tanQ J(a)+a a tanh a
Ct
2 l+a
a = cos2etan2~, J
~ ~

Q
t
= sin 2Q - (~ - c) cos 2Q (6)

Q = tan l(tanecos~), C = cos2~.

Eqs. (6) show that for initial conditions e =~ 0 < ~ <cos-1/A/4 the
resulting motion is oscillatory: for ~ > cos-1/A/4, ~ eventually reaches
TI/Z [3]. At this point, the solution becomes a rapidly separating kink and
antikink. The breather form is now inappropriate and the analysis has to be
repeated using kink parameters. A suitable parametrization is now
i e -q YIp
r
sl -
-
"2 = cothq e
ZSl

the equations of motion being

P 2 _ C2 _ ~ p cothp sinhq, C (7)

4 '
A P J(a)+a
Ct ="4 ~

We have numerically integrated (6) and (7) for the initial conditions
1 -1
e = 0, = "4 A (1 - E)

with E = 0.1 or -0.1 and A = 1.0. When E = -0.1 the motion is oscil-
latory as shown in Fig. 3. For E = 0.1 break-up occurs (Fig. 4), re-
quiring a change of parametrization when C becomes negative. A comparison
of Fig. 4 with Fig. Z shows good agreement for break-up in this no-continuum
approximation. However, eventually truncation errors cause the parametriza-
tion to break down due to the asymptotic properties of the hyperbolic func-
tions in (7). For the oscillatory regime, agreement is almost perfect until

50
u first reaches its maximum amplitude. Subsequently, a slight periodic
emission of ripples may be discerned representing energy being pumped into
the continuum. The net result is to shorten the period of the breather-like
oscillation.

Fig.3 The oscillatory regime Fig.4 Break-up described by (6) and

described by (6) (7)

We are currently investigating this discrepancy by trying to account for

the growth of the continuum. The procedure is however extremely complicated
and seems unlikely to yield a set of equations with the relative simplicity
of (6) or (7). Despite this anomaly there seems to be good agreement even
for A = 1. O.

References

1. D.J. Kaup, A.C. Newell: "Solitons as Particles, Oscillators and in

Slowly-Varying Media: A Singular Perturbation Theory", Proc. Roy. Soc.
(London, 1978) (to be published)
2. S. Duckworth: Ph.D. Thesis, University of Manchester (1976)
3. A.L. Mason: In Proc. of the NATO Advanced Study Inst., "Nonlinear
Equations in Physics and Mathematics", ed. by A.D. Barut (Istanbul,
August 1977) (to be published)

51
Soliton Perturbations and Nonlinear Focussing

Alan C. Newell
Department of Mathematics, Clarkson College of Technology
Potsdam, NY 13676, USA

Abstract
A method for analyzing equations which are close to soliton equations is
given and several physical examples are discussed in detail. The possible
connection between certain focussing properties on nonlinear waves and
turbulent bursts observed in many fluid dynamical situations is discussed.
O. Preamble
The first aim of these lectures is to survey the method by which one can
analyze equations which are perturbations of soliton equations. The
method is simply to apply the same transformation which carries the coor-
dinates of the unperturbed equation to action-angle variables and then to
compute the slow rate of change of the action variables and angular phases
due to the cumulative effects of the perturbation. The approach we
describe is completely equivalent to the apparently more straightforward
but in fact more difficult method of perturbing the equation directly.
The second aim is to use the method to examine the effects of applying
typical types of perturbations to the canonical and ubiquitous Korteweg
deVries (KdV), nonlinear Schrodinger (NLS) and sine-Gordon (SG) equations.
In each case an attempt is made to show that these results can also be
found by the judicious use of certain shortcuts. The particular problems
we study are (a) the reversal in polarity due to depth changes of an
internal solitary wave travelling on a thermocline, (b) the synchronous
response of an envelope solitary wave to external forcing and the ram-
ifications of the result to one-dimensional condensates, and (c) the
tunnelling of envelope pulses through barriers created by regions of
increased density. Along the way, several important points concerning
the nature of NLS and SG solitons, which do not appear to be generally
appreciated are brought out. Most of these results were obtained in
collaboration with DAVID KAUP and are presented in more detail in ref-
erences [1-4]. The final aim of this survey is to introduce some new
ideas in connection with the behavior of waves which are created by the
onset of an instability. The tendency of such a system to reach an
ordered, cooperative state (the presence of a most unstable wave inhibits
the growth of neighboring waves) on the one hand is counterbalanced on
the other by the tendency of the modulational instability to spread the
energy in wavenumber space and therefore cause focussing to occur in real
space. A criterion which expresses this balance is given. It is sug-
gested that when dispersive effects dominate and when the dimension of
the system is more than one, the resulting focussing may be the mech-
anism for producing the intense irregular bursts often observed in these

52
situations.
1. Introduction and General Discussion
The discovery of the soliton [5,6,7) has had a profound impact in the
mathematical sciences. Quite simply it is a new type of normal mode of
a special class of nonlinear, infinite dimensional, mechanical systems,
systems often described by partial differential equations of evolution.
The special property of these equations is that each is integrable;
namely, one can find a canonical transformation, the inverse scattering
transform or 1ST which carries the coordinates in which the equation is
originally written to new coordinates which are action-angle variables.
The 1ST is the nonlinear analogue of the Fourier transform. Consider the
KdV equation,
( 1.1)

and associate with a real q(x,t) the eigenvalue problem

vxx + (s2+q(x,t))v=0, -oo<x<oo. (1.2)
Define for real S, the solutions ¢(x,t;s) and ~(x,t;s) with the following
asymptotic properties:

¢ ~

x= _00 e- isx a(s,t)e1Sx_b*(s,t)e-isx (1. 3)

x= +00 a(s,t)e-isx+b(s,t)eisx e isx

We call a(s,t) (a(-s,t)=a * (s,t)) and b(s,t) (b(-s,t)=b * (s,t)) the scat-
tering functions. In the linear limit a is 1 and
i
sb(s,t) = f~oo q(y)e- 2iSY ds is simply the ordinary Fourier transform of
q(x,t). In the physicist's terminology, a- l is the transmission coefficient
and ba- l the reflection coefficient. If q(x,t) decays sufficiently
rapidly as X"": a(s,t) may be extended to Ims>O and its zeros there,
00 ,

sk = ink' k=l, ... ,N, are the discrete eigenvalues of (1.2) and at s=sk'
-n x *
¢k->-bke, k ~k+-bkE!nkx as x:oo respectively. The 1ST is a mapping from
q(x,t) to either of the two sets.
b bk . 2nkxk N
S+{a' s real; {sk = ink' Yk = ao- = 21 nke }l} or
k
*
SJ~ ,sreal;{sk (1 .4)

called the scattering data, and from a knowledge of S one can reconstruct
* _. _. - 2n l xl ,
q(x,t). In particular if ba :=0, N=l, s-l nl ' 8 1 --21 nl e then
q(x,t) = 2n~ sech2nl(x-xl) one of the reflectionless potentials of (1.2)

53
and the soliton of (1.1).
As q(x,t) evolves in time so does S, usually in a most complicated way,
*
in which the time behavior of ~ (~,t) at one s=~ depends on this quantity
at all other values of s. Theasystem remains strongly coupled. However
for a very special, and fortunately for us, a very useful class of evo-
lutions of q(x,t), the time rate of change of S is simple and suitable
functions of the quantities [S,9,10]
* *
I~ I, Ilk' k=l, ... ,N, Arg(~), xk k=l, ... ,N are the action and angle
variables respectively. For (1.1),
* *
I~I = 0, (Arg ~) = - SO1S3 , (1 • 5a )
a t a t
- 2
Ilkt = 0, \ t = 41lk· (1 • 5b)

It is very natural, then, to analyze the effects of perturbations to

(1.1),

(1.6)

by writing down expressions for the corresponding change in the action-

angle variables. We find [1] that

Ilkt = - -2iE:-
Ilk
r _00
F YkljJk2dx (1.7a)

a "
13 kt +- (~+
ak L)S k r;kt
r;k
(1 • 7b)

b*
a t a
. 3 b*
(-) = -Sl s - + - -
2i r;a2
E
r
_00
2
F ljJ dx (1.7c)

and
at -- 2ir;
E r _00
F¢ljJdx. (1.7d)

The program at this stage is to solve (1.7) iteratively and obtain

asymptotic expansions in appropriate powers of E for the quantities nk ,
13 k, ~*. In general these expansions will be nonuniform in time but
by choosing the zeroth order approximations to be slowly varying functions
of time the nonuniformities can be removed. A long time t~-a, some a>O,
description of the system is thereby obtained.
Whereas in principle the method holds very generally, in practice it
only works (at this time) when the zeroth order approximation is close
to a multi sol iton state. The reason for this is that althouqh the squared
eigenfunctions can be calculated in principle, their form does not
facilitate the calculations in (1.7) unless the continuous
spectrum is ignored to leading order. However the cont1nuous spectrum
still can be excited by the perturbation even though the initial state

54
consists only of solitons. In certain of these cases, its presence can be
handled and in fact we show that without it our understanding of the
problem would be quite incomplete. The energy is transferred to the
continuous spectrum by resonance. If the interaction term
fF~2dx in (1.7c) contains a time dependence e ia ~)t, then it is possible
*
that a(s) matches the natural frequency _8i~3 of ~either (a) for no ~,
(b) for a particular s=t say or (c) for a continuous range (perhaps all) of
s. If (a), then the continuous spectrum is not excited to any appreciable
*
extent. If (b), then ~ developes a Dirac delta function behavior about
A a
s=s in long time and results in an order E nondecaying contribution from
the continuous spectrum to q(x,t). If (c), then the continuous spectrum
will grow to order one on the liE time scale and should therefore have
been incorporated in the leading order approximation, a task, as we have
mentioned, at present beyond the practical application of the theory.
It is natural to consider the alternative approach of perturbing (1.6)
directly; if q=qo + dq + .•• ,
qOt + (3q~+qoxx)x = 0 (1.8a)

(1. 8b)

The disadvantage of this approach is that the problem of finding the

appropriate basis in which to resolve oq and therefore separate the left
hand side of (1.8b) is as difficult as (and in fact directly connected
with) the problem of finding the appropriate eigenvalue problem to solve
(l.8a) in the first place. From the scattering theory we know that [see
ref. [10]] the appropriate basis for oq is
~ ~ 'i¢2 ... .
E = {ax ,I:; real; (ax 'axal:; )k} WhlCh lS adJolnt to the set
G= {~2, I:; real; (~2, ~ )k. Now the squared eigenfunction function
~2 satisfies,
12 12222
-;f( ~ )xxx - 2" qx~ - q (~ ) x = I:; (~ )x (1 .9)

and if qO(x,t) satisfies (1.8a),

2 . 3 2 2 2 (1 .10)
(~O)t = ( 2q OX- 8l l:; )~O + (41:; - 2q O) (~O)x·

Multiply (1.8b) by ~~, use (1.9,10) and integrate to find,

ar'"
at J_oooq~odx .3 f oo_oooq~odx
2 + 811:; 2 = J_
r'" F(qo) 2 (1.11)
oo ~Odx.

But rOq~02dX is simply the infinitesimal change o(L)

_00
*
a in the reflection
coefficient times 2il:;a 2 (see (1.7c); there the infinitesimal change is
a~) and thus (1.11) is precisely what one obtains at 0(£) from (1.7c) by

55
 * * *
setting ba = (ba )0+ e:c5 (ba ). Therefore the two methods are precisely
equivalent and in fact if one employs the second and direct method. one is
led back to (1.7). A Green's function method suggested by KEENER AND
Me LAUGHLIN [11] is very closely related to these ideas.
At this stage we also state two other useful facts. First even though
the set G is not a basis (for L2; -f-2
does not tend to zero as x+_ oo ). we
can still expand the special function q(x.t) as
200 b2 N 2 ()
q(x.t) = in Lao I:;, a:W dl:;,-\~lYk1,;kWk . l.12

Second we emphasize that the soliton shape assumes the form of the square
t
of a hyperbolic secant i.e •• Wk = e-nkxk sechnk(x-xk) only when it is
separate from all the other normal modes. We will find that even the
simplest of perturbations excites the 1:;,=0 wavenumber of the continuous
spectrum and this contribution attaches itself to the soliton which as a
result is distorted from the hyperbolic secant shape (see section 2).
Before we go on to discuss the results of perturbations we list and
comment on some of the features of the soliton solutions for the other
equations we shall study. The nonlinear Schredinger (NLS) equation
"
qt-1qxx- 2"lq 2q* = 0 (1.13a)
has the one soliton solution
q(x.t) = 2n sech 2n(x-x) exp(-2il:;,x-4i(~-n2)t+i~). xt =-41:;,. (l.13b)
The single kink (2n pulse) solution of the sine-Gordon equation
2
~T-c uxx = -u xt = -n
2"
smUt x = X+T
2 ' t = X-T2 (1.14a)

is 2
u(X,T)=4 tan- l expny(X-VT). y = (1 _ V2)-1/2 • (1.14b)
c
and the breather (On pulse) solution is

u(X.T) = 4tan- l 2ric sech2ny(X-VT) COSRy(T-V~) (1.14c)

with frequency (A .: 2riC) c
2 2 2 2 2 n2
n = n -4n c or n =:-::2". (l.14d)
l+A
The energies of (1.14b) and (1.14c) are Mc 2• M=8y and 2Mc 2(1 n2)1/2 '
-:r
n
respectively. We note that in the small amplitude limit. the solution to
(1.14a) can be written as the product of a fast sinusoidal oscillation and

56
an envelope function which satisfies the NLS equation (1.13a). There is a
close connection between the solutions of (1.13a) and the small amplitude,
almost monochromatic, solutions of equations like (1.14a) which in the
weak nonlinear limit have a soft spring potential.
Let me make two very important points. Note first that the natural
frequency of the breather is less than n (the "plasma" or "pinning"
frequency). This distinguishes these normal modes from those associated
with the continuous spectrum whose frequencies lie above n (cf. the
linear limit w2 = n2+c 2k2). Now there are travelling wave solutions of
(1.14a) whose natural frequencies are reduced below n by the n9nlinear
modification. For example, in the small amplitude limit, u=Ae 1 (kX-wT)+(*)
where w2=n2(1_~*)+c2k2, is a solution and has a frequency, for sufficient-
ly large A and small k, below n. The minus sign in front of AA* comes from
the fact that the force sinu derives from a soft spring potential. However
this solution corresponds to the x independent solution of (1.13a) and is
unstable in the BENJAMIN-FEIR [12] sense. Thus below the plasma frequency
the only stable modes are the breathers, just as the only stable modes of
positive frequency of (1.13a) are the solitons (1.13b). - -
The second point concerns the nature of the soliton (1.13b). Notice
that whereas its frequency is modified by amplitude, the velocity of its
envelope phase is not, a fact which does not faithfully reflect the nature
of the equations such as (1.14a) from which it derives. In fact the
soliton (1.14b) of the NLS equation moves with the linear group velocity
k-2E~ of the underlying carrier wave. This weakness can be corrected by
including a term iEoqxt, usually neglected, on the R.H.S. of (1.13a) which
leads to xt = -4~+4En20 + O(E~2). While this inclusion may not be
important in many situations in which the NLS equation is used as a model,
it is very important when we examine the tunnelling of soliton pulses.
2. A List of Results and Their Applications
lao The effect of damping on the KdV equation. Consider
qt+6qq/qxxx = -r(t)q, r(t)« 1 , (2.1)
where r(t)=o, q(x,t) = 2n6 sech2no(x-x), xt =4n 2 , t<tO' Let q(x,t) be
its subsequent evolution and let its scattering data be (1.4). Using
(1. 7) we fi nd
2 - 2 r
nt = -1' rn, xt = 4n + 3n (2.2)
from which n and x can be found. Using (1.7c) we compute the excitation
of the continuous spectrum and find that ~* behaves like a Dirac delta
function at ~=O and gives a nonvanishing O(r) contribution qc(x,t) to
q(x,t). Using formula (1.12), _ 2
r 2 _ { sin2~(x-x)-sin2~(x-xO+4~ t) ) 1
qc(x,t)= 61Tn tanh n(x-x) roo
-----..2=~--....::--- d(2~»+O(.f)'
)
~ - ~n ' xO<x<x and zero otherwise. (2.3)

57
Unable to balance both the total mass and energy flux relations
d foo qdx = -r
at r'" d
at r'" 2 r'" 2 .
-00 1_ 00 qdx and 1_ 00 q dx = -2r 1_00 q dx, the sollton changes

n in such a way to satisfy the latter and develops a shelf in order to

compensate for its i nabil ity to sati s fy the former. Note that d~ rooqdx =

d~ r: qsdx + d~ L: qc dx = -%r(4n) - t r(4n) = -r(4n), the exact result. Thus

the changing soliton accounts for t of the rate of mass loss (r>O) while
the shelf supplies the other t. When the soliton is at x=x(t)at time t,
the shelf q (x, t) = -
c fn exp(ftt rdt)
which may be written as a function of
_
2
x through x=x(t), xt =4n the soliton position. The total area Ac(t) in the
shelf at time t is 4nO exp(-f~ rdt) - 4nO exp(-t f~ rdt) and in the
o- 0
soliton is As(t) = 4nO exp (-t f~ordt), which agrees with the exact result

A(t) = A(t O) exp (-f~ rdt). We can also compute the distortion of the
o
soliton,
qs(x,t) = -4YlI;;11J!12 = 2n 2sech 2n(x-x)
- ( 1- -2
r (x-xO
- ) tanhn (-))
x-x, (2.4)
6n
which means that the soliton is steeper in front if r>o. Hence it travels
faster. We may also verify (2.2) by looking at the exact relation for
the motion of the center of gravity d~ roo xqdx = 3roo q2dX -r f~ooxqdx.
Using (2.4) and the expression qc for the shelf height 2 r
_ we find xt = 4n + 3n •

*
Defining y = fxq/ fq we find for small t-t o ' Y'"x - r_x_ + O(x 2) whence
- 12 2
- 2 d- 2 n 3E 3E O
Yt~4n which agrees with the exact result = fq / fq = M = M exp
o
t
(- f to rdt)~ 4n
2
for t~tO' Then there is no sudden change in velocity of
the total center of mass. In order to examine what happens on the time
sca 1es on whi ch the perturbati on theory breaks down 2n 2 = 0 ( I), we must
n
take account of the change in the nature of a(l;;,t) which developes an
irregular singular point at s=O, a(l;;,t) = I;;+-~n exp 23 i frndt.
I;; In I;;
Note that by a judi ci ous use of the exact "conservati on" 1aws we woul d
have been able to derive all the results (2.2) and (2.3). Could we have
simply used these shortcuts initially? As it is with many expedient
devices, the answer is yes (although we would have to calculate the soliton
distortion by direct methods) with the proviso that one understands the
nature of the problem. Certainly the range and role of the shelf was a

58
major controversy unti 1 resolved in [1] by the "exact" perturba ti on theory.
However once one understands the reason for the shelf we can use the idea
directly. For example if we wish to perturb the s~liton solution q =
2nsech2n(x-x), for the modified KdV equation qt+6q qx+qxxx = -rq, then

d~ Jq2dX = -rJq2dx implies nt = -2rn, the shelf height when initially

. - 1Ir
created 1S Xt ' -xt 'V= 4n,
2 ( and then at roo qdx ~ -r1l ) an d at tlme
d 1_
00
.

t is -11~ expJ~ rdt, t=t(x). For the next correction to xt (it turns out
4n t
there is none [1]) we would need to calculate the soliton distortion.
lb. Application to Internal Waves. This work is being done in collab-
oration with COLLEN KNICKERBOCKER and more detailed results will be
published shortly. We consider a lake model consisting of two layers of
water with slightly different (approx. 6%) densities. Initially the bottom
layer is fairly deep and thus long disturbances will travel along facing
into the deeper water. However as the pulse enters the shallow portion of
the lake, it reaches a point at which the deeper layer is the one on top.
This situation is modelled by the equation ux+6f(x)uut+Uttt = O. The
coefficient f(x) is assumed to be a slowly varying function of position
1 +1
and changes sign at x=xl (e.g. f(x) = 0-1, xl = £}. If we set u=T q,
-f
then we obtain (2.1) with x and t interchanged with r(x) = ~ , and we may
apply the previous results. If f= 0-1, then from (2.2) if nO=l, n=(_f)2/3,
qmax = 2(_f)4/3 and umax =_2(_f)1/3. The area Jqsdt under qs is 4(_f)2/3

and under us' Jusdt, is_4(_f)-1/3 and under the shelf Jqcdt is 4(_f)_4(_f)2/3
or under uc ' JUcdt is-4+4(-f)-1/3. As a function of x, uc(x,t) =-} (l-ft<x-xo)~/7 ,
xO<x<x a fact confirmed by numerical experiment. Furthermore, the shelf
faces into the top layer. The theory breaks down when 2n 2 = O(r/n) or
(-f) = (f)1/3 at which point Us = O(€1/9). However the shelf is small and
slowly varying in x and thus propagates from Xl (l_(f)l/3) to Xl (l+(f)l/3)
without significant distortion. At this point, it is in a region where it
is a potential well and it has a positive area equal to that of the
original solito~Hence it creates solitons which face into the top and
now deeper layer. The net result is that the original soliton appears to
have changed its polarity at the turning point with (a) a change in phase,
(b) slightly less amplitude and (c) trailing oscillations. Our numerical
calculations agree closely with these predictions. We expect a similar
qualitative behavior to obtain in many parallel situations.
2. The Synchronous Response of a NLS Soliton to the Combined Effects of
Damping and Forcing.
Cons i der

59
 (2.5)
for small f, E and suppose
q ~ 2n sech2n(x_x)e-2i(a+rr/4) (2.6)
where a = I;x+o. The detailed rates of change of the action n, I; and angle
i
variables X, a are written in [1]. Here we will assume the solution
retains the form (2.6) and present a shortcut to the main result. From the
a
exact law ~t foo (q *q -qq * )= -2f f 00
(q *qx-qq *x)dx we find straight away that
o _00 x x _00

~I;n) = -2f(l;n) which implies that, depending on the initial ratio ~ ,

either I; or n will tend to zero. If the former occurs, the forcing field
loses the soliton. Let us assume then that (~)O is such that I;~ and
formally set 1;=0. Next use the exact relation ata foo qq *dx= -2f foo
_00 _00

* . t
qq dx-[Ee'wO L: q*dx+(*)], from which we find nt =-2fn+¥-Sinx, x=wOt+20.
Now to 1eadi ng order 0t = -2n 2 and th us Xt = wo-4n2. The poi nts n = ~ ,
2f.TwQ
X = 2nn+Xo' sinXo = ~ , are stable nodes or spirals. Thus a
synchronized sol iton q = rwo sech.TwQ(x-x)e iwOt- i Xo- in/ 2 is achieved
2 2
provided the forcing frequency lies in the window O<wO< n ~ . Note that
4f
wO>O. For WO<O, only the continuous spectrum is excited. Conversely if
wO>O is given, and the forcing amplitude slowly increased, the system does
not respond until E > ~ and then once formed the soliton amplitude is
independent of the forcing amplitude.
2b. A lications to one-dimensional condensates. In a recent letter [13],
RICE, BISHOP, KRUMHANSEL AND TRULLINGER RBKT suggested that the solitons
(1.14b) of (1.14a) may result in the condensate having a large d.c.
conductivity at low temperatures. As a natural extension of these ideas,
KAUP and NEWELL [14] have suggested that the breather solutions (1.14c)
may contribute to the a.c. conductivity of the condensate. In the RBKT
model, a breather would correspond to an oscillating electric dipole. It
has the advantage over the kink that it needs no minimum activation energy.
We expect these modes to be excited by an a.c. source with frequency below
n, the pinning frequency of the condensate. Consider the RBKT model and
include the effects of forcing and damping; here u(x,t) is the phase of
the order parameter and satisfies
2 2 dV
Utt-C uxx + n du = - fUt - 2n£cosrlt (2.7)

60
where E = !~~,e is electric charge, M= Bnsm*n/cq6 and ns ' m*, qO=2kF are
the density of condensed conduction electrons, the condensed electron
effective mass and the fundamental periodicity of the undeformed condensate;
r! = n-o wand V is a soft spring periodic potential. Now look for solutions

u= t ~e-int+(*) and find (here we take V = l-cosu), to a first approximation,

i c2 i 2* 1 . i 0 wt ( )
~t - 2il ~xx - 16 n~ ~ = - '2 r~ - lEe 2.B
to which we apply our earlier results. First we predict a significant
absorption peak should occur for values of the external frequency just
below the pinning frequency n of the condensate. Second we suggest that
2
the width of the window !:::.w = ;2 (Y) 2n may provide a means of measuring
the phenomenological damping. Using representative values for TTF-TCNQ,
We f ln n
· d!:::.w ~ (10- 4 T
En ) 2' wh·lC h f or -E = 10 3 V/cm and r~lO -1 n would be order

one. Third we find the average power absorbed from a single frequency
<p> = -<E foo j(x,t) coslltdx>, j=n s eqo-1U t , is 2rMc 2 (20w
_00 n )1/2, independent
of the applied field. Finally, we calculate the power absorbed over a
range of frequencies and using a breather distribution nb = ~ (kT2)
c Mc
suggested by CURRIE [15] find the total power absorbed fnb<p>d(ow) is
n -5- 3
nskT(!:::.w E ) (5.9xlO E) where T is the temperature, k Boltzmann's constant
and !:::.wE»!:::.w is the bandwidth of the applied field, a distinctly non-ohmic
response.
3. NLS Sol itons in Dens ity Gradients.
3a. Consider
.
qt -lq 2. 2 *
xx - lq q F, F«l. (2.9)
There are two cases of interest. First if F = -rq we already have given
the result. Second if F = -2iaxq-2iax 2q, write
X = x+a(t), T=t, q(x,t) = Q(X,T)eixE+i(aE+F)
where a'+2E = 0, (F+aE)' = _E2+2aa 2 , E' = -2a+4aa and find QT- iQ _2iQ2Q* =
-2iaX2Q. Then if Q(X,T) ~ 2~sech 2~(X-X) e-2i~X-4i fe2dT+4in2T+~~
and &
A A

is small, the appl ication of perturbation theory shows that nT=O, XT= -4~ ,
t;T=2aX. If a=O, the soliton is trapped if &>0 and repelled if &<0.

61
If &:;0, then E=-2at, a=2at 2 and the path of the soliton is described
by x'" -2at2; namely if a>O (the case if the pl asma dens ity increases as
x increases), all right going solution components decelerate, turn around
and travel in the direction of the decreasing density gradient.
3b. Nonlinear Tunnelling. Consider the model provided by (1.14a) in the
case where the parameter n, the plasma frequency (or the pinning frequency
in a condensate) is a slowly varying function of X which increases from
n6 at some X=XO to w2 at X=O. Near X=O n2 = w2(1+20£2 X). Suppose a
finite amplitude signal of frequency w is initiated at some X>XO.
Because of the modulational (Benjamin-Feir) instability a signal of long
but finite extent will break into a sequence of breather pulses (which for
sufficiently small amplitudes are solitons of an appropriate NLS equation).
Let us suppose that the frequency of the breather as seen by the observer
V2 _ n2
at rest is w. Then yn=w and from (1. 14d), ~ - 1 - 2 2. Now let
c w (l+A )
us assume that this relation will continue to hold if n2(X) is slowly
varying. If near X=O, n2 = w2(1+2o£2 X), then we might argue that the
soliton will penetrate to a point X given by 2o£2X = A2. Furthermore if
p P
n2 were to decrease before x=x p' say n2 = w2(1+2o£2X-ol(£2X)2), and
£2 X ~ , then we might suggest that the soliton penetrates the barrier
p 1
in a lossless manner. We verified these conjectures [3] in a more precise
way by examining the propagation of a NLS soliton in a situation where
n2 (X) is slowly varying. By setting u(X,T) = ~k £B(T,y)exp(i!kdX-iwt)
1 2 2 aw £2 n2 .
+ (*), k= c Iw -n (X), T=£(T-!kldX), kl = ar'y = 2""!1r,'3" dX we obta1n
c k
a perturbed NLS equation for B(T,y) which is
. . 2 * k. 2
B -lB -21B B =....1... B - ~ B + 0(£ ) (2.10)
Y TT 2k c2k2 YT '
where the first term on the RHS reflects the change in n(X) and the second
allows the envelope phase of the soliton to depend on amplitude. The
solution obtained in this way is valid close to but not at X=O. Near X=O,
we take n2 = w2(1+2o£2 X) and u(X,T) = £C(X,T)e- iwT +(*), and obtain to leading
order -2iwCT-C XX = -2aw2£2 XC + t n£2C2C* which may be solved exactly. The
soliton is matched to the incoming pulse of (2.10) and the penetration
depth calculated. It agrees exactly with that predicted by the earlier
argument. Therefore we might conjecture that in general nonlinear solitary
pulses penetrate past the linear cut-off until they reach zero velocity at
which point the local frequency of the pulse is equal to its initial
external frequency.

62
3. Focussing and the Behavior of Unstable Waves.
There are many physical situations which are not simply conservative and,
in addition to the balance of forces which give rise to wave motions,
involve nonconservative effects such as diffusion and external influences
which provide a reservoir of potential energy. The balance between the
external forces and dissipation forces is usually characterized in the
form of a parameter, such as the Reynolds, Rayleigh or Taylor numbers in
fluid mechanics, the inversion number in lasers, the temperature in
superconductors, the time step in a finite difference algorithm. At
certain critical values of these parameters, a fundamental and nonana1ytic
change in the nature of the solution occurs; for example the change can be
from a stationary state to a steady or wavelike, regular or irregular
motion. When the resulting motion is ordered, like in a fluid heated from
below (convection cells), or in a laser (synchronized light emission) or
a superconductor (boson like behavior of superconducting electrons (Cooper
pairs)), then we call the new state a cooperative phenomenon [15]. On
the other hand, it is often the case (e.g., instability of the Blasius flow,
certain plasma instabilities) that while at the onset of instability the
flow appears regular, it quickly degenerates and exhibits local turbulent
bursts. It is the goal of this section to suggest a possible explanation
for the concentrated patches of irregular behavior.
Consider the model equation
d d d d 2 3
L(at'ax.-,R) u=N[at' ax.-,R] (u ,u , •.. ) ( 3.1)
J J
which describes the departure u(Xj,t) of some quantity from the background
state. L is a linear operator WhlCh depends on the time, the spatial
derivatives and certain potentially critical parameters in an analytic way.
For simplicity and in fact without loss of generality in the resulting
equation we take L to be independent of x and t and the range of each Xj to
be infinite. A linear stability analysis on the solution u=o is carried out
by setting u(xj,t)ocexpi(kjxj-iat), a = w+iv which leads to the generalized
dispersion relation L(-iw+v, ik., R) = 0 which defines both the growth rate
J
v=v(kJ.,R) and the dispersion w=w(k. R). The neutral stability surface is
J,
v(k j ,R) = 0 and the critical wave-number kjc is defined by ~. = -~ ~~. = O.
J J
At k{kjc' R= Rc and we assume this val ue of R is lower than the one
obtained by setting w =0, the principle of exchange of stabilities; namely
the instability first sets in as a growing oscillation. Next suppose
R=Rc (1+t: 2x). Following NEWELL AND WHITEHEAD [16] we analyze the response
of the system by writing down the behavior of the envelope t:A(X.,T), X.
J J
dw
d XJ. - Tt) , T = t: 2t of the most unstal:, 1: mode i.e., u 'V t:A(X., T) exp
J - J
(ikjcxj-iw(kjc,R)t)+(*),

63
 2 2
aft. _ 1L (i a w + av Cl R) (3.2)
aT 2.' JI, Clk.akJl, ClR ak.ClkJl,
J, J J
where we have assumed in this case that the only nonlinear response is a
self interaction. One might expect that if x,Sr>O, the amplitude A will
grow to saturation IAI = If-
modified by a frequency adjustment exp-iS i ~T.
~r .r
We have shown however, that this solution is stable only if the matrix

(~)
Si akjClkJl, R fixed
+ av-.i!L
Sr aRClkjakJl,
(3.3)

is positive definite. This criterion reflects the battle between the

"cooperative" tendency of the system (namely, an initial power spectrum
of the spatial correlation <A(~)A(~+r»-<A>2 narrows around k = kc and the
system develops an ordered structure, and the "dispersive" tendency in which
case (3.3) is not positive definite and the spectrum broadens. This
broadening in k space is simply a result of the modulational instability
and is accompanied by a narrowing of the pulse in ~ space. In one dimension,
and in the absence of the real terms in (3.2), (3.2) is the NLS equation
and the pulses form solitons. In the presence of the real terms, we would
expect the formation of some kind of solitary pulse.
However if the dimension of the system is greater than one, then the
work of ZAKHAROV AND SYNAKH [17] suggests that a much more dramatic
phenomenon can occur. If again we neglect the terms with real coefficients
in (3.2), we obtain the higher dimensional NLS equation [19] of which the
canonical form in two dimensions is ~t-i(a~xx+b~yy)-2ic~2~* = O. If
a=b=c=l, the solution ~(x,y,t) collapses in a finite time in a self-similar
manner for a sufficiently large value of the motion constant foorl~12 dr,
~ 0
r = Ix +y • If a= -b=c=l, the case of deep water gravity waves, then y
independent solutions are unstable essentially as a result of the linear
quartet resonance mechanism of the underlying carrier wave. In fact by
setting x=r cosha, y=r sinha, x2>y2 we see that the generai solution has
a tendency to collapse onto the resonance curves x2=y2 (the portion of
the PHILLIPS [l9] figure of eight near the vertex). If a=b=-c=l, then
the system disperses and its long time behavior is given by the two-
dimensional version of the BENNEY-NEWELL [18] similarity solution
ii 2i 2()
r) 4f+T B r+···
~(x,y, t) ~ r
1
B(t" e
(3.4)

How are these results likely to apply to (3.2)? The first point we
make is that if the dispersion tensor (3.3) is nonpositive, the long
time behavior of the system is not cooperative in the usual sense. The
second and strongest point is that if (3.3) is negative, then the solution

64
begins to collapse. If Sr>O, then the collapse is eventually stopped but
the system begins to oscillate and can create local collapsing spots else-
where and the process is repeated. Furthermore if Sr is very small (as in
the case of BLASIUS flow [20]), then the large local amplitudes can give
rise to secondary (e.g., inflexional point) instabilities. IfBr<O, then
even though X<O, the subcritical case, the collapse can overcome the
initial damping of the system and reach amplitudes at which the nonlinear
instability occurs. Focussing provides a mechanism whereb* the am~litude
of linearly stable but nonlinearly unstable waves can reac the crltical
amplitude without the benefit of imperfections, end effects or large
initial perturbations.

Focussing may also enhance the tunnelling capabilities of these non-

linear pulses in systems with dimension greater than one.

Finally we mention that we are not in agreement with arguments in [17]

for the 2/3 collapse law. Let ¢t-iV2¢-i¢2¢2 = 0, ¢ = f-1Bcxpie,

x=rf- l , z=t and f(t) = 0(1) as t+to. We find

t xf2(B2)z + (xB2 (e x-xff'/2))x = 0, (3.5a)

1
x
(xB)
x
+ B3 = (~(e z-if,2/ 4 ) + (e _Xff'/2)2)B.
x
(3.5b)

Note the exact solution e=g+iff'/4, B=B(x), and cr=f2 g', a=f3 fll / 4 are
constants. However, as pointed out in [17], this solution has infinite
energy. Next take e = g + iff' /4 + ¢, neglect the second term on the RHS of

(3.5b), define J l = f oo0 xB 2dx, J 2 = foo0 xBxdx,

2 J 3 = foo0 xB 4 dx and

J4 = f~ x3B2 dx and show (by multiplying (3.5b) by xB and x2Bx respectively)

that
J 2 = J 3-yJ l -aJ 4 , J 3 = 2yJ l + 4aJ 4 ' (3.6)

where a, y are as given above but may no longer be constant. Now the
equation has the exact laws [17],

dI l dI 2 dI 4
( i t = 0, ( i t = 0, (it = 8I 2 (3.7)

where Il = f~ xB 2dx, I2 = ;2 f; (XB~ + xB2e~ -} XB 4 )dx and

I 4 = f2foox 3B2dx But from (3.6)

0 •

1 2
= ~f
II
I2 ) J4 (3.8)
and thus I 2++00 as t+to unless fa:(to-t) 1/2 or (to-t)\ A~ 1. In [17],
the authors set B=Bo + fBl with Bl constant in I 2 , and balance the term

65
 f,2/4I~ X3B~dx with 2Bl/f f~ XB~dx, an argument clearly inconsistent
with the results stated above. Hence, to date, the theoretical arguments
for the 2/3 law are not convincing. Nevertheless there are good arguments
for some kind of quasi self-similar solution (exact solutions for
f oc (to-t}1/2 have infinite energy). One argument is to note that the wave-
length A of maximum growth rate of the Benjamin-Feir instability of
<Pt-iv2<p-il<p12a<p=0 is Aa or AClA- l / a . Thus a recurring Benjamin-Feir
instability would suggest a solution of the form f -l/a ~ (r)
f .

References
1. Kaup, D.J. and Newell, A.C. (1978), Solitons as particles, oscillators
and in slowly varying media: A singular perturbation theory. To
appear, Proc. Roy. Soc. Lon. Ser. A.
2. Kaup, D.J. and Newell, A.C. (1978), A theory of nonlinear, oscillating,
dipolar excitations in one-dimensional condensates. Preprint.
3. Newell, A.C. (1978), Nonlinear tunnelling, J. Math. Phys., li, 1126.
4. Newell, A.C. (1978), Near integrable systems. Proc. Symposium on
"Inverse Spectral Transform", Rome, June 1977. To be published 1978
Research Notes in Mathematics. Pitman, London, Ed. F. Calogero.
5. Zabusky, N.J. and Kruskal, M.D. (1965), Interaction of solitons in a
collisionless plasma, Phys. Rev. Lett., 15, 240.
6. Gardner, C.S., Greene, J.M., Kruskal, M.~ and Miura, R.M. (1967),
Method for solving the Korteweg-deVries equation, Phys. Rev. Lett.,
19, 1095.
7. Gardner, C.S., Greene, J.M., Kruskal, M.D. and Miura, R.M. (1974),
Korteweg-deVries equation and generalizations. VI. Methods for
exact solution, Comm. Pure Appl. Math., ~, 97.
8. Zakharov, V.E. and Faddeev, L.D. (1972), Korteweg-deVries equation:
a completely integrable Hamiltonian system, Functional Anal. Appl.,
5, 280.
9. Flaschka, H. and Newell, A.C. (1975), Integrable systems of nonlinear
evolution equations, Dynamical Systems, Theory and Applications,
J. Moser, ed., Springer-Verlag, New York, 355.
10. Newell, A.C. (1978), The inverse scattering transform. A review
article to appear 1978 in a Springer-Verlag volume on solitons.
Editors: R. Bullough and P. Caudrey.
11. Keener, J. and McLaughlin, D.W. (1977), A Green's function method for
perturbed solitons, Phys. Rev. A, 16, 777.
12. Benjamin, T.B. and Feir, J.E. (1966), The disintegration of wave
trains on deep water. Pt. I, J. Fluid Mech., 27, 417.
13. Rice, M.J., Bishop, A.R., Krumhansel, J.A. andTrullinger, S.E. (1976),
Weakly pinned Frohlich charge-density-wave condensates: A new non-
linear current-carrying elementary excitation, Phys. Rev. Lett., 36,
432. -
14. Currie, J. F. (1977), Ph. D. Thesi s, Cornell Uni versi ty.
15. Haken, H. (1975), Cooperative phenomena in systems far from thermal
equilibrium and in non-physical systems, Rev. Mod. Phys., 47, 67.
16. Newell, A.C. (1974), Envelope equations. Lectures in Appl ied
Mathematics, 15, 157. Published by American Math. Society. '
17. Zakharov, V.E~and Synakh, V.S. (1976), The nature of the self-focusing
singularity, Soviet Physics JETP !L, 485.

66
18. Benney, D.J. and Newell, A.C. (1967), The propagation of nonlinear
wave envelopes, J. of Math. and Phys. (Stud. in Appl. Math.), 46, 133.
19. Phillips, O.M. (1960), On the dynamics of unsteady gravity waves of
finite amplitude, J. Fluid Mech., 9, 193.
20. Gaster, M. and Grant, 1. (1975), An experimental investigation of the
formation and development of a wave packet in a laminar boundary layer,
Proc. Roy. Soc. Lon. A. 347, 253.

67
Novel Class of Nonlinear Evolution Equations
Solvable by the Spectral Transfonn Technique, Including the
So-Called Cylindrical KdV Equation

F. Calogero and A. Degasperis

Istituto di Fisica, Universita di Roma Istituto Nazionale di Fisica Nucleare
Sezione di Roma, 1-00185 Roma, Italy

1. Inverse Spectral Problem for the One-Dimensional Schroedin-

ger Equation with an Additional Linear Potential

1.1 Spectral Problem

-'Yx x ( x, z ) + [x +u ( x)] "r( x ,z ) =zIi x , z ) , -ro< X(+oo, u(2:.00 )=0 I (1)

+ct>
Jdx/U(x)j/(1+!x/)1/2 < 00 (2)

1.2 -'"
Definition of Jost Function f(~

f (z ) =7T [F x ( x, z ) ¢ (x, z ) - F( x, z) ~x ( x, z ) ] (3)

where F and <p are the solutions of (1) characterized by the

boundary conditions

*~+Q? [cp(x,Z)/Ai(X-Z)] =1 (4)

*~-oo [F (x,Z)/Ei(X-Z)] =1 (5)

with

Ei(y)=Bi(y)-iAi(y) (6)

and Ai (y), Bi (y) the standard Airy functions.

1.3 Solution of Inverse Problem

jJ(z)= /f(Z) I -2_ 1 (7)

M( x, y ) = r: d z '( z ) Ai ( x- z ) Ai (y - z ) (8)

K(x,y)+M(x,y)+ f:;ZK(X,Z)M(Z,y)=o, x ::; y (9)

~(x)=-2dK(x,x)/dx ( 10 )

68
1.4 Remark

There is a biunivocous correspondence between u(x) and f(z);

and a constructive procedure is now available, involvin0 only
linear calculus. to construct f from u (direct problem; see
eqs.(l-S)) and u from f (inverse problem; see eqs.(7-10)).

2. Solution by the Spectral Transform Technique of a Nonlinear

Evolution Equation

2.1 Nonlinear Evolution Equation

ut =O(o( t ) ux+0(1 ( t) [u. xxx -6 uxu- 4 xux- 2u J, u == u ( x, t ) ( 11 )

2.2 Corresponding Evolution of f(z,t)

f t (z, t )= [O(J t ) - 4 zo(l ( t ) ] fZ( Z , t ) ( 12 )

f ( z , t ) =f (z ex p [ - 4)~ dt 0(1 ( t +

>
o
I I ) ]
( 13 )

+) ~ 0 d t o(o( t
I I ) ex p [ - 4 ) ~ t "0( 1 ( t" ) ] ' to) .

2.3 Evolution Equations Related to Eq.(ll) by Ch~f Varia-

bles

q(y,t)=jl(t)u(x,t), y= d'(t)x ( 14 )

qt=aq yyy +Aq y +Bq y q+(Ct/C)yq y +(Dt/D)q ( 1S )

a(t)=-~B(t)D(t)E(t)/C2(t) ( 16 )

E( t ) = )~ dt B( tI I ) C( t I ) D( t I ) (17)

fl = DE- l /3, r=E 2/3 IC, o(o=ACE- 2/3, ( 18 )

Remark: in Eq.(lS), A,B,C and D are essentially arbitrary fu~

ctions of t, while a is given by Eq.(16) and Eq.(17). A spe-
cial case of Eq.(lS) (A=O, B=6, C=12, D=t- 1/2 , a=-l) is the
"cylindrical KdV equation"

qt+ qyyy -6q y q+(2t)

-1
q=O, q = q(y,t) ( 19 )

In this special case Eq.(13) simplifies to

f(z,t)=f(z(t/t o )l/3, to). ( 20 )

69
3. Conservation Laws for the Evolution Equation (11)

An (infinite) sequence of conserved quantities (with explicitly

time-dependent coefficients) is associated to the flow (11).
It can be displayed by expressing the quantities f (t) (being
the coefficients of the asymptotic expansion as z ~ -~ of
f(z,t)) in terms of u. A constructive procedure to do this is
available. The first few formulae read

( 21 a )

( 21 b)
)

f 3 (t)= ~ [fl(t)]3_t~xu(x,t)[ 2x+u(x,t)] ( 21 c )

f4 (t)=f l (t)f 3 (t)-i [fl(t) J4 ( 21 d )

On the other hand the explicit time-dependence of the f 's

n

( 22)
( n- 2 ) ! !
m
2 m!(n-2m-2)!!

A to, t ) =4 ;: d t ' o<o( t ' ) e x p [ 4 ): ' d t " c(i ( t" ) ]

0 ( 0 ( 23 )

Not e t hat Eq . ( 2 2) s imp 1 i f i e s i f i n Eq . ( 1 1) 0(0 == 0 , reading

in such case simply

f n ( t ) =f n ( to) ex p [2 n J: 0 d t ' o(i( t ' ) ] . (24)

Exchanging in Eq.(22) t with to, and using Eq.(21), one

gets explicit constants of the motion for the flow (11), for
instance

(25 a)

( 25 b)

More details on all the results reported here will be publi-

shed elsewhere.

70
The Complex Modified Korteweg-de Vries Equation,
a Non-Integrable Evolution Equation

Charles F.F. Karney

Plasma Physics Laboratory, Princeton University, Princeton, New Jersey, USA
and Abhijit Sen
Plasma Fusion Center, Massachusetts Institute of Technology Cambridge, Mass., USA
and Flora V.F. Chu
Electronic Systems Laboratory and Department of Electrical Engineering and
Computer Science, Massachusetts Institute of Technology Cambridge, Mass., USA

I. Derivation of Equation

Some anisotropic media, e.g. a magnetized plasma, can support propagating

electrostatic waves. We wish to study the dispersion and nonl inear self-
modulation of such waves when the system is two-dimensional Cx and y, x
being the principal axis) and has reached a steady state [all field quanti-
ties - expC- iwt) , w = constantJ. We wi II take the medium to be homogeneous
and non-dissipative, and wi II assume that the dielectric tensor depends on
the electric field ampl itude squared. The equation for the complex elec-
tric potential, cP, is

CI)
2
We expand K about V = 0 , IVCPi = 0 Cthe long-wavelength, I inear limit)
- 2 - 2- - 2 2
KCV, Ivcpl) =K+sCI/2)C3K/3V3V):VV+sC3K/3Ivcpl )Ivcpl, (2)

where s is a formal expansion parameter. Assuming Kyy > 0 > Kxx then to
2 2 2 2
order SO CI) becomes the wave equat i on - I Kxx 13 CP/3x + Kyi CP/3y = 0 •
Expanding cP about the right-going solution we let cP = CPCT,~), where
liz
T <X Y and ~ <X X - CI~xl/Kyy) y and 3cp/3~ = OCI), 3CP/3T = OCs). To
order s we obta i n

where v <X 3cp/3~, and subscripts denote differentiation. We call (3) the
Complex Modified Korteweg-deVries equa+ion [IJ.

2. Constants of the Motion

Only four constants of the motion are known. This is in contrast to equa-
tions soluble by the inverse scattering method which have an infinite num-
ber of constants of the motion. The four constants are:
00 2
/" v d~ , (4) J I v I d~ C5)

00 4 2 00 2
I3 J_00 I v I /2 - I v E, I d~ (6) I
4
J Ivl /kdk (7)

71
In (7) k is the Fourier-transform variable conjugate to 1;,. Three of
these constants have phys i ca I i nterpretat ions. 11 = const. states that the
electric field is derivable from a potential. 12 = const. and 14 = const.
give the conservation of momentum and energy, i.e. the force and power ba-
lances.

3. Soluble Limits

Although (3) is not analytically soluble, it is closely related to the

modified Korteweg-deVries equation,
2
v1 + V1;,1;,1;, + K Iv I vI;, = 0 , (8)

which is soluble by the inverse scattering method [2J. To see this relation
we rewrite (3) in two ways,
2 2
V1 + vl;,l;,l;, + 31vl vI;, = 2i Ivl v81;, , (9)

v1 + vI;, 1;,1;, + Ivl2Vl;, =-vI V2 11;, , (10)

where 8 = arg(v). In the I imits of slow and rapid phase variation, the
right hand sides of (9) and (10) respectively are negl igible and in these
limits (3) reduces to (8), although the strength of the nonl inear term, K,
is different. When neither limit appl ies, we must solve (3) numerically.

4. Numerical Solution

We choose initial conditions of the form,

V(l=O,I;,) = Asech(l;,)exp(ikol;,) . (II)

Figs. I and 2 show two examples of the evolution. We see that there are
two types of sol itary pulses produced; one had a constant phase (Fig. I),
whi Ie the other is an envelope pulse (Fig. 2).
5. Constant Phase Pulses

The constant phase pulses are a special case of the sol itons of (8). Their
form is

v = 12 a sech[a(1;, - I;, o - a 21)Jexp( i8 0 ) ( 12)

The area of these pu I ses is 12 'IT • However, these pu Ises do not behave as
sol itons in (3). Fig. 3 shows the col I ision of two of these pulses which
have different phases, 8 . We see that after the col I ision the phase and
ampl itude of the pulses ~ave changed and some "radiation" is produced.

6. Envelope Pulses

The general form of the envelope pulses is V(l, i;,) = VU;)exp(ikol;,- iWol),
where V is complex, Z; = I;, - Cl, C = a 2 - 3k 2 , and W = k (3il 2 _ k2 )
o 0 0 0
Here ~ is the decay rate of V as 1z;1 + (Xl. V satisfies
2 2
(\z;+IVI V-a 2V'z; + ik o <3Vz;z;+lvl V~3a2V) = 0 . (13)

72
 For ko = a, we recover the constant phase pulses. Numerically Integrating
(13), we find that for a < Ikol::: a.5a, v does not form a pulse, while for
Ikol ~ a.5a we do obtain a pulse. The asymptotic form of the pulse for ko
large Is

V = 16 a sech(al;){ 1+ ie: tanh(ar,;) + e: 2 [tanh 2 (ar,;) - 1/2J/3 + O(e: 3 ) } , (14)

1.1
1.1

-20 20 -20 20
\I,
I,
I,
\I
-2
-6

Ibl
Ibl

-20 20 -20 20

-2
-6

Ie)
leI
T'=o.&

-20 \ 20 -20 20
\ I
""\I\I
-2
•
-6

~ (left) Evolution of (3) with initial conditions given by (I I) with

A = 3, ko = I The sol id I ine, long dashes, and short dashes denote lvi,
Re(v) , and Im(v) respectively

Fig. 2 (right) Same as Fig. I, except A 3

73
 1,1
T '0.2

,\
"
1\
1,1

0
1\ T ~

,
I \ -20 20
I \
I \ /,,,\ II
Ir
Ii
,
I
I \
I \
\
\
/
,/
/ \
\
\"
III
f,
/ \\. -,,' ......... - -8
15 15

Ibl
T :0.6

Ibl

-20 ,I 20
\I
1/
\I
\I
1/

-8
!
i5
lei
T =0.8

~
Col I ision of two constant phase
pulses

-8

5 Constant Phose Fig. 4 (above)

Same as Fig. 2, except A 6
4

2
A
3
\ ; L - - - - - - - ~./6

I Envelope
~.f2 ~-----A

"------------t./6 Fig. 5 (left)

t.f2I----J o (Radiation) Schematic showing numbers and
types of pulses produced with
o initial conditions given by (II)
ko

74
where E: = alk . In the limit E: ->- 0 we recover the sol iton of (8) (with
K = 3). The ,Pabsol ute" area of Vis (cf. the area of the constant phase
pulses)
2 4
!6 TI[ I + E: 14 + 0 (E: )] ( 15)

7. Transition From Envelope to Constant Phase Pulses

Since sol itary pulses do not exist for the ful I range of kola it is clear
that there cannot be a continuous transition from envelope to constant phase
pulses. If we take an envelope pulse and alter the initial conditions in such
a way that Ikol < 0.5a, we see from (15) that it has sufficient area to break
up into about 3 constant phase pulses. For initial conditions of the form of
(II) this happens when A 2ko (see Fig. 4). Fig. 5 shows schematically what
0'

pulses are produced for different initial conditions. The equivalent figure
for (8) would consist of a single set of I ines at A(K/6)1I2 = N - 1/2.

Acknowledgments
We wish to thank J. D. Meiss and N. R. Pereira for making avai lable their pro-
gram for solving nonl inear evolution equations. One of us (FYFC) thanks the
Theoretical Physics Group of Lawrence Berkeley Laboratory for its hospital ity.
This work was supported by USDoE (Contracts EY-76-C-02-3073 and EG-77-G-01-
4107), NSF (Grant ENG-77-00340), and US Air Force (Grant AFOSR 77-3321).

References
I. C. F. F. Karney, A. Sen, F. Y. F. Chu: Plasma Physics Laboratory Report
PPPL-1452, Princeton University (1978)
2. M. J. Ablowitz, D. J. Kaup, A. C. Newell, H. Segur: Stud. Appl. Math.~,
249-315 (1974)

75
III. Statistical Mechanics
and Solid-State Physics
Soliton-Bound States in the Magnetic Gap

A. Luther
Nordita, Blegdamsvej 17, DK-ZI00 Copenhagen, Denmark

Abstract

General spin correlation functions for the Ising model are re-
viewed and used to study the Ising model in a field. The quantum
double sine-Gordon equation results. At the critical tempera-
ture this equation leads to six bound states within the mag-
netic gap, one at threshold, for a total of seven. It is con-
jectured that the number of bound states is a universal property
of critical points.
Soliton States in the Ising Hodel

An interestina example of the quantum sine-Gordon equation

appears in the statistical mechanics of the Ising model in a
field. At first sight, there appears to be a little in common
between these two problems. One is after all quantum mechanical
in nature, while the other is a classical statistical mechanics
problem. The relation between the t,,'o is provided by the trans-
fer matrix method [1] , which leads to a Hamiltonian formulation
or the Isinq model. An analysis of the transfer Hamiltonian in-
cludinq the field provides some insight into the oriainal Isinq
model. This analysis is not yet complete, for solutions of this
Hamiltonian, the double sine· Cordon eauation, are lackina, but
the solutions which have been found are quite interestinq, SUq-
qestinq new physics.

There are many technical questions to deal with in deriving

these relationships. Most have been treated before [2] , and
it is only necessary to build the additional field into pre-
vious analysese. There is one caveat to this exercise however,
because, to be precise, the double sine-Gordon equation corres-
ponds to the eight vertex model [3] in the decoupling limit in
an electric field. Now the decoupling limit is known to be
trivially related to two non-interacting Ising models [4] , so
solutions to this problem are solutions to Ising, once the alge-
bra is untangled. However, in a field, the algebra is difficult
and has not yet been carried through satisfactorily. Thus the
information which is obtainable from the double sine-Gordon
problem is of direct interest for the eiqht vertex problem in
the electric field, and it is conjectured that many of the
features survive the algebraic un tanglement to the two sub-
lattice Ising case.

78
 The striking new feature of this problem, which make the
exercise interesting, is the appearance of bound states in the
transfer Hamiltonian in a field. It is not clear at present
how to interpret these states, but it is intriguing to dis-
cuss their origins. It is well known on phenomenological grounds
that an external field modifies the critical properties near
a second order phase transition. The critical region is charac-
terized by a length, the correlation length,which diverges as
the critical temperature is approached. In the quantum mechani-
cal formulation, this length corresponds to the reciprocal of
a gap energy in the eigenvalue spectrum. Vanishing gap means
infinite length, and the gap for the Ising case turns out to
be proportional to the difference in temperature and the criti-
cal temperature, T - Tc , in the absence of a field.

In a similar manner, if the temperature is precisely equal

to the critical temperature, but a field is present, the field
induces a gap, the so-called magnetic gap. It is shown here
that, in addition, there are bound states within this magnetic
gap, states which in principle can modify the behavior of the
critical fluctuations. The question of the appearance of these
bound states in various correlation functions is not yet known,
but their presence is established definately in the spectrum.
For the model of interest here, there are six bound states with-
in the gap, and a seventh at threshold.

The starting point for this discussion is the Ising model

transfer Hamiltonian in zero field. Next the steps to the
continuum limit field theory are traced to illustrate the rele-
vance of the sine-Gordon equation for the critical behavior.
It is emphasized that there are many operator equivalences in
this limit, which permit the establishment of equivalent re-
presentations of a sine-Gordon problem, fermion problem, or
the spin ~ one dimensional chain. This property suggests the
phrase "operator democracy" reflecting the absence of prefe-
rence among the representations. The field is added, and the
double sine-Gordon, in that particular language, results.

The transfer Hamiltonian is given by the one dimensional

Ising model in a transverse field [1]

J.(I (1)

where the Gi are Pauli matrices, (~-l) is a measure of the

temperature, and near the critical temperature,Tc,is proportional
to T - Tc . The sum runs over the N sites of a chain. It should
be recognized that this Hamiltonian contains two sublattices,
since the even sites are not connected with the odd sites. The
single sublattice problem, and correlation functions within a
single sublattice, are in fact needed to solve the Ising model,
and these will be extracted subsequently.

First a dual transformation on this Hamiltonian leads to

another member of the democratic representation. Introduce

79
additional spin ~ operators to be associated with bond variables
by the relations s.z=oizoi~l . It then follows that the first
part of Eq. (1), Oi~i+2 becomes SizSI+l , and the transverse
part 0i x has the same matrix elements in the set of states
defined by the bond variables as the operator product SixSi+!
A rotation about the x-axis then leads to the x-y model:

GIl Y Y x x (2)
~.xy = iL Si Si+l + ~Sl' Si+l

This spin ~ representation bears a close relation to the eight

vertex model in the decoupling limit, that is, the limit of
two non-interacting Ising sublattices (corresponding to the two
sublattices in Eq. (1) obviously).

Through the magic of the Jordan-Wigner transformation, the

spin problem can be transformed into a free fermion problem and
solved [5] . The continuum limit of this problem is the charge
density wave problem of a one-dimensional conductor, the non-
interacting Luttinger model with a gap, the Dirac equation, or
a special case of sine-Gordon problem. These steps have been
discussed elsewhere, and a quick review of important results [2]
is sufficient here.

The fermi problem at the critical temperature (it is impor-

tant to emphasize that the temperature in the fermi system is
always zero, as in Eq. (1), and the parameter A specifies the
temperature in the statistical mechanics problem) is given by:

(3)

where the spinor fermi operators satisfy the canonical anti-

commutation relations [W ,Wl+]+ = [W2,W2+]+ = 8(x-x') ,[Wl,W2]=0.
Using the boson represen~atlon for these fermi fields, it is ~
readily verified that the following bose problem results:

(4)

Here the phase operator is related to the fermi operators

through the relation
x
lP(x) ilIT fdy [Pl(y)+P2(y)] (5)

The momentum II = 4> , while and are the

fermion density operators.

80
 It is now simple to follow these steps for the anisotropy,
when 6 ~ 1 (or T ~ Tc) in Eq. (2), and therefore have the
fermionic equivalent to the temperature field. The results of
these manipulations are [6]:

J.l = I dx2
--
[( a~ ,n)
X
~
2 + IT 2 + (6-l)a -2 cosv4n
n:= (/l]
(6 )

where a is a short distance cutoff, corresponding to a lattice

spacing in Eq. (2).

This Hamiltonian, called sine-Gordon, describes the critical

region of the Ising model. The correlation functions are ob-
tained from this directly, working the transformations back-
wards to Eq. (1). An Ising correlation function along a line
is given by the equivalent object, for ~ay) even spacings,

(7)

It is now desired to obtain the equivalent operator in the sine-

Gordon language. Following the dual transformation and sub-
sequent rotation leads to the intermediate result:

(8)

while the continuum limit of the Jordan-Wigner transformation

[2] leads to the following identifications of the spin operators

-.!.
sy(x) a" cos/IT((/l(x)-(/l(O) (9)

and this operator, at least along a line (corresponding to equal

time), gives the correct Ising functions (squared).

There is an argument to find these functions correct at

arbitrary positions. This, in essence, involves the use of
Lorentz invariance together with the proper operator expres-
sion along a space direction (equal time). Lorentz invariance
implies that a Wick rotation to the Euclidean region will give
the proper correlation functions for the Ising model. It is
easy to prove that the operator of Eq. (9) does not quite
satisfy the Lorentz property, and therefore doesn't give the
correct general functions. In the original construction of
this operator, it was observed that an operator ambiguity was
removed by the imposition of Lorentz symmetry. The ambiguity [2]
arises from the (/l(O) operator, or in fact, any operator which
has no spatial dependence. But if the equations of motion are

81
to be Lorentz invariant, this operator must be dropped. Then
from Eq. (8), the square of the proper Ising functions are
given by the operator a- l / 4 coslIT~(x,t)~ p(x,t), as:

(10)

and t +iy after the calculations are performed.

It is also possible to prove this operator to be unique

(modulo the dual transform), for the coslIT ~ for 6 = 1
can be transformed to a current (or density operator) for a
massless Thirring (Luttinger) model, at a coupling strength
chosen to make the dimension of the current to be equal to
the dimension of the coslIT~ in the free case (which is ~).
Then the theorems about uniquess of currents [7] can be used
to prove that there are only two operators which are Lorentz
invariant and have a given anomalous dimension. One is the
operator coslIT~ , above, while the other is the sinlIT~
Not surprising, since the dual transform implies that at Tc
where the one dimensional models are self dual, that two
equivalent representations must exist.

The result for the four spin correlation function [8] was
confirmed using the construction of correlation functions on
a lattice before taking the continuum limit. The present
method for calculating the functions is actually trivial, for
at Tc ' correlation functions are Gaussian integrals, and the
virtues of the "continuum limit first" espoused here, become
obvious. It has been shown that for the special case of all
spins on a line, the correlation functions calculated with
this method [9] agree with previous results [10].
It is possible to identify the operator Si Y with the electric
field order parameter of the eight vertex model [2] , (sometimes
called the arrow order parameter) and to add an electric field
c·p to the transfer Hamiltonian of Eq. (6). The result is
the double sine-Gordon problem:

where the electric field, E , has been chosen in the y-direc-

tion. At the critical temperature 6 = 1, it is possible to
read off the spectrum of states, because the sine-Gordon
equation results with the cos/4n ~ replaced by cos lIT ~ The
spectrum is given by [11] ,

(12)

for n 1,2 .. 7, and the gap 6 0 is given by [3]

82
 (13)

Only the power law behaviour in Eq. (13) has been given; the
numerical prefactor is non-universal and not of interest for
the present discussion. It is remarkable that precisely seven
bound states occur for the Ising case, and it is natural to
ask if these states can appear in the spin pair correlation
function.

However, the difficulty occurs in sorting out the algebra

to check whether a relation analogous to Eq. (8) is valid.
Eq. (8) has been derived only in the zero field limit. Since
it has been used to calculate all spin correlation functions,
it is possible to use these to derive the perturbation theory
in the Ising model external field, as in fact done elsewhere
[2] . In this way, there is a close relation between the Ising
and eight-vertex models in their respective external fields,
although it is clearly more complicated than Eq. (8). Further
support for an equivalence between the two problems in an
external field comes from the fact that the magnetic gap in
the Ising case can be calculated from Eq. (8), giving the
dependence H 8/15, the correct operator dimension. The same
operator, P(x), determines the exponent 8/15 and 4/7 in both
problems, the difference arising from the fact that the "square"
of one is equal to the other.

The analytic structure of the Ising model in a field near Tc

has been extensively analyzed, [12] although results precisely
at Tc are lacking. The relationship between these two complemen-
tary viewpoints involves knowing the precise generalization of
Eq. (8) including the external field.

In this language, the critical region is described by two

gaps, one from the temperature field, T - Tc , the other from
the external field, E 477 . The analytic structure of correla-
tion functions near the critical point is obviously a compli-
cated problem, particularly since the corresponding operators
are composite operators, that is, they have matrix elements
between multi-particle states. The arrow correlation function
has the operator coslIT~ which can have such matrix elements.
If this operator has matrix elements between the ground state
and single bound state, the arrow correlation function will
have simple poles at the appropriate binding energies. Matrix
elements to pairs of these bound states will introduce a thresh-
old, and the first such threshold occurs at twice the energy
of the lowest state, 2~1 ' which is still greater than ~2'

It is interesting to consider the possibility that these

bound states are a universal feature of critical phenomena. In
other solveable models, such as the gaussian or spherical
models, there are no bound states within the gap generated by
an external field. This is not surprising, since these models

83
 are completely free field models, that is, the order parameter
is a free field and the Hamiltonian is bilinear in these fields.
Depending on the representation, the Ising case is either a
free particle Hamiltonian with complicated order parameter, or
a free field order parameter with an interacting particle
Hamiltonian. Certainly, the general expectations are that real
three dimensional models are truly interacting systems, and it
is then possible that there are bound states within the scale
gap. If so, the number of these would be a universal quantity,
dependent only upon the symmetry of the order parameter.

The importance of the number seven has often been suggested

on theological and historical grounds, and it is with some
satisfaction that this number is now singled out as special in
the Ising model. Perhaps the converse is true as well.

References

1. T.D. Schultz, D.C. Mattis, and E.H. Lieb, Rev. Mod. Phys.
36, 856 (1964)

2. A. Luther and I. Peschel, Phys. Rev. B12, 3908 (1975);

A. Luther, Phys. Rev. B14, 2153 (1976-)--

3. R.J. Baxter, Phys. Rev. Lett. 26, 834 (1971); Ann. Phys.
(N.Y.) 70, 323 (1972); J.D. Johnson, S. Krinsky, and
B. McCo~ Phys. Rev. A8, 2526 (1973)

4. L.P. Kadanoff and F. Wegner, Phys. Rev~ B4, 3989 (1971)

5. B. McCoy, Phys. Rev. 173, 531 (1968)

6. The sine-Gordon equation with a2 =4n is plagued with diver-

gence problems in the continuum limit. The discussion here
strictly refers to the interacting fermi problem on a
lattice which does not, but does have the same infrared
behavior as the sine-Gordon equation.

7. B. Klaiber, Lectures in Theoretical Physics, edited by

A. Barut and W. Britten (Gordon and Breach, New York, 1968)
Vol. XA

8. Helen Au-Yang, to be published

9. M. Bander and J. Richardson, Phys. Rev. B, to be published

10. L.P. Kadanoff and H. Ceva, Phys. Rev. B3, 3918 (1970)

11. L.D. Faddeev and L.A. Takhtajan, Theor. Mat. Fiz. 21, 160
(1974); R.F. Dashen, B. Hasslacher, and A. Neveu, Phys.
Rev. 011, 3424 (1975)

12. B. McCoy and T.T. Wu, to be published

Statistical Mechanics of Nonlinear Dispersive Systems 1

Alan Bishop
Physics Department, Queen Mary College, University of London
Mile End Road, London E1 4NS, Great Britain
and
IBM Zurich Research Laboratory, CH-8803 RUschlikon, Switzerland 2

1. Introduction

The primary recurrent theme in 'soliton physics' as it is now evolving [1J

in statistical physics, critical phenomena and condensed-matter physics
generally (as well as quantum-field theory and gravity, etc.), is the simple
recognition that linearization schemes are frequently misleading and (as
witnessed by the contents of this Symposium) sometimes unnecessary. It is
increasingly appreciated that we need to go beyond (linear) normal modes and
finite-order perturbation theories in a spectrum of intrinsically nonlinear
problems for which distinctive nonlinear, e.g., soliton, sectors of solution
space are possible. Since the physical signatures of nonlinear modes, par-
ticularly spatially-limited ones, are often so crucial, it is essential that
they are fully represented. In this situation, there is a growing tendency
to develop theories which ab initio include all fundamental modes (including
nonlinear ones) even if this is only possible semi-quantitatively. Such an
approach has well-established precedents, e.g., phonon-roton gas models in
liquid-helium-4 or vortices in superconductors.

It is precisely this idea which will motivate our discussion of the clas-
sical statistical mechanisms for a class of model, nonlinear, dispersive
Hamiltonian systems below. Essentially, we argue that a path-integral par-
tition function (Z) representation for a Hamiltonian density ~(~), i.e.,

Z = fq;[q,J exp[ -s f ~(x) dxJ (1)

[with field q,(x,t): x = space-variables; t = time; S-l= kBTJ, will be domi-

nated by extremal paths minimizing f~(x)dx, i.e., by all solutions of the
Euler-Lagrange equation of motion for q,(x,r). We then try to build a
'configurational phenomenology' based upon a mixture Df these dominant tra-
jectories with an appropriate thermal weighting. This is entirely analagous
to the procedure adopted (when practicable [2J) in quantum-field or critical
phenomena, where the dominant contributions are classical or mean-field
approximations. To make this kind of approach rigorous it is important to
have some knowledge of mode stability and of mode-mode interactions generally.
In steepest-descent (or stationary-phase) theory these arise in first order
as 'Gaussian corrections' from the interference effects of trajectori~s
neighbouring the dominant ones, 'dressing' the bare mode with quantum [2J,

ISupported in part by SRC (U.K.). 2Guest Scientist, July 1978.

85
critical [3J or thermal (§3) fluctuations as appropriate.

In a system supporting nontinear modes, superposition is, of course, not

possible. Thus 'dressing' includes contributions from interaations between
modes and we will emphasize that it can be vitally important to include these
to ensure a physically satisfying phenomenology. Specifically, we will need
to consider the effects of spatially-limited nonlinear modes (e.g., solitons)
on the phase space available to extended ones (e.g., phonons) which are
affected by boundary conditions. In fact, this problem is the natural cor-
ollary to the nonlinear conceptual revolution, namely, how to renormalize a
nonlinear ground state, mean-field configuration, classical solution, etc.,
especially in view of symmetries (e.g., translation) broken by the nonlinear
configurations. Apart from technical differences, exactly this same question
is presently posed in statistical mechanics and Renormalization-Group ap-
proaches to critical phenomena (e.g., interfaces in mixtures [3J; discommen-
surations [4J; vortices in isotropic 2-D planar spin models [5J), quantiza-
tion theories [2J, etc. As yet aomptete understanding is limited to special
models, particularly those in one-space dimension which we consider in §2.
These examples have analytically tractable features which have led to their
popularity in all the areas mentioned above, but it is important to stress
that the basic physical features and nonlinear consequences are not limited
to these special models. Certainly, numerical studies [6J and models sup-
porting topological or non-topological defect structures [l,2,5J have strongly
supported the broad concept of a configurational approach.

The parallel roles of critical, quantum and thermal fluctuations in

statistical mechanics are widely appreciated (see [lJ for a summary). Indeed,
quantum fluctuations can sometimes actually destroy a classical phase transi-
tion. Although we will deal here only with classical statistical mechanics,
it is possible to include quantum fluctuations [2,7J and conclude that at
sufficiently low T the concept of 'ideal' gases of nonlinear excitations
is still valid. It is also worthwhile noting the formal similarity between
'instantons' in quantum-field theory [2,8J and the classical statistical
mechanics of §3, although instantons are quantum-mechanical tunneling
phenomena between classical vacuum states.

The term 'soliton' is used with a variety of connotations in the natural

sciences, and even in the context of this Symposium. We should be clear
about the terminology now used in statistical mechanics. To the mathema~
tician[9,lOJ the term is usually reserved for particle-like kink, pulse or
envelope excitations appearing in totally integrable Hamiltonian systems.
These examples are extremely rare and are associated with equally special
properties (§2.3). Real physics is never concerned with such strict soliton
systems. Fortunately, many of the physiaatty important soliton character-
istics (e.g., the existence of topologically stable particle-like excita-
tions) do not demand the esoteric strict soliton" environment. Consequently,
the delightfully tractable strict soliton systems are ideal and quite repre-
sentative models if realistic physical limitations are introduced [l,llJ.
(They can usefully be considered as the nonlinear generalization of a purely
harmonic system.) This is particularly so for perturbation studies [lO,llJ
and statistical mechanics where they provide natural models for a configura-
tional approach if ergodicity-restoring interactions are introduced (§3).
Again, their simplifying role in quantization schemes is well-documented [2J.

86
In field theory and condensed matter it is now fashionable to use 'soliton'
to describe a wide variety of stable, finite-energy, particle-like field
patterns in various dimensions, which retain some central physical nonlinear
properties but few of the precise mathematical ones. We will adopt this
weaker nomenclature, referring to the mathematicians' by 'strict' (or
'aristocratic' [12]) soliton. The need to identify the more general non-
linear soliton modes has cultivated powerful generic tools which now form
equally central foundations to 'soliton physics' - the concept of 'frustra-
tion' in the presence of competing interactions [1] (e.g., spin-glass, in-
commensurate structures), the use of classical techniques of algebraic
topology (e.g., homotomy groups) to classify topologically stable 'order-
parameter defects' allowed w.r.t. a given ordered medium [13] (e.g., liquid
crystals, helium -3 and -4), etc. Topological arguments are usually possible
if degenerate ground states are available, when homotopic classification
reduces to considerations of boundary conditions. Alternatively, non-topo-
logical defects are still possible without degeneracy if non-trivivial con-
servations are imposed (of angular momentum, isospin, baryon number ,etc.)
[14]. Ultimately, specific minimization of a free energy, Lagrangian, etc.
is necessary to locate metastable states or the detailed form of stable ones.
Special models (§2) and numerical techniques are then valuable.

Although we will not be considering limiting critical effects in dimension

D > 1 in this article, it is useful to place our nonlinear concerns in the
wider framework of nonlinearity and critical or subcritica1 phenomena. Else-
where [1] we have suggested four categories: (i) first-order phase transi-
tions, including comparisons with droplet models and metastable states in
field theory, or the thermodynamics of mixtures [8]; (ii) continuous phase
transitions, and the emphasis on short-ranged, long-lived coherent cluster
excitations in addition to conventional continuum modes and associated criti-
cal exponents and time scales [15]; (iii) defected mean fields where the mean-
field solution (viewed as a, say, temperature-dependent ground state) is
itself nonlinear or 'defected', requiring critical fluctuations to be chosen
carefully (c.f. above). Examples include [4,16,17] incommensurate epi-
taxial, spin- or charge-density-wave structures; intermediate phases in
w-transforming alloys; interfacial structures, etc., and (iv) defect-induced
transitions, where phase transitions may be intrinsically associated with the
appearance of inhomogeneous (defect) states. Examples here are dislocations
at ordinary melting transitions, discommensurations at commensurate-incommen-
surate transitions, transitions between various phases in liquid crystals
[13], vortices in the isotropic 2-D XY model [5]. Note that inhomogeneous
states need not always induce a transition, although they will seriously
affect dynamics and correlations (e.g., the isotropic 2-D Heisenberg system
[18]) .

The examples we treat below fall into class (ii) so that nonlinearity
enters as fluctuation structure. Nevertheless, the concept of a configura-
tional approach and precursor effects is central to all the above classes. We
will also confine ourselves to Hamiltonians with degeneracy in their bare
form (e.g., through a phase or spin variable), rather than being generated by
renorma1izations as in a Ginzburg-Landau-Wilson approach. In fact, the
relationship between these two sources of nonlinearity is not entirely
clear, especially in isotropic systems [15]. Topological arguments are
available in our cases (degenerate ground states). The models are actually

87
quite directly relevant to some quasi-one-dimensional electronic [19J and
magnetic [20J materials - low dimensionality frequently enhances nonlinear
features, providing a substantial regime of low-dimensional fluctuations
(with respect to an ordered condensate if below a long-range ordering
temperature) outside any ultimate limiting 3-D critical region. Most
generally, we are interested in nonlinear lattice dynamics both critically
and subcritically, and our examples illustrate possible intrinsic mechanisms
for soft modes and central peaks [15J.

We reiterate that the primary emphasis is on configurational phenomenol-

ogies: both experimentalists and theorists should be at least as concerned
with order-parameter structure, as with critical exponents. Specific con-
figurations can imply very physical structural, response, energetic, transport
properties. It should be emphasized that finite lifetime modes are a very
relevant concern in statistical mechanics. Although we will concentrate on
dispersive systems it will be appreciated that nonlinear dissipative equations
are an equally challenging area of great physical importance - e.g., in
diffusive critical phenomena, population dynamics, spinodal decomposition,
chemical reaction fronts, interface dynamics, etc.

2. Model Hamiltonians and Mode Analysis

The models we will discuss below are chosen because of their analytical con-
venience but are also considered representative of very general nonlinear
statistical mechanical systems (§l). In particular, we will emphasize the
idea of a real space configurational approach and therefore need a rather
detailed analysis of the nonlinear modes available and of how they interact.
This is one reason why the models chosen are in one space dimension (D = 1),
where the information is most accessible, but also in l-D we have powerful
alternative schemes for evaluating a partition function (§3). Thus the
configurational phenomenology can be carefully monitored. The generality of
the basic lessons are, however, becoming appreciated much more widely (see
§l and elsewhere in these Proceedings), particularly through the roles of
'inhomogeneous states'. An extremely popular example is provided by the 2-D
isotropic planar spin model [5J (an example with two-component order-para-
meter dimension: n = 2). Here, topological vortex excitations are essential
to the curious non-ordering phase transition which occurs. In fact, the
partition function can be specifically decomposed into vortex and spin-wave
configuration contributions [5J. Symmetry-breaking terms restore a con-
ventional continuous phase transition [5J but then different nonlinear exci-
tations are operative - in fact, 2-D generalizations of the l-D, n = 2
'walls' we describe in §2.2. As we mentioned in §l, all of our models will
have a degenerate local potential in the bare Hamiltonian. In real systems,
this may appear through a complex order-parameter phase dependence, or from
microscopic energy calculations or because we operate well below a mean-field
scale temperature, so that broken symmetry has already developed in a mean-
field effective Hamiltonian. We will omit any Hamiltonian dependence on
thermodynamic fields (temperature, pressure, magnetic field, etc.), although
this is easily restored and essential in some problems (e.g., mUltiphase
equilibria [lJ, w-phase-transforming alloys [16J; incommensurate phases in
epitaxial, spin-phonon, electron-phonon systems [4,17J, etc.).

88
2.1. Generalized Klein-Gordon Models (D = 1)

These have motivated much recent field-theory [2] and statistica1-mechanics

[21] literature. We consider the classical Hamiltonian density [22,23]

(2)

where ~(x,t) is a dimensionless, one-component field variable in one space

(x) and time (~= ~). 'A' sets the energy scale, Wo and Co are charac-
teristic frequen~y and velocity, and the only constraint on tfie local (on-
site) potential V(~) is the topological one that it has at least two
degenerate minima (~= ~1 and ~ , say) and is bounded on (~1' ~2)' Topo-
logically stable, single-soliton (~ink, solitary wave, etc.) solutions will
then follow from the equation of motion for ~

(3)

We simply look for a travelling-wave solution and apply boundary conditions

[22] for a kink, ~~ with velocity v: using s = x - vt, the solution
describing a single kink is found as [V(~l 2) = 0]
,
2
= + [ 2wO (4)
- 2
Co
or

s =+ (5)

Here, Ivl < Co and the (static) kink 'width' is 2d = 2c O/wO' The Te1a-
tivistica11y invariant' form of (2) or (3) guarantees the covariant form of
(5) and similarly assures the velocity dependence of the energy - direct
calculation of the excess energy, E~, associated with a bare kink gives

v 0 ( v 2 )-1/2 (02 2 2 )1/ 2 (6)

EK = EK 1 - c~ = EK + p Co '

with 'relativistic momentum' p - ~_KV (1 - v

2/ Co2)-1/2 ' where
the 'rest mass' is given by
1/2
(7)

Popular analytic examples of this class include the ,~4,:v=(~2_1)2/8 [21J;

sine-Gordon (SG): V=l-cos~ [9]~ multiple SG: e.g., V=~(cos~+cos~/2) [24];
doub1e 4quadratic: V=1/2( 1~1-1) [17,25]; rectangular wells and others. SG
and ~ are representative of sub-classes in that they display bounded
period and unbounded non-periodic local potentials, respectively, with some
different physical consequences [22]. Explicit solutions follow readily
from (5) and (7): e.g.,

89
 ,4
'I' (8)

SG (9)

Having found bare (classical) solutions the next question is invariably

(§l) how these are 'dressed' with small fluctuations. Mathematica~ly this
coincides with familiar linear stability analysis [11]. SG and <P enjoy
some special simplifications in this regard (below) which is a major reason
for their popularity. However, we emphasize again that these models are only
singularly simple mathematically - the important physical conclusions almost
all generalize completely, even though analytical convenience may be partial-
ly lost.

Linearization about any of the constant vacuum states in (2) (e.g., <P
<Pl or <P 2) produces familiar extended periodic excitations (harmonic in the
l1near limit) - quite different from the particle-like, large-amplitude
solitary waves. These extended modes will correspond to phonons, mesons,
spin-waves, etc. and their unsuitability as a basis for expanding solitary
waves is precisely the theme of §l. Specifically, if (3) is linearized
about a potential minimum <PI 2 and we assume V(<P) is symmetric, the phonon
modes are '

<P(x,t) - <P l ,2 oc exp [i(Kx - wKt)] (10)

with dispersion relationship

(11)

where V(<P) has been scaled so that

A possibility always remains that linearized modes of this sort are modula-
tionally unstable if higher-order effects a~e included. Thus, expanding to
third order about a potential minima for <P is known [1] to lead to the
possibility of localized envelope solitons. Whilst such questions do not
affect the validity of formal expansions and term-by-term comparisons in
statistical mechanics (§3), they are important in interpreting precisely what
field patterns are observed in numerical simulations (or experimentally if
the pure model is directly relevant), or for establishing realistic con-
figurational ~henomenologies (§3). Thus, molecular-dynamics simulations
[26] for a <p chain exhibit evidence for pulse-soliton configurations as
well as kinks (or domain walls). The pulse-soliton modes can be viewed [1]
as (third-order) approximathons to the finite-lifetime 'quasi-breather' modes
found numerically in the <p system. However, the infinite potential walls
of this system preclude any total instability. On the other hand, the
bounded potential in SG may mean that linear phonons are modulationally un-
stable to fully developed 'breathing' modes, known to be 'normal modes' of

90
that very special system [9]. Such conclusions are subject to boundary con-
ditions - e.g., natural normal modes of finite SG, Korteweg-de Vries, etc.,
systems with periodic boundary conditions appear as cnoidal waves [27] -
although physical differences should be small in the large system limit.

Complete classes of travelling-wave solutions can generally be found as

ellip~ic functions from which kinks or phonons appear as limits. For example,
in ~ three classes are necessary [28] : the cn, dn and sn Jacobi
elliptic functions, describing oscillations above the double-well structure,
about the potential well minima, and about the potential hump, respectively.
Naturally, they have different regimes of stability.

We have implicitly assumed a continuum limit in (2). Typical physical

contexts will carry an intrinsic discreteness length scale, £. The simplest
generalization is the Hamiltonian

1 • 1
H L A £ {Z <Pi +-
2
(12)
i
with the ensuing coupled differential-difference equations of motion replac-
ing (3). Kinks remain stable excitations, although their energy, structure
and dynamics are modified [29]. Similarly, phonons are allowed with modified
dispersion
2
W (13)
K

We will need to use (13) in §3.2, rather than the Debye limit (11). The
continuum or 'displacive' limit is valid for d» £, i.e., when the coupling
between neighbours is large compared to the on-site potential so that <P-
variations from site to site are gradual.

Linear kink stability analysis proceeds [11] by substituting <P = <P + ~

in the governing equation of motion (2) and linearizing in ~ , which fhen
satisfies a 'scattering' equation with solutions of the form ~ = f(x) exp
(iwt). In all cases there will be at least one bound state (B) with OW = 0,
representing (to linear order) rigid translation of the kink:fB(x)~d<PK/~x.
This may be termed a Goldstone translation mode. There may be further finite
frequency bound states describing localized kink oscillations. In addition
t~ere will be a scattering continuum f (x) with w = w. In fact for SG or
<P the scattering problem is purely re~lectionless [30]~ The phonon dis-
persion (11) is unchanged by the presence of a kink, whose only effects are
to produce an asymptotic phonon phase shift and trap the finite number of
modes described above. The number of modes trapped is related to the asymp-
totic phase shift in a Friedel sense, (c.f., impurities in an electron gas).
In the SG case, for example [11], fB(x) = (2/d) sech(x/d) and there are no
other bound states. The continuum modes are

Examining the asymptotic forms of fK(!oo) we find that the phonons have
suffered an asymptotic phase shift

91
 O(K} = nK/IK I - 2 tan
-1
(Kd) (14)

due to the (static) kink. The nonlinear modes satisfy orthogonality and
completeness relations and the energy, E, is given to this linear order by

E = ~co2 + 21 ~ v
2
+
J dK p(K} wK • (15)

Here v is the kink velocity and p(K} the density of phonon states. In
these senses the 'kink-phonons' and kinks are 'nonlinear normal modes'.

The most important point is that the phonon density of states is ahanged
by the phase shift, since these modes are extended. If we introduce periodic
(Born-von Karman) boundary conditions on a chain of finite length L, then
the allowed wave-vectors are

L Kn + O(K n } = 2nn (n = 0, + I, ..• ) (16)

and

(17)

From (14) and (16) we see that the KO = 0 mode is no longer allowed and all
other modes are shifted tow2rds2 ~V2 0 slightly. (For a kink of velocity
v, the mode at K d = -vi (c O- v ) is lost.} As required by degree of

number of bound states: fdK~p(K} = -n 0(0+} = -1, where +

freedom conservatIon, the total number_£f phonon states is decreased by the
signifies
Cauchy Principal value. We see here the essentiaZ mechanism by which the
kink's creation and translation are compensated for by two degrees of free-
dom in the phonon sea.
4
For ~ a second mode is removed from the continuum at frequency wI =
V2 13 Wo and the maximum asymptotic phase shift, 0(0+}, is 2n. The multi-
ple SG and double quadratic systems are examples without the simplification
of a reflectionless potential. In fact, the latter case presents a textbook
o - fn, scattering potential [25J, so that again only one mode (w B = O)
is trapped. The asymptotic phase shift is actually the same as for SG, i.e.,
(14). [f (x) is a superpos!tion of + K phonons.J These circumstances
explain the popularity of ~ and SG, but again demonstrate that the central
physical conclusions are much more general, even if potentials are reflecting
or if numerical implementation of stability analysis is necessary. The
results for class (2) are given in complete generality in [23J.

2.2. Complex Scalar Fields with Phase Anisotropy

Multi-component order-parameter Hamiltonians (n > I) are also important in

many physical problems, e.g., with coupled order parameters, complex order
parameters (n = 2), spin systems, etc. For D > 1 various defect types are
possible [13,14,18J but even in D = 1 stable solitons are possible if we
introduce anisotropy in n-space.

92
 Several models have been studied but an n = 2 system with Ising (uni-
axial) symmetry is especially tractable [31J: the Lagrangian density is
cp 11 2 1 2 1 2 1 4 I 2
~~oc - ~ I - -I~ I + -AI~I - -BI~I - DI~I (1 - cos 2~) (IS)
2 t 2 x 2 4

with = u(x,t)exp[i~(x,t)J and A,B,D positive constants. The la~t

~(x,t)
th~ee term2 in (IS) can be thought of as the local potential V(~) = -Au /2 +
Bu /4 + Du (1 - cos 2~) wh~ch corresponds to the shape of the 'bottom of a
wine bottle' (c.f., the ~ model,§2.l) whose continuous rotational symmetry
has been broken and replaced with a (2-fold) discrete one. There are thus
two degenerate potential minima and solitary-wave profiles will now corre-
spond to coupled variations in both u and ~ as the field evolves between
the two vacuum states. The amount of coupling should clearly depend on the
relative magnitudes of the central hump and phase 'dimple' potentials, or
equivalently the ratio D/A.

Linearization about a ground state readily produces decoupled amplitude-

phonon and phase-phonon ('amplitudon' and 'phason') approximate, extended
modes analogous to those for class (2). The complex kink solutions are best
derived by transforming to the variables S = Re(~), n = Im(~) in which the
onl~ asymmetry derives from symmetry-breaking, i:e., D # o. Two classes ~f_V2
so11tary waves can be found [31J upon transform1ng to s = (x - vt)(l - v )
clvl< 1)

sl(s) = + U tanh [(lA) V2 sJ

- 0 2
(19)

and
V2
= ! U o tanh (2D s)
(20)
SD V2 V2
=! uO(l -~) sech(2D s).

Solution 2(20) is admissable only if A > SD. Eqs. (19) and (20) coincide
when A = SD. We therefore have a classic example of ground state (or mean
field) 'bifurcation' (or equivalent)2D phase transition [1,31J. Although
both solutions exist for A > SD only type 2(20) is stable, as can be proven
[31J by the same linear stability analysis described above. The same con-
clusion is suggested by the relative magnitude of kink energies [31J

(21)

N~tice that because of the Ising symmetry, type 1 (19) is no more than the
~ solution (§2.l) and D-independent. Similarly, in the limit D« A,
amplitude variations are costly in energy and the solitary-wave evolves at
essentially constant amplitude. This 'phase-dominated' regime [19J in fact
corresponds to the SG equation, as we see from (20). Model (IS) nicely
demonstrates how the amount of phase-amplitude coupling is determined by the
relative magnitudes of the phase and amplitude potential components.

93
2.3. Integrable Hamiltonian Systems

Clearly, travelling waves do not represent a complete solution to class (2).

They are unable for example to describe a multi-kink configuration. In some
cases more general special solutions are known, together with special tech-
niques for generating them, e.g., Backland transformation, Hirota's technique,
the inverse scattering transform, etc. [9,10]. In practice, all of these
devices and special solutions apply (almost invariably) to a very speeiaZ
subset of nonlinear equations, namely those deriving from fully integrable
Hamiltonian systems. It is well known [9] that examples of this type are
extraordinarily generic physically but also highly singular mathematically.
We will not describe examples in detail (see elsewhere in these Proceedings)
except to point out that SG is the only member in class (2). Fully integrable
systems make very beautiful idealized theoretical models (c.f. the role of
the harmonic system in linear physics), primarily because a eompZete know-
ledge of solutions to the corresponding equations of motion is possible and
(nonlinear) normal modes can be identified, lending themselves directly to
the idea of a configurational basis. In these special cases, pulse, kink or
envelope solutions correspond to our terminology 'strict soliton'. Such
soliton-bearing equations are associated with a variety of related and quite
remarkable special properties. For example, with appropriate boundary condi-
tion assumptions, they are eompZeteZy soluble, via (for example) the 'inverse
scattering transform' [9,10], a canonical transform to functions of general-
ized action-angle variables which serve to label allowed types of excitation
('nonlinear normal modes') in the system. The allowed excitations are infi-
nitely long-lived, retaining their asymptotic identity exactly under 'colli-
sion' with other excitations. They simply suffer pair-wise add~tive asymp-
totic shifts. (c.f., Zinear order results of §2.l for SG and ~.) Thus,
general solutions can be labelled by their nonlinear normal mode components,
and indeed can be expressed in the form of a generalized Fourier transform
with respect to these [10]. A number of related properties also follow
including recurrence phenomena, an infinite number of conservation laws, etc.
For our statistical mechanics purposes, the most suggestive consequence
[c.f., the linear order result (15)], is that the associated Hamiltonian (and
momenta) are exaetZy separabZe [32] into contributions from the nonlinear
normal modes. (See also parallel uses of these ideal models in perturbation
studies [10] and quantum field theory [2].)

It should be noted that the most frequently cited integrable system forms
apply strictly only to the continuum, infinite system limit with boundary
condition ~ + 0 sufficiently rapidly as Ixl + 00. (e.g., the SG 'normal
modes' are t~en [1,32] solitons (antisolitons), 'breathers' and continuum
excitations.) This may be important since partition function or quantization
calculations are regularly performed (§3) on a finite, discrete system and
with a variety of boundary conditions; continuum and infinite system limits
are taken last. In fact there is a good deal of interest from mathematicians
in rigorous studies of finite versions of, e.g., SG with periodic boundary
conditions [27]. Again, several discrete integrable models soluble by in-
verse scattering are available, although it must be noted that these do not
coincide with the local discretizations usually posed in condensed matter
[i.e., (12)].

94
3. Equilibrium Statistical Mechanics: Exact Results and Configuration Pheno-
menology

The ideas described in §l have prompted attempts [21,23] to construct con-

figurational phenomenologies emphasizing elementary nonlinear excitations.
We describe here how this approach can be made rigorous for the low-T limit
of class (2) if mechanisms for degrees of freedom conservation and free
energy sharing are incorporated, arising from the non-superpositional inter-
actions of §2.l. An extended discussion of these results will be given
elsewhere [23], but the principal ideas are summarized below. We first
describe an exact calculation of the classical partition function Z in the
original variables (~,~,~ ), with which we then compare phenomenology.
x

3.1. Exact Partition Function Calculation

For the discrete system governed by (12) we can use a transfer integral
technique to calculate Z, details of which can be found elsewhere [2~]. We
ultimately compare results with a phenomenology based on §2.l, so that we
will primarily consider the continuum limit ~/d + O. Classically, Z
factors as Z = Z~ Z~ with [23] (L = chain length)

z· (2TI A ~/s h2)N/2 (22)

~
(23)
n
-1
where S = (k T) , h is Planck's constant and L = N~. Eq. (25) assumes
periodic boun~ary conditions ~N 1= ~l' but others are readily possible [23].
{(} in (23) are the eigenvalue; of the transfer integral operator [21],
wh~ch can be solved by various techniques. However, in the displacive limit,
much more intuition is possible from a differential approximation valid to
O(~/d) [21,23]

(24)

with
1 d2
H(~) + V(~) + Vo (25)
* d~2
2m
* A2 2 2 S2 (SE O)2
m Wo Co ~ (26)
K
2 -1 2
Vo (2Sw O ~A) In (A cOS/2TI~). (27)

We are thus faced with ~ pseudo-Schrodinger equation for a single particle of

dimensionless 'mass' m, in l-D, moving in the nonlinear potential V(~)
[33].

Physical interpretations are possible at both high and low T. Here we

concentrate on m*» 1. Note first from (23) that in the thermodynamic
limit (L,N + ~ = const.),
00 ; Z is dominated by the lowest eigenvalue EO.
Thus the configurational (i.e., $otential energy) free energy density con-

95
tribution is

(28)

We therefore need to determine €O. At low T, i.e., SE~» 1, the low-

lying eigenspectrum will be 'tunnel-split'. Specifically, if EO is the
lowest level in a single isolated well of v(~), then
(29)

where to is the tunneling component. For large m* ,EO is appr~~~mated by

the 2 10west harmonic oscillator level (suppressing VO): E = 1/2 m* +
O(T). The splitting is given reasonably accurately by tRe WKB tunneling
formula [34J

(30)

(31)

Here determine turning points for the WKB integral.

Periodic potentials (e.g., SG) differ from unbounded ones in several ways.
For example, the pseudo-Schrodinger equation (24) poses a l-D 'band struc-
ture' problem (Hill equation). Thus the eigenfunctions have Bloch form and
the eigenvalues lie in a sequence of continuous bands - despite the absence
of a discrete spectrum, result (28) still holds [23J. For the SG case, the
eigenproblem (24) is actually equivalent to the Mathieu equation, which is
extensively documented [35J: asymptotic expansions in the low T ('tight-
binding') limit give the centre of each band as (n = 0,1,2, ... )
o -1 -
En = 4(2n + l)(SE K)
2 0 -2
[(2n + 1) + IJ (SE K) + ... (32)

The linear T term is the harmonic oscillator level EO. Eigenvalues in the
n-th band differ only by tunneling. An improved WKB calculation [34,35J
[c.f. (30)J gives the bandwidth

2 tn(T) ~ 24(n+l)(n!)-1(2/TI)1/2(tSE~)n-1/2 exp(-SE~). (33)

Contrast this with the band gap T-dependence

(34)

3.2. Configurational Phenomonology

We now ask whether the kink (5) and phonon (13) excitations appear as
natural elementary modes in the formal statistical mechanics of §3.l.

Follow~ng the qualitative suggestions of KRUMHANSL and SCHRIEFFER [21J

in the ~ example, it is not difficult to demonstrate that the component
EO of €O' taken together with aZl of F" [from (22)J and FV '
~ 0

96
corresponds exactly to the free energy, FO' of a set of one-dimensional
classical (kBT» hw ) harmonic phonons t13) - calculated to O(£/d) to
be consistent with (zg). It is essential that the discrete dispersion rela-
tion (13) is used in order to be consistent with the transfer operator
technique. We find
kBT +rr/£
Fo /L -- -Zrr f dK In (Shw K)
-rr/£

£~ kBT {£-l In (TIwOSd/£) + (Zd)-l}

Aw; EO - kB T L- l In Z¢ + AW; VO' (35)

This identification of EO leads naturally to the speculation [21J that

to corresponds to the free energy of a gas of independent kinks. It is easy
to sBoW in general [22J that as T + 0, 6~1 2 + 0, EO + 0, and so I(T) + 1(0)
= SE , from (7) and (31). This result certa~nly suggests [21J an exponen-
tial~y low kink density but includes no thermal kink dressing (§l) and con-
sequent kink-phonon non-superpositional interactions. Indeed in the strict
limit T + 0 no modes survive at all! Note similarly that all of the dynam-
ical free energy F' is apparently taken up by the phonon modes. Mode
interactions are es~ential to account for kink dynamics (§Z.l), but formally
(§3.l) we must calculate 6~ to the same order as EO to be consistent.
Explicit calculations [23J to linear T (harmonic order) show that periodic
members of class (2) have the 'universal' [22J form

The numerical proportionality constant depends on the particular V(~) and

the accuracy of the WKB formula (30) [34J.

To understand these formal results phenomenologically, we must include

kink-phonon interactions explicitly, as found in §2.l. The simple lesson was
that the density of states of extended modes (e.g., phonons) are changed by
the presence of local defects (e.g., solitons). This results in a change in
the phonon free energy which we view below as a 'kink self-energy'.

At low T only low velocity (v« cO) kinks need be considered. The
change, ~F, in phonon free energy is thus [c.f. (35)J

~F = kBT ++rr/£dK In (STIW ) 6p(K)

-rr/£ K
(37)

£+0
+ - kBT NB In

r
(38)

+ kBT (2rr)-1
0+

where we have used (17) and NB is the number of bound states for the par-
ticular V(~) (see §2.l). From (38), ~F is generally negative [c.f.(14)J
- the phonon free energy is reduced by the kink's presence, as expected from
the reduction in the number of allowed phonon states. We interpret this as

97
a contribution to the kink self-(free) energy, which is similar in spirit to
that adopted [2] in the quantum renormalization of the kink mass from its
effect on the vacuum zero-point energy [36]. Similarly, another contribution
to the kink self-free energy will come from any localized internal harmonic
oscillations corresponding to additional 'bound' states (§2.l). The total
kink self-(free) energy is thus
N
LK(T) = ~F(T) + kBT LB In (eflw ),
n
(39)
n=2

where {w} are the internal 0Bcillation frequencies. (If NB= 1 there is
no secondnterm.) For kBT« EK, the kink density is exponentlally small and
we can associate the full L~ with each kink (or antikink) as well as
neglecting kink-kink interactlons or ordering restrictions. For periodic
boundary conditions the number of kinks (NK) and antikinks (N~) are equal,
but_tparticularly f~r periodic V~~)] the net 'winding number densi:y:
= 1 (NK- Ni)' can ltself be physlcally relevant and serve as an addltlonal
thermodynamlc variable [37]. We restrict ourselves here to periodic boundary
conditions for brevity.
r--1
We can now construct a grand canonical partition function ~ in mode
representation. Simple factorization (as in a linear system) is not possible
because of the (phase-shifting) interactions. However, ~F (total) is pro-
portional to the number of kinks (at low T), which leads us to write [38]

.-------, ,------, (0) ,---, ,...--, (40a)

I----t I----t I--t t-1
L---.....J P K L-..J K
~ '---'
~o
I---i exp(-j3 FO) (40b)
'-------' P

r---r0
Here~ is the free phonon result and ZK(N K) is the classical
partition ¥unction for an ideal gas of NK indistinguishable 'relativistic'
particles with self-energy LK• The dimensionless constant B is a T-
independent phase space factor (see later). Since there are no external
constraints (in the thermodynamic limit) we eventually set ~K= ~K = O.
Similarly, there are no constraints on 'phonon' numbers (1 ~ 00).

(41)

The grand canonical potential density is The

average total kink number density is then

98
 n.~OT
K.
= L-l(N K + N-K) = -
(")/')
0 " 0 flK T, L

(evaluated at fl = 0), and the free energy density is F/L

We find explicit~y from (35), (40) and (41) that

~OT(T) o 0 V2
(2E K/Bhc O) (2n/SE K)
0
exp[-S(E K + I K)] (42)
N w
-~T In ( Sfiw O) - kBT lB In (--.9.)
w
n=2 n (43)

-1 oof do 2 2
+ kBT (2n) 0+ dK (dK) In (1 + d K ).

From (43) the T-dependence ofO IK is quite simple and yields a 'universal'
[22] dependence of ~ on (SE K):

~OT(T) = (2n)-V2 (2eo/Bd)(SE~)V2 exp(-SE~). (44)

where 0 = -s I (T) - In(Sfiw o) F o(T). The numerical factor B has to be

obtained by cowp~ring the phenomenological and exact free energies [23] -
B(SG) = 1; B(~ ) =44; B (double quadratic) = 21:2 ; o(SG) = 0 (double quad-
ratic) = In2; o(~) = In(413). The exact agreement for SG is presumably a
reflection of its strict soliton interactions. Other thermo~rnamic functions
now follow. For example, the internal energy density u = L a(SF)/as can
be written in the suggestive form [38]

u = (~-l - NB ~OT) kBT + ~OT [E~ + ~ kBT + (NB-l) kBT]. (45)

. 1S
Th1s . s1mp
. 1y the 1nternalTO~ergy
. . h (L/ ~ - NBL n TOT) coassi-
of a system W1t K
cal harmonic modes, and L~ 'Newtonian particles' of rest energy EK
each having kBT/2 translational energy and thermal energy kBT for each of
the (NB- 1) 1nternal oscillation modes.

At higher temperatures the complete phenomenology described above is not

available (except in principle for integrable systems). Higher-order terms
in EO [c.f.(32)] are purely anharmonic phonon contributions and include any
'breather' contributions. Mode-mode interactions (virial, etc.) are in-
creasingly complex and have only been included approximately, althoueh with
some success even near the incipient first-order transition [22] SEK ~ 2.

It will be clear from §3.2 and the above discussion that fulZy integrabZe
systems are in principle ideal for configurational phenomenologies at all
temperatures - the energy is exactly separable into 'normal modes' whose
interactions are purely asymptotically phase-shifting and pair-wise additive
(not requiring low T). Thus, the formulation (40) immediately generalizes to
the complete basis. The familiar SG example (with decaying boundary condi-
tions) has been investigated in detail [23]. However, the 'breather' con-
tribution for that example produces some anomalies [23] when compared with
the transfer integral results, which have yet to be fully understood. They
may arise from the discrete-lattice-finite L limits or the different

99
boundary conditions - an ideal comparison should be with a separable, finite,
discrete model with, e.g., periodic boundary conditions. Intensive thermo-
dynamic quantities are insensitive to these questions but appropriate modes
for a configurational phenomenology may be more so (see also §Z.3).

We have already mentioned (§l) the separable Hamiltonian form for the Z-
D XY model partition function in terms of spin waves and vortices, which
emphasizes the conceptual similarity with the l-D models above. This is
again a nonlinear separation as is shown by Renormalization-Group studies
[5] which self-consistently generate mode interactions.

4. Correlation Functions

Appropriate static correlation functions (see below) are sensitive tests of

kinks in particular, and dynamic correlations can be even more so, leading to
central-peak and soft-mode descriptions [15]. The latter are best discussed
elsewhere in these Proceedings but we note that again mode interactions are
crucial and should be totally manageable in integrable systems. In practice,
this can be complex and is frequently omitted [39], although approximate
phenomenology can be very satisfactory.

Thermodynamic functions such as F or u are dominated by phonon con-

tributions at low T, since the kink density is exponentially small. However,
kinks will dominate those properties insensitive to phonons - e.g., DC
transport in periodic V(~) models [40] or appropriate correlations. Static
correlations follow formally [21] from the transfer integral technique (§3.l):

C(x)= <g[Hx) ]g[HO)]>

(46)
L l<nlg(~)lo>lZ exp[-BAW~ (En - to)x]
n
Z
Typical correlations consider g(~) = o~,o~ , exp(i~),etc., where o~(x)-
~(x) - <~>, etc., and exp(i~) is most relevant if v(~) is periodic.

At large distances C(x) will be dominated by the state with the smallest
eigenvalue which is coupled to the n = 0 level [i.e., has non-zero matrix
element in (46)]. For periodic V(~4 these are determined from a 'Bragg
condition' if g = exp(i~). The ~ -type example is very familiar [21,23].
Here the order-parameter correlation length (g = o~) is dominated by the
lowest tunnel split levels and is thus proportional to anO'activation' term
exp(BE ), where E is the kink creation energy: E = EK (T + 0). This is
expect~d physicall~ since ~ is changed by the pres~nce of a kink (from
~ + 1 to - 1). In fact the correlation lengthzis then essentially the
ln~erse kin~ density [21]. On the other hand o~ is insensitive except in
the immediate kink vicinity and shows no long-range correlations, which we
see formally since the lowest coupled levels in C(x) then differ by a term
~ T, as T + 0+.

The same ideas apply [31] to the complex order-parameter system §Z.Z. There
the 'Schrodinger' problem (24) has a 'band-structure' distribution of eigen-
values (as for SG). There are no long-range correlations as in olwl 2 (i.e.,

100
correlation length f as T ~ 0+) because the lowest coupled levels are
in different bands and differ by ~ T [c.f. (34)J. Physically, 1~12 is
unchanged by the kinks and again the correlation length is characterized by
the defect width rather than average separation. However, complex kink
energies will characterize activated behaviour for correlations in the field
itself [<o~(x)o~*(O»J: ~ changes locally from + u (§2.2) in passing
through a kink and correspondingly the lowest couple~ levels are the top and
bottom of the first band, whose width is exponentially small at low T. In
view of the crossover from SG to ~4 as D/A ~ VB described in §2.2, it is
interesting to isolate the activation energy E as a function of D/A and
compare it with the complex kink creation energ~. A numerical determination
of E as T ~ 0 has been found [31J to agree excellently with the kink energy
(21)~

Finally, we emphasize the general importance of order-parameter defects

for correlation behaviour. To choose one example, the non-topological in-
homogeneous states in the 2-D isotropic Heisenberg model [18J do not result
in a transition (as for 2D XY [5J) but do change temperature dependence from
power law to exponential in the order-parameter correlation length.

5. Summary

The roles of nonlinearity in lattice dynamics, critical phenomena and statis-

tical mechanics are numerous and at a point of rapid development as will be
seen from other contributions to these Proceedings. I have tried in this
article to focus on a few simple models which can be handled in some analytic
detail, but nevertheless illustrate very general nonlinear lessons - see, e.g.,
the molecular-dynamics results reported by SCHNEIDER and STOLL. [In addition
the models of §2 are directly useful for some physically accessible quasi-I-D
systems - see, e.g., articles of M.J. RICE and M. STEINER.J I have emphasized
especially the usefulness of a configurational phenomenology as a conceptual
and practical scheme. As a corollary, it was necessary to consider (non-
linear) mode-mode interactions very carefully to properly account for conser-
vation of degrees of freedom and essential mechanisms of free-energy sharing
(§3).
It has not been possible to describe recent activity [40J in non-equi-
librium statistical mechanics for the nonlinear models in class (2) of SG
type, although these properly complement the equilibrium motivations above.
One approach has been to employ a Fokker-Planck formulation and then the
analogue of steepest descent trajectories in the partition function are the
most probable evolution paths in phase space. The transfer integral tech-
nique is useful here [40J as for the equilibrium problem (§3.1). The driven,
damped SG system in the presence of thermal noise is generically interesting
as a model for nonlinear transport or phase slippage [lJ. Once again, identi-
fying configurations leads to an appealing physical interpretation of formal
results and phenomenological nucleation descriptions have been suggested [41J.
In particular, at low constant driving fields and temperatures, solitons are
found to be the dominant transport mechanism, as confirmed by molecular-
dynamics simulations [42J.
I am grateful to many colleagues, especially J.A. Krumhansl for his fore-
sight, and S.E. Trullinger for most enjoyable collaborations.

101
References

1. A.R. Bishop: Chalmers Soliton Symposium (Goteborg, June 1978, Sweden)

(to appear in Physica Scripta)
2. R.F. Dashen, et al.: Phys. Rev. D 11, 3424 (1975); S. Deser, et al.:
Nucl. Phys. B 114, 29 (1976); R. Jackiw: Rev. Mod. Phys. 49, 681 (1977)
3. J. Rudnick, D.-Y;snow: Phys. Rev. B 17, 1351 (1978) --
4. See, e.g., P. Bak: these Proceedings--
5. J.V. Jose, et al.: Phys. Rev. B 16, 1217 (1977); see also T. Schneider,
E. Stoll: these Proceedings --
6. See, e.g., T. Schneider, E. Stoll: these Proceedings
7. S.E. Trullinger: Preprint
8. See, e.g., D.J. Wallace: these Proceedings
9. A.C. Scott, et al.: Proc. IEEE ~, 1443 (1973); R. Bullough: these
Proceedings
10. See A.C. Newell: these Proceedings
11. A.R. Bishop: In Solitons in Action, ed. by K. Longgren, A.C. Scott,
(Academic Press, New York 1978)
12. I am indebted to T.D. Lee for the imaginative terminology!
13. M. Kleman, L. Michel: J. de Phys. Lett. ~, L29 (1978); see also K.
Maki: these Proceedings
14. R. Friedberg, et al.: Phys. Rev. D 13, 2739 (1976); G. Toulouse, M.
Kleman: J. de Phys. 12, L149 (1976)
15. See, e.g., A.D. Bruce: these Proceedings
16. See B. Horovitz: these Proceedings
17. See S. Aubry: these Proceedings
18. A.A. Belavin, A.M. Polyakov: JETP Lett. 22, 245 (1975)
19. See, e.g., M.J. Rice: these Proceedings--
20. See M. Steiner: these Proceedings
21. J.A. Krumhansl, J.R. Schrieffer: Phys. Rev. B 1l, 3535 (1975)
22. A.R. Bishop: Physica (in press)
23. J.F. Currie, et al.: in preparation
24. See, e.g., P.W. Kitchenside et al.: these Proceedings
25. S.E. Trullinger, R.M. DeLeonardis: Preprint
26. T .. R. Koehler: (unpublished results 1975)
27. S.P. Novikov: JETP 40, 1058 (1975); G. Costabile, et al.: Appl. Phys.
Lett. 32, 587 (1978~
28. S. Aubry: J. Chem. Phys. 64, 3392 (1976)
29. J.F. Currie, et al.: Phys-:-Rev. B 15, 5567 (1977)
30. L.D. Landau, E.M. Lifshitz: In Quantum Mechanics - Non-relativistic
Theory (Pergamon Press, Oxford 1965), pp.78-80
31. S. Sarker, et al.: Phys. Lett. 59A, 255 (1976); J.F. Currie, et al.:
Preprint; see also J. Lajzerowi~ these Proceedings
32. V.E. Korepin, L.D. Fadeev: Theor. Mat. Phys. 25, 1039 (1975)
33. The formal analogy with a zero-dimensional field theory and instantons
is striking here (see Refs. [2J and [8J) V2
34. Formula (30) should be corrected by the universal factor (rr/e) for
symmetric members of class (2) (R.M. De Leonardis: Private communication)
[see also (33)J
35. M. Abramowitz, I.A. Stegun: In Handbook of Mathematical Functions (U.S.
Department of Commerce)
36. Although the rigorous procedure is to identify 'collective coordinates'
and orthogonal modes (Ref. [2J), Fadeev has shown that, to each order in

102
 quantum perturbation theory, free continuum modes can be used (no
phase shifts) if the zero frequency modes are excluded to avoid
double-counting
37. J.F. Currie, et al.: Phys. Rev. A 16, 796 (1977)
38. These decompositions are not unique but aesthetically satisfying in
view of our configurational approach or comparisons with quantum field-
theory treatments (Ref. [2])
39. K. Kawasaki: Prog. Theor. Phys. 55, 2029 (1976); Y. Matsuno: Phys.
Lett. 64A, 14 (1977); H.J. Mikeska: J. Phys. C 11, L29 (1978)
40. S.E. T~linger, et al.: Phys. Rev. Lett. 40, ,206 (1978)
41. R. Landauer: Private communication
42. T. Schneider, E. Stoll: these Proceedings

103
Some Applications of Instantons in Statistical Mechanics

D.J. Wallace
Physics Department, The University, Southampton S09 5NH, Great Britain

Abstract

Instantons are solutions of non-linear equations associated with tunnelling

phenomena. We discuss how instantons characterise the behaviour of diverg-
ent perturbation series and how this information can be used to obtain con-
vergent results from these series. Applications to c-expansions and phase
transitions are reviewed. We indicate how the same instanton techniques
apply to the problem of the electron in a random potential and to metasta-
bility.

1. Introduction

This talk is concerned with handling problems in statistical mechanics where

the Hamiltonian looks unbounded below. The basic idea is to define the par-
tition function in this case by analytic continuation from a well-defined
problem with a bounded Hamiltonian. An example which we will consider in
some detail is the Landau-Ginzburg-Wi1son Hamiltonian of a fluctuating
field +(x)
f.I.."= Idd xU (V4') 2 + ~m24'2 + !g4'4}, (1)

in the case in which the coupling g is negative; this is defined by con-

tinuation from the usual theory for positive g. (Here if means the reduced
Hamiltonian, containing l/k BT.) As we shall see, a typical result of the
continuation is that for negative g the partition function, which is real
for positive g, picks up an imaginary part. This talk is about techniques
for calculating this imaginary part.

The main justification for including this topic in this symposium is

that non-trivial, extended solutions of the classical field equation

(2)

playa central role in these techniques. These solutions are the instantons
referred to in the title. These calculations are to be contrasted with the
standard perturbation expansion in the anharmonic coupling g by Feynman
graph expansion; at any order in this perturbation theory, the partition
function is real for real coupling g (regardless of its sign). The result
of the instanton calculations is an imaginary part (for g<O) which is
exponentially small in g (_ . exp(const/g) ).

Despite one's anticipation that the partition function should be real in

any physical problem, the imaginary part provides extremely interesting
information for a number of problems in physics. In this talk I shall
indicate its importance in three areas: (i) to obtain the asymptotic form
of the perturbation in g i.e. the behaviour of the high order terms;
(ii) to obtain the density of states in the tail of the band for an electron

104
in a random potential; (iii) to elucidate metastability and the nature of
the singularity on the coexistence curve (H~, T<Tc in magnetic notation) of
Ising - like systems.

Some remarks about the historical development of this subject are probably
worth-while. The use of non-trivial solutions oJ field equations in problems
(ii) and (iii) above was recognised over ten years ago [1, 2 and references
thereinJ. Identical techniques were developed independently by quantum field
theorists to discuss the tunnelling phenomena which occur in Non-Abelian
Gauge Theories of elementary particle interactions [3,4,5,6J ; the word
instanton was coined in this work. The particle physicists also revisited
[7J the problem of condensation considered by Langer [2J . Following some
pioneering work by Bender and Wu [8 ; see also 9J it was recognised that the
same techniques could be used to obtain the asymptotic behaviour of pertur-
bation expansions in quantum mechanics, quantum field theory and statistical
mechanics [10,11].

The form of instanton solutions will be clarified later. At this point

it is sufficient to say that they have some features in common with kinks
and solitons; here I am characterising instantons as objects associated
with tunnelling in quantum physics, or analogous phenomena in statistical
physics.

2. An Example in Complex Variable Theory

In order to illustrate the basic ideas behind instanton techniques, let us

consider the one-dimensional integral

(3)

it may be considered as the partition function of a zero-dimensional field

theory (with real field ~ defined at a single point - the integral on ~
represents the statistical average). We want to show that Z has an imaginary
part for g <0, and to calculate it.

For the "physical" case g> 0, it is understood that the integral runs
from - 00 to 00 and Z(g) is real. By expansion in g within the integral it is
then straightforward to obtain the perturbation expansion
_ ~ K K 1 ~ 4K
Z(g) L (-g) CD -. jd¢' ~ exp-!¢'2
K=O K. -00
(4)
~ K r::! r(2K+
L.(-g) • ..,2. r K+l
K=O

by the substitution t = _!~2 and the definition of Euler's gamma-function.

This is the analogue of the Feynman graph expansion. It is clear that the
perturbation series (4) has zero radius of convergence and the function
Z(g) therefore has a singularity at g = O.

In order to elucidate the nature of this singularity it is useful to

consider Z for complex values of g. Clearly the integral I~ in (3) converges
whenever Reg > 0 and the function Z is defined as an analytic function of g
in the right hand plane. By the standard method of rotating the integration
contour in the ¢' plane, we can extend this definition of Z to all complex
values of g. In particular for g < 0 the two cases argg = ±rr are defined

105
by contour integrals rotated by +rr/4 respectively (so that g~4 remains
positive and the integral always converges). These two contours of ~
integration are the complex conjugate of one-another and thus Z(g) has
equal and opposite imaginary parts for the cases argg = ±rr, indicating that
it is analytic in the g plane cut from -ooto O.

We can now proceed to evaluate the imaginar~ parts for small (negative) g
by the method of steepest descents [See e.g.12J. This consists of finding
the saddle points with respect to ~ of the function

(5)

and the lines in the complex ~ plane of constant (in this case zero)
imaginary part of~. The contours of integration can then be distorted to
follow the path of steepest descent along these lines. For the case g < 0,
the saddle points are:

(6)

They are shown together with lines of zero imaginary part in Fig.l

... 9 =]t
arg
,
\

IL.. _ _ _ _ _ _ _ _ _ _ _ _ ..JI
r----- ------,
,J ,,
I
\

,
I
~

arg g =-yt
(ig.l Saddle points and lines of constant imaginary part of the function ,.
Eq.5) for g < O. The integration contours of steepest descent are shown
as dotted lines

Since the integrand is entire, we can distort the original contours at

angles ±rr/4 into the steepest descent contours shown in Fig.l. Clearly the
part of the integral between the saddle points±~ is purely real; the
largest contribution occurs near ~O, and this pirt yields the usual
perturbation expansion of Z as in (4). The imaginary part has its largest
contribution from the saddle points ±~. After the change of variables
~ = ~ + iT, Taylor expansion in T yielas
c

106
 1m Z (g)
arg g = 11

= -/11 exp 4~ [1 + O(g)] (7)

(Note that since g < 0 this is indeed exponentially small for small g).

We discuss the generalisation of this kind of calculation to field theories

in the next section and complete this one by outlining how the result (7)
can be used to obtain the behaviour of late terms in the series (4). The
function Z(g) is analytic in the g plane cut from -~ to O. We can therefore
write a dispersion relation

Z(g) = _l-S~
211i g' -g
d'
g
C
where C is the contour shown in Fig.2.

Fig.2 Dispersion relation contour C used to obtain high order estimates

The contour at infinity may be discarded and the real parts of Z cancel
between the contours -~ + iO to 0 and 0 to -~ -iO. Therefore we have

Z( )
g
= !JO
11
~mz(g')
. g' -g
dg' (8)
-40+,0
Expanding this integral in g and using the result (7) yields

Z(g) - ygK(;;) !dg' (g,)-K-l exp 1/4g' [l+O(g')]

K=O ~ -~ K [ -1 ]
!(_g)K 1 r(K).4 .1 + O(K ) (9)
K=O Tn

107
We have used the change of variables t =-1/4g to obtain again the Euler gamma
function. Note particularly that the (uncalculated) corrections of order g'
produce a f(K-l) instead of a r(K) and are hence negligible for large K. The
reader may verify by Stirling's formula that expression (9) agrees with the
form for large K of the exact series (4).

To summarise, these calculations show that Z(g) has an imaginary part for
g < O. For small g, the value of ImZ(g) is determined by a non-trivial
saddle point,~ , of the functionl+(Eq.5). The asymptotic behaviour of the
perturbation s~ries in g can then be obtained by means of a dispersion
relation in g.

3. Inst:'antons, and Field Theory Calculations

It is clearly desirable to be able to extend the above calculations to

the partition function of a field theory in d space dimensions,

Z (g) !D<f' exp -~W (10)

!D<f' exp -tt;(<f')

Hamiltonian 'W- is as in (1), for example, and *

where !D<f' denotes the functional integral over all field configurations. The
is the free Hamiltonian
(g = 0), so that Z is normalised by Z(O) = 1. °Unfortunately, we do not have
sufficient control of steepest descent methods in function space to follow
completely the calculation of section 2. However, we can extend sufficient
key components of that calculation to enable us to obtain the analogue of
(7) for many field theories.

The first key component is the analogue of the saddle point,<f'c' (Eq.6)
from which the imaginary part receives its dominant contribution. The equi-
valent in field theory is a solution of o~/o<f'= 0 i.e. a solution ~ (x) of
the classical field equation (2). From the complex variable analy~is, it is
clear that the trivial solution ¢(x) = 0 will not do. We need a non-trivial
solution, the instanton, ~ (x). We can then formally follow Eq.(7).
Writing <f'(x) = ~ (x) + ~ {x) in the numerator and expanding U(<f') =~(~ ) +
~m¢ +0(~3), whefe 1Vl.= 82W8<f'2(<f'c)' we obtain c

ImZ(g) ~ exp -~(<f'c)·

!D¢ exp-n~~ I ~
!D<f' exp~!!<f'~o~ l+O(g~
(g<O)

= exp -+l-(<f'c)-[~:~mJ -!. [1 + O(g)] (11)

For the Hamiltonian (l),~(<f' ) is always of the form -l~ag) (a = constant)

and the determinants of the ~ifferential operators m and mare obtained
by expanding the fields ~ and <f' in terms of their respectivg eigenfunctions.
The corrections of order g come from the neglected anharmonic terms 0($3) in
the numerator, as in (7).

Of course the non-linear equation (2) has many non-trivial solutions and
we must select the appropriate solution <f' (x) for expression (11). Since
1m Z(g) is always exponentially small in ~, we obtain the leading behaviour
of ImZ for small g by taking the non-trivial solution with minimum ~(<f'c)
(and we thereby assume it is the first saddle point we meet in a proper
steepest descent calculation). In particular,~(~ ) must be finite, so
that ~ (x) ~ 0 aslxl ~ ~ in order for the integralc(l) to converge, i.e. for
large fxl the field <f' is constrained to the "metastable" value <f' = O.

108
One may further show [13J that for large classes of Hamiltonians, including
(1) for d < 4, the solutions with minimum ~are radially symmetric. In this
case the field equation (2) becomes an ordinary differential equation

(12)

where

(13)

The form of V for m2 > 0, g < 0 is shown in Fig.3 •

N

Fig.3 Potential V for the motion in one dimension (¢') described by

equations (12) and (13)

Clearly (12) can be interpreted as an equation of motion in onedimen~ion ¢'

with "time" variable r, damping depending on d¢'/dr, and in potential V.
One can anticipate a solution starting at some value ~(O) =¢'o with sufficient
potential energy to overcome the damping and climb to ¢'= 0 at r = ~ . This
is the instanton solution required because other solutions starting at a
larger value of ¢' and performing two oscillations before coming to rest at
¢' = 0 will have a greater value of ~ . (This argument clearly breaks down
if the damping is too great and in fact the case d = 4 requires special
treatment). In many cases the instanton solution cannot be obtained analytic-
ally in terms of simple functions, but it is clear from the above discussion
that it can always be constructed numerically.

Having identified the relevant classical solution, we turn to some techni-

cal problems in evaluating the determinants in (11). F or the Hamiltonian
(1), the differential operator m
takes the form

m = -V 2 + m2 + 3g¢'2 (x)
c
(g < 0) (14)

since ¢' (x) is localised about x = 0 and vanishes at infinity, the calculat-
ion of Ehe determinant of~is equivalent to finding the eigenvalues (whose
product gives the determinant) in a quantum mechanical problem with potential
V = m2 + 3g¢'2(x) whose qualitative form is shown in Fig.4.
c

109
 v

Fig.4 Qualitative form of the potential whose eigenvalues give det]R in (14)

Clearly V is an attractive potential and one must pay particular attention

to the possible existence of bound states which have eigenvalue < O. The
following analysis resolves this problem in a very simple way. 'We have
remarked that ~ (x) is radially symmetric about x = O. Clearly, by the
translation inv~riance of the original Hamiltonian, the function ~ (x - x )
is also a solution of the field equations with the same value of ~c for aRy
fixed x; it just corresponds to an instanton centred at x. Now if we take
d/dXP ~p = 1,2, .•• ,d) of eq. (2), obeyed by ~ (x ~ x ), we 8btain
o c 0

+ 3g~2)
c
o

Therefore the differential operator

d~ /dX~ with zero eigenvalue.
m has a d-fold degenerate eigenfunction
This eigenfunction transforms under the ~ = 1
re~resentation of the rotation group Oed); one must therefore expect that
for the attractive potential V in Fig.4 there is at least one ~ = 0 eigen-
function with negative eigenvalue. Now, for negative eigenvalues oflYt
the steepest descent contours of integration in (11) for each of the
corresponding eigenmodes must branch into the complex plane, in presumably
the way indicated in Fig.l. I t is precisely this imaginary "D~" which
produces the imaginary part we wish to calculate. Thus there is a further
constraint on ~ in these calculations; it should produce a differential
operator~whigh has one and only one negative-energy bound state. In fact
this condition is probably automatically satisfied in most cases by the
solution with minimum ~because the solutions with higher ~ correspond to
larger values of ~ and hence more and deeper potential wells in V.

Thus the existence of one bound state of ntis precisely what ensures we
are calculating the imaginary part of the partition function Z.

Other problems remain for the evaluation of the determinant of nt.

We have recognised the existence of zero modes inlYl. These exist for every
continuous symmetry of m broken by ~. (They are Goldstone modes of a
spontaneously broken continuous symmerry). For the corresponding mode
integration in (11), the Gaussian approximation there is inadequate because

110
the 0(~2) terms vanish, and in fact we have to integrate °y.~ctly over these
modes. This can be done by the method of collective coordinaLes ~,14]
The basic idea is to take the centre Xo of the instanton as an integration
variable, writing
Hx) = ~c
(x-xo La ~n (x-x0 )
) +n
n
where ~ are (normalised) eigenfunctions of YVl,so that
n
a~
dHx) = dxIJo • ~--[l+o(/g)] + Lda c/' (x-x.,)
axo n n
Since ac/'d ax~ is the zero mode, we can exclude it from the sum on eigenmodes
~ and still span the space of all modes by varying x • If the eigenfunctions
~n are normalised one picks up simply a Jacobian fact8r:
n
'l,p12
Dc/' .. d? TCda . [I(~p lax)J [l+O(g)]
o nco
(15)

The integrals on a now give the determinant with the zero modes removed and
the integral on x nrestores the original translation invariance of the theory.
In particular foro the partition function Z, the integrand is independent of
x and one obtains simply an extensive factor proportional to the volume of
tRe system. (In fact the effect of a dilute gas of instantons is to
produce exponentiation of the result (11), so that this extensive factor does
indeed appear in R.nZ as one should anticipate; c.f. the appearance of the
free energy density in (22).)

Collective coordinates must also be introduced if there is a contin-

uous internal symmetry (such as in the Heisenberg model) or a dilatation
invariance (such as in four dimensions).

There is a further problem in evaluating det1n in that the contribution

to it from modes with high angular momentum R. contains the usual ultraviolet
divergences of the Hamiltonian (I), which can be removed by conventional
renormalisation. We refer the reader to the literature for further details
on these technicalities [4,lO,l1,15,16,17J.

Let us complete this rather technical section by noting that the general
form for the imaginary part of any N-point correlation function for the
Hamiltonian (1) is

1m G(N) exp llag .(_g)"b • c [1 + O(g)J

(g < 0) (16)

Since ~c '" (-g) -!, the constant b is determined by the number s of spontane-
ously broken continuous symmetries (from the Jacobian factor such as (15»
and the number N of external fields according to

b Hs + N) (17)

The constant a comes from the classical calculation of ij(~ ) and is positive;
the hard work of calculating the determinant is required t8 evaluate c. We
remark finally that in four dimensions the form (16) is complicated by the
dilatation invariance which is broken by renormalisation.

111
4. Some Applications of High Order Estimates

The result (16) for the imaginary part can be used directly to obtain high
order estimates via the dispersion technique in (8). Following equation (9)
one obtains

(18)

This indicates the asymptotic nature of the Feynman graph expansion

anticipated by Dyson [18J.

The question naturally arises of how bestNjo use the information (18) to
reconstruct convergent approximations for G from the first few terms of
the Feynman graph expansion which have been calculated explicitly for any
given problem. The natural starting point is to use the Borel transform
technique based on the usual integral representation of the gamma function.
If one formally interchanges the sum and integral representation in (18),
one has

(19)

The sum on the right hand side has now finite radius of convergence [gt=l/aJ.
Provided the analytic continuation of the sum has no singularities on the
positive taxis, Pade methods or conformal map~~~g techniques may be used

note that for g > °

to obtain a unique convergent expression for G • It is instructive to
the oscillatory nature of the series (18) implies that
the singularity closest to the origin is indeed on the negative t axis (at
-l/ag). Indeed it has been proved [19J that in the anharmonic oscillator
in quantum mechanics, Pade-Borel resummation does converge to the exact
answer. Conversely when g < 0, the series (18) is not oscillatory and the
Pade-Borel method fails because there is a singularity in the region of t
integration (at -l/ag); this is just a reflection of the fact that the
perturbation theory is incomplete when exponentially small terms such as
(16) are present. The requirement that the perturbation series should
oscillate is very important for models of physics.

The coefficients a, band c in (16) have been calculated for many models
in quantum and statistical physics [8,10,11,ZOJ. SO far the most useful
application in statistical physics has been to the problem of phase transitions
in three dimensions. The coefficients a, band c have been obtained
numerically [ZlJ for various quantities arising in the field-theoretic re-
normalisation group approach to phase transitions. Resummation techniques
based on the Borel transform and using the information in (18) and (19) can
then be applied to the explicit sixth order calculations of Baker et al tzzJ .
The results [ZlJ for various critical exponents in Ising-like systems are
shown in Table 1; similar results exist for n-component Heisenberg-like
models [ZlJ.

High order estimates have also been made fors-expansions of critical ex-

dilatation invariance of the Hamiltonian (1) for ref=°

ponents in d = 4 - s dimensions [11, Houghton et al 20J. Because of the
and d = 4, much of
the calculations may be done analytically - for example the instanton solution
in this case is the function
<P-: (x) = j;f. 1 ~ x2

112
Technical problems have so far prohibited the calculation of the coefficient
c. The results for the ~-expansiRns of critical exponents of n-component
Heisenberg models, in the form LE (-a) r(K+b) are
4 + n/2 for n
a = 3/ (n + 8) b
(20)
5 + n/2 for I/v
wi th the appearance of explicit fourth order calculations [24J one should be
able to obtain another set of independent results with errors similar to
those in Table 1.

Table 1 Comparison of field theory results [Le Guil10u and Zinn-Justin 2lJ
~first column), series expansion results ~!3] (second column) and experiment
l24J (third column) for Ising-like critical exponents in three dimensions.

y: X'ex(T-T )-y 1. 2402 ± 0.0009 1 250+0·003 1.240 ± 0.002

c • -0.007

n: G(2) exq -2+n 0.0315 ± 0.0025 0.018 ± 0.015

v: E; ex (T-T )-v 0.6300 ± 0.0008 o 638+0 •002 0.625 ± 0.006

c • -0.008

5. Other ApElications

One of the earliest problems to which these techniques were applied was that
of the electron in a random potential [IJ. The problem is to find the
average density of states and conductivity for electrons in a random potential
Vex) with a probability distribution P{V}. For a Gaussian distribution
p{V} ex exp-i1d d xV 2 (x) , one may show that the average density of states feE)
for electrons of energy E is given in terms of the imaginary part of the
two point correlation function of an n-component LGW Hamiltonian (1), in the
limit n + O. The parameter m2 in (1) is given by the energy E, and the
couplin~ g by the width y of the potential distribution p{V}: in units in
which n/(electron mass) = 2 one has

-E g -!y (21)

Thus for a sensible Gaussian distribution with y>O, the coupling g is negative.
For low lying states with negative E it is trivial to read off the y depend-
ence of the imaginary part of G(2) from equations (16) and (17). We can
identify N = 2 and s = d-l Jacobian factors (for the translation modes and
n-l+-1 Goldstone modes of the O(n) symmetry) and hence write down the average
density of states in the form
1
= exp- ay y -(d-l)/2 • c [ l+O(y)J (22)

The E dependence which we really want can now be obtained by careful dimens-
ional analysis from equations (1) and (2) and the iden~ification d2I). For
example (kfcJr
/~s a length dimension, we have [E] = [LJ - [y] = [L] ( -4) so that
y /(-E) is dimensionless. The final result is '

const. (-E) d(5-d) / \ - (d+l) /2exp_ [const (-E) (4~d) /2 /yJ

(23)
• [1 + O(y/(-E) (4-d)/2)]

113
This is the result quoted in a recent paper by Cardy [26]. The powers of E
differ apparently from those in Zittartz and Langer rl]; I believe that (21)
is correct (For example the reader may check that ~has dimensions number/
unit volume/unit energy). Note finally that for d < 4 (21) is valid for any
distribution width yprovided we are deep enough in the band. Correspondingly,
the approximation breaks down if we increase E towards the mobility edge;
this is an extremely interesting strong-coupling problem to which the re-
normalisation group cannot be immediately applied because there is no available
infra-red stable fixed point [27J.

Finally I remark that the same technique was also applied by Langer [2J
to the problem of the singularity at a first order transition. In magnetic
lan§uage this involves the Hamiltonian (1) with an external field term
HId x~(x) added. For T < T (m2 < 0 in the mean-field approximation) the
problem is to find the natufe of the singularity of the free energy as a
function of H as H goes through zero (The coupling g is now held fixed).
The idea is that for H > 0 say, the configurations of the field t are cons-
trained to have the positive expectation value +M at infinity. If we
attempt to continue analytically this free energy as a function of H to
negative values of H, the continuation of the functional integral must be
performed for fixed boundary conditions on the field ~ • Thus for H < 0, the
fluctuations are still constrained to tend to the positive expectation value
at infinity. In this circumstance there is an instanton, which is convention-
ally called the critical droplet, which gives rise to an imaginary part of
this free energy analytically continued to negative H. The general form of
the result can be obtained from a~le droplet-model picture, with a
critical droplet of radius r - I -m2 'M/IHI. The small oscillations have
indeed a single negative ene~gy state. The final result for the imaginary
part of the free energy density is
2 2 2
1m'} "'" const IHI-(2-d+d )/2(g)-<2+d+d )/4(-m2) (6+3d+2d )/4
( H-+O-)
• exp - [cons t (-m2 ) (2d+l) /21H 1- (d-l) (g) -(d+l) /2J

• (1 + O(g/(-m2) (4-d)/2) ) (24)

The factors in the first line are from2 the Jacobian and the eigenvalue of
the negative energy bound state of O(H). A discussion of the lon~ wave-
length singularities occurring in det~ can be found in reference L2].
The form of the singularity in '} as H -+ 0+ (Le. in the stable state) is
obtained by a dispersion relation in H as in equation (8). Since a range of
H contribute in this dispersion relation, "virtual" droplets of many scale
sizes contribute to this singularity, and this is a problem which one may
attempt to tackle by lattice renormalisation group methods [28]. The strength
of the instanton calculation of the imaginary part (24) is that the single
length scale r of the critical droplet dominates the problem.
o
Thus the form of the essential singularity on the coexistence curve can be
elucidated; the interpretation of the imaginary part as a decay rate of a
metastable state requires further clarification.

ACKNOWLEDGEMENTS

I am very grateful to M.Stone and to all colleagues at Southampton for

illuminating discussions.

114
REFERENCES

1. J.Zittartz and J.S.Langer: Phys. Rev. 148, 741 (1966)

2. J.S.Langer: Ann.Phys. 41, 108 (1967)
3. A.A.Belavin et al: Phy~ Lett. 59B, 85 (1975)
4. G. 't Hooft: Phys. Rev. Lett.37~ (1976) ; Phys.Rev. D14,3432 (1976)
5. A.A.Belavin and A.M.Polyakov:--Nordita preprint 77/1
6. C.G.Callan, R.Dashen and D.J.Gross; Princeton preprint (1977);
S.Colernan: Harvard preprint and references therein (1978)
7. M.B.Voloshin, I.Yu Kobzarev and L.B.Okun: Sov.J. Nucl.Phys. 20, 644
(1975); S.Coleman: Phys. Rev. DIS, 2929 (1977); C.Callan an~S.Colernan:
Phys. Rev. D16,1762 (1977); M.Stone: Phys Lett. 67B, IB6 (1977);
H.J.Katz: Phys Rev. D17, 1056 (1978) -
8. C.M.Bender and T.T.WU:- Phys. Rev. 184,1231 (1976); Phys. Rev. Lett.~,
461 (1971); Phys. Rev. D7, 1620 (1973)
9. C.S.Lam: Nuovo Cim. 55A,258 (1968)
10. L.N.Lipatov: Sov. Phys. JETP 44, 1055 (1976) and~, 216 (197n
11. E.Brezin, J.C. Le Guillou and ~Zinn-Justin: Phys. Rev. DIS, 1544 and
1558 (1977); E. Brezin: Proc. European Conference on Particle Physics,
Bucharest, and references therein (1977)
12. E.T.Copson: Theory of Functions of a Complex Variable (Oxford University
Press) p.330
13. S.Colernan, V.Glaser and A.Martin: Comm. Math. Phys. 58, 211 (1978)
14. J-L.Gervais and B.Sakita: Phys. Rev. Dll, 2943 (1975);-N.H.Christ and
T.D.Lee: Phys Rev. D12, 1606 (1975) -
15. J.Zinn-Justin: Carg~ Lectures 1977; G.Parisi: Cargese Lectures 1977;
R.Jackiw: Rev. Mod.Phys. 49, 681 (1977)
16. S.Chadha et al: Phys. Lett. 72B, 103 (1977)
17. A.J.McKane and D.J.Wallace: ~thampton preprint THEP 77/78-18
18. F.J.Dyson: Phys. Rev. 85, 631 (1952)
19. S.Graffi, V.Grecchi an~B.Simon: Phys.Lett. 32B, 631 (1970)
20. G.Parisi: Phys. Lett. u6B, 167 (1977); E.Brezin, G.Parisi and J.Zinn-
Justin: Phys. Rev. D16, 408 (1977); C.ltzykson, G.Parisi and J-B.
Zuber: Phys. Rev. D16, 996 (1977); A. Houghton, J.S.Reeve and D.J.
Wallace: Phys. Rev:-B18, 2956 (1978); J.C.Co11ins and D.E. Soper:
Ann. Phys. (NY) 112,-z09 (1978)
21. J.C.Le Guil10u a~J.Zinn-Justin: Phys. Rev. Lett. ~, 95 (1977);
G. Parisi and E. Brezin, to be published
22. G.A.Baker et a1: Phys. Rev. Lett. 36, 1351 (1976); B.G.Nicke1, to be
published --
23. W.J.Camp et a1: Phys. Rev. B14, 3990 (1976)
24. R.F.Chang et a1: Phys. Rev.-rett. 37, 1481 (1976); R.F. Chang, H.Burstyn
and J.V.Sengers: Maryland preprint--
25. F.M.Dittes, Yu.A.Kubyushin and O.V.Tarasov: Dubna preprint 1977
26. J.L.Cardy: Santa Barbara preprint UCSA TH-1 1978
27. D.J.Thou1ess: J.Phys. C: Solid State Phys. 8, 1803 (1975); M.V.
Sadovskii: Sov. Phys. JETP 43, 1008 (1976) -
28. W.K1ein, D.J.Wa11ace and R.K:P.Zia: Phys.Rev. Lett. 1I, 639 (1976)

115
The Theory of Structural Phase Transitions:
Cluster Walls and Phonons
A.D. Bruce
Department of Physics, University of Edinburgh
Edinburgh EH14 4AS, Scotland

Abstract
The essential problems posed by systems undergoing a structural phase trans-
ition are discussed in the context of a simple double-well potential model.
The traditional conceptions of the contrasting behaviour at the model's
order-disorder and displacive limits are reviewed, and set alongside a con-
temporary view of the displacive system, which assigns a central role to the
formation of precursor-clusters of short range order, displaying a two-time-
scale, order-disorder behaviour. An assessment is made of the extent to
which this view is substantiated, directly or indirectly, by dynamic and
static theory (analytic, Renormalisation Group and molecular dynamics methods),
by experiment ("central peaks") and speculative analogy (precursor effects
at lock-in phase transitions).

1. Introducti on
Systems undergoing structural phase transitions are essentially and inescap-
ably anharmonic. This principle is established by the most rudimentary
phenomenology of phase transitions, and is expressed in the essential
ingredients of any microscopic model. Nevertheless, until relatively
recently, the full subtlety of its consequences has tended to be lost in the
inadequacies of Landau phenomenology, and the inconsistencies of the class-
ical approximation schemes traditionally invoked in the solution of micro-
scopic models.
Much of the recent activity is attributable to the motivating impact of
two key experiments [1 ,2J , which revealed the need for techniques specifi-
cally designed to accommodate the effects of strong non-linearity [3,4J.
Firstly, the observation [2] of non-classical order-parameter behaviour
near the 105 K structural phase transition (s.p.t.) in SrTi03 initiated
interest in the distinctive problems of the critical regime, and motivated
extensive application of Renormalisation Group ( R.G.) techniques. Though
employing a conventional (harmonic) basis, these methods provide a prescript-
ion for the systematic treatment of the non-linear coupling among critical
fluctuations: their explicit predictions for the universal quantities
(chiefly exponents) characterising (principally) static [4,5J and (less
extensively) dynamic [6] behaviour, near the critical point, represent the
most clearly-identifiable successes of recent years.
These techniques, however, have (as yet) offered little to illuminate the
underlying character of the spectrum of excitations promoting the (calculable
116
but opaque) deviations from classical exponent values. In revealing an un-
expected double-peaked structure in the soft mode spectral function, above
the critical temperature in SrTi03, the second key experiment [1] focussed
attention on this issue, thereby motivating a series of studies, probing the
dynamical consequences of strong non-linearity - either by direct computer
simulation [7,8], or by analytic techniques shunning the customary assumpt-
ions of perturbation theory about a phonon basis [3,8-12]. Partly because
the productive questions are less evident in this area, and partly because
those questions that are evident are also hard to answer, the success of
these efforts has, pernaps, been more muted than that of exponent-generat-
ing R.G. calculations. Nevertheless, there has emerged a qua1itative1y-
compelling (if, as yet, largely unsubstantiated) picture of the grass-roots
"physics" of systems undergoing s.p.t. 's - a picture which I shall develop
here, in the context of the ubiquitous double-well potential model.

2. Model and 8ackground

In keeping with the bulk of the literature I will discuss the properties of
the simple d-dimensiona1 model (Fig.1) defined by the classical Hamiltonian
H =T + V
n.n.
V =~ E (u(~) - u(~,))2 + EV (u(~)) (1 )
~ ~,~' ~ s

describing the behaviour of a set of atoms of mass m, each located in its

own double-well potential

7f:<0 (2)

and coupled to its 2d immediate neighbours by harmonic springs. It is evi-

dent that the model's impressive array of simplifying features (the supposed
perfection of the crystal lattice, the rigidity of the sub1attice setting
up the local potential V , the short range forces, and the scalar character
of the coordinates, ref1@cting a uniaxial anisotropy) limit its relevance to
real (dirty, compressible, frequently ferroelectric and mu1ti-component-
order-parameter) systems. Nevertheless the model embodies the one key
feature common (in some form) to all systems undergoing s.p.t's. - the local
double-well potential which (for d > 1 : c.f. Section 6) may be shown [13]
to constitute a sufficient condition for a phase transition, at a non-zero
Tc ' to a homogeneously-ordered phase (u =<u(~» # 0)
The model offers, moreover, a unified context in which to examine the
behaviour of the two limiting ("order-disorder" and "disp1acive") classes of
systems undergoing s.p.t.'s, which traditionally, and perhaps unfortunately,
have tended to be treated in isolation from one another. This classification
may be made according to the ratio

(3)

of the two energies char~cteristic of (1) - the depth of the local potential
well Ew = IVs(uo)1 = 17f:1 /48, and a representative bond energy

117
E = dCuo2 = C IAI lB. Si nce. in the d=3 case (to be unders tood in the
a§sence of any contrary indication) the ratio kBTc/EB is of order unity.
and only mildly model-dependent. the ratio h closely reflects the parameter-
dependence of the conceptually more transparent quantity

(4)

In the order-disorder limit h » 1 it is then clear that. except at temper-

atur~T » Tc. each atom will be predominantly localised in the neighbour-
hood of one or other of the well minima (± Uo = ± ~). as envisaged in
the Ising model; within the harmonic approximation one finds that all modes
of vibration about the high-temperature (h.t.) phase equilibrium poSTtions
(h.t. phonons) are unstable and the inescapably non-linear character of the
dynamics is immediately apparent.
In contrast. in the displacive limit. h « 1. the critical thermal energy
is large in comparison with the local barrier height; within the harmonic
approximation only a small group of long-wavelength h.t. phonons are unstable
andaquasi-harmonic view of the dynamics seems tenable.
Remarkably. these fundamental differences are not reflected in the
majority of exactly-established properties of (1). Specifically it is well-
known that there are ("universal") features of the model's static critical
behaviour - principally the critical exponents - which are quite independent
of the system's location on the order-disorder/displacive spectrum. and
which therefore necessarily coincide with those of the Ising model of ferro-
magnetism (now known with great accuracy). Similarly. it has been shown
[14] that there are universal features of the dynamic behaviour. which are
those of a model with a scalar non-conserved order parameter. and conserved
energy ("model C" in the nomenclature of [6.15J). In particular the relax-
ation time [6]

I
)

f\
IV
('-. .a/\
~
VV V\J \J

~ Schematic representation of the d=l form of the model (1)

118
 -1
T :: X(O) ~I (5)
u ~ ~=O

characterising the critical slowing down displayed by the frequency-depend-

dent order parameter response function X(~) satisfies [14]

T- l ;:: 0 2 t 2v +a. (6)

u TK -

where t is the reduced temperature, DT is the thermal diffusivity whose

inverse has the critical behaviour of the specific heat (c _ t- a ) , and K
is the inverse of the correlation length (s _ t-V). v
This result is explicit, but conceptually opaque, giving no hint as to
the physical character of the collective mode in which the critical slowing
down occurs. Traditionally, the critical slowing down in order-disorder
and displacive systems has been envisaged in quite different terms. The
traditional picture of the order-disorder system (Section 3) emerges
immediately from a recognition of the strongly non-linear character of the
deep-well limit; qualitatively this picture is not in doubt. The trad-
itional picture of the displacive system has emerged from classical approx-
imation schemes (Section 4): its basis is therefore less secure - the
classical schemes are known to yield incorrect critical exponents - and,
indeed, it has come to be replaced (Section 5) by a picture (that alluded
to in Section 1) which has much in common with the traditional conception
of the order-disorder system, and which illuminates the origins of univers-
ality.

3. Order-disorder Systems: A Qualitative Discussion

In the order-disorder (deep-well) limit it is clear that the fundamental
dynamical processes (i .e. those that maintain/express the thermodynamic
equilibrium of the high-temperature phase) are the collective analogues of
single-particle (thermally activated) hopping across or (in a quantum theory)
tunnelling through the large potential barrier separating the two wells. It
seems little more than a truism to assert that these processes represent
the motion of domain walls, whereupon it becomes natural to interpret the
relaxation time (6) for the order parameter degrees of freedom as the time
taken for a domain wall, with "diffusion constant" DD.W.::: DT [161 to diffuse
the distance s = l/K.
The established treatments of the order-disorder regime - the (classical)
Glauber model [17] and the (quantum-mechanical) pseudo-spin model [18,19J
focus exclusively on these "inter-well" degrees of freedom, which must
evidently dominate the critical dynamics in general and, in particular, the
spectral response S(~) = kT 1m X(~)/~ in which they will manifest themselves
as a quasi-elastic peak of width l/Tu' However, these treatments do not,
in the first instance, accommodate the additional "intra-well" degrees of
freedom, possessed by the Hamiltonian (1), which will be reflected in
additional collective modes - the phonons of a quasi-statically-disordered
lattice. (A direct "independent site approximation" (i .s.a.) to the h» 1
limit of (1) does some justice to these degrees of freedom, but the non-
ergodic character of the isolated classical double-well, to whose properties

119
the problem is reduced in i.s.a., leads to an artificially narrow-o-
function! - representation of the "inter-well" contribution to S(S'l) [20J).
The spectral function S(S'l) should therefore exhibit additional (resonant)
response at the frequency of this collective bottom-of-the-well motion,
which should itself remain largely unaffected by the approach to criticality.
In effect, then, the order parameter degrees of freedom have not one
but two distinct time scales - one long (and critical) time scale associated
wi th coll ective inter-well moti on ("domain wall di ffusi on") and the other
generally much shorter time scale associated with collective intra-well
motion (low temperature. t.t., phonons). Indeed, the existence of both low-
frequency (diffusive) and high frequency (resonant) components in the spectral
response of a "single mode" (c.f. the schematic Fig.2) is an experimentally
well-established (though not invariably appreciated) feature of order-disorder
systems, a prime example being NaN02 where dielectric measurements reveal a
(critical) central response at frequencies lower than 109 Hz. [21J in
addition to a non-critical response at the frequencies (10 12 Hz.) of infra-
red phonons [22].

4. Displacive Systems: Classical Approximation Schemes and the SoftPhonon

The history of the dynamical properties of displacive systems is dominated on
both theoretical and experimental fronts by the concept of the soft phonon
mode [23J. Thus, revelations of the idea's limitations have, if anything,
come as more of a cultural shock to the s.p.t. community than the discovery
of non-classical behaviour in equilibrium properties. Both the idea itself,
and some pOinters to its limitations, emerge from the two classical decoupling
schemes which I shall now briefly outline, beginning with the independent
site approximation [20,5].
The essential feature of the i.s.a. is the neglect of inter-site correl-
ation - and thence of short range order (s.r.o). Explicitly, one makes

~i9.2 Schematic representation of a spectral function S(S'l) with both low-

requency (diffusive) and high-frequency (resonant) components

120
the replacement

u(R.)u(R.') + u(R.)<u(R.'»i.s. = u(R.)ui.s. (7)

in the Hamiltonian (1) when the T > Tc i .s. (u i .s. = 0) response function may
be written as

x(n) = xi.s·(n) _ Xo(n) (8)

1 - 4dCXo(n)
where Xa(n) is the analytically-tractable [20] response function for an
isolated atom in the effective potential

vi.S·(u) = ~u) + 2dCu 2 (9)

In the displacive limit C/I~I » 1, so that the effective potential Vi . s .

has only a single minimum, and the strongly non-linear character of the
underlying system is suppressed. The dynamic response function xi.s.(n)
is then found to have a single pole (and the spectral function S(n) a single
peak) located by the ("soft mode") frequency [20J

(10)

T=15 Tc=2.5

(T-TelITe

Fig.3 The imaginary part of x(n) as a function of frequency (left) and the
squares of soft mode frequencies as a function of reduced temperature, in
the i.s.a. for the model (1), in the displacive limit [20J.

121
whose approach to zero as T ->- Tci.s ::: 4C11ti/B (c.f. Fig.3) is the dynamic
signal of the divergence in x(n=o), and is the explicit realisation of a
critical slowing down with relaxation time (y =~(n)1 )
n=o
(11 )

Below the phase transition (not shown in Fig.3) the frequency of this mode
(then a ~.t. phonon) hardens from zero, under the stabilising influence of
order parameter growth.
A qualitatively similar picture emerges from the "self-consistent phonon",
alias "independent mode", approximation (i.m.a.) in which variational tech-
niques (see, e.g. [24J) are used to establish the optimum harmonic represent-
ation for the equilibrium properties of (1), thus eliminating explicitly non-
linear effects by fiat. In effect (in characterising the h.t. phase) one
makes the replacement
u4(~) ->- 3u2(~)<u2(~»i.m. (12)

in (1); a self-consistent treatment of the resulting harmonic equations of

motion then agaiQ leads to a spectral response with a.soft phonon resonance
at a frequency w,·m:, te2ding to zero (Fig.4) as l hT c,·m·:::2.6 CI1tI/B. (In
i.m.a., however, (w,·m.) vanishes as (T-T c,·m.)2, and the soft phonon
,is undamped in the simplest version of the theory; more importantly, see
also Section 5.2.1.)

Ti .m. Ti .m. T
c a

~i9.4 The qualitative temperature-dependence Qf the square of the soft mode

requency in the i.m.a. for the model (1); To,·m. locates the (first order)
phase transition

122
 Both approximation schemes thus imply (or assume) the same underlying
view of the T > Tc dynamics: each atom exhibits an essentially harmonic
motion about its high-temperature phase equilibrium position (the top of its
well), while the collective mode in which the critical slowing down resides
is a (generally) damped long wavelength "well-to-well", h.t. phonon.
There can be no doubt that this picture gives a qualitatively correct
account of the gross features of the behaviour of displacive systems, of
which SrTi03 is the accepted prototype: the experimental evidence support-
ing the soft phonon concept is compellingly extensive (c.f. Fig.5, and [25J)
altogether dwarfing the tenuous evidence for its deficiencies. It is,
nevertheless, to the latter that the remainder of this paper is devoted.

2.5,--~-~-_--,-_---, _ _ _--,

2.0

~ I,~

J:

N
.-
Fig.5 The squares of soft mode
frequencies in SrTi03' as determined
by neutron scattering techniques [25J

I!)() 200 250 300

TEMPERATURE oK

5. Displacive Systems: Beyond Classical Theory

5.1 A Contemporary View
In the last few years the soft phonon picture sketched above has been called
in question, not merely in detail ("With what exponent does the soft phonon
frequency vanish as T ..,. Tc?") but in its very essence ("Does a soft phonon
mode frequency vanish at Te?"). From these questioningstfi"ere has emerged
a new view - I believe it to be the consensus of those whose names appear in
the references - of the essential physics of displacive systems near their
critical points: this view I shall now describe in outline, before assembling
the evidence in which it originated, and through which it may be refined and
assessed.
The essential idea is that, with the onset of criticality, the growth in
short range order promotes a crossover from a regime in which the collective
behaviour has the classical displacive form, to a regime in which the collect-
ive behaviour displays features that are better described in the language

123
traditiona 11y reserved for order di sorder sys terns. Thi s "di spl acive-order-
disorder" (or"phonon-Ising" [26]). crossover, occurring at some temperature TI'
is signalled by the appearance of "microdomains" or "clusters" - groups of
atoms whose motion is strongly correlated - within which the average atomic
displacement ("cluster order parameter") is non-zero: each atom then
oscillates about the equilibriUIP ...position set by the value of the "order para-
meter" of the cluster of which it is a member, this position of local equilibrium
itself changing sign as a cluster wall diffuses by.
Thus, while the collective behaviour within the displacive regime has but
a single time scale, set by the soft high-temperature phonon (giving a
single region of possibly overdamped response in S(n)) the collective
behaviour in the regime Tc < T ~ TI has the two time scales characteristic
of an order-disorder system. The crHical time scale is set by the diffusion
of the walls of the clusters, giving rise to a central response in S(n) -
the s.r.o.-induced dynamic precursor of the long-range-order (£.r.o)-induced
static (Bragg) peak to appear below Tc' The second, non critical, time
scale is set by the collective oscillations about the positions of local
equilibrium, giving rise to a (possibly overdamped) phonon side band in S(n) -
the residue of the high temperature phonon, and the natural precursor of the
low temperature phonon, into which it evolves without ever exhibiting a
complete softening.
Evidently this picture constitutes a graphic and aesthetically appealing
expression of universality - a perspective I shall develop, in discussing
its origins and assessing its validity.

5.2 The Theoretical Background

A balanced assessment of the theoretlcal basis for the picture I have describ-
ed must discriminate between the evidence that is purely circumstantial, and
the evidence that is explicit: the following discussion is lO'osely subdivid-
ed along these lines.
5.2.1 Circumstantial Evidence
The failure of the soft mode concept· is illuminated by the inadequacies of
the classical theories from which it emerges. The formal inconsistencies of
f
both the i .s.a. and the i.m.a., close enough to T and in systems with
dimensionality d < 4, are well known (see e.g. [5 ).
The i .s.a. fails because, though predicting a divergent correlation length,
it takes no account of correlations: its failure emphasizes the need to do
justice to the behaviour's growingly coherent - and therefore growingly non-
linear-character. (It will be recalled that it is in the assumption of in-
coherent, uncorrelated motion that the double-well character of the local
potential is lost.)
The i.m.a. fails because, though allowing for the growing coherence of the
motion, it requires that the increasing coherence be expressed simply through
a growing population of soft harmonic phonons: its failure emphasizes that
the behaviour of the (long wavelength) collective degrees of freedom is non
-linear in an essential way.
In fact, the i .m.a. is more informative than the discussion of Section 4

124
would suggest: when extended to the ordered phase. the i .m.a. is found to
imply that the complete softening of the high-temperature phonon mode is
preempted by a first order phase transition (see e.g. [24J. [5J) at a temp-
erature To,·m. > Tc' .m. (Fig.4). Though it is clear that this prediction is
wrong - there is no doubt that (1) exhibits a continuous phase transition -
it is not so clear that it is devoid of physical significance. In ~ffect
the i .m.a. analysis demonstrates that there exists a temperature To' .m.
below which it becomes better (in a variational sense) to describe the
behaviour in terms of a static homogeneous ordering. together with the corr-
esponding ~.t. phonons. than in terms of h.t. phonons (whose softening is.
at that stage. incomplete). It is then tempting [27.28] to regard T ' .m.
as a measure of the displacive-order-disorder crossover temperature ?c. below
which (in the picture sketched earlier) the (softening) high-temperature
phonon is indeed superseded by a (hardening) low-temperature phonon. Within
this interpretation the i .m.a: 's principal failure is that it identifies the
crossover temperature TI.~ To' .m. with the macroscopic ordering temperature
Tc. when in fact Tc < To' .m. . This failure it is natural to attribute to
the i.m.a.'s inability to accommodate the other feature of the order-disorder
regime - the strongly non-linear order parameter configurations associated
with cluster walls. against whose formation (one might surmise) a homo7
geneously ordered phase is actually unstable in the regime Tc < T < To' .m.
L12J. the consequent break up into microdomains destroying the ~.r.o. but
leaving the s.r.o. necessary to sustain the low temperature phonons.
A formal ~emonstration that the true transition temperature Tc does indeed
lie below To,·m. would lend further plausibility to these conjectures. (It
is known that the stability limit Tc' .m·represents a lower bound for Tr
[29.13J. the ratio Tc/Tci .m. tending to unity in the displacive limit ~[30.5J.)
A second. substantially richer. vein of circumstantial support for the
cluster picture is to be found in the properties of the analytically-tract-
able one-dimensional form of the Hamiltonian (1). which has been the subject
of a number of extensive studies [3.8.9J. Since the explicit results of these
studies will no doubt figure in the presentations of other speakers at this
Symposium I shall simply discuss their implications. briefly and qualitatively.
Perhaps the most immediately-instructive consequence of the analytic
tractability of the d=l model. in the displacive limit. is that exact solu-
tions to the non-linear equations of motion may be identified. anaiSnown to
constitute explicit realisations of the three excitations prominent in the
picture promoted in Section 5.1 - low and high temperature phonons. and
cluster walls bounding microdomains. These solutions have been shown to
constitute a successful phenomenology both of the low-temperature equilibrium
properties (exactly calculable by transfer-integral techniques [3]). and of
the spectral response S(n) (itself not analytically tractable. but calculable
by molecular dynamics methods [8J). which displays the expected features:
a softening h.t. phonon response evolves gradually (with reducing temperature)
into the response of a hardening ~.t. phonon. and a critically-narrowing cent-
ral component that can be traced to microdomain-induced s.r.o.
That such a crossover should occur at low enough temperature is. of
course. evident from the nature of the ground state of (1): thus. in a
sense. the principal open question is whether (or to what extent) the cross-
over occurs before (at temperatures higher than) the onset of ~.r.o .• at Tc -
a question which is interesting (non-trivial!) only in systems which (unlike
the d=l model; c.f. also [11]) possess ~.r.o. at a finite temperature.

125
 Nevertheless, the temperature TI locating the d=l crossover does deserve
attention. It has been characterised according to a number of conceptually
different but (in d=l) practically equivalent ways - as the temperature of
the i.m.a. 's first order phase transition (c.f. above), as the temperature
at which the specific heat exhibits its maximum I8], kink-excitations (domain
walls) become thermally unstable [1 OJ , and the single-particle distribution
functi on begi ns to exhi bit two peaks [28J. Though additi ona 1 criteri a may
seem less than essential, the following argument is, I think, illuminating.
As the temperature is lowered the correlation length of the d=l displacive
system grows (within the displacive regime) as

(13)

where a is the lattice spacing. While the growth in s will be expressed in

the appearance of clusters of atoms moving coherently, the motion of the
clusters will not reveal a distinctively Ising behaviour (and (12) will
remain valid) until the potential barrier Ew s seen by the cluster (during
the collective inter-well motion of its member atoms) becomes comparable
with the thermal energy kT = kTI' Evidently Ew,s - sEw/a, whence, using (13),
d 1111 3(2
kTI - B (14)
in accord with the results based upon the other criteria, cited above, and
with the crossover temperature located by d=l molecular dynamics [8J.
Since the form of this result is actuall~ directly implied by the homo-
geneity properties of the displacive limit 18] it would be dangerous to acc-
ord great significance to the apparent "success" of the criteria from which
it may be derived. Nevertheless, the argument presented above may be rigor-
ously grounded in the methods of the R.G., which provide a framework within
which, for any dimensionality d, a precise meaning may be attached to the
idea of a "cluster coordinate" and the "potential" it sees.
Cluster coordinates describe the collective behaviour of atoms within a
volume of linear dimension S, and are thus plausibly represented bylRthe local
degrees of freedom left after application of a R.G. transformation bf4] with
scale factor b = s/a; the potential seen by a cluster (of the physicat
system) may thus be identified with the potential experienced by a local co-
ordinate in the transformed system, with configurational energy Vb = nKbV'
In the limit in which criticality is aeproached the potential seen by a
cluster tends to a fixed point form, Vs ' with universal features: Figure 6
displays the form of V~ for various dimensionalities. (In d=l the fixed
point potential presumably consists of two infinitely sharp and deep Ising-
like wells.) A natural measure of their order-disorder/displacive character
is afforded by the fixed point value g* (c.f. (4» of the ratio of well to
critical thermal energy. In d=3, Wilson's approximate recursion formula
[4] gives g* ::: 0.55; for d=4-s, the result is g* = s/4 + 0(£2). (The fixed
point value h* of the ratio of well to bond energies displays similar behav-
iour - though the R.G. 's use of a spherical Brillouin Zone makes it difficult
to determine h* in a way that is entirely consistent with (3).)
These results should be viewed with a modicum of caution: it is not clear

126
 u

(i~.6 The fixed point form of the single particle potential [~

a ford=3.9 (b) ford=3 (c) ford=2

that the quantity g* will be found to be strictly universal (cut-off-independ-

ent) in calculations going beyond the levels of approximation inherent in
Wilson's recursion formula and the leading order €-expansion; neither is it
clear to what extent the universality of features of the fixed point con-
figurational energy will be reflected in universal features of the critical
dynamics.
Nevertheless these results do illuminate the nature and, potentially, the
extent of the displacive-order-disorder crossover. The behaviour - at the
very least, the equilibrium properties - of the long wavelength, "cluster",
degrees of freedom in the near critical system is determined by a fixed-
point potential which itself has some definite location on the order-disorder
displacive spectrum - a location that is independent of the system's micro-
scopic parameters (e.g. h or g) but which reflects, principally, those
essential features (here, simply the dimensionality) that are instrumental
in determining universal quantities. Thus, while the parameters g and h
form a crude basis for a displacive/order-disorder subdivision in the non-
critical regime of weak correlations (where, e.g., h determines the character
of the potential seen by an atom: c.f. (9)), close to criticality it seems
more plausible to base this subdivision on the fixed-point quantities g* and
h*, whose values incorporate the all-important effects of correlations.
The quantitative inferences to be drawn from these arguments are not yet
apparent; qualitatively, however, it seems safe to observe that the order-
disorder "content" of the near critical behaviour (as measured, say, by g*)
decreases from a large value in d=2 (an infinite value in d=l) to zero in
d=4, and is by no means insignificant in d=3.
This trend mirrors the dimensionality-dependence of the deviations of
critical exponents from their mean field (i .s.a.) values, making it tempting
to correlate the occurrence and extent of non-classical behaviour, with the
formation and order-disorder content of precursor clusters, and to visualise
the universality of the critical region as the universality of cluster
behaviour.

127
5.2.2 Direct Evidence
As witnessed by the brevity of this section, the direct theoretical evidence
for two-time-scale, order-disorder features in the near-critical behaviour
of displacive systems, is not extensive. Such evidence as there is falls
into two categories.
Firstly there are the results (arguably more appropriate to the
"experimental" Section below) of a number of illuminating molecular dynamics
investigations of the Hamiltonian (1) [71. These studies will be reviewed
elsewhere at this Symposium [31] and so 1 shall remark only that they provide
considerable support for the cluster picture - having, indeed, played a
major role in its development: specifically, calculations on a 2-dimensional
"displacive" system (h::; 0.13; g = 0.19) display, in the near-critical
wavevector-dependent spectral response (Fig.7) both an (incompletely-soften-
ed?) overdamped h.t. phonon, and a critical central component (shifted to
finite energy at finite wavevector?) which has been traced, quite explicitly,
to the dynamics of Ising-like clusters.
There have, secondly, been a number of studies of (3-dimensional) dis-
placive systems which, while employing a phonon basis, have attempted to
treat non-linearity with greater care than classical (or low-order perturb-
ation) theory. Infinite order perturbation theories (ad-hoc, but physically-
motivated in the calculations of [26J; systematlcally-controlled by the
dictates of the R.G. and the E-expansion in [14]) hove vindicated an earlier
conjecture, expressed in a number of different guises [32], that in systems
with the essential symmetry envisaged in the model (1), the critical slowing
down (and thus the central response in S(~)) will be dominated by a contri-
bution to the soft h.t. phonon self-energy that arises from non-linear
coupling to a thermal diffusion mode (c.f. Eq.6) built from some "ladder" series
of phonon diagrams. The R.G. studies described in [14] seemingly capture

- ( 3,1
50
--- (2.95,0.95)
-.- (2.85,0.85)
40 ----- (2.65,0.65)

30

20

~ The frequency- and wave-vector-

dependence of the spectral response of a
d=2 displacive model at t ::; 0.06, a~
determined by molecular dynamics [7J
0.2 0.4 0.6 0.8

128
6nl* thi s central response; however the cal cul ations reported in [21i] reveal
ot central and h.t. phonon responses, the former,at its first appearance,
being broader than the latter, but narrowing and growing in intensity with the
approach to Tc'
These calculations provide the most immediate analytic evidence for a
two-time-scale dynamics in displacive systems; however, they vindicate the
cluster picture itself, only to the extent that microdomain walls may be
regarded as the principal microscopic realisations of phonon-density (temp-
erature) fl uctuati ons, and the di ffusi on of "thermal fl uctuati ons" effect-
ively synonymous with the diffusion of domain walls (as suggested, implicitly,
in Section 3).
5.3 The Experimental Background
The experimental support for the cluster picture of displacive systems is
sparse, resting largely, and somewhat insecurely, upon the key inelastic
neutron-scattering study of SrTi03 [lJ referred to in Section 1. This experi-
ment revealed that the softening ~.t. phonon observed in earlier, less- 12
detailed, studies (Fig.5) in fact saturates at a finite frequency (~0.13xlO
Hz) at Tc; a central component in the spectral response appears some 50 K
above the phase transition, its intensity diverging with the approach to
criticality (Fig.8).
While these observations are in qualitative accord with the cluster
picture (and, indeed, constituted the major motivation for the theoretical
efforts reviewed above) it now seems unlikely that they can be understood
entirely within this framework. The principal problem lies in the frequency-
width of the central component - less than the instrumental resolution, both
in the original neutron-scattering experiments [11, and in subsequent investi-
gations employing a high-resolution back-scattering spect,ometer [33.]: even
on its first appearance the width is no greater than 2xlO Hz - a M8ssbauer
y-ray scattering experiment establishes an even smaller upper bound of 3xl06Hz.
[34] •
It seems highly unlikely [14,16] that such a narrow response does, in fact,
originate in the cluster dynamics of an ideal crystal: molecular dynamics
computations [7] and the calculations reported in [261 both suggest a central
component with a width not much different from that ol the soft phonon -
except, of course, for reduced temperatures t« 1, where critical slowing
down produces an arbitrarily narrow response. It is possible that a response
much narrower than that of the scalar (n=l), spatially-isotropic, models
investigated so far may arise in systems which, like SrTi03, have an n=3-
component order parameter (c.f. the pathological features of the critical
dynamics of such systems alluded to in ~5]), or, again like SrTi03, have a
highly-anisotropic soft-phonon dispersion (c.f. the conjectures in r36]).
Nevertheless, it seems likely that a large proportion of the central component
observed in SrTi03 reflects non-ideal crystal behaviour: perhaps the influ-
ence of a strain-induced ordering of a surface layer, setting in well above
the bulk Te [34J; perhaps the influence of mobile impurities [16] acting
as nucleatlon centres for clusters, and extending their life-time. The
latter seems the more attractive possibility (since it salvages the central
ideas propagated here!) and the more plausible, since the former mechanism
would not (in itself) account for the saturation of the phonon side band,
observed in the neutron scattering experiments [1].

129
 7~PO
350 SrTi0:3 1750

Q~(~H l-r25
300 T=llooK
1500
Tc ~99.5 ±o.3 OK
250 E O=4.9meV 1250

200 1000

150 750

100

50
-- 500

250
-!~--
0
-3.0 -2.0 -1.0 0
w(meVl

Fig.8 The soft mode seectral response in SrTi0 3 , as determined by neutron

scattering techniques ll]

It is arguable that this feature deserves as much attention as the central

peak itself, since it is indisputably a bulk phenomenon, and (unlike the
central component!) is apparently sample-independent. In this context it
is worth observing that, if the cluster mechanism is, at least partially,
responsible for the observed behaviour in SrTi0 3 (- note that the experi-
ments do not rule out the existence of an additlonal central component,
with a width compatible with the cluster picture -) the inelastic response
at Tc might reasonably be expected to display precursor traces of the two
(A and E) soft ~.t. phonons that evolve from the 3-fold-degenerate h.t. soft
mode. (c.f. Fig.5)

5.4 A Suggestive Analogy: the Lock-in Phase Transition

Stepping briefly outside the framework of the Hamiltonian (1), I wish,
finally, to make reference to a class of s.p.t. at which the elements of the
cluster picture are demonstrably present, even within Landau theory [37,3!D.
Consider the double-well potential model, modified by the inclusion of
additional, longer-range, interactions between atoms, such that the minimum
in the h.t. phonon dispersion occurs at some wave-vector qs' lying close to
a wave-vector q that is commensurate with the h.t. lattice
Je.g. qc = 2rr/a(1/3,0,0)) but shifted from it by some small amount t, so that
qs = 211/a(1/3+o o '0,0) is effectively incommensurate [39J with the h.t. lattice.
On cooling, this system will undergo a phase transition to an incommen-
surate phase (Fig.9), characterised by the spontaneous ordering of the
fourier coordinates of wavevectors ±qs: the physics of this s.p.t. is not
very different from that of the Hamiltonian (1).
On further cooling, however, such a system typically undergoes a further
transformation (Fig.9) to a phase where the ordering is characterised by
non-zero expectation values for the fourier coordinates of wavevectors ±qc,

130
 T
DI SORDERED

ORDERED
COMMENSURATE
Fig.9 Schematic phase diagram for a
system exhibiting an incommensurate
phase and a lock-in s~.t.

and thus is commensurate with the h.t. lattice: this "lock-in" s.p.t. has
some remarkable features, the most notable of which is that precursor effects
are inherent in strictl~ equilibrium properties, and hence are evident with-
in Landau theory [37,38j.
The Landau theory itself (though effectively one-dimensional) is not
trivial, however, because the nature of the incommensurate phase structure,
close to the lock-in temperature, TL' is not simple - at least, it does not
appear to be simple when viewed within the usual framework of an h.t. phonon-
basis: non-linearities ensure tha4 once there exists a distortion with an
incommensurate wavevector qc + t, a whole series of harmonics (with wave-
vectors qc - 28, qc + 38 .... ) are induced (so that t need not coincide with
80 l, the series being only slo~lly convergent near TL.

The Landau theory becomes tractable when the problem is viewed in direct
space [37,38J. Close to TL it transpires that the incommensurate structure
can best be described as consisting of a series of regions that already
possess a commensurate order (within these the equilibrium displacement
field <u(x» varies as cos(qs.x+¢l) separated by "domain walls" ("discommens-
urations" [37], "phase solitons" [38}) , which are perpendicular to the [100}-
direction, and at which there is a mismatch of the phases of the commensurate
ordering in the adjoining regions (the phase angle ¢ evolves through 2TI/3
within the domain wall): c.f. Fig.10. The concentration of domain walls
diminishes with the temperature, vanishing continuously at TL' where true
long-range (commensurate) ordering sets in.
The analogy with the situation envisaged in the cluster picture of (1)
is apparent, and is reflected also in the scattering properties [401. The
increase, with the approach to TL' of the mean separation between tne domain
walls is expressed as a decrease in the value of the parameter 0 controlling
the location of the Bragg-satellites (associated with the series of fourier
components with wavevectors qc+8,qc-2l ..... ), which thus converge upon the
commensurate-phase Bragg point qc. Thus, just as in the cluster picture of
(1), the growth of the scattering in the vicinity of the incipient Bragg
point (as the s.p.t. is approached from above) reflects a decreasing pop-
ulation of domain walls, and a consequent growth in the coherence of the
ordering of the precursor microdomains which they separate. In this case,

131
u(xl

<I>(x)

Fig.10 The spatial variation of the displacement field <u(x», and the
phase field ¢(x), in an incommensurate phase near a lock-in s.p.t.: the
dashed line shows the displacement field in the absence of the phase
soliton [37,38J

however, the domain walls are essential ingredients of the equilibrium con-
figuration in the h.t. phase, and the scattering to which they give rise is
elastic; in the cluster picture of (1), the domain walls form part of the
excitation spectrum of the h.t. phase, and hence issue in scattering that
is inelastic.
Lastly it is illuminating to notice that the lock-in s.p.t. occurs when
the ~.t., commensurate, phase becomes unstable against the creation of
domain walls, which destroy Lr.o: in a real sense a domain wall is the
"soft mode" for this phase transition.

6. Concluding Comments
The last observation prompts one final question which embodies, picturesquely,
if imprecisely, the central issues raised here: is a displacive structural
phase transition more properly regarded as an instability of the low-temper-
ature phase against the creation of a ~.r.o.-destroying domain wall, than as
an instability of the high-temperature phase against the condensation of a
~.r.o.-creating phonon? The answer to this question exhibits the dimension-
ality-deper.dence which we have already seen in other properties of the
Hamiltonian (1), and which, therefore, merits a final resume.
In one dimension the R.G. fixed-point potential lies at the order-disorder
end of the spectrum; two-time scales will be clearly identifiable in the near-
critical dynamic behaviour; deviations from classical theory are at their
largest; the transition temperature Tc=O does indeed reflect the fact that
(for any T>Tc=O) the ~.t. (ordered) phase is dynamically unstable against
domain-wall formation [9].
In two dimensions the fixed-point potential is still quite strongly

132
order-disorder; two time scales should be evident in the dynamic behaviour
in a range of reduced temperature reflecting the size of some characteristic
parameter such as h; critical exponents are strongly non-classical; it
seems possible that the idea of the s.p.t. as an instability against a
domain wall could be made the basis of as successful a phenomenology of the
transition temperature in the displacive case as it has in the order-disorder
limit (see, e.g. [41J).
In four dimensions the fixed point Hamiltonian is displacive ("gaussian");
a quasi-harmonic, single-time scale dynamics is expected; critical exponents
have their classical values; the s.p.t. is best viewed as an instability
against the soft h.t. phonon.
The stark differences between the low (1 and 2) and high (4) dimensional
cases presumably reflect, at heart, the dimensionality-dependence of cluster
topology. In d=l,2 clusters are "mutually exclusive": at anyone time a
given region of the crystal can be occupied by only a single cluster (though
in d=2 pockets of "anti-ordering" can be embedded in larger clusters of
"ordering"); the conc~.pt of a cluster wall seems viable here. In d=4 (indeed
in d > 2) spongelike [42J clusters of ordering and anti-ordering can inter-
penetrate one another and the idea of a cluster wall seems less potent.
Appropriate use of the Renormalisation Group, together with further
experiments (laboratory or computer) on displacive systems offer the best
hope of establishing the picture appropriate for three-dimensional systems
(doubtless a compromise between the d=2 and d=4 situations) - in particular,
the relative importance of the soft phonon and the soft domain wall.

Acknowledgements
I am happy to acknowledge numerous illuminating conversations with, among
others, S. Aubry, H. Beck, A.R. Bishop, R.A. Cowley, J.A. Krumhansl,
T. Schneider, E. Stoll and H. Thomas.

References
1. T. Riste, E.J. Samuelsen, K. Otnes, J. Feder: Solid St. Commun. 9, 1455,
1971; S.~. Shapiro, J.D. Axe, G. Shirane, T. Riste: Phys. Rev. ff6,
4332, 1972
2. K.A. Muller, W. Berlinger: Phys. Rev. Lett. 26, 13, 1971
3. J.A. Krumhansl, J.R. Schrieffer: Phys. Rev. Bll, 3535, 1975
4. K.G. Wilson, J. Kogut: Phys. Rep. fl£, 77, 1974
5. For a review see A.D. Bruce: Ferroelectrics 1£, 21 1976; and to be
published
6. P.C. Hohenberg, B.I. Halperin: Revs. Mod. Phys. 49, 523, 1977
7. T. Schneider, E. Stoll: Phys. Rev. Lett. 31,1254,1973; Phys. Rev.
Lett. 35, 296, 1975; Phys. Rev. ~, 1216:-1976; Phys. Rev. ~, 1302,
1978 -
8. S. Aubry: J. Chem. Phys, 62, 3217, 1975; 64,3392,1976
9. C.M. Varma: Phys. Rev. ~, 244, 1976

133
10. A.R. Bishop, J.A. Krumhansl: Phys. Rev. ~. 2824. 1975
11. H. Beck: J. Phys. £. ~. 33. 1976
12. A.R. Bishop: Proc. Int. Conf. on Lattice Dynamics (1977). 144
(Flammarion, Paris. 1978)
13. S. Sarbach: Phys. Rev. ~. 2694. 1977
14. R. Bausch and B.I. Halperin: 1978 preprint
15: B.I. Halperin. P.C. Hohenberg. S. Ma: Phys. Rev. ~. 139. 1974;
Phys. Rev. ~. 4119. 1976
16. B.I. Halperin. C.M. Varma: Phys. Rev. ~. 4030. 1976
17. R.J. Glauber: J. Math. Phys. ,1.294. 1963
18. R. Brout. K.A. MOller. H. Thomas: Solid St. Commun. ,1. 507, 1966
19. R. Blinc. B. Zeks: Adv. in Phys. 2l. 693. 1972
20. Y. Onodera: Prog. Theor. Phys. 44, 1477, 1970
21. I. Hatta: J. Phys. Soc. Japan 24. 1043, 1968
22. M.K. Barnoski. J.M. Ballantyne: Phys. Rev. lZi, 946. 1968
23. W. Cochran: Phys. Rev. Lett. 1. 412. 1959; Adv. in Phys. !. 387,1961
24. E. Eisenriegler: Phys. Rev. B9. 1029. 1974
25. R.A. Cowley. W.J.L. Buyers, G. Dolling: Solid st. Commun,. 7. 181. 1969.
For a review of the evidence for soft phonons see J.F. Scott: Revs.
Mod. Phys. 46. 83. 1974
26. K.K. Murata: Phys. Rev. ~. 462. 1975
27. S. Aubry. private communication
28. A.D. Bruce. T. Schneider: Phys. Rev. ~. 3991. 1977
29. J. Fr6hlich. B. Simon. T. Spencer. Commun. Math. Phys. 50. 79. 1976
30. R. Oppermann. H. Thomas: Z. Phys. B22. 387. 1975
31. T. Schneider: this Symposium
32. R. Silberglitt: Solid St. Commun. 9. 2021. 1971; J. Feder: Solid St.
Commun .• 9. 2021.1971; R.A. Cowley. G.J. Coombs: J. Phys. C6. 143.
1973 -
33. J. T6pler. B. Alefeld. A. Heidmann: J. Phys. flQ. 635. 1977
34. C.N.W. Darlington, D.A. O'Connor: J. Phys. C9. 3561.1976
35. K.K. Murata: Phys. Rev. ~ 2028. 1976
36. S. Aubry: Ferroelectrics.!.§.. 313. 1977
37. W.L. McMillan: Phys. Rev. ~. 1496. 1976
38. P. Bak. V.J. Emery: Phys. Rev. Lett. 36. 978.1976; P. Bak: this
Symposium
39. See. e.g .• A.D. Bruce. R.A. Cowley: J. Phys. C (in press)
40. A.D. Bruce. R.A. Cowley. A.F. Murray: J. Phys. C. (in press)
41. M.E. Fisher: Essays in Physics (Academic Press) ,1. 43. 1972
42. E. Stoll: Phys. Lett. 58A. 121. 1976

134
Nonlinear Lattice Dynamics: Molecular Dynamics Studies

T. Schneider and E. Stoll

IBM Zurich Research Laboratory, CH-8803 RUschlikon, Switzerland

1. Introduction

Nonlinearity combined with dispersion were essential aspects in various

contexts for many decades but have enjoyed remarkable revival of interest in
the last few years [1,2]. The topic we wish to discuss here is structur-
ally-commensurate lattice systems reducing in the continuum limit to the
~4 Hamiltonian and exhibiting a distortive phase transition. This transi-
tion takes the system from an ordered low-temperature displacement configu-
ration to a different high-temperature pattern, where the mean displace-
ments, representing the order parameter, vanish. The occurrence of a phase
transition brings about novel aspects which were absent in previous studies
of nonlinear systems such as the FERMI-PASTA-ULAM problem [3] or the TODA
[4] lattice. In fact, a molecular-dynamics study [5] of a two-dimensional
(2-d) model revealed, close to the transition temperature T, the formation
and motion of clusters separated by walls, representing conn~cted particles
whose displacements have a sign opposite to that expected from the zero-
temperature arrangement. The existence of these clusters and their dynamics
has important consequences. It reveals that a given particle not only
vibrates around a mean position, but also that the particles perform large
amplitude motions which cannot be treated within the framework of conven-
tional lattice dynamics. Subsequently, these numerical results have been
confirmed [6] extended to l-d [7-9] and 3-d systems [10] including models
where the displacement vector is no longer a scalar but has two components
(n = 2) [11]. The l-d systems are special in the sense that for any finite
temperature T no ordered phase occurs. Nevertheless, because T = 0 can
be considered as the critical temperature and considerable analytic progress
was possible, the l-d models are also of considerable interest. The work
of KRUMHANSL and SCHRIEFFER [12] on the l-d ~4 problem, demonstrated that
at sufficiently low temperatures the free energy is, to a good approxima-
tion, that of a gas of noninteracting quasiparticles of two types, namely
phonons and kink excitations. The presence of kinks with a statistical
weight is gratifying since it puts the numerical evidence for cluster walls
(kinks) onto a firm physical ground. These analytic procedures are substan-
tially limited to l-d and low T. Nevertheless, numerical studies revealed
that close to the transition temperature, clusters and cluster walls also
occur in 2- and 3-d ~4 systems [5,6,10,11], and that the static and dy-
namic critical exponents are consistent with those of the corresponding
Ising model.

135
 It is also interesting to note that the partition function of the w4
system reduces in a certain limit [13] to that of the Ising model. More-
over, according to the universality hypothesis for static critical phenomena
[14], the w4 system belongs to the Ising universality class with critical
exponents which depend on d only, becoming classical for d > 4 and
identical to the Ising model exponents. This expectation is c~nsistent with
the numerical results for d = 2 [5,6] and d = 3 [10]. Recognizing that
in the Ising limit, where the scalar displacement can assume two values
only, let us say +1 and -1, cluster and cluster walls represent the only
possible displacement pattern, one might question whether or not cluster and
cluster walls dominate the static and dynamic properties close to the phase
transition. If so, we would have a pictorial explanation of universality,
according to which the w4 system belongs to the Ising universality class,
in terms of the dominating displacement configurations, the clusters being
separated by walls.

It is the main purpose of this paper to shed some light onto this ques-
tion. In Sec. 2 we define the models, a conventional 3-d w4 system and an
extension of this model, which also exhibits an acoustic mode. Section 3 is
devoted to a sketch of our molecular-dynamics technique. Some numerical
results, demonstrating the dominating role of clusters and cluster walls,
and giving the temperature dependence of the zero wave-vector excitation
spectrum are presented in Sec. 4.

2. Simple Models Exhibiting Distortive Phase Transitions

Distortive phase transitions include structural, ferroelectric and anti-

ferroelectric phase transitions in crystals. An essential property is the
existence of an order parameter which represents a linear combination of
particular displacements. Examples are the angular displacements of the
oxygen octahedra in perovskite-type crystals and of hydrogen octahedra in
ammonium halides. Such complicated systems have many other degrees of
freedom. Consequently, a detailed theoretical description, taking all
degrees of freedom into account, would lead to insurmountable difficulties.
In this case, it is not surprising that the model systems considered so far
only take into account the order parameter and those degrees of freedom
strongly coupled to it [15].

Here we consider two 3-d models with a scalar (n = 1) order parameter.

Model I corresponds to a discrete w4 system having one particle per unit
cell. Model II is an extension with two particles per unit cell which
reduce in a certain limit to Model I.

Model I

The Hamiltonian of the ferrodistortive model is [10]

MXZ
JI. A-12C B x4 +.f L (X - XJI. ,) 2 .
%= L -2-+ -2-- L x2 +-
JI. 4 L 2<JI.,JI.'> JI.
(1)
JI. JI. JI. JI.

136
£ labels the particle with mass M in the £-th unit cell. ~L ~nd
are momentum and displacement with respect to a rigid, cubic prlmltlve
reference lattice with lattice constant a. M, A, Band C are model
parameters chosen as [10]

A = -1, B = V3, C = V6, M = 1. (2)

parameters guarantees that at T = °

<£,£'> denotes nearest-neighbor interactions. This choice of the model
the order parameter given by

l2C-A
(3)
B

does not vanish. Consequently, the system will undergo a ferrodistortive

phase transition at some T = Tc > 0, where <X?= 0.

The equation of motion reads

- M X£ = (A - l2C)X£ + B X~ + 2C L (X£- X£,) . (4)

£'
Introducing the scalar displacement field f(R,t) this equation becomes, in
the continuum limit, identical to the ~4 equation of motion

- M f(R, t) = (A - l2C)f + B f3 - 2C a 2 t; f. (5)

Linearization of (3) and (5) appears to be adequate at low temperatures

where, at most, all particles perform small amplitude oscillations in one of
the double wells defined by

(6)

which double well is preferred depends on the preparation of the system at T

0. The frequencies of the small amplitude oscillations are given by
2
Mw (7)

where q is the wave vector. As pointed out in this context by KRUMHANSL

and SCHRIEFFER [12], Eq. (5) also has other important particular solutions,
namely the kink and anti-kink SOlutio(n~x_ Uxt )
f(R ,R = 0, R = O,t) =
x y z
~ Uo tanh
<;
x
n (8)

where
U = ~12C-A' <; 2 =
° B
x l2C-A
(9)

This kink takes f fromU O to -Uo; and antikink takes -f to f

Between +f and -f there is a wall which we call the cluster wall of
approximate thickness 2n
sx . The cluster wall moves with velocity Vx

137
 The partition function

+00 +00
z ~oodXl··· d~ ~oodXl··· d~ exp (10)

of this system reduces in the limit

A + - 00, B ->- + 00 , _AB 1 (ll)

to that of the 3-d Ising model [13J.

Approach~ng this limit, the phonon frequency (7) tends to infinity

because <f >->- 1 [see (3)J while the cluster wall thickness tends to zero
and U to 1 [see (9)J. This behavior reveals that by approaching the
Ising 2imit, the phonons cease to be of relevance. What remains are kink-
like large amplitude motions giving rise to the formation of clusters
separated by thin walls.

There is another interesting limiting case, i.e., the displacive limit

[16J defined by l2C = A where the order parameter vanishes even at T = 0
[see (3)J. By approaching this limit the cluster wall thickness increases,
the amplitude of the kinks decreases (9), and the phonon frequency (7)
remains finite. Accordingly, near this limit both small and large amplitude
motions seem to play an important role.

The model considered here and specified by the parameters in (2) is quite
far from the two limits. In fact, according to (2),(3) and (7), the phonon
frequency is at low temperature and small q, given by
2 1 2 2
w (q) = 6 + 3 a q, (12)

and
A/12C = - 1/2 (13)

being far from both the value at the Ising limit (A/12C -00) and the dis-
placive limit (A/12C = 1).

Model II

The Hamiltonian of this model is

~X .Q,K2 A 2 B 4
:1C= L ---+
2 2 L (X
.Q,K
- X.Q,'K') +-
4 L (X
tK
- X.Q,'K~)
.Q"K nn nn
C
+-
2 I (X
.Q,K
- X.Q,'K ) 2. (14)
nnn

.Q"K label the K-th particle with mass K in the .Q,-th u~i~ cell: MKX.Q,K
and X.Q,K are m~mentum and displacement.with r~spect to a rLgLd cubLc
reference lattLce. There are two partLcles wLth mass Ml and M2 per
unit cell. nn denotes nearest-neighbor and nnn next-nearest-neighbor

138
interactions. The arrangement of the particles is sketched in Fig. 1.

Fig.l Schematic sketch of the

high temperature phase. X denotes
particles with mass MI and 0
particles with mass M2

At zero temperature, the force acting on a given particle has to vanish.

This condition determines the T = 0 order parameter given here by
A
(15)
B
The ordered phase will be antiferrodistortive because the mean displacement
in sublattice 1 is positive and in sublattice 2 negative. Some additional
insight might be obtained by treating the model in the Hartree approximation
where the quartic term is linearized according to

B
(16)
4" nn

In this approximation, the Hamiltonian can be diagonalized in terms of two

normal coordinates. One corresponds to the order parameter, describing the
counter-motion within the ceIl and giving rise to an optic-phonon branch.
The other normal coordinate is associated with an acoustic branch (Fig. 2).
The expectation that large amplitude motions will also occur can be sub-
stantiated as follows: i) in the limit where the mass of the particles in
one sublattice, say M -1' tends to infinity, Model II obviously reduces to
Model I; ii) moreover,K- it can be shown that the partition function of the
potential energy reduces in the continuum limit to Model I. This is achiev-
ed by reducing the unit cell by a factor of two and relabeling the particles
from 1 - 2N. Accordingly, the static critical properties will be identical
to those in Model I. To describe the dynamic properties, however, the
kinetic energy involving the two different masses must be taken into account.
This is easiest seen in the Hartree approximation where the phonon frequen-
cies are given by

139
 Z
(Ml+MZ)S + ~Ml-MZ)ZBZ+ 4MlMZaZCOS Z ~ q
w (q) (17)
ZMlMZ
where
Z
a = 6 A + 18 B «XII - XlZ ) >
S = a + 6 G (1 - cos a q) (18)
-+
q (1, 1, 1) q

The model parameters have been chosen as

A = -1, B = 1, c= 1, 1. (19)

Fig.Z Sketch of the phonon-

dispersion law in Model I I in
the Hartree approximation
q

Dynamic Variables and Spectral Densities

To investigate the excitation spectrum, we consider the spectral densities

of the dynamic variables of interest

-+ -iwt <A(-~,t) A(q,O»

SAA (q,w) = f dt e (ZO)
<A(-q,O) A(q,O»

Variables A of particular interest are the displacement (order parameter)

and energy fluctuations, defined by

Model I
• -+ -+
1
L
~q'R
X (;1) <5XR, e R" <5XR, XR, - <XR,> (Zl)
IN R,

1 (;J·lt R"
.~(q) L <5%R, e <5~= ~-<~> (ZZ)
IN R,

140
 ·2
MXR, A-12C
~=-2-+ (23)
2

+
q is the wave vector and the vectors define the direct lattice.

Model II

(24)

(25)

(26)

It is important to note in this context that the energy density ~(q) is a

conserved variable because

d + • +
dt ~(q = O,t) =!TC (q = O,t) = 0, (27)

as it should be for a Hamiltonian system.

3. The Molecular-Dynamics Technique

In the conventional molecular-dynamics technique [17,18] , one solves the

set of coupled Newton's equations associated with a given Hamiltonian ac-
cording to a set of difference equations with a time increment. This set of
difference equations approximates Newton's equations. Starting from given
initial conditions for the positions and velocities, the particles are then
allowed to move, and the time evolution of their canonical variables
(XR,' XR,) are calculated. Assuming the system is ergodic, estimates for
microcanonical ensemble averages may be obtained in terms of time averages.

It would be preferable to have a molecular-dynamics technique, simulating

a canonical ensemble, as most experiments are performed at constant global
temperature. This was achieved by considering, in place of the Newton's
equations, the coupled set of Langevin equations [10]

_ d~_
~ r M Xn + nn(t), (28)
oXR, J<. J<.

where
(29)

141
Here, it is assumed that the particles suffer collisions with much lighter
ones which represent the heat bath defining the temperature T. The colli-
sions are described by the friction r M X
and a random force n (t). It
may be shown that the stationary solution 5f the associated Fokker~Planck
equation is the canonical distribution function

P eq (Xl' ... , *N, Xl' ... , l),)

1
$. (30)

Starting from initial values for pos~t~ons and velocities, the particles are
then allowed to move under the influence of the computer-generated random
force. The temporal evolution of the variables are then calculated with a
set of difference equations approximating the Langevin equations (28). On
this basis, one obtains
(31)

For a detailed description of the algorithm and the random force generation,
we refer to [10]. The system is then allowed to age, or, i~ other words, to
reach equilibrium. After this interval, the subsequent 10 steps are used
to perform time averages representing canonical ensemble averages.

From the Langevin equations (28), it is obvious that the dynamic proper-
ties will be modified, in particular, owing to the damping term. To reduce
this modification, r must be chosen in such a way that
1
r »Tc' (32)

where T denotes the characteristic time of the dynamics. This guarantees

that theCexcitations do not become overdamped owing to the friction term.
Another important constraint on r evolves from the energy conservation of
a Hamiltonian system. Since our system evolves according to the Langevin
equation, it follows that

d%"= ~ M X9, a%" •

dt z:
9, MX
+ ax9,X9,

• 2
-I [r M X9, - x9,n9,(t)]. (33)
9,

Consequently, energy is not conserved because the Hamiltonian system is in

contact with the heat bath. To avoid artificial features due to the random-
noise pulses, the mean time between two pulses must be small compared to
TC. In this case, we may average (33) over some pulses. This leads to
d%"
(34)
dt

With the Ansatz

-t/
T
= a e (35)

142
and
d 6 Ek d 6 Ek 2 C d 6 Ek
d%= d% d% dT 'V v
---= 'V (36)
dt d 6 Ek dt dT d 6 Ek dt kB dt

we find
C
V 1
T = (37)
kB r '

where C is the specific heat. Accordingly, energy is almost conserved

within d;'e characteristic time Tc' provided that
C
v 1
T = » T • (38)
kB r c

Moreover, owing to the fact that the system evolves according to the
Langevin equation, the time interval Tch over which an evolution is follow-
ed must be larger than T so that
T « (39)

Combining inequalities (38) and (39), we finally obtain

C
V
T» » T • (40)
ch kB r c

From this relation, it becomes evident that energy can be almost conserved
provided r and the chain length T h are appropriately chosen. An ex-
ception is very close to T, where Ene characteristic time T becomes
very long. c c

In the calculations presented here, we have considered systems of 8000

particles defined by Hamiltonians (1) or (14) and model parameters specified
in (2) and (19), respectively. The systems were subjected to periodic
boundary conditions. In the time interval, where time averages were perfor-
med, r was chosen as
r = 0.005. (41)

4. The Excitation Spectrum: Numerical Results

Before examining the explicit results, it is helpful to summarize those

features of the excitation spectrum which will be expected on general
grounds. At low temperatures T« T , the particles will oscillate about a
mean pos~t~on. The displacements ar~ small and anharmonic perturbation
theory should be adequate. We therefore expect an excitation spectrum
exhibiting phonons and, by virtue of energy conservation, heat diffusion or
second sound. The resulting Rayleigh or second-sound peaks may occur in
both the energy and displacement spectral densities, due to the coupling
between order parameter and energy fluctuation below Tc.

With increasing T, anharmonic effects become mor~ important and a

softening of the optic branch will occur for small q, as expected from the

143
Hartree approximation. Moreover, large amplitude motions will become more
important, because it is much easier for the particles to overcome the
potential barrier. This will result in the formation and dynamics of
clusters. A cluster represents particles connected by nearest-neighbor
bonds, having displacements with a sign opposite to that expected from zero
temperature. A cluster is surrounded by a cluster wall where the displace-
ments change sign. This phenomenon will illustrate the importance of the
particular kink and anti-kink solutions of the underlying equation of motion
in the continuum limit. The formation and dynamics of these clusters will
certainly affect the excitation spectrum producing, in particular, a central
peak. This peak will be superimposed, at least below T and for q # 0,
on the heat-diffusion peak. Above T, energy and displ~cement fluctuations
are no longer directly coupled. Neve?theless, central peak will be expected
in the displacement spectral density due to the cluster dynamics and in ad-
dition, phonons, due to the small amplitude motions.

To summarize, the excitation spectrum is expected to be rather rich and

complicated due to the presence of large and small amplitude oscillations
and the implications of energ4 conservation. In fact, by approaching T
from below we expect in S (q,w) a softening of the optic-phonon resonagce
for q = 0, a heat-diffusi~~ central peak for small but q # 0, and a
central peak due to the cluster dynamics. Above T, the excitation spec-
trum of the displacement spectral density should becsimpler since the heat-
diffusion central peak cannot occur by means of a direct coupling between
energy and displacement fluctuations. Accordingly, by approaching T from
above, we expect in S (q,w) a softening of the optic-phonon resonange for
q = 0 and a central p~ik due to the cluster dynamics.

Model I

For a detailed discussion of the numerical results we refer to [lOJ. Here

we only present some of the crucial results. As mentioned above, we expect
the formation of clusters separated by walls. Figure 3 shows snapshots of
the particles in a plane perpendicular to (1,0,0). Only those particles are
marked where sgn XQ, = - sgn <X>T=O.

... . . ......
.... .
...........
-.-- --_. ---
.....
.............
...............
....
....... .......
.......... .
...... . .......
- ---- ------------- ...........
...
....... ..........
......... . ....
............
Fig.3 Snapshots of instan-
- -- - ... ..
............ ... ..
taneous cluster configura-

----- -:--- (0)

.....
.1·· ••1 • al. -: 1-.:
'-_-_-_-__-_--_-_-_-_-__-_-....J (b)
tions (a) kBT = 6; (b) kBT
8

Above Tc ~ 7.2 the clusters of positive and negative XQ, must, of course,
be equal on the average so that the order parameter vanishes. The formation
of these clusters and the associated cluster walls has important consequen-
ces. Their formation implies the presence of large amplitude motions which
will certainly affect the excitation spectrum. Here it might be appropriate
to start with an overview of the excitation spectrum.

144
 ~ +
In Fig. 4 we have sketched the essential features appearing in S (q,w)
~nd g~q,w) for vari~us temperatures at wave vectors q = (O,O,O)x~nd
q = (rr/10a,0,0),respect1vely. At k T = 15 and 30, there is a broad phonon
peak in S ,but no central peak (C~) occurs. ~~~ only exhibits a CP
peak with xXhalf-width nw ~ q2, due to heat diffusion. This phenomenon

LLL
!u(q''")uuuJ~~LL~~LLL qO{tOa· O•O}

o 0.25
•
0.5 2 6 8 15 30 T

<8x81t>= 0 INDEPENDENT
ANHARMONIC PERTURBATION THfORY QUARTIC
OSCillATORS

°
Sketch of the essential features of the excitation spectrum in
~i.4
t!_____1+
+xx q,w) and ~jt,q,w) at various temperatures for wave vectors q =
+ +
and
q = (rr/10a,0,0). HD denotes the central peak due to heat diffusion, and C
the central peak ar1s1ng from the cluster dynamics. ss denotes the second-
sound peak, and oss overdamped second sound

does not occur in S ,as expected, because the displacement-energy coup-

ling vanishes above XXT • Moreover, the broad phonon resonance shifts to
lower frequencies with aecreasing temperature. At k T = 8, which is close

°
to kBT ~ 7.1, new features appear: a CP at q = (O,B,O) in S and the
soft mgd~ becomes overdamped. At q = 0, heat diffusion does notXexist.
Consequently, the q = CP must be attributed to the cluster dynamics. In
S~~, it is superimposed by the Rayleigh peak.

Below T , the coupling between displacement and energy fluctuations is

nonzero. Tte Rayleigh peak will appear4 therefore, in both S and S~~
Accordingly, we have at k T = 6 and q = 0, the Rayleigh andx~luster cen-
B + +
tral peaks superimposed. The cluster CP also appears, as is seen from q = 0.
The soft mode is overdamped. At k T = 4, the soft mode becomes underdamped
and the height of the cluster CP isBreduced, as expected. In fact, for the
formation of clusters, it is necessary that particles overcome the potential
barrier. This process becomes less probable as the temperature is lowered.
This fact is illustrated at kBT = 2, where the cluster CP no longer appears
(q = 0). At q # 0
there is a CP due to the energy fluctuations in both
+
S~x and S~~, but it is weak.in Sxx. The q de~ende~ce of the half
w1dth reveals, however, that th1s CP cannot be expla1ned 1n terms of heat
diffusion. Since this peak splits at lower temperatures (k T = 1), it must
. B .
be attr1buted to overdamped second sound. The phonon resonance 1S now
defined, indicating that we enter the regime where anharmonic perturbation
theory should work. At kBT = 1, the phonon is very well defined. It

145
dominates S ,but at small w the weak second-sound resonance appears.
xx .
This peak domLnates S~~, where the phonon peak LS very weak. At the
lower temperature kBT = 0.5, the features are similar to those at kBT = 1.
The only difference LS a zero-frequency tail of the second-sound resonance
in S~~. This tail must be attributed to two phonon processes. Finally,
at T = 0, where the spectrum can be calculated exactly, only the phonon
resonance at the frequency given by (12) survives.

An essential feature of the q °

CP is, that by approaching T from

°
above or below, it appears in addition to the optic soft-mode resonagce.
->- -+
This feature, illustrated in Fig. 5, clearly reveals that the CP at q =
does not evolve from the small amplitude oscillations but is associated with
the large amplitude motions originating from the formation and the dynamics
of clusters and cluster walls.

-+
O,w) at
OL-_L----IL----1_-----L_-----L_---L_--.J
o 0.4 0.8 1.0 w
- ->- ->-
These results reveal that the excitation spectrum appearing in S (q = O,w)
is close to T dominated by a central peak originating from theXfarge
amplitude moti6ns associated with the dynamics of clusters and cluster
walls. Moreover, the strength of the soft-mode resonance evolving from
small amplitude motions, decreases in this regime by approaching T.
Accordingly, what remains are the motions reminiscent of the Ising timit.

Model II

As pointed out in Sec. 2, the partition function of this system can also be
reduced to the ~4 model. Accordingly, here we also expect the formation
of clusters and cluster walls. This .expectation is confirmed in Fig. 6

°.
showing snapshots of the particles in a plane perpendicular to (1,0,0). Only
those ·particles are marked where sgn X9.,/' sgn < X9., K>T=

::- . :: .:; .
• .:. -.11 .. _.
.... . .. ......
... ....
II-;5:55-.-:
•• :.:.
.....1....... .
III- 1.-.
;:-.1.
... I: •
•::.. :.;1:1::11=;1
Fig.6 Snapshots of instan-
taneous cluster configura-
:....... : ..: I i:· .1.i·::I~: :I::~:
.1 • II .c- .-.-.1
tions at kBT = 1.75 (a) and
L ___:_·_ _ _i . : (0) L-·_·_"_·_·_·_·_·_.__·_·_·~(b) kBT = 2.1, where Tc ~ 2

146
 We expect that the large amplitude motions associated with the formation
of the cluster walls will, in analogy to Model I, affect the excitation
spectrum close to T.
c
To test this expectation, we next discuss the temperature dependence of
the excitation spectrum above T. Figures 7,8 and 9 show the w depen-
dence of S (q;+
O,w) at kBT; 2.3, 2.2 and 2.1, respect~vely.
. It sh~uld
be emphasiz~a that acoustic-branch resonances no longer occur at q; 0 .
At kBT; 2.3, the spectrum is dominated by the optic-phonon resonance at
w 'V 0.62.
max 'V

6
'3
'9x
x
(if)

D,w) at

By lowering the temperature to k T ; 2.2 (Fig. 8), the peak position of

the soft optic mode remains essentia~ly the same, but the strength of the
resonance is considerably lowered and a new feature appears in addition,
namely a CPo By lowering the temperature further to kBT; 2.1 (Fig. 9),
the strength of the soft optic-mode resonance becomes so small that the
spectrum is dominated by the CPo

These results again reveal that by approaching T the excitation

spectrum of § (q; O,w) is exhausted by a central ~eak originating from
motions which ~~e distinct from the small amplitude oscillations, giving
rise to the soft optic-mode resonance. In fact, the strength of the optic-
mode resonance is seen to decrease by approaching T and over a certain
temperature range, both this resonance and the CP ar~ present. In view of
Fig. 6, we are then led to conclude that the large amplitude motions ~sso~
ciated with the dynamics of clusters and cluster walls dominate the q; 0
excitation spectrum close to Tc.

147
 4
'3
'9><
><
(<fl

2
->-
D,w)
~
Fig.8 S (q at
kBT = 2.2x

0
0 0.8 1.6 w

8
'3
'9
><
><
(<fl

4
->-
D,w)
~

Fig.9 S (q at
k BT = 2.lj:X
1

5. Summary and Conclusions

The present discussion has revealed that in 3-d ~4 systems, nonlinearity,

dispersion and energy conservation led to a rather rich excitation spectrum.
Phonon resonances, second sound, heat diffusion and a CP due to large ampli-
tude motions, associated with the formation and the dynamics of clusters do
occur. At zero wave vector and close to T, however, the excitation
spectrum of the displacement spectral densiEy becomes simple. It is ex-
hausted by the central peak arising from the cluster dynamics. The numer-
ical results on 2-d ~4 systems indicate that this conclusion even holds
close to the displacive limit. Accordingly, the cluster displacement
patterns and the associated large amplitude motions seem to dominate the
partition function and the zero wave-vector dynamics of the displacement
spectral density close to T. If one accepts this conclusion, then it
offers a pictorial explanatign of universality according to which the ~4
system belongs to the Ising universality class [14J. Conversely, it does
not suggest trivial critical behavior for d > 4, where the critical expo-
nents are expected to be classical [14J.

148
References

1. A.C. Scott, F.Y.F. Chu, D.W. McLaughlin: Proc. IEEE 61, 1443 (1973)
2. R.K. Bullough, R.K. Dodd: In "SyneY'getics", ed. by H~Haken (Springer,
Heidelberg 1977)p.92
3. E. Fermi, J.R. Pasta, S.M. U1arn: LASL Report No.LA-1940, 1955; To be
published
4. M. Toda: Progr. Theoret. Phys. (Kyoto) Suppl. 59, 1 (1976)
5. T. Schneider, E. Stoll: Phys. Rev. Lett. 31, 1257 (1973); Phys. Rev. B
ll, 1216 (1976) --
6. M.L.P. Bartolome, W.C. Kerr: Solid State Commun. 21, 253 (1977)
7. S. Aubry: J. Chern. Phys. 62, 3217 (1974); 64, 339Z-(1976)
8. T. Schneider, E. Stoll: Phys. Rev. Lett. 3~ 296 (1975)
9. T.R. Koehler, A.R. Bishop, J.A. Krurnhans1:-J.R. Schrieffer: Solid
State Commun. 15, 1515 (1975)
10. T. Schneider, ~ Stoll: Phys. Rev. B 17, 1302 (1978)
11. T. Schneider, E. Stoll: Phys. Rev. Lett. 36, 1501 (1976)
12. J.A. Krumhans1, J.R. Schrieffer: Phys. Re~ B 11, 3535 (1975)
13. St. Sarbach, T. Schneider: Phys. Rev. B 13, 46~(1976)
14. M.E. Fisher: Rev. Mod. Phys. 46, 597 (1974)
15. A.D. Bruce: Ferroe1ectrics 12:-21 (1976)
16. T. Schneider, E. Stoll: Phy~ Rev. B lQ, 2004 (1974)
17. A. Rahman: Phys. Rev. 136, A405 (1964)
18. A. Ver1et: Phys. Rev. 159, 98 (1967)

149
Computer Simulation of Structural Phase Transitions

Will iam C. Kerr

Centre Europeen de Calcul Atomique et Moleculaire
Universite de Paris XI, Orsay, France,and
Department of Physics, Wake Forest University
Winston-Salem, NC 27109, USAl,2

Phenomena associated with structural phase transitions (SPTs) have been in-
tensively studied in recent years, ever since the discovery of the "central
peak" in the neutron and light scattering spectra of the soft mode in many
diverse materials [lJ. These lattice dynamical systems lend themselves to
study by molecular dynamics (MD) and several such investigations have been
carried out. The results presented here are from a MD study of a two-dimen-
sional 1600 particle system which exhibits a structural transition.
The high temperature phase is a square lattice with parameter a. At the
transition the unit cell doubles, giving a square structure for the low
temperature phase also. At the transition each atom shifts parallel to the
[OlJ direction and adjacent atoms move in opposite direct~ons, giving an
antiferrodistortive transition with soft mode wavevector qc at the corner
of the Brillouin zone [2J. The mechanism producing the transition is that the
mode with wavevector qc and polarization parallel to [OlJ is unstable in the
bare harmonic approximation. The lattice is then stabilized by the anhar-
monic potential energy, which has the form
~ (a) = ~(l1if!o/aa )L:-:t{
Ruo[if
2
x (it) + ify (1)J + V ox
[u4 (1) + if' (it)J} ,
y (1)

where ci(R) is the atomic displacement at site R, M is the particle mass, ~

is a nearest neighbor harmonic force constant, and uo and vo are constants. o
This form for ~(a) has been used for renormalization group analyses of the
phase diagrams of perovskite crystals [3J.

Some features of the atomic motions in this system can be understood on the
basis of the continuum approximation to the equations of motion. To obtain
slowly varying quantities for which continuum equations are appropriate, the
staggered displacement field S(R,t), defined by reversing the displacements on
alternate lattice sites, is introduced. The equation for S is
y
(00 a)-202S /Ot2 - [-(1/2) - (3/8)r2 + (1/8)r4 J'12 S + (2+r2 - r4 - g)S y
o y y
- (1/8)(r2 - r
4 )(';72 S -
X Y
vaYSY) - g'J2 S -
x Y
~';72 S
5 xy X
(2)

+ [u (S2 + sa) S +
o x y y
V
0
S3 J
y
= O.
There is a similar equation for Sx. The constants r2 ,r4 ,r5 characterize the
harmonic part of the potential energy; g is an anisotropy parameter arising
from an external stress on the system (the same as used in [3J). The terms of
the first line of (2) form the Klein-Gordan equation; the coefficient (2+r2 +
r 4 -g) is negative, which gives the instability of the harmonic approximation.
The terms of the second line enter because the symmetry of the lattice is
1
~Permanent address
work at Wake Forest University supported by a Research Corporation grant
150
square rather than isotropic, there is an exte5nal stress on the system, and
there is a coupling between the components of $. The terms of the third line
describe the non-linear forces. For the special conditions of propagation
parallel to the y-axis with only a y-component to the motion, these equations
reduce to equations of one-dimensional models studied by other authors [4,5J,
and thus this model possesses the propagating domain wall solutions found
by those authors.

Several thermodynamic functions and correlation functions have been ob-

tained from the MD calculations on this model. The most interesting results
therefrom are presented here; a complete report is in preparation. The re-
sults for the specific heat, order parameter, and susceptibility show that
the model has a SPT with the critical temperature at Tc = (5.90±0.03)XlO-3 •

The instantaneous order parameter, its correlation function, and frequency

spectrum are defined by
,.. -iwt
D(w) = jdt e D(t). (3)

Some of the results obtained for D(w) are shown in Fig.l. At T = 7.357XlO-3
the soft mode peak and a central peak of comparable intensity are evident. As
T decreases, the central peak grows rapidly; the soft mode peak keeps about
the same intensity while decreasing in frequency. Two curves are shown at T
= 6.022xlO- 3 in order to bring out the soft mode peak at w = 0.012. At this
temperature the central peak intensity is about 450 times that of the soft
mode. Over the temperature range shown in the figure, the unnormalized value
of D(w=O) increases by a factor of 8000. As T goes below Tc ' the sequence of
changes of D(w) reverses: the central peak intensity decreases and the soft
mode moves out to higher frequency.

·B
For T outside an interval between 5.4xl0-3 and 6.2X1O- 3 the squa~e of the
soft mode frequency follows a linear dependence on temperature, ur (qc) ~

3 57

6000 Fig.l Spectral function of the

10 order parameter correlation function.
The numbers identifying each curve
200 are 103T, where T is the temperature.
6.145 Throughout the paper temperature is
o 0.1 Q2 w expressed in units of ~a2/~ and
04000 frequency is expressed in unifs of
" Wo
a
"-
~
a
2000 w

151
IT ~ Tol, with To> Tc' For temperatures within this interval around Tc '
w2(qc) deviates from this linear behavior, and it appears to remain finite
at Tc' This behavior is in agreement with the observed behavior of strontium
titanate [6J and lead germanate [7J.

The width of the central peak is a rapidly varying function of T. Exclud-

ing the voint at T = 6.022XI0- S , the values of the width can be fit by a
liT - Tcl dependence, both above and below Tc' These are nearly the same tem-
peratures where w2 (Qc) follows a linear dependence on T. However, the width
at T = 6.022xI0-S is considerably narrower than is given by this square root
dependence, so there may be a crossover between two different dependences here.

The correlation function for the spatial Fourier components of the staggered
displacement field is defined by

(4)

and its spectral function D(q,w) is defined as in (3). For q = 0, D(q,t) re-
duces to the order parameter correlation function, so this function gives the
wa~evector dependence of the central peak characteristics. The results for
D(q,w) at the temperature at which D(w) has the most intense central peak are
shown in Fig.2. As q moves away from zero, the central peak splits so that
the maximum is at a small finite frequency. A similar splitting has been
found by SCHNEIDER and STOLL [8J. This splitting occurs in a smaller temper-
ature interval around Tc than the interval for which there is a central peak
in~D(w). As q
increases further, the central peak intensity decreases, and
D(q,w) reverts to familiar anharmonic lineshapes.

The final results are for the energy density correlation function. The en-
ergy density h(R,t) is the function whose~sum over the lattice is the total
energy. Its spatial Fourier transform h(q,t), correlation function E(q,t),
and spectral function E(q,w) are defined analogously to (4) and (3), respec-
~
tively. Results for q = qc' and for T> Tc are shown in Fig.3. The most

t
5~=6'022 x10'3
100 (0,1) (1,0)

10~ Fig.2 Spectral functions for the

o 00.1 wOO .5 w correlation functions of the Fourier

9~~
components of the staggered displace-
ment field. The integer labels with
each curve identify the wavevector
i~ (0,2) 00 4WI0L 5
.5 w
in the usual scheme for periodic
'- (1,1) boundary conditions
'3 0 .I .2w
Ti20~
o
0
100 .I .2 OJ

00 .I .2w ~,2)
10~ 00 .5 W

oo .2 .4w 10~,3)
Il:,=:~6) O~w
00 .5 w 10~
00 .5 w

152
noticeable feature in E(qc'w) is the growth of a well-defined high frequency
peak near w = 0.9 as T approaches Tc from above. This feature is unlikely to
be a second sound peak because it occurs in a high temperature regime and it
is a short wavelength phenomenon. It is more likely to be a propagating two

b
phonon resonance.

g~'5 1 W
7.357 Fig.3 Spectral functions for the
(3 energy density ~orrelation function
.". °
Ta' 5~'::.'-'-'-""""5~~""'-\-1 w
=0'7
for wavevector qc' The numbers with
each curve identify the temperature
~ 6.469 in the same way as for Fig.l
"-
'3
.°
~ 5K"'O............~~.*5~~(\'-+-,16.....=.1457:~W

~.~.. ,.~.C
g~
~
° .5 1 w

In conclusion these MD calculations have shown that it is possible to ob-

tain a central peak which is an intrinsic anharmonic effect and which has the
same features as those found experimentally. In addition they have found a
splitting of the central peak for non-zero wavevectors and a high frequency
propagating mode in the energy density correlation function.

References
1. G. Shirane: Revs. Mod. Physics 46, 437 (1974)
2. M. L. P. Bartolome and W. C. Kerr: Solid State Commun. 21, 253 (1977)
3. A. D. Bruce and A. Aharony: Phys. Rev. B 11, 478 (1975)--
4. S. Aubry: J. Chern. Phys. 62, 3217 (1975); ibid. 64, 3392 (1976)
5. J. A. Krumhansl and J. R~ Schrieffer: Phys. Rev. B 11, 3535 (1975)
6. s. M. Shapiro, J. D. Axe, G. Shirane, T. Riste: Phy~ Rev. B ~, 4332
(1972)
7. K. B. Lyons and P. A. Fleury: Phys. Rev. B 17, 2403 (1978)
8. T. Schneider and E. Stoll: Phys. Rev. B 1l,-r2l6 (1976)

153
Soliton-Like Features in a Two-Dimensional XYModel
with Quartic Anisotropy

E. Stoll and T. Schneider

IBM Zurich Research Laboratory, CH-8803 RUschlikon, Switzerland

1. Introduction

Recently [1,2] we have shown that in a two-dimensional XY model with quartic

anisotropy, soliton-like features do occur. Here we extend this study to
include a variable anisotropy. Our results include: (i) the cluster wall
thickness multiplied by T seems to remain constant, (ii) the rather sharp
cluster walls prevent the formation of vortices [3], and (iii) the phase
diagram agrees quite well with predictions as obtained by the real-space
renormalization group [4].

2. The Model

The Hamiltonian of this model reads

$=!:! L (x? L
+ y·2 ) + -A (X 2 + y2) + ~ (X 2 + y2)2 L
2 ~ ~ ~ 2 ~ ~ ~ 8 ~ ~ ~

Bl
+-
4 L (X 4~ + y4) - C
~ L (X~ XHm+ y~ y~+m)· (1)
~ ~,m

+ + ••
The m~mentum and displacement vector X~ has two components, X~ (X~,Y~)
and X~ = (X~,y~), respectively. We consider nearest-neighbor interactions
only. For B = 0, the model would be rotationally invariant. The rigid
reference lattice is assumed to be a simple square lattice with lattice
constant a. The equations of motion read for the X~ component

- M it~ (A - 8C) X~ + ~ (X~ + y~) X~ + Bl X~ + 2C L (X~ - X~+m). (2)

m
Introducing polar coordinates, the equations of motion (2) become in the
continuous limit for the angular field
2
02 '" - -
M - ••
Bl P
v T <p sin 4 <po (3)
2C }
The model parameters have been chosen as follows

A =- 1, B = 2/3 - 2 B1 , 1/240 < - C 1/4. (4)

154
Of particular interest close to T is the kink solution [5J of the sine-
c
Gordon equation (3)
-1
",(Rx ' \ = O,t) = + tan exp (~a ~), (5)

which is a soliton (+) or antisoliton (-), respectively. Moreover,

2
a R V t (6)
x x

Of particular interest is the case where IBll becomes small, because in

the limit IBlt= 0 the system reduces to the two-dimensional XY model.
Although evidence from high-temperature series expansions [6J suggests that
the susceptibility of the isotropic model diverges at some finite T, the
lack of long-range order at any temperature has been proven rigorousty [7J.
KOSTERLIZ [3J suggested that below T bound vortex-anti vortex pairs should
occur, which dissociate above T. R~cently, JOSE et al. [4J studied the
effect of symmetry-breaking pertSrbation, such as quartic anisotropy using
the Migdal recursion scheme. Their results suggest that quartic anisotropy
is a strongly relevant variable, leading to a conventional continuous phase
transition but with non-universal exponents.

3. Molecular-Dynamics Results

For quantitative discussions we solved (2) by means of molecular dynamics

[1,8J. Systems of 1600 particles subjected to periodic boundary conditions
were studied. We found that in a temperature window below and above T,
clusters are formed, built up of particles connected by a nearest-neighBor
bond with positions lying in the same quadrant. Figure 1 shows a snapshot
of cluster configurations in the model with parameters given by (4) and
Bl = - 1/240 and T close above Tc.

Fig.l Snapshot of a cluster

pattern for T/Tc= 1.3 and
Bl = -1/:4? The arrows mark
tfie pos~t~ons of the particles,
and the clusters are separated
by lines

155
 Following the arrow directions, it is seen that the clusters are separa-
ted by rather narrow TI/2 walls. In fact, Fig. 2 reveals that for clusters
consisting of more than five particles the cluster wall thickness l/a is
less than two lattice constants, even for small IBll.

2 ,....., 2
8 1:- 11240 .3
....--, .... _---_ ... -
..2. \._._._._- -1/15
~
...... -2/5 ~
r-

(0) (b)

00
I 1 00 0.2 0.4
50 100 -8 1
n
Fig.2 Cluster wall thickness l/a depending on (a) the cluster size n
and (b) for a cluster consisting of more than five particles on the ani so-
tropy -B l , in units of lattice constants a

These results do not agree with the predictions of (6), where a ~ ~.

We note, however, that according to Fig. 3 the product T x (l/a) tends

c
to a constant for small IBll.

cy 6 r-----------------,
'--'

Fig.3 Product of cluster wall

thick~ess l/a and T depen-
ding on the anisotropyC -B l

0.2 0.4
-8 1
From these features and the absence of vortices (Fig. 1), we expect that for
the anisotropies considered here (4) Ising-like critical behavior will
dominate close to T. The non-universal behavior of the critical exponents
as proposed by JOSE ~t at. [4J appears to be restricted to very small aniso-
tropies. Nevertheless, their predictions of the anisotropy dependence of
T
c

156
 (7)

agrees quite well with our numerical estimates (Fig. 4) .

•
l- .
j
0.5 1.0 ~ 1.5
- 0.1 I- k BTc /(Cl) • Fig.4 Dependence of T on

. Bl . - - - after (7) [ZJ •

• molecular-dynamics results
- 0.2

4. Conclusions

To summarize, we have presented numerical evidence that close to T kink

and anti-kink sol~tons are relevant even in a two-dimensional discr~te
lattice model. We find sharp cluster walls preventing the system from form-
ing vortices. Finally, the anisotropy dependence of the critical tempera-
ture agrees quite well with the prediction of a real-space renormalization-
group study.

References

1. T. Schneider, E. Stoll: Phys. Rev. Lett 36, 1501 (1976)

2. E. Stoll, T. Schneider: In "Proceedings of the International Conference
on Lattice Dynamics", ed. by M. Balkanski (Flammarion, Paris 1978)
p.609
3. J.M. Kosterlitz, D.J. Thouless: J. Phys. C ~, 1181 (1973); J.M. Koster-
litz: J. Phys. C 7, 1046 (1974)
4. V. Jose, L.P. Kad~noff, S. Kirkpatrick, D.R. Nelson: Phys. Rev. B~,
1217 (1977)
5. A.C. Scott, F.Y.F. Chu, D.W. McLaughlin: Proc. IEEE 61, 1443 (1973)
6. R.E. Stanley, T.A. Kaplan: Phys. Rev. Lett. 17, 913 (1966); R.E. Stan-
ley: Phys. Rev. Lett. 20, 589 (1968); M.A. Moore: Phys. Rev. Lett. 23,
861 (1969) - -
7. N.D. Mermin, R. Wagner: Phys. Rev. Lett. 17, 1133 (1966); P.C. Rohen-
berg: Phys. Rev. 158, 383 (1967); D. Jasnow, M.E. Fisher: Phys. Rev.
Lett. 23, 286 (1969)
8. T. Schneider, E. Stoll: Phys. Rev. B l!, 1302 (1978)

157
Behavior of a 1/>4-Kink in the Presence of an Inhomogeneous
Perturbation

N. Theodorakopoulos, S. Hanna, and R. Klein

Fachbereich Physik der Universitat Konstanz
0-7750 Konstanz, Fed. Rep. of Germany

1. I ntroducti on

In recent years, the ~4 field theory became of interest as a one-dimensional

model for structural phase transitions. In particular, it was shown that in
a linear chain of harmonically coupled particles, which in addition sit in
local double minimum potentials, ordered clusters exist. The walls separat-
ing ordered clusters are the kinks of ~4-theory [lJ. The possible relevance
of an ensemble of independently moving kinks to the spectrum of scattered
neutrons and light has been investigated [1,2].
In t;,is contribution the response of the system to an external field of
wave number K and frequency ~ in the presence of constant damping n is stu-
died. We are mainly interested in the polarisation of a kink at rest (cal-
culated in section 2) and in the absorption spectrum of an ideal gas of mo-
ving kinks, treated in section 3.
2. Polarisation of a Kink at Rest
In properly chosen units, the equation of motion for tne displacement field
~ is
( 1)

The unperturbed ~4 field is known to pocess excited states (j) which are so-
lutions to (1) (~ith n=O=E) linearized in the neighborhood of a static kink
solution ~k(~) = tanh ~. These excited states are characterized by fj(~)
exp(iw<T), where fj(~) form a complete orthonormal set of functions. These
repres~nt the translation mode (j=T), the localized mode (j=L) and a conti-
nuum (j=c) of scatterin~ states {q}; the corresponding energies are w2 r =0,
w2 L = 3/2 and w2 q = 2+q /2 [3].
In the presence of damping and external field ~Ie follol1 a linearization
procedure similar to that employed by FOGEL et al. [4 1 for a homogeneous time
dependent field in the case of the Sine-Gordon (SG) equation. Accordingly,
~(~,T) ~k(~) + ~(~,T) (2)

~(~,T) ~ Bj(T) fj(~) = ~ ~j(~'T) (3)

J J

The expansion coefficients Bj(T) can be calculated in the framework of line-

158
ar response theory and are given by
Bj ( .) = [w 2 j - n2 - innJ-1 Fj (K) Eexp ( - i n.) (4)

Fj(K) = fd~ exp(iK~) f;(~) (5)

The Fj(K)'s are given in Table 1 for both the SG and <jl'+ cases.
Table 1 The Fourier coefficients Fj(K) of fj(~) for SG and <jl'+ fields. For
K=O the SG results reduce to those aerived in [4}. P stands for
principal value

Mode j SG <jl'+

2 ~ 1T K/2
T COSh1T K/2 .3 slnh1T K/2

L - A COSh1T
i 1T K
K/2

{q} - 1 -1 [ 21Tq o(q-K)+ 1 [21T(2-q2)o(q-K) +

/271 Wq [21T(q'++5q2+ 4)J V2
1T , - P 1 + 31T~K-T) ]
+ P sinh1T{K-q)72 J ~1Tq sinh 2(G-K) sin-1T -q-K)72

From (3) - (54 and Table 1 w(~,.) can be calculated. In the SG case, as
well as in the <jl case, the scattering states produce a contribution wvac
to w(~,.), which is plane-wave-like:
wvac(~") = [w 2K - n2 - innr1 E exp(iK~ - in.) (6)
The vacuum corresponds in this context to a crystal with optical phonons but
no kink. This part is subtracted out in the following definition of a mea-
sure for the kink polarizability. The total kink polarization is taken as
P = L p. (7)
j J
In the <jl'+ case the various contributions can be shown to be

(8)

(9)

PL =f d~ L =0
W (lQ)
-co

The definition used in (8) ensures that only effects are considered, \~hici1
are riirectly connected with the presence of the kink. It represents a gene-
ralization of the definition used in [4]. Note that the localized mode does
not contribute.

159
3. Absorption Spectrum of Moving Kinks
We may obtain additional information about the dynamics of the nonlinear
elastic continuum underlying (1) by considering the absorption spectrum of
a moving kink. At a first stage we may then average over all possible kink
positions, assumed uniformly distributed over all space; at a second stage
we can average over kink velocities, introducing on appropriate distribu-
tion. The resulting spectrum contains, apart from a "vacuum" contribution
due to absorption by optical phonon modes, new features due to the presence
of a kink. To the extent that kinks may be considered as non-interacting
entities such "corrections" to the spectrum will be additive and thus pro-
portional to the total number of available kinks.
Let sa and va be the (initial) position and velocity of the ath kink in
the laboratory coordinate system. We may then introduce coordinates in the
kink rest frame via a Lorentz transformation:
xa = ya (s - Sa - Va T) t = y LT - 2v (s - S )]
r
( 11)
a a a a
where y = [1 - 2V2]-lf2 (Note that the maximum kink velocity in the units
prescriged by (1) ~s 1/12). ~Je shall now make the assumption (a) that the
perturbing field is a scalar under Lorentz transformations and (b) that a
di spers ion re 1a ti on of the type n=cK holds, vlhere cis of the order of 10 5 ,
corresponding to the magnitude of the light velocity in the dimensionless
units chosen. ~Je may now compute the quantity IjJ in the kink's rest frame u-
sing (3)-(5) and considering that on account of the scalar property of the
external field the kink now sees a field, in its rest frame,with wavevector
Ka = y (K-2v n) and frequency n = y (n-v K). Upon transforming back to la-
a coordinates
boratory a a over
and averag1ng a al"1 PO:;l."'-.
clCroS
{E;a} i'/e ob"'-a"
c' n

(12)

(13)

where L = y L, L being the length of the chain. In the following we shall

conside~ th~ limit n+O. The contributions of the various modes are:
(i) j=T becomes relevant only in the "pathological" case n=vaK which.is ex-
cluded by the assumption (b) made above; the latter acts as a select10n rule;
note, however, that for a static kink the above condition corresponds to n=O,
and this is precisely the case in which energy can be absorbed by the static
kink - as predicted by (9) above.
(ii) j=L gives rise to absorption at n =wL; note again that !h~ oscillator.
strength FL(K), which would be zero fo~ a static kink (K 2NI0 ~ .for n=wL) 1S
in fact non-negligible if we consider a distribution of veloc1t1es P(v) =
Ay3 exp(-yE /T), where Eo is the rest energy of the kink, T the temperature
and A a nor~alisation constant [5J. Upon performing the thermal average over
the velocity distribution we obtain an absorption band for n<wL
( 2 n2)32 E wL
Imx~(n) = (A 1[3/L) wL~1 -1[ (2 2) 92 exp(- 2~n ) (14)
cos 72 wL- n
(iii) j=c. The continuum modes give rise, besides the vacuum absorption (ob-
tainable from (6) as well), to two distinct kink-related contributions.

160
Firstly we obtain absorption at frequency WK and secondly a diffuse back-
round which vanishes if we go far from wK'
ex 1
Imxc(n) = + r- B(Kex)n 8(n2 - wK)
2

ex
9n 2 K2
+ P 1 ex (15)
~ (K2+1)(K2+4)
ex ex
11here Qex= [2(n~-2)j lf2 and B(Kex) turns out to be, for long-wavelength phonons,
equal to -3.
The qualitative structure of the absorption spectrum is shown in Fig. 1.

[mx Fig.1 Schematic plot of the

kink absorption spectrum. The
dotted line represents the va-
cuum contribution; also shown
are the band arising from the
localised modes and the 8-peak
II as well as the diffuse back-
round due to the scattering
modes

4. Conclusion
In section 2 we have considered the total polarization of a single static
kink. The approach of section 3 should be regarded as complementary, since
it averages over all kink positions and therefore restores translational in-
variance. It snould be pointed out, as a further common characteristic of the
two approaches, that they both explicitly exhibit the number of phonon modes
removed by the kink. This can be seen by considering the K~O limit of (8)
and (15) where the strength of the phonon absorption peaks is in both cases
equal to lim d8/dK = -3, where 8(K) is the phase shift of a phonon with wave-
vector K K~O [3] .
References
J.A. Krumhansl and J.R. Schrieffer, Phys.Rev. B11, 3535 (1975)
S. Aubry, J.Chem.Phys. 62, 3217(1975); 64, 3392(1976)
2 W. Hasenfratz, R. Klein,N. TheodorakopoUTos,Solid State Comm.18,893(1976)
3 See for instance R. Rajaraman, Physics Reports 21,229(1975) or
W. Hasenfratz and ~. Klein, Physica 89A, 191(1977)
4 M.B. Fogel, S.E. Trullinger and A.R.-sTshop, Phys.Lett.59A, 81(1976)
5 L.u. Landau and E.M. Lifshitz, Statistical Physics, (Pergamon Press 1969)

161
Solitary Wave Solutions in a Diatomic Lattice

H. BUttner
Physika1isches Institut, Universitat Bayreuth
D-8580 Bayreuth, Fed. Rep. of Germany, and
H. Bi1z
Max-P1anck-Institut fUr Festkorperforschung
D-7000 Stuttgart, Fed. Rep. of Germany

Recently, it has been shown that the non-linear and anisotropic oxygen po-
1arizabi1ity explains successfully the unusual dynamical properties of oxi-
dic perovskites (I) . In particular the temperature dependence of the fer-
roelectric (transverse optical) soft mode and its coupling to the transverse
acoustical modes in incipient ferroe1ectrics (SrTiO and KTaO ) has been
described by a single quartic electron-phonon coup1fng parameter (localised
at the oxygen lattice site) in the renorma1ised harmonic approximation
(RHA (2) ). This coupling leads, using the adiabatic condition, to a 10ng-
range anharmonic interionic potential along the (IOO)-direction where we
have diatomic chains of alternating transition metal and oxygen ions. In
view of the well-known limits of the RHA (3) we are interested in more ge-
neral solutions of the problem. The structure of the Hamiltonian su~gests
a simplified version which allows analytical solutions while keeping the
essential features of the underlying physics. The simplest model consists
of a quasi-one dimensional diatomic lattice with harmonic and quartic nea-
rest-neighbour interactions. It is shown that this lattice has at least two
kinds of solitary wave solutions differing in their acoustical or optica1-
mode character. The low-energy acoustical excitation obeys a modified Korte-
weg-de Vries equation similar to the 4s01iton in a monoatomic chain (41 The
optical mode is the solution of a ¢ -type wave equation and represents a
'kink-soliton'; it exhibits a close relation to the Krumhansl-Schrieffer
model (5) for structural phase transitions. The group velocity of this soli-
tary wave function is determined by the dispersion of the (harmonic) opti-
cal-phonon branch. This is clearly seen in a lattice with additional next
nearest-neighbour interaction which makes closer contact to the experimen-
tal phonon dispersion curves.
The problem of nonlinear lattices has been studied in recent years quite
extensively (61. The lattices studied so far have, to our knowledge, only
one atom per unit cell. In the following we study a diatomic chain with two
different atoms of masses M, and rv,? The nearest-neighbour harmonic force
constant is denoted by k2 and the nonlinear one by k4 . The classical equa-
tions of motion for the displacements uln and u2n of atoms t11 and M2 are:

MIUln=k2(u2n-uln)+k2(u2n_l-uln)+k4(u2n-uln)3+k4(u2n_l-u ln )3 (I)

In the following we focus on long-wavelength solutions and therefore we use

the continuum-limit approximation in both sublattices (i=I,2):

162
 (3)

A general solution of the resulting partial differential equations would lead

to a definite relation between u, and u2 . The main point in our solution
procedure is to try a general ansRtz for 02n as a function of u1n which, for
k4~' reduces to the harmonic limit and, for equal masses, describes the
k~own solitary wave solution in a monatomic chain. This ansatz reads:

u2n ,v=a v { u1n,v+B1,vduin,v+ B2 , v(d 2/2)U 1n ,v+ ... } ,v=1,2 (4)

with a1 = 1 (acoustical mode) and a 2=-M 1/M 2 (optical mode). The coefficients
Bi v are determined by the condition that both equations, (1) and (2), should
re~ult in the same differential equation for u1
n,v .
For the acoustical mode, we obtain a modified Korteweg-de Vries equation
which is ldentlcal to the one found in (4) for the monoatomic chain except
for the different meaning of the parameters:
2AM 1v1+ k2(B4,1/12)d4vl' + 3k 4 8 2,1 d4vivi = 0 (5)
with v1(x,t)ii.OU1n,1(x,t)/<5X and the parameters: B2 ,1= 2M1/(~11+M2)

= 211/M 2; B4 ,1=4B 2,1 (1-311/(M 1+M 2») ; ;\2=2k 2d2/(M 1+"l2)· p, propagating wave
solution of (5) is given by
v1(x-ct) = ~ (2A~/b)1/2sech ((2;\+c)1/2(x-ct)+ e) (6)
wher: .A+=12A/(k 4B2 1d 4); b=3k 4B2 1/2k2B4 l' and c,e determined by boundary
condl t l O n s . ' "
For t~e optical mode, u1n 2,we find (up to second order in d) the following
equatlOn: '
U1+(2k 2/IJ )u 1+A2u1 +2(k 4/11) (M 1/11 )2 u1 3 =
=-6(k 4/IJ)(M 1/M 2)2 d2[uiul+(M1/M2)(Ui)2U11 (7)

The l.h.s. is id~ntical with the continuum approximation of the Krumhansl-

Schrieffer-mode l5) for displacive phase transitions. The relation to their
parameters is given by (see (6) of( 5) ):
A=2k 2M1/ 11,• B=(k 4/4k 32)A,
3
Co2 =A 2 • (8)

Note that k? is a negative constant in our model. The condition for a dis-
placive phase transition given in (5) of (5] reads, with our parameters,
(9)

This means that, in this simple model, the phase transition is determined
by the mass ratio, only. In our calculation we find additional terms which,
for not too large amplitudes u1 ' define renormalized harmonic force con-
stants and sound velocities:

163
 k2= k2+ 3k 4 (M1/M2)2d 2 (ui)2 , (10)

;:2= 1. 2 + 6 (k4/l.t) (M/M 2 )d2U i. (11 )

These equations may be regarded as the first step in a self-consistent proce-

dure reminiscent of the RHA used in (1) . In the 'paraelectric' regime above
the phase transition, the values of the second terms, in the r.h.s. of (10)
and (1) or their self-consistent analogues, overcompensate the negative va-
lues of the first terms thus leading to positive effective force constants
and sound velocities. In a more general solution these renormalising terms
correspond to a residual coupling between the solitary optical and acousti-
cal solutions. The result of this coupling near the phase transition des-
erves particular interest.
Vienote that the sol itary solutions which are usually visual ized as origina-
ting from double-well potentials at the lattice sites of a monoatomic lattice
(5) may be derived from the translationally-invariant quartic interaction
potential of a diatomic chain.
Furthermore, we have investigated the effect of additional next nearest-neigh-
bour harmonic force constants,a1 and a2. With these force constants we si-
mulate the compensation between the strong attractive Coulomb forces which,
in the bare harmonic approximation, would lead to an instable ferroelectric
transverse mode, (k?<o), and the effective repulsive forces. We can then re-
adjust the effective phonon frequency,~o' of this mode so that it is in ag-
reement with realistic values, say, for srTiO~ or KTa0 1 in the paraelectric
phase at a given temperature. Repeating the aOove calcalations for this
new case leads to the following equation for the acoustic mode:
2>: M1v1+k2(B4,1!12)d4vl'+3k4 2,ld4vivi = 0 (12)
with the new parameters
-A 2= A2 (\1+2(1Y. +a )/k ) (13 )
1 2 2

S 4,1= 64 ,1 ( 1+(2t12/k2)(a2/"-12-~/M1)) (14)

In contrast to the ansatz in (4), here we have used a non-linear relation
between u2n ,1 and u1n ,l:
u2n ,l= t. ~,lul~~i
1
+ 0 (d4/24)(u~2)ul ' (15)

where
(16)

This extension of (4) re-establishes the form of the original modified Kor-
teweg-de Vries equation except for a renormalization of the acoustic sound
velocity and of the parameter 6 4 ,1.
For the optical mode we obtain instead of (7), with (10) and (11),
Ul+(2k2/~) ul+~~Pt u1 +2(k4Mi/~3) ui = 0, (17)

164
where
(18)

Here, we have not considered renormalising terms analogous to those in (10)

and (11). (18) is different from (13) for the sound velocity A. Thus, it
becomes obvious that, for the optical soliton, Ahas no physical relation to
the sound velocity of the system but merely describes the dispersion of the
optical-soliton branch at very long vlavelengths for small ratios k4/k2:
W~Pt(q) = 2k 2fJl _A ~!)t q2+... (19)
I-lith an arbitrary choice of the force constants k?( < 0) ,0:1 (>0) and 0:2(>0) we
may, for example, describe a rather strong disperSion of W t(q) lXopt>A)
while the sound velocity could be pretty flat (~ > A)' op
We note that the condition (9) for a displasice phase transition is now mo-
dified:
(20)
This condition indicate~ that strong mass differences (v-/ith '11+'1?»]J) or
flat optic dispersion (A t«A;0:]0:2>0). favors order-disorder transitions
inde!)endent of the strena~n of tne non-linear coupling parameter k4 .
As our discussion has shown, one might expect that in the acoustical and op-
tical branches of diatomic crystals with short-range forth-order anharmoni-
city solitary wave solutions may appear simultaneously, one acoustical of
the Korteweg-de Vries type (pulse-soliton) and a second 'optical' one of
the Krumhansl-Schrieffer type ('kink'-soliton). As the calculation has ex-
hibited, the two solutions are not completely decoupled but, in particular
for the optical soliton, terms remain which may be treated as renormalisation
effects only for small values of k4' as has been done in (10) and (11).
In summary, it has been shown that a simple one-dimensional diatomic lattice
with quartic nearest-neighbour anharmonicity leads to the simultaneous ap-
pearance of acoustical (pulse) and optical-mode (kink) solitary waves. Since
the model keeps several important features of the dynamical behaviour of
displasive ferroelectrics, such as SrTi0 1 , it is argued that the results ob-
tained in this paper may be of importance for the understanding of phase
transitions in those crystals.
The authors are grateful for discussions and critical comments to A. Bishop,
W. Dieterich. R. Klein, ~!. Kleppmann and R. Zeyher.

References

1 R. Migoni, H. Bilz and D. Bauerle, Phys. Rev. Lett. 37, 1155 (1976),
2 R. ~ligoni, H. Bilz and D. Bauerle, Proc. Int. Conf. Lattice Dynamics,
Paris 1977, Ed. M. Balkanski, Flammarion, p. 650
3 N.S. Gillis, in: Dynamical Properties of Solids, Vol .2., Ed. G.K.Horton
and A.A. Maradudin,(North-Holland, Amsterdam, 1975) p. 105
4 M. Wadati, J. Phys. Soc. Jap. 38, 673 (1975) and 38, 681 (1975)
5 J.A. Krumhansl and J.R. Schrieffer, Phys. Rev. BIT: 3535 (1975)
6 For recent reviews see: ~t. Toda, Phys Reports m, 1 (1975),
A. R. Bishop in: Springer Tracts Mod. Phys., Vo-l,-on Solitons,197S,
Heidelberg

165
Lattice Models of High Velocity Dislocation Motion *

N. Flytzanis
University of Virginia, Charlottesville, Virginia, USA

Abstract
The high velocity motion of a screw dislocation is critically affected by
the discreteness of the medium. The motion is accompanied by the emission
of sound waves following the defect as a wake. The dynamic Peierls stress
s, required to maintain the dislocation in uniform motion is a structured
function of velocity for v2 0.5c but increases monotonically from v:0.5c up
to a critical v:0.9c. The external stress needed deoends very strongly on
the interatomic force law and can correspond to a strain as low as 10-5
for a straight screw dislocation with a piecewise linear force law. The
interaction of the dislocation with phonons is strongly non-linear for
phonons with phase velocity equal to the dislocation velocity. They can
form a phonon-dislocation complex moving with no external stress.

1. Introduction

Plastic flow takes place through the propagation of dislocations which can
be ingrown in the crystal or can be nucleated under the influence of the
applied stress [1,2]. While many flow phenomena can be understood by phe-
nomenological models using continuum elasticity, plastic deformation is
controlled by processes happening on the atomic level, in particular near
the core of the defects where continuum elasticity theory is inadequate.
A realistic description, therefore, of a dislocation in a crystal must take
into account the discrete nature of the medium.

It is clear in principle how to construct a realistic model of a dislo-

cation in a pure crystal. We can separate the crystal in two regions:

*Supported by the National Science Foundation Grant DMR-76-01059

166
(i) the core of the defect that consists of a finite numher of atoms inter-
acting with non-linear forces and (ii) the rest of the crystal. Outside
the core the atomic motions are described using lattice dynamics (3).
All the necessary information is in the frequency spectrum w(R,a) and the
polarization vectorsea. The frequency spectrum can be evaluated directly
from the elastic constants and neutron scattering data. In the core where
the atomic displacements are outside the region of linearity we must solve
the equations of motion with full interatomic forces or use some simple
phenomenological model. For static dislocations many realistic calcula-
tions have been made using various methods [4,5,6). From these one can de-
termine the core energy of the dislocation, the static Peierls barrier that
must be overcome to move a straight dislocation segment from valley to valley
in the periodic structure of the crystal, and the atomic configuration of
the defect. These parameters depend very strongly on the type of inter-
atomic forces used.

In most experimental measurements for the dislocation velocity as a func-

tion of the applied stress one can distinguish three regions [2,7,8): i) the
low velocity region v<10-4c (c is the appropriate speed of sound) where the
motion occurs by kink nucleation and diffusion, ii) an intermediate region
where the velocity increases rapidly with applied stress, and iii) the high
velocity region v>O.lc. It is in this region that steady state motion is
possible and we will concentrate on it in the rest of the paper. The ap-
propriate experiments to study high velocity motion are those using the
tensile test, where dynamic effects due to long range dislocation propagation
are observed in the form of bands of deformation (9). The most direct method
to study single dislocation motion is the use of the etch pit technique [2,10]
on a pulse-stressed crystal, but this measures an average velocity over the
duration of the pulse. On the other hand ultrasonic attenuation experiments
[11,12) sample small displacements of the dislocation lines from equilibrium
and are more relevant to study damping mechanisms at low dislocation veloci-
ties.

There are many mechanisms that can damp the propagation of dislocations.
In fact almost 95 percent of the plastic work done on aluminum crystals for
a large range of strains at room temperature is converted into heat [13].
These processes include (14): damping due to the emission and scattering of
lattice waves, electronic damping, anelastic relaxation as well as interaction

167
with impurities and other defects. In a pure crystal at low temperature
the dominant mechanism is expected to be the creation of phonons because the
lattice presents in general a high barrier that the dislocation must over-
come. This viscous force depends very non-linearly on the dislocation ve-
l ocity.

The theoretical study of moving dislocations is complicated by the fact

that the core of the dislocation is being displaced. There are techniques
to deal with this. We assume an arbitrary motion for the non-linear part
for a given interaction force law. We find the linear motion of the rest of
the crystal, and then we determine the non-linear motions in the core self
consistently.

Here we shall limit ourselves to some simple interatomic force laws in

order to study the effect of the discrete nature of the crystal. In section
2 we will introduce the snapping bond model that has been used in one dimen-
sional [l5) and simple three dimensional dislocation models [l6i17). In the
next section this will be extended to include any force law consisting of
linear segments. The influence of phonons will be examined and the last
section will include a summary of the results as well as a discussion for
future work.

2. Snapping Bond Model (SB)

We generalized the model used for a static screw dislocation in a discrete

lattice [l8) to the ca.se of a moving dislocation with the appropriate cor-
rections for the nonlinearity of the forces in the core region. The crystal
is simple cubic (not a necessary restriction) with nearest neighbor inter-
actions. The only component of the atomic displacements is parallel to the
dislocation axis and win be denoted bv. I.~ m,n where m and n are half integers,
labeling the positions of the atoms in the lattice. The force law between
nearest neighbor atoms (actually rows of atoms) is assumed to be a piece-
wise linear function of the relative displacement D i.e.

F{D) = -A{D-\!b) (2.1 )

where b is the lattice spacing in the direction parallel to the dislocation

(also the magnitude of the Burgers vector) and \! an integer such that

168
 ID-vbl~. As the dislocation moves in the crystal from one valley A to the
next A'~ it breaks all the bonds connecting the two rows between A and A'
across the slip plane and the atoms of these rows are connected to new equi-
librium positions displaced by a lattice spacing. In the process the dislo-
cation creates a disturbance which asymptotically consists of two main con-
tributions [19] (i) a contribution from the k~O waves that is equal to that
of continuum elasticity and has a l/r dependence for the strain,(ii) a con-
tribution due to the emission of phonons (from the core) that have phase
velocity matching that of the dislocation. This radiation field falls off
h
like 1/r 2 and creates oscillations within a Cerenkov anqle behind the dis-
location, while along the caustics the field falls off slowly like 1/r l / 3.
A plot of the atomic displacement at different planes as a function of ~-vt
is shown in Fig. 1.

Fig. 1 Atomic displacements

v~
Wm n for different planes
parallel to the slip plane.
0.92

The crosses (+++) represent

the asymptotic approximation

-14 -12 -10 -8 -6 -4 -2 0

I'-(~ m-vt)

The solution can be written in terms of the Green's function for phonons
with momentum k perpendicular to the dislocation axis, whose Fourier trans-
form is:

(2.2)

where f(k) is the dispersion relation for two dimensional transverse waves.
Due to the coherence of the motion only those phonons will be most effective
that obey the relation

f(k) = v.k (2.3)

169
The Fourier transform of the displacement across the slip plane as a function
of ll=m~vt is

with

G(k ,k )
x y
= G(k,w)1 w -_ +v. ~". (2.5)

and the factor (k _i£)l is the Fourier transform of the step function e(m-vt)
x
(since the sources of the radiation (the snapping bonds) are in front of the
dislocation.) The solution for the displacement can be easily written down

where s is the strain due to the externally aoplied stress and can be ob-
tained from the energy balance condition, i.e. the work done by the external
stress is equal to the energy emitted in the lattice waves. It is given by

(2.7}

where Im(iG(x)) is the density of states of possible emitted Dhonons which

satisfyeq. (2.3) and comes from the vanishing of the denominator in (2.6).
For details of this derivation the reader is referred to reference [16].

For the simple S8 model the strain s can be calculated from the phonon
spectrum of the perfect crystal. A plot of the strain as a function of the
dislocation velocity for a cubic crystal is shown in Fig. 2. The main fea-
ture is that the stress is a nonlinear function of the velocity. There is
a broad minimum near v:.5c. For low velocities there are many unstable
regions, and because of the high strains required it is more likely that
the motion is of the thermally activated type or with large oscillations
in the average velocity. For v>c the strains are finite but enough to cause
a breakdown of the uniform motion. In this phenomenon there are new dis-
locations being created in the wake as was first evidenced in computer simu-
lations of the one-dimensional modified Frenkel-Kontorova model [2n] (MFKM).

170
 0.9r--r-"....--r-+-..,.-......,.-""T""----.-""T""-.--"'"T"-..---r~

A···. ._·.· ·.· · · . . · ·

0.8

0.7
Fig. 2 Plot of the strain
0.6
5
~the 5B model) as a
0.5 function of the dislocation
velocity V (in units of the
0 .• speed of sound c)
0.3

0.2 I
0.1

This is shown in Fig. 3 where the bonds behind the dislocation snap again
for the velocity v=1.2c (i.e. ID-ll>~). While the strains of Fig. 1 are
much higher than possible applied strains (the static Peierls stress is also
high), many of the features of the plot will survive in the extension to a
more general force law to be discussed in the next section. The advantage
from this calculation is that we have the Green's function for the moving
defect, from which in principle we can solve analytically the problem for
any piecewise linear force law.

2.0 1 -------------------,
D
--- - f =0.5

.... + =0.9
- - f=1.2

~ Plot of 0 vs
1.0 ~(=m-vt) for three dis-
'-'
I
location velocities

0.5

0.0 ,7-~;;--__!_;;_-_!;;-~f::_-__:~-_::l:::_--~-~
-6.0 -5.0 -4.0 -3.0 -2.0 -1.0 0.0 1.0. 2.0

m-vt

171
3. Piecewise-Linear (PWL) Force Law

The extension of the lattice dynamical theory of dislocation motion to allow

for arbitrary interatomic force laws in the core reqion, is a non-trivial
problem if one wants to treat self-consistently the couoling between the
highly non-linear motions in the core of the dislocation and the motions of
the rest of the crystal, which are adequately described by discrete linear
elasticity theory. The problem has been treated numerically for the one
dimensional Frenkel-Kontorova model by ISHIOKA [21] for PWL and sinusoidal
substrate forces. More extensive computer simulation studies were performed
by EARMME and WEINER [22] for the same model with emohasis on breakdown
phenomena, and by WEINER and PEAR [23] for a simolifipd two dimensional model
of an edge dislocation. Using a method similar to that applied by l-IEERTMAN
[24] to a smeared out dislocation in a continuum, we can treat the problem
analytically in terms of a shape function h(u) [25,26] (analogous to Weertman's
distribution function) that will be defined below.

For simplicity we limit the discussion to the P\'!L force law, shown in
Fig. 4a, with only 3 linear segments, where Sb is the maximum displacement
for which Hooke's law is valid. The problem is solved once we can determine
the stress-strain relationship across the slip plane as a function of time.
To accomplish this we write the force for the bonds that are stretched beyond
Sb in terms of a dislocation shape function h(u) (different for each non-
equivalent soft bond):
z
F [ D(ll) ] = AbD(ll) + Abj/Uh(U) e(u-ll) ( 3.1)

where ll=m-vt is the distance of the dislocation at time t from the m-th atom
on the slip plane. The force (3.1) is identical to the one in Fig. 4a, once
h(u) is normalized to unity over the range (-z,z) (2z is the dislocation
width) and for Illl<z satisfies the self consistency condition
z
Eh(U)=fdU'g(U-u' )h(u') ,Illl<z (3.2)
-z
where E=-(1-2s)(~/2) and g(u-u') can be obtained from the Green's function,
and in fact equals (~/2) times the derivative of the SB displacement
across the slip plane. Eq. (3.2) is solved numerically by expanding h(u)
in polynomials [25]. Once h(u) is known we can obtain the atomic displacements

172
from the convolution of the source function h(u) with the Green's function
for the defect, D (u-u')
o

D(u)=(s-so)bn
z
+/
h(u') Do(u-u') du'
-z
(3.3)

where Do(u-u') and So are obtained from the solution of the SB model. From
energy balance we can evaluate the external strain s required to maintain
uniform motion,

_ lfdk x 1 H(k )1 2 Im(iG(k ) (3.4)

s-";;7-k- x x
x
where H(k ) is the Fourier transform of h(u). The dependence of the strain
x 2 2
s on 8 comes through IH(kx)l. In particular, s will be small if IH(kx)1
is small for a wave vector kx which corresponds
. to a high density of
states for emitted phonons, i.e. to a large value of Im(iG(k x))'

,,,- (a)

A/3b ,, ,
,
I
/3 112 11-/3) I 0
-A/3b ,
I
Fig. 4 Interatomic force laws and bond
,
I
~ ....I configurations. (A) The solid line is
the PWL (piecewise linear) force law;
the dashed line is the SB (snapping
(b)
bond) force law. (B) The DB (dangling
Allb bond) force law. (C) Schematic dia-
gram of the bonds near the slip plane
I 0
-Allb

(e)

173
 Calculations reported previously [25] show that the stress field of a
moving dislocation for a PWL interatomic force law in the velocity range
0.7c<v<0.92c is not markedly different from the stress field for the snapping-
bond model. Thus the analysis of high-speed dislocation phenomena in this
range can be based on the much simpler SB model with the expectation that the
results will be at least qualitatively correct. It is important that the
calculations be carried out self-consistently within the model chosen; it is
not permissible to choose a velocity - independent dislocation shape function
h{u) from continuum elasticity as done by ISHIOKA [28] and implicitly by
BOFFI et al. [29]. However, this simpl ifying assumption may be justified at
low velocities. The major effect of the softening of the bond law is a phase
shift of the oscillatory displacement field in the wake of the moving dislo-
cation; however the height of the oscillations and the magnitude of the ex-
ternal strain required to maintain the dislocation in uniform motion are
not much changed from the predictions of the snapping bond model.

For 0.92c<v<0.99c it is difficult to achieve complete self-consistency

by the methods used in this paper, since the bonds across the slip plane
are repeatedly stretched beyond the linear limit in the wake of the dislo-
cation. The most interesting questions in this re~ion concern the critical
velocity at which breakdown will occur [20] and which bond will be respon-
sible for the breakdown [28]. The results indicate that breakdown occurs
first for bon"d A (see Fig. 4c) and that the critical velocity is in the
range of velocities indicated, depending nonmonotonically on the value of
S, because of the phase shift of the oscillatory wake field.

For v<0.7 there are noticeable deviations from the SB model. In Fig. 5
we show a graph of s for a velocity v = 0.5; the roints computed for several
values of lIs are connected by straight line segments. In curve Al only the
bonds across the slip plane are treated self consistently. We see that the
strain decreases by orders of magnitude from the SB value (s = 0.5) and dis-
plays sharp minima, which correspond to almost symmetric core configurations
for h{u). The structure, however, disappears when the bond parallel to the
slip plane are treated self-consistently (curve A2 in Fig. 5). We have also
considered a crystal with preferred slip anisotropy, with the bonds perpendi-
cular to the slip plane weaker by a factor K than the bonds parallel to the
sl ip plane [27]. For K = 1/3 the minima in s are even sharper (curve B) and
the motion near the minima is self-consistent. In Fiq. 6 we show a olot of

174
 10 1
" V-0.50
..........-::::.::.,-------
..... , --------
16 2 ,
,
\
' , __ (A2)

\
10 3 \ /--- .....
\ I "
en

10-4
\1
\ 1 '
\ /"'----'"
\/ " (AI)

'-'"
165

106

1/(3

Fig. 5 The external strain s vs. the parameter lis for a dislocation
velocity v = 0.5 and two values for the anisotropy parameter K:
(Al) K = 1, only bonds across the slip plane treated consistently;
(A2) K = 1, all bonds treated consistently; (B) K = 1/3, only bonds
across the slip plane need to be treated consistently

I.~

1.0

.:-c
Q

o.~

0.0

0.4

. .,
_0.2
~
c
~o.o

-0.2

-12 -8 o 4

Fig. 6 Plot of the strain DA(D AA ) of bonds A(AA) (Fig. 4C) vs for
~25 with S = .158 (solid Tine) and S = .500 (dashed line)

175
the displacement of the bonds A and AA (see Fig. 4c) for the SB model and
for the PWL force law with S corresponding to a minimum in the s vs lis
plot, and we see that there are no residual oscillations behind the defect.

The same analysis was also made for a "dangling bond" force law (DB) [27]
(see Fig. 4b) as an approximation to interatomic forces that are negligibly
small outside a small region from the atomic core. The results are similar
to those in the PWL force law. In this way we can bracket a reasonable
range of interatomic force laws with the DB and PWL force laws. The PWL
method can also be extended to fit any desired potential with a linearized
force law containing an arbitrary number of segments, since the computa-
tional effort involved increases slowly with the number of segments used.

We also examined the width of the dislocation core as the force law shape
was modified. The core width (2z) increases almost linearly with S-l. The
slope depends on the velocity and the anisotropy constant K (for
v=0.5 : K=l, 2z ~ 1/(2S); K=1/3, 2z=1/(3S)). Thus the s vs lis plot
(Fig. 5) can also be interpreted as an s vs z plot. The minimum shear
stress required to advance a straight dislocation in a crystal has been
examined by several authors [30,34] with respect to the associated width
w of the dislocation. In general one sees an exponential decrease for the
strain-width relationship Sa exp(-Aw) although A depends'on the particular
model. Discrete models also show a superimposed non-monotonic decrease
in the strain-width relationship for both short range [35] and long range
[36] interatomic forces.

r'lost of the discussion above was at a fixed velocity. But as seen in

Fig. 7 for a fixed interatomic force law there are preferential velocities
for dislocation propagation. Motion at velocities slightly higher than the
one corresponding to the minimum requires a significantly higher stress.

. .. . . . .
I ••

Fig. 7 The external strain s in an

anlsotropic crystal with K = 1/3 as
a function of the dislocation velo-
city v. The circles (0 0 0 0) are
+ + - for S = .500 (lis = 2.00 and the
crosses (+ + + +) for S = .255
(lis = 3.92)
-

I 0-' L-_...I..-_.....L.._--'-_--'
0.0 02 04 0.6 0.8
V

176
4. Oi s 1oca t i on- Phonon Interact ion

Various phonon mechanisms of dislocation damping have been studied: thermo-

elastic losses [37.38]. phonon viscosity [39]. phonon wind [40.41] and flut-
ter effect [42]. A review on this subject gives a critical discussion of
calculations based on these different mechanisms [43]. Calculations of pho-
non scattering from the dislocation in a continuum approach taking into ac-
count the phonon relaxation which includes as special cases the effects of the
phonon wind. phonon viscosity and thermoelastic losses have been performed
[44.45] and it was concluded that at low temperatures relax~tion processes
can be neglected as compared to scattering processes. At high velocities.
however. the scattering mechanism must be included along with the emission
of phonons (since this is the dominant) necessitating thus the study of a
discrete model.

In the framework of perturbation theory FLYTZANIS and CELLI [46] studied

the propagation of a dislocation in a crystal at low temperature under the
action of a constant applied stress to balance the frictional stress due to
the emission and scattering of phonons. which causes a nonuniformity in the
motion. It was found that the response of the dislocation to the scattered
waves depends on the difference of IVp-vl between the phase velocity vp of
the waves and the velocity of the dislocation. If Ivo-vl/v 20.1 the response
of the dislocation can be treated as a perturbation and the frictional stress
consists of two terms: the emission term modified by a Oebye-Waller factor
and a term proportional to the temperature. due to the scattering of the pho-
nons which act as a damping mechanism. For phonons that are phase matched
i.e. with vp=v. however. the response is divergent in the perturbation ap-
proach and must be considered in a fully nonlinear treatment [47]. The set
of phase-matched phonons is identical with the set that is radiated by the
dislocation. carrying away energy in the wake behind the dislocation. An
analysis of the contributions to this radiative field is given in [19]. The
predominant phonon field inside the wake for v = 0.50 and K=1/3 has wave
number (k .k ) = (4. l75.~). which will be used for the remainder of this dis-
x y
cussion. These phonons can inhibit or help the dislocation propagation de-
pending on their phase. This could be particularly important near the minima
found in the previous section for the applied strain (s~10-5) which could be
supplied by lattice waves having an amplitude ~ s. obtaining thus a disloca-
tion phonon complex. The problem is treated along the lines of section 3.

177
except that now the displacement across the slip plane D also includes that
due to the lattice waves (~P),

~P=-4IAkl sin(llk/¢k) sin(k/2) (4.1)

where 1Akl and ¢k are the amplitude and phase of the lattice wave (kx,ky ).
Then we must solve the consistency conditions for h(u), which includes im-
p1icity, but in a correct way, the nonlinear interaction between the wave and
the dislocation. Some sample calculations are presented for 1/6=3.33(5=0.0052
at IAkl=O). Fig. 8 shows that for fixed phase, the external strain 5 depends

15 .75
V=0.50
10 K= 1/3 (8) .70
/3=0.30 00
o. 0

.65
., (A)
N
+
o 0 0 0
o .60
><
0
•? •
en
-5 .55
0
000

-10
</>=1.4 7r A= .0052
.50

o 10 0.0 0.5 1.0 1.5 2.0

</>/.".

Fig. 8 Plot of the strain 5 and the half-width z vs phonon amplitude

(~Iith fixed phase) and vs phase (for fixed amplitude)

linearly on the phonon amplitude and for fixed amplitude the strain has a
roughly sinusoidal dependence on ¢ and can be driven to zero if IAKI is large
enough. A similar dependence is seen for the core half-width z. Table 1
shows that the influence of the phonon as measured by ~s/2 and ~z/2 is strong-
ly affected by the parameter 6 [~s(~z) is the maximum difference in strain
(half-wi dth) values generated by varyi ng ¢ from 0 to 2'ITJ. A phonon of fi xed
amplitude IAKI = 0.001 can change the strain by ~s/2=127.47 x 10- 5 for S =.300,
but by only ~s/2=8.18 x 10- 5 for B = .255, corresponding to the first minimum
in the solid curve (K = 1/3) of Fig. 5. In this preliminary work, consid-
erations are limited to the single dominant phonon, but the method is easily
extended to any superposition of phonons of various amplitudes, phases and
wave vectors (as long as w = vk). The combination of perturbative and non
x
linear methods described leave only the treatment of near resonance phonons
to provide a complete picture of the dislocation-phonon interaction.

178
 Table 1 Effect of phonon on external strain

Sxl0 5 z
(no (no
lis phonon) ~ x 10 5 t:,z
ohonon) 2 2"
2.00 4494.7 .0000 400.00 .0000
3.33 520.11 .60425 127.47 .0159
3.77 27.55 .83540 22.59 .0069
3.85 7.23 .8814 7.53 .0026
3.88 3.19 .9007 3.07 .0011
3.92 1. 47 .9299 8.18 .0030
4.00 7.78 .9793 21.00 .0082
4.55 91.32 1. 2517 54.61 .0245
5.00 51.99 1.4430 36.57 .0237
5.88 11.48 1 .8954 20.93 .0270
6.35 24.63 2.1008 23.07 .0363

5. Discussion

Linear continuum elasticity predicts no radiation loss for a velocity v<l

(in units of the speed of sound), but this is simply because all effects due
to the discreteness of the medium (includinq the Peierls stress) are absent
in this theory. It is not surprising therefore, that attempts by ESHELBY
[48] using the Peierls-Nabarro model with the inclusion of dispersion pre-
dicted the effect is small. The Peierls-Nabarro model allows for the periodic
nature of the force across the slip plane that joins two elastic half-spaces
held together by non-Hookean forces. When the dislocation width is small the
defect behaves like a particle whose potential energy varies periodically.
Calculations to evaluate the damping due to the non-uniformity of the motion
[49,50] are also based on continuum elasticity. The relevant conclusion is
that for velocities v>O.l this radiation is small.

The radiation losses of a moving dislocation in a perfect crystal at zero

temperature are due to the emission of phonons resulting from the rearrange-
ment of atoms in the core of the dislocation. We find that the radiation
losses depend very strongly on the velocity (for v < 0.5) and on the inter-
atomic force law. We can conclude that for a wide ranqe of interatomic forces

179
there are velocities where nearly stress-free modes of motion can occur which
are stable and will be preferentially selected by the dislocation, if high
enough velocities can be attained. The one-dimensional Frenkel-Kontorova
model gives rise to the Sine-Gordon equation, if displacement differences
are replaced by derivatives; this equation has "soliton" solutions [51,52]
that represent a dislocation in uniform, loss-free motion at zero stress for
any v < 1. Keeping finite differences, EARMME and WEINER [53] discovered
that loss-free motion can still occur in a modified Frenkel-Kontorova model.
Wake-free motion is also observed in the computer simulations of HOOVER et al.
[54] for the motion of a dislocation in a 2-dimensional triangular lattice.

Even the single phonon considered has a pronounced non-linear effect on

the resulting motion. The results presented indicate that a phonon can have
a "lubricating" effect, carrying along the dislocation in the absence of an
applied stress or even against the direction of motion favored by such a
stress. (In such a case, the moving ope absorbs energy from the phonon
field, leaving behind an "antiwake".) Such results might be expected on
simple intuitive grounds, but have never been shown to follow in a self-
consistent way from a lattice model. One can describe the ope as a type
of elementary excitation of the non-linear, discrete medium, analogous to
soliton solutions in non-linear continuous media. Recent computer simula-
tions show that temperature can have an enhancing effect on the mobility
of a dislocation segment, perhaps due to the phenomenon of thermal energy
trapping by the moving dislocation. [55J

It should be pointed out that there is a characteristic difference of the

damping mechanisms in the low velocity and high velocity regions of disloca-
tion motion. In general it is not easy to demonstrate the relative importance
of a given mechanism and in most cases more than one mechanism might be con-
tributing to the resistance to uniform motion. For a real crystal one must
use the macroscopic description of kinetic rate theory with phenomenological
parameters obtained from microscopic model calculations. Whether the high
velocity motion studied here occurs will ultimately depend on the structure
of a real crystal that can favor activation of highly correlated slip pro-
cesses or complex dislocation interactions due to cross slip. The materials
to look for high velocity motion are the high yield alloys [9J. To under-
stand these experiments, however, we must study the interaction of disloca-
tions in moving arrays and the influence on their motion of substitutional
impurities. The former has been studied extensively in continuum models
but some discrete models have also been considered [56J. In [56J it was
found that it is much easier to move two dislocations because one can ride
on the wake of the leading dislocation. Several studies of the impurity
damping have been based on the string model, which considers the dislocation
line as a flexible string being pinned at the impurity sites. This model
has met both successes [57J and failures [58J in the low velocity region.
If the rate of unpinning, however, is high one must consider the dynamic
interaction of the defect with the impurity [59,60J.

180
References

1. W.T. Read Jr., Dislocations in Crystals (McGraw-Hill, New York, 1953).

2. W.G. Johnston and J.J. Gilman, J. Appl. Phys. 30, 129 (1959).
3. A.A. Maradudin, E.W. Montroll and G.H. Weiss, Solid State Physics Supple-
ment 3, Theory of Lattice Dynamics in the Harmonic Approximation (Aca-
demic Press. New York 1963).
4. Z.S. Basinski, M.S. Duesbery and R. Taylor, Phil. Mag. n, 1201 (1970).
5. V. Vitek, R.C. Perrin and D.K. Bowen, Phil. Mag. n, 1049 (1970).
6. J.W. Christian and V. Vitek, Rep. Prog. Phys. 33, 307 (1970).
7. E. Yu. Gutmanans, E.M. Nadgornyi and A.V. Stepanov, Soviet Phys. Solid
State 11, 3081 (1970).
8. H. Araki and T. Ninomiya, J. Phys. Soc. Japan il, 1684 (1976).
9. R.B. Schwartz and J.W. Michell, Phys. Rev. B9, 3292 (1974).
10. K.M. Jassby and T. Vreeland, Jr., Acta Met. 20, 611 (1972).
11. A. V. Granato and K. L~' cke, J. App 1. Phys. 'fl., 583 (1956).
12. A. Hikata., R.A. Johnson and C. Elbaum Phys. Rev. Letters 24, 215 (1970).
13. W.S. Farren and G.I. Taylor, Proc. Roy. Soc. lQI, 422 (1925).
14. F.R.N. Nabarro, Theory of Crystal Dislocations (Clarendon Press, Oxford,
1964) .
15. W. Atkinson and N. Cabrera, Phys. Rev. 138A, 763 (1965).
16. V. Celli, and N. Flytzanis, J. Appl. Phys. 11,443 (1970).
17. S. Ishioka, J. Phys. Soc. Japan, 30, 232 (1971).
18. A.A. Maradudin, J. Phys. Chem. Solids 2, 1 (1958).
19. V. Celli, N. Flytzanis and S. Crowley, J. Phys. Chem. Solids E, 1125
(1976) .
20. Y.Y. Earmme and J.H. Weiner, Phys. Rev. Lett. 11, 1055 (1973).
21. S. Ishioka J. Phys. Soc. Japan, 34 462 (1973).
22. Y.Y. Earmme and J.H. Weiner, J. Apol. Phys. 45,603 (1974).
23. J.H. Weiner and M. Pear, Phil. Mag. 11, 1055 (1975).
24. S. Weertman, Symp. Math. Theory Dislocation. American Society of Mechan-
ical Engineers. Northwestern University (1969).
25. N. Flytzanis, S. Crowley, and V. Celli, J. Phys. Chem. Solids, 38,
539 (1976).
26. N. Flytzani s, S. Crowl ey and V. Celli, Phys. Rev. Lett. 12., 891 (1977).
27. S. Crowley, N. Flytzanis and V. Celli, J. Phys. Chem. Solids, to be
published.
28. S. Ishioka, J. Phys. Chem. Solids ~, 427 (1935).

181
29. S. Boffi, G. Caglioti, G. Rizzi, and F. Rossitto, J. Appl. Phys., 45
3220 (1974).
30. A.J. Foreman, M.A. Jaswon, and J.K. '4ood, Proc. Phys. Soc. (London)
A64, 156 (1951).
31. A.H. Cottrell, Dislocations and Plastic Flow in Crystals (Oxford Univer-
sity Press, New York 1953) o. 64.
32. R. Hobart, Jour. Appl. Phys., 36, 1944 (1965).
33. V.L. Indenbom, Soviet Phys. - Cryst. 1, 193 (1958).
34. W.T.O. Sanders, Phys. Rev., ~, 1540 (1962).
35. S. Ishioka, J. Phys. Soc. Japan, ]§., 187 (1974).
36. T. Kurosawa, J. Phys. Soc. Japan 11 153 (1958).
37. J.D. Eshelby, Proc. Roy. Soc., A197, 369 (1949).
38. J.H. Weiner, J. Appl. Phys. 29, 1305 (1958).
39. W.P. Mason, J. Aopl. Phys. ~, 2779 (1964).
40. G. Leibfried, Zs. Phys. 1lI, 344 (1950).
41. J. Lothe, J. App1. Phys. 33, 2116 (1962).
42. T. ;linomiya, J. Phys. Soc. Japan, 25,830 (1968).
43. V.I. Al'shitz and V.L. Indenbcm Sov. Phys. Usp. ~, 1 (1975).
44. A.D. Brailsford, J. Appl. Phys. 43, 1380 (1972).
45. V.I. Al 'shitz, A.G. Malshukov, ZETP Ql, 1849 (1972).
46. N. Flytzanis and V. Celli, J. Appl. Phys. 43 3301 (1972).
47. S. Crowley, N. Flytzanis and V. Celli, submitted to J. Phys. Chern. Solids.
48. J.D. Eshelby, Proc. Phys.Soc. London A62, 307 (1949).
49. E.W. Hart, Phys. Rev. 98 1775 (1955).
50. V.I. Al 'shitz, V.L. Indenbom and A.A. Shtol 'berg JETP, 33, 1240 (1971).
51. A. Kochendorfer, A. Seeger and H. Donth Zeits. Phys.,~, 533 (1950);
DQ, 321 (1951); lli, 173 (1953).
52. A.C. Scott, Am. Journ. Phys. 1[, 52 (1969).
53. Y.Y. Earmme and J.H. Weiner, Phys. Rev. Lett. 33, 1550 (1974).
54. 14.G. Hoover, N.E. Hoover and 1-i.C. Moss, Phys. Lett. 63A, 324 (1977).
55. J.H. Weiner, A. Hikata and C. Elbaum, Phys. Rev. ~, 531 (1976).
56. N. Flytzanis and V. Celli, J. Aop1. Phys. 45, 5176 (1974).
57. A.V. Granato, Phys. Rev. Lett. 'fl, 660 (1971).
58. A. Hikata, and C. Elbaum, Phys. Rev. B9 4529 (1974).
59. J. Takamura and T. Morimoto, J. Phys. Soc. Japan~, Suppl. 1, 28 (1963).
60. A. Ookawa and K. Yazu, J. Phys. Soc. Japan ~, Suppl. 1, 36 (1963).

182
Grain Boundaries as Solitary Waves

Ralph J. Harrison and George H. Bishop, Jr.

Army Materials and Mechanics Research Center, Watertown, MA 02172, USA
and
Sidney Yip and Thomas Kwok
Nuclear Engineering Department, Massachusetts Institute of Technology
Cambridge, MA 02139, USA

In computer molecular dynamics "experiments" on high angle symmetric

tilt grain boundaries we have discovered that these boundaries exhibit many
features attributed to solitons. For example, the grain boundaries have a
geometrically stable motion which is nonlinear in that it requires a finite
temperature to activate and it persists over times long compared to other
relaxation times. The analogy with dislocation motion naturally comes to
mind since low angle grain boundaries may be regarded as rows of dislocations
and there is extensive literature [1] on the soliton aspects of dislocations.
If we had started out to look for soliton motion we might well have started
out with low angle grain boundaries; however, we were originally interested
in using dynamic simulation methods to obtain the thermodynamic properties
of the metallurgically more important high angle grain boundaries [2]. We
attempted to measure the excess boundary entropy by observing the change
relative to the perfect crystal in the spectrum of the velocity autocorrela-
tion function for atoms near the boundary. However, we found changes far
from the boundary as well, and in tracking this down we discovered that the
two crystals were undergoing a relative sliding motion which was coupled with
migration of the boundary [3]. This is the motion we call soliton-like and
we shall describe this motion in some detail in the following.
Figure 1 gives the arrangement of the two crysta11ographica11y distinct
layers of atoms in our model bicrysta1 simulating a coincidence boundary in
an fcc crystal. Coincidence boundaries are formed at those special angles
of rotation such that a super lattice is formed by crystal lattice points
common to a crystal lattice and the rotated lattice. The corresponding
y

+ +
.+' + + Fig. 1 Projection on a {l00} plane
+ + of the {310} fcc twin grain boundary
+ ' ...... .......
t't-'
...... ~. + atomic sites in successive {100}

"-r\ -+ . . . ~.1 ............. layers are shown by dots and crosses .

.'" "'.
---1 The line patterns drawn, the "kite"
. .'" +.
I ' ,\
+. , ~. , 1"'. ~
and zig-zag are mnemonics useful for
.;'
recognition of the boundary structure
+ +
+ + +
+ + x

183
superlattice is the coincidence site lattice (CSL) [4]. The boundary is
characterized by periodicity in the transverse x direction. Periodicity in
the z direction is associated with the ABABA .. [001] stacking sequence of
the fcc lattice. The grain boundary pattern of Fig. 1 is one which results
from intuitive geometrical concepts supported by static energy minimization
procedures [5]; it also resembles "bubble raft" models of boundaries. Fig. 2a
represents essentially the same pattern except that circles are drawn to
represent the atoms and the centers are at the actual positions obtained by
the static relaxation method. In Fig. 2 one has an additional "pattern rec-
ognition" aid, namely the shapes of the interstices. Fig. 2b-d represent the
time development of the motion after random velocities were imparted to each
atom in Fig. 2a resulting in a simulated temperature of 0.2 energy units
defined by the depth of the Lennard-Jones potential. Short time averaging
has been done in order to eliminate some of the high frequency motions. One
notes that individual atoms in the top crystal move to the right, while those
in the bottom move to the left. In the course of this motion the boundary
has moved downward, although individual atoms have not. That is, the two
crystals have undergone sliding relative to each other, while the crystal has
migrated downward. In this particular case the coupled sliding and migration
has produced a migratory motion equal in magnitude to the sliding motion.
The fine grid drawn in Fig. 1 is the DSC lattice introduced by Bollman [6].
It is the coarsest lattice which contains both the crystal lattice and the
rotated lattice as sublattices and it is relevant to the description of the
periodic character of the observed grain boundary motions. When the two bi-
crystals have slid relative to one another by [100] DSC, the boundary has
also migrated exactly [010] DSC. Fig. 2b shows the motion after about 1/2
[100] DSC as one may judge from the interstices near the boundary; if one
would pinpoint the boundary it seems to be about half way between the two
distorted kites. We have computed the static Peierls barrier to the boundary
motion for Lennard-Jones potentials. The barrier exhibits the periodicity
of the DSC lattice. It has a shallow minimum at the symmetric position, a
deeper minimum at a displacement of 1/8 [100] DSC and a maximum at 1/2 [100]
DSC. Apart from the exact details, the noteworthy features of the barrier
are that the heights of the peaks and depths of the valleys are much less
than for the barrier to shear in a perfect crystal because of the fact that
the DSC lattice has a much finer grid than does the crystal lattice, so that
the boundary atoms never have to move very far from their minimum energy po-
sitions. Another consideration is that there is some "free volume" at the
boundary. The fact of thermal activation of boundary motion is consistent
with the computed static Peierls barrier.
Figure 3 shows the time dependence in another computer run of the relative
coordinates of the centers of mass of each crystal. The large sliding motion
is in the x coordinate, whereas the y and z coordinates simply show oscilla-
tions associated with phonon-like modes of the computational cell. Other
similar runs show: an incubation period before the boundary starts to move;
the coupled sliding and migration mode with motion in a direction statistic-
ally chosen; the motion stops for a while; then starts again in the same or
reverse dlrection. The velocity when moving is a few percent of the velocity
of sound, with the temperature dependence not well established yet, although
at very low temperatures there is no sliding or migration at all. The com-
putational time steps shown in the figure are 0.015 using the dimensionless
unit of time which is approximately the longitudinal speed of sound divided
by the lattice parameter. Characteristic times for reestablishing "equilibrium"
in computer molecular dynamics runs on single crystals after a parameter such

184
 b

~2 Atom positions in the vicinity

or-tne grain boundary. The two layers
in <100> fcc stacking sequence are
shown separately (left and right).
(a) Starting structure. (b)-(d)
Positions at selected later times

~ The time variation in the x, y,

and z coordinates of the center of mass
of groups of 30 atoms in the center of
each bicrystal. The sliding is revealed
by the drift in the x coordinates. The
y and z coordinates show ordinary thermal ·2 2 4

fluctuations " X In unIts of [100J DSC .

185
as volume or temperature is changed are typically several hundred computa-
tional time steps [7J, i.e., much shorter than the typical time for persis-
tence of the grain boundary motion. The boundary motions may be compared to
Brownian motion [llJ, resembling also that of domain walls [12J.
In addition to the boundary described above we have studied two-dimensional
grain boundaries [8J which also exhibit thermally activated motions. We have
also carried out some exploratory studies where boundary motion was initiated
without random thermal motions simply by starting the crystals with relative
sliding motions of various magnitudes. We may also mention that the motion
we observed of the grain boundary between a pair of three-dimensional fcc
crystals is crystallographically the same as in the formation of a deformation
twin and similar to that in the martensitic transformation [9J. Special inter-
facial dislocations (e.g., the 1/6 [112J twinning dislocation) have been
thought to be involved in effecting the associated lattice rotations locally.
One would expect a martensitic interphase soliton to differ geometrically
from the interphase soliton proposed by HOROVITZ et al. [lOJ for the bcc to
w phase transformation in Zr-Nb alloys. There are many new and fascinating
phenomena to be explored by computer molecular dynamics techniques having
relevance to soliton theory as well as to metallurgy.
Acknowledgment S.Y. and T.K. acknowledge support for part of their work by
the United States Army Research Office under Grant No. DAA29-78-C-0006.

References

1. Some basic and some more recent papers in this area are: A. Kochend8rfer
and A. Seeger, Z. Phys., 127, 533 (1950); A. Seeger and A. Kochendorfer,
ibid. 130, 321 (1951); A. Seeger, H. Donth, and A. Kochendorfer, ibid.
134, 173 (1953); N. Flytzanis, J. Crowley, and V. Celli, Phys. Rev. Levtt.,
39,891 (1977); Y. Y. Earmme and J. H. Weiner, J. Appl. Phys., 48, 3317
(1977).
2. R. J. Harrison, J. A. Cox, G. H. Bishop, Jr., and S. Yip, Nucl. Metal.,
20,604 (1976); G. H. Bishop, Jr., G. A. Bruggeman, R. J. Harrison,
J. A. Cox, and S. Yip, Nucl. Metal., 20, 522 (1976).
3. G. H. Bishop, Jr., R. J. Harrison, T. Kwok, and S. Yip, Trans. Amer. Nucl.
Soc., 27, 323 (1977).
4. G. Friedel, Lecons de Crystallographic, Gauthier-Villars, Paris (1926);
M. L. Kronberg and F. H. Wilson, Trans. AIME, 185, 501 (1949).
5. R. J. Harrison, G. A. Bruggeman, and G. H. Bishop, Jr., Grain Boundary
Structure and Properties, ed., G. A. Chadwick and D. A. Smith, p. 45,
Academic Press, N. Y. and London (1976); G. A. Bruggeman, G. H. Bishop, Jr.,
J. A. Cox, and R. J. Harrison, Nucl. Metal., 20, 450 (1976).
6. W. Bollman, Crystal Defects and Cr~stal Interfaces, Springer Verlag,
New York, Heidelberg, Berlin (1970).
7. O. Deutsch, Ph.D. Thesis, Dept. of Nuclear Engineering, M.I.T. (1975).
8. T. Kwok, M.S. Thesis, Dept. of Nuclear Engineering, M.I.T. (1978).
9. S. Mahajan and D. F. Williams, Int. Metall. Rev., 18, 43 (1973); J. W.
Cahn, Act. Met., 25, 721 and 1021 (1977).
10. B. Horovitz, J. L. Murray, and J. A. Krumhansl, Bull. Amer. Phys. Soc.,
23, 274 (1978).
11. R. J. Harrison, G. H. Bishop, Jr., S. Yip, and T. Kwok, Bull. Amer. Phys.
Soc., 23, 253 (1978).
12. T. R. Koehler, A. R. Bishop, J. A. Krumhansl, and J. R. Schrieffer,
Solid State Comm., 17, 1515 (1975).

186
The Relation of Solitons to Polaritons in Coupled Systems

D.F. Nelson
Bell Laboratories, Murray Hill, NJ 07974, USA

Soliton solutions in acoustic modes [1] and in optic modes [IJ of solids have been studied in the
past. Envelope soliton solutions of coupled exciton-electromagnetic mode systems [2J and coupled
two-level atom-electromagnetic mode systems [IJ have also been studied. In this paper we wish to
present soliton solutions for coupled optic mode-electromagnetic mode systems (for which linear
sinusoidal excitations are called po/anlolls) and for coupled optic mode-acoustic mode systems.
Since the mathematical derivations are similar in the two cases, we will present only the former one
in detail.

The physical origin of the soliton solutIOn in either system can be described as follows. Con-
sider an optic mode whose potential energy consists of a term quadratic in the optic mode ampli-
tude and a term quartic in the amplitude. We assume each term produces a restorIng force though
the restoring force from the quadratic term is regarded as weak. In other words, we are considering
a soft mode above the phase transition temperature. If the transition is a ferroelectric-paraelectric
phase transition, we will consider an electromagnetic wave coupled to the optic mode; if the transi-
tion is a ferroelastic-paraelastic phase transition, we will consider an acoustic wave coupled to the
optic mode. In each case the coupling term to the traveling wave can cause the quadratic term to
reverse its sign and so become non-restoring. Thus during the interaction the effective potential
energy of the optic mode has two minima. As is well known, such a double-well potential energy
allows soliton solutions.

Optic Mode-Electromagnetic Mode Solitons

We model the interaction of an optic mode with an electromagnetic wave by the electric field wave
equation for the scalar amplitude E of an electric field eigenmode,
a 2E Kh a 2E q a 2y (J)
fhT-?ii?~ E"C 2 fi72'
and by the force equation for the scalar normal coordinate y of an optic mode,
lI1a 2y/al 2~ -A mY - A 40y J + qE . (2)

Here K h is the dielectric constant at frequencies high compared to the resonant frequency of the
optic mode, q and 111 are the charge and mass densities associated with the Optlc mode, and A 20 and
A 40 are constants characterizing the linear and nonlinear restoring forces of the optic mode. We
assume A 20 and A 40 are both positive; hence the potential energy of the optic mode has a single
minimum at y ~ 0 and the unperturbed state of the crystal has y ~ 0 and E ~ O.

We now search for a stationary pulse solution by transforming to a traveling coordinate sys-
tem characterized by i; ~ Z - vI. The wave equation can then be integrated twice to yield
E ~ v2qY/E,,(c2_KhV2) (3)

since the integration constants must be taken as zero to satisfy the boundary conditions,
(4)

187
The optic mode equation then becomes

mv 2a2y/ae = -A;uY - A 4(tY J (S)

where
A 20 == A 20V;;( v/- v2)/ v/( v;;- v2) (6)

v;; == c2/ K" ' v/ == c2/ K I (7)

K I == K" + q 2/ E "A 20 . (S)

Here KI is the dielectric constant at frequencies low compared to the optic mode resonance.

For soliton solutions of (S) to exist it is necessary for

A 20 < 0 (9)

which requires
(10)

This allowed range o./soliton velocities coinCides with the.forbldden range o./polariton phase velocities.

Eq. (S) can now be integrated twice with the aid of (4) and (9) to obtain the soliton solution
y=y"sechk(~-~"l , (lJ)

E qy" () (12)
= (2 2) sechk ~-~"
Eo V - VI!

with the optic mode amplitude y" given by

y" == Akv/1r , (13)
the pseudo-wavenumber k given by
k = (wdv)(v2-v/)1/2/(V;;-V2) III (14)

and ~" being an integration constant. Here A is an area constant,

A == f_:= ydt = 1r(2m/A 4ol l /2 , (IS)

depending on the nonlinearity parameter A 40 and w Land w T are the longitudinal and transverse
optic frequencies
(16)
[wdwrr = (VI/VI)2= K/K"
(17)
Wr = (A 2(/m) 112

Eq. (16) is the Lyddane-Sachs- Teller relation. OrdinarilY the velocities v" and VI are
regarded as limiting velocities measured on the high and low frequency sides of the optic mode
resonances which produces a transmission stop-band for frequencies w such that w T < w < w L'
Examination of the soliton solution, (1 J)-(l4), reveals an alternate interpretation of the LST rela-
tion, namely, that v" is the velocity of the largest amplitude, narrowest width soliton and that VI is
the velocity of the lowest amplitude, broadest width soliton.

Eq. (14) is the dispersion relation of these solitons; it relates the pseudo-wavenumber k and
the velocity v. Surprisingly we find that this relation depends only on linear properties of the
medium. Thus the dispersion relation oIan inherently nonlinear wave contains no nonlinear properties.

The dispersion relation can be put into a more suggestive form if we define a pseudo-
./i"equency by w == kv. Eq. (14) can then be rearranged into
k 2 = (w/v,,)2(wl+w 2)/(w i-+w 2) (IS)

188
This should be compared to the dispersion relation. for single frequency plane wave solutions of (I)
and (2) (polaritons) in the absence of nonlinearity (A 41J = 0) which is

(J 9)

where the wavenumber K and frequency H p of the polariton have their usual meaning
"
[E - expi(Kpz-nl't)l. It can be seen that the replacement Kp~ ik, np~ iw in the polariton
dispersion relation yields the soliton dispersion relation. Thus, if Kp and n p in (J 9) are regarded
as complex, the soliton dispersion relation corresponds to the imaginary wavenumber, imaginary frequency
soilitions qlthe polariton dispersion relation The figure illustrates the soliton and polariton dispersion
relations.

/
UPPER POLARITON /
/
/
Vi /
/ Plot of absolute values of complex
/
/ wavenumber and complex fre-
/
-r- - - quency. For the polariton branches
/
,.--,--SOLITON BRANCH
IKI = K p , Inl = np with K p , HI'
governed by (J 9); for the soliton
branches IKI = k, Inl = w with k,
w governed by (18)

u;;----LOWER POLARITON BRANCH

Optic Mode-Acoustic Mode Solitons

The interaction of an optic mode and an acoustic mode can be modeled by the acoustic wave equa-
tion,
(20)
and the optic mode force equation,
ma 2y/at 2 == -A 2oY - A40Y J - A Ilau/ax (21)
In these equations u is the scalar amplitude of an acoustic eigenmode, X is the material coordinate
in the direction of propagation, p is the mass density of the medium, A 02 is the elastic stiffness
from all sources except the optic mode normal coordinate y, A II is the coupling constant between y
and the displacement gradient au/ax, and all other quantities have the same meanings as in the last
section. Piezoelectric coupling to an electric field E could be added to each of these equations along
with the addition of the divergence of electric displacement equation. Since these additions make
no qualitatively change in the conclusions, we omit them for simplicity.

By manipulations analogous to those in the last section the soliton solution of these equa-
tions is found to be
(22)

au AllY" () (23)
!lX = (2 2) sechk ~-~ u
V p V - v"
with ~ == X-vtand (10), (J3)-(J6) (with the second equality in (J6) dropped) holding once again.
The defini tions in (7) must be replaced by
(24)

189
and the definition in (! 7) must be replaced by
WL == (A 2r/ m) 1/2 . (25)

The replacement of (! 7) by (25) is initially very puzzling because (a) (A 2r/ m) 1/2 was called
the transverse optic frequency in the last section since it gave a pole in the dielectric constant and
(b) here (A 2r/ m) 1/2 is called the longitudinal optic frequency even though it gives a pole in the
elastic stiffness. The resolution of this follows from the fact that the elastic compliance (which is
the inverse of the elastic stiffness) is analogous to the dielectric constant since the linear wave velo-
city is proportional to the inverse square root of each of these quantities in the respective cases.
Hence (25) is the frequency at which the compliance is zero and so is rightfully identified as a long-
itudinal optic frequency.

All remarks concerning the soliton dispersion relation in the last section apply also to the
coupled optic mode-acoustic mode system here.

References

1. See references contained in A. C. Scott, F. Y. F. Chu and D. W. McLaughlin: Proc. IEEE 61,
1443-1483 (1973)
2. S. A. Moskalenko, V. A. Sinyak and P. I. Khadzhi: Sov. J. Quant. Electron. 6, 464-465
(J 976)

190
Solitons in CsNiF3: Their Experimental Evidence
and Their Thermodynamics

M. Steiner
Hahn-Meitner-Institut, Glienicker Str. 100
0-1000 Berlin 39, Fed. Rep. of Germany, and
J.K. Kjems
Research Establishment Ris¢, DK-4000 Roskilde, Denmark

Abstract

An inelastic neutron scattering study of the 1-D-ferromagnet CSNiF3 in a

magnetic field yields strong evidence for the existence of thermally
activated solitons for 14 K ~ T ~ 5 K and 2 kG ~ H ~ 36 kG. Comparing the
experimental data with a classical continuum-soliton theory developed by
Mikeska for this particular system very good agreement was found, e.g. the
activation energy was found experimentally as 28 K and theoretically as
34 K.

1. Introduction

In the last few years it has been shown, that soliton or kink solutions of
the equations of motions of the constituents of a system undergoing a phase
transition might be of importance [1,2,3,4,5] . Most of these theoretical
treatments have been done for onedimensional systems since exact solutions
exist in this case for the nonlinear equation of motions, the Sine-Gordon
equation for example. It is possible to show that the dynamics of a 1-D-
ferromagnet with planar anisotropy can be described by the Sine-Gordon
equation in the classical continuum limit if a magnetic field is applied
perpendicular to the chain direction (Mikeska 1978 [1]).

In a series of papers it has been shown that CSNiF3 is an ideal model sy~
tem to study the predicted solitons [6]. Its magnetic properties are deter-
mined by magnetic chains of Ni 2 + running along the hexagonal c-axis, the in-
trachain interaction being a hundred times the interchain interaction. The
magnetic moments are kept in the a-b-plane by a strong single site ani-
sotropy. There is no anisotropy measurable in the a-b-plane. The following
Hamiltonian is apprcpriat.e for CSNiF 3 in the temperature range T ~ 3 K,
TN being 2.7 K:
....., x
?f; -2J ~ S.
:L
S.
:L
(1)

with the following constants as determined by inelastic neutron scattering:

J k ; 11.8 K; A/k ; 9 K and S ; 1 [7].

191
 2. Experiments

We studied a single crystal of about 1 cm3 on a cold source - triple axis

spectrometer at the DR3-reactor in Ris¢ for 3.1 K ~ T ~ 14 K and
o ~ H ~ 36 kG. In order to reduce incoherent background the crystal con-
tained only the 58Ni-isotrope. Measurements were performed in constant
Q-scans around the (0,0,2) reciprocal lattice point especially at (0,0,1.9)
or at 0.1 r.l.u. In order to extract the magnetic contribution one has to
subtract the still existent incoherent background. The determination of this
background is crucial. We used as background a measurement at 3.1 K and
36 kG at (0,0,1.9), under which conditions no solitons should exist. This
background measurement was then subtracted from the different measurements
at constant H = 5 kG for 3.1 K ~ T ~ 14 K and at constant T = 9.3 K and
o ~ H ~ 20 kG. The remaining spectrum consists of three peaks: two spin-
wave peaks (creation and anihilation of spin waves) at ± 0.55 meV and a
peak centered around zero energy. In order to extract the reliable para-
meters we tried to describe the experimental spectra by a line shape of the
following form: Two spinwave peaks as a convolution of the resolution and
a Lorentzian line centered around the spinwave energy and in addition a
Gaussian line centered around zero energy. The parameters like widths,
position and intensity of the peaks were determined by a least squares
fitting. Some of the results for H = 5 kG and T = 6.3 K and 10.1 K are
shown in Fig. 1 demonstrating the good agreement between the theoretical
line shape and the experimental results. It can be seen as well how
strongly the peak around zero energy increases with increasing temperature.

SOD DO

sou.
Fig.l Experimental spectra after
background subtraction for H = 5 kG,
200." q = 0.1 r.l.u., T = 6.3 K (upper
half) and T = 10.1 K (lower half).
The abszissa and ordinates are given
"".to
in units of meV and counts per
25 min respectively

stUO

.ouo

300.DO

..
200.11

lOD.OO

...
'.10
-1.10 0,00 1.50 1.10

192
3. Discussion

Since we want to check the soliton theory of Mikeska, we concentrate on the

temperature and field dependence of the peak around zero energy. The
neutron cross-section for scattering from solitons is given by Mikeska as

S (Q , c.') ".,
/3e."'"(.d.L
2 C;9 1T e
-~:'f" ""9/2," )2.
~ sin h11'$,1", (2 )
with m = (g~ H/2J) 1/2, c = S (4AJ) 1/2 and g = 2.4, S 1, J/k 11.8 K,
A/k = 4.5 K ~since the theory is a classical one).

This formula is analogous to scattering for neutrons from a gas of real

particles with a thermal velocity distribution and does not describe the
excitation of a single soliton. Thus the solitons show up in a neutron
scattering experiment as a gas of noninteracting quasiparticles moving
along the ferromagnetic chain with a velocity v ~ c, (c being the slope
of the spinwave dispersion). We see from formula (2) that the intensity
is governed by the activation energy 8 m and the width for the peak can
be calculated without adjustable parameter. The last term reflects the
spatial extension of the solitons. Generally, all our results agree at
least qualitatively with Mikeska's result: intensity J versus temperature
at fixed field (H = 5 kG), J versus field at field temperature (9.3 K),
halfwidth versus temperature and field. We want to present here the
J(T)-results only, the other results will be publis~ed in another paper
shortly. The intensity is simply given by fif exp- 8m l:1 for H = 5 kG,
q = 0.1 r.l.u •• Thus by plotting ln (J versus l/T one should find a VT)
straight line with slope 8 m. In Fig.2 the experimental results are shown
together with a line representing the theoretical slope (full line). The
best line through the experimental points is given by the broken line.
It is obvious that the experiments show the expected exponential tem-
perature dependence for T ? 5 K but the activation enery is slightly
less then the theoretical value as calculated from the known parameters:

8 mexp = 27 K; 8 mtheor = 34 K

CsNiF,
Q; H=5kG
"6 10
u 8 ~

-e.
,
V1 \

•
6 \
0
\
l;: Fig.2 Semi log-plot of the
x
\
Z:- integrated intensity times ~
Vi
c ~ versus l/T. Full line: theory
.l!! \
~ \ (scaled at 14 K); broken line:
1 \~ best straight line through
".l!!
0 .8 \ experimental points
t.f; .6
\
\
•
0
Jnverse Temperature
IK-' )

193
We thus conclude, that solitons are an important part of the dynamics in
CSNiF 3 and can be regarded as noninteracting quasiparticles which are
thermally activated. We were able to show that the thermodynamics of the
solitons are as predicted. These first direct experimental studies of
solitons hopefully stimulate theoretical treatments which look for the
discrepancies we found between the theory and the experimental results.

References

1. H.J. Mikeska: J. Phys. C 11, L 29 (1978)

2. J.A. Krummhansl, J.R. Schrieffer: Phys. Rev. B~, 3535 (1975)
3. N. Gupta, B. Sutherland: Phys. Rev. A 14, 1790 (1976)
4. R.A. Guyer, M.D. Miller: Phys. Rev. A 17, 1205 (1978)
5. J.F. CUrrie, M.B. Fogl, F.L. Palmer: Phys. Rev. A 16, 796 (1977)
6. M. Steiner, J. Villain, C. Windsor: Adv. Phys. ~,137 (1976)
and references therein
7. M. Steiner, J.K. Kjems: J. Phys. C~, 2665 (1977)

194
 Structure and Stability of Domain Walls - Phase Transition

J. Lajzerowicz and J.J. Niez*

Laboratoire de Spectrometrie Physique, B.P. 53
F-38041 Grenoble Cedex, France
*a 1so with
Departement de Recherche Fondamentale,
Centre d'Etudes Nucleaires de Grenoble, B.P. 85 X
F-38041 Grenoble Cedex, France

1. Introduction

We have shown [IJ that a ferromagnetic domain wall can have two configura-
tions, one corresponding to an ISing domain wall (for large uniaxial aniso-
tropy) and the other corresponding to a Bloch type d~m~in wall. This change
of configuration is a two dimensional phase transition, the order parameter
being the chirality. In this paper, we want to study the structure and sta-
bility of a ferroelectric domain wall when two types of order are competiti-
ve, for example Ferroelectric and Antiferroelectric phas.e transitions.

2. The Model
One uses a Landau Free Energy [2,3J to describe the occurence of Ferroelec-
tricity and Antiferroelectricity. Let us call PI and P2 the polarizations of
the two sublattices and take an energy of the form :

-+-
( 'Pol.
I
-pl
+ ;J.
)~ (--1)

Let us introduce two new variables, the Ferroelectric and Antiferroelectric

ones :

/J.. = I (1)
)
V2.

The new form of the Free Energy is :

cf = ~ [(vFf + (VA)t] +
-I- t ('P~-t At.)2. (3)

The phase diagram is given in fig. 1 where in the ferroelectric phase A = 0;

F = ± 1- !a+bYd and where in the antiferroelectric one F = 0; A = ± /b-=a/d.

195
So if b varies we have a first order phase transition from a ferroelectric
to an antiferroelectric at b = 0. Such a situation can be found in a crystal
like Pb(Zq_x Ti x)03 [2].

3. Structure of a Domain Wall

We can study it in the ferro or in the antiferroelectric phase but the si-
tuations are symmetric so we will study the ferroelectric domain wall. The
EULER-LAGRANGE equations are :
C \/2 F _ d (F.l-t p,2.) ~ = 0
(If-)
C. '12. A _ d (1="".1.+ A~) A =0
-t

If A = 0. we have a ferroelectric domain wall

Fo (x) ~ ~ - ~+ b t nh
0. X V_~ t-;
4. Stability of the Domain Wall
We study small excitations around the static solution (5). If we associate
an effective density for the ferroelectric and the anti ferroelectric compo-
nents. Pf and Pa (for simplicity we take Pf = Pa = p) and if we write

J--
We have in first order

PW2. F", = _ c r F, + (a.+b)F, _3(Cl+b) F, L 12

[<>''''1}'' --
co. ...

2. <:.
b
(")
f W 1.. A \ = _ c \l Z.A I -+ (a.-b)A, _ (a.+ b ) A, to.. '" 1,2 x \y.1_ o..? b
::2.. C.

It is easy to find the spectra of these Shrodinger like equations. The sys-
tem will be stable if the eigenvalues of these equations are positive. For
the first one. they are all positive. for the second one we can easily write
the lowest eigenvalue

So the domain wall is stable against antiferroelectric fluctuations if a > 3b

and is unstable against such fluctuations if 3 b > a. For woo we have a soft
mode behaviour near the threshold 3 b = a. therefore the domain wall changes
its configuration in a cooperative way.

196
5. Structure of the Domain Wall in the New Configuration

We know an exact solution of the system (4) including non linear terms [4].
We have two cases

F = '-/_ ~+Jr. taro h X '1- ~+cb

A-:. 0

_~) 3b - 0.. > \) t=' -:. '1--a:b Ca.ro ~ )l. y_.2~

A-:. y3~-~
1c~x F-~
We see that the amplitude of the anti ferroelectric component is the order
parameter in our problem, so the domain wall exhibits an anti ferroelectric
component.
It is interesting to note that when b goes to zero, near the ferroelectric-
anti ferroelectric transition the width of the antiferroelectric component
diverges showing that, at this first order phase transition, the nuclea-
tion of the new phase occures in the domain wall.
Remark I
The situation is equivalent in the antiferroelectric phase where we have a
ferroelectric component in the antiferroelectric domain wall. Such a struc-
ture must be sensitive when subjected to a field gradient. If we want to
compare it with what happens in the ferroelectric it gives an effective
field

E =

when ~ is the width of the domain wall le/-2b. This gives very low values
for the effective field.
Remark II

Such a situation can, of course, be found for ferromagnetic-antiferromagne-

tic transition.

197
 b~

Figure

Region F corresponds to the ferroelectric phase

Region A corresponds to the aotiferroelectric phase
Region P corresponds to the oar~electric phase.
Tn the region I the domain wall of the ferroelectric phase exhibits an
anti ferroelectric character.
In the region II. the domain wall of the antiferroelectric phase exhibits
a ferroelectric character

References

J. Lajzerowicz and J.J. Niez

Phase transition in a domain wall (to be oublishedl

M.E. Lines and A.M. Glass

PrinCiples and Applications of Ferroelectrics and Related Materials
Clarendon Press. Oxford (19771

[3] C. Kittel
Phys. Rev .• ~. 729 [19511

[4] S. Sarker. S.c. Trullinger. A.R. Bishop

Phys. Lett. A, 59. 255 [19761.

198
Periodic Lattice Distortions and Charge Density Waves
in One- and Two-Dimensional Systems

R.H. Friend
Cavendish Laboratory, Cambridge, Great Britain

At low temperatures, low dimp.nsional metals are commonly found to exhibit

superlattices incommensurate with the primitive lattice (periodic lattice
distortion coupled to charge density waves, PLD - CDhT). Transitions to
such states are driven by strong electron phonon coupling and favoured by
the particular forms of the Fermi surfaces of these metals. Experimental
investigations of the structural distortions, and of the changes in elec-
tronic structure around the Fermi level produced by the CDhT - PLD state,
will be discussed for the layered dichalcogenides, the linear chain plati-
num complex KCP, and the organic charge transfer salts.

The basic instability of a one-dimensional metal to a distortion with a

wavevector that opens an energy gap at the Fermi level, the Peierls distor-
tion, has been known for some time (1,2). However, it is only in the past
few years that physical systems that appear to meet the conditions for such
a distortion have been discovered. The study of these systems has revealed
a fascinating wealth of phenomena, including the importance of one-dimension-
al fluctuations, the role of three-dimensional coupling, and the interaction
between the lattice periodicity and that of the Peierls supperlattice.

In this review paper I shall first review the basic theoretical concepts
and then discuss the properties of the three groups of materials that have
been most intensively investigated, the linear chain platinum complexes,
(KCP), the organic charge transfer salts (TTF-TCNQ) and the layered transition
metal dichalcogenides. These compounds have been extensively reviewed (3-8).

In the case of a one-dimensional metal, for which the Fermi surface consists
of two planes at wavevectors -k F and +kF' an applied potential, Vq with wave-
vector 2kF, will introduce an energy gap in the electronic spectrum at the
Fermi level. Since occupied states will have their energies lowered by this
potential, and only unoccupied states are raised in energy, the overall elec-
tronic kinetic energy is lowered. hThen the periodic potential, Vq is derived
from a periodic distortion of the lattice, at T;O the lowering of the electronic
kinetic energy will dominate the lattice distortion energy to give a distorted
semiconducting state with a superlattice of wavevector 2kF.

Following the mean field treatment of RICE and STRASSLER (9) if the electron-
phonon coupling is of the form
1 L L + (1)
H c (b -b + )
e-p IN p q g(q) cp+q p q -q

in which the c's and b's denote annihilation and creation operators

199
for the electronic Bloch states and phonon states, and g(q) denotes the elec-
tron-phonon coupling constant, the effect of the conduction electrons on the
bare normal mode frequency ~(q) is to change it to
2 2 2 (2)
w (q) = wo(q) (1 - 2 (g (q)/*wo(q)X(q,T»

where X(q,T) is the Lindhard function,

E fk f k _q
X(q,T) (3)
k

(f k is the Fermi function for Bloch state k and Ek is the energy of state k).

For the case of a one-dimensional conductor, X (2kF ,T) diverges as

N/EF) R,n(EF/kT) at low temperatures, where N(E F) is the density of states
at the Fermi level. At a temperature Tp ' the phonon frequency is renormalized
to zero, and for T<Tp ' a static distortion of the lattice will be present.

kT 'V EF exp (- I / A) (4)

p
where A, the dimensionless electron-phonon coupling constant is given by
A g2(q) N(E F)
(5)
\>I wo(q)
Below Tp ' the energy gap in the electronic spectrum builds up, following a
BCS type temperature dependence, and at T=O,

2 t:, =3.5kT p (6)

In this model the energy gap in the electronic spectrum is produced by a

periodic lattice distortion (PLD). The conduction electrons will attempt
to screen the periodic potential set up by the atomic displacements, creating
a charge density wave (CDW) in the conduction electron density. As emphas-
ised by FRIEDEL (10), the charge density wave stiffens the lattice and makes
its instability less easy to produce. Thus if vo(ext.) is the potential
due to a phonon, the total pot~~tial acting on th; electrons is

Vq = Vq (ext.) + Vq (int) (7)

defining the dielectric constant, Eq such that VqE q Vq (ext.)
4 e2
E = 1 - -2- X (q,T) (8)
q q
The effect of the CDW on the formation of the PLD-CDW state is to modify the
stability criterion (I.t(q) o in (2» to
2
2 g (q) >
1
\>I "b (q) £
q
X(q,T) (9)

Including the effect of electron-electron repulsion for the conduction elec-

trons further stiffens the phonon. If U is a short range Coul,omb repulsion,
equation (9) is modified to

2g2(q)
- U > (10)
X(q,T)

200
A similar criterion, including also an exchange interaxion between conduc-
tion electrons, has been obtained by CHAN and HEINE (11) using a variat-
ional technique.

Any real physical compound must of course be three dimensional, and even
though extremely anisotropic electronic properties are possible in chain-
like structures, as is the case for KCP and TTF-TCNQ, interchain coupling
will distort the Fermi surface planes of a one-dimensional metal. The
warping of the planar Fermi surface will tend to prevent any single super-
lattice wavevector introducing a gap over the whole Fermi surface, and will
eventually suppress the Peierls transition (12). Characterising the inter-
chain coupling by a transfer integral SL ' it is clear that if kTp>tL ' the
formation of the PLD-CDW state is unaffected by t L . This criterion is
satisfied for KCP and TTF-TCNQ, although other related charge transfer
salts show distortions which are depressed by tL ' and are further weakened
by pressure induced increases in tL ' (13).

The other class of compounds showing PLD-CDW formation that I shall discuss
here is the layered structure transition metal dichalcogenides. These
materials have effectively two-dimensional Fermi surfaces, with little band
dispersion out of the plane of the layer. Large contributions to X(q,T)
in (3) can arise from areas of Fermi surface which are separated from each
other by a single wavevector. This condition is termed Fermi surface
"nesting". Calculations ofX (q, T) (I5, 16) based on band str~cture calcul-
ations for the layer compounds (16,17) indicate only weak nesting, and as
stressed by DORAN et al (18), it is the strong electron phonon coupling in
these compounds that is important in driving the distortion. It is no
coincidence that many of these compounds which distort but remain metallic
below the distortion are superconductors (19).

The mean field treatment of the Peierls distortion outlined above produced
a phase transition at finite temperature in spite of the theorem that one-
dimensional systems without long range forces cannot undergo phase transi-
tions at finite temperatures. This is because,by considering one phonon mode
only, the short range electron-phonon interation was effectively turned
into a long range sinusoidal force of wavevector 2kF. If all other Fourier
components are included, the short range character of the force is restored
and the phase transition is suppressed. This has been treated by LEE et al
(20) using a Ginzburg-Landau expansion of the free energy. One-dimensional
fluctuations are very sensitive to interchain, three dimensional coupling,
and the effectiveness of interchain tunnelling coupling in suppressing fluc-
tuatio~s has been considered by RICE and STRASSLER (21) and HOROVITZ et al
(12). Coulomb coupling between CDW's on adjacent chains has also been pro-
posed as a means of establishing three-dimensional long range order (22).
Of the three series of compounds discussed later, only in KCP is behaviour
strongly influenced by fluctuations, al though TTF-TCNQ remains controver-
sial (HEEGER in (4) presents the case for their importance).

In this model of the Peierls distortio~ the zero-temperature Peierls gap

is destroyed at high temperatures by thermal excitation of electrons across
the gap, giving the relationship between 6 and Tp in eqn. (6). For the
layer compounds, where this relation is not obeyed (the factor of 3.5 in
(6) is found to be much larger), McMILLAN (23) has developed a short coher-
ence length model ( ~ ~superlattice period), consistent with the poorly

201
defined peaks calculated in X(q) and the very broad Kohn anomaly in the
phonon spectrum, extending over", 1/3 Brillouin zone (24). Since mrmy modes
are involved, the phonon entropy dominates the electron entropy at finite tem-
perature and determines the value of T .
P
One of the interesting aspects of the Peierls distortion is that the super-
lattice wavevector is determined by the details of the Fermi surface, and
may bear no simple relation to the lattice periodicity. If the superlat-
tice wavelength A = na where a is the lattice dimension and n integer, the
PLD-CDW is said to be a commensurate superlattice, and will have preferred
positions with respect to the lattice and will be unable to move without
surmounting an energy barrier. If however A f na, the super-lattice is
termed incommensurate, and if the PLD-CDW is a simple sinusoidal distortion
then it has no preferred position in the lattice, and if not impeded by too
many impurities, it will be free to move. Such movement, which would carry
current, is the collective mode envi.saged by FROHLICH as a mechanism for
superconductivity (2). It appears that any contribution to the conductivity
from the FrUhlich collective mode is unimportant in the systems I shall
discuss in this paper, however measurements of non-linear conductivity in
the quasi one-dimensional system NbSe3 (25), in which an incommensurate
superlattice has recently been discovered (26) have rekindled interest
in this conduction mechanism.

WhenA is close, but not equal to na, there will be competition between
the electronic bandstructure energy which favours the best nesting wave-
vector, and the commensurability energy gained when the PLD-CDW sits in pre-
ferred positions in the lattice. The commensurate superlattice is fav-
oured for large values of the distortion amplitude, and a sequence of
transitions, first from the undistorted stated to an incommensurate super-
lattice, then at a lower temperature to a commensurate superlattice is
commonly observed for the layered compounds, when the incommensurate
superlattice is close to commensurability, and the weak Fermi surface
nesting reduces the band structure energy penalty in moving away from the
best nesting wavevector. For the quasi one-dimensional systems where
nesting is better, and the distortion amplitudes smaller, a lock-in to the
commmensurate superlattice is less likely, although it has recently been
observed in TTF-TCNQ under pressure (27). This sequence of phase transi-
tions has been modelled using a Landau theory by McMILLAN (28).

As an alternative to the incommensurate phase characterised by a simple

sinusoidal distortion, McMILLAN (29) has proposed that whenA is close to
na the PLD-CDW may dissociate into regions of locally commensurate
superlattice and regions where the phase changes rapidly with respect
to the lattice, introducing harmonics of the superlattice period. The
regions of rapidly changing phase, termed discommensurations by McMILLAN
can be regarded as solitons. Evidence for their existence in the layer
compounds will be discussed later.

202
K2 (Pt(CN)4) Br O. 30 . 3H 20, usually abbreviated to KCP, is one of a family
of non integral oxidation state platinum complexes of cyanide or oxalate
investigated by Krogmann and co-workers. Their properties are extensively
reviewed in (3).
A much simplified sketch of the backbone of the structure is shown in Fig. 1.

KCP

1C)14:---_~:----::-_ _=':_----l
o 10 20 30 40
'OOO/T 1<"'

Fig. 1 Idealised sketches of a Fig. 2 Conductivity parallel

~- group, and of the stacking and perpendicular to the chains
of Pt(CN)~- groups in a KCP crystal, for KCP, after (30)
showing the overlap of the d z 2 atomic
orbitals based on the Pt atoms

The planar arrangement of cyanide anions around the platinum cations allows
stacking along the c axis and overlap between the atomic d z 2 platinum
orbitals on adjacent ions. Not shown are the positions of the K+,Br-
and H20; however the presence of 0.30 bromide ions per formula unit is
crucial for the conduction properties, as they take electrons out of

203
what would have been the filled platinum d z 2 band, and leave it with only
1.7 electrons per platinum ion. Thus in the absence of any distortion
the material should be metallic along the chain axis. The conductivity
measured parallel to the chains is several hundred (!l cm) -I, and the ani-
sotropy ratioCf,,/(J
1 is about 5 x 10 4 at room temperature (30). Below
room temperature toe conductivity falls rapidly, as is shown in Fig.2, and
X-ray and neutron diffraction studies indicate the appearance of an incom-
mensurate superlattice, with a chain axis repeat of 6.7 lattice units, which
is the value of 2kF expected for the 1.7. electron filled band (31). A
large Kohn anomaly can be seen in the longitudinal acoustic phonon branch
at room temperature, shown in Fig. 3.

3.0

Fig. 3 Acoustic phonon branches in KCP at room temperature (31). An

enhanced Kohn anomaly appears along the line q=2kF=const

Elastic neutron scattering is also observed at all off symmetry directions

studied which have a 2kF component of the wavevector parallel to the chains,
demonstrating the presence of a PLD together with the Kohn anomaly. Below
140K, the intensity of the superlattice point (rr/d1 , rr /d ,2kF) increases
rapidly and then saturates at around 80K. The intensiti~s of other 2kF
points, such as (0,0,2kF) fall in this temperature region, but do not reach
zero. This indicates that whilst the coherence of the distortion with
transverse period 2di increase rapidly below 140K, the system never attains
long range order. X-ray experiments measure the low temperature coherence
length to be some 60 Pt-Pt spacings. The choice of (rr/d~, rr/di, 2kF)
suits both Coulomb (22) and tunnelling (12) interchain energy considerations.

Measurements of the infra-red reflectivity show what is interpreted as the

Peierls gap, at an energy of~ 0.2 eV (30). While this gap is most appar-
ent at low temperatures, it is still present at room temperature. The value
of 0.2eV implies a mean field transition of some 800K, from equation (6)

204
and it is evident that KCP is a system where the transition is suppressed
by one-dimensional fluctuations. The random potential produced by the
partial occupation of the Br- and H20 lattice sites may be responsible for
the absence of true long range order at low temperatures, but it is inter-
esting to note that under pressure, when presumably interchain coupling is
strengthened, a sharp anomaly in the resistivity as a function of tempera-
ture is observed (32). The anomaly is similar to that seen at the three-
dimensional ordering transitions in the organic charge transfer salts
(see Fig. 5), and indicates that there is a transition to a three-dimensional
ordered state at low temperatures at pressures in excess of 20kbar.

ORGANIC CHARGE TRANSFER SALTS

The organic charge transfer salts, d which tetrathiofulvalenium tetra-

cyanoquinodimethanide (TTF-TCNQ) is the most studied, arehighly anisotropic,
or quasi one-dimensional metals in which overlap between TI orbitals of
planar molecules stacked one on top of another results in band formation,
and del.ocalisation of electrons along these stacks. TTF and TCNQ molecules
are shown in Fig. 4a, and the three-dimensional structure of TTF-TCNQ is
shown schematically in Fig. 4b. Segregrated stacks of TTF and TCNQ are
formed along the b axis, and as determined from the value of Peierls dis-
tortion wavevector, there is charge transfer of .59 electrons per molecule
from the TTF to the TCNQ stack.
N N
C C

1\ Vc
10lDI
5,,/5

i--- ~------\ \-'"'

5/ "'5
W c
1\C
c

1001]
N N
reNQ
TTF iljjllilli'IPfiittP
TTF T C NQ

Fig 4a Structure of TTF & TCNQ molecules Fig 4b Schematic representation

of the TTF-TCNQ structure, show-
ing the tilted stacks of TTF &
TCNQ molecules along the b axis

In a one electron one-dimensional band model, the Fermi wavevector, and

therefore the charge transfer, is determined by the value of k at which
the bonding TCNQ band crosses the antibonding TTF band. The transfer
of charge puts .59 electrons into the TCNQ band, leaving the TTF band with

205
1.41 electrons, and with both bands partially filled, both stacks contEi-
bute to the metallic conductivity, at room temperature some 500 (n cm) I
along the b axis (33). Coupling between stacks is weak, the conductivity
anisotropy is of order 500 at room temperature (33), and the tunnelling
integral between stacks is estimated from the NMR inters tack relaxation
time to be some 5meV (34). The conductivity increases on cooling, reach-
ing a maximum for most crystals of between 10 and 25 times the room temp-
erature value at around 56K (35). Larger measured increases have been
interpreted in terms of a FrBhlich mode contribution to the conductivity
(36), however these measurements have been criticised by Schafer et al
(37), and Thomas et al (35) consider that the majority of conductivity
measurements are consistent with single particle conduction mechanisms.

20 TCK)

TTF - TCNQ 1.67

TS.F-TCNQ

E
u
c::

HMTSF _TCNO---
P=lIokbar

SN.
k~~~7~2.6~2~.5~2.'~2~.3~2.2~2.~1-2~
,+-----~~~~~~~ 1.4 1.2 1 0.8 0.6
10 20 30 40 SO 60 70 80 h.- 3.6 , CONSTANT)

Fig. 5 Temperature dependence of the Fig. 6 Microdensitometer

resistivity along the high conductivity readings of a series of X-ray
axis for several Id conductors (34) patterns between 60K.and room
temperature for TTF-TCNQ. Note
the different temperature depend-
ence of the 2kF and 4kF scattering
(38)

206
Below the conductivity maximum, TTF-TCNQ becomes semiconducting, showing
a series of phase transitions, two of which, at 53K dnd 38K are visible as
peaks in the slope of the resistivity versus temperature curve shown in
Fig. 5. X-ray and neutron scattering measurements have been hampered by
the small size of crystals available, the elegant measurements of the struc-
tural distortions taking place are reviewed by Comes (38). Diffuse X-ray
scattering at temperatures above the distortions show appreciable phonon
softening at . :'Y.5b t " which is interpreted as the 2kF phonon. Scattering
at twice this wavevector is also seen in the X-ray pattern shown in Fig. 6
(.59b*, or 41b* in the reduced zone). This 4kF scattering has a different
temperature dependence from the 2kF, being relatively stronger at room
temperature, and it cannot be considered as a harmonic of the 2kF distortion.
The sequence of phase transitions seen at low temperatures is associated with
varying stages of the three-dimensional ordering of the Peierls distortions
on the two stacks. At 53K, the 2kF scattering condenses to give a super-
lattice of (2a,3.~b,c). Many measurements including ESR (39) and C13NMR (40)
indicate that this corresponds to the formation of a Peierls distortion on
the TCNQ stacks. The transverse period of 2a can leave the TTF stacks un-
distorted (41) and it is only below 49K that the distortion builds up on them,
and the transverse period increases from 2a. Finally at 38K there is a
first order transition where the transverse period locks in to a value of
4a.

A controversial issue for TTF-TCNQ is the validity of the free electron

model. The maximum calculated values of the bandwidths of either stack
is only !eV (42), whereas the value of the on site Coulomb repulsion, U is
10 times larger (43). Explanations for the 4kF scattering are usually
based on a large U model for one or both stacks (38,44,45). It should
be noted that 4kF scattering is not observed in most other TCNQ salts,
such as the selenium analogue TSF-TCNQ (46), although it has b~en recently
reported to be present in NMP-TCNQ (47) which is also considered to be a
system where correlation effects are important.

At atmospheric pressure none of the charge transfer salts have a value of

2kF close enough to a commensurate superstructure for there to be a lock-in
transition. However, recent high pressure measurements (27) indicate that
at 20kbar, charge transfer in TTF-TCNQ has increased to 2/3, giving a super-
lattice with period 3b. The range over which the PLD-CDW is commensurate
is limited to bk/k < 1%, which is consistent with the almost perfect one-
dimensional Fermi surface (kTp '\, t 1)' and contrasts with the larger commen-
surability ranges seen in the layer compounds.

LAYERED TRANSITION ~lliTAL DICHALCOGENIDES

The group V transition metal dichalcogenides, MX2 are layered structure

materials, composed of close packed sheets of metal atoms sandwiched by
layers of chalcogen atoms. Bonding between chalcogen layers in neighbour-
ing sandwiches is weak, and these compounds, which are d band metals, show
two dimensional metallic properties, although the anisotropy between the
in-plane and out-of-plane conductivities is typically only 30 (48).

Coordination of the metal by the chalcogens can be either octahedral

207
or trigonal prismatic; many ways of stacking the sandwiches, of either
or both coordination types, are possible, and there are many polytypes found.
Fig. 7 shows some of the more common of them, the simplest octahedral co-
ordination polytype, with only one layer per unit cell, IT, the simplest
trigonal prismatic polytype with two layers per unit cell, 2H, and the
mixed coordination polytype, 4Hb •

lP lP ~
~

Ii!
1T
~
~
•
0
s,Se

~
Ta,Nb
2H

X
4Hb

Fig. 7 Structures of some of the Group V dichalocogenide polytypes

Physical properties are mainly determined by the coordination type, in

fact the properties of IT and 2H materials are sufficiently different for
it to be simplest to treat them separately. The Fermi level in 2H mater-
ials lies in the middle of a d subband of d z 2 , dxy and dx 2-y2 character
with a hybridisation gap between it and the rest of the d band. This
results in a high N(EF) and a complicated Fermi surface, whereas the d
band for the IT polytypes is not split, N(EF) is lower and the Fermi
surface simpler (16,17).

Structural distortions, not necessarily related to Fermi surface nesting,

are a general feature of this class of materials; all the Te compounds
are distorted from the hexagonal layer structure (49) and even the Group
IV material TiSe2 which has only a low carrier density from a small p-d
band overlap exhibits a low temperature superstructure in which the in-
plane and out-of-plane cell constants are doubled (50). The common
feature for these systems is strong electron-phonon coupling, seen for
instance in the high infra-red effective charge in TiSe2 (51). Nevertheless
the Sand Se Group V compounds show a series of incommensurate and commen-
surate distortions for which it does seem possible to relate the superlat-
tice period to Fermi surface nesting in the calculated band structures (14,
15).

208
 10- 1 r-----------------------------------,
IT-Ta~

2H -TaSt2

10-7~----~~--~~----~L-----~----~
o 100 200 300 400
T(K)

Fig 8. The electrical resist~v~ty parallel to the layers of several

layered conpounds (F.J. Di Salvo in (4»

Resistivity versus temperature curves for both IT and 2H polytypes are

shown in Fig. 8. The resistivity anomalies clearly visible all corres-
pond to transitions in the PLD-CDW structure; since not all the Fermi
surface is destroyed by PLD-CDW formation they remain metallic at all
temperatures, in contrast to KCP and TTF-TCNQ (the rise in the low temp-
erature resisitivity in IT TaS2 is the result of falling carrier mobility
from impurity induced localisation rather than falling carrier concentrat-
ion (52).)

IT POLYTYPES

The two IT polytypes of Ta, TaS2 and TaSe2 both show strong distortions,
with both commensurate and incommensurate structures at high and low temp-
eratures. Fig. 9 shows electron diffraction photographs of IT TaS2 at
various temperatures. The simple hexagonal diffraction pattern of the
undistorted lattice is present at 600K (by which temperature there is an
irreversible transition to the 2H polytype). The three phases of super-
lattice shown in the other three photographs are bounded by the two resis-
tivity transitions in Fig. 8, one at 350K, the other centred at 200K with
~ 20K hysteresis. Above 350K, in the ITI phas~ an incommensurate structure

209
 200
.. K . 320K
All ..

320K
ill ..
1T1 - 600K 2H
Metal II • Metal

Fig.9 Transmission electron diffraction patterns for the 3 superlattice

phase~, and undistorted phase of IT TaS 2 (53)

210
with superlattice spots in line with the principal spots is seen. As for
all the layer compounds, with hexagonal or trigonal symmetry in the layer,
a triple PLD-CDW is observed, with three plane wave PLD-CDW's at 1200 to
one another. Between 350K and 200K, in the IT2 phase, the superlattice
reflections have rotated away from the line of the principal reflections.
Although it is the tendency to form the commensurate superstructure, the
1T3 phase, that must rotate the superlattice in this regime, the super-
lattice is not commensurate. The commensurate structure is only achieved
below 200K, by rotation of the superlattice by 13 0 54', so that the axes of
the hexagonal supercell are 3a + b (a and b in-plane lattice vectors),
or 113 lal. IT TaSe2 does not exhibIt the-IT2 phase, and the lock-in
from a similar ITI type phase to thel13 superlattice occurs at 475 K (19)

158 179 209 230

14

\
40 209 230 329
--
371

Fig. 10 XPS from Ta 4f7/2 levels in IT TaS2 taken during a cooling

and heating cycle, showing hysteresis at the 200K transition. The
spitting between the two peaks at the lowest temperatures is ~ leV (56)

The distootions in the IT polytypes are large, with atomic displacements

of ~ .25 A (54). One of the more spectacular indications of the changes
taking place is X-ray photoemission from the Ta 4f5/2 and 4f7/2 levels (55,
56). Changes in the charge density around the Ta sites create splittings
in the energy levels of up tOleV. WERTHEIM et al (55) estimate that such
a large splitting corresponds to a charge redistribution between sites of
as much as one electron. Fig. 10 shows measurements at various tempera-
tures through the two transitions in IT TaS2. In the IT3 phase HUGHES and
f .AR (56) argue that the 13 metal sites in the supercell are distributed
into 3 inequivalent types, in the ratio 6:6:1, and the two peaks arise
from the two more common sites. What is intriguing is that even in the
1T2 phase, when the superlattice is incommensurate, the two peaks are still
resolved; this must indicate that the superlattice is locally commensurate,

211
and invites a description similar to the discommensurate picture developed
by McMILLAN (Z9) for the ZH polytypes.

ZH POLYTYPES

PLD-CDW formation in the ZH polytypes is weaker than in IT polytypes, and

the characteristic commensurate superlattice is 3a x 3b. The two best
studied compounds are ZH TaSeZ and ZH NbSez. The-neutron diffraction inves-
tigation by MONCTON et al (Z4) shows that both compounds undergo a distort-
ion to an incommensurate structure with ~ = 1/3~* (1-0) ando ~Z%, TaSeZ
at IZZK and NbSeZ at 33K. Below these temperatures,o decreases, and at
90K, TaSeZ locks into the commensurate phase. NbSez however remains
incommensurate down to low temperatures. In the incommensurate regime
of TaSeZ MONCTON et al noticed in addition to satellites at 1/3a*(I-o)
weaker satellites at 1/3~*(I+Zo), as is shown in Fig II. -

2M-Ta So2

M
0
0"

"" I 2 ~ 4 5
[too) -

>-
....
iii
z T '901<
w
....
! T:IOOK

& 28
l~ ~ 4000 ~ 4050
[too]-

Fig. 11 Elastic neutron scattering measurements of the satellite peak

intensity and position in ZH TaSeZ versus temperature (Z4)
The presence of harmonics in the PLD-CDW is consistent with the discommensur-
ation picture (29), although the treatment by MONCTON et al of the lock-in
energy gain is somewhat different.

Nb 93 is suitable for NMR (I = 9/Z), and the formation of the PLD-CDW in

ZH NbSez produces changes in both the Knight shift and the electric field
gradient (57). Fig. IZ shows the (!,-!) central line at various tempera-
tures. The broadening of the line below the transition at 33K is com-
pared to the calculated lineshape for a triple incommensurate CDW (shown
as the solid line). BERTHIER et al (57) consider the general agreement
with the experimental line to be good, but the peak in the experimental
line at the high field edge is not consistent with the step that is
calculated, and may be an indication for a tendency towards local

212
Fig. 12 NMR in NbSez showing evolution of the central line (!,-!)
below 33K. The solid line is the calculated lineshape corresponding
to a Knight shift distribution associated with a triple incommensurate
CDW (57)

commensurability in this incommensurate phase. BERTHIER et al (57)

estimate the change in charge density below the transition to be 10%,
a much smaller change than in the IT polytypes.

I would like to thank D. Jerome and A.D. Yoffe for many helpful discussions.

213
REFERENCES

I. R.E. Peierls. 'Quantum Theory of Solids', O.U.P. (1955)

2. H. FrHhlich, Proc. Roy. Soc. A223, 296 (1955)
3. 'Low-Dimensional Cooperative Phenomena', Ed. H.J.Keller, Plenum (1975)
4. 'Chemistry & Physics of One Dimensional Metals', Ed. H.J.Keller,
Plenum (1977)
5. 'Electron-Phonon Interactions and Phase Transitions', Ed. T. Riste,
Plenum (1977)
6. A.D. Yaffe, Chern. Soc. Rev. 5, 51 (1976)
7. A.J. Berlinsky,Contemp. Phys.-17, 331 (1976)
8. G.A. Toombs, Phys. Reports C 401, 181 (1978)
9. M.J. Rice and S. StrMssler, Solid State Commun. ll, 125 (1973)
10. J. Friedel in (5)
II. S-K. Chan and V. Heine, J. Phys. F 3, 795 (1973)
12. B. Horovitz, H. Gutfreund and M. Weger, Phys. Rev. B 12, 3174 (1975)
13. B. Mogensen, R.H. Friend, D. Jerome, K Bechgaard and ~Carneiro,
Solid State Commun. (1978)
14. N.J. Doran, B. Ricco, M. Schreiber, D. Titterington and G. Wexler,
J. Phys C II, 699 (1978)
15. H.W. Myron~J.Rath and A.J. Freeman, Phys. Rev. B 15, 885 (1977)
16. G. Wexler and A.M. Woolley. J.Phys. C 9. 1185 (1976)
17. H.W. Myron and A.J. Freeman, Phys. Rev. B II, 2735 (1975)
18. N.J. Doran, G. Wexler, V. HeIne and B. Ric~, Nuovo Cimento B 38, 544
( 1977)
19. J.A. Wilson, F.J. Di Salvo and S. Mahajan, Adv. Phys. 24, 117 (1975)
20. P.A. Lee, T.M. Rice and P.W. Anderson, Phys. Rev. Lett:-31, 462 (1973)
21. M.J. Rice and S. Strgssler, Solid State Commu~ 13, 1389 (1973)
22. K. Saub, S.Barisi( and J. Friedel, Phys. Lett. 56A, 302 (1976~
A. Bjelis and S. Barisi~, Lecture Notes in Physics 65, 291 (1977)
23. W.L. McMillan, Phys. Rev. B 16, 643 (1977) -
24. D.E. Moncton, J.D. Axe and F:J. Di Salvo, Phys. Rev. Lett. 34 734,
(1975), Phys. Rev. B 16801 (1977)
25. N.P. Ong and P. Monceau, Phys. Rev. B 16, 3443 (1977)
26. K. Tsutsumi, T. Takagaki, M. Yamamoto,-Y. Shiozaki, M. Ido, T. Sambongi,
K. Yamaya and Y. Abe, Phys. Rev. Lett. 39, 1675 (1977)
27. R.H. Friend, M. Miljak and D. Jerome, Phys. Rev. Lett. 40, 1048 (1978)
28. W.L. McMillan, Phys. Rev. B 12, 1187 (1975)
29. W.L. McMillan, Phys Rev. B I~ 1496, (1976)
30. H.R. Zeller in (3) -
31. B. Renker and R. Comes in (3)
32. M. Thielemans, R. Deltour, D. Jerome and J.R. Cooper, Solia State
Commun. 19, 21 (1976)
33. A.J. Heeger and A.F. Garito in (3)
34. D. Jerome and M. Weger in (4)
35. G.A. Thomas et aI, Phys, Rev. B 13, 5105 (1976)
36. M.J. Cohen, L.B. Coleman, A.F. Garito and A.J. Heeger, Phys. Rev.
B ll, 5111 (1976)
37. D.E.Schafer, F. Wudl, G.A. Thomas, J.P. Ferraris and D.O. Cowan,
Solid State Commun. 14, 347 (1974)
38. R. Comes in (4) -
39. Y. Tomkiewicz, A.R. Taranko and J .B. Torrance, Phys. Rev.Lett. 36
751, (1976), Phys. Rev. B 15 , 1017 (1977)

214
40. E.F. Rybeczewski, L.S. Smith, A.F. Garito, A.J. Heeger and
B.G. Silbernagel, Phys. Rev. B 14, 2746 (1976).
41. P. Bak and V.J. Emery, Phys. Re~ Lett, 36, 978 (1976)
42. F. Herman, D.R. Salahub and R.P. Messmer:-Phys. Rev. B ~, 2453 (1977)
43. J. Hubbard, Phys. Rev. B 17, 494 (1978).
44. J.B.Torrance in (4). --
45. V.J. Emery, Phys. Rev. Lett. 37, 107 (1976), and in (4).
46. C. Weyl, E.M. Engler,K. Bechgaard, G. Jehanno and S. Etemad,
Solid State Commun. 19, 925 (1976).
47. J.P. Pouget, S. Megtert, and R. Gomes, Bull. Am. Phys. Soc. ~, 380
(1978) .
48 R.F. Frindt, R.B. Murray, G.D. Pitt and A.D. Yoffe, J. Phys. C 5 LI54
(1972) .
49 J.A. Wilson and A.D. Yoffe, Adv. Phys. ~, 193 (1969).
50. F.J. Di Salvo, D.E. Moncton and J.V. Waszczak, Phys. Rev. B~, 4321
(1976).
51. R.M. White and G. Lucovsky, Nuovo Cimento B 38, 280 (1977)
52. F.J. Di Salvo and J.E. Graebner, Solid State-Commun. 23, 825 (1977).
53. P.M. Williams, G.S. Parry and C.B. Scruby, Phil. Mag.-Z9, 695 (1974)
ibid. 31, 255 (1975). --
54. R. Bro~er and F. Jellinek, 5th International Conference on Solid
Compounds of Transition Elements, Uppsala, Sweden (1976).
55. G.K.Wertheim, F.J. Di Salvo and S. Chiang, Phys. Lett. 54 A, 304 (1975)
56. H.P. Hughes and R.A. Pollak, Phil. Mag. 34, 1025 (1976):-
57. C. Berthier, D. Jerome and P. Molinie, J:-Phys. C ll, 797 (1978)

215
Solitons in Incommensurate Systems

Per Bak
Nordita, Blegdamsvej 17, DK-2100 Copenhagen, Denmark

1. Introduction

Many interesting physical systems underqo phase transformations

to periodically ordered phases vlhich are incornrnensura te with
the underlying lattices, i.e. the wave vector describing the
modulation cannot be formed by simple rational fractions of the
reciprocal lattice vectors. The ordeced structure may be a sta-
tic charge density wave (COW), a modulated lattice distortion,
a spin densitv wave (SOW) or helical magnetic structure, or
even a separate atomic lattice or "nass density wave" (MO"n.

The periodic potential of the lattice may cause complicated

non-linear distortions of the condensed wave. If the coupling
is strong enough this potential may even drive a separate phase
transition, the commensurate-incommensurate transition, where
the period of the modulated phase becomes commensurate with the
lattice. The transition takes place as a consequence of the
competition between local "Umklapp" terms in the Hamiltonian
which favour the commensurate phase, and the remaining "ela-
stic" terms which must favour a periodicity slightly displaced
from the commensurate one. This transition is the subject of
the present article.

It is rather easy to show that a simple Landau theory, in

which the order parameter is assuned to be homogeneous through-
out the system, leads to a first order transition. In a very
elegant paper [lJ, however, r·ic!1illan demonstrated that by
allowing the phase of the order parameter to fluctuate while
keeping the amplitude constant one finds in fact a second order
transition. Accordinq to this theory, the incommensurate phase
near the phase transition consists of very large regions which
are essentially commensurate ".lith the lattice, separated by re-
latively narrow domain vTalls or "discornrnensurations" where the
phase of the order parameter changes rapidly. Bak and Emery
[~ subsequently demonstrated that the ddmain walls in fact
appear as the solutions of the Id sine-Gordon equation; hence
the name "solitons". The order parameter of the transition is
the soliton density, which in turn is proportional to the dis-
placement of the wave vector from the commensurate one.

Here the theory of the cOl!1Jnensurate-incomrnensurate transi-

tion will be reviewed and extended to describe a variety of

216
widely different physical systeMs. McMillan~ theory in its
original form applies to a one-diMensional modulated struct~re,
where the modulation is described by a single wave vector, q.
On the other hand, most systems which have been observed to ex-
hibit commensurate-incommensurate (I-C) transitions are in fact
two-dimensionally modulated, i.e. the Modulation is described
by two or three symmetric q-vectors in a common plane. It will
be demonstrated that a second order transition is still possjble
with a linear temperature dependence of the order parameter
(S=l) in contrast to the logarithDic behaviour (S=O) in the Id
case. This difference is caused by the topologically different
nature of the incommensurate phases in the two cases. This
theory applies to the charge ~ensitv ~ave (CO~) in the layered
~ ~

metal chalqoqenides [3J, the I-C transition in the "triple g"

modulated magnetic structure of neodymium [4J, and also to the
registry-non-registry transitions in a monolayer of rare gas or
deuterium atoms adsorbed on graphoil or graphite [5, 6J. vIe,
shall also consider the coupling to alternative degrees of free-
dom. It will be shown that a coupling to a macroscopic strain
does not change the nature of the transition, and the induced
strain ~ill be of a inhomogeneous soliton-like nature.

We start with a brief review and discussion of the nature of

the modulated phases in various physical systems.

2. Modulated Structures. Id and 2d Modulations

The most well known modulated structures are probably the sinu-
soidal or helical maqnetic structures observed in the rare earth
metals [71. Near TO; where one wave vector is predominant, the
magnetic structures may be vlritten on the form

~qexp ( 19or + complex conjugate

' ~ ~)
(2.1)

where ~~ is the so-called order parameter. By the term "incom-

mensura~e structure" we shall here understand a structure on
the more general form

(2.2)

where Mx (q 0/1 is periodic with period 2 n. The non-commensurate

wave vector g is defined by this equation. This structure in-
cludes all higher harmonics. A charae density wave, or a modu-
lated lattice distortion may be written on a similar form. Un-
til now, no I-C transitions~have been reported in such systems.

There exist, however, a number ~f Modulated systems charac-

terised by two or three symmetric q-vectors. For exaMple, the
COW in 2H-TaSe2 [3] is given by
3
p(r) = 2:
i=l
~.exp(iC.
1 ~l
or) + tI!... exp(-iq. or)
1 ~l
(2.3)

217
with 1~~);lw21=lw31 and ql = q(l,O,~), q2 = q(-~, /rlL,0), q3=
q(-~,-. ,0). This is the "triple q" structure. At TO=122 K,
2H-TaSe2 undergoes an I-C transition to a co~ensurate phase
where q = 2n/3a. In the one-dimensional conductor TTF-TCNQ
there are three ordered phases, two of which are characterised
by single incommensurate wave vectors [2,8J. The third phase
includes two co~~ensurate wave vectors [9J.

Quite recently it was discov~red that the magnetic structure

of neodymium is also a "triple q" structure [4J. Near TO, the
magnetic moments in one of the hexagonal layers are given by

~1(r) = ",,,.qA.exp(ia
4,.~ ~ . '~.• r) + cc
l.
(2.4)

This structure is shown in Fig. 1. The magnetic structure is

.~~'t~A~"r~"'~~'.'~'
t""'/~~"""/~-.""
" " t ///~ \ " " .~~/',
I"~"~ .~~/, ""
'/// t '"
\'~~f,~~,~,~~,,~~t'~~
",...
t ..... "1 \ , ..... , .. "" \ ......... "
"11' \\ ••. ',I! ~\\ .. ·.1
··"l'H··"\~'II··,q\ ...... - •• - . . . . . . . c . . . . . . . . -.-....
··'\l/', .. ··'\ll"···'\
,~~~~,~~"'~~~.~~'\_A~'
,/-~.~""/-~ ~"'/~- ...
' / / / ' , ' " \ ~/// t " " , / / /

_ ...._---
,/'//' " ' , . -"~/, " " f __~/'
.4~~I"~~·.~~~t'~~··~~~
\,~'f'~',\'~~"~~/I'~~
1\\""'1'\\""·'/1\\"·
,'11/ ..'.....\'\' \'1 1,'..''', '.',,'\'\'I' t' .....
I'~"'·~\'~~~"~~"~~~
.~~"'~~~ .. ~"'~~~ .. ~,'

Fig. 1 "'l'riple q" magnetic structure of neodymium (left) and

the accompanyinn lattice distortions (right)

accompanied by a similarly modulated lattice distortion. This

distortion is very similar to the one found in the layered me-
tal chalgogenides. The modulation period is temperature depen-
dent, and there are strona indications that the structure locks
into a periodicity of 8 lattice units at TO = 8 K [l~. The
modulated structure forms a hexagonal pattern on top of the ba-
sic lattice.

The last group of systems that will be considered seems at

first glance to be of fundamentallv dif£erent nature. A mono-
layer of rare gases, or deuterium,-adsorbed on graphoil [5,Q
may undergo transitions fro~ a liquid state to a two-dimensio-
nal crystal. The two-dimensional graphoil surface here plays
the role of the basic lattice. This solidification may be de-

218
scribed as the formation of a "mass density wave"
3
~ .->->
~0.exp(-lq. ·r) + cc. (2.5)
i=ll 1

When the pressure is changed, some of these systems undergg

phase transitions to the commensurate "/3 structure" with qOl
21T /3 .
ii(O,;r) or the two s~~etrlc vectors. The symmetry, therefore
is essentially the same as in the preceeding systems.
->
In general, the incommensurate state of the triple q system
also includes higher harmonics. He shall assume that these
harmonics will not lOv!er the sYMmetry further by destroying the
hexagonal pattern. This requirement can be formulated in a
very elegant way usinq symmetry groups of dimensionality 3+n,
where n is the number of vlave vectors involved [llJ.

3. The Commensurate - Incommensurate Transition. McMillan's

Theory for Id-modulated Systems

In this section we shall briefly review McMillan's theory for

the I-C transition in a system where the ordering is character-
ised by a single wave vector. In general, the order parameter
has several components corresponding to the equivalent, symme-
tric, wave vectors. For the system that we consider there are
··
SlX 1ent vectors ±ql,
equlva -> ->
±q2' ->
and ±q3. Th e or d er parameter
therefore has six components ~±l' ~±2 and ~±3. In the Landau
theory the free energy is expanded in terms of these component~
According to Md1illan flj the appropriate expansion for the
commensurate order parameter in 2H-TaSe2 is

(3.1)

3
3 3
+ d 2 ~ (~i+~-i)
i=l

3 2
+ e' L
i=l
(V-iti)~i

219
~ere 8 i is a small vector along the commensurate wave vector
gOi. The term with coefficient e' favours an order parameter
Wi = Aexp(i8 ior), Le. p(r) = Aexp :i(QOi+8i)·r , according to
(2.3). This elastic term thus tends to induce an incommensur~e
CDW with wave vector qi = qOi+8i. On the other hand, the Um-
klapp term with coefficient d2 favours the commensurate wave
vector given by Wi = -A. The phase transition takes place when
18il becomes small enough for the Umklapp term to compensate
the elastic term. For the other system that we have in mind
the Umklapp terms are of a slightly different form leading to
different rational wave vectors. It is a trivial matter to
generalize the theory to these cases.

Depending upon the values of the parameters in (3.1) eithe~

a single g wave, with 1Wli f 0, !w21 = Iw31 = 0, or a triple q
wave, with IWll ~ Iw2! ~ w31 is stable. The basic assumption
entering McMillan's theory is that near the I-C transition the
amplitude, A, of the order parameter is essentiallv constant,
but the phase, <jl, is allm.red to accomodate to the ~nderlying
lattice by fluctuating in space. Inserting

WI = Aexpi<jl(x)'W 2 = W3 = 0

we get the following energy density (apart from a constant term

and a constant factor):

F(x) k(d<jl(x) _ 1))2 _ v(cos{3"'(x) }-l) _ ~1)2 (3.2)

2 dx '"

where x is measured along the ordering wave vector and 0=1811.

The actual phase function is the one which minimizes the inte-
gral of F(x). The function <jl(x) =0 yields the commensurate
phase with energy normalized to zero. The incommensurate phase
is thus formed by space-dependent local displacements of the
commensurate phase. The functions <jl(x) for which this free
energy has an extremum satisfy the sine-Gordon equation,

<jl" (x) - 3vsin{3<jl(x)} = 0 (3.3)

2'Tf
with periodic boundarv conditions <jl(x+L) = <jl(x) +~. The
average wave vector, q,
which may be measured in a diffraction
experiment is

21T
q
3L (3.4)

according to the definition (2.2). The solutions of this equa-

tion are the "solitons" ~J or domain walls separating almost
commensurate regions with <jl= 231Tn and w= ~(n+l), n integer.
. 3

220
Inserting the solutions into (3.2) we find the free energy den-
sity

F(x) = 4~ q [1 + 4exp(- 2;IV)] - q6. (3.5)

Minimizing (3.5) vlith respect to q one finds q versus 6. The

commensurate-incommensurate transition takes place when 4/V/~
=6. The terms in (3.5) may be interpreted in a simple way.
Noting that q is proportional to the soliton density we may
interpret the coefficient of q as the energy required to create
a soliton. The remaining term depends exponentially on the
distance between solitons and is also proportional to the num-
ber of solitons, and may thus be interpreted as a repulsive
interaction between solitons.

The equilibrium soliton density results from a balance be-

tween the energy gain from forming solitons and the energy loss
from the repulsive interaction. In section 5 we shall see that
this picture is substantially different for the 2d-modulated
case because of a different topological structure of the soli-
tons! At the transition the density of solitons goes smoothly
to zero, and the transition is thus of second order. The den-
sity of solitons may serve as the order parameter for the tran-
sition. The soliton density depends loqarithmically on 6 and
hence also upon the temperature

- -1 T-T O
q '\, -In

- -1 p-P O
or q '\, -In

if the pressure is the field variable.

4. Coupling to Strain or "how to count solitons"

Before proceeding to consider the triple q case, let us cOffiider

the coupling to alternative degrees of freedom. To be specific,
we consider the coupling to an elastic strain. It turns out
that the transition remains second order; in fact one may argue
that the soliton picture and the second order transition are
stable against any such local coupling.

The theory presented in this section closely follows the cal-

culation of .Timonen and the author r12]. Recently Bruce, CoWLey
and Murray [13] considered the coupting to a uniform strain n.
They argued that such a coupling will necessarily drive the
transition first order. Here it will be demonstrated that the
induced strain will in fact be of a very inhomogeneous, solfton-
like nature, and that the transition remains second order.
Cowley, Bruce and Murray showed that the symmetry of the system

221
allows a linear coupling between the macroscopic strain nand
the phase gradient. This coupling gives rise to additional
terms in the free energy density

F (x)
n
= wn~
dx
+ ~cn2 (4.1)

Proceeding as before they found that the free energy density

now includes a negative quadratic term - w2 - 2 which clearly im-
plies a first order transition. 2c q-

In the calculation presented in [12J ' n is allo,,'ed to be po-

sition dependent. To simplify the calculation without changinq
the physical picture the cosine term in (3.2) is replaced by
the potential (Fig. 2)

7T 37T
15 P

Fig. 2 Potential V(¢) used in the calculation. [l~

9 2
V(¢) "2v¢ (x) -

V(¢) V(¢ + 271) (4.2)

3

This is a periodic parabolic potential. In general the actual

potential for a real physical system is unknown. The present
calculation may thus also serve the purpose of illustrating
that the soliton picture does not depend on the particular
choice of potential. The complete free energy is

222
 f
L

F t F (x)dx,

L
-"2

(4.3)

where a term quadratic in the strain gradient has been included

for completeness. The Euler equations determining the extr~ma
of this functional are

¢" (x) - r¢ (x) + \\'ll' (x) = 0 (4 • 4)

dll"{X) - Cll{X) - w¢' (x) = 0

These equations have the complete solutions

L< <L
¢(x) = Asinh{Kx), -"2 -x-"2

~(x+L) = ¢(x) + 2; (4.5)

L< <L
II (x) Bcosh{Kx), - "2 - x-"2

II (x+L) II (x)

Here

A 1T/3sinh{KL/2)

2
B AWK/{dK -c)

If the coupling is unphysically strong, {!rd+Vl)2>c, the phase

soliton picture breaks dovm and this type of solution ceases to
exist. The parameter K qives the width of the solitons. The
free energy versus q is now

223
 F = ''TIu
6K -q [ l+2exp (- ~)J
3q --
qo (4.6)

where

u = r + dK 2 (~) 2.

This free energy is of exactly the same form as (3.5) and in-
cludes no quadratic term in q. The transition therefore re-
mains second order. The main effect of the coupling to the
strain is to Modify the soliton width 11K, and to change the
transition temperature given by 0= 'TIu.
6K
The phase soli tons and the induced strain are sho ....'n in Fi'l.
3. The strain is hiqhly inho~ooeneous, beinq localized near

47T
P
37T
P
27T
P
7T
P

0.5 1.0 1.5 2.0 2.5 VL

1}

1.0

Fiq. 3 Phase solitons and induced strain calculated for K=lO

(After l12] )

the phase solitons. The cusps originate from our special crnice
of potential which also has CUSDS. The macroscopic, measurable
strain, n,
is given by the inte;ral

224
 L

-
n i SdXL
n (x)
-q B
KA
(4.7)

-"2

This is an interestinq relation. The strain is proportional

to the soliton density~! To count the number of solitons in
a crystal one may simply measure its distortion near TO' i.e.

- - -1
n 'V q 'V -In I T-T O! . (4.8)

One may argue that any physical quantity coupling to the phase
gradient or higher derivatives gives a contribution proportio-
nal to the number of solitons, or q, since these derivatives
are small except in the very narrow soliton regions. It is
therefore expected that the coupling to such degrees of free-
dom will not change the soliton picture and drive the transi-
tion first order.

Our calculation points out a simple analogy: just as the

assumption of a uniform phase ¢ leads to a first order transi-
tion which becomes second order of the phase is allowed to
fluctuate in space, the assumptions of a constant strain leads
to a first order transition which becomes second order if one
allows it to be position dependent.

5. Phase Solitons in 2d-modulated Systems

For the physical systems which are known to exhibit I-C transi-
tions, the modulated structure cannot be described by one wave
vector but is formed as a superposition of waves with two or
three symmetric wave vectors. To be specific we shall study a
system characterized by three symmetric vectors gl' ~2' and g3·
This is the situation for the layered metal chalgogenides, for
the magnetic structure of Nd, and for the rare gases adsorbed
on substrates.

The starting point will be again the free energy expansion

(3.1), but now we assume that the parameters favour the "triple
g" structures. Inspired by the ld-modulated case we assume the
amplitudes of the three waves to be fixed, but allow the phases
to fluctuate,

(5.1)

225
to obtain the followinq free energy functional

F Sd3r[ucOS(~1+~2+~3) + v(cos3~1 + cos3~2 + cos3~3)

+ e~(~<pi-ti)2 + ft (qiXV~i)21. (5.2)
~:-l i=l

Again we have ignored constant terms and factors. The first

cosine term favours the phases ~1=~2=~3 = 0, and the equivalent
combinations both in the commensuraEe and in the incommensurate
phases. Just as we expect the amplitude fluctations to be in-
essential and therefore keep the amplitudes constant, we now
make the additional ansatz that the sum of the phases remains
zero. It turns out then to be convenient to define new phase
variables

~2 ~3
ux ~l - 2"- 2"

uy 11 (5.3)
2"(~2-~3)

~l + ~2 + ~3 O.

The variables u x and u y describe local displacements of the

commensurate ~odulated structure in the perpendicular x- and
y-directions in the plane spanned by ql, Q2' and q3. This de-
scription is quite analogous to the ld-situation, where the in-
commensurate phase was described by local displacements of the
commensurate structure alonq q. In fact, the continuous symme-
try which is broken at the I-C transition is the translation of
the CDW with respect to the lattice. In the ld case the trans-
lational symmetry is broken in one direction, in the 2d-triple
q case it is broken in b'o perpendicular directions. Our free
energy functional thus seems to include the appropriate degrees
of freedom.

The free energy functional becomes

F
1
J
1i c l (
dux du
~)2
dx + dy

(5.4 )

+ V[3 - cos(2u x ) - cos(l3u -u ) - cos(l3u +u

y x y xj~
226
where cl=c/3, c 2=c/6 + f/6, c 3=f/3. This free energy has a
very inEerestinq interpretation which is most clearly under-
stood in terms of the two-dimensional monolayers on graphoil.
The two first terms are the elastic energies associated with
the two irreducible strains defined for a 2d-crystal. For the
CDW these terms give the ener~ies associated with the analogous
elastic strains of the wave. The third term with coefficient
c3 is the elastic energy of an infinitesimal rotation of the
2d crystal. This term, of course, is zero for a free crystal.
Here it represents an interaction with the lattice. If this
term is negative, the equilibrium rare qas lattice or cm~ \oTill
be rotated relative to the underlying lattice, and \Ole \O!ould get
the "orientational epitaxi" discussed by Novaco and McTague
[15[.
In fact, the functional (5.2) is one that one would intui~
tively apl1y directly to the rare gas lattice on the hexagonal
graphoil surface. Our transformation has thus illustrated in a
simple and efficient \olay the equivalence between this system
and the "normal" triple q structures in 2H-TaSe 2 and Nd.

The actual distortions ux(x,y) and u y (x, y) are the ones whjch
minimize the free energy functional. The co~~ensurate phase
is given by ux(x,y) ~ 0, uy(x,y) ~ O.

In the ld-case it ~las assumed that the solution is neriodic,

i.e. corresponds to a single q vector. In a similar m~nner it
will be assUmed here that the-functions ux(x,y) and uy(x,y)
have hexagonal symmetry, i.e. the condensed wave will always
have sixfold symmetry, consistent with our general understand-
ing of the incommensurate phase described in section 2. It can
be shown that this symmetry will be conserved to any order in
perturbation theory. [15J. The diffraction pattern consists of
a hexagonal array of satellites situated around each reciprocal
lattice point. For mathematical simplicity the cosine potenti-
al is again replaced by a truncated parabolic potential:

o~ ux ~u
y
11"3 (5.5)

The potential defined in this way within the irreducible part

of the hexagonal cell is continued periodically to cover the
whole (ux,u y ) space. (See F iq. 4).

In summary, the free energy density that we shall consider

is

227
 v ux=cosG uy=sinG
v

Fig. 4 Potential V(u ,u ) used in the calculation of two-dimen-

sional phase gol~tons

F 8/3
L2
rVc 1 (~
j
du
dx
+
d
~)2
dy
+ c
2
[ d x
dx
d
(u _ ~)2 +
dy
du
(--Y
dx
+
dux
dy)
2J

du du 2 2
+ c3( d~ -~) + AU y 2 + Bu x (5.6)

du du y
- 0( d~ + dy)

where the integration area is described in Fia. 5.

y
L
, -_ _2i======t\( '/2'13 L, Ll2)

L/.fjx Fiq. 5 Inteqration area

used in (5.6)

228
 The Euler equations are

d 2u X d 2u d 2u
x
(c 1 +c 2) - - 2 + (c 2 +c 3 ) dy2 + (C l -c 3 )CiXdY -v
Bu x a (5.7)
dx
d 2u
d 2u d 2u x
x
(c l +c 2 )--2 + (c 2 +c 3 )---1 + (c l -c 3 ) dxdy
- Au
Y
a
dy dx

with the boundary conditions dictated by the haxaqonal symme-

try:

Uy(X,~) = 1T/13
u y (y/l3,y) = l3 ux (y/l3,y)

(5.8)

These equations can probably not be solved analytically in ge-

neral, but for B=3A the complete solution can be found. There
seems to be little reason to believe that this particular choi-
ce will affect the physical picture.

The solution is

1T sinh(i3Kx)
u
x 3" sinh(KL/2)

1T sinh (KY)
uy
73 sinh (KL/2) (5.9)

with K2(c +c 2 ) = A. For larqe L this solution represents a

network o~ solitons (Fig. 6). The incommensurate phase con-
sists of large areas (or volumes) of essentially commensurate
phase separated by a hexagonal lattice of solitons where the
phases u and u change rapidly. This is consistent with the
numerical result obtained by Nakanishi and Ehiba r15]. For the
2d-monolayers these solitons are in fact the "mis~it ~isJoca­
tions" introduced by Venables and Schabes-Retchkiman L16J.
By feeding the solution back into (5.6) we find

41TO
F /3"L

(5.10)

229
Fig. 6 Hexagonal soliton network near the I-C transition. The
numbers in the brackets are the commensurate phases
approached in the almost co~~ensurate regions

21T
The average wave vector, q 73L is inserted to yield

2A c l -c 2 - 2
F 2cS)q + -(--)q (5.11)
K2 c l +c 2 -

+ 4A1T ~exp(- 21TK)

73K - I3q .
By minimlzlng this function \'.1i th respect to q one finds q vs cS
(or T, or P). The transition is first order for neqative coe-
fficient of q2, second order for positive coefficient. For the
specific model studied by Nakanishi and Shiba the transition
seems to be first order, but the present calculation lends
support to the conclusion that it may indeed be second order.

The various terms in (5.11) ~ay be interpreted in a simple

way. The first terms is the usual soliton formation energy
which becomes zero at the transition temperature given by
A1T=I3KcS. The third term is the positive repulsion between so-
litons. The second term is proportional to the square of the
number of solitons, which in turn is proportional to the number
of intersections in Fig. 6. This term thus gives the soliton
intersection energy, which of course does not exist in the Id-

230
case. One finds

(5.12)

in contrast to

in the ld-case (Fig. 7).

I~

Fig. 7 Temperature depen-

dence of average wave vector
q near the I-C transition
for single q and for triple
q structures, respectively

This fundamental difference has a very simple topological

explanation. The structure is stabilized by the competition
between soliton formation energy and soliton intersection ener-
gy in the 2d-case, and not by the com~etition between the soli-
ton energy and the repulsion energy as in the ld-case.

The prediction (5.12) should be tested against experiment.

The transition in 2H-TaSe2 is first order [3J, which of course
is possible within the theory. More inter€sting in this re-
spect is the LEED experiment performed by Chinn and Fain (Fig.
8) on krypton on qra~hite. The experiment does indeed show a
second order I-C transition when the oressure is varied, and
the lattice constant near TO seems to-be consistent with the
linear form (5.12). The neutron scattering experiment by
McTague and "Nielsen on graphoil was performed at constant cove-
rage and not at constant pressure. The continuous result (Fig.
9) is however consistent with a second order transition. Other-
wise a phase separation would have taken place near TO. Pre-
liminary neutron scattering experiments on Nd indicate a lock-
in at a period of eight lattice units at TO=8 K [lOJ. It would
be interesting to study the temperature dependence of the
wave-vector in this material and compare with (5.12).

231
 I. 12

1.08
Ch;". F;:;;: 2.2 C

/
(~o tl. 04
2.1
~
:=- 2.0

;W~59K
~
-;:; 1.9

1.6
1.00
1.7
10- 7 10- 4 1.0 112 1.2
IQI
(Torr)

Fig. 8 Lattice parameter mea- Fig. 9 Neutron scattering

surements for krypton on ara- measurements of lattice para-
phite. [5J meter of D2 on graphoil." [~

There is one interesting complication. A positive intersec-

tion ener~y is required for a second order transition to take
place. One may therefore ask whether an arrangement of parallel
solitons might be energetically favourable at not too large o.
To test this, the hexagonal boundary conditions (5.8) are re-
placed by the conditions

(5.13)

These conditions break the hexagonal symmetry. The free energy

becomes

F = ~ r (~; - 20)q + ~:qexp(- 5;~)1 (5.14)

l ~ J
For 0 small enough, this free energy ~~y be lower than that ob-
tained by minimizing (5.11). The equilibrium soliton density
is more than twice that given by (5.11) and goes logarithmical-
ly near T g.
At larger 0, the repulsive interaction will com-
pensate t e intersection energy and a first order transition to
a symmetric state will take place. The picture that emerges
thus includes tvlO transitions and an intermediate 101-7 symmetry
phase. This sug0ests that the second order transition may take
place only through metastable states, and one should investi-
gate this· problem for example by looking for hysteresis effects.

232
Acknowledgements

I would like to thank J. Timonen and B. Lebech for a very plea-

sant collaboration on much of the research reported here. Also
thanks are due to J.D. Axe, J. van Bohm, R.A. COvlley, V.J. Emery
J.P. McTague, D. M.ukamel, M.Nielsen, A.D. Novaco, V. Prokrovsky,
and J.A. Venables for many interesting discussions. I am grate-
ful to s.c. Fain, Jr. and M. Nielsen for their permission to re-
produce Figs. 8 and 9, and to E. Riedel and G. Grinstein for a
careful reading of the manuscript.
References

1. W.L. McMillan, Phys. Rev. B14, 1496 (1976).

2. P. Bak and V.J. Emery, PhyS:-Rev. Lett. 36, 978 (1976).
3. D.E. Honcton, J.D. Axe, and P.J. DiSalvo~Phys. Rev. Lett.
34, 734 (1975). Phys. Rev. B16, 801 (1977); J.A. Wilson,
~J. DiSalvo, and s. Mahajan~dvances in Phys. ~, 117
(1975) .
4. P. Bak and B. Lebech, Phys. Rev. Lett. 40, 800 (1978).
5. S.c. Fain, Jr. and M.D. Chinn, Journal de Physique, 38,
C4-99 (1978), Phys. Rev. Lett. 39, 146 (1977). --
6. M. Nielsen, J.P. McTague, and vl-:-Ellenson, Journal de
Physique, 38, C4-10 (1978).
7. See, for example, V7.C. Koehler in "Magnetic properties at
rare earth metals", R.J. Elliott, ed., Plenum, New York,
1972 p. 8l.
R. R. Comes, S.M. Shapiro, G. Shirane, A.F. Garito, and A.J.
Heeger, Phys. Rev. Lett. 35, 1518 (1975), W.D. Ellenson,
R. Comes, S.M. Shapiro, G-:-Shirane, A.P. Garito, and A.J.
Heeger, Solid. State Commun. 30, 53 (1976).
9. P. Bak, Phys. Rev. Lett. 37, 1071 (1976); A. Bielis and
S. Barisic, Phys. Rev. Lett. 37, 1515 (1976). -
10. B. Lebech, Private communications.
11. A. Janner and T. Janssen, Phvs. Rev. B15, 643 (1977);
P. Bak and T. Janssen, Phys.-Rev. B17:-436 (1978).
12. P. Bak and J. Timonen, J. Phys. C, to be published.
13. A.D. Bruce, R.A. Cowley, and A.F. Murray; A.D. Bruce and
R.A. Cowley, J. Phys. C, to be published.
14. A.D. Novaco and J.P. McTague, J. Physique 10, C4-116 Q977).
15. K. Nakanishi and H. Shiba, J. Phys. Soc. Japan, 44, 1465
(1978) . - - -
16. J.A. Venables and P.S. Schabes-Retchkiman, Surface Science
71, 27 (1978); Journal de Physique ~, C4-105 (1978).

233
Fluctuations and Freezing in a One-Dimensional Liquid: Hg3-~AsF6

J.D. Axe
Department ~f Physics, Brookhaven National Laboratory
Upton, NY 11973, USA

1. Introduction
In this talk we examine some of the properties which may arise in solids
composed of two (or more) interpenetrating sUblattices with spacings which
are incommensurate one with another. The reason to suspect something out
of the ordinary is shown by the following simple considerations. Imagine
the two sublattices to be perfectly periodic and write their interaction
ene~~y as a product of the charge density, 0A(r) ,of one times the potential,
wB(r),of the other,
VAB = fOA(r)wB(r)dr

= I ° (G)w (G')
GG' A B
f eHG-G,).rdr (1)

= ~G' °A(G)wB(-G')oG,G'
This shows that the two sub1attices interact only by virtue of common reci-
procal lattice vectors. Suppose that both sublattices can be thought of
as two-dimensional arrays of chains arranged on a common rectan9ular lattice,
but with different and incommensurate interatomic spacings along the common
chain direction, z. It then follows trivially that the only common recipro-
z
cal lattice vectors have Gz = G = 0 and the resulting forces, while con-
straining the chains in the x,y plane, do not fix the relative positions
of the two sub1attices along z. (In fact, the system can gain additional
interaction energy by a mutual modulation of the natural period of one
chain type with the period of the other, but this is a small effect and
does not change the qualitative conclusion that the forces which act to
localize the atoms or. their chains are, at best, abnormally weak.)
1.1 H9 3_oAsF 6 • Perhaps the best studied example to date of the type of
structure we have in mind is the mercury chain compound H9 3_oAsF 6 . It
6
consists of an ordered body-centered tetragonal (bct) lattice of AsF anions
(the host lattice) through which pass linear chains of po1ymercury cations
arr~nged in two idenlica1 perpendicular nonintersection arrays, one parallel
to aT' the other to b. See Fig. 1. These will be referred to as the x-
and y-arrays, respectTve1y. Room temperature diffraction studies have
shown in addition to the expected Bragg reflections, strong diffuse scat-
tering arranged into series of thin sheets in reciprocal space[1-3]. Fig.2

234
~ Structure of H9 3_ nAsF 6. The octahedral AsF 6 groups carry one nega-
tlve charge. The Hg-atOms on the chains are shown schematically. Above
Tc=120 K the average H9 density is uniform along the chains. After A. J.
SCHUL TZ et a1. (Ref. 3)

is a sketch of the (HKO) scattering plane. It is established that the dif-

fuse sheets arise from the Hg-atoms and the narrow width of the sheets shows
that the intrachain Hg-Hg distance, d, is well defined, and the nearly uni-
form distribution of intensity within a sheet shows that there is little or
no interference between scattering from different chains [3,4]. Thus posi-
tions ofte atoms along the chains are virtually uncorre1ated from one chain
to the next. Finally, fromothe spacing of the diffuse sheets, the inter-
chain Hg distance d = 2.67 A, which is incommensurate with al = 7.53 A.
This results from a non-stoichiometric composition Hg AsF with
3-0 = (al/d) = 2.82. (A puzzling fact is that chemicaf-Rna1Yses consistently
find 0 = O. Whether this is due to "pools" of excess Hg, to random vacancies
on the host lattice, or neither, is at present unresolved.)
Further work by HASTINGS et a1.[4] extended the diffraction study to low
temperatures and concentrated on the behavior of the diffuse scattering in
the m = 1 sheets. Fig. 3 shows that what is essentially a uniform distribu-
tion of intensity within a sheet at room temperature evolves into a pro-
nounced modulation at 180 K. The modulation was interpreted as arising from
short range correlations between the position of Hg atoms on nearby parallel
chains. In the vicinity of Te = 120 Ksharp Bragg peaks grow out of the
sheet of diffuse scattering wlth a temperature dependence typical of a con-
tinuous second order transformation {see inset, Fig. 3} and which must be
associated with interchain ordering. Very peculiar, however, is the fact
that the Bragg peaks do not develop at the positions on the sheets where
the modulated diffuse intensity is strongest, but grow instead from regions

235
 K

r
• 0 ~.
'111/1I111/111111!1I1/1I111(/lIHIII/II/II/III/I~'WI///IM 111)

~
~
~
o • ~ 0
21r
d
I~
~

1
• 0 ~.
~
~
""~
~--~o-----~--~o-~H
~
lS
~

~ A schematic representation of the diffraction pattern of H9 3 &AsF 6

at room temperature. The straight lines represent the intersection Of
sheets of diffuse scattering lying perpendicular to the figure with the HKO
scattering plane

3000 >-
I-
Ui

1:1
Z

"'
I-

>- 10,000 ---

f-
U)
Z
w 5.000
f-
Z
o 50 100
1000 TIK)

-0,5 o 0.5
(h,3-0,0)

~ Temperature dependent evolution of short and long range order as

seen by thE. modulation of the m=l diffuse sheet. Parallel chain interactions
are responsible for the broad peaks at h ~ ± 0.4. The long range order ap-
pears in the sharp Bragg peaks at h = (1-8) % 0.82. The inset shows the
temperature dependent growth of the Bragg scattering below Tc. After J. M.
HASTINGS et al. (Ref. 4)

236
of low intensity, i.e. the Bragg peaks are preceded by little or no "criti-
cal" scattering. The nature of the resulting ordering was deduced by
HASTINGS et a1. by noting that the positions of the Bragg peaks on the
sheets were such that a reciprocal lattice vector from the x-array coincided
with one from the y-array (at a point on the intersection of the two m=l
sheets). This fact, in conjunction with the theorem of the first paragraph,
strongly implicates interactions between perpendicular chains as the domi-
nant factor in the ordering. Unexplained, however, was the apparent sudden
reversal of the relative importance of the parallel chain interactions,
responsible for the short range order, and perpendicular chain interactions
responsible for the long range order.
At high temperatures HASTINGS et a1. also found that emanating from all
points of the diffuse scattering sheets are inelastic scattering surfaces
with linear dispersion depending only upon the component, Q, of momentum
along the chain direction. That is w = ± v IQ-Qml and Q = 2nm/d specifies
the posltion of thE: m'th diffuse sheet. They ascribed tWis scattering to
1-d longitudinal phonons propagating along the independent Hg chains and
found that v = 4.4 X 10 5 cm/sec.
The remain~er of this talk is devoted to a discussion of a simple model
developed and analyzed by EMERY and AXE[5] for H93_oAsF6 (although with
little modification it should be useful in thinking abolit other linear in-
commensurate phases as well.) It incorporates competing parallel and per-
pendicular chain interactions, predicts correctly the long range order and
clarifies the apparent failure of the system to anticipate this ordering in
the fluctuations above Tc. In addition, it treats carefully the effects of
one-dimensional fluctuations, and predicts that the Hg chains at high tem-
peratures behave as a one-dimensional liquid. The subsequent phase trans-
formation can be thought of as a freezing of the 1-d Hg liquid, and can be
discussed in terms of self-consistent solutions of the sine-Gordon
Hamiltonian.
2. The Model Hamiltonian

assumes harmonic interactions, ')( intra = L

The Hamiltonian is the sum of intra- and inter-chain contributions'?{intra
'jt ~i where for example
",1 0 1 N rr 2(R.x,a)
-n. ="2"L + K(x(R.x,a+1) - x(R.x,a)-d)2 (2)
R.x a=l m
where (rr(R.x,a), x(R.x,ci)) are the components of the momentum and position
vectors of the a'th particle on the R.~th chain. (The subscript i = x,y is
to be used to specify the x- or y- array of chai ns.) The effecti ve near-
neighbor stiffness constant K = mv 2/d 2 is chosen to give the measured 1-d
phonon velocity. m is the bare Hg atom mass.
The configuration of the R.i'th chain is specified by the particle density
operators, for example
PR.x(x) = L o(x-X(R.x,a)). (3)
a
In the disordered (high temperature) phase the Hg density is uniformly dis-
tributed along the chains, i.e. the thermodynamic averages

237
<P£x{x» = <P£y{y»= constant. In terms of the Fourier transformed variables,
<P£x{Q» = <P£y{P» = 0 except for P = Q = O. The quantities «P£x{Qm»'
<P£y{P m» for (Pm,Qm) = 2~m/d can be taken as a complete set of order para-
meters specifying the chain ordering transformation. We will see that the
instability is associated with the primary order parameters
{<Pi (Q,», <P£ (PI»)' Note that we have retained the notion of a local
chalfl variable ~y Fourier transforming only the position along the chain
direction. Although it is useful in what follows to introduce wave vector
components perpendicular to the chain directions as well, it is still im-
portant to distinguish between parallel and perpendicular components, as
the latter are conjugate to discrete chain positions and can be restricted
to the first Brillouin zone, whereas the former is associated with a con-
tinuous distribution along the chains and are thus unrestricted.
We introduce local coupling between chains of the form
'){inter = Y-xx + ~y + 1<y where

~t xx = ~ £X,£x'
I J dx JdX'V~x £x,{x-x')P£x{x)P£x{x')
,
(4a)

~~ xy = ~ £x,£y
L J dx JdYV~x , £y{x-X~y,y-y~x)p£x{x)p£y{y) (4b)

These equations can be rewritten in terms of their Fourier transforms, e.g.

P£x{Q) = N- 1/ 2 L e-iQx{£x,a)
a
and for example

(4c)

where N is the number of atoms per chain.

2.1 Range of Interactions· We find that only rather near-neighbor coupling
is necessary to explain the observed behavior of Hg 3_8AsF 6 . The short range
of the interchain coupling is understandable. If we associate a charge
density, CJ£x{x)= e*P£x{x) with the atomic density and calculate the Coulomb
coupling between two parallel chains (£x,£x') separated by a distance R, we
find
00

VII ,(Q) = 2e*tJ cosQ{x-x') dx = 2e*2 K (QR)

x, x d [(x-x')2+R 2Jl/2
_00 d 0
(5a)

~ 2e*2 {__~__ )1/2 e- QR (QR» 1) (5b)

d 2QR

where Ko{z) is a Bessel function. This shows that the coupling between charge

238
modulations on parallel chains is exponentially small if the wave vector of
the modulation is large compared to the inverse interchain spacing, R_l.
The coupling between perpendicular chains shows similar behavior. The im-
portant charge fluctuations are at multiples of Ql = (2TI/d) and for near-
neighbor parallel chains Q]R = 2TI(aL/d) = 2TI(3-8). Although neighbor
perpendicular chain~ are closer, we are justified not only in neglecting
interactions between widely separated chains, but also in neglecting inter-
actions involving harmonics of the fundamental chain spacing even on nearby
chains. That is, the secondary order parameters <P~f(Q », etc. with m > 1
playa vanishingly small role in the interchain coup in~.
3. High Temperature Properties (T > Tc)

We discuss the thermodynamics using a generalized mean field theory in which

the interchain coupling is approximated by a mean field but the resulting
one-dimensional chain problem is solved exactly [6]. At low temperatures,
where the full nonlinear response of the chains is important, this formula-
tion leads to a sine-Gordon Hamiltonian, and thus is of most direct relevance
for this conference. It is worthwhile, however, to sketch some results for
T > Tc since they display several unusual features of this system and es-
tablish much of the necessary justification for the model itself. For
T > Tc we need only the linear response, Xo, of the harmonic chain, so that
<Px(q» = xO(q)heff(q) (6a)

heffCq) = hOCq) - VII Cq)<p (q» - I v.L(-q,p)lICp-q)<p/p» (6b)

x p
where we have now introduced Fourier components perpendicular to the chain
di recti ons, so that for the x-array q := (Q ,q ,qz) and for the y-array
p := (p ,P,pz)' The notation emphasizes the fuixed nature of the momentum
variabfes, with the components represented by lower case symbols being de-
fined modulo a reciprocal lattice vector and thus reducible to the first
Brillouin zone. This mixed momentum representation is also in evidence
through the function
lI(p-q) := 1 if Px = Q(mOdG);qy = P(modG'); Pz = qz;
:= 0 otherwise.

where G(G') is a reciprocal lattice vector of the x(y)-array.

Eq.(6), together with a ~imilar s~t defining <Py(p» are to be solved
for the coupled response X(q) := <Px(q»/hO(q), or equivalently the pair
correlation functions <Px(q)px(-q» = kTx(q). (We will justify shortly the
use of the classical form of the fluctuation-dissipation theorem.) Because
of the umklapp momentum terms, the solutions can only be developed pertur-
batively. Their character depends upon the relationship of the momenta
components along the two chain directions.
3.1 Uncoupled Solutions In regions of reciprocal space such that P and Q
are not approximately equal the two chain arrays are effectively decoupled,
and for the x-array

239
 <p (~)p (_~» = SO(Q) (7)
x x l+Sv" Cq)So(Q)
where S = (kT)-l and a similar expression holds for the y-array.
SO(Q) = kTXO(Q) is the pair correlation function for an independent one-
dimensional harmonic chain. It is given by
iQ(xo-xo) iOu -iQu iQda - ~2«U -u )2>
SO (d) = I e a ° <e . a e 0> = lee a ° (8)

and«ua -u o)2> may be evaluated as an ensemble average over the single chain
Hamiltonian, }to,
Tr/a
«u -u )2> = ~ f
dq (l-cosgd) = lal0 2 (9)
a ° 4Trmv2 s i n2 (.9i)
-Tr/a 2
where 0 = ~ d is the mean square fluctuation in nearest neighbor distance.
2 2
As is well WXown, even though there is a well-defined average spacing, ad,for
a'th neighbors, the harmonic l-d chain lacks long range order since the mean
square fluctuation about ad increases linearly with lal. Substituting (9)
into (8) yields a geometric series which can easily be summed to give
sinh(~2Q2)
SO (Q) = 1 (10)
cosh(~2Q2)-cos(Qd)

This is a typical liquid-like scattering function (see Fig. 4). Using the
measured phonon velocity, we find for H93-oAsF6, (0/a)2 = 6.4 x 10- 4 at
room temperature, which justifies the use of the harmonic approximation
within the chain.
In the high temperature limit (somewhat above room temperature for
H93_oAsF6) that we may set the denominator of (7) to unity and we recover
the independent chain limit. For Qd » (0/d)2 which is easily fulfilled in
this case, SO(Q) consists of a series of nearly Lorentzian peaks (the sheets
of scattering) centered at Qm = 2Trm/d with a half width at half maximum, Km,
given by Kmd = 2Tr2(0/d)2m 2.
The predicted value of K) is consistent with a resolution limited width
for the m = 1 sheet, but it should be possible to measure the widths of the
higher order sheets and to verify whether or not they are proportional to
m2 T [7].

As the temperature is lowered, the form of (7) and (8) shows that the ef-
fect of parallel chain intEraction is first evidenced near Q = Ql since
successive maxima in SO(Qm) are weaker, SO(Qm) = SO(Q))/m 2 . (This explains
the failure to observe modulation on the m = 2 sheet at temperatures where
such modulation was oronounced at m = 1.) The modulation along the sheet
is determined by vII Cq) and the existing data car; be fit semiquantitatively
with contributions from near neigh~or and next near neighbor chains only,
with vnnn ~ -2vnn ~ 0.14 K. (The interaction seems other than direct

240
 10

<T=.i.d
'10

Qd/2rt

~ The scattering function, SO(Q), for a l-d harmonic model (see Eq.(lO)
shows a typical liquid-like pattern. For this case, aid = 1/10, the corre-
lations are weak

Coulomb as vnn is the wrong sign and both are ~ 50 too large.) Although the
interactions are weak, they are sufficiently enhanced by the long coherence
length within a chain as to tend toward an ordered state only a few degrees
below Tc = 120 K.
3.2 Coupled Solutions. The character of the solutions of (6), together
with the corresponding ones for Py(p) are of a different character if the
momenta along the two chains are nearly equal, P = Q. For these momenta,
the x- and~-arrays are str~ngly coupled, giving rise to new fluctuation
modes, p±(q) = [px(q) ± Py(q)] and the fluctuation scattering is proportio-
nal to

( 11)

which for reasons discussed above is entanced for Q = Q ,that is along the
line of intersection of the m'th sheets (a reflection o~ the simple physics
of (1)) and most enhanced for m = 1. Whether an instability first
arises on the m = 1 sheet at (Qj,Qj,qz)due to perpendicular chain coupling
or at a more general position (Qj,qy,qz) due to parallel chain coupling
depends upon whether the denominator 1S smaller in' (11) or (7); the
former is the case for H93- oAsF6. We believe that the apparent failure to
observe critical scattering above Tc is the result of the fact that the
region of enhanced scattering is restricted to a linear dimension of order
2Kj in the (aT,bT) plane. Since this width is below the existing experi-
mental resolution, the basal plane scans should have thE appearance ot weak
Bragg scattering persisting above Tc. Just such scattering has been

241
observed, and it should be possible to establish its true character by de-
termining whether 1!he scattering is broad or narrow in the z direction, per-
pendicular to both chain arrays. In both sense (repulsive) and magnitude
v seems roughly consistent with Coulombic interactions.
4. Long Range Order
As usual, we associate the order parameter with the mode glvlng rise to the
divergent fluctuations (i.e. with the coupled ~@de soluJ.ions discu~sed above)
and thus define a complex orger parameter, nle 1 = <Px(qc»= ±<Py(qc».
The arbitrary phase factor e1W plays no role in determining the energetics
of the system and is associ a ted with a zero energy "s 1i di ng mode," famil i ar
in incommensurate systems. Fr·r convenience, we set W= O. nl specifies the
amplitude of the sinusoidal mcJulai-ion of the mean atomic density on a
chain, e.g. <PQ,x(Ql» = nlei·¢~x' where ¢~x is a phase associated with
the perpendicular components ot qc and can be made to vanish by an appro-
priate choice of origin for each chain. Using this convention the mean
field potential V, obtained by replacing one of the density operators in (4)
by its mean value, is identical for each chain. This allows us in what
follows to suppress ~e chain index, Q,i, and we are left with the problem
of a l-d harmonic chain in a (commensurate) staggered field,

v(nl) = ~ { ~(ua+l-ua)2 + nlhcosQlu a } (12 )

where h = 2(v" (q ) + vJ.(q )). To discuss the evolution of the low tempera-
ture phase we mus~ calcula~e the growth of all of the Fourier components of
the atomic density on a chain. This can be done classically using transfer
matrix techniques [8] since except at very low temperatures the effect of
zero point fluctuations are negligible. The long intrachain coherence
length, Ki l ~ 200 d, allows us to pass to the continuum limit
((ua+l - ua ) ~ d(au(x)/ax))and (12) reduces to the classical sine-Gordon
potential and we must calculate

( 13a)

= <~ o IcosQm
ul~ o
>/<~
o 0
I~ > (13b)

where ~ = ceo(q,v) is the lowest eigen vector of the transfer matrix and
satisfigs the Mathieu equation.

[ ~dv2 + (a -2qCOS2V)]
0
~ 0
(v) = 0 (14a)

q = 4KS2hni = -2 (~ )2 nl (14b)
Q~ Sc

where 2v = QIU. The transformation temperat~re lc = (kSc)-l is obtained by

setting the denominator of (9) to zero for q = qc.

242
 Eq. (13) can be readily evaluated by developing ~o(v) in a Fourier
series. When m = 1 (13) must be solved self-consistently with (14b). The
temperature dependence of the first three Fourier components of the atomic
density are shown in Fig. 5. For small III (T fI:I Tc)

(15)

and llm is proportional to lllm, while near T = 0

llm = III m % 1
2 m2 (L)
T
,18 c

",,,,
tOfll':::'"-------------,
\
\
\

, ,,,
\
\
\

\
\
\
\
\
\
\

o
TIT.:..

~ Self-consistent solutions of the sine;Gordon Hamiltonian for the

order parameter, llm. associated with the first three density wave
components

As T + O. the density distribution on the chain approaches that of a sum of

Gaussians centered at xa = nd with a mean square fluctuation
«X-Xa )2> = (TIl'l Q~Tc)'
The temperature dependence of the Bragg scattering associated with Q] as
studied by HASTINGS et al (see inset Fig. 3). rises much more quickly than
predicted by (15). We believe that the reason for this is that the fluctu-
ations associated with Vii may not be neglected for T % Tc because they are
divergent at T = Tl a few degrees below Tc' The coupling of the two types
of order parameters not only has the effect of promoting a more rapid
growth in 1l1(T) but also in suppressing the v" fluctuations below Tc. a
feature which is very noticeable in Fig. 3. As this aspect of the theory
is specialized to H9 3_oAsF 6 we will not pursue it further here.

243
5. Dynamics
We conclude with a brief discussion of the dynamical properties that are to
be expected in a system of loosely coupled harmonic chains. The dynamics
are readily susceptible to calculation and contain several novel features
which one can compare with neutron scattering experiments in progress.
In the high temperature limit, it is possible to redo the calculations
summarized in (8) and (9) for the time dependent pair correlations
°
S (Q,w) = (2n)
-1 J
dte
iwt
I <e
iQx (t) iQx (0)
a eO> = kTxo(Q,w)
a

The result, in the vicinity' of the m'th diffuse sheet (i.e.

I~Qm/ = /Q-Qml « d- 1 ) is

(16)

Eq.(16) deserves several comments.

1) The unusual product-of-Lorentzian form is characteristic of corre-
lation functions of one-dimensional problems [9].
2) In deriving (16) one cannot proceed through the familiar separation
into a product of a time dependent and time independent parts, as the
latter (Debye-Waller) term vanishes while the time dependent fluctuations
diverge. Similarly, there is no separation into one- and multiphonon
terms. Eq. (16) represent.s the total density response.
3) When (16) is integrated over frequency one recovers (10), and as
with a 3-d liquid, there is no truly elastic scattering (i.e. no term pro-
portional to o(w)).
It is possible to extend the above results to include interchain coupling
in the random phase approximation. The dynamical analogs of (7) and (11)
are obtained by replacing <Pi(q)Pj(q» by
Sij (->-)
q,w = ( 2n )-lJ e iwt <Pi(q,t)Pj(-qO»dt
->- ->-

and SO(Q) by SO(Q,w) in those expressions.

Below T the dynamics can be discussed in terms of weakly coupled sine-
Gordon sys~ems. For an individual chain there are two sorts of excitations
to consider [10]. The first are free solitons for which
16mv2hnl
w2 = ~2 + V2Q2; ~2 = ___
S S s (nfI)2
At low temperatures, the minimum energy necessary to create a soliton is so
large (-I\~s/k 'V 700 K) that these are not important thermal excitations, but
the gap vanishes as nl 1/ 2 near Tc. It may be possible to observe these
excitations directly by neutron spectroscopy. The second kind of

244
excitations can be described as bound soliton-antisoliton pairs or doublets
for which
w2 = 62 + V2 Q2. 62 = 46 2sin2(rr8v)
v v 'v s 2
where v = 1,2, ... 8- 1 and 8 = (rr~/2mvd). The maximum value of v is the
boundary of stability for breakup into a free soliton-anti soliton pair,
whereas for small v, Wv %3t(X1~1) 1/2 V and the excitations can be thought
of as ordinary phonons near the bottom of the sinusoidal potential. These
single-chain excitations form the basis for coupled collective modes which
satisfy the lattice translational symmetry. Of course at low temperatures,
the results may be obtained by making an harmonic approximation to the full
Hamiltonian, including the interchain coupling, but near Tc the interactions
are highly anharmonic.
6. Acknowledgments
It is a pleasure to acknowledge the collaboration of V. J. EMERY, who
brought to this endeavor not only many of the ideas presented here, but a
considerable sophistication which is missing in this account of it. We
both profited from discussions of the experiments with J. M. HASTINGS,
I. U. HEILMANN, J. P. POUGET, and G. SHIRANE. Research at Brookhaven was
supported by the Division of Basic Energy Sciences, U. S. Department of
Energy, under Contract No. EY-76-C-02-0016.
7. References
1. I. D. Brown, B. D. Cutforth, C. G. Davies, R. J. Gillespie, P. R.
Ireland, and J. E. Verkris, Can. J. Chern. 52, 791 (1974).
2. C. K. Chiang, R. Spal, A. Denenstein, A. J:-Heeger, N. D. Miro, and
A. G. MacDi armi d, Soli d State Commun. 22,293 (1977).
3. A. J. Schultz, J. M. Williams, N. D. Miro, A. G. MacDiarmid, and A. J.
Heeger, Inorg. Chern., March, 1978.
4. J. M. Hastings, J. P. Pouget, G. Shirane, A. J. Heeger, N. D. Miro, and
A. G. MacDiarmid, Phys. Rev. Lett. 39, 1484 (1977). .
5. V. J. Emery and J. D. Axe, Phys. Re~ Lett. 40, 1507 (1978).
6. This approximation was first described in a systematic way by D. J.
Scalapino, Y. Imry, and P. Pincus, Phys. Rev. B 11, 2042 (1975).
7. Neutron scattering measurements have verified these predictions. (Pri-
vate communication, I. U. Heilmann, G. Shirane, and J. D. Axe).
8. S. F. Edwards and A. Lenard, J. Math. Phys. 3, 778 (1962); N. Gupta and
B. Sutherland, Phys. Rev. A 14, 790 (1976). -
9. See, for example, S. A. Brazovskii and I. E. Dzyaloshinskii, Zh. Eksp.
Teor. Fiz. 71, 2338 (1976). (English transl., Sov. Phys. JETP 44, 1233
(1977)); A.Luther and I. Peschel, Phys. Rev. B 9, 2911 (1974):-
10. R. F. Dashen, B. Hasslacher, and A. Neveu, Phys.-Rev. D 11,3424 (1975);
A. Luther, Phys. Rev. B li, 2153 (1976). -

245
Charge Density Wave Systems: Thecp-Particle Model

M.J. Rice
Xerox Webster Research Center, Webster, NY 14580, USA

1. Introducti on
In this lecture I would like to introduce the subject of quasi-one-
dimensional (l-d) charge density wave (COW) condensates [1]; and to discuss
the ¢-particle model [2] which is a model of a charged non-linear excita-
tion of a weakly-pinned COW condensate. Such COW condensates exist at low
temperature in a number of well-known organic [1] and inorganic [3] linear
chain conductors which at room temperature exhibit quasi-one-dimensional
metallic behavior. It is intended that this lecture will provide useful
background material for the following lectures of Drs. luther, Friend and
Horovitz and also provide the non-specialist with a simple introduction to
the physical ideas basic to COW phenomena in quasi-one-dimensional systems.
2. The Idea of Peierls (1955)
In 1955 Peierls [4] pointed out that a hypothetical l-d metal would prob-
ably never exhibit metallic behavior because at absolute zero it would be
energetically more favorable for the linear lattice to develop a periodic
distortion with wavevector qo equal to twice the Fermi wavevector kF of the·
conduction electrons. Such a distortion leads to an insulating state at
absolute zero.
The argument is quite simple. In the uniform (metallic) linear chain
system at absolute zero (T=O) all the conduction electron states with wave-
vectors Ikl ~ kF are occupied, while those with Ikl > kF are unoccupied.
The density of states at the Fermi energy is finite. Now suppose the
uniform chain is subject to a periodic distortion in which the j th atom or
molecule is displaced from its equilibrium position Rj by an amount Uj given
by Uj = u cos(qoRj). Then, because of the electron-ion interaction, the
conduction electrons will be subject to a new periodic potential with wave-
vector 2kF. Such a periodic potential will open up an energy gap of magni-
tude 26 in the electron spectrum precisely at the wavevector kF. The den-
sity of states at the Fermi energy will be zero. Since the electronic
states below kF will now have been pushed lower, it follows immediately
that the electronic energy of the distorted chain will be lower than that
of the undistorted chain. If, for the distorted chain, the increase in
elastic distortion energy eEL is smaller than this decrease in the electron-
ic energy eEel, then the distorted linear chain will be the stable structure.
For the linear chain this will always be the case at absolute zero since
eEL will be quadratic in the amplitude u of the periodic lattice distortion
(PlO) whereas eEel will be of the form eEel = b u2 log(l/Iul) where b is a
constant. The logarithmic term is peculiar to one-dimension.

246
3. The Ideas of Frohlich (1954)
Independent studies of the l-d metal were made by Frohlich [5J. He
stressed the electronic response to the PLD and pointed out a series of
interesting consequences. The main points are as follows:
(a) The PLD is balanced by a periodic modulation in the macroscopic con-
duction electron density n of the form n(x) = no + on(x) where no is
the uniform density of the metallic state and on(x) = no A cos qox.
on(x) describes a CDW with wavevector qo and its amplitude A may be
related to the amplitude u of the PLD by use of force balance consid-
erations [5,6J. The CDW and the PLD sustain each other and may be
termed the condensate. In particular, it is to be borne in mind that
the CDW is tied to the PLD and not to the equilibrium positions of the
underlying atoms or molecules. ---
(b) The condensate will have a phase ~ specifying its position relative to
an external laboratory frame, so that in general the PLD and CDW are
described by the equations
u. = u cos (q x + ~)
J 0
on(x) = noA cos (qox + ~)

A variation o~ in the phase translates the entire condensate - and

hence no electrons per unit length - over the distance oX = o~/qo.
(c) For a translationally invariant system, such as was considered by
Frohlich, the condensate energy Ec is independent of the phase ~,
Ec t Ec(~): consequently the condensate may be set into a state of
uniform motion with velocity v. This corresponds to the time-
dependent phase ~(t) = ~o + vt and a furnishes the collective current
i = noe ~/qo

In the presence of an electric field E the equation of motion of the

condensate may be written
M* ~/qo = e*E
where M* and e* are the total inertial mass and charge of the conden-
sate, respectively. Consequently Frohlich proposed the translationally
invariant CDW condensate of the l-d metal as a mathematical model for
the then unexplained phenomenon of superconductivity.
4. The Ideas of Lee, Rice & Anderson (1974)
Now I should like to come to the ideas of Lee, Rice and Anderson (LRA) [7J
who with the advent of the experimental realization of quasi-l-d metals,
pointed out that, of course, in real systems translational invariance will
be broken by a number of factors. The principle factors will be:
(a) The discreteness of the lattice (commensurability).
(b) Interchain interaction (presence of CDW's on neighbouring chains).
(c) Lattice defects - particularly charged impurities.

247
The condensate energy Ec = E(A.4» will now depend upon the phase 4> as well
as on the amplitude A of the COW. Consequently the condensate will now be
subject to a finite restoring force
. aE
FR = - _cax = - q0 aE c/a""'.
Thus a preferred position. denoted by 4> = 4>0' may be expected for the con-
densate corresponding to FR(4)o) = o. The condensate may be said to be
"pinned". For small displacements 64> from 4>0' we may assume the harmonic
form FR = Ko4>. and writing the spring constant K as K = M*w~. where wF has
the dimensions of frequency. the equation of motion of the pinned conden-
sate wi 11 be
M* o~/qo = e*E - (w~M*/qo)64>
This equation describes simple harmonic motion of the phase with frequency
wF and leads to a frequency dependent conductivity of the form
1 e*2 2
a(w) = [ iwM* -f---2
w
wF-
and a static dielectric constant ES equal to
411e*2 1
ES "~-2-
wF

which diverges as wF + O. Here L denotes the length of the system. Weak

pinning. that is small wF' leads to a low frequency IR absorption and to a
large static dielectric constant. For the quasi-l-d Pt-complex compound
KCP [3]. wF is observed to be of the order of 2 meV and ES ~3000.
5. Pinning Potentials
I would now like to discuss briefly the form of the various pinning
pontentials [7].
(a) Commensurability potential: this arises when the ratio of the linear
chain reciprocal lattice vector G and qo is equal to an inte~er MM
There will then arise in Ec (A.4». a term of the form V(4)) ex (l'I/W)
(l-cos(M4») where Wdenotes the electronic bandwidth. The COW thus
moves in a periodic potential with phase period 211/M.
(b) Interchain interaction: The Coulomb interaction V12 between COW's on
adjacent parallel chains will be a function of the relative phases
4>1-4>2 of the two COW's:
V12 ex ~1~2 cos(4)1-4>2)
This interaction leads to ordering of the relative phases at low-
temperatures and gives rise to an effective periodic pinning potential
for a single chain V(4)) ex cos 4> with phase period 211.
(c) Oefect pinning: this is more complicated and in general leads to a
pinning potential which is not a single periodic function of 4>. E.g .•
for a random distribution of charged impurities V(4)) takes the form

248
 V(~) = L.1
V. COS (q 01
x. +
1 where x.1 describe the locations of the
~(x.))

impurit~es and Vi the magnitude of the impurity potential. As dis-

cussed by Fukuyama and Lee [8] this leads to the phase of the CDW
becoming pinned locally and consequently to a CDW ground state with an
inhomogeneous phase!
6. Order Parameter and its Behavior
I should now like to make some general remarks concerning the order param-
eter of the CDW state and the determination of its behavior.
the order parameter is complex and we may define it to be
Clea~ly,
~ = A e1~,
where A denotes the amplitude and ~ the phase of the CDW. In
general, both A and ~ may be functions of space and time, so that
op(x,t) = A(x,t) cos (qox + ~(x,t))

The behavior of ~ can be obtained if Ec[~] is known. A microscopic theory

is required to provide the latter. Unfortunately microscopic theory is
available only for simple models [1]. Thus one must proceed phenomenolog-
ically.
One way to proceed is to construct a Lagrangian density L(x,t) =
L[~(x,t)] which is argued to have a physically resonab1e form. Then mini-
mization of the classical action

o J f
tl
t2
dt dx L[~] = 0
yields the equations of motion for A and~. These provide the ground state
amplitude and phase profiles, and of course, also the excited state profiles.
Also, for fields with one space dimension, the thermodynamic properties of
the Lagrangian field may be rigorously investigated by functional integral
techniques such, as for example, those fol1ov/ed by Sca1apino, Sears and
Ferrell [9].
A physically resonable form for L[~] is
L[~] = cl;1 2 + t 1~12 - ~ 1~14 - dlvx~12 - Vpin[~]
where a, b, c and d are positive constants. In the absence of space and
time dependences and the pinning energy this Lagrangian generates a double-
well potential as a function of the amplitude A and is a form well-known in
other order-parameter contexts [10].
In general, however, the present Lagrangian describes two-coupled non-
linear fields [11]. Nevertheless if the magnitude of the pinning potential
is very small by comparison to the depth of the double-well potential for A,
the low lying excitations of the condensate will be dominated by deforma- .
tions in the phase~. These may be investigated by substituting into the
Lagrangian the form ~(x,t) ~ Aoexp(i~(x,t)) where Ao is the uniform ground
state amplitude. In this case the Lagrangian becomes a function of the
phase ~ only. This brings us to the ~-particle model [2] which is a model
of non-linear phase deformations for cases involving such weak pinning
potentials.
249
7. p-Particle Model
In the ~-particle model, introduced by Rice, Bishop, Krumhansl and
Trullinger [2J, the Lagrangian is constructed as follows. ns electrons per
unit length are assumed to have condensed to form the CDW, each with a mass
m*. A ground state with a perfectly uniform phase ~ is assumed and in
excited states the local phase ~(x,t) is measured re9ative to ~o' The
kinetic energy density is then equal to 1 m* ns~(x,t)2/q2. Spacial varia-
tion in the phase leads to a strain pote~tial energy den~ity and this is
taken to be PE = ns ~ K(q~lVx~)2. Re-expressing the spring constant K by

t
K = m* c~, where Co has dimensions of velocity, we have
PE = m* ns q~2 c~lvx~12. The pinning potential is written as ns Vp(~)
where Vp(~) is an even function of ~ and periodic in ~ with period 2n.
Without loss of ~enerality, Vp(~) is expressed in the reduced form
Vp(~) = m* q~2 wF V(~). For small ~, V(~) is assumed to possess the
harmonic form V(~) = ~ ~2. Combining these three energy densities we
obtain the Lagrangian density

0
{l
L[~J = L(~ o ) + ns m*q-2 2
~2
The Lagrangian is valid for small gradients in the phase but arbitrary
large variations in the magnitude of the phase.

It is useful to supplement this Lagrangian with the following physical

information [2J. The time derivative of the local phase generates the
local current density js(x,t) according to js(x,t) = ns e qo-l~(x,t). Also,
a spacial gradient in the local phase leads to a modulation en~ in the local
condensed electron density according to ens(x,t) = -ns q -lVx~lx,t). This
important consideration follows from the physical defini~ion of the local
phase. It is evident that js and ens satisfy the conservation equation
e ens(x,t) + vxjs(x,t) = O.
It follows from the Lagrangian that the equation of motion for the local
phase is ~ - c& V~~ + w~ (dV/d~) = O. This is a non-linear wave equation
of the Sine-Gordon type. Small amplitude extended wave solutions of this
equation correspond to the linear phase ~honons obtained by LRA [7J. These
have the dispersion relation w2 = w~ + Co q2. Here, of course, we shall be
interested in the solitary wav~ solutions.

8. Solitary Wave Solutions: p-Particles

The solitary wave solutions ~±(x-vt) are given by

±(x_vt)/(1_v2/c~)1/2 = (t/l2) ~~± d~V(~)-1/2

± n

in which the length t = co/wF' These describe a local shift in phase of

± 2n propagating at uniform velocity v (Ivl < co), The distance Dover
which the phase variation takes place is D ~ 2t Il-vZ/c2. Evidently the
o

250
solutions describe propagating domain walls which separate segments of the
condensate having common uniform phase. However, in.view of the relation
between the local condensed electron density and phase gradient, they also
imply propagating localized compressions or rarefactions in the local con-
densed electron density ns(x,t). The solitary wave solutions can thus be
regarded as describing pseudo relativistic particles of (rest) mass Mand
charge ± e*; these may be termed ~ and anti-~ particles.
The excitation energy of the ~-partic1es may be computed from the
Hamiltonian density to be E (v) = M c2//1-v2/c2 where the rest mass Mis
given by ± o 0

M= 8n s wF M*G/q~ Co
in which G is a numerical factor depending on the shape of the periodic
pinning potential (G = 1 for a SG potential):
TT

G= l S d~V{~) 1/2
18 0

Note that M+ 0 as wF + O.
The ~-partic1e charge e* is obtained by an integration over the phase
gradient

yielding e* = + 2e{n
_ s In). At absolute zero, for a non-gap1ess condensate,
ns = nand e* = + 2e.
Since the total number of condensed electrons is a conserved quantity,
the spacial integral over ens must be zero at all times. Thus, ~-partic1es
are created in pairs of positively and negatively charged particles.
9. p-Partic1es as Elementary Excitations
Can the non-linear ~-partic1es be regarded as elementary excitations of the
condensate? The answer appears [2J to be yes for kBT small compared to the
rest energy Eo = Mc 2. The functional integral technique [9J may be
employed to compute ~he classical free energy F. In the limit T + 0 one
finds a term in the free energy equal to Ft = -2kBT (L/2£) exp (-Eo/kBT).
This is just what we expect for a "lattice gas" of N{T) = (L/2£) exp-
(-Eo/kBT) ~-particle pairs distributed at random over L/2£ ~-particle sites
and (L/2£) anti-~-partic1e sites.
It is of interest to know the magnitudes of the ~-partic1e parameters
for assumed typical values of the phenomenological constants. The results
of a representative ga1cu1ation are shown in the Table below. There ~e 1
have assumed Co ~ 10 cm sec- 1 , m*/m ~ 10 2 , wF ~ 1.5 meV, qo ~ 5 x 10 cm- ,
ns = no and G ~ 1.

251
Table 1 Representative values of the ~-partic1e constants

e* 2t Mc02 M/m
0
+ 2e 88A 220 0 K 11.5

10. Conductivity
It is interesting that although the condensate is pinned, and therefore
unable to contribute a collective Frohlich conductivity [5J, the charged
~-partic1es now render it conducting. Indeed, for weak pinning the
~-partic1e excitation energy Eo = Mc~ may be anticipated to be much smaller
than the single-electron energy gap. in which case ~-partic1es will be the
dominant current carriers at low-temperatures. In this context we mention
that prior to the formulation of the ~-partic1e model the possibility that
non-linear phase deformations could contribute a conductivity was investi-
gated variationally by Pietronero, Strass1er and Toombs (1975). If we
introduce a phenomenological transport lifetime T for the ~-partic1es the
conductivity due to them will be given by [2J
1[2 nseT
OF = 2G Iii*"" exp(-E/kBT)
This describes a thermally activated conductivity with activation energy
equal to Eo'
In recent work Larkin and Lee [13J have shown that if charged impurities
are present ahd pin the COW locally the transport lifetime will be of the
form T = TO exp (-b Im*/m)where b is a constant. The exponential factor
represents the probability amplitude for tunneling of a ~-partic1e through
the impurity-pinned phase and severely reduces the ~-partic1e transport
time. In other recent work Maki [14J has shown that at absolute zero an
applied electric field F can lead to ~ anti-~ pair-production. This leads
to an interesting non-linear conductivity of the form o(F) ~ Fo/F exp(-Fo/F)
where Fo is a characteristic (de-pinning) field.
The low-temperature, thermally activated, conductivities of the various
organic quasi-1-dimensiona1 metals have been tentatively [15J interpreted
by the University of Pennsylvania group as arising from ~-partic1e trans-
port. The experimental investigations in this direction, however, are
still in an early phase of development.

11. Analogy with Schottky Defects

In conclusion, I would like to point out that the creation of ~-partic1es
in the COW condensate is analogous to the creation of Schottky defects in
an ionic crystal. This is illustrated in the figures drawn below which
depict the condensed electron density ns as a function of x. At T = 0, ns
is everywhere uniform and corresponds to the perfectly formed COW
(Figure 1).

252
 T=O

o ~-------------------------------+x
~ The condensate at absolute zero
At small but finite T thermodynamic considerations guarantee that there
will always be a finite number of ~-particle pairs, N(T) f O. These dis-
order the otherwise perfect CDW (Fig. 2).

e* T f 0
-e*

o L--------------------------------+x
~ Disordered condensate at T f 0
In this respect we may regard the ~-particles as the inevitable Schottky
defects of the perfect "ionic crystal" of which the perfectly formed CDW is
representative.
References
1. For a recent general review see G. A. Toombs: Phys. Reports 40C, 181
(1978)
2. M. J. Rice, A. R. Bishop, J. A. Krumhansl, S. E. Trullinger: Phys. Rev.
Lett. 36, 432 (1976)
3. P. Brue5ch, S. Strassler, H. R. Zeller: Phys. Rev. B12, 219 (1975).
4. R. E... Peierls: Quantum Theory of Solids (O.U.P., London 1955) p. 108
5. H. Frohlich: Proc. Roy. Soc. A223, 296 (1954)
6. M. J. Rice, S. Strassler, W. ~chneider: Lecture Notes in Physics 34,
282 (1975) --
7. P. A. Lee, T. M. Rice, P. W. Anderson: Solid State Comm. 14, 703 (1974).
8. H. Fukuyama, P. A. Lee: Phys. Rev. B17, 535 (1978) --
9. D. J. Scalapino, M. Sears, R. S. Ferrell: Phys. Rev. B6, 3409 (1972).
10. For example, see J. A. Krumhansl, J. R. Schriefer: Phys. Rev. Bll,
3535 (1975) -
11. For example, see S. Sarker, S. E. Trullinger, A. R. Bishop: Phys. Letts.
59A, 255 (1976)
12. ~Pietronero, S. Strassler, G. A. Toombs: Phys. Rev. B12, 5213 (1975)
13. A. I. Larkin, P. A. Lee: Phys. Rev. B17, 1596 (1978) -
14. K. Maki: Phys. Rev. Lett. 39,46 (19m
15. W. J. Gunning, S. K. Khanna; A. F. Garito, A. J. Heeger: Solid State
Comm. £1, 765 (1977)

253
The Soliton Lattice: Application to the w Phase

Baruch Horovitz
Laboratory of Atomic and Solid State Physics and Materials Science Center
Cornell University, Ithaca, NY 14853, USA

1. Introduction
In the present work we study the connection among three subjects. 1) Dislo-
cation defects in crystals. 2) The physics and thermodynamics of solitons.
3) Displacive phase transitions and the quasi-elastic scattering in the
disordered phase (the "central peak"). The idea that connects these subjects
is that a displacive phase transition also involves a change in the lattice
constant, i.e. a volume change. If a domain of the ordered phase appears
within the disordered phase, it will suffer a strain due to the discrepancy
in lattice constants. The interaction of these incommensurate structures
in a one dimensional (10) model leads to the solitons as described in some
of the previous contributions in this volume, (see A.D. BRUCE; P. BAK; J.D.
AXE) .

An excellent case for the study of this problem is the S-w transfor-
mation in Zr-Nb and similar alloys [1-5J. These alloys transform from a
bcc (S) into a hcp structure below ~610°C. However, when alloys with 5-30
wt% Nb are quenched from their high temperature bcc phase into room temper-
ature, regions of the so called "w-phase" appear, imbedded in a bcc matrix.

The w phase can be described by considering the [1,1,1] planes of the

bcc lattice, which form a sequence of the type ABCABCA ... (Fig.l). The A-A
separation is the nearest neighbor distance in the bcc-3a, where a is the
nearest [1,1 ,1J plane distance in the bcc. The ~l subvariant [6] is formed
by collapsing the B-C planes towards their midplane while the A planes re-
main unshifted. The w2 and w3 subvariants are formed by retaining the B or
C planes respectively and collapsing the other pairs of planes.

The w structure corresponds to a longitudinal displacement modulation in

the [1,1,1] direction of the form u sin(2nn/3 +~), where n indexes the
[1,1,1] planes and tp= 0, 2n/3, 4n/3 for the three possible subvariants.
In the ideal w phase u = um = a/3 and pairs of planes are fully collapsed to
form a hexagonal structure. For a partial collapse u<u m and the structure
is trigonal.

The w-structure has been studied extensively by neutron [1], electron

[2,5] x-ray [3] and Mossbauer [4] techniques. The interesting features of

254
 A

c
30
B

~---I-~~
@E~----~~-----~~A

~ The bcc structure viewed along its [1,1,1] direction. For con-
venience the [1,1,1] axis is expanded by a factor of 3 relative to the
perpendicular axes. The nearest neighbor distance is 3a. The w phase is
obtained by collapsing planes B, C, while planes A are unshifted

the experimental data are the following: a) As Nb concentration increases

the w phase reflections become more diffuse and tend to elongate perpen-
dicular to one of the [1,1,1] directions. Intensity ratios show [6] that
u is close to um for the 8% alloy, decreases to ~~ um for the 20% alloy, and
remains ~ ~um for the 30% alloy. b) In the 5-15% wt Nb alloys the w re-
flections appear at (or very close to) the expected positions of the w
structure, while in the 20-30% wt Nb the diffuse peaks are shifted away from
these positions corresponding to a larger modulation wavevector. This
suggests some kind of transition at about 17% wt Nb. c) Preliminary results
[7] show that as temperature is lowered to ~5°K the peak position of the
30% alloy is hardly affected, while the peak shift in the 20% alloy becomes
somewhat smaller but it is definitely present. d) The diffuse peaks are
elastic within 3.10- 9 eV [4] (at room temperature).

The shape of the w phase reflections (feature a) shows that the w phase
appears in rod shaped clusters [2,3] alon~ the [1 ,l,l]~ direction. Thus we
assume that the w clusters are one-dimensional objects which interact with
the surrounding 8 matrix. We do not use a microscopic theory for the S-w
transition. Instead we assume that the w phase prefers a smaller lattice

255
parameter than that of the S phase. It will turn out that this information
is enough to construct an effective Hamiltonian for an appropriate phase
variable in terms of which the scattering can be explained. This effective
Hamiltonian leads to the well known non-linear soliton solutions which we
call here stacking solitons. We then show that stacking solitons are stabi-
lized if the discrepancy 0 in the lattice constant is larger than some 0c
and then the ground state is an array of solitons. Thus for o<oc the sand
0' phase are commensurate, while for o>oc they are incommensurate. We suggest
that 0 changes with Nb concentration, and that a commensurate-incommensurate
transition occurs at ~17% Nb concentration.

In the next section we define the stacking solitions, following [8J. In

the third section we study the thermodynamics of solitons, compare with
experiment and suggest other experiments.

2. Stacking Solitons
Let us first examine the defect proposed by BORIE, SASS and ANDREASSEN
(BSA) [6J. They found that the correct shifts in the diffuse scattering can
be reproduced if a very particular sequence of subvariants, WI' w3' w2' wI' ...
exists along the W cluster. (Fig.2a). If the positions of the [1,1,1]
planes are described by
x = na + u sin (q na + cp ) (1 )
non
where q = 2rr/3a, then as n increases the BSA defect sequence implies that
C¥\, jumpg by + 2rr/3 from one subvariant to the next. Since subvariants are
degenerate ground states this defect may be thought of as a soliton [9]. An
antisoliton, which would vary locally from 0 to -2rr/3 is not compatible with
the BSA sequence. In fact, the form (1) implies immediately that a monot-
onically increasing function ~ leads to a local wavevector which is larger
n
than qo' and this leads to the observed shifts in the diffuse peak positions.

The BSA sequence was developed as an ad-hoc explanation for the shifts
in the diffuse peaks. This sequence, viewed as a soliton, is a proper exci-
tation mode, however it is not acceptable here for the following reasons:
a) individual solitons can be thermally excited but are not a property of
of the ground state. However, experimentally the peak shifts are present
even at low temperatures and the defects should be described by the ground
state. b) There is no reason for the system to prefer solitons or anti-
solitons; if present in equal numbers the peaks would not be shifted.

It was suggested [10] that the bcc instability is associated with a

phonon mode whose wavevector q differs from q. Such a modulation q is
.
lncommensurate with the latticem and solitons could
0
result [9.
] m
However,
the Mossbauer spectroscopy [3,4] shows that the inelastic scattering is
centered at the exact w positions. although the elastic portion is shifted
from these positions. Therefore the phonon intensity is centered at qo and
not at qm'

256
 '\ I

, ( )

\ \

\ \

(0) ( b)
~ Defects in an w phase cluster embedded in the bcc [1,1,1] planes.
a) The BSA structure [6] representing a defect in the modulation pattern.
b) The stacking soliton - the w modulation pattern is maintained but there
is a local density defect. Note that away from the defects the structures
(a) and (b) are in exact registry. Hence the structure factors of (a) and
(b) are very similar

In the present theory we concentrate on static properties; the Nb atoms

are smaller than the Zr atoms and act as a pressure source. This favours
the appearance of w clusters with a natural lattice parameter a - 8 embedded
in a matrix with a lattice parameter a.

The average positions of the atoms in the chain are specificed by a

sequence < x ) from which the displacement modulation is defined. The
values (x) ~re not necessarily separated by a - 8, since the matrix strains
the chainnand prefers values of (xJ separated by a. The actual positions
of the atoms are x = ex ) + u sin(q (x) + ; ), and in general u and;
n n non n n n
are position dependent amplitude and phase of the order parameter. Here we
consider the simpler case un = const. = U. [8]
The important physics is that variations in ; are coupled to variations
in the density. This is derived from the identif~cation of the local wave-
vector with the local lattice constant by the relationship
qoa = qloc • a loc (2)
where a loc =(Xn+ l) - (Xn) and

257
 (3)
qloc = qo + ( ~n+l - ~n)/aloc
Note that the modulation pattern is defined by r = q a/2n(for the w phase
r = 1/3), and is independent of <x). Eq. (2) meansOthat this pattern on
the chain is indeed locked into it~ own lattice, in spite of possible shifts
in the positions <x) Eqs. (2.3) lead to
n
<xn) = na - ~n/ q0 (4)

a loc =a - (~n+l - ~n)/qo (5)

The relation (5) between variations in the phase and the local density in
a continuum limit is well known for charge density wave systems [11,12].
From Eq. (4) we obtain the following equivalent relations:

xn = na + ii sin (q o
<nx ) +~ n"'n
) - «i / q0 (6)

x = na +
n
u sin (2nrn) - co/q
n 0
(7)

These results are different from the form (1). In particular, a solution
for ~ which varies from 0 to q a leads to the BSA defect in (1) while in
(7) tHere is no defect in the m8dulation pattern; it is the density which
is modified locally. (Fig.2). This demonstrates the physics behind (2):
The phase is locked to its own lattice i.e. the chain, but may be unlocked
relative to the surrounding matrix. Therefore the BSA type solitons are
not allowed in this scheme, but instead we can obtain "stacking sol itons",
that is, solitons in the phase of the chain relative to the surrounding
matrix.
The locking energy Vlock is a periodic function of the shift <x n)- na
with periodicity a. A simple form is ~cos [2n«x )- na)/a] so that
n

Vl k = w 2 [1 - cos(~ /r)] (8)

oc 0 n
The competing phase deoendent energy is the elastic energy of the form
~C [<xn+l)-<x n)- (a-o)]2/i. Thus we are led to the following phase
dependent Hamiltonian
:K = E{_C_ (1)1 _1)1)2 + /(l-cos1)l )}- 2C~ -ao (1)IN - 1)I_N) (9)
n 8n2 n+l non "

where 1)1 = ~ Jr. The effects of time dependence are discussed later on.
n n
In the last term of (9) 1)1 - 1)1 is the phase difference of the first and
last atom in the chain and m~asure~ the length change (Eq.4). Therefore 0
is identified as the pressure on the chain, and positive pressure (0)0)

258
favors increasing solutions ~ and length contraction.
n
The Hamiltonian (9) can be derived directly from Eq. (4) without identi-
fying ~ as a phase. This actually has been done [13] and should be classi-
fied asnan epitaxy problem. [14]. Our approach shows that change~ in the
lattice constant are coupled intrinsically to the phase of a complex order
parameter and therefore the solutions of (9) appear in the nucleation process
of a phase transition.

The equation of motion for (9) is the sine Gordon equation. The soliton
solution, when inserted in Eq. (6), leads to the defect in Fig.2b which we
call a stacking soliton. On the other hand a soliton solution in (1) leads
to the BSA defect of Fig.2a. Far from the defect centers the two structures
are in exact registry; thus for a narrow soliton the x-ray scattering from
the two types of defects are very similar to each other and to the experi-
mental data [6,8]. In the next section we show why the stacking solitons
should be the proper type of defect- as they explain all the other experi-
mental features.

3. The Soliton Lattice

Denote the soliton mass of (9) (for 8=0) by Es' (For explicit expressions
see [8,14].) The last term of (9) reduces this mass by ~ = C 8/a, and if
~>Es the ground state \1i11 have a finite density p of solitons. This is
described by the solutions of the soliton lattice [13-16]. Thus we obtain
a transition in the ground state properties of the chain. For 181 < oc(oc
aE /C) there no solitons and the phase is locked ~(X) = const., while
fOf a > 8c a soliton lattice appears. As a grows the soliton density in-
creases, until for %c 7 00 the solitons merge into a line ~(x) = 2nox/a2
which is the free chain with its lattice constant a-8. (Clearly, if
a < 0c an array of anti-solitons will appear.)
The behaviour of the density p as function of temperature T can be studied
in the continuum limit using the transfer integral technique. [15J. In this
limit the mass of an isolated soliton is Es = 4w !C/n and its width is
£ = a!C/(2nw ). As T 7 00 the locking potential i~ less effective and
o
p 7 o/a 2 as for the free chain. For a finite T one can use various ex-
pansions. For T« Es -~, [17]

p = (2E/n T) exp [-(E s - ~)/T] ( 10)

Eq. (126) of [15J is a high T expansion; the same method can be used to
obtain an expansion in l/~ valid for all T. The result is
8E 4
p = _2~_ [1- (~)4 s
n
2 £E s \. 8

The function p(~) is shown in Fig.3 at various temperatures. There are

three qualitatively different regions: 1) For ~ «E the ground state has
p = D, while at low temperatures (compared with Es-~sand ~) solitons are

259
thermally excited [17] as described by the leading term of ~ p in (10).
2) At ~ ~ Es temperature has the most dramatic effect. At T = 0 there is a
sharp phase transition - the commensurate-incommensurate transition at
~ = E. It is a continuous transition, but "almost first order" since
p ~ [s~(~ _ Es)]-l near the critical~. For T t 0 the transition is smeared,
as expected in a one dimensional system. Therefore p is a sensitive function
of T when ~ ~ Es' 3) In the third region ~ »E temperature has a small
effect on p- see (11). The ground state has a finite p which is close to
the high T value.
pL~--------------~
t
0.3
a} T=O

0.2

0.1

0.5 1.0 1.5

-fLIEs
~ Soliton density p(~) in the continuum limit at various temperatuares.
Curve (b) is obtained from (10) and (11) which are valid at the lower and
upper regions respectively; the dashed portion is an extrapolation between
these regions. The mass and width of an isolated soliton are Es and l
respectively

In a discrete model the exact solution p(~,T) is different from Fig.3

[14], however the qualitative features should be the same. These features
can explain the experimental data on the w phase systems. This can be done
if we assume: a) the w cluster is long enough compared with the lattice
constant and the soliton width. Therefore the effects of the end atoms on
the amplitude u and the soliton solution can be neglected. From the fit to
the x-ray data on the 20% and 30% alloys [8] an w cluster has ~40 planes, the
soliton width is ~a and each cluster has ~l soliton. b) Each plane within
the w cluster consists of a small and equal number of atoms so that the
locking potential is constant within this plane and has the periodicity a.

260
This is consistent with the observed rod shaped w cluster [3,5].

We proceed now to show that the w phase systems are a laboratory test for
the features of Fig.3. As Nb concentration increases the discrepancy
o(~~) in lattice constants increases; 'therefore the ~ axis in Fig.3 is a
measure of the Nb concentration. The stacking solitons shift the scattering
intensity away from the commensurate wavevector q ; therefor the p axis in
Fig. 3 measures the magnitude of the peak shift. °The commensurate-incom-
mensurate transition is at ~17% Nb; in higher %Nb alloys stacking solitons
appear in the ground state, partially unlocking the w cluster from the S
matrix.

The temperature dependence is consistent with Fig.3, even if the ampli-

u (or the coefficients in (9)) do not change with temperature [7,8]; the
temperature dependence comes from the liT factor in the partition function.
The 30% alloy correspond to o»oc (~»Es) and its peak shift should depend
weakly on temperature. The 20% alloy has 0 closer to 0 and its peak shift
should depend more strongly on temperature. These results were indeed
observed in a preliminary experiment [7]. We predict that the alloy with
~17% Nb is the most sensitive to temperature change since then 0 ~ 0c (~ ~ Es)
and that the slight peak shift of the 15% alloy disappears at low temper-
ature (~ < Es )'

We also suggest the following experiments: Apply high pressure on the

quenched Zr-Nb alloys at room temperature and then release the pressure. As
in the pure Zr case [18], the alloy should transform into a 3D w phase, and
this phase should be retained after the pressure release. We predict that
the resulting w phase will have a smaller lattice constant in its [0,0,1]
direction than the previous bcc lattice in its [1,1,1] direction. The
change 0 in the lattice constant should increase with Nb concentration and
the critical 0c at ~17% Nb can be determined.

In a discrete system the solitons are free to move only if their width is
large compared with the lattice constant [14]. Otherwise the solitons are
locked and the vibration spectrum is above some finite frequency. In the
w clusters the solitons are narrow and therefor they form a static con-
figuration. This is consistent with the extreme narrowness of the quasi-
elastic peak. [4].

Finally we suggest that stacking solitons are relevant to the general

problem of the central peak. Evidently we look for transitions which in-
volve a change in the lattice constant, but this is probably true for all
displacive transitions. Also one should study the appearance of stacking
solitons on a surface of 3D clusters [19].

Models for the central peak [20,21] consider 10 systems, and it is still
not understood why domains in 3D systems are so stable and quasi-static.
These domains can either move around (phase fluctuations) or disappear

261
(amplitude fluctuations). Both phenomena can be seen by computer simulations
[22], and they tend to broaden the scattering intensity S(q o ,w). Consider
now the effect of stacking solitons: If the soliton is not too wide compared
with the lattice spacinq, translation requires finite energy and phase fluctu-
ations are suppressed; this can happen in a discrete model also without
stackinq solitons. In addition however. the stackinq soliton suppresses also
amplitude fluctuations, since by taking the amplitude to zero the density
defect is not removed (Fig.2b). Thus there is an energy barrier against
motion or decay of domains, and then the central peak becomes narrower.
Experimentally stacking solitons are seen by a shift of the peak in
S(q ,w) away from q , and we suggest that this shift correlates with in-
o . 0
creaslng narrowness of the peak around w = o.

A well known example are the A15 compounds Nb 3Sn and V3Si [23,24] which
transform from cubic to tetragonal at 45°K and 210K respectively. Evidently,
this transition involves changes in the lattice constant and stacking solitons
are favored. In fact, the central peak was found to exist also away q = 0
which is analogous to the shift in the w reflection peaks. Various properties
of this transition were explained by postulating the existence of defects
[25J. Stacking solitons are natural and intrinsic structures which can
represent these defects.

ACknowledgments
A large part of this work was carried out with J.A. Krumhansl and J.L.
Murray. I am very grateful to D.H. Bilderback and B.W. Batterman for
presenting me their unpublished data. I also wish to thank S. Aubry, T.S.
Kuan and S.L. Sass for very useful discussions. This work has been supported
by the Cornell University Materials Science Center, Grant #DMR-76-81083,
technical report #3042.

References
1. D.T. Keating and S.J. LaPlaca, J. Phys. Chern. Solids 35, 879 (1973).
2. S.L. Sass, J. of Less Common Metals 28, 157 (1972); C.W. Dawson and
S.L. Sass, Metall. Trans. 1, 2225 (1970).
3. W. Lin, H. Spalt and B.W. Batterman, Phys. Rev. B11, 5158 (1976).
4. S.K. Andersen and B.W. Batterman, (to be published).
5. T.S. Kuan and S.L. Sass, Phil. Mag. 36, 1473 (1977).
6. B. Borie, S.L. Sass and A. Andreassen. Acta Crystallographica A29.
594 (1973). -
7. D.H. Bilderback and B.W. Batterman, (private communication).
8. B. Horovitz, J.L. Murray and J.A. Krumhansl, Phys. Rev. B, (to be
published) .
9. This has also been noticed by R. Pynn, J. Phys. F~, 1 (1978).
10. H.E. Cook, Acta Metal. 23, 1041 (1975); J.M. Sanchez and D. DeFontaine,
J. Appl. Cryst. 10, 220\1977).
11. T.M. Rice, Solid-State Commun. ]2, 1055 (1975).
12. B. Horovi tz and J.A. Krumhansl. Sol id State Commun. 26, 81 (1978).

262
13. F .C. Frank and J.H. van der Merwe, Proc. Roy. Soc. A198, 205 (1949).
14. S. Aubry, J. Math. Phys. (to te published). See alsoln this volume.
15. N. Gupta and B. Sutherland, Phys. Rev. A14, 1790 (1976).
16. W.L. McMillan, Phys. Rev. B14, 1496 (1976); P. Bak and V.J. Emery,
Phys. Rev. Lett. 36, 978 (1976).
17. J.F. Currie, M.B. Fogel and F.L. Palmer, Phys. Rev. A1§., 796 (1977).
18. J.C. Jamieson, Science 140, 72 (1963).
19. B. Horovitz, J.A. Krumhansl and E. Domany, Phys. Rev. Lett. 38, 778
(1977). -
20. J.A. Krumhansl and J.R. Schrieffer, Phys. Rev. Bll, 3535 (1975).
21. S. Aubry, J. Chem. Phys. g, 3217 (1975); 64, 3392 (1976).
22. T. Schneider and E. Stoll, Phys. Rev. BJ.l, 1216 (1976). See also T.
Schneider in this volume.
21 S.M. Shapiro, J.D. Axe, G. Shirane and. T. Riste, Phys. Rev. B~, 4332
(1972) .
24. J.D. Axe arid G. Shirane, Phys. Rev. B8, 1965 (1973).
25. C.M. Varma, J.C. Phillips and S. T. Chui, Phys. Rev. Lett. 33, 1223
(1974); J.C. Phillips, Solid State Commun. ]§., 831 (1976);K.L. Nagi
and T.L.Reinecke, Phys. Rev. B1§., 1077 (1977).

263
The New Concept of Transitions by Breaking of Analyticity
in a Crystallographic Model

Serge Aubry*
Laboratoire Leon Brillouin, Orme de Merisiers, CEN-SACLAY, B.P. 2
F-91190 Gif-sur-Yvette, France

1. Introducti on

Phase transitions such as for example, incommensurate to commensurate mo-

dulated cristal U] metal-insulator in Peierls quasi-one dimensional metals
[2] , non registered to registered in adsorbed (or intersticial) atomic
monolayers [3] exhibit new and unexpected features. Despite other problems
like the zero-width central peak phenomena [4] , the Anderson-Mott metal-
insulator transition [5] , the spin-glass transitions [6] look very diffe-
rent, we believe that the ideas that we develop in this paper, should be
relevant to suggest a new approach to them. The main topic of this paper is
the description of the phase transitions versus parameters of a very simple
one-dimensional model for epitaxy at zero degree K. Most results are rigo-
rous but their proofs which are too lonq, are omitted. We try to emphasize
the role of two coupled concepts of defectibility and of frustration in the
origin of this new kind of phase transitions.
These ones are not usual symmetry breaking but involve analytical properties
of some characteristic functions. Despite mathematical objects such as dis-
continuous or non differentiable functions,Cantor set [7] have not very
much been considered by physicists until now, it is suggested that they play
a fundamental role in the conceptual description of many physical phenomena,
such as those mentioned above. In addition, it is shown that they have
indirect observable consequences.
2. The model.
We choose to study a very old model used to describe many different phenome-
na such as the dislocation motion in crystals [8.a] or adsorbed atomic mono-
layer [S.b] . It is likely the Simplest one for our purpose. Its potential
energy at 0 K is

~ ({u.})
1 r
= \ L.
1
( La)

with

where u· is the atomic position of the ith atom of the chain. The potential
V(u) with amplitude A is choosen analytical and periodic with period 2a :
V(u + 2a) = V(u) and symmetric V( -u) = V(u). The potential W(u) - P x u is
ana lyti ca 1 and convex: its second deri vati ve versus u: Woo (u Lis
*Charge de Recherche au CNRS.
264
strictly positive. P is a pressure term choosen such that the minimum of
W(u) is obtained at u = 0 (or equivalently at a multiple 2na of the period
of V).
Conditions on V and Ware not only convenient to get rigorous proofs but
avoid the occurrence of more complicated features, which should be conside-
red in further studies.
without periodic potential A = 0, the consecutive atoms, in the classical
ground-state, would be at a constant distance £0 defined by

(2)

This distance is generally different and incommensurate with the period 2a

of the substrate potential A V. When A is not zero, a conflict arises
between these two energy terms W( u) - Pu and AV(u) Whl ch we ca I I by defi-
nition frustration, by analogy with spin-glasses. Here the frustration is
determined by two parameters: the pressure P which controle as we will see,
the creation of defects, and the amplitude A which determines the relative
force of the two conflicting terms and by this way, the interaction between
defects.
We study the classical ground state of this model at fixed volume
<ui+l - ui> = ~ and at fixed pressured P. We discuss first the e~istence
and the properties of defects in (1) and show that they are essentlally a
consequence of stochastic features produced by the strong non-linearities
of (1). Defectibility makes non trivial the frustration effects.
3. Deflectibility and stochasticity
The stationary (time-independent) configuration family of classical crys-
tallographic model may contain other configurations than the ground state
if defects exist (i.e.) if the system is defectible.
3.1. The formalism
Stationary configuration of (1) are given by

(3)

In (1), ¢ has been written as an action with the Lagrangian Li with a dis-
crete time (L replaces the usual time integral f dt).
i
A conjugate variable Pi to ui can be defined as
aL
Pi aU i - W' (u i - ui - 1) (4)
We rewrite (3) and (4) using an operator TA defined as

(5)

which is the evolution operator of the canonical system associated with

265
the Lagrangian (1.b) (for details. see ref. [9]). This operator T).. is
inversible and preserve the measure dp A du in the phase space (P.u). All
the solutions of ' equation (3) are generated by the successive applications
of the operators T).. and T\l to all the initial points (uo'Po) and are
then represented by some trajectory in the phase space (u.p). (This forma-
lism. generalisable to most crystallographic models is the transfer matrix
method at 0 K).
ui is~generally unbounded.~For convenience and since V(Ui) is periodic. we
set ui = ui (mod 2a) and T)..

Pi V'(U~)
( ~i+1
+)..
) (6)
u +
i 1
( ( W-1 (p. 1) + ~.) mod 2a )
I
1+ 1

The ~nalyticity of V and l~ impl ies that \ which appl ies the cyl inder
(Pi.Ui) onto itself is analytical.

3.2. Trajectory properties

Fig4!e (1) shows several examples of mapping generated by the trajectories
of T).. (we choose).. = 1. V(u) = (Eo/2)(1-cos(1Iu/a)) W(u) = 1/2 C u2 •
and plotted more conveniently. the sequence of points Pi = ((ui/a).(ui_1/a))
mod 2 for i = 1 to 1000 instead of the sequence (Pi (ui - ui_1)/C • ui mod
2a).
We observe
1. qth order fixed point F (called qth order cycle) such that ....Tq(Fl = F.
For example. the first order fixed points H =1').. (H) and Fo = T)..(Fo and the
the second ord~r ~ixed pOint set FZ = TA (F 1). Fl = TZ(F 1). Such a trajec-
tory {u i } sat1sf1es for any

u.
1+q
= U.+2pa
1
(7)

where p is some integer independent on i. The corresponding stationnary

configuration has a unit cellof q atoms and of length 2pa. They are perio-
dic commensurate configurations of model (1).
2. Trajectories dense on q qth order smooth and regular orbits ~ich are
mapped for example. by the iterated sequence for 1 : n : 1 000 : Tr (M 4 )
(see the closed curve around Mo which is first order) or the seque'nce
!r(M?) (see the two closed curves around Fi and F2 which are second order)
Tne existence of such kind oflr)..-invariant curve 1S proved by the
Kolmogorov-Arnold t·1oser theorem (called K.A.M.) (see reJ. [9al theorem 21.11
or ref.[12] theorem 2.11). In addition on these curves T~ (where q is the
order of the curve) reduces to a simple rotation with an angle q k incom-
mensurate with 211 . (dependent on the considered orbit).
The corresponding configuration can be considered as a periodic commensurate
configuration (7) modulated by a surimposed periodic modulation with a wave
vector k/a incommensurate with 211/a (see Ref. [10])
A specially important kind of smooth trajectory (which does not exist in

266
the case CaZ/E = 4 but which exists for larger values of this parameter)
map closed orb9t on the two dimensional cylinder (Pi, ui) which are not
homotopic to zero. On these trajectories ui can be written as

u·1 f(i£ + a) = i£ + a + g(i£ + a) (8)

where f and g are analytical functions,f is monotonous increasing and g is

periodic with period Za, £ is incommensurate with Za and a is an arbitrary
phase.
These configurations are periodic and called incommensurate with period £.
(see section 4).
3. Erratic trajectories which are mapped for example by the sequences
~~(Ml) or l~(Mz). The resulting clouds of points are almost not distingui-
shable on figure 1. The closure of these sets of point are not analytical
curves, but are complic~ted shape set, called in mathematics Cantor sets.
The existence of these TA invariant non analytical sets is proved when
some conditions are satisfied which, as we shall see in the next, is the
localized defect existence. The corresponding configurations are (generally)
disordered without any defined period
Fi
u~-I(mod 2)1-:--·--,----..-:..7C<~~.':v.i.----r-J;~...-..- ..-.·-"-'.:-::~~:~'~:;~...."">-::-":~~.'-"~:::---'.:'~',-:C1.' H
:;~.::.: ~~:~., .t:' .. '.~. ~:./:,:'~.:': .::. ':'

:. :. .:~.:"J'i;flS,",:_,' ..• ,,:,';

-:,. . .. . ..............\
.,
,= :.- ~.::.: ~'.:' "

\~~;t::.. :: .
:: ,.: . :. ~;::.
:.

i· ...:· -.,'

Figure 1
Such a mapping behavior can be considered as the characteristic consequence
of strong non-linearities of model (1). For example, the TA mapping behavior
becomes gradient-like and smooth when V(u) is replaced by a non-linear but
single well potential

267
3.3. TA -stability of trajectories
The stochastic property of each trajectory can be simply related to its
linear stability.Giving a TA trajectory ~i = ~(~o) = (Pi'u i ) which is not
a set of fixed p01nts, a small perturb~tion ~o ~n ~o is ~ultiplied by the
linearized part J A of the operator TA in Mo ,M 1 ,M 2 ,·· "~n (called
Lyapounov matrices). Because TA is. measure-preserving, their determinant
are unity. Then we set at the point IT , the matrix definition:
o
=- J (l4 n_2 )
In(Mo) J(~ln_l) x x J(M a ) (9a)

(
with A V" (u.)

)
/
1

J(M.1 ) , A V"(u i )
1+ (9b)
W"(u i +1 - ui ) W" (u i +1 - ui )

If this matrices product (9.a) diverges for n going either to +00 or to -00
this trajectory ~; is linearly TA unstable. In fact, it is unstable even
taking into account the non-linearized transformation and this trajectory
turns to be generally erratic.
If the matrices product (9.a) is bounded for any n, the trajectory M is
linearly stable and in fact turns to be generally stable by the non-~inea­
rized transformation TA . Then its closure is a set of q closed analytical
curves.
When the trajectory M. is stable and represented by a set of q fixed points,
these fixed points ar~ called elliptic if the eigen-values of the Jacobian
matrix Jq OfT~ are complex conjugate with modulus unity e±i8. Then jq is
equivalent to a rotation on an ellipse. The case where 8 = 0, is marginal
and the fixed point is then parabolic.
If the trajectory is unstable, the qth order fixed point is hyperbolic. The
corresponding eigen-values are real as well the eigen-directions. The eigen-
direction which corresponds the eigen-value larger than 1, is dilating and
the second one is contracting.
3.4. Homoclinic, heteroclinic point trajectories and defects
Let us consider now a qth order hyperbolic fixed point HI' Its two real
eigen-directions are parallel in HI to the dilating and contracting sheets
~f Hl . The first one is defined as the set of points P such that the sequence
TAqn (P) converges to Hl when goes to +00. The second are has the same defi-
nition for n going to (see ref. 10 for example).
-00

When these two lines intersect transversally at some point h called a homo-
clinic point, then it is proved that erratic trajectories are generated
(see for example in ref. [12] theorem 3.7 and figure 17). Their closure is
a Cantor .~t. Such a situation does exist close to most elliptic fixed
point of TA (ref. [12] theorem 3.9). By definition, the trajectory genera-
ted by h is identical to the trajectory of HI for large i, i going to +00
or _00 • This trajectory represents physically nothing else than a localized

268
defect configuration of the periodic and commensurate represented by HI'
More generally a localized defect is referred to tW2 periodic configurations
which are necessarily both represented by unstable TA trajectories. These
ones cannot be erratic and then are both represented by hyperbolic fixed
points HI and HZ' with some orders equal to their periodicity at +00 and -00.
This localized defect must be represented by the trajectory of a point belon-
ging to the intersection of the dilating sheet of HI and of the contracting
sheet of HZ' (see Ref. [10]). These pOints are called heteroclinic points.
The configuration represented by HZ may be the same as the one represented
by HI apart from a phase shift. Then the obtained configuration is by defi-
nition a phase defect of a periodic and commensurate configuration.
In continuous models (where the index i is choosen continuous), for example
the well known Sine-Gordon equation, these phase defects are easily found
as moving kinks or better as solitons.
In fact, in discrete lattice, solitons or moving kinks are exceptional. In
general they are locked.
We shew in ref. [10] that this is the fundamental property which explains
physically the stochasticity which occur in solutions of equations similar
to (3). A defect is submitted to a locking force from the lattice and to
interacting forces from the other defects. If the distance between neighbor-
ing defects is sufficient, the locking force is larger than the interacting
forces. Then to a random distribution on the lattice of defects, corresponds
a stationary configuration. It is represented,by its definition, by an erra-
tic trajectory. This is the physical translation of the theorem of (ref.[lZ]
theorem 3.7) on homoclinic points.
For larger concentrations of defects, the resulting interacting forces may
be stronger than the lockinq forces. The defects cannot be maintained at the
chosen position in the lattice. If no other lattice effect intervenes,
(commensurability of the defect mean distance with Za) after rearrangement
the defect must be regularly spaced. Then the lattice locking forces vanish
and the interacting forces balance exactly. The resulting trajectory (repre-
senting this configuration)is dense on a set of analytical closed curves.
This is the physical interpretation of the K.A.:~. theorem.
More generally, the (stationary) defect properties of any classical model
at O.K. are exhibited by the stochastic features of the transformation (5)
of this model. (This remark is generalizable to continuous models, but
since there is no lattice, stochasticity describe only the interactions
between the defects).
3.5. Gap in the stationary configurations spectrum and dynamical stability
Assuming a mass m for each atom of the chain the time-Fourier transform of the
linearized dynamical equation, of the small motions {Ei(w)} around a statio-
nary configuration {ui} is

(>.. V"(ui) + W"(u i +1 - ui ) + W"(ui - ui - 1) - m wZ) Ei

o (10 )

A solution of (10) can be obtained for any w , by choosing an arbitrary

initial vector (E 1(w) , EO(W)) . Ei+l (or Ei - 1) is then calculated recur-
sively, as a linear combination, given by (10) of E.1 and E.1 - 1 (or E·+
1 l
)

269
for positive and negative i. However, since £i is a perturbation, a solu-
tion is physically acceptable if and only if £i(w) is uniformly bounded for
any i. For w = 0, omi~;ing some calculations for brevity, this condition
is equivalent to the fA -stability condition of 3.3.
Then we can conclude : The stationary configurations with a zero frequency
mode (gapless configurations) are represented byl'A-stable trajectories.
(which generally are dense on closed analytical curves). This stability im-
plies that no other configuration can be asymptotic to one of them at
i = ±oo and locally different. Such configurations are called undefectible.
If no zero-frequency mode eXists, the stationary configuration has~a finite
frequency gap (except in marginal cases) and is represented by a TA uns-
table trajectory. These ones are either a set of hyperbolic fixed points or
erratic and dense inside a ~ invariant Cantor set. With the above defini-
tion,this configuration 1S defectible when its corresponding trajectory
belongs to such a Cantor set (but is not necessarily dense in it).
To have a dynamically stable configuration (such that all the frequencies
w determined by equation (10) are real) we found that a necessary and
sufficient condition is that for w = 0, there is at most one change of sign
in the sequence of £i generated by equation (10) from an arbitrary (E1 EO)
Applying this criteria to the different kinds of configurations, we found
that
1. Configurations represented by qth order elliptic fixed pOints and
2. Configurations represented by qth order analytical curves, which are
homotopic to zero are both dynamically unstable (see another proof in ref.
[10] appendice A and 8).
As a result, the dynamically stable configurations of model (1) are neces-
sarily represented either by an hyperbolic fixed points set, an erratic
trajectory or an analytical orbit which is non-homotopic to zero as descri-
bed by (8). This last configuration only is gapless. However, these condi-
tions are not sufficient.
4. Transition by breaking of analyticity versus the amplitude of the perio-
dic potential
Fixing the boundary conditions as
(N-N°)~
( 11)
for N + +00 and N° + -00 the atomic mean distance along the chain is fixed
to an arbitrary value ~ . We study in this section, the classical ground-
state of model (1) with condition (11).
First, we examine the case where t is commensurate with the period 2a of
V(u) ~ = 2a (p/q) where p and q are two irreducible integers. We proved
(see ref. (16) that the ground state {u i } satisfies (7) and is then repre-
sented by a set of hyperbolic fixed points. In addition,we prove [15] the
existence of elementary phase defects. These are defect configurations
obtained by boundary conditions, generally such:
(N_N°)2a*±2qa ( 12)
and which has the minimum energy (1). They are represented by heteroclinic
points (see 3.4).

270
4.1. Proof of a phase transition versus A for incommensurate £
We proved [16] that when £ is incommensurate with 2a

i£ + a + g(i£ + a) (13 )

where f is a monotonous increasing function and g is periodic and antisym-

metric with period 2a. These functions depend on £ , A V and W but are in-
dependent on the arbitrary phase a .
Moreover, we proved [16] the existence of a constant Al independent on
£ (for £ in some interval £1 < £ < £2) such that for

(14)
functions f or g are necessarily discontinuous. The discontinuity point
set A is countable but dense on the real axis. With the same condition (14)
we succeeded to prove that the derivative of f is zero except on the set A
and that the variation of f is the sum of its only discontinuity jumps. As
a result f takes its values inside a Cantor set with zero measure which
proves partially the conjecture of ref. [16]. Then the closure of the re-
presenting trajectory of (13) is a Cantor s~t. (In addition, this Cantor
set is an invariant minimal closed set of lA which means that it contains
no other closed set which is invariant by TA ).
For small values of £ when £ is sufficiently incommensurate with 2a, the
K. A. M. theorem can be applied (ref. [g.a] theorem 21.11 or ref. [12]
theorem 2.11).
(The condition on £ means that £/2a is not a Liouville number, in other
words there exists a real positive number S such that the rational numbers
set which satisfy

1 ( 15 )
< 2+S
q
is finite. The measure of the Liouville numbers is zero which means that
almost any real number is not a Liouville number. This condition is useful
to know the behavior of series, the terms of which contains denominators like
1 - cos(2nn£/2a). (small denominators). There exists similar small denomi-
nator problems in other physical situations e.g. ref. [5] ).
When the parameter A is zero, the canonical system defined by (6) is inte-
grab 1e, since Pi is not dependent on i. It is a moti on i nvari ant for "fA

ui + 1 - ui = £ for any i,
To is a rotation with angle £ on the analytical circle, defined by Pi = P .
Slnce £ is incommensurate with 2a, the K.A.M. theorem states that for sma9l
enough perturbation A V on this integrable system, defined by some A2(£)
and
( 16)

there exists an invariant analytical curve close to this circle, on which

271
-
T\ is conjugated to a rotation with mean
gurations are described by (8)
angle~. The corresponding confi-

I'~ext, we prove that if for a gi ven £ such a traj ectory exi sts, it is the
only stationary configuration with atomic mean distance £ (apart phase
shifts). Therefore it represents the ground state with boundary conditions
(11 ) .
These two results show that when \ varies between \2(~) and \" some tran-
sition by breaking of analyticity of the hull functlon f must occur.
Numerical test [19] on the ground state configurations with condition (11)
shows a well-defined second order transition at some A = AC(£)' For
A > AC(£) we clearly see the existence of discontinuities on the function
f and a finite gap in the phonon spectrum frequencies which both disappear
atA=Ac(£).
4.2. Physical interpretation of the analyticity breaking transition
Fixing £ is equivalent to fix the number of phase excitations referred to
the state ui = 0 (epitaxy dislocations).Increasing \ is equivalent to increase
the locking force on the excitations while the interacting force which
depends mainly on Wchange very few.
When A < AC(£) the ground-state perturbation spectrum is gapless (see 3.5)
Equation (10) exhibits the translation mode Ei (at w = 0) defined as

Ei = f'(i£ + a) l+g'(H+a) ( 17)

Ei does exist because f is analytical and therefore differentiable.

The atomic monolayer is unlocked on the substrate periodic potential V(u)
From 3.5, this ground-state is undefectible which mean that there is no stable
rearrangement of the epitaxy dislocations building this ground-state.
When A > AC(£)' fO as conjectured in ref. [16] is zero and the gap is nume-
rically observable. The atomic monolayer is locked. This state is defectible
which means that there is many possible rearrangements of the epitaxy dis-
locations. The atoms of the chain, are trapped close to the minima of the
periodic potential V(u) and requires, to be moved, to jump its potential
barriers.
These features suggest an analogy with those of a metal insulator transi-
tion. The metallic phase is gapless, conducting and does not exhibit any
charge defect. The insulating phase possesses a gap, is locked and is defec-
tible by exitons.
More precisely, we believe that this model is not too bad to interpret the
metal-insulator transition of Peierls systems [2] . A substrate potential
AV with wave vector 2kF and amplitude A , could represent the electronic
potential on the atoms, the parameter A being non zero and increasing for
decreasing temperatures below To' However more sophisticated models, con-
sidering also the quantum motioh of the electron in the quasi-periodic lat-
tice potential [19] , could exhibit simultaneously a transformation of the
density of state into a purely singular measure. (Such measure has a zero
measure support as for example ~"(£) in the following). Simultaneously
the electron wave function should localize. There is other arguments sug-
gesting that analyticity breakings and "Cantorization" are involved in

272
metal-insulator transitions of various types [5],[17],[18]. (see [22] )
5. Continuous non-differentiable transformations versus pressure
The mean energy per atom of model (1) at P = 0 with the atomic mean distance
£ in ~(~). Finding the ground state of (1) at a given pressure P requires
to find the minimum of ~(£) - P£. The solution of the equation

1/J'(t)-P = 0 (18)
determines t(P).
5.1. Differentiability of p(t) and defect energy
We proved first that 1/J(t) is a convex function versus t . Its left deriva-
tive 1jJ~(t) and its right derivative 1/J+(t) are both defined. Although,
1/J+(t) ~ 1/J~(t) they are generally unequal when t is commensurate with 2a.
Let us assume that tr = (p/q) 2a with p and q two irreducible integers. We
saw that the elementary phase defect (12) of this configuration corresponds
generally to a phase shift which is ± (2a/q). Changing t in tr + ot(ot > 0)
is equivalent to add q(ot/2a) phase defects per atom. The change of energy
per atom is 1/J(tr + ot) - 1/J(t r ). For ot ~ 0+, the interacting energy between
phase defect goes exponentially to zero. and is negligible. The required
energy to add a single phase defect is
2a ,I,' (2a.2.) (19a)
q '1'+ q

and identically the energy to suppress a phase defect is

We prove [15] that generally e+(t r ) 1 e_(t r ).

Then, the condition (18) shows that t(P) is constant in some interval when
t = tr.
The ground-state with mean distance tr becomes unstable when the energy to
add an elementary phase excitation
e (t ) _ 2a P
+ r q

becomes negative (or similarly to suppress an elementary phase excitation).

This condition is also given by (18).
5.2. The devil'~ stair transition and the infinitely many phases diagram
Function 1/J'(t) which is monotonous increasing, has then infinitely many
jumps (1/J~(t\) - 1/J:(tr )) for any t = tr commensurate with 2a. The inverse
function t(p) defined by (18) has infinitely many steps and is called a
devil's stair [7].
We proved [15] that when the condition (14) is satisfied for t1 < t < t 2 ,
the second derivative~'{t) is zero for almost any t incommensurate with 2a.
(~his.fu~ction.must be understood as generalized [20J). Since ~"(£) is posi-
tlve lt lS a slngular measure. {Its support has a zero usual measure. This

273
implies that ~(P) is entirely composed of steps. The devil 's stair is called
complete and then the ground state is commensurate for almost any pressure.
For smaller values of A < AI' we conjecture that the devil 's stair l(P)
is not complete. In other word it is possible to find incommensurate ground
states for a finite measure set of pressure. Equivalently, this means that
~"(1) is not purely singular. It contains in its decomposition, an absolu-
tely continuous component, and a singular component. The support of the
continuous part has however a very complicated shape since it has a finite
usual measure and is included inside th~ set of 1 incommensurate with 2a.
When A goes to zero, the singular part of ~'(1) goes to zero. In addition
at fixed pressure, the incommensurate ground-states are always unlocked.
(See section 3.5).
We did not yet prove this conjecture, but reliable arguments based on mea-
sure estimations and also on the conditions of the proof of the singularity
of lji"(l) for large A > AI' are obtained.
5.3. An exactly calculable model
All the above results are valid when choosing V and Was analytical func-
tions.(They are still likely qualitatively exact if V and Ware at least
three time differentiable) [15] Choosing in model (1)
A V(u) = z1 wo(u
2
- 2m a)
2
(20.a)

with
m = integer part of (u 2: a) (20.b)
and
W(u) - Pu 1 2 (20.c)
= Z (u - b)
Potential V is not everywhere differentiable and then, this model does not
exhibit all the features mentioned above. Particularly, it is easy to show
that all the stationary configurations of this model are locked. From
reference [16] theorem 1,
u. +a
Int (-+a-) ---za .
Int(il+a)

Equation (3) provides u. as a linear combination of mj, and omitting some

calculations, we get readily:

(21. a)

with

fJ
(21.b)

and
( 1 + Int kl ) (kl _ Int kl) (21.c)
2a a 2a

274
~k(~) is continuous and linear by parts. When ~ is incommensurate with 2a
~"(~) = O. The ground-state is commensurate for almost any pressure and mo-
del parameters. This is consistent with 5.2 because incommensurate unlocked
configuration does not exist and then, cannot be represented in the phase
diagram.
Figure (2) sho~s an idea of the exact phase diagram of this model. The
point (l//C/w~, b/2a) lies in the region where the commensurability ~/2a
of the ground state is indicated. The plane should be completely filled by
such commensurate phase regions, without any first order line (compare with
ref. [8bJ). The grey area corresponds to the high order commensurate phases.

4 1
tC/w ~'

3 REGISTERED PHASE

0
0 11 1 1 t 1 2 1 2 1 3 2 34 1
10 9"8 7 "6 5 9 4 "1 3 "8 "5 79 "2
b/2a

Figure 2

The next figure shows a picture of some steps of the devil 's stair trans-
formation £(P) in this model [20] .
For analytical models (1), the phase diagram looks qualitatively the same
(the pressure P is on the x-axis and ~ on the y-axis). However in the
grey area of figure 2 there exists close to the x-axis a finite measure
but infinitely disconnected region in-between the comm~nsurate regions
where the ground-state is incommensurate, and gapless '

275
 RJo

C/W5 =6

0'------ b/o
Figure 3

5.4. Physical interpretation of the devil' s stair transfomation

Of course, it is impossible to distinguish experimentally the details of
the curve t(P). However, there is observable effects. If the devil 's
stair is complete, all the ground-states t(P) are locked. The transforma-
tion of these configurations requires that the atoms or more precisely the
defects, jumps the potential barrier of AV(U). An estimation of the effec-
tive barrier shows that when A is large enough (i .e. A > A1) this barrier
is weakly dependent on t. There is a global hysteresis schematized by
the formula
t
tobs.(P) "" t(P- 6) (22a)
and
(22b)

The observed value of t : t b ,should depend on the history of the system.

It is t6bs' if P has been i ngr~ased before the experiment and t;bs.i f it has
been decreased.
Such features have been observed in several experiments for apparently
continuous transformations (see for example [21] ).
For smaller value of A , the devil 's stair becomes incomplete; 6 in formu-
la (22) becomes more dependent on t and vanish for unlocked ground-states,
more precisely for incommensurate one such that ~ "(t) f O.

276
When A is small enough, 0 is practically zero and the transformation £(P)
is second order, without almost any hysteresis. At the opposite limit: A
large, £(P) varies sharply only on very small pressure intervals with an
important hysteresis. Physically this transformation establishes the conti-
nuity between a second and a first order transition which may be interpre-
ted also as blurred first order transition, often observed in various
experiments.

Acknwoledgments.
I thank for their encouragement and their hospitality J.A. Krumhansl at
Cornell University and V. Emery and M. Blume at the Brookhaven National
Laboratory where parts of this work have been done. I thank also V. Pomeau
for many useful discussions.

References
[1] r~. Izumi, J.D. Axe and G. Shirane, K. Shimaoka, Phys. Rev. B.!2, 4392,
(1967).
[2] R.E. Peierls, Quantum Theory of Solids (Clarendon, Oxford 1955, p. 108).
R. Comes, M. Lambert, i~. Launois and H.R. Zeller, Phys. Rev. B 8,571
(1973), -
[3J M.D. Chinn and S.C. Fain, Phys. Rev. Letters 39, 146 (1977);
A. Novaco and J.P. f4c Tague, Phys. Rev. Letters 38, 1286 (1977).
[4J T. Riste, E.J. Samuelsen, K. Otnes and J. Feder,-Solid State Communi-
cation 9, 1455 (1971).
s.r~. Shapiro, J.D. Axe, G. Shirane and T. Riste, Phys. Rev. B6, 4332
(1972). -
[5] D.J. Thouless, Physics Reports 13, 93 (1974).
[6] K. Binder, Advance in Solid State Physics, XVII 55 (1977).
G. Toulouse, Commun. Phys. 2, 115 (1977). -
J. Villain, J. Phys. C 10,4793 (1977).
[7J Benoit B. Mandelbrot, Form, chance and dimension, W.H. Freeman
and Company, San Francisco (1977).
[8a] F.C. Frank and J.H. Van der Merve, Proc. Royal Soc. (London) A198,
205 (1949). ----
[8b] S. C. Ving, Phys. Rev. B 3,4160 (1971).
[9a] V.I. Arnold and A. Avez,-Ergodic problems of classical mechanics,
W.A. Benjamin Inc. (1968).
[9b] L. Landau and E. Lifchitz, Mechanics p. 143, Pergamon (1960).
[10] S. Aubry, preprint (to be published in J. of Mathematical Physics).
[11] M. Henon, Quat. Appl. Math. 27, 291 (1970).
[12] J. 14oser, Stable and Random motions in Dynamical Systems, Princeton
University Press, N.J. (1973).
[13] D. Ruelle, preprint (1977).
[14] G. Benettin, L. Galgani, J.i'1. Strelcyn, Phys. Rev. Ali, 2338 (1976).
[15] S. Aubry,in preparation
[16] S. Aubry, preprint.
[17] E.I. Dinaburg and V.G. Sinai, Functional analysis 9, 279 (1976).
[18] D.R. Hofstadter, Phys. Rev. B 14, 2239 (1976). -
[19] G. Andre and S. Aubry, in preparation.
[20J I.M. Gel 'Fand and G.E. Shilov, Generalized functions, Academic Press
(1964) .
[21] W.D. Ellenson, S.i~. Shapiro, G. Shirane and A.F. Garito, Phys. Rev. B
16, 3244 (1977).
[22] w:- Rudin, Real and Complex Analysis, ~lc Graw Hill (1970).
277
Textures in Superfiuid 3He

Kazumi Maki
Department of Physics, University of Southern California
Los Angeles, CA 90007, USA

1. Introducti on
Superfluid 3He is a new phase of liquid 3He at extremely low temperatures
(below 3 mK). It consists of two distinct phases; 3He-A and 3He-B (although
the third phase 3He-Al appears in the presence of magnetic fields). The
phase diagram is shown in Fi~. 1. Superfluid 3He arises from the pair con-

30
P(6~
20

10

o
1.

TCmK)

~ The phase diagram of liquid 3He in the absence of magnetic field

densation of 3He-atoms just like the superconductivity in metals is due to

the electronic pair condensation. However, unlike the case of a supercon-

278
ductor, the pa1r1ng takes place in the spin triplet P wave state. This in-
dicates that the condensate is characterized by a nine-component complex
order parameter A~i (~, i = 1,2,3) where ~ and i are spin and orbital indices.
This large degree of freedom associated with the order parameter implies
high degeneracy in the ground state of superfluid 3He in the absence of ex-
ternal perturbation. In particular superfluid 3He possesses three distinct
classes of Goldstone bosons; zero sound, spin wave and orbital wave. Further-
more, superfluid 3He exhibits rich phenomena associated with textures like
liquid crystals.
Before going into classification of textures, it is necessary to specify
the order parameters for 3He-A and 3He-B and then to look into energies,
which control formation of textures. The order parameters of 3He-A and
3He-B are given by [1]

A~i = (A lIo(Tl d~ (6 1 + i6 2)i

1 i'l' ....
13 11 0(T)e R~i(n,a) (1)

respectively, where 1I0(T) is the temperature dependent amplitude. Here d

is a unit vector spec1fying the spin configuration of the 3He-A condensate,
~l' ~2 and i (= ~l x ~2) are a triad of unit vectors describing the orbital
configuration. In particular i designates the direction of the symmetry
axis of the quasi-particle energy gap as shown in Fig.2. The condensate of
3He-B is characterized by a phase function'!' and a rotation matrix R~i(n, a);
R~ l·(n, a) = cosa 0 .
~1
+ (1 - cosa) n~ n.1 + sina E
~
l·k nk (2)

~ig.2 The orbital triad

51' 62> i) and anisotropic
energy gap in 3He-A are shown

where n is the axis of rotation and a the rotation angle [2].

In general we mean by textures [~ the spatial conformation of and i a
for 3He-A and that of ~ for 3He-B. Textures are controlled by a variety of
energies. Of most importance is the kinetic energy associated with spatial

279
distortion of textures, the nuclear dipole energy and the magnetic aniso~
tropyenergy [lJ. In the Ginzburg-Landau regime the kinetic energy is given
by [1]
_1 3 * * *
Fk,·n - -2 K J d r {a,. AjJ ,. a.J AjJJ. + a·, AjJJ. a.J AjJl. + a., AjJJ. a., AjJJ.}

where

(3)

and m* is the effective mass of the quasi-particle. The nuclear dipole

t
energy ED and the magnetic anisotropy energy EH (which appears only in the
presence of magnetic fields) are given by

ED = - 2l xN Q2A J d3r '

(~ . '2
d)

8 2 3 1 2
15 xB QB J d r (cose + 4) (4)

and

(5)

for 3He- and 3He-B respectively, where xN' xB' and 6XA are the spin suscep-
tibility of the normal liquid, of 3He-B and anisotropic part of x in 3He-A,
QA and QB are Legett frequencies for 3He-A and 3He-B, Wo (= Yo~) is the
Larmor frequency, QH (-~O/kBTc) QB)is a small frequency, and H is the unit
vector parallel to ~.
Comparing Fkin with ED, we obtain a characteristic length ~(= C~/QA - 10jJ)
which we shall call the dipolar coherence length. Here C~ is the spin wave
velocity in 3He-A. In 3He-B we have also a similar length ~ = C~/QB' where
C~ is now the spin wave velocity in 3He-B. These lengths play an extremely
important role in the following consideration. Furthermore, in 3He-A
ED ~ EH yields H = HO (= 20 De).
Finally, in the presence of confining walls, i in 3He-A has to be orien-
ted normal to the wa 11 at the wa 11 surface [4]. For 3He-B on the other t
hand, there are two superweak surface energies Fs and FsH 2J;

Fs = - a J d2r (n+ • s)
'2 +' 2
(1 - g (n • s) )

FsH = - b J d2 r ~'s · ~
R +
(n· e0) • +~2
H (6)

where the two-dimensional integrals have to be done on the wa~l surface!

Here a ~ ED ~O(T), b ~ Yb (XN - XB) ~O(T) and ~O(T)(- 2 x 10- (l-T/Tc)-~ em)
is the microscopic coherence distance. Putting these terms together, we
have the total free energy

280
 (7)

Stable textures in superfluid 3He are determined as local minimum of F.

2. Classification
In classifying a variety of textures the homotopy theory is extremely use-
fu1. In superfluid 3He most of the analysis has been done by TOULOUSE and
KLEMAN [5] and by VOLoviK and MINEEV [6].
a. 3He-A
Textures in 3He-A are first classified depending on the importanc~ of the
dipole energy into three categories; 1) dipole-locked textures (til ± d all
over space), 2) dipole-generated textures (planar solitons where i = ± a is
broken locally), and 3) dipole-delocked textures. The first two textures
are found in an open geometry (0 > ~, where D is the linear dimension of the
system), while the last texture is found in a confined geometry (D < ~),
a
where is practically constant allover space. The internal spaces of
cases 1 and 3 are the same as SU(2). The homotopy theory tells us then
that there are two classes of textures with different dimensionality; linear
textures and point-like objects, although the latter appears to be dynamic-
ally unstable due to the orbital viscosity [7]. Linear textures consist of
singularities like vortex lines and disgyrations with Frank index ± 1 and
linear solitons (i.e., coreless vortices of MERMIN and HO [8J and of
ANDERSON and TOULOUSE [9J). Some of the disgyrations are shown in Fig.3.

Fi a:3 Typical textures with disgyration in 3He-A are shown. Solid lines
in lcate direction of t
Radial and circular disgyrations have Frank index 1, while hyperbolic dis-
gyration -1. In a long tube with circular cross-section, if t lies in a
plane normal to the axis of the tube, sum of Frank indices of all disgyra-
tions in the tube has to be 1. A coreless vortex of ANDERSON and TOULOUSE
[9J is shown schematically in Fig.4. If one circles around the periphery

281
Fig.4 The coreless vortices of Anderson and Toulouse
of the vortex, the orbital triad completes two full rotations. The coreless
vortex carries a topological charge

where the integral is performed over an area, which encloses the total cross-
section of the linear soliton as shown in Fig.S. For a MERMIN-HO vortex

Fig.S The surface over which the

integral (8) must be done

N2 = ± 1, while for an ANDERSON-TOULOUSE vortex N2 = ± 2. Indeed N2 is

proportional to the strength of vorticity associated with the linear soliton,
as is seen from the MERMIN-HO relation [8J

(9)

282
where Vs is the superfluid velocity.
When a coreless vortex terminates at a point on the surface of the wall,
this termination point is called boojum [1 OJ ' which can be either the sink
or the source of vorticity. Booja may play an important role in the stabil-
ity of superflow in 3He-A.
All the dipole-generated textures are planar solitons, which are mobile
domain walls. There are three distinct planar solitoQs; a-solitons, £-
solitons, and composite-solitons, which involve both d and i-fields. In an
open system the composite solitons are the only stable solitons. However,
i-soliton can exist trapped at the wall of the container in the presence of
a magnetic field. In particular in the presence of a magnetic field B,. the
a
composite iolitons are quite stable, as both £ and are confined in a plane
normal to H III I. This stability is guaranteed by a topological charge:
_ 1 Q
Nl - n fp dX i H
' • r:~ "-
x di £
~
d x di~ (10)

where the integral is done along a path connecting two points P and Q lying
on opposite sides of the soliton as shown in Fig.6. The planar solitons

1//J ~ J#-J
P
~
~
Fig.6 The integral path in (1O) is

,Q
?3
schematically shown
II

~
~
~
have Nl = ± 1. These solitons are most abundant in 3He -A.
b. 3He-B
In 3He-B textures are classified into two categories; 8-s01itons and ~­
solitons. 8-s01itons are domain walls generated by the dipole interaction,
whereas ~-solitons are generated by the superweak interaction. There are
two types of 8-S0 1itons; type I and type II [12J, although type I I 8-S0 1iton
is unstable and decays into ~-soliton Q1J. ~-solitons can be, on the other
hand, planar, linear, or point-like, although only planar solitons are stu-
died in great detail [13, 14].
3. Nuclear Magnetic Resonance
Perhaps nuclear magneti§ resonance provides the most sensitive means to study
textures in superfluid He, since it can explore small regions of dimension
of the order of t;. In l?articular spin waves are trapped by textures [15,16J
and inhomogeneities [17J in superfluid 3He. These spin wave bound states

283
can be seen as satellites in the nmr experiments since the composite solitons
in 3He-A, and n-solitons in 3He-B have a well-defined set of bound states,
nmr is extremely useful in identifying these textures. On the other hand,
to explore the dipole-locked textures in 3He-A,nmr is almost useless and one
has to rely on other techniques like ultrasonic attenuation.
3
a. He-A
The linear spin dynamics in the ~resence of textures is most readily formu-
lated in terms of Lagrangian [18J;
L = T - V

T = 21 xN f d3r (+w 2 - 2+ Wo . +)
w

v= of (11 )

where

Wy = COSa St + sina sins Yt

->
(M = - YOX ~) (12 )

and a, S, and yare Euclerian angles describing the rotation of d and YO is

the gyromagnetic ratio of the 3He nucleus. Note that the kinetic energy is
the same as that for a spherical top. of is t~e spin fluctuation free energy,
that is obtained fro~ (7) by ~ubstituting d = dO + oa and ~eeping only the
quadratic terms in od, where dO,describes the equilibrium d configuration
in the texture. Furthermore, ad can be expressed in terms of Euclerian
angles as

d(t) (= d(O) + ad) = ~ (a, S, y) d(O) (13 )

In particular for a composite soliton D6, 19J, we can parameterize d as

d = cosg {sin(~ + f) ~ + cos(~ + f) y} + sing ~

= sin~ x
h ,

dO + cos~ y
h ,

~O = sinx x + cosx y (14)

The resulting Euler-Lagrange equation is solved in two steps; first to de-

termine the eigenvalues

284
 Al t;2 ~ r 1 2 l 2
= - ds ~l - "2 a ) f~ + (1 - 2sin v)f

Agg = - t;2 ~
ds ~l - "21 a2) g~ . 2
+ (1 - Sl n v - (1 - - a ) t;
1 2
2
2 </I)
2
s g (15 )

where
+
s k x, a = kl cosx + k2 sinx, and v = X - </I (16 )

Then the satellite frequencies are given in terms of the above eigenvalues
as

w£ = (Af)
k
2 ~A
r2
~~
and wt = ~O + Ag ~AJ ( 17)

for longitudinal and transverse mode respectively. In particular for the

twi s t compos ite soliton [16J, we have

Aff = - t;2 f ss + (1 - 2 sech 2 (15 s/2t;))f

Agg = - t; 2
gss +6
(1 - 52
sech (15 s/2t;))g (18 )

where we made use of the relation

x + 4</1 = const
v = 2 tan - 1 ~xp (15 s/ t; 8
Eq. (14) yi e 1ds
13
1 f [sech [15 S/2t;Jr(15 - 1)
Af = "2 (/65 - ?) <X

A
g
= i5 g <X {sech [15 S/2t;] r/ 5 (19 )

More generally when the magnetic field makes an angle e to k the normal
direction of the domain wall as shown in Fig.?, the composite soliton changes
continuously from pure twist to splay like as the angle e increases from 0
to w/2. The corresponding eigenvalues Af and Ag are! calculated numerically
and shown in Fig.8 [19J. Here plotted are R£ = (Af)'2 and Rt = (Ag)~ instead
of Af and Ag. The longitudinal satellite in 3He-A was first observed by
AVENEL et al., [20J. The satellite frequency was reported to be
1
w/~A - 12

More recently an extensive study of these satellites in both the longitudinal

and the transverse case has been carri ed out by GOULD and LEE [21J. Thei r
results are summarized as

285
 R£ = 0.74 - 0.35 (1 - T/Tc)
and
Rt = 0.835 (20)

~ The general configuration is

shown

O.9r-_____ Rt

0.8

0.7

a6~------~~------~L-------~
o ~ ~ ~
e
Fig.8 Rt and R£ which appear in the satellite frequencies are shown

286
These as well as the earlier Orsay-Saclay result [2 OJ , are in excellent
agreement with the theoretical prediction, if we discard for a moment a
small temperature dependent term and if we assume that their longitudinal
satellite is due to the twist soliton (8 = 0) whereas their transverse sat-
ellite is due to the splay soliton (8 = n/2). Indeed in the Cornell experi-
ment, [21J they rotated the static magnetic field by 90 0 rather than the rf
field ln order to measure both the transverse and the longitudinal resonance,
which justified the assumption that longitudianl and transverse satellites
arise from ~elated but different textures. Furthermore, the observed tem-
perature dependence in R2 is interpreted to be due to deviation from the
Ginzburg-Landau regime [22J, which allows us to express Fla the Fermi 1iquid
coefficient in terms of the coefficient of the temperature dependent term in
(20). This yields Fla = - 1.03, which is remarkably close to the value de-
duced by OSHEROFF et al., Q5] from the analysis of their spin wave data in
superfluid 3He-B.
b. 3He-B
In superfluid 3He-B the most interesting textures are n-solitons in open
geometry [13, 14]. In the presence of a magnetic field the n-vector has two
preferred directlons;
->-
n/l ± H

The n-soliton is the domain-wall associated with these two equilibrium n

configurations. In particular in the limit wO » nB the spin wave spectrum
associated with a n-soliton is given by

(21)

where

S = k ->-x , ±
M ~ f± and MZ ~ fZ
->- ->- ->-
and nO is the n vector at the equilibrium configuration. For an-soliton
nOz is well approximated by nOz = tanh (ns), where

n - ~H- 1 (_= nH/C~) and ~H

( - 1 mm)

is the magnetic coherence length in 3He-B. Then (21) tells us that a ~­

soliton provides a repulsive potential to the transverse mode and an attrac-
tive potential to the longitudinal mode. Indeed, the longitudinal mode has
a series of bound states, which can be solved exactly in terms of the hyper-
geometric function [l4]. In particular the spin wave spectrum is given by

(w£)n = nB 0 ~2n + 1) ~ + -!- K~Y, - -!- K G2n + 1)2 + lJf

K = (n~) (1 - ~ k~) (22)

and n integer.

287
 The homogeneous rf field couples only with even n mode. The intensities
of these modes are given by

(23)

where B(x, y) is the beta function and we normalized the intensities by the
width of the domain wall n-l. In the usual circumstance K - 10- 2 we expect
about 10 satellites associated with n-soliton. As seen from (21), there is
no bound state for the transverse mode. However, in the presence of a field
gradient (this effect is included by replacing wo by wO + Gs in (21)) a
series of bound states do appear in the transverse mode as first shown by
ENGELSBERG et a 1., [23J. The energy spectrum is gi ven by D4J

(Wt) n = Wo + (2n) -1 G Qn (4n >l~/ wOG) + lJ

1 2 ~
+ C.L (Gn/2wO (1 - 4" k.L)) 2 (2n + 1) (24)

where again only even n-mode couples with rf field. This provides the most
convincing interpretation of one of the unusual nmr's in 3He-B observed by
OSHEROFF [24J.
4. Creation of Planar Solitons
So far we have described properties of solitons which are well understood.
We shall now embark on problems of which further clarification is clearly
desirable. It is now well established that planar solitons are created
either by turning off a localized magnetic field or by tipping the local mag-
netization by 180°. In particular a scenario of creation of composite sol-
itons in 3He:'A run~ as follo'1s [25J: 1) by localized magnetic perturbation
a large number of d-soliton-d-antisoliton pairs are created. (If i is assumed
to be fixed, this problem is solved by the inverse scattering method by
ABLOWITZ et al., f26J in the case of a turn-off experiment [27J.) 2) Then
these solitons and antisolitons run away from their location of creation and
stop somewhere due to the spin diffusion within 10- 1 m sec. 3) Then a-sol-
itons transform into composite solitons due to the orbital viscosity in the
time scale of 10 (1 - T/Tc)Y, m sec. More precisely, if the turned off field
is of a rectangular shape,
H(z) = HO e(a 2 - z2)
with e(x) the step function, the total number of created soliton-antisoliton
pairs are given by
=1 2~ 2 a (24)
N - 2 (1 - (>lA/wO)) NO' NO = ; (wO/>lA) ~
where wo = YO HO' In usual circumstances where a - 1 cm, solitons are
created by hundreds when wO > >lA'
The predicted threshold behavior as well as the number of created com-
posite solitons appear to be consistent with recent experiments [28, 29J.
However, details of the scenario may have to be modified. For example, A

KRUSIUS et al., [3 OJ , observed ultrasonically disturbance caused in the i-

field associated with creation of the composite solitons after a magnetic

288
field is suddenly turned off. Indeed, they have observed the threshold be-
havior consist~nt with the above scenario. However, the time required for
appearance of i-disturbances is by a factor of 10Z larger (i.e., - 102 m sec)
than that expected from the scenario. More recently, BOZLER, et al., [30J
recorded free induction decay of magnetization after 180 0 tipping., First
of all they have not seen any signal indicating the appearance of d-solitons
(i.e., unshifted Larmor frequency). Furthermore, before the appearance of
the splay composite soliton signal with Ag ~ 0.7, a large portion of the
system passes through a metastable state with well defined resonance signal
(with A ~ 0.5) after a lapse of time of the order of 10 m sec. Only about
1 secong later a clearly defined composite soliton signal does appear.
These results indicate the need for further work on this subject.
Most of the results reported here were obtained in collaboration with
P. Kumar and Y. R. Lin-Liu. The present work is supported by the National
Science Foundation under grant number DMR76-21032.
References
1. A. J. Leggett, Rev. Mod. Phys. 47, 331 (1975).
2. W. F. Brinkman, H. Smith, D. D.lOsheroff, and E. I. Blount, Phys. Rev.
Lett. 33, 624 (1974).
3. P. G. de Gennes, Phys. Lett. A44, 271 (1973); in Proc. Nobel Symposium
Vol.24 (Academic Press, New York, 1975).
4. V. Ambegaokar, P. G. de Gennes, and D. Rainer, Phys. Rev. A9, 2676
(1974). -
5. G. Toulouse and M. Kleman, Journal de Physique 37, L-149 (1976).
6. G. E. Volovik and V. P. Mineev, Pisma ZETF 24, 605 (1976), and ZETF 72,
2256 (1977). - -
7. M. C. Cross and P. W. Anderson, Proc. of the 14th International Confer-
ence on Low Temperature Physics, Otaniemi, Finland, 1975, edited by
M. Krusius and M. Vuorio (North Holland, Amsterdam, 1975) Vol.l, p.29.
8. N. D. Mermin and T. L. Ho, Phys. Rev. Lett. 35, 594 (1976). -
9. P. W. Anderson and G. Toulouse, Phys. Rev. Lett. 38, 508 {l977}.
10. N. D. Mermin, in uantum Fluids and Solids, editedlby S. B. Trickey,
E. D. Adams, and J. W. Dufty Plenum Press, New York, 1977}.
11. G. E. Volovik and V. P. Mineev, Phys. Rev. B (to be published).
12. K. Maki and P. Kumar, Phys. Rev. B14, 118 (1976).
13. K. Maki and P. Kumar, Phys. Rev. B16, 4805 (1977).
14. K. Maki and Y. R. Lin-Liu, Phys. Rev. B (in press).
15. D. D. Osheroff, W. van Roosbroeck, H. Smith and W. F. Brinkman, Phys.
Rev. Lett. 38, 134 (1977).
16. K. Maki and!'. Kumar, Phys. Rev. Lett. 38, 557 (1977); Phys. Rev. B~,
182 (1977). -
17. K. Maki and T. Tsuneto, J. Low Temp. Phys. 27, 537 (1977).
18. K. Maki, Phys. Rev. Bll, 4262 (1975). -
19. K. Maki and P. Kumar, Phys. Rev. B17, 1088 (1978).
20. o. Avenel, M. E. Bernier, E. J. Varoquaux and C. Vibet,in Proc. of
the 14th International Conference on Low Temperature Physics, Otaniemi,
Finland, 1975, edited by M. Krusius and M. Vuorio. (North Holland,
Amsterdam, 1975) Vol.5, p.429.
21. C. M. Gould and D. M.-Lee, Phys. Rev. Lett. 37, 1223 (1976).
22. M. C. Cross, J. Low Temp. Phys. 21, 525 (1975J.
23. S. Engelsberg, W. P. Brinkman, and D. D. Osheroff, J. Low Temp. Phys.
29, 29 (1977).
24. 0: D. Osheroff, in Quantum Fluids and Solids, edited by S. B. Trickey,

289
 E. D. Adams, and J. W. Dufty (Plenum Press, New York, 1977).
25. K. Maki, in uantum Fluids and Solids edited by S. B. Trickey, E. D.
Adams, and J. W. Dufty Plenum Press, New York, 1977).
26. M. J. Ab1owitz, D. J. Kaup, A. C. Newell, and H. Segur, Phys. Rev.
Lett. 30, 1262 (1973).
27. K. Makr-and P. Kumar, Phys. Rev. B14, 3920 (1976).
28. R. W. Giannetta, C. M. Gould, E. N~Smith, and D. M. Lee, in Quantum
Fluids and Solids edited by S. B. Trickey, E. D. Adams, and J. W. Dufty
(Plenum Press, New York, 1977).
29. J. Kokko, M. A. Paa1anen, R. C. Richardson, and Y. Takano, to be pub-
lished.
30. M. Krusius, D. N. Paulson, and J. C. Wheatley, preprint.
31. H. M. Boz1er, T. Barto1ac, and K. Luey, (in preparation).

290
Creation of Spin Waves in 3HeB

P.W. Kitchenside, R.K. Bullough, and P.J. Caudrey

Department of Mathematics, U.M.l.S.T., P.O. Box 88
Manchester M60 lQD, Great Britain

This article is largely concerned with the following problems: the solution
of the doubZe sine-Gordon equations [1,2,3]

u
xx
- u
tt
= -+ (sin u + l2 A sin !2 u) (1)

for boundary conditions u,u ,u ,etc. + 0, as Ixl + 00, and initial data
u(x,O) = 0; u (x,O) = a, Ixlx<!~ u (x,O) = 0, Ixl > t. Two cas~s are
directly re1e~ant to the spin-waves in the superf1uid phases of He below
2.6 mK: these are the -ve ·sign and A = 1, and the +ve sign and A = O.
The latter is an initial value problem for the sine-Gordon equation and can
be solved by an inverse scattering method [4]. The double sine-Gordon
equations which arise for A # 0 are not soluble by any of the techniques
presently available for solving nonlinear evolution equations [2,3,5]: there
are, for example, apparently only three conservation laws [2] and the systems
are not 'integrable'. Evidently only singular perturbation theory [6] and
numerical integration are available to solve this problem. Despite success
with perturbation theories for the case of the positive sign [6], we do not
yet know how to handle similar perturbation theory for the negative sign:
u - u = - sin u is unstable and its mu1tiso1iton solutions are unstable.
T~~s pa~~r therefore confines its report to the results of numerical work.

A survey of the physics to ~hich these results are relevant, namely to

spin waves in the A-phase of He (sine-Gordon case) and the B-phase (double
sine-Gordon case) must necessarily be brief. We have elsewhere [3,5,7] intro-
duced an 'adiabatic' Hamiltonian density ~ which generalizes that proposed
by LEGGETT [8] to describe the unusual NMR behaviours observed. We derive
from this the equations of motion
IS~ 2 -1 (2)
6 t = 6r = - y B + y X r3
3
IS~ -2 -2 IS~
r 3 ,t -'66 y X c 6
xx - '66
and therefore
-2 2 -1
6 tt - c 6 Y X ISH DIIS6 (3)
xx
in one space dimension x (we confine discussion to this case). The para-
meters are: y = elm c = gyromagnetic ratio, X = static susceptibility, ~
p

291
a velocity ~ vF ; 8 and r are a pair of canonically conjugate variables
such that yr 3 describes th~ magnetization (r 3 is the expectation value of
a spin one operator oW: Ow has the eigenvalues Sw = ! 1,0); B is a
homogeneous external magnet~c field imposed along x. Important 2results are
[7,8J that the dipole interactions Hare H = -3/5 g (T) cos e (A-phase)
a~d HD = + 4/5 gD(T). (cos 8 + cos 2 Y)
(B-pha~e). ~he Pirst case yie:ds the
s~ne-Gordon equat~on ~n u = -2 8; the second case y~elds the double s~ne­
Gordon equation, with negative sign and A = 1, in u = -2 8.

A Hamiltonia~ density for this second case is [7] .<fl-= 1/2 u 2+ 1/2 u 2+ 2
(cos 1/2 u + 1/4) . The dipole interactio~lenergy minimum is atteither ~f the
two roots 6 and 4rr - 6 of u = 2 cos (-1/4). There are two solitary wave
solutions (kinks) which take u from u = 6 to 4rr - 6 ('4rr - 26' kink) and
from 4rr - 6 to 4rr + 6 = 6 (mode 4rr) ('26' kink). These take the forms

u = 2rr + 4 tan- l (13/5 tanh 1/2 8), u = 4 tan -1 (/5/3 tanh 1/2 8) (4)

for the 4rr - 26 and 26 kinks respec2i~v~y: the arguments 8 are 8

K(X - vt) + 8 and K 115/16 (1 - v) . In the rest frame, these kinks
have 'masses,05.l097 and 11.3929 units. One easily realises (and see [7J)
(i) that with boundary conditions u + 6 (mod 4rr) the 4rr - 26 kink must
be followed by the 26 kink, the order cannot be reversed and such kinks
will bump in collision; (ii) with the same boundary conditions a 4rr - 26
kink-antikink pair can convert to a -26 antikink-kink pair if the colliding
4rr 28. pair has enough kinetic energy to produce the 26-pair rest mass. The
threshold for this is v = 0.8938. Below threshold the 4rr - 26 kink-ant i-
kink pair bumps [3,5,7J. The behaviour just above threshold is particularly
interesting [3,5,7J because of the loss of energy by radiation.

We turn to the creation of spin waves and their behaviours. We suppose

t > °
that an additional inhomogeneous field 6B O(x) is removed at t = 0. For
the field B is a constant field BO' It is easy to see from (2)
that there is a jump induced in r 3 , 6r 3 , and a motion of e which satisfy

(5)
2 -2
6r = - 6 HD/6 8 + Y Xc 8
3,t xx

with 8 = 80 (a minimum of HD), 8 t = -y 6B O(x) , 6r 3 = 0, 6r = °

t = °
and 8 + 80, tk-l+ 0, etc. , as Ixl + for all t » 0.3,t This
= _y2 X 8 ~/6 8 + c 2 8 for t >
00

, with 8 =
equivalent to 8
xx
-y 6B O(x) at t £;to, and 8 + 80' 8x + 0, etc. , as 1;1 + ° 00.

The A-phase (sine-Gordon) problem has a homogeneous case (recall u - -28;

Q~A is the longitudinal NMR frequency [7,8J)
2
-Q~A sin U; u = 0, u t 2y 6B O at t = 0. (6)

However, for u + 0, U + etc., as Ixl +° for all t > 0, the solution

00

does not remain homogen~ous. It does so for the open-ended boundary condi-
tio~lux = 0,
cos cn(wt,k), w = y 6B O and °
Ixl = ~ for all t ~ 0. The solutions in ~?is case ~le u =
< k < 1; u = rr + 2 cos dn(wt,k ),

292
 -1 -1
w = Q~A and k > 1; -1 =Qt Y ~BO is the modulus of It he Jacobian elliptic
funct~on cn and k of !n. When k = I, u = 2 cos sech wt. The mag-
netization therefore rings with the frequ~ycies of the Jacobian elliptic -1
functions. These frequencies are rry~BoK for y~BO > QtA and rrQtA(2K)
for y~BO < Q in which K(k) is the complete ell~ptic ~ntegral or the
first kind [K(It = 00, K(O) = rr/2l. For small k, linearized theory, the
longitudinal NMR has a steady oscillation at the frequency QtA which is
independent of BO and ~BO' This is in agreement with the unusual trans-
verse and longitudinal NMR behaviour observed in the A-phase [8,9J. (The
transverse theory is not contained in the present analysis.)
-1
Notice that when k = 1 the trajectory u = 2 cos sech QtAt , u t = 2Q tA
sech QtA t, does not ring. This defines a critical field for no ringing
y ~BOc = ~tA' WHEATLEY [9J has measured ~BOc follow~ng an. analysis similar
to tliat g~ven here: agreement between theory and exper~ment ~s not perfect.
However, the assumption of homogeneity and the boundary condition u = 0,
Ixl = t were not discussed. Physical boundary conditions certainl~ involve
the effect" of ~1 vector textures [10J consistent with the condition ~t
normal to the end surfaces. This problem is too hard to consider here. To
determine one possible effect of boundary conditions we impose the fixed ends
condition u = 0 at Ixl = t.

The sine-Gordon equation has been solved in a form readily applicable at

present only for solutions u(x,t) defined on the reat tine -00 < x < 00
We therefore approximate the problem of fixed ends by

KAUP [4J has used the inverse scattering method partially to solve this
problem. The essential point is that for T < 1 only stationary breathers
are created. Each such breather is of the form

u = 4 tan- l {cot~ sin(t sin~)sech(x cos~ )}. (7)

The number of such breathers is the integral part of (tT/rr) + V2. The value
T = 1 where y~BO= y~BOc = QtA remai~s a cri:ical.fiel~. For T > 1
kink-antikink pa~rs beg~n to travel out ~n oppos~~e d~rect~ons. In terms of
the scattering data, pairs of eigenvalues S, - s move successively onto
the imaginary axis and split-up there. Figure I, shows four such radiating
kink-antikink pairs; two standing wave breathers remain in the central region
Ixl ~< to

Corresponding results for the B-phase double sine-Gordon case are more
complicated. The homogeneous problem has phase plane trajectories
1 2
-u
2 t
(cos u + cos 21 u) + C. (8)

There are two ringing free trajectories namely when the constant C = 0 ~d
when C = 2. The critical fields are y~BOCl /3/5 QtB and y~BOC2= /5/3
Q . Further details are given in [7J. Preliminary experiments designed to
ofi~erve the two critical fields are reported by WHEATLEY [9J.

293
 Fig.l Four outgoing kink-antikink
pairs of the sine-Gordon equation
generated by initial data u (x,O)=
2.6,lxl<15. Two standing wa~e
breathers remain in Ixl<15. Note
the initial homogeneous region

The arguments surrounding the boundary conditions for the A-Phase problem
apply in similar form to the B-phase problem; but the physical problem of
textures is certainly different [10J. The initial value problem (1) (with
negative sign and A = 1) is not soluble by any known scattering method.
Nevertheless, for fields ~BO < ~BO we can expect that standing wave
breathers may be formed. The Fig. 21 for u = 1 (less than the critical
values u = 3/2) is evidence supporting this expectation.
t

Fig.2 Breather-like solutions of

the double sine-Gordon equation
generated by initial data u (x,O)=
1, Ix I<15 t

For boundary conditions u + 0, Ixl + 00, the 4n - 20 kink cannot be

followed by a second such kink. The Fig. 3 for u = 3/2 shows that two
kink-antikink pairs radiate out in opposite directions. The Fig. 3 starts
at t = 40 in time units and is followed for the longer time shown. The
pairs continue to emit radiation but hold together well. This result
~uggests that for u = 3/2 at a point x > £ in a specimen of B-phase
He, it might be pos~ible to detect magnetically an emitted kink-antikink
pair: in contrast in the A-phase a single 2n-kink would be observed for the
critical value u 2.
t

294
 The Fig. 4 shows the result of crossing the second critical threshold
at u = 5/2. The boundary condition is u = 6, Ixl ~ 00, for the right;
6 ~ 4*- 6 ~ 4n + 6 ~ 8n- 6 ~ 4n+ 6 ~ 8n- 6. This sequence involves three
4n - 26 kinks, one (4n - 26) antikink and one 26 kink! The conclusion
from these figures is that below threshold at u = 3/2 in the B-phase the
homogeneous problem is unstable to break up intotbreathers; at u = 3/2 it
is unstable to emission of two 4n - 26 kink-antikink pairs; at t Ut = 5/2 it
is unstable to further emission involving one 26 kink and one 26 anti-
kink.

Fig.3 Two 4n-26 kink antikink

pairs of the double sine-Gordon
equation generated by the criti-
cal initial data u (x,O)=3/2,
Ix l<15. Time start~ at t=40 in
this figure

Fig.4 The combination of three

4n-26 kinks, one (4n-26) anti-
kink and one 26 kink of the
double sine-Gordon equation
travelling to the right, and the
corresponding combination of
antikinks and kinks travelling
to the left, generated by initial
data u (x,O)=2.6. This is above
the critical value u =2.5 for
creating 26 kinks t

A conclusion important to the physics is that the critical fields for

no ringing NMR remain critical for break up into solitons both in the A-phase
and in the B-phase. However, if break up into solitons occurs, as it surely
must for most physically realizable boundary conditions, such break up would
be accompanied by an apparent ringing signal at least at specific points
x > ~ outside the original region of inhomogeneous magnetic field. It is

295
not easy to see that any averaged signal will necessarily be ringing free and
this may have some bearing on the imperfect agreement between theory and
experiment observed for the A-phase [9].

It seems possible to detect magnetically the different kinks emitted in

the A- and B-phases and so obtain further evidence on the correctness of the
symmetries assigned to the order parameters in the two different phases.

References

1. S. Duckworth, R.K. Bullough, P.J. Caudrey, J.D. Gibbon: Phys. Lett. 57A,
19 (1976)
2. R.K. Dodd, R.K. Bullough: Proc. Roy. Soc. London A 351, 499 (1976); 352
481 (1977) -
3. R.K. Bullough, P.J. Caudrey: 3The Multiple sine-Gordon Equations in Non-
linear Optics and in Liquid He" In Proceedings of the Syrrrposiwn on
Nonlinear Evolution Equations Solvable by the Inverse Spectral Transform
(Academia dei Lincei, Rome, June 1977) ed. by F. Calogero (in press)
4. D.J. Kaup: "Studies in App!. Maths.", LlV, 165 (1975) (Massachusetts
Inst. of Technology) --
5. R.K. Bullough, R.K. Dodd: "Solitons I. Basic Concepts II. Mathematical
Structure" In Synergetics A Workshop, ed. by H. Haken (Springer-Verlag,
Berlin, Heidelberg, New York, 1977) pp.92-ll9; R.K. Bullough: Solitons
in Physics (Lectures given at NATO Advanced Study Institute on Nonlinear
Equations in Physics and Mathematics, Istanbul, August 1977) ed. by
A.O. Barut (in press)
6. P.W. Kitchenside, A.L. Mason, R.K. Bullough, P.J. Caudrey: "Perturbation
Theory for the Double sine-Gordon Equation", these Proceedings; A.C.
Newell: "Perturbed Soliton Systems", these Proceedings
7. R.K. Bullough, P.J. Cau1rey, P.W. Kitchenside: "Bumping Spin Waves in
the B-phase of Liquid He", to be published in J. Phys. C: Solid State
Physics (1978)
8. A.J. Leggett: Rev. Mod. Phys. 47, 331 (1975)
9. J. Wheatley: Rev. Mod. 3hys. 47, 415 (1975)
10. K. Maki: "Textures in He", these Proceedings

296
The Interaction of Spin Waves in Liquid He 3
in Several Dimensions

John Gibbon
Department of Mathematics*, U.M.I.S.T., P.O. Box 88
Manchester M60 lQD, Great Britain

Nonlinear Klein-Gordon equations in two or more spatial dimensions of the type

Oq, = F(q,) (1 a)
32 (3 2 32 32 ) (1 b )
D=W- l~+~+w
persistently occur in theoretical physics as multi-dimensional models of wave
propagation in nonlinear media. One important area where equations of this
type arise and where multidimensionality arises naturally is in studies of the
interaction of spin waves in liquid He 3 [1,2,3]. In the A-phase of He 3 , spin
waves of magnetization are governed by the sine-Gordon (SG) equation
Oq, = -Q2Asinq, (2)
The constant Q is an NMR frequency and the variable q, is an angle which
defines the spi~ orientation [2]. In this case the metric is really (1,-1,
-2,-1) for the [J operator but it is easier to rescale the y variable so
that we shall always consider Cl to be in the form given in (lb). Equation
(2) also occurs in flux propagation across a Josephson junction [4].
The double sine-Gordon (DSG) equation arises in the B-phase of He 3
(3 )

Note the change of sign on the r.h.s. of (3) in comparison to (2). For the
detailed physics of these equations we refer the reader to refs. [1-3]. A
further equation of great interest is the q,4 equation
(4)
which is of interest in particle physics [5] and studies in lattice dynamics
[ 6].
In 2 dimensions (1 space; 1 time), of all these equations only the SG
equation shows perfect soliton behaviour i.e. no ripples are emitted when the
solitons collide. The DSG with the opposite sign to (3) governs the propaga-
tion of pulses in degenerate SIT in optics [2] and wobb1ing pulses occur.
However for both the bumping spin waves in the He 3 B-phase governed by (3)
and for the hyperbolic tangent "kink" solutions of (4) ripples are emitted
when one wave collides with another [2,3]. When more spatial dimensions are
added as in (1) the situation is no better as far as soliton preservation is
concerned. The SG equation has multiple soliton "wavefront" behaviour [7]
but special conditions occur on the motion which force the wavefronts to move
in triangles whose areas must remain constant in time. Wavefront type solu-

*Address after 1st Oc. 1978: Department of Mathematical Physics, University

College, Belfield, Dublin 4, Eire.

297
tions are useless in particle physics but are obviously more useful in solid
state physics when one is studying the interaction at differing angles of
lines of spins. One of the few equations which have truly four dimensional
(Euclidean), analytic, non singular solutions which go to zero everywhere at
1nfinity, 1S the ~~ equation
tJ~ = C~3 (5)
which arises in self-dual Yang-Mills field theories. A solution of (5) is

(6)

This solution has become known as the Po1yakov instanton [5].

In order to salvage something from the difficulties which arises in any
attempt to find multiple wave solutions of equations such as (1); it is
easier to consider solutions of (1) as depending on a single variable
g=g(x,y,z,t) [8,9]. Equation (1) is now
(CJ9)~g + (~g)2~gg = F(~) (7)

where 8 plays the role of the gradient operator with an appropriate metric
such that ~.~ = LJ. There is a result in classical field theory which says
that if Og/(~g)2= f(g) where f is an arbitrary function then the "sur-
faces" 9 are equi potenti a1 surfaces. It therefore seems appropri ate to
take
CJ9 = A(g) (~g)2 = B(g) (8)
As it stands (7) is not integrable but if we take A = ~dB/dg. then (7)
becomes
(9)

The function V plays the role of a potential since CV=O. We can pick
anything for B except (8) may not have any solutions. However it is appro-
priate to take B(g) = _g2n2 so nV = log 9 and
og = -nga (10)
where n is an appropriate constant
One set of solutions of (10) are
N
9 = L exp 8. (11 a)
i =1 1

8i = Pix + qi Y + riz - wit + 0i p.2 + q.2 + r.2 - w. 2 = n2 (11 b)

1 1 1 1
With NC conditions on the motion
2
(p. - p.)2 + (q. - q.)2 + (r. - r.)2 - (w. - w,)2 = 0 (11 c)
1 J 1 J 1 J 1 J

The set of equations (llb,c) is overdetermined for N>7. As long as (9) can
be integrated then solutions can be found trivially. For example the SG
kink-type solution is
(12 )

298
which takes ~ from 0 to 2TI.
It is worth making a comparison at this point with the full multiple soli-
ton solution of the SG equation. Lack of space precludes showing the full
formulae but a graph of sin~~ against x and y at t=O for three soli-
tons is shown in figure 1. Details are given in (7). The middle and outer
wedge shape areas of the graph are areas in which ~=O (spin down) and the
other areas are ~=2TI (spin up). The figure therefore shows areas of differ-
ing spins propagating in the (x,y) plane the main result being that the
central triangle area os spin propagates without change of form. It was
shown in (7) that the area of the triangle formed by the three kinks must
remain constant in time. The connection between the full soliton solutions
and those given in (11,12) is simple. The conditions on the motion (llc) are
exactly those which force all the soliton phase shifts to have infinite value.
Miles investigated such solutions for the 2 dimensional KdV equation and
called them resonant solutions. It seems therefore that the DSG and ~4 equa-
tions have mult,ple resonant soliton solutions although not full ones. These
solutions are essentially very simple and can be thought of as representing a
configuration of localised travelling wave solutions. Equation (llc) has a
simple meaning. The dispersion relation p.2 + q.2 + r.2 - w. 2 = n2 expresses
the mass' energy relationship for a set of soliton~ of mAss n~ The condition
(llc) shows that two solitons of momentum (w.,p.,q.) yield a third of zero
mass. This is the classical analogue of Compton' scAttering.

Plot of sin~~ against

x (continuous) and y
(discrete) at t=O for three
full solitons (non resonant)
for the sine-Gordon equation.

There are two types for the DSG equation (2), which are associated with the
zeroes of the r.h.s. of (3) which occur at 0 = 2cos- 1(-1). There are no
real solutions associated with the zeroes which occur at 2NTI. One pulse,
called the 20 pulse takes ~ from -0 to =0:
~=4tan-1[~[f++)] n=n B (13)
A second solution is the 4TI-2 0 pulse which takes ~ from 0 to 4TI-0:
(14 )
Each of these pulses (12,13,14) are spin waves of magnetization where ~
switches from one value or level to another. Figure 2 represents the 20
299
pulse in the (x,y) plane at t=O for three waves. The configuration is
moving down the figure, the upper level representing an area of one spin
orientation and the lower level another.

Fi g 2
A 26 pulse solution
of the DSG equation (t=O)
for two waves (>lR=l).
Upper area: ~=6 anC lower
area ~=-6

One further set of solutions in 2 dimensions for the SG equation are of

interest. If we take g = >lA2 and (~g)2 = g>lA 2 in (7) then a solution
is
g = !(x 2 - t 2)>lA2 (15a)
where
(15b)
gg + ~ g = s i ncp
g~

The solution for g is equivalent to a similarity solution. The o.d.e. in

(15a) can be transformed into the third Painleve transcendent which has no
essential singularities or branch points. Numerical computation [10] shows
that ~ now moves between 0 and TI and has the characteristic of being
an oscillating function.

References
1. A.J. Leggett: Rev. Mod. Phys. 47, 331, 1975
2. R.K. Bullough and P.J. Caudrey: Proceedings of the Symposium "Nonlinear
Evolution equations solvable by the Inverse Spectral Transform" Pitman
Research Notes 1978
3. R.K. Bullough and P.J. Caudrey: "Optical Solitons and their spin wave
analogues in He 310 Proc. of 4th Rochester Conference on Quantum Optics: ed.
L. Mandel and E. Wolf (Plenum N.Y.) 1978
4. A. Barone, F. Esposito, C. Magee, A. Scott: Riv. Nuovo Cim., 1,227, 1971
5. R. Jackiw: Rev. Mod. Phys. 49, 681, 1977
6. T. Schneider and E. Stoll: Phys. Rev. Letts, 35, 296, 1975; J.A. Krumhans1
and J.R. Schrieffer, Phys. Rev. Bll, 3535, 19~
7. J.D. Gibbon and G. Zambotti: Nuovo Cimento, 28B, 1, 1975
8. J.D. Gibbon, N.C. Freeman and R.S. Johnson: Phys. Letts. 65A, 380, 1978
9. J.D. Gibbon, N.C. Freeman and A. Davey, J.Phys.A, 11, L93, 1978
10. G.L. Lamb: Rev. Mod. Phys. 43, 99, 1971
300
Josephson 'fiansmission Line Oscillators

Alwyn C. Scott
Department of Electrical and Computer Engineering, University of Wisconsin
Madison, WI 53706, USA

1. Introduction
In this paper, the term "Josephson transmission 1ine" (or JTL) will refer to
a wave structure composed of two superconductingometal strips separated by
an insulating barrier that is thin enough (~25A) to permit transverse
(Josephson type) tunneling of electrons in the superconducting state [1].
JTl's as long as 35 cm have been fabricated by a process employing rf sput-
tering of the metal layers and photoresist definition of all patterns [2,3].
A diagram of the basic structure to be considered is shown in Fig. 1 where a
TEM wave is assumed to propagate in the x-direction. The dependent variables
of this wave can be viewed as the transverse voltage v(x,t) across the in-
sulating barrier and the logitudinal current i(x,tl flowing parallel to the
insulating barrier in the x-direction.

DISTRIBUTED BIAS CURRENT (Y)

~ Sketch of the Josephson transmission line (JTL) considered in this

paper

If dissipative effects are neglected, the magnetic flux of the wave

301
 <I> =: Jv dt
is governed by the partial differential equation
--2 )-2.
<I> xx - (c 1 <I> t t = lA J s 1 n(211 <I> 1<!> 0 J (1 )

where c is a characteristic (or wavefront) velocity for the TEM wave, A

is a "Josephson penetration length" for magnetic fields into the insulatin~
barrier, and

~ 2.0678 x 10- 15 volt-seconds

is the quantum unit of magnetic flux. Typical values for care 1/15 to
1/20 the velocity of 1ight I3,4J and AJ ranges from .01 to 1 mm [3J.
If x is measured in units of AJ , time in units of AJ/c and magnetic
flux (<1» in units of <1>/211, 0 l becomes

~xx - ~tt - sin~ =a (2)

which is readily recognized as a normalized form of the sine-Gordon equation

[5J. The "kink" solutions of this equation

x-ut-x )]
~±lx,t;u,xo) = 4 tan- l [ exp ( ± ~o (3)

represent either a fluxon l~+J traveling in the + x direction or an anti-

fluxon (~) traveling in the + x direction (see Fig. 1). If the trans-
verse voltage is measured in units of oC/211A J , then <l>

~t = v•

Also if the longitudinal current is measured in units of joAJ I lwhere jo

is the maximum Josephson current per unit length in the x-direction), then

Although several interesting applications have been proposed to employ

fluxons in information processing systems [6J, attention here will be focused
upon the fluxon as a generator of electromagnetic radiation. To this end we
will be Interested In analyzing not the "perfect" sine-Gordon equation given
in (2} but structural perturbations of it that permit input and dissipation
of energy. Energy input to a fluxon is readily obtained through the distri-
buted bias current (y) indicated in Fig. 1. The effect of y is to exert
a Lorentz force which accelerates a fluxon in the + x direction and an anti-
fluxon in the - x direction. Energy dissipation will be represented by add-
ing phenomenological loss terms of the form rl~tl~t or a~t to the right
hand side of (2). Thus we shall investigate either
~xx - ~tt - sin ~ = rl~tl~t + y (4a)

302
or

~xx - ~tt - sin~ = a~t + y l4b)

depending upon which is more convenient for the problem at hand.

2. Oscillations on a Finite JTL

In this section we assume our JTL to be of a length L that is long compared
with its Josephson penetration length. AJ • but "finite" so open circuit
boundary conditions at the ends influence the oscillations. Thus in the ab-
sence of dissipative effects we seek solutions of (2) together with the
boundary conditions: ~ (O.t) = ~ (L.t) = O. The qualitative features of
such solutions have bee~ describe~ in a recent survey by Fulton [7]. Anal-
ytic expressions can be obtained from the ansatz suggested by Lamb [8].
~(x.t) = 4 tan- 1[h(x)g(t)] where 2 h a2d ~e~ in general. Jacobian 4e1-
1i~tic functions defined by (h') = ah + (l+b)h - c and (gl)2 = cg +
bg - a with a. band c arbitrary constants. Constabile et a1. [9] have
displayed solutions for the three principal types of oscillatory behavior.
Plasma oscf1lation. This is an oscillation about ~ = 0 described by
~ = 4 tan- l [A cnCBxikf) cn(nt;k g )] (5)

where

and n. B and A are related by the nonlinear dispersion equation

(6)

Imposition of the open circuit boundary condition at the ends of the JTL
fixes the spatial periodicity as

where N = 1.2 •••. is the number of nodes in the standing wave and K(k f )
is the complete elliptic integral of the first kind.
Bion oscillations. These correspond to bound state oscillations of f1ux-
on-antifluxon pairs (they might also be called "breather" oscillations).
They are described by

~ = 4 tan -1 {A dn[B(x-xo);k f ] sn(ntik g )} (8)

where

303
and the nonlinear dispersion equation is
S = r:lA. 19)

The open circuit boundary conditions now require

R..=!!.K(k)
~N L f (1 a)

with two possible values for x: a) SXa = KCk f ) and N even (this cor-
responds to bions located near ~he center of the JTL) and b) xa = a (this
corresponds to fluxons bound to virtual antifluxons at both ends of the JTL).
Fluxon oscillations. Such an oscillation can be qualitatively described
as a fluxon that moves to one end of the JTL, is reflected as an antifluxon
that moves to the other end, is reflected ••. etc. It differs from bion and
plasma oscillations in that at each point (x) of the JTL, ~ increases
without bound as t increases. Thus a fluxon oscillation can be driven by
a distributed bias current (y) and may serve as a useful ac generator.
The generator frequency will be equal to that of the Josephson current,
10 sin(2n~/~0). Thus if the dc component of the voltage across the junction
is V, the corresponding frequency is
f = V/~a t 11 )

where ~ -1.1S equa 1 to 483.59 megacyc 1es per microvolt. Fluxon oscilla-
tions arg described by

(12)

The nonlinear dispersion equation is as in (9), and the open circuit bound-
ary conditions again require (la). A monotone increasing behavior of ~
with t is obtained from (12) by noting that ~ ~ 2n as tn ~ 00 and
r:lt ~ t2m-l) K(k) where m = a,±l, ••• ; thus we can switch branches of the
arctan functiongat these instants.
Since fluxon oscillations exhibit a nonzero average value of ~t' a cor-
responding finite voltage should be measured. The resulting volt-ampere
characteristic can be calculated by introducing bias current and dissipation
as indicated in (4a). Assuming that rand yare small enough so ~(x,t)
can be approximated by (12), the total input power is the time and space av-
erage of y~t while the power dissipation is the time and space average of
rl~tl~~. Equating these two power terms yields the relation

304
 y ( 1 3a)

where E(k f ) is the complete elliptic integral of the second kind, and the
time and space average of the normalized voltage is

(l3b)

From (13) a plot is readily obtained for y vs. <W t >, the volt-ampere
characteristic associated with a fluxon oscillation. This relation is plot-
ted in Fig. 2a for an increasing number (N) of fluxons involved in the oscil-

N=
0.5

r O.4

0.3

>-0.2
La)
0.1

°o~==~~====~~~~
~2
__~~~~
~3 ~4
__-J
<~I)/ 2fT - - - -...

--. ...
--+ I I I
to. 12m. ,,-2
6

.., ..,
--+ T • 4.Z K
H·O
5
-+ -+

4 - ,,-,0
ct
e -+

3
",'2
-+
Z ,.
n-'4
(b)

- -
_)-1- J
L I
so 100 150

v (/kV I

~ (a) Plot of current (y) vs. average voltage from (13). (b) Current-
voltage measurements on a lead-oxide-lead Josephson oscillator with dimen-
sions 1.5 mm x 0.2 mm by Chen and Langenberg [10]

305
lation. For comparison we reproduce, in Fig. 2b, some measurements on a
Pb-oxide-Pb rectangular Josephson junction oscillator which have been pub-
lished by Chen and Langenberg [10]. While the qualitative similarity between
Figs. 2a and 2b is encouraging, several reservations and explanatory com-
ments should be made.
a) A major qualitative difference is that the curves of Fig. 2a approach
infinity while those of Fig. 2b terminate at some value which is less than
the maximum Josephson current. This discrepancy is an artifact of the power
balance calculation leading to (13). It was assumed that the structure of
the nonlinear mode, i.e. <p(x,t), remains equal to that given in (12) as y
becomes large. Study of an exact solution of (4a) under periodic boundary
conditions by Parmentier and Costabile [11] shows that this is not the case.
Their calculations yield b~anches that terminate at y = 1 and display some-
what sharper "elbows".
b) The JTl structure studied in Fig. 2b has dimensions of approxi~ate1y
1.5 mlD-X 0.2 mm. Thus the branch 1abel ed "m = 2" corresponds to the' owest
transverse oscillation.
c) The branch labeling in Fig. 2b indicates integer multiples of the
lowest resonant frequencies of the rectangular Josephson junction cavity
which occur at frequencies fn = nc/2l. In Fig. 2a, on the other hand, the
tndex N indicates the number of f1uxons engaged in the nonlinear oscilla-
tion. For one f1uxon (N = 1) a period is the time required for a f1uxon to
bounce from one end of the junction to the other at velocity e. Thus the
limiting frequencies to be expected are fN = Nell. The fact that no branches
are found for odd values of n in Fig. 2b tends to confirm the theoretical
picture presented here.
d) Measurements of power output from the.osci11ator of Fig. 2b are much
less than that indicated by the power balance calculation. Along the n = T1
branch, for example, the maximum power 9utput was found to be only 2 x 10-
watts over a very narrow range (2 x 10- volts) of junction voltage. Out-
side of this narrow range the power output drop~ed by two orders of magnitude
and the volt-ampere characteristic displayed a fine structure" with hyster-
esis between adjacent branches. The power balance calculation, on the other
hand, indicates that power output should rise as a monotone function of vol-
tage and reach a broad maximum along the vertical segment of the character-
istic. It is likely that this discrepancy appears because the "parasitic"
effects of plasma and bion modes have not been taken into account. To pro-
ceed analytically in this direction will require the more general multiply
periodic solutions of the sine-Gordon recently given by Matveev [12].

3. Radiation from a "Wobbling" Fluxon

The oscillator structure to be considered in this section is shown in Fig. 3.
Here a JTl of the sort indicated in Fig. 1 is formed as a ring and periodic-
ally loaded with localized regions of large Josephson current (the "micro-
shorts") that are separated by an equal spacnng a. Means for supplying a
distributed current bias (y) are included so a 1F1uxon can be forced around
the circumference of the ring. As the f1uxon passes a microshort, it will
exchange kinetic energy for locally stored magnetic energy and slow down.
On the average, the f1uxon will execute a "wobb1 ing" motion at a frequency
Wo = 21Tum/a (14)

306
where um is the mean circumferential velocity of the fluxon. Radiation
from the fluxon can be expected at the appropriate Doppler shift of frequency
Wo

l~diation
-- •--...

y---

-y
CPo
~ A "wobbl ing" fluxon oscillator structure

A convenient analytic tool to study this problem is a soliton perturbation

theory in the form recently suggested by Keener and McLaughlin [13]. Only a
brief r€sum€ of this approach will be presented here; the reader is referred
to [14] for full analytical details. We begin the analysis by writing a
structurally perturbed version of the sine-Gordon equation (2) in the form

(15)

where the structural perturbation

f = -cp t - 1.ex ex L
- H. 0(x - na) sin cp (16)
n

The first two terms in (16) represent dissipation and current bias as in
(4b); the last (summation) term takes account of the increased Josephson
current localized at the microshorts. Then a series expansion in the small
parameter ex takes the form
W " Wo + exWl + ...

307
where 'W::: col (¢'¢t)' Wo:: col(¢ ,¢ t)!. and ¢ is a solution of the
perfect sine-Gordon equation (2). ~lsg W l ::: col(¥l'¢lt) is a first order
correction governed by the linear equation

LW1 = f (17)

where f::: col(O,f) and the operator

.L'::: (:a- -:!c~s-~- -~- ~

xx 't'o' t
-). (18)

If it is assumed that ¢(x,t) is an N-soliton wave at t 0, 07) can be

so1ved as
t co
Wl = J J G(x,tlx',t') f(x',t')dx'dt' (19)
a -co
where G(x,tlx',t') is a Green's function that can be explicitly constructed
[14]. Secular terms in (19) arise from resonances between soliton contribu-
tions to the source (1) and their corresponding contributions to the
Green's function G. These can be avoided by allowing the positions and
speeds of the f1uxons to modulate slowly with time. To see this, suppose ¢o
is a single f1uxon

¢o = 4 tan- 1 [ex p (x-X )]

~
t
where X::: f u(t')dt' + xo(t) is the f1uxon's position. Then the source
a
(at order a) becomes augmented to

(20)

. .
If u and Xo are chosen so that F is orthogonal to the rows of G, secu.
1arities are avoided and we have the o.d.e.'s

308
 11 2 2 (x-na) (x-na ) (21 )
+ "2 (l-u.) ~ sech A_u 2 tanh A_u 2 •

•
Xo = 11 u ~ (X-na) sech 2(~) tanh
-"2 D (~)
D

Integrating these to obtain the periodic motion u(t) = u(t + 2n/w 1 then
determines the wobble frequency and the mean fluxon velocity, urn. 0 At
this point, nonradiated dc power input to the fluxon can be calculated as
the product of the Lorentz force (y) on the fluxon times its mean velocity.
Thus

Pnr = YUm • (22)

Also the augmented source in (20) can be written as a Fourier series in Wo

+ iw t'
F(x',t') = FO(X'-umt') + Fl(X'-umt')e 0

(23)

+ higher harmonics

and (19) can be rewritten in the form

t 00

141 =} 1 Gr(x,tlx',t') nx',t')dx'dt' (24)

o -00

where G is the "radiative" component of the Green's function which was not
made orthogonal to F. Thus (24) computes the first order radiation from--a--
fluxon as it moves under the influence of (21). The radiative part of the
Greens function takes the form*

G(x,tlx',t') = 1.
-4 J
00
d>'
[911
m _00 g21

where the gij g .. lx-utlx'-ut), I;; = i[(1+u)/(l-u)]1/2, k(>'):= 2A - 1/8>.,

lJ

* Explicit expressions for the gij are given in [14].

309
and weAl = 2A + 1/8A. Thus w~ = k~ + 1 and dA/A = dw/k = dk/w.
The "dc" source term from (23) causes a first order correction to the
shape of ¢o [15] which can be neglected when ¢o is used for other first
order calculations. The steady state response to the fundamental harmonic
in F can be defined as

Wlss(x,t) = __1__. lim ,00 dA ,t dt" dx ' exp{-i[kCA)Lx-xl)+w(A)lt-t ' )]

4nl t~ -00 -t ACS 2_A2)2
iw t I --*
. t'
-lW O
[ g. . ] [-F1 ('
x -u t
I )
e
0 + Fl (x I -umt I ) e .]
lJ m

Integration over x' introduces a factor exp{iku t'l so the t' integral
involves only m

t
lim j exp{i[w(A)+ktA)um±wo]tl}dt' 2:rro(w+ku m±w o).
t ... 00 -t

This drives radiation at the Doppler shifted wobble frequencies -(ku ± wp)'
For oscillator applications the higher fre2uency (w+) is of primaryminte -
est. Here w+ = -CwO + k+u) m and k+2 = w+ - 1 so

1 +
and (26)

Note that as u ... 1, w ... -w /(l-u) so a substantial increase in the

radiation frequ~ncy is obtainedOthrou~h the Doppler shift. Changing the
variable of integration from A to w then yields

exp{-i[k+x + w+t + ; ]}
128
l+u m + tk + w )2~ 2
k __
+ l-u m + +

(27)

+ cc.

310
Away from the fluxon this radiation is governed by the linear Klein-Gordon
equation wxx - Wtt = w for which the corresponding energy (Hamiltonian)
density is H = ~(w2 + w~ + i~. U~in~ the dispersi~n equation (w 2 = k2 + 1)
and the plane wave r~lation: w = k wand wt = w+w2, this can be
written H = wi = w;w2. This e~ergy+propagates at the group velocity dw/dk
-k+/w+. Thus the radiated power is

2
w • (28)

The radiated power will be absorbed by the losses (a) of the JTL and by
a useful load. Suppose a is composed of two components: i} a' represents
internal losses and ii) a" represents power absorbed by a shunt load. Then
a = a' + a" and the power abosrbed by the load would be
a"
P =--,P (29)
Oar

The above discussion has been presented entirely in normalized units. In

laboratory units, the limiting velocity of a fluxon (e) is about 1/15 of the
velocity of light or 2 x 10 7 meters/second. Settin~ a = 1 means that the
microshorts are separated by a distance equal to the Josephson length"
AJ ~ 5 x 10-4/;,r- microns where J is the Josephson current density meas-
c 2 c
ured in amperes/meter, AJ ~ 5 microns and the wobble frequency fo = Um/AJ
would be 4 x 1012(u Ie) cps. As previously noted, the corresponding Doppler
shifted radiation fr~quency would then approach

as um + c, the relativistic limit. For um/e = 1/2, this corresponds to

a free space wave length of .075 mm.

References
1. A. C. Scott, Nuovo Cimento B~, 241(1970).
2. L. S. Hoel, et al., Solid-State Electron. 15, 1167(1972); and
P. Rissman and 'f':"" Palhomen, ibid • .!l, 611 (f974).
3. A. C. Scott, et ~., J. Appl. Phys. 47, 3272 (1976).
4. H. T. Yuan dnd A. C. Scott, Solid-State Electron. ~, 1149 (1966).
5. A. Barone, !l~., Rivista del Nuovo Cimento 1, 227 (1971).
6. T. A. Fulton, et al., Proc. IEEE 61, 28 (1973); K. K. Likharev, IEEE
Transactions oniMagnetics MAG-13, 245 (1977); P. Gu~ret, ibid. MAG-ll,
751 (1975); K. Nakajima et al., J. Appl. Phys. 47, 1620 (1976).--
311
 7. T. A. Fulton, in Superconductor AtPlications: S~IDi and Machines,
B. B. Schwartz and S. Foner eds. Plenum, New Yor, 97"iT p. 125.
8. G. L. Lamb, Jr., Rev. Mod. Phys. 43, 99 (1971).

9. G. Costabile, et ~., App1. Phys. Lett. B., 587 (1978).

10. J. T. Chen and D. N. Langenberg, in Low Temperature PhfsiCS, K. D.
Timmerhaus, et ~. eds. vol. 3 (Plen~ New York, 1974 p. 289.
11. R. D. Parmentier and G. Costabile, Rocky Mountain J. Math. 8, no. 1
(in Press); R. D. Parmentier, in Solitons in Action, K. E: Lonngren
and A. C. Scott, eds. (Academic Press, New Yor""k,i978).
12. v. B. Matveev, Preprint No. 373, Inst. of Theoret. Physics, Univ. of
Warsaw, 1976 (unpublished).
13. J. P. Keener and D. W. McLaughlin, Phys. Rev. A ~, 777 (1977);
J. Math. Phys. ~, 922 (1977).
14. D. W. McLaughlin and A. C. Scott, Appl. Phys. Lett. 30, 545 (1977);
see also Solitons in Action, K. E. Lonngren and A. C:-Scott, eds.
(Academi c Press, New York, 1978).
15. M. B. Fogel, S. E. Trullinger, A. R•. Bishop and J. Ao Krumhansl, Phys.
Rev. Lett. l§., 1411 (1976); Phys. Rev. B.:!i, 1578 (1977).

312
Dissipaove Structures in Quasi-One-Dimensional
Superconductors

A. Baratoff
IBM Zurich Research Laboratory, CH-8803 RUschlikon, Switzerland

1. Introduction

Consider a superconducting filament of length d» ~(T), the coherence

length, close to its transition temperature T, and fed by a dc current
source. If the filament is connected to massi~e superconducting electrodes
and if its cross-section is sufficiently small, the current density j is
uniformly distributed, while the total phase difference ~ and voltage drop
V, if any, are effectively localized within the filament. According to the
basic JOSEPHSON relation

~ = 2eV/ft (1)

Such systems can sustain a time average Vd corresponding to the normal

resistance of a segment much shorter than a.
This has been attributed [1]
to phase-sZip aenters: the increase in ~ is compensated by -2~ jumps in
the local phase of the complex order parameter W= fe~X as the amplitude
f periodically vanishes over a section of the filament.

Recently, KRAMER and the author [2] showed that such states appear among
one-dimensional solutions of the simplest time-dependent GINZBURG-LANDAU
(TDGL) equations. A classification of possible steady states and of thresh-
old solutions signaling global instabilities has been achieved, both for
infinite filaments [2] and finite links [3]. In this report, we summarize
the main conclusions of those investigations, fit the kink soZutions first
identified by LIKHAREV [4] into that scheme, and illustrate their role in
transitions between (globally) different steady states.

2. One-Dimensional Time-Dependent GINZBURG-LANDAU Equations

Although they ignore quasiparticle relaxation processes which are important

in real superconductors, the TDGL equations are valid for sufficiently high
j [5]. Moreover, they seem to be qualitatively correct at low j and
represent the superconducting state quite well close to T. We are mainly
interested in states with nonzero electrical potential V cand nontrivial x-
dependence, and must rely on numerical computations. The complex form of the
equations

j (2)

313
is more convenient in that connection, but the physics involved becomes more
obvious if one separates amplitude and phase, i.e.,

Tf = f" + (1 - (X), 2 - f2)f (3)

Tf2 (X + iJ) = (f2 X I ) I , f2 X I - iJ I j. (4)

Thus T describes the relaxation of f, while (4) im~lies that the charac-
teristic spatial and temporal scales for iJ are 1/!T and 1, respectively,
in terms of the reduced units defined in [2,3J. Finally, the local JOSEPHSON
relation, iJ + X= 0, holds in superconducting regions. Contacts to massive
electrodes with a possibly different transition temperature TcO are simu-
lated [6J by imposing

iJ (0) v= ~ , iJ(d) 0, (5)

-i~
lj! (0) fO e , lj!(d) (6)

The location where X = iJ = 0, can be chosen to optimize computational

accuracy. The results presented here refer to T = 5.79, a special value
singled out by microscopic theory [2-5J.

3. Steady-States: Stability Ranges

Unless specified, stable henceforth refers to local stability. In an infin-

ite filament [2J two simple stationary states are known:

- the normal (N) state: lj! = 0, iJI = -j (stable for all j # 0);
·
- t h e superco nd ~c t &ng (S) state: lj! = f e iqx ,q 2 1 _ f2 ,J. = f2 q
00 00 00

(stable for q < 1/3 i.e., Ij I < j = 0.385).

m

A particular state is globally unstable if at least one stationary or

periodic (limit cycle) solution exists which only deviates locally from that
state. Stationary solutions representing critical S-nuclei in the N-state
are found for j < j , where j decreases with T and approaches j
for T ~ 2 (the exagt behavior in that range is difficult to determine~. A
typical example is shown in Fig. 1. As j ~ j these threshold solutions
grow in width and degenerate into two separatea, stable, stationary SN-
interfaces. According to [4J such kinks also exist for !j I < jm; they move
towards the N-side if Ij I < j . Once exceeded, a critical S-nucleus
actually splits into a pair of ~uch kinks moving in opposite directions.

Besides the unstable stationary saddle-point solution [7J

lj! = {!2q + i(l - 3q2)V2 tanh [(1 - 3q2)V2 (x - x o)/!:2 J}e iqx , (7)

which merges with the S-state at q2 1/3 and exists for Ij I < jm: we
also find a stable oscillatory solution localized about the S-state Ln a
narrow range j < jj I < j . ,where j . (0.326 for T = 5.79) also de-
creases with T~ This is ~fi~ phase-slipL~tate mentioned earlier. Below
J. min' solution (7) returns to the S-state after one or several phase slips.

314
 Fig.l Spatial dependence of the
0.8
\/-L
\
real and imaginary parts of the
complex order parameter ~ and of
0.6 \ the electrical potential ~ for a
\ stationary critical superconduc-
0.4 \
\ ting nucleus at j=0.30 < jc (0.335
0.2 for T=5.79). Only the left half
is shown; ~=O at the center x=O;
0 distances are measured in coher-
ence lengths
-0.2

The existence of (7) in that range is therefore not connected with a signifi-
cant global instability.

In the presence of well-separated boundaries, (7) or its analog with coth

can be matched to f < foo or > f oo ' respectively. Such solutions corre-
spond to stable S-s~ates unless Xo is inside the filament. The N-state is
then represented by [5J

~ = g(x) exp [-i~(t)J *

+ g (d - x), (8)

both g's being stationary solutions localized near the boundaries which
are stable as long as they do not exhibit maxima, i.e., for j > j , where
j = j if fO > f ( j ) and j < j otherwise, (it then cor~~sponds to
t~~t thfeshold solution ~hose maxi~ijm f C= f O)' Finally, stable phase-slip
solutions localized near each boundary are found if jLA(qO) < Ij i < j ;
jLA(qO)- being the current at which ~7) has a minimum r~ght at the bougdary,
i.e., /2qo = f O' This occurs for fO < 0.34 if T = 5.79; no properly
matched S-state can then be found if Ij I > jLA(qO)~

4. Transitions between Steady States

The situation sketched has all the trappings of first-order phase transi-
tions: globally different states with overlapping ranges of metastability and
critical nuclei strongly affected by boundary conditions, as well as inter-
phase boundaries. Moving kinks effect transitions, as illustrated in Fig. 2.
A long filament connected to electrodes with T 0 = T is initially in the
N-state (j = 0.34 > j )(a). In the absence ofcnoise; transitions occur only
when the metastabilityCbounds of a particular state are exceeded; thus j is
switched to j = 0.3 < j . Characteristic stages of evolution are shown.
First an NS interface de~aches itself from each boundary and moves inwards
(b). The velocity v of each kink, inferred from the slope of the resulting
ramp in Vet) (e) compares well with a direct calculation of dV/dj I [4J.
Phase slips of growing amplitude develop as the kinks overlap (c); tfiey stop
once ~ swings over the saddle-point solution (d). If j is switched into
the range between j . and j before the subsequent irreversible com-
pletion of the S-sta~~ntakes pl~ce, phase slips eventually stabilize as shown
on the inset (f). As pointed out by LANDAUER [8J on a simpler analogue of

315
a current-driven transition, noise is only likely to affect the nucleation
of kinks during the initial stage. Since the kink velocity vanishes at j ,
transitions and transient readjustme~ts in response to changes in j take c
a long time close to that value. CritiaaZ sZowing-down is also noticeable
in the initial stage of the transition from the state near jm'

M',fL,
"
(a)
f
lLJ(b)
1

" ""
1
L(e)
" x2
o 0 ~
X d X d X d

~1: (fb~
UU 600
X dO 0 100

Fig.2 Profiles of f (---) and ~ (---) are shown at different stages

(a-d) of a transition from the normal state in a filament of length d=20
coherence lengths (only the right half is shown) switched to j=0.3 at t=O;
in (c) curves are shown for a typical sequence: phase slip, maximum v,
maximum f at center, minimum V; in (d) both the final S-state and the
corresponding saddle point (dash-dotted curve) are exhibited. Time depen-
dence of total voltage (e) - times corresponding to (a-d) are indicated by
arrows. Stable phase slips localized in the center achieved via timely
switching to j=0.33 (f) - the resulting oscillations develop gradually and
are not displayed initially

References

1. W.J. Skocpol, M.R. Beasley, M. Tinkham: J.Low Temp. Phys. 16, 145
(1974)
2. L. Kramer, A. Baratoff: Phys. Rev. Lett. 38, 518 (1977)
3. A. Baratoff, L. Kramer: In Superaonduating-Quantum Deviaes and their
AppZiaations, ed. by H.D. Hahlbohm and H. Lilbbig (Walter de Gruyter,
Berlin 1977) p.62, and to be published
4. K.K. Likharev: ZhETF Pis. Red. (Soviet) 20, 730 (1974); Sov. Phys.
JETP Lett. 20, 338 (1974) --
5. A. Baratoff~L. Kramer: To be presented at the 15th International Conf-
erence on Low-Temperature Physics, Grenoble, August 23-29, 1978
6. K.K. Likharev, L.A. Yakobson: Zh. Eksp. Teor. Fis. (Soviet) 68, 1150
(1975); Sov. Phys. JETP 41, 570 (1975) --
7. J.S. Langer, V. Ambegoaker: Phys. Rev. 164, 489 (1967)
8. R. Landauer: Phys. Rev. A15, 2117 (1977Y;-other references cited there-
in

316
Solitary Phenomena in Finite Dissipative Discrete Systems*

E. Ben Jacob
Tel Aviv University, Department of Physics and Astronomy
Ramat Aviv, Israel and
Y. Imry
IBM Zurich Research Laboratory, CH-8803 RUschlikon, Switzerland and
Tel Aviv University, Department of Physics and Astronomy
Ramat Aviv, Israel

1. Introduction

Idealized nonlinear systems which are assumed to be infinite, continuous

and n9ndissipative are well known to exhibit a variety of solitary-wave
phenomena [1]. For example, the space-time dependence of the relative
phase ~(x,t) in an extended Josephson junction [2] is governed by a sine-
Gordon-type equation. In the idealized case, the sine-Gordon equation is
known to possess solitary solutions where a localized change of ~(x,t) by
2n propagates in an unchanged form with an arbitrary velocity v < ~ where
~ is the velocity of light in the junction. However, the existenZe and
properties of such solitary solutions for a realistic system is an important
unsolved problem [3,4]. Finite dissipation always exists in real systems,
presumably balanced by driving forces. Also, finite boundaries may be
important and the continuum approximation not always valid for discrete
Josephson-junction arrays or for the pendulum systems used for analog simu-
lations. The problem is not only of fundamental interest: a knowledge of the
dynamic I-V characteristics and their dependencies on external magnetic
fields and external electromagnetic waves is germane to the understanding of
Josephson-junction devices [5] and SQUIDS [6]. An interesting case occurs
when the driving (e.g., current) source is localized in space. Energy is
injected into the system at some point and it has to be fed into the solitary
mode at long distances.

We have used qualitative mathematical [7] and numerical [4,8,9] methods to

study these problems for models with N < 4 point junctions. While solitary
waves are found to exist in such systems, they do differ in many important
respects [10] from the idealized situations, as will be discussed. We also
find solitary-like phenomena when the system is not coupled strongly enough
to justify a continuum approximation.

2. The Model and Results

We consider a chain of N Josephson point junctions [4,7,11] where nearest-

neighbor junctions are coupled inductively and where each of the junctions

*Partially supported by the Commission for Basic Research of the Israeli

Academy of Sciences

317
is described by a simple lumped circuit model where the quasi-particle con-
ductance is approximated as ohmic [12,13J. In addition to the interest in
the specific case where N = 2, which is the simplest one where the spatial
dependence of the phase can exist, the model may be viewed as a numerical
simulation of a continuous junction of a length L. This will be discussed
later but it should be a reasonable representation of the continuum junc-
tion if N is taken as ~ L/AJ' where AJ is the Jo~ephson penetration
depth. The equations of the model are (see inset to FLg. 1)

IiI + G 8 1 + sin 8 1 II + K [8 2 - 8 1 - 8ex J

Ii. + G 8. + sin 8. I. + K [8. 1+ 8j+l - 28.J; 2 < j < N-l (1)

J J J J r J
eN + G 8N + sin 8N IN + K [8 N- l - 8 + 8 exJ,
N

for an open chain. Here 8. is the phase difference of the jth junction
K = 2 'V
~O / rrL I J , ~O
• J 'V.
LS the flux quantum, L the Lnductance and 8 ex the
flux (in units of ~O/2rr) through the circuit between the two consecutive
junctions, I J the Josephson current amplitude, I. the ~iiving current
Ehrough the Jth junction in units of I J , and G =J(w RC) where w =
C/A J is the Josephson plasma frequency. R is the juRction quasipart~cle
(or shunt) resistance and C its capacitance.

In the case of two junctions [4J, three general types of solution exist,
resulting in three branches of the dc I-V characteristics: a static solution
where V = <8 1> = < 82> = 0, a running solution where 8 and 8 2 increase
steadily, which has the highest voltage, and a beating solution of inter-
mediate voltage where the two junctions interchange static and running
roles. The beating solution can be viewed as being of a solitary type, a
phase change of an integer multiple of 2n (which can also be regarded as
n flux units) running across the system. The beating solution exists only
when K is not too large. For K > 1 the two junctions move together and
the solitary mode does not exist fo~ N = 2. More generally, for suffi-
ciently large N, K determines a correlation length among the junctions
which is equal to IK
(in units of the distance between the junctions) and
the solitary mode is only visible for N > IK.
In fact, the continuum
approximation for the discrete system should become exact for K» 1, which
will therefore necessitate N» 1. However, for K < 1 the N 'V 2-4 case
does exhibit solitary modes, which, as long as K is not too small, appear
as qualitatively-valid approximations to the continuum case. The discrete
case with K« 1, is very interesting; here, in the beating mode one pen-
dulum can make a number, n , of revolutions of the order of l/K, while the
others are approximately stationary. This is propagated as a function of
time along the chain and can be regarded as an n soliton state.

In Fig. l(a), a bound double soliton-like solution is presented. It is

seen that the phase change of 4n is propagated along the chain. The form
of this solution is not exactly invariant with the motion and its propa~a­
tion velocity is not fixed, but a strict periodicity in the motion of the
whole system exists. After one soliton crosses the chain, a new one is gen-
erated at the first junction and launched along the chain. Other types of
solitary solutions are possible. A double (phase change of 2 x 2rr) soliton

318
with a single peak in e., a bound pair of solitons, etc. In the latter
case two solitons move at each instant along the chain: see Fig. l(b).

Fig.1 Junction voltages, N=4.

(a) A bound two-soliton pair for
G=0.8, K=1.8/24n, 1 1=3.21 J , 12=
13=14=0; (b) Two correlated soli-
tons, for the same parameters
except K=1.8/6n. The straight
lines are drawn to guide the eye
through each soliton

We emphasize that energy is only fed through the driven junction No. 1
into the system. The solitary modes are launched from junction No. 1 and
leave the system at the open end - junction No. 4 - with no reflection.
While we cannot rule out the possibility that due to energy dissipation,
(finite G) the motion will eventually decay for N + 00 this is apparently
,

not suggested by the fact that within an accuracy of about 1%, no decrease
in amplitude is observed at N = 4. It appears that, in fact, the periodic
train of solitons is generated at junction No. 1 and that somehow the energy
is propagated along the train and thus fed to the leading pulse.

We have also treated a closed ring with N = 4, driven at a point. Here,

solitons are launched in two opposite directions with the same sense of
rotation. These two modes seem to pass through each other later without any
significant changes in shape. Of course, it may also be assumed that they
collide head-on and reflect backwards, thus not providing proof of the non-
interacting nature of these modes.

Each type of solution corresponds, in the Josephson case, to a branch of

the I-V characteristics (Fig. 2) where hysteretic jumps among the branches

11/ 1;)
5
4.5
Fig.2 dc I-V characteristics,
4 traced with increasing or de-
3.5 creasing current (see arrows).
1=1 1 , 12=1 3=1 4=0. The smooth
3 lines are for ~ = 0, the dashed
one for ~ = ~ f~~ (a) Same para-
meters asefn l£b); (b) G=0.8,
o <V1> 1.0 K=1.8/12n; the voltage is in units
of l1w /2e
p

319
may occur. From our investigation, an understanding of the conditions for
obtaining the various solutions and their limits of stability (including
cases where only multiple solitons are possible) are obtained. Also, the
dependence on external fields is understood and can be estimated analyti-
cally in agreement with the numerical data.

3. Conclusions

Solitary-type solutions are indicated as existing in dissipative driven

finite discrete systems. However, they differ in important respects from
the ideal case and their behavior does not seem to be just a small modifi-
cation of the latter one. For example, the propagation velocity is fixed
for a given set of parameters of the system [10J, the modes are periodically
created at the driven end and seem to propagate without any overall dissipa-
tion but with changes of shape and velocity, while the discrete system has,
for small K, modes that do not exist for the continuous one.
Experimentally, the most direct observation of these effects in the
Josephson case would be through the voltage peaks that occur upon the pass-
ing of the solitary mode at a given junction. A less direct method to
observe these effects would be via the low V branches of the I-V character-
istics including their limits of existence and dependence on external fields.
More work is needed on the questions of the interaction among the solitary
waves, the finite-temperature noise effects and the evaluation of space-
time correlation functions of the phase [14J. The behavior as a function of
the dissipation parameter G will be discussed elsewhere. The G + 0 limit
does not appear to be a simple regular one.
One of the authors (YI) would like to thank Drs. A. Baratoff, P.M. Marcus,
T. Schneider, and E. Stoll for useful discussions and for the hospitality at
the IBM Zurich Research Laboratory.

References

1. A.C. Scott, F.Y.F. Chu, D.W. McLaughlin: Proc. IEEE 61, 1443 (1973)
2. B.D. Josephson: Adv. Phys. 14, 419 (1965) --
3. T.A. Fulton, R.C. Dynes: Solid State Commun. 12, 57 (1973)
4. Y. Imry, P.M. Marcus: IEEE Trans. MAG-13 , 868~1977)
5. H.H. Zappe: IEEE Trans. MAG-13 , 41 (1977)
6. D.C. Tesche, J. Clarke: J. L~ Temp. Phys. 29, 301 (1977)
7. V. Imry, L.S. Schulman: J. Appl. Phys. 49, 749 (1978)
8. E. Ben Jacob, Y. Imry: Proc. LT15: To b;-published
9. E. Ben Jacob, Y. Imry: To be published
10. K. Nakajima, V. Onodera, T. Nakamura, R. Sato: J. App1. Phys. 45, 4095
(1974)
11. J.A. Blackburn, J.P. Leslie, H.J. Smith: J. App1. Phys. ~, 1097 (1971)
12. W.C. Stewart: Appl. Phys. Lett. 12, 277 (1968)
13. D.E. McCumber: J. App1. Phys. 39:-3113 (1968)
14. In quasi-lD magnetic systems, the correlation functions cose.(O)
cose.(t) lead to the observabi1ity of soliton-related effect~ by
ine1kstic neutron scattering. H.J. Mikeska: J. Phys. Cll, L29 (1978);
J.K. Kjems: Private communication

320
Stability of Nonunifonn States in Systems Exhibiting
Continuous Bifurcation

M. BUttiker and H. Thomas

Institut fUr Physik der Universitat Basel, KlingelbergstraBe 82
CH-4056 Basel, Switzerland

Many systems which are driven away from equilibrium by pumping or forcing
show an instability of a stationary uniform state to a new stationary or
travelling-wave state with broken translational symmetry. In continuous
systems the loss of stability of the uniform state, which is usually as-
sociated with a destabilization of a normal mode, gives rise to a bifur-
cation of a whole manifold of new solutions. The task is then to select
the members of this manifold according to their stability properties, and
thus to find the candidates which may be physically realised. The travel-
ling-wave case can be reduced to the stationary case by a transformation
to a moving frame. In this paper, we therefore focus attention to sta-
tionary states.

Introduction

Consider a system described by a set of fields ~(r,t) = {~1(r,t)'~2(r,t),

•••. } and a set of parameters cr = {crl'cr2' •... } which can be externally
controlled. Let the time evolution of the fields ~(r,t) given by

a~/at = B(~,cr) (1.1)

where B is a nonlinear partial differential operator acting on ~ and is

assumed to be translationally invariant in rand t. A linear stability
analysis of a given stationary solution ~s(r) leads to the eigenvalue
problem

(1. 2)

for the perturbation 8~(r) exp(+ At). The state ~ is called (linearly)
stable if Re A(cr)~O for all eigenvalues, and if tRe eigenvalues A = 0
belong to hydrodynamic modes or to symmetry-restoring modes. If the state
is uniform, the eigenfunctions are plane waves, 8~q~ exp(iqr) and the
eigenvalues Aq can be characterized by the wavenumber q.

We are interested in the case that the uniform state becomes unstable for
cr=cr against a soft mode with critical wavenumber q • In the neighbour-
hooa of the instability we assume a dispersion law c

Re A(q,cr)=-r(cr-cr )+D(q-q )2; ImA(q,cr)=O. (1. 3)

c c

321
With increasing a, successively more and more modes become undamped /1,2/,
each q mode generating at its point of bifurcation ab>a c ' after splitting
off a factor exp(iq r),a small-amplitude ~eriodic solution with period
Ab=2nID/r(a~-acIV2 ~nd amplitude A~(a-ab)2. With increasing a-ab' the ampli-
tude will p~ck up higher harmonics but the states will remain periodic.

Method

The object of the present paper is a general stability analysis of this

family of solutions ~s(r,a,ab)' Our method is based on the following con-
siderations:

1. Because the nonuniform states ~ (r) are states of broken translational

symmetry, a set of equivalent stat~s ¢ (r+a) is generated by an infinitesi-
mal translation. Therefore there alway~ exists a Goldstone mode (GM)
o¢=a¢ (r)/ar with egenvalue zero.
2. Th~ linear operator L(¢ ) has at least the symmetry of the stationary
state ¢. If ¢ is periodtc with period A, then L(¢ ) is invariant under
the discfete gr5up TA of translations nA. ThereforeSthe eigenvalues are
multivalued functions An(q) of the reduced wave vector q over a Brillouin
zone (BZ) - T(/ A<q,; T(! 1\
3. For the zero-amplitude solution A->D, ab=a, the spectrum is given by (1.3).
For the nonuniform states ab<a (a fixed, a b decreasing), the width of the
BZ changes a l/A(a b ), and with increasing amplitudes A(ab) gaps develop at
q = 0, n/A separating the bands A (q).
4. For a general q in the BZ, no cros~ing of eigenvalues can occur as A
increases.

We show in the following examples that depending on the wavenumber q of the

GM, these properties together with k'p perturbation theory around the GM
lead to general conclusions about the stability of the family of the peri-
odic states.

Examples

In recent years several methods have been developed to describe systems near
the point a of instability by time-dependent Ginzburg-Landau equations
/3,4/, withConly a few order parameter fields varying slowly in space
and time being kept to describe the instability.

A. Symmetry-Conserving Instability

The first example is related to current instabilities /5/. The excess elec-
tric field ¢(¢ real) satisfies the time evolution equation

(2.1)

which has two stationary uniform states ¢ 1=0, ¢u 2=a with spectrum A =
a+q2 and A=-a+q2, respectively. The statignary nonuniform states for a>O
are represented in the (¢,¢, )-plane in Fig. lao The operator L is invari-
ant under TA, and the BZ hasXtherefore width 2n/A. The GM belongs to q=o
and lies in the center of the BZ (Fig. lb). For a ¥a, all eigenvalues
A (q) are nondegenerate and belong to the unit rgpresentation of the small
g¥oup even at q=O. Therefore, the unstable band cannot cross the GM fixed
at A=O, q=O, i.e. all periodic states A>Ab including the solitary state
A=oo are unstable.

322
 a b

Fig.la Stationary nonuniform solutions of Eq. (2.1). A solitary solution

separates periodic solutions (closed curves) and unbounded solutions
(open curves)
.Fig.lb Spectrum of the unstable state ¢=a (broken line) and of a periodic
solution with period A of Eq. (2.1)

B. Symmetry-Breaking Instability

As second example we consider the $4 potential with a real field ¢ with

a¢/at=-av(¢)/a¢+¢,xx; v(¢)=- ~2~', (2.2)

The softening of the normal modes \=-a+q2 of the uniform state ¢=O leads
to the bifurcation of two uniform states ¢ =±aV2 and the familiy of nonuni-
form states shown in Fig. 2a. u

a b

,, ,
'---
~

Fig.2a Stationary nonuniform states of Eq. (2.2). Two solitary solutions

separate the periodic states (closed curves) from the unbounded
states (open curves)
Fig.2b Spectrum of the unstable state $=0, a>O ,(broken line) and of a peri-
odic solution with period A of Eq. (2.2)

Because a translation by A/2 changes the stationary solution $ to -$ , and

L depends only on $2, L is invariant under TA . The width of ~he BZ ~s thus
4n/A and the GM lie~ at the BZ-boundary (Fig./~D). Therefore the lowest
band A (q,A) has the possibility to became positive above a critical A.
To sho~ that this is not the case, we apply a k.p perturbation around the
GM (A=O). One finds A(k)={-12aA exp(-AIQT2)} k 2+0(lkl 3) <0, k=2n/A-q. Thus
for all finite A there exist modes with A(q~<O, and therefore all periodic
solutions are unstable. Only the stationary uniform states $ =±~ and the
solitary solutions are stable. u

As third example we consider the time-dependent Ginzburg-Landau equation for

a complex field with $4 potential

(2.3)

which describes the onset of convection in the Rayleigh-Benard instability

/3,6,7hthe Laser threshold /7/,as well as chemical instabilities /8/. It
is also related to the superconducting phase in thin wires /9,10/. For a>o

323
 a b

q
-K K

Fig.3a Stationary states of Eq. (2.4). The broken lines represent the
~ "rolls". In each halfplane R>O, J#O the unstable roll is surrounded
by s.mall amplitude periodic solutions. From the stable roll a soli-
tary solution emerges and returns
Fig.3b Spectrum of an unstable roll in the extended BZ

one has in addition to ~=O.the uniform stationary states ~ =~ exp(i8o),

(8 arbitrarYh The nonuniform states ~ =R exp(i8), R, 8 re~l, are given by
thg equations s
J2
R -( -3 )+aR-R 3 =0 J=8'xRL=const (2.4)
'xx R '
and are represented in Fig. 3a. The stability of these solutions is deter-
mined by a selfadjoint operator

(2.5)

acting on a spinor ~. The uniform stationary states have spectra Al=A2=-a+q2

for ~ =0 and Al=2a+q2, A2=q2 for ~ =~ exp(i8 ) (a>O). The simplest nonuni-
form ~tates are the constant-ampli'fude stateso("rolls") ~ =(a-K 2 )V 2 exp(iKx)
(broken line in Fig. 3a) belonging to J=K(a-K2). ForSthese states, H
is still fully 6ranslationally invariant, and the spectra are A=(a-K2+q2)±
(4q2K2+(a-K2)2) 2 /111 showing that the rolls for K>ra[3 are unstable(Fig.3b).
Around each unstable "roll" there exist doubly periodic small-amplitude so-
lutions with 8(x)=Qx+~(x), R(x) and ~(x) simply periodic with period A.
Phase shifts with respect to each of the two periods generate a two-para-
meter family of equivalent states, giving rise to two A=O modes, ~l=(R,x'
R~,x) and ~2 = (O,R). ~1+Q~2 corresponds to a~s/ax.

a b

Fig.4a Spectrum of an unstable roll (broken line) and of a (doubly) perio-

------ dic solution J#O (solide line) with gap below A=O
Fig.4b Same situation as in Fig. 4a but with gap above A=O

324
For J=O (2.5) decouples into two independent eigenvalue problems for u and
v. The eigenvalue problem for u is identical with that for real ~ (2.2).
The eigenvalue problem for v has a band A (q)<O . Thus, the periodic solu-
tionsare unstable under an additional braRch of modes leading in the limit
A+oo to an isolated eigenvalue AO<O of the solitary solution /3,10/.

Whereas for J=O (2.5) is invariant under T AI2 , for J~O the translation
group is T~ From Fig. 3b it is seen that tne unstable rolls have three ei-
genmoaes w1th eigenvalue ~=O. Two of these are the A=O modes discussed
above, the third one will have A3~0 as one moves away from the constant-
amplitude states, and the stability would depend on the sign of A3' However,
even if A3<0 initially, the situation of Fig. 4a might change at a critical
A into that of Fig. 4b, because A3 belongs to a different representation
and is therefore not prevented from crossing A=O. A k.p perturbation shows
that the A=O modes vary as A=cq2, and the stability depends on the sign of
c, but this calculation has not been carried through. In Ref. /10/ the
solitary solution was shown to be unstable. This suggests that the gap
opens downwards(Fig. 4a),rendering all periodic solutions unstable.

References

1. K. Kirchgassner in Synergetics edited by H. Haken, Springer Verlag

Heidelberg, New York 1977
2. H. Thomas in the proceedings of the V lnternationale Conference on
Noise in Physical Systems, Bad Nauheim, March 1978
3. A.C. Newell, J.A. Whitehead, J. Fluid, Mech. 38, 279, (1969)
4. H. Haken, Z. Physik B21, 105 (1975) --
5. M. Blittiker, H. Thoma~ Phys. Rev. Lett. 38, 78 (1977)
M. Blittiker, H. Thomas, Solid-State Electronics 21, 95 (1978)
6. L.A. Segel, J. Fluid Mech. 38, 203 (1969) --
7. R. Graham, in Fluctuations,-rnstabilities arid Phase Transitions, ed.
by T. Riste, Plenum Press, New York and London (1975)
8. Y. Kuramoto, T. Tsuzuki, Prog. Theor. Physics, 54, 687 (1975)
9. J.S. Langer, V. Ambegaokar, Phys. Rev. 164, 498~1967)
10. D.E. McCumber, B.I. Halperin, Phys. Rev~l, 1054 (1970)
11. W. Eckhaus, Studies in Non-Linear Stability Theory, Springer Verlag,
Berlin, Heidelberg, New York 1965

325
The Sine-Gordon Chain: Mass Diffusion

T. Schneider and E. Stoll

IBM Zurich Research Laboratory, CH-8803 RUschlikon, Switzerland

In recent years, there has been growing interest in the theory of Brownian
motion of interacting and non-interacting particles in a periodic potential
driven by an external field [1-4J. Interesting aspects are: a close analogy
between the non-equilibrium phenomena occurring close to the critical field,
and continuous phase transitions in thermal equilibrium [3J; the prediction,
that the average velocity at small fields and low temperatures might be
associated with thermalized sine-Gordon solitons [IJ.

In this work, we present and discuss some molecular-dynamics results

aimed at studying the response on a small field of a sine-Gordon chain in
the presence of damping, and subjected to a random force. On the basis of
these results we hope to elucidate the relevance of thermalized sine-Gordon
solitons in the diffusion process.

The equations of motion of the sine-Gordon chain, using the mechanical

analogy, are the Langevin equations

M X. av
- --- r M X. + 11. (t) , i 1, ... , N, (1)
~ ax.
~
~ ~

where
A
V ;
I [B(l - cos X.a) + - (X i + l - X.) 2
~ 2 ~
- E X. J.
~
(2)
i

The system is a linear array of N-interacting particles of mass M moving

in a periodic potential and subjected to an external field. We assume
periodic boundary conditions so that

(3)

The system is in contact with much lighter particles, treated as a heat bath
of temperature T. The collisions with these particles are described by a
friction - r M X.
and a random force 11.(t) which acts on the particles
independently, i.~., ~

<11. (t» ; 0
~
(4)
2 r M kBT 0 .. o(t
~J

326
B, A and a are model parameters. The problem is thus equivalent to the
Brownian motion of N-particles.

Using a molecular-dynamics technique, as described in [5J, we solved the

coupled set of Langevin equations (1) numerically. In the present study, we
considered systems of 10 3 particles and the model parameters were chosen as
follows:

M I, a = I, r = 4, E 0.1, 0.05 }
(5)
B I, A 50, 100 15, 20, 25.

This choice matches the regime where the mass transport characterized by the
. 1
L
average velocity < X> = liN <X.> might be dominated by solitons. This
regime is defined by 1

E « I, A» B (6)

where EO is the rest energy of the sine-Gordon (S-G)-kink solution in the

undr1ven system and treated in the continuum approximation. In this regime
and the large damping limit, TRULLINGERet al. [lJ predicted that the mean
velocity should be dominated by thermalized S-G-kink solitons. They found in
terms of an approximate solution of the Smoluchowski equation,

r M<X> (7)
E

where the kink rest energy is given by (6).

To test these predictions, we calculated the average velocity and followed

the time evolution of the system in the steady state. Figure 1 shows snap-
shots of the instantaneous particle positions at two times.

600 ~=2000:
';
400 r-

200f--
I; Fig.l Snapshots of the instan-
taneous particle positions at
two instances of time for B=l,
A=lOO, E=O.l, k BT=20 and r=4.
The particles are labeled by ~
and X denotes the magnitude
of their displacement

327
These results clearly demonstrate that the mass transport in the regime
where (7) is expected to hold is governed by moving kinks and antikinks.
For a more detailed pictorial presentation of the kink dynamics, we refer
to our motion picture. In particular, it illustrates the creation of kink
and antikink pairs by thermal fluctuations. Subsequently, both partners
move in opposite directions under the action of the applied field. More-
over, a kink may annihilate an antikink created at a later time, or vice
versa. This picture is consistent with that of LANDAUER [6]. In Table 1,
we compare our results for the mean velocity with (7). From the results
for A = 100 and kBT = 20 at different fields, it is seen that we are in,
or close to, the linear response regime. Accordingly, comparison with (7)
makes sense. For this purpose we divided our estimates for r <X> /E by
the temperature-dependent terms of (7). The resulting constants are listed
in the last but one column of Table 1. In the last column, we used an
alternative expression where the prefactor (EO/k BT)3/2 in (7) is replaced
by (E O/k BT)2.

Table 1 Numerical estimates for the average velocity for different temper-
atures and fields. The other model parameters are given by (5)

A kBT E r<x> E k T~ -3/2 exp

r <x>C o
B
(~)
k T
B
-<x~o t2 ~EO 1
r E - kBT exp kBT

100 15 0.1 0.023 + 0.002 3.86 + 0.37 1.67 + 0.17

100 20 0.1 0.049 + 0.007 3.34 + 0.45 1.67 + 0.17
100 25 0.1 0.071 + 0.005 3.04 + 0.21 1.70 + 0.17
100 20 0.05 0.029 + 0.003 3.96 + 0.42 1.98 + 0.20
50 15 0.1 0.051 + 0.007 3.02 + 0.40 1.56 + 0.16

The quite good agreement between these constants supports the functional
relationship (7), as obtained from an approximate solution of the Smoluch-
owski equation. The modified prefactor leads, however, to a better agree-
ment.

To summarize, we have shown that the average velocity or the diffusion

coefficient is, in the appropriate regime (6), dominated by kink and anti-
kink solitons of the sine-Gordon equation.

328
References

1. S.E. Trullinger, M.D. Miller, R.A. Guyer, A.R. Bishop, F. Palmer,

J.A. Krumhansl: Phys. Rev. Lett. 40, 206 (1978); Phys. Rev. Lett. 40,
1603 (1978)
2. R. Festa, E. Galleani d'Agliano: Physica 90A, 229 (1978)
3. T. Schneider, E. Stoll, R. Morf: To be published, Phys. Rev. B 18,
(1978)
4. R.A. Guyer, M.D. Miller: Phys. Rev. A 12, 1205 (1978); Phys. Rev. A
12, 1774 (1978)
5. T. Schneider, E. Stoll: Phys. Rev. B 17, 1302 (1978)
6. R. Landauer: Private communication

329
Solitary Wave Propagation as a Model for Poling in PVFz

A.J. Hopfinger, A.J. Lewanski, T.J. Sluckin, and P.L. Taylor l

Case Western Reserve University, Cleveland, OH 44106, USA

1. Introduction

A piezoelectric and pyroelectric material of great technological importance

is the phase-I, or a-phase, form of poly(vinylidene fluoride), also known
as PVF Z' In this phase the crystal consists of a parallel array of planar
zigzag macromolecules, each constructed of alternating and oppositelY_30
oriented CF Z and CH Z units. The large electric dipole moment of 7xlO Cm
of each CF Z unit ana their parallel orientation in the unit cell make this
material p~ezoelectric. I2§ orthorhombic unit cell, shown in Fig. 1, has
a dipole moment of 1.4xlO Cm.

The a-phase material is most commonly produced by stretching films of

the more readily crystallizable a phase. Because the chain backbone fol-
lows a more tortuous path through
the unit cell of the a phase than
the planar zig-zag of the a phase,
a longitudinal stress favors forma-
tion of the a phase. In such
films the chain direction is gen-
erally aligned with the direction
of strain but no net macroscopic
electric polarization is exhibited;
presumably the polarization of
individual crystallites lies in
random directions perpendicular to
the strain direction. Electrically
active material may, however, be
produced by annealing a sample in
a sycrmetry-breaking electric field
of sufficient strength, a process
known as poling. The microscopic
description of the poling process
and of the mechanism of electrical
activity remains a matter of some
controversy at the time of
writing[lJ.

The present paper describes one aspect of a research program directed

towards an elucidation of the microscopic behavior of a PVF Z' We here

1. Hork supported by the U. S. Army Research Office and in part by the

U.S. National Science Foundation.

330
explore a model in which the poling process involves a reversal of the
direction of polarization of the chains within a crystallite through the
propagation of a 180 0 twist along the chain length.

2. The Polymer in Equilibrium

It is, in principle, possible to treat crystalline polymers merely as lat-

tices with a rather large number of atoms per unit cell. In practice, on
the other hand, it is natural to take advantage of the relative strength
and rigidity of the intrachain covalent forces in comparison with the
much weaker van-der-Waals forces of the interchain interaction. This can
be accomplished by first calculating the properties of an isolated chain
and then adding the effects of interchain forces in a mean-field formalism.
Such a program has been applied[2) to systems consisting of only CH 2
units (polymethylene) or only CF 2 units (poly tetrafluoroethylene) , and is
currently being applied[3) to PVF 2 . The central quantities in these cal-
culations are the distribution n(8) of torsional angles along a chain,
and its temperature variation in equilibrium. The idealized Hamiltonian
that we use is

(1)

where U is an intrachain interaction and W is an interchain interaction.

These interactions were computed using conformational analysis[4). The
free energy f of this system per monomer is determined from the smallest
eigenvalue of the transfer-integral equation

exp(-Bf)1jJ(8) = (21r)-1 lIT exp-B[U(8,8')+\2h(8)+\2h(8')H(8')d8' (2)

o

where the interchain forces act through the mean field h = oW/on(8) and
where B (kBT)-l. The distribution of torsional angles is given by

n(8) ljJ(8)x(8) (3)

with X(8) the eigenfunction of the equation adjoint to (2). This formal-
ism allows the prediction of the limits of stability and metastability of
various crystalline and melted phases.

3. The Poling Process

Under conditions in which B-PVF 2 is stable or metastable, the interchain

potential energy was found to favor a parallel or antiparallel alignment
of each CF 2 unit with the local crystalline b-axis. Upon taking some
liberties with the geometry of the permissible excursions of torsional
angles we are led to consider a Hamiltonian of the form
\' • 2 2
H L . [~I8. +'~K(8.-8.+l) -A 2 cos28.-Alcos8.) (4)
~ ~ ~ ~ ~ ~

Here 8. represents the orientation of the ith monomer unit relative to the
crystalline b-axis, I is its moment of inertia, K is a torsional parameter
of the intrachain interaction, O<A = pE + ~ « A2 with ~ and A inter-
chain energy constants and pE the ~roduct of monomeric electric 2dipole
moment and electric field. The addition of temperature-dependent
Brownian-motion forces Fi(t) and a phenomenological damping mechanism

331
Fig.2 In this sequence a wave of polarization is seen to proPagate as
the chain rotates 180 0 about its axis. The electric field is applied down-
wards to a sample at intermediate temperature

332
characterized by a constant n then result in an equation of motion of the
form

181.' = -2A2sin28l.' - Al sin8.+K(8. 1-28.+8. 1)-In 8.+F .(t) (5)

1. 1.+ 1. ~ 1. 1.

Similarities between this system and those studied by other authors [5]
are immediately apparent.

In the poling mechanism envisaged for this model, the high temperature
and applied field induce the rotation of a chain from a conformation in
which all 8. are in the vicinity of TI to one in which all 8. are close to
zero. The Birth and propagation of a solitary wave of pOlafization are
then predicted by the solution of (5).

In the continuum limit, for low T, the velocity v, wanedobthtaiCl.'nkness xo'

of the solitary wave can be determined analytically.

(6)

where a is the polymer repeat unit. However, the low damping and discrete
nature of the system limits the applicability of (6), and so a numerical
study was made of this system. The results were displayed as a computer-
generated cine-film, sample frames of which are shown in Fig. 2.

110re details of this work will be given in a future publication.

References

1. A review of the properties of PVF 2 is given by Y. Wada and R.

Hayakawa in Japan. J. Appl. Phys, 15 2041 (1976).
2. F. P. Boyle, P. L. Taylor and A. J~Hopfinger, J. App. Phys. 46 4218
(1975), J. Chem. Phys., ~ 353 (1977), J. Chem. Phys. to be published.
3. F. P. Boyle, A. J. Hopfinger, T. J. Sluckin and P. L. Taylor, Bull.
Am. Phys. Soc.,ll 379 (1978).
4. A. J. Hopfinger, Conformational Properties of Macromolecules (Academi~
New York 1973).
5. S. E. Trullinger et al. Phys. Rev. Lett., 40 206 (1978) and references
therein. In the undamped, zero-temperature, continuum limit one
finds the double sine-Gordon equation discussed by S. Duckworth
et al. (Phys. Lett. 58A, 19, 1976) and J. D. Gibbon et al. (Phys.
Lett: 65A, 380, I978~

333
Theory of One-Dimensional Ionic and Solitary Wave Conduction
in Potassium Hollandite

J.B. Sokoloff and A. Widom

Northeastern University, Boston, MA 02115, USA

Potassium hOllandite is an ionic conductor which exhibits one dimension-

al character in the sense that the mobile ions move in one dimensional
channels provided by the remaining ions in the crystal [1]. There is a
theorem which states that the conductivity of a one dimensional conductor
should be zero if it has impurities, because all the states of the current
carrier are localised [2]. The reason for this can be roughly understood on
the basis of the following simple picture: if the distribution of impurities
is represented by a row of n potential barriers, each with reflection co-
efficient Rl' the transmission coefficient tor a particle incident on this n
barrier system approaches zero in the thermodynamic limit when all coherence
effects between barn.ers are neglected (which is good for random impurities).
In fact, the mean free path (i.e. the distance through the system of barriers
over which the wave function of a particle incident on the barriers drops to
a small fraction of its Incident value) is (C Rl)-l where C is the number of
impurities per unIt length and Rl is the transmission coefficient of one
barrier. For the potential barrier
Vo (1 + eX/b)-l

where Vo is the barrier height and b its width, the reflection coefficient
is given by
sinh7T(k l -k 2)b 2
Rl = ( Sinh7T(k l +k 2)b] ~

k2 [2m(E-Vo) J" Ih

where E and m are the incident particle energy and mass l3]. For m equal to
the potassium ion mass, the mean free path is already ~O.l mm when E-Vo =
lO-4 eV. That is, the potassium ions are essentially classical and travel
over the barriers without reflection. The apparent resistenceless current
flow implied by thIS result does not occur for the high concentrations of
potassium ions actually found in potassium hollandite because the ions can
dIssipate momentum by particle-particle scattering in which momentum is not
conserved because of the lattice potential due to the other ions.

For high ionic concentrations (i.e. most channel sites occupied by

potassium ions), conduction is best described by the solitary wave like
motion of vacancies in the potassium ion lattice. For applied electric

334
fields small compared with the value necessary to push a vacancy over the
potential barrier which must be overcome to move the vacancy to the next
lattice site, electrical conduction proceeds by thermally activated hopping,
in which thermal equilibrium is re-established between hops. In this case,
the method of Weiner and Sanders [4], originallY proposed for describing one
dimensional dislocation motion, can be applied because the vacancy is a type
of dislocation in one dimension. The vacancy hopping rate is given by
a2 ' 6E ~ ~eaE
f
~
~11 (m-J' v exp [- a kT
where a 2 is the force constant of the potential well in which a potassium
ion sits, ~Ea is the activation energy, a is a lattice constant of the
channel which holds the potassium ions, E is the applied field and v is a
dimensionless quantlty deflned in reference 4. Then, the current density
and conductivity are found from
2b 2 a2 , -6Eal kT
J = oE = nea If -f ) = [~1.- (_)2 V e ] E,
+ - kT 211 m

where b is the mean potassium ion spacing and n is the number of vacancies
per unit volume. The quantity 6Ea is found by solving the equilibrium
equations for the ground state of the one vacancy case and with the vacancy
in the activated position and taking the difference of the energies for both
configurations. The result is an activation energy much smaller than the
observed 0.2 eV [5]. In real i ty, there are always breaks in the channels
in which the potassium ions reside and the actually observed activation
energy is the energy to get around such a break [6]. The calculated
activation energy and conductivity should be found, however, in the higher
frequency A.C. conductivity.

Vie would like to thank the U.S. Department of Energy (Contract Number
BG-77-5-02-4432) and the Natlonal Science Foundation (Grant Number (DMR 75-
06789) for financial support.

References

1. H.U. Beyeler, Phys. Rev. Lett. 37, 1557 (1976); H.U. Beyeler,
L. Pietronero, S. Strassler and H.J. Wiesmann, Phys. Rev. Lett. 38,
1532 (1977).
2. N.F. Mott, Advances in Physics 16, 49 (1967).
3. L.D. Landau and 1. Lifshitz, Quantum Mechanics - Nonrelativistic Theory,
2nd edition (Pergamon Press, Oxford, 1965), p.78.
4. J.H. Weiner and W.T. Sanders, Phys. Rev. 134, Al007 (1964).
5. J. Singer, H.E. Kautz, W.L. Fielder, J .S. Fordyce, "Fast Ion Transport
in Solids", ed. W. van Gool (American Elsevier Publishing Company Inc.
New York, 1973), p.653.
6. D. Kuse and H.R. Zeller, Phys. Rev. Lett. 27, 1060 (1971).

335
IV. Summary
Summary: Where Do Solitons Go from Here?

S.E. Trull inger

Department of Physics, University of Southern California
Los Angeles, CA 90007, USA

When Alan Bishop asked me to help organize this symposium, he gave me two
very difficult tasks. The first was to obtain travel support money from the
National Science Foundation to assist the U.S. participants. The completion
of this task was made possible largely due to the efforts of John Connolly
and Lew Nosanow in the Division of Materials Science at NSF. These gentlemen
realized the importance of the symposium and worked very hard to insure that
a decision on the travel proposal would be made in time. We are all in-
debted to them and the National Science Foundation without whose generous
assistance this symposium would not have been possible.
The second task which Alan gave to me was to give a summarizing talk at
the end of this symposium. One obvious purpose of such a talk is to remind
the participants of the key points and important findings that have been
discussed during the symposium. I have spent several hours trying to prepare
an adequate summary and this has proven to be impossible. Although I attri-
bute this failure mostly to my own inability, I would like to place at
least part of the blame on the enormous success of the symposium. We have
listened to almost forty talks in the last three days and each one has con-
tained very important results and concepts. The talks have exhibited in a
dramatic way the breadth of applications of solitons and other nonlinear
structures in condensed matter.
Although it does seem impossible to do justice to forty talks in the
forty minutes I am allotted, I think it is important for all of us to at
least briefly review what has transpired. For this purpose I would like
to ask each of you for your help, by relying on your own memories of the
last three days, rather than my imposing on you my own recollections. In
order to stimulate your memories, I shall recite a list of some of the key
words or phrases that have cropped up (perhaps more than once) during the
symposium. After each item on the list I will pause for a second or two
to allow you to reflect on those features connected with the word or phrase
that impressed ~ the most. Let us begin:
Integrable Systems Wobblers
Conservation Laws Shelves
Inverse Scattering Transform Fluxons
Canals, Barges, Horses Parasitic Modes
Solitons (aristocratic and plebian) Twists
Phonons, Mesons, Radiation (or what- Cylindrical Solitons
ever you want to call them) Phase Slips
Kinks Vortices
Bions Phase Boundaries
Breathers Domai n Walls
Pulsons Clusters

338
Mexican Hats Lunch Bags
Microdomains Help-wanted ads
Soft Modes Transfer Integral
Instantons Partition Function
Droplets Fokker-Planck
"Langer did it all ten years ago" Smoluchowski
Spin Density Waves Ionic Conduction
Charge Density Waves Vacancy Motion
Mess Density Waves Soliton Lattices
Discommensurations Stacking Solitons
Dislocations The w-phase
Etch Pits Epitaxy
Polariton Solitons Frustration
Pendulum Chains Clouds of Erratic
Periodic Distortions Traj ectori es
Bifurcations Fixed Points
Instabil ities Cantor Sets
A Poem Devils' Staircase
</>-particles Textures
Schottky Defects Spin Triplet p-Waves
Operator Democracy Goldstone Modes
Universality Homotopy Theory
Fermions/Bosons Linear Textures
Electron-Hole Pairs Pointlike Solitons
Incommensurate to Vortex Lines
Commensurate Transitions Disgyrations
Kohn Anomalies Boojums
Precursors Snarks
Longitudinal Lock-in d-Solitons
Fermi Surface Nesting Twist Solitons
/Tj Superlattice Splay Solitons
2 kF and 4 kF Composite Solitons
Thermalized Solitons 1T'S and c's
Poling Flash Bulbs
I hope this list has reminded you of the amazing explosion which has taken
place in the physics community and condensed matter physics in particular.
Such diversity of applications of solitons and related objects is a sure
sign of the fundamental importance of a new pattern of thought or paradigm,
to use Jim Krumhansl's phrase. Once the uninitiated can bring himself to
revolt from his "1 inear" upbringing (via the harmonic oscillator and
Schrodinger's equation) and accept the fact that large-amplitude localized
objects can exist which do not spread in time, then he stands on the threshold
of developing that wonderfully simple and beautiful pattern of thought that
emphasizes the remarkable stability of such objects, their very natural use
as elementary excitations or as fundamental objects present in the ground
state, their coexistence in many cases with extended linearized solutions,
their striking particle-like behavior in the presence of perturbations, and
perhaps most important of all, their essential role in describing so many
physical phenomena that cannot be explained in any other way. Now that we
have all been exposed to~paradi~m, who among us will be able to ignore
the admonishment by G. B. WHITHAM [lJ in the last sentence of his book on
Linear and Nonlineat' Waves: " ... one should not always turn too quickly to
a search for the E."?
Now we turn to the question, "Where do solitons go from here?" Let me

339
preface my retreat from this question by remarking that no area of physics
seems to have been left untouched by solitons, and as we have witnessed here,
condensed matter physics is no exception. It seems folly to try to predict
the specific areas where solitons will crop up in the years ahead. Who
could have predicted three years ago the items on the above list? Indeed,
if such predictions could be made for three years hence, it would be a sure
sign that the renaissance of nonlinearity is complete. This symposium augurs
the contrary.
Nevertheless, we can be certain of some aspects of the soliton's future.
It won't be long before the soliton becomes as natural and familiar a tool
in nonlinear physics as the harmonic oscillator has become in linear physics.
Indeed, because of its richness of structure and ubiquity in applications,
the soliton will become one of the cornerstones of modern physical science.
For this reason, solitons will also find their way into the classroom and
textbooks as deliciously simple and beautiful examples of what nonlinear
physics has to offer and of what it requires.
Never before has a single concept brought together physicists of such
varied backgrounds and enabled them to communicate foreign ideas and appli-
cations using a common language. The student of solitonics has an unparalled
opportunity to venture from his own field of interest and rekindle his love
for all areas of physics. This opportunity is one of the major legacies of
this~~posium and others like it, and I have no doubt that solitons will
peri odi ca lly converge at future symposi a where we can 1earn where solitons,
have gone, and gain insights and intuition to further refine and enrich the
emerging pattern of thought.
In closing, I would like to ask the participants to join with me in asking
Chris Eilbeck and Al Scott to organize a conference in 1984 at Heriot-Watt
University in Edinburgh to commemorate the 150th anniversary of Scott-Russell 's
first observation of the soliton. I sincerely hope that they will be able
to commandeer a flat·prowed barge and some horses in order to recreate the
momentous event and I'm looking forward to joining all of you in a horseback
ride along that famous canal in Scotland to observe that "singular and
beautiful phenomenon".
References
1. G. B. Whitham, Linear and Nonlinear Waves (Wiley and Sons, New York,
1974) .

340
Index of Contributors

Aubry. S. 264 Gibbon. J.D. 44.297 Nelson, D.F. 187

Axe. J. D. 234 Newell, A.C. 52
Hanna. S. 158 Niez, J.J. 195
Bak. P. 216 Harrison, R.J. 183
Baratoff, A. 313 Hopfinger, A.J. 330 Rice, t4.J. 246
Ben Jacob, E. 317 Horovitz, B. 254
Bilz, H. 162 Schneider, T. 135,154,326
Bishop, A.R. 85 Imry, Y. 317 Scott, A.C. 301
Bishop. G.H.,Jr. 183 Sen. A. 71
Bruce. A.D. 116 Karney. C.F.F. 71 S1 uc kin. T. J . 330
BUttiker, M. 321 Kerr, W.C. 150 Sokoloff. J.B. 334
BUttner, H. 162 Kitchenside, P .W. 48,291 Steiner, M. 191
Bullough. R.K. 2,48,291 Kjems, J.K. 191 Stoll, E. 135,154,326
Klein, R. 158
Calogero. R. 68 Krumhansl, J .A. 22 Taylor. P.L. 330
Caudrey. P.J. 48,291 Kwok, T. 183
Chu, F.Y.F. 71 Theodorakopou los. N. 158
Lajerowi cz, J. 195 Thomas, H. 321
Degasperis, A. 68 Trullinger, S.E. 338
Dodd, R.K. 2 Lewanski, A.J. 330
Luther, A. 78 Wallace, D.J. 104
Eilbeck. J.C. 28 Widom, A. 334
Flytzanis. N. 166 Maki, K. 278
Friend, R. 199 Mason. A.L. 48 Yip. S. 183

341
 H.Haken

Synergetics
An Introduction

Nonequilibrium Phase Transitions and Self-Organi-

zation in Physics, Chemistry and Biology

2nd enlarged edition. 1978. 153 figures. Approx.

360 pages
(Springer Series in Synergetics, Volume I)
ISBN 3-540-08866-0

Synergetics, deals with profound and striking ana-

logies recently discovered between the self-orga-
nized behavior of seemingly quite different systems
in physics, chemistry, biology, sociology and other
fields. The cooperation of many subsystems such as
atoms, molecules, cells, animals, or humans may
produce spatial, temporal or functional structures.
Their spontaneous formation out of chaos is often
strongly reminiscent of phase transitions.

This book, written by the founder of synergetics,

provides an elementary introduction into the basic
concepts and mathematical tools. Numerous exer-
cises, figures and simple examples greatly facilitate
the understanding. The basic analogies are demon-
strated by various realistic examples from fluid dyna-
mics, lasers, mechanics, engineering, chemical and
biochemical systems, ecology, sociology and theo-
ries of evolution and morphogenesis.
Springer-Verlag
Berlin The second edition differs from the first by an ad-
ditional chapter on chaotic motion, a rapidly grow-
Heidelberg ing field, and by new sections on laser pulses and on
New York morphogenesis.
Springer Volume 81
G. Leibfried, N. Breuer

Tracts Point Defects in Metals I

Introduction to the Theory
• 1978. 138 figures, 22 tables. XIV, 342 pages

In ISBN 3-540-08375-8
Contents: Introduction and Survey. - HarmonicAapproximation

Modem and Linear Response (Green's Function) of an Arbitary

System. - Lattice Theory. - Continuum Theory. - Transition
from Lattice to Continuum Theory. - Statics and Dynamics of

Physics Simple Single Point Defects. - Scattering of Neutrons and

X-Rays by Crystals. - Probability, Distributions and Statistics;-
Properties of Crystals with Defects in Small Concentration. -
Ergebnisse der exakten Appendix.
~aturwissenschaften
Volume 82
Editor: G. Hohler
Associate Editor:
Electronic Structure of Noble Metals and
E.A Niekisch Polariton-Mediated Light Scattering
With contributions by B. Bendow, B. Lengeler
1978.42 figures, 20 tables. VI, 114 pages
ISBN 3-540-08814-8
Contents: B. Lengeler: de Haas-Van Alphen Studies ofthe
Electronic Structure of the Noble Metals and Their Dilute
Alloys. - B. Bendow: Polariton Theory of Resonance Raman
Scattering in Solids.

Volume 83
E. Arnaldi, S. Fubini, G. Furlan
Pion-Electroproduction
Electroproduction at Low Energy and Hadron Form Factors
1978. 47 figures. Approx. 180 pages
ISB~ 3-540-08998-5

Contents: Quantities ofPhysicaJ Interest - Theoretical

Approaches. - Main Features ofthe Experiments, Preliminary
Tests and Measurements. - Hadron Form Factors from Electro-
production. - Other Developments. - Appendices.

Volume 84
Collective Ion Acceleration
Springer-Verlag With contributions by C. L. Oison, U. Schumacher
1978. 63 figures. Approx. 240 pages
Berlin· ISBN 3-540-09066-5
Heidelberg Contents: U. Schumacher: Collective Ion Acceleration With
Electron Rings. - c.L. Olson: Collective Ion Acceleration With
New York Linear Electron Beams.