1: \documentclass[11pt]{article}
2: \usepackage{amsfonts,epsfig,latexsym}
3:
4: \topmargin=0in
5: \oddsidemargin=0.7cm
6: \textheight=8.5in
7: \textwidth=6in
8:
9: %Martin's newcommands: (perturbed in the direction of Paul's)
10: \newcommand{\R}{{\mathbb{R}}}
11: \newcommand{\Z}{{\mathbb{Z}}}
12: \newcommand{\N}{{\mathbb{N}}}
13: \newcommand{\C}{{\mathbb{C}}}
14: \newcommand{\beq}{\begin{equation}}
15: \newcommand{\eeq}{\end{equation}}
16: \newcommand{\ra}{\rightarrow}
17: \newcommand{\ds}{\displaystyle}
18: \newcommand{\cd}{\partial}
19: \newcommand{\wt}{\widetilde}
20: \newcommand{\wh}{\widehat}
21: \newcommand{\nvec}{{\bf n}}
22: \newcommand{\evec}{{\bf e}}
23: \newcommand{\uvec}{{\bf u}}
24: \newcommand{\xvec}{{\bf x}}
25: \newcommand{\zerovec}{\mbox{\boldmath{$0$}}}
26: \newcommand{\epsvec}{\mbox{\boldmath{$\varepsilon$}}}
27: \newcommand{\cs}{\cos\theta}
28: \newcommand{\sn}{\sin\theta}
29: \newcommand{\fl}{{\cal F}\ell}
30: \newcommand{\spec}{{\rm spec}\, }
31: \newcommand{\spn}{{\rm span}\, }
32: \newcommand{\diag}{{\rm diag}\, }
33: \newcommand{\re}{{\rm Re}\, }
34: \newcommand{\rot}{{\cal R}}
35:
36: \newtheorem{thm}{Theorem}
37:
38: %Paul's newcommands:
39: \newcommand{\be}{\begin{equation}}
40: \newcommand{\ee}{\end{equation}}
41: \newcommand{\bea}{\begin{eqnarray}}
42: \newcommand{\eea}{\end{eqnarray}}
43: \newcommand{\bean}{\begin{eqnarray*}}
44: \newcommand{\eean}{\end{eqnarray*}}
45: \renewcommand{\theequation}{\arabic{section}.\arabic{equation}}
46: \newcommand{\news}{\setcounter{equation}{0}\hskip 0.5cm}
47: \newcommand{\bn}{{\bf n}}
48: \newcommand{\vac}{{\bf e}_3}
49:
50: \begin{document}
51:
52: \title{Discrete Breathers in Anisotropic Ferromagnetic Spin Chains}
53: \author{J.M. Speight\thanks{Email : J.M.Speight@leeds.ac.uk} \\
54: Department of Pure Mathematics,
55: University of Leeds, \\ Leeds LS2 9JT, U.K. \\
56: \\
57: \\ P.M. Sutcliffe\thanks{Email : P.M.Sutcliffe@ukc.ac.uk} \\
58: Institute of Mathematics, University of Kent at Canterbury,\\
59: Canterbury, CT2 7NF, U.K.}
60: \date{}
61: \maketitle
62:
63:
64: \begin{abstract}
65: \noindent
66: We prove the existence of discrete breathers (time-periodic, spatially localized solutions)
67: in weakly coupled ferromagnetic spin chains with easy-axis anisotropy. Using numerical
68: methods we then investigate the continuation of discrete breather solutions as the intersite
69: coupling is increased. We find a band of frequencies for which the 1-site
70: breather continues all the way to the soliton solution in the continuum. There is a second
71: band, which abuts the first, in which the 1-site breather does not continue to the
72: soliton solution, but a certain multi-site breather does. This banded structure continues,
73: so that in each band there is a particular multi-site breather which continues to the soliton
74: solution. A detailed analysis is presented, including an exposition of how the bifurcation
75: pattern changes as a band is crossed. The linear stability of breathers is analyzed. It is
76: proved that 1-site breathers are stable at small coupling, provided a non-resonance condition
77: holds, and an extensive numerical stability analysis of 1-site and multisite breathers is
78: performed. The results show alternating bands of stability and instability as the coupling
79: increases.
80:
81: \end{abstract}
82: \newpage
83: \section{Introduction}\news
84:
85: Discrete breathers are time periodic, spatially localized solutions in
86: networks of coupled oscillators (including rotors and spins). They arise in
87: a variety of very general systems \cite{aubmac,SM} due to the interplay between nonlinear and
88: discrete effects, and therefore there is an enormous potential for their application
89: in many areas of physics, particularly condensed matter and biophysics, where physical
90: systems are often inherently discrete. Generally one finds that as the system is moved
91: closer to its continuum limit, by increasing a coupling constant in the theory,
92: there comes a point at which
93: a discrete breather solution no longer exists. This is to be expected since an
94: increase in the coupling constant results in an expansion of the phonon frequency band
95: and eventually this band captures a harmonic of the breather frequency. In fact, it is
96: often the case that the discrete breather cannot be continued even up to the point at
97: which the above resonance argument applies, and this is not well understood.
98:
99: In this paper we study the classical equations of motion for a ferromagnetic
100: spin chain with an easy-axis anisotropy. Some numerical studies of discrete breathers
101: (which are known as intrinsic localized modes in the condensed matter physics
102: literature) have been performed in the cases of easy-plane ferromagnets \cite{WMB} and
103: easy axis antiferromagnets \cite{laisie}. However,
104: both the perspective and the results of the current investigation are quite different.
105: We prove the existence of discrete breathers in the weakly coupled case by starting
106: from the anti-continuum limit (zero coupling constant) and applying an implicit
107: function theorem argument to the trivial 1-site breather. We then apply numerical
108: methods to investigate the continuation of this solution as the coupling constant
109: is increased, with the novel result that it continues for all values of the coupling.
110: This is explained by an examination of the spin waves.
111: Depending upon the breather frequency, the continuation results in either the soliton
112: solution of the continuum model or a trivial static solution in which all spins are
113: either aligned or anti-aligned with the vacuum. Similar results are found to apply
114: to multi-site breathers, which fit into an intricate pattern, and lead us to conjecture
115: that the continuum soliton solution (of any allowed frequency) can always be obtained
116: from the continuation of a particular form of multi-site breather. Evidence in support
117: of this conjecture is presented.
118: The numerical results are in some ways reminiscent of
119: those found \cite{eil1,dnls} for the discrete nonlinear Schr\"odinger (DNLS) system, but have
120: not previously been seen in spin chains.
121:
122: We then go on to perform a linear stability analysis of the breathers within the Krein theory
123: framework pioneered by Aubry \cite{aubstab}. We prove that nonresonant continued
124: 1-site breathers must be stable for sufficiently weak coupling, and numerically investigate
125: the stability of both 1-site and multisite breathers. The results suggest that those families
126: which tend to the continuum soliton experience regular repeating bands of instability of
127: diminishing strength as the coupling increases.
128:
129: During the refereeing process for this paper, a very interesting
130: paper by Flach, Zolotaryuk and Fleurov
131: on breathers in very similar systems appeared \cite{flazol}. The differences between their
132: results and ours are, briefly, as follows. They consider both easy-axis and easy-plane
133: anisotropy, but in the case where the exchange interaction is anisotropic also (our exchange
134: interaction maintains isotropy). Breather existence is proved by continuing from a limit in
135: which two of the exchange integrals vanish, but the third remains nonzero. Since the
136: continuation is for small values of the continuation parameters (the two small exchange
137: integrals) their existence result applies to strongly anisotropic exchange interaction, as
138: opposed to ours, which applies to isotropic exchange. More importantly, in this paper we
139: exploit the isotropic exchange to reduce the breather equations to a purely algebraic system,
140: which greatly simplifies the analysis in comparison with theirs.
141: Consequently our existence result requires none of the nonresonance hypotheses of Flach
142: et al's. Also we are able to build spatial localization of breathers directly into our
143: analysis, whereas Flach et al do not consider this issue at all (although standard results
144: of MacKay on exponential breather localization should apply to their system just as they do to
145: ours - see section \ref{sec-analytic}).
146: Similarly, we exploit the exchange isotropy to simplify the linear stability analysis,
147: again reducing it to a purely algebraic problem, a reduction not possible for the systems
148: analyzed in \cite{flazol}. We are able, therefore, to perform a very detailed numerical
149: existence and
150: stability analysis at minimal computational cost. To summarize, Flach et al prove a result
151: of rather more general applicability and physical relevance than ours,
152: though their context does not, strictly
153: speaking, include ours. We consider a somewhat more idealized system, but obtain correspondingly
154: stronger results.
155:
156: \section{Anisotropic ferromagnetic spin chains}\news
157: \label{defns}
158:
159: The classical formulation of a spin chain involves a three-component
160: unit vector ${\bf n}_i$, giving the spin at each lattice site $i\in\Z.$
161: The type of spin chain is defined by its Hamiltonian, which we take to be
162: \be
163: H=\sum_i\{\alpha(1-{\bf n}_i\cdot{\bf n}_{i+1})+\frac{A}{2}[1-({\bf n}_i\cdot {\bf e}_3)^2]\}.
164: \label{ham}\ee
165: Here $\alpha\ge 0$ is the coupling constant (exchange integral), which is positive
166: in the case of a ferromagnet, and $A>0$ is the anisotropy constant, which is also
167: positive since we wish to consider an easy-axis anisotropy,
168: with ${\bf e}_3=(0,0,1)$ being the easy-axis. In this case the
169: minimum of the Hamiltonian is zero, obtained by the vacuum configuration
170: $\bn_i=\pm \vac.$ In the following we shall choose $+\vac$ to be the vacuum
171: configuration and refer to spins which take the values $+\vac$ and $-\vac$
172: as being spin up and down respectively. Furthermore, by applying a scaling symmetry
173: we can, without loss of generality, set $A=1.$
174:
175: The equation of motion is obtained from the Hamiltonian as
176: \be
177: \dot\bn_i=-\bn_i\times\frac{\partial H}{\partial \bn_i}
178: \label{eom1}
179: \ee
180: where a dot denotes differentiation with respect to time.
181: Using (\ref{ham}) we obtain
182: \be
183: \dot\bn_i=\alpha\bn_i\times(\bn_{i+1}+\bn_{i-1})+(\bn_i\cdot\vac)(\bn_i\times\vac).
184: \label{eom2}
185: \ee
186: Discrete breather solutions have the form
187: \be
188: \bn_i(t)=(\sin\theta_i\cos\omega t,-\sin\theta_i\sin\omega t,\cos\theta_i)
189: \label{ansatz}
190: \ee
191: where $\omega$ is the frequency. Substituting this ansatz into the equation of
192: motion (\ref{eom2}) yields the following
193: nonlinear second order difference equation for the angles $\theta_i$,
194: \be
195: \alpha\{\cos\theta_i(\sin\theta_{i+1}+\sin\theta_{i-1})
196: -\sin\theta_i(\cos\theta_{i+1}+\cos\theta_{i-1})\}
197: =\sin\theta_i(\cos\theta_i-\omega).
198: \label{angeom}
199: \ee
200: Such a periodic solution may properly be called a discrete breather if it is
201: spatially localized, that is, $\lim_{i\ra\pm\infty}\theta_i=0$.
202:
203: In section \ref{sec-analytic} we shall prove the existence of discrete
204: breathers for $\alpha$ sufficiently small and
205: in section \ref{sec-numeric} we shall study them numerically for a range
206: of values of $\alpha.$
207:
208: Note that for all $\alpha$ there is a set of static solutions in which
209: each spin can independently be chosen to point either up or down, since
210: in each case we have that $\sin\theta_i=0$ for all $i,$ which clearly solves equation
211: (\ref{angeom}). These solutions will play an important role later in our discussion.
212:
213:
214: Finally, in this section, we address the continuum limit $\alpha\rightarrow\infty.$
215: Write $\alpha=1/h^2$ and regard $\bn_i$ as the value of a continuous function
216: $\bn(x)$ sampled at the lattice points $x=ih.$ Taking the continuum limit
217: $h\rightarrow 0$, the Hamiltonian (\ref{ham}) becomes
218: \be
219: H=\int\{ \frac{1}{2}\bn'\cdot\bn'+ \frac{1}{2}[1-({\bf n}_i\cdot {\bf e}_3)^2]\} \ dx
220: \label{hamc}\ee
221: and the equation of motion is
222: \be
223: \dot\bn=\bn\times\bn''+(\bn\cdot\vac)(\bn\times\vac).
224: \label{eom2c}
225: \ee
226: where prime denotes differentiation with respect to $x.$
227:
228: This partial differential equation has time periodic, exponentially localized solutions
229: known as magnetic solitons \cite{KIK}. At this point it is perhaps worth pointing out that
230: there is an unfortunate difference in nomenclature, in that in the discrete case
231: time periodic localized solutions are known as breathers, whereas in the continuum model
232: they are termed solitons. We shall continue to use the term soliton when referring to the
233: continuum limit, but the reader should be aware that it has exactly the same time
234: dependence as the discrete breather and moreover, as we shall see later, the soliton
235: can be obtained from the discrete breather in the continuum limit.
236:
237: Using the same form for the time dependence as in the discrete case (\ref{ansatz})
238: i.e.
239: \be
240: \bn(x,t)=(\sin\theta(x)\cos\omega t,-\sin\theta(x)\sin\omega t,\cos\theta(x))
241: \label{ansatzc}
242: \ee
243: equation (\ref{eom2c}) yields the second order ordinary differential equation
244: \be
245: \theta''=(\cos\theta-\omega)\sin\theta
246: \label{angeomc}
247: \ee
248: which, of course, is also obtained from the continuum limit of (\ref{angeom}).
249: The boundary conditions for a soliton solution, located at the origin,
250: are $\theta'(0)=\theta(\infty)=0.$ Equation (\ref{angeomc}) can be integrated
251: explicitly and the solution satisfying the correct boundary conditions is given
252: by
253: \be
254: \theta(x)=\cos^{-1}\{
255: \frac{2\omega}{1-(1-\omega)\tanh^2(x\sqrt{1-\omega})}-1\}.
256: \label{soliton}\ee
257: Since we require the soliton to be a smooth exponentially localized solution,
258: this formula shows that the frequency must be restricted to the range $\omega\in(0,1).$
259: There is a second type of soliton solution \cite{KIK} for which $\omega<0$,
260: but this has a different structure from (\ref{soliton}); in particular
261: $\bn=-\vac$ at the centre of the soliton for all $\omega.$ This negative frequency
262: soliton will not arise in our discussion, so in this paper when we refer to
263: a soliton we shall mean the solution (\ref{soliton}).
264:
265:
266: \section{Analytic results on discrete breathers}\news
267: \label{sec-analytic}
268: Following
269: the ``homoclinic orbit'' approach of Flach \cite{fla}, one would like to
270: prove directly the existence of solutions of (\ref{angeom}) with the correct
271: boundary behaviour, using techniques of dynamical systems theory.
272: Unfortunately, equation (\ref{angeom}) does not determine a well-defined
273: homeomorphism of the torus, $(\theta_{i-1},\theta_i)\mapsto (\theta_i,
274: \theta_{i+1})$, so the direct approach is not convenient here.
275:
276: However, we may still prove the existence of breathers in this system by
277: continuation of 1-site breathers from the decoupled limit ($\alpha=0$),
278: in the manner of Aubry and MacKay's work on oscillator networks
279: \cite{aubmac}.
280: Here we have an algebraic system rather than an infinite system of ODEs so the details
281: are considerably less technical (compare, for example, with
282: \cite{flazol}), and will be treated with corresponding brevity.
283: The idea is that when $\alpha=0$, $\omega\in(-1,1)$,
284: (\ref{angeom}) supports the almost trivial solution
285: \beq
286: \label{m4}
287: \theta^b_i=\left\{\begin{array}{cl}
288: 0 & i\neq 0 \\
289: \cos^{-1}\omega & i=0
290: \end{array}\right.
291: \eeq
292: in which every spin remains pointing up except one (whose
293: location we have chosen to be $i=0$) which precesses with frequency
294: $\omega$ around
295: a circle of fixed latitude. Note that there is no reason to assume that
296: $\omega>0$ as we had to in the continuum system. Keeping $\omega$ fixed,
297: the existence of breathers for all $\alpha$ sufficiently small will follow
298: from an implicit function theorem argument.
299:
300: To be precise, let $F:\ell_2\oplus\R\ra\ell_2$ such that
301: \beq
302: \label{m5}
303: [F(\theta,\alpha)]_i=(\cs_i-\omega)\sn_i-\alpha[\cs_i(\sn_{i+1}
304: +\sn_{i-1})-\sn_i(\cs_{i+1}+\cs_{i-1})],
305: \eeq
306: where $\ell_2$ is the Banach space of sequences $\theta:\Z\ra\R$ with finite
307: norm
308: \beq
309: \label{m6}
310: ||\theta||_{\ell_2}=\left[\sum_i\theta_i^2\right]^{\frac{1}{2}}.
311: \eeq
312: Note that $\theta\in\ell_2$ is a solution of (\ref{angeom}) at coupling $\alpha$
313: if and only if $F(\theta,\alpha)=0$. Note also that every $\theta\in\ell_2$
314: converges to $0$ as $i\ra\pm\infty$, so such a zero of $F(\cdot,\alpha)$ is
315: a discrete breather. For all $\omega\in(-1,1)$, $F(\theta^b,0)=0$, where
316: $\theta^b$ is the 1-site breather defined in (\ref{m4}), and the partial
317: differential of $F$ with respect to the $\ell_2$ factor at $(\theta,\alpha)=
318: (\theta^b,0)$ is easily seen to be an isomorphism
319: $DF_{(\theta^b,0)}:\ell_2\ra\ell_2$, that is, a bijection with bounded
320: inverse. In fact this linear map is diagonal:
321: \beq
322: \label{m7}
323: [DF_{(\theta^b,0)}\, \delta\theta]_i=\left\{
324: \begin{array}{rl}
325: (1-\omega)\delta\theta_i & i\neq 0 \\
326: -(1-\omega^2)\delta\theta_0 & i=0.
327: \end{array}
328: \right.
329: \eeq
330: Hence we may apply the implicit function theorem \cite{ift} to obtain the
331: following
332: %\vspace{0.5cm}
333: %\newline
334: %{\bf Theorem}
335: %{\it
336: \begin{thm}\label{exist}
337: For all $\omega\in(-1,1)$ there exist $\epsilon>0$ and a
338: $C^1$ map $[0,\epsilon)\ra\ell_2$, denoted $\alpha\mapsto \theta^\alpha$,
339: such that $\theta^\alpha$ is a frequency $\omega$, coupling $\alpha$ solution
340: of (\ref{angeom}) and $\theta^0=\theta^b$. The map $\alpha\mapsto\theta^\alpha$
341: is unique provided $\epsilon$ is chosen sufficiently small.
342: %}
343: %\vspace{0.5cm}
344: \end{thm}
345:
346: As noted above, since the continuation occurs in $\ell_2$, it has weak
347: spatial localization built in. This can be improved to exponential
348: spatial localization (that is, there exist $C,\lambda>1$ such that
349: $|\theta_i|<C\lambda^{-|i|}$) by applying some standard results of
350: Baesens and MacKay. The idea is to differentiate the continuation equation
351: $F(\theta^\alpha,\alpha)=0$ with respect to $\alpha$ to obtain an ODE
352: on $\ell_2$, namely,
353: \beq
354: \label{m8}
355: \frac{d\theta^\alpha}{d\alpha}=[DF_{(\theta^\alpha,\alpha)}]^{-1}
356: \Delta(\theta^\alpha),
357: \eeq
358: where $\Delta:\ell_2\ra\ell_2$ such that
359: \beq
360: \label{m9}
361: [\Delta(\theta)]_i=\cs_i(\sn_{i+1}
362: +\sn_{i-1})-\sn_i(\cs_{i+1}+\cs_{i-1}).
363: \eeq
364: The theorem above may be interpreted as asserting the local existence and
365: uniqueness of a solution to the initial value problem $\theta^0=\theta^b$
366: for (\ref{m8}). Now $\Delta(\theta^\alpha)$ is clearly exponentially
367: localized provided $\theta^\alpha$ is, as is
368: $[DF_{(\theta^\alpha,\alpha)}]^{-1}
369: \Delta(\theta^\alpha)$ by theorem 3 of \cite{baemac}. Thus one obtains
370: an exponential decay estimate for $d\theta^\alpha/d\alpha$ at $\alpha_0$
371: given a decay estimate for $\theta^{\alpha_0}$. This may be integrated to
372: show that the estimate for $\theta^{\alpha_0}$, if chosen correctly,
373: remains valid in a finite interval containing $\alpha_0$
374: (see the proof of theorem 2 of \cite{baemac}, for example).
375: It remains then to
376: note that the initial datum $\theta^b$ trivially satisfies {\em all}
377: exponential decay criteria (i.e.\ one can choose any $\lambda>1$),
378: so exponential localization persists for $\alpha$ sufficiently small.
379:
380: We may deduce the optimal exponent $\lambda$ for given $\omega$ and
381: $\alpha$ by linearizing equation (\ref{angeom}) about $\theta=0$:
382: \beq
383: \label{m9.1}
384: \delta\theta_{i+1}-\left[2+\frac{1-\omega}{\alpha}\right]\delta\theta_i
385: +\delta\theta_{i-1}=0.
386: \eeq
387: The general solution to (\ref{m9.1}) is $\delta\theta_i=A\lambda^i+
388: B\lambda^{-i}$ where
389: \beq
390: \label{m9.2}
391: \lambda=1+\frac{1-\omega}{2\alpha}+\left[\frac{1-\omega}{\alpha}+
392: \left(\frac{1-\omega}{2\alpha}\right)^2\right]^\frac{1}{2}.
393: \eeq
394: It is this type of exponential decay, $|\theta_i|\sim\lambda^{-|i|}$, which
395: we will encounter in our numerical analysis.
396:
397: Another interesting piece of information can be deduced from equation
398: (\ref{m8}), namely the tangent vector to the continuation curve in $\ell_2$,
399: that is, the direction of continuation away from $\alpha=0$. Substituting
400: (\ref{m4}) and (\ref{m7}) into (\ref{m8}) at $\alpha=0$ yields
401: \beq
402: \label{m10}
403: \left[\left.\frac{d\theta^\alpha}{d\alpha}\right|_{\alpha=0}\right]_i=
404: \left\{\begin{array}{cc}
405: 0 & |i|>1 \\
406: \frac{1+\omega}{\sqrt{1-\omega^2}} & |i|=1 \\
407: \frac{2}{\sqrt{1-\omega^2}} & i=0.
408: \end{array}\right.
409: \eeq
410: So the continuation initially proceeds by pulling the central spin, $\bn_0$, further
411: away from the vacuum $\vac$, and pulling its nearest neighbours, $\bn_{\pm 1}$ away from
412: $\vac$ in the same direction (as $\nvec_0$ and each other), while leaving
413: all other spins fixed at $\evec_3$, to first order.
414:
415: We have proved the existence of a continuation of the one site breather
416: $\theta^b$, but the same argument can be applied to any zero of
417: $F(\cdot,0):\ell_2\ra\ell_2$ to prove existence of more general discrete
418: breathers at small $\alpha$. Clearly, if $F(\theta,0)=0$ then
419: $\theta_i\in\pi\Z\cup\cos^{-1}\{\omega\}$ for all $i$, and given that
420: $\theta\in\ell_2$, $\theta_i=0$ for all $|i|$ sufficiently large. So
421: one can construct frequency $\omega$ solutions at $\alpha=0$ like
422: \beq
423: \label{m11}
424: \wt{\theta}=(\ldots,0,0,\cos^{-1}\omega,\pi,\pi,\pi,\cos^{-1}\omega,0,0,
425: \ldots)
426: \eeq
427: for example. It is easy to check that $DF_{(\wt{\theta},0)}$ is an
428: isomorphism, so an analogous result to the theorem above applies here
429: also. Of particular relevance to the numerical work in section \ref{sec-numeric}
430: will be those periodic solutions obtained by continuing $\alpha=0$ solutions
431: with even reflection symmetry about
432: $i=0$, that is, $\theta_{-i}\equiv\theta_i$. Such a solution may be
433: specified by a finite coding sequence \cite{aubseq} representing the
434: values of $\theta_i$ for $i=0,1,2,\ldots$ by $+$ if $\theta_i=0$,
435: $o$ if $\theta_i=\cos^{-1}\omega$ and $-$ if $\theta_i=\pi$, with the
436: convention that the last symbol represents the constant tail of the
437: sequence. Of course, since $\theta\in\ell_2$, each coding sequence must end
438: in $+$, so the last symbol is redundant. Nevertheless, we include it in
439: order to make the notation more suggestive. The symbols themselves are
440: meant to represent the directions of the central spins:
441: up is $+$, down $-$, while $o$ is somewhere ($\omega$
442: dependent) in between. We shall refer to those breathers obtained by
443: continuing such a solution by the same coding sequence. Hence
444: $\theta^\alpha$, the continuation of $\theta^b$ is a $(o+)$-breather, while
445: $\wt{\theta}$ and its continuation are $(--o+)$-breathers, for example.
446:
447: \section{Numerical results on discrete breathers}\news
448: \label{sec-numeric}
449: In this section we present the results of an extensive numerical investigation
450: of the continuation of 1-site and various multi-site breathers from the
451: anti-continuum limit. To construct numerical solutions we must first truncate
452: to a finite number of lattice sites, which we implement by fixing vacuum boundary
453: conditions, $\theta_i=0$ for $i=\pm N$, and restricting to the interior of the
454: lattice $|i|<N.$ Furthermore, the solutions we consider are symmetric about the
455: central lattice site, that is, $\theta_i=\theta_{-i}$, so only the sites $i\ge 0$
456: need be considered, with an appropriate modification at the central site $i=0$.
457: Explicitly, the task of finding a numerical solution reduces
458: to finding a zero of the following $(N+1)$-component vector
459: \be
460: \label{f}
461: F_i=
462: \left\{
463: \begin{array}{ll}
464: \alpha[\cs_i(\sn_{i+1}+\sn_{i-1})-\sn_i(\cs_{i+1}+\cs_{i-1})]+\sn_i(\omega-\cs_i)
465: & 0<i<N \\
466: 2\alpha[\cs_0\sn_1-\sn_0\cs_1]+\sn_0(\omega-\cs_0) & \ i=0 \\
467: \theta_N & \ i=N
468: \end{array}\right.
469: \end{equation}
470: as a function of the $(N+1)$ values $\theta_i$, $i=0,..,N.$
471:
472: To find a zero of the vector $F_i$, for a given value of $\alpha$, we begin with
473: the value $\alpha=0$, where we have the explicit solution corresponding to the
474: trivial 1-site breather (\ref{m4}) (and later we use other multi-site
475: breathers).
476: We then increase $\alpha$ by a small amount and use a Newton-Raphson scheme
477: to converge to the new solution. This process is repeated until the desired
478: value of $\alpha$
479: has been obtained. During the calculation the determinant of the Jacobian matrix
480: $J_{ij}={\partial F_i}/{\partial \theta_j}$ is monitored to ensure that it is
481: non-zero,
482: as a check that the scheme is not accidentally jumping to a different solution
483: branch.
484: The results presented below were obtained from the value $N=20$, though different size
485: lattices were checked to confirm that, providing $N$ is sufficiently large, the results
486: are not sensitive to the number of sites.
487:
488: \begin{figure}[ht]
489: \centerline{
490: \epsfxsize=7cm
491: \epsfbox[50 50 407 301]{figA.ps}}
492: \caption{Plots of $\cs_0$ and $\cs_1$ for the continuation of the $(o+)$-breather
493: with $\omega=0.2$, from $\alpha=0$ to $\alpha=0.5$.}
494: \label{fig1}
495: \end{figure}
496:
497: In Figure 1 we display the results of the continuation of the $(o+)$-breather,
498: with frequency $\omega=0.2$, from $\alpha=0$ to $\alpha=0.5.$ Presented are
499: plots of $\cs_0$ and $\cs_1$, that is, the third component of the spin
500: for the central and next to central sites. From this plot it is clear that the
501: $(o+)$-breather continues until it joins the static $(-+)$ solution. This is true
502: for all
503: frequencies in the band $\omega\in(-1,0.424]$, where we have computed the edge
504: of the band to within the numerical accuracy given. It is not surprising that
505: for $\omega$ close to $-1$ the $(o+)$-breather continues to the static $(-+)$
506: solution
507: since for $\alpha=0$ the $(o+)$-breather tends to the $(-+)$ solution as
508: $\omega\rightarrow -1.$
509:
510: \begin{figure}[ht]
511: \centerline{
512: \epsfxsize=7cm
513: \epsfbox[50 50 407 301]{figB.ps}}
514: \caption{Plots of $\cs_0$ and $\cs_1$ for the continuation of the $(o+)$-breather
515: (solid curves) with $\omega=0.5$, from $\alpha=0$ to $\alpha=2$.
516: The dashed curves are the corresponding quantities for the soliton solution.}
517: \label{fig2}
518:
519: \end{figure}
520: \begin{figure}[ht]
521: \centerline{
522: \epsfxsize=7cm
523: \epsfbox[50 50 407 301]{figC.ps}}
524: \caption{
525: $\cs_i$ for all the lattice sites of the $\omega=0.5$ $(o+)$-breather
526: with $\alpha=2$ (diamonds). The dashed curve is the corresponding soliton solution.
527: }
528: \label{fig3}
529:
530: \end{figure}
531: In Figure 2 we present the results of a similar calculation, but this time the
532: initial $(o+)$-breather has a frequency $\omega=0.5.$ The values of $\cs_0$ and
533: $\cs_1$ are shown as solid curves and the dashed curves represent the same
534: quantities for the continuum soliton solution (\ref{soliton}) sampled at lattice
535: sites $x=0$ and $x=h$, where $h=1/\sqrt{\alpha}.$ As the coupling $\alpha$
536: is increased, the values at these sites tend towards those of the continuum
537: soliton solution, demonstrating that the $(o+)$-breather approaches the
538: soliton solution in the continuum limit. As further evidence, in Figure 3 we plot
539: $\cs_i$ for all lattice sites, for the value $\alpha=2$, superimposed with the soliton
540: solution given in (\ref{soliton}), with $x=i/\sqrt{2}.$ As a final check, starting
541: with initial conditions obtained from sampling the soliton solution at the lattice
542: sites and continuing backwards to $\alpha=0$, we find that we recover the $(o+)$-breather.
543: We have also verified that the decay rate of the $(o+)$-breather shown in
544: Figure 3 fits extremely well to the derived exponential decay constant given in (\ref{m9.2}).
545:
546: Similar results apply for all frequencies in the band $\omega\in(0.424,1)$,
547: with the $(o+)$-breather continuing to the soliton solution, although the region in
548: which $\omega$ is close to 1 is numerically inaccessible within the present scheme
549: since the soliton has a relatively slow spatial decay in this region.
550:
551: Just below the interface of the two bands, that is for $0\le 0.424-\omega\ll 1$,
552: there is a more complicated bifurcation structure between the $(o+)$-breather
553: and the static $(-+)$ solution than in the bulk of the lower band $(-1,0.424].$
554: To examine this region requires a slightly more sophisticated numerical approach,
555: as we now describe, by illustrating the case $\omega=0.42.$
556:
557: \begin{figure}[ht]
558: \centerline{
559: \epsfxsize=7cm
560: \epsfbox[50 50 407 301]{figD.ps}}
561: \caption{
562: $\cs_0$ for the continuation of the $(o+)$-breather (dashed curve), the
563: static $(-+)$-solution (solid line), and the perturbed $(-+)$-solution
564: (dotted curve).
565: }
566: \label{fig4}
567:
568: \end{figure}
569: \begin{figure}[ht]
570: \centerline{
571: \epsfxsize=7cm
572: \epsfbox[150 230 470 550]{table.ps}}
573: \caption{
574: A table displaying the results, as $\omega$ is varied, of continuing
575: breathers of type $(o+)$,$(-o+)$ and $(--o+).$ The symbol at the foot
576: of each column indicates which breather is being continued and the
577: symbol inside the column represents the end result of the continuation
578: at that frequency.
579: }
580: \label{fig5}
581:
582: \end{figure}
583: \begin{figure}[ht]
584: \centerline{
585: \epsfxsize=12cm
586: \epsfbox[25 270 580 510]{bifurcation.ps}}
587: \caption{
588: A schematic to indicate how the bifurcation pattern
589: between the $(-+)$ solution and the $(o+)$ and $(-o+)$
590: breathers varies with $\omega.$
591: }
592: \label{fig6}
593:
594: \end{figure}
595: For $\omega=0.42$ the continuation of the $(o+)$-breather is presented in Figure 4,
596: where the dashed line represents the value of $\cs_0.$ The continuation fails at
597: $\alpha=\alpha^*=0.4085$, where the determinant of the Jacobian matrix $J$ is zero.
598: We expect that this solution still bifurcates from the static $(-+)$ solution,
599: but that it first turns around in $\alpha$ space. To confirm this expectation
600: we need to find the value $\alpha=\widehat\alpha$, at which the bifurcation of the $(-+)$
601: solution takes place, and compute the tangent vector in the direction of this
602: bifurcation. The Jacobian of the $(-+)$ solution is the tridiagonal matrix
603: \be
604: J=\pmatrix{
605: 2\alpha-\omega-1 & -2\alpha & & & & \cr
606: -\alpha & \omega-1 & \alpha & & & \cr
607: & \alpha & -2\alpha+\omega-1 & \alpha & & \cr
608: & & \ldots & \ldots & \ldots & \cr
609: & & & \alpha & -2\alpha+\omega-1 & \alpha \cr
610: & & & &0 & 1\cr
611: }\ee
612: Computing the determinant of this matrix (for $N=20$) and setting $\omega=0.42$, we find that
613: $\mbox{det}\,J$ has only one real and positive root, which occurs at
614: $\alpha=\widehat\alpha=0.3858.$ Note that this calculation provides the first check on
615: our bifurcation assumption since we require that $\widehat\alpha<\alpha^*$, which is indeed
616: true. Substituting $\alpha=\widehat\alpha$ into $J$ we then compute the eigenvector,
617: $\delta\theta$, corresponding to the zero eigenvalue, normalized so that its norm
618: $|\delta\theta|^2=\epsilon\ll 1.$ Finally, we take the $(-+)$ solution, in which $\theta_0=\pi$ and
619: $\theta_i=0$ for $i>0$, and add the perturbation $\delta\theta$, to create an initial condition
620: which we then iterate to converge to a solution
621: for a value of $\alpha$ obtained by incrementing $\widehat\alpha$ by a small amount.
622: Following this solution for increasing $\alpha$ we find that it continues until $\alpha=\alpha^*$,
623: where it meets the end point of the continued $(o+)$-breather. This is illustrated in
624: Figure 4, where we plot $\cs_0$ (dotted curve) for the solution branch obtained by
625: perturbation of the static $(-+)$ solution, as just described. Following the end point of this
626: solution branch backwards in $\alpha$ confirms that it joins the static $(-+)$ solution at
627: $\alpha=\widehat\alpha.$
628:
629: To summarize, the results of the continuation of the $(o+)$-breather, as $\omega$ is varied,
630: are represented in the first column of Figure 5, and presented schematically in the upper
631: half of Figure 6. For $\omega\in(0.424,1)$ the $(o+)$-breather continues to the soliton,
632: for $\omega$ outside this range, but close to $0.424$, it turns back and joins the
633: static $(-+)$ solution, and in the bulk of the band $(-1,0.424]$ it joins the
634: static $(-+)$ solution without turning back.
635:
636: \begin{figure}[ht]
637: \centerline{
638: \epsfxsize=7cm
639: \epsfbox[50 50 407 301]{figE.ps}}
640: \caption{
641: $\cs_0,\cs_1,\cs_2$ for the continuation of the $\omega=-0.2$
642: $(-o+)$-breather for $\alpha\in[0,0.8].$
643: }
644: \label{fig7}
645:
646: \end{figure}
647: For $\omega$ above $0.424$ there is no longer a bifurcation of the static $(-+)$
648: solution with the $(o+)$-breather, but to fill in the full bifurcation pattern
649: of the $(-+)$ solution we must now consider the continuation of the $(-o+)$-breather.
650: Note that now the start point of the continuation, $\alpha=0$, is already a multi-site breather,
651: since we have applied reflection symmetric conditions, $\theta_i=\theta_{-i}$,
652: and the breathing site is not located at $i=0.$
653:
654: \begin{figure}[ht]
655: \centerline{
656: \epsfxsize=7cm
657: \epsfbox[50 50 407 301]{figF.ps}}
658: \caption{
659: $\cs_0,\cs_1,\cs_2$ (solid curves) for the continuation of the $\omega=0.2$
660: $(-o+)$-breather for $\alpha\in[0,2].$
661: The dashed curves are the corresponding quantities for the soliton solution.
662: }
663: \label{fig8}
664:
665: \end{figure}
666: \begin{figure}[ht]
667: \centerline{
668: \epsfxsize=7cm
669: \epsfbox[50 50 407 301]{figG.ps}}
670: \caption{
671: $\cs_0,\cs_1$ for the continuation of the $\omega=0.55$
672: $(-o+)$-breather for $\alpha\in[0,0.5].$
673: }
674: \label{fig9}
675:
676: \end{figure}
677: In Figure 7 we plot $\cs_i$ for the first three sites, $i=0,1,2$, for the continuation
678: of the $(-o+)$-breather with frequency $\omega=-0.2.$ Clearly, for this frequency, the
679: $(-o+)$-breather continues to the static $(--+)$ solution, and this is true for all
680: $\omega\in(-1,0.133).$ From a similar earlier discussion, it follows that this result
681: is to be expected, at least for $\omega$ near $-1.$
682: For $\omega\in[0.133,0.424]$ the $(-o+)$-breather approaches the soliton
683: solution in the continuum limit. As an example, for $\omega=0.2$, we present in
684: Figure 8, a plot of $\cs_i$ for the first three sites (solid curves) and also
685: the corresponding quantities for the sampled soliton solution (dashed curves).
686: For $\omega\in(0.424,1)$ the $(-o+)$-breather continues to the static $(-+)$ solution,
687: as demonstrated in Figure 9 for $\omega=0.55.$ These results are summarized in the
688: second column of Figure 5.
689:
690: It is, of course, no accident that the frequency ($\omega=0.424$) at which the $(o+)$-breather
691: fails to continue to the soliton is precisely that at which the $(-o+)$-breather begins
692: to continue to the soliton. This can be understood by completing the bifurcation pattern
693: of the $(-+)$ solution, using the above results on the continuation of the
694: $(-o+)$-breather, to fill in the bottom half of the schematic in Figure 6.
695: As presented in the schematic, there is always a bifurcation of the $(-+)$ solution,
696: but it switches over from the $(-o+)$-breather to the $(o+)$-breather as the
697: frequency is decreased to the critical value $\omega=0.424.$ At this point the
698: $(o+)$-breather no longer continues to the soliton, but the $(-o+)$-breather now does.
699: We have verified this structure with a number of detailed further calculations, for example,
700: we have confirmed the turning back of the continuation of the $(-o+)$-breather by
701: perturbing the $(-+)$ solution as described earlier.
702:
703: The lower edge of the band $\omega\in[0.133,0.424]$ can also be understood in a similar
704: fashion by an analysis of the bifurcation pattern of the $(--+)$ solution, which requires
705: computations of the continuation of the $(--o+)$-breather. These calculations have been
706: performed and the results are summarized in the third column of Figure 5.
707: Again there is a soliton band, which begins at the frequency $\omega=0.045$,
708: and ends at the start of the $(-o+)$-breather soliton band where
709: $\omega=0.133.$ Below the soliton band the $(--o+)$-breather continues to the
710: $(---+)$ solution and above the soliton band it continues to the $(--+)$ solution.
711: Computations confirm that the bifurcation pattern of the $(--+)$ solution
712: is analogous to that of the $(-+)$ solution depicted in Figure 6, with an extra
713: $-$ sign inserted into the coding sequence of each solution together with a shift in the frequencies.
714:
715: Given the above results we are naturally led to the conjecture that the pattern
716: of soliton bands continues, with each band covering a smaller range of frequencies,
717: but such that for any frequency $\omega\in(0,1)$ there is a discrete breather solution
718: whose continuation from $\alpha=0$ tends to the soliton solution
719: in the continuum limit $\alpha\rightarrow\infty.$
720:
721: Generally, as mentioned in the introduction, discrete breathers in networks of coupled oscillators
722: fail to continue beyond a certain coupling due to resonance with phonons. In the system
723: considered here, the role of phonons is played by spin waves,
724: \beq
725: \delta\nvec_i(t)=\left(\begin{array}{c}
726: \cos(ki-\omega_p t) \\ \sin(ki-\omega_p t) \\ 0
727: \end{array}\right),
728: \eeq
729: which are travelling wave solutions of (\ref{eom2}) linearized about the vacuum
730: $\nvec=\evec_3$. These have dispersion relation
731: \be
732: \omega_p=1+4\alpha\sin^2(k/2),
733: \ee
734: so the spin waves form a frequency band with $\omega_p\in[1,1+4\alpha].$
735: Note that the maximal frequency spin wave ($k=\pi$) is standing, hence of the form
736: (\ref{ansatz}) with frequency $\omega_p=1+4\alpha$. It may equally well be regarded as
737: periodic of frequency $\omega_p/n$ for any $n\in\Z^+$. Mathematically, a bifurcation in
738: the continuation of breathers may only occur where $DF$ acquires nontrivial kernel, that is
739: the equation of ``motion'' (\ref{angeom}) linearized about the breather supports a nonzero
740: solution. Of course phonons, being linearized solutions about the {\em vacuum},
741: never lie in
742: ${\rm ker}\, DF$ strictly speaking.
743: Nevertheless, since the breather approaches the vacuum exponentially fast as
744: $|i|\ra\infty$, it is generally accepted that the existence of a standing phonon of
745: frequency $\omega_p=n\omega$ generically implies nontriviality of ${\rm ker}\, DF$, that
746: is, the phonon is close to a tangent vector in ${\rm ker}\, DF$, approaching it
747: asymptotically as $|i|\ra\infty$. In the present case, the standing spin wave lies outside
748: $\ell_2$, so cannot be close to ${\rm ker}\, DF$, and hence cannot cause a bifurcation.
749: In practice this technical point is irrelevant: the numerics are performed on
750: a finite chain where such distinctions are impossible to make. Why then do the standing
751: spin waves cause no bifurcations in the numerical continuation? The reason is that our
752: map $F$ contains explicit parametric dependence on $\omega$, so only the standing spin
753: wave of frequency $\omega_p=\omega$ is in the kernel of $DF$ at the vacuum. Higher
754: harmonics, $\omega_p=n\omega$ are {\em not} linearized
755: solutions of our
756: ansatz derived equation (\ref{angeom}), and so cannot cause bifurcations.
757: This is entirely consistent with physical intuition. These breathers, like those
758: in the DNLS system, are monochromatic: their time dependence contains no higher harmonics so one
759: does not expect them to resonate with the spin waves. It is precisely the imposition of
760: a monochromatic ansatz (\ref{ansatz}) which introduced the parametric $\omega$ dependence
761: discussed above.
762:
763: \section{Linear stability}\news
764: \label{sec-stability}
765:
766: Linear stability of breathers in networks of anharmonic oscillators was studied by
767: Aubry in \cite{aubstab}. Despite the many superficial differences between spin chains and
768: oscillator networks, we will find that the analytic framework developed by Aubry readily
769: adapts to this new setting, yielding similar results. In particular, we will prove that
770: $(o+)$-breathers of all frequencies $\omega\in(-1,0)\cup(0,1)$
771: must be linearly stable for sufficiently small $\alpha$ provided $\omega^{-1}\notin\Z$.
772: Just how small $\alpha$ must be, and the stability properties of the various other
773: breather types will be investigated numerically.
774:
775: As in \cite{aubstab}, it is technically convenient to truncate the lattice to (large but)
776: finite size $N$ (the results will be independent of $N$).
777: Existence of discrete breathers in such a system follows from an identical argument to that of
778: theorem \ref{exist}, but with $\ell_2$ replaced by $\R^N$ with its usual norm.
779: Equation (\ref{eom2}) along
780: with fixed endpoints $\nvec_0=\nvec_{N+1}=0$ then define a flow on phase space
781: $M=(S^2)^N$. Crucial to the analysis is the fact that this flow is Hamiltonian with
782: respect to the natural symplectic structure on $M$, namely
783: \beq
784: \label{sympform}
785: \Omega=\sum_i\Omega_i
786: \eeq
787: where $\Omega_i$ is the area form on the $i$-th 2-sphere. The Hamiltonian
788: $H:M\ra\R$ is the truncation of (\ref{ham}), again with fixed endpoints. It follows that
789: the flow is symplectomorphic.
790:
791: A period $T$ solution of the system is a fixed point of the return map $P_T:M\ra M$,
792: $P_T:n(0)\mapsto n(T)$. Such a solution is said to be linearly stable if the
793: spectrum of its associated
794: Floquet map $\fl=(dP_T)_{n(0)}:T_{n(0)}M\ra T_{n(T)=n(0)}M$ lies within the closed unit
795: disk $D=\{z\in\C:|z|\leq 1\}$. Recall that $\fl:\delta n(0)\mapsto\delta n(T)$ where $\delta n(t)$ is
796: the solution of the linearization of (\ref{eom2}) about the solution $n(t)$, explicitly,
797: \beq
798: \label{leom}
799: \delta\dot{\nvec}_i=\alpha[\nvec_i\times(\delta\nvec_{i+1}+\delta\nvec_{i-1})+\delta\nvec_i
800: \times(\nvec_{i+1}+\nvec_{i-1})]+(\delta\nvec_i\cdot\evec_3)\nvec_i\times\evec_3.
801: \eeq
802: Since $P_T$ is a symplectomorphism, $\fl$ is a symplectic map, that is
803: \beq
804: \Omega(\fl\, \delta n,\fl\, \delta n')\equiv\Omega(\delta n,\delta n').
805: \eeq
806: It follows that if $\lambda\in\spec\fl$, so are $\bar{\lambda}$, $1/\lambda$ and
807: $1/\bar{\lambda}$. Hence the solution is linearly stable if and only if $\spec\fl$ lies
808: on the unit circle $\cd D=\{z\in\C:|z|=1\}$.
809:
810: It is straightforward to construct $\fl$ explicitly for all the breathers considered
811: in sections \ref{sec-analytic} and \ref{sec-numeric}
812: in the uncoupled limit, $\alpha=0$. For each $\nvec_i(0)
813: =(\sin\theta_i,0,\cos\theta_i)$ define the ordered orthonormal basis
814: \beq
815: \label{basis}
816: (\epsvec_i=(\cos\theta_i,0,\sin\theta_i), \evec_2)
817: \eeq
818: for $T_{\nvec_i(0)}S^2$, so that
819: $T_{n(0)}M=\bigoplus_{i}\spn\langle\epsvec_i,\evec_2\rangle.$ Relative to this basis,
820: the symplectic form has the usual block matrix expression, namely
821: \beq
822: \Omega=\diag(\ldots,R(\frac{\pi}{2}), R(\frac{\pi}{2}),\ldots)
823: \eeq
824: where $R(\psi)$ is the $SO(2)$ matrix which performs a clockwise rotation through angle
825: $\psi$, explicitly,
826: \beq
827: \label{Rdef}
828: R(\psi)=\left(\begin{array}{cc}
829: \cos \psi & \sin \psi \\
830: -\sin \psi & \cos \psi \end{array}\right).
831: \eeq
832: The
833: Floquet matrix of the frequency $\omega$ $(o+)$-breather is
834: \beq
835: \label{flo+}
836: \fl=\diag(\ldots,R(T),R(T),S,R(T),R(T),\ldots)
837: \eeq
838: where $T=2\pi/\omega$ and $S$ is the matrix
839: \beq
840: S=\left(\begin{array}{cc}
841: 1 & 0 \\ (1-\omega^2)T & 1 \end{array}\right).
842: \eeq
843: Similarly, the $(-o+)$-breather has Floquet matrix
844: \beq
845: \label{fl-o+}
846: \fl=\diag(\ldots,R(T),R(T),S,R(T),S,R(T),R(T),\ldots)
847: \eeq
848: and so on. Clearly all these matrices have $\spec\fl=\{e^{\pm iT},1\}\subset\cd D$ so the
849: uncoupled breathers of all types are linearly stable.
850:
851: As $\alpha$ increases from 0, theorem \ref{exist} guarantees the existence of a continuous
852: family of symplectic maps $\fl_\alpha$ (recall that $\alpha\mapsto\theta_\alpha$ is
853: $C^1$) whose eigenvalues depend continuously on $\alpha$. Do these eigenvalues remain on
854: $\cd D$? Clearly only coincident pairs may leave $\cd D$ in tandem, by the symmetry
855: properties of $\spec\fl$. Krein theory \cite{aubstab} states that an eigenvalue
856: $\lambda\in \cd D$ may bifurcate off the unit circle only if its associated Krein
857: signature is indefinite. This signature is defined as follows. Let $E_\lambda\subset
858: \C^{2N}$ be the $\lambda$ eigenspace of $\fl$ and $V_\lambda:=\re(E_\lambda\oplus
859: E_{1/\lambda})\subset\R^{2N}$. Then on $V_\lambda$ one defines a bilinear form
860: $Q_\lambda:V_\lambda\oplus V_\lambda\ra\R$ such that
861: \beq
862: Q_\lambda(\xi,\eta):=\Omega(\xi,\fl\, \eta).
863: \eeq
864: The Krein signature of $\lambda$ is the signature of this bilinear form. Hence, $\lambda$
865: may bifurcate off $\cd D$ only if $Q_\lambda$ is indefinite, that is, the diagonal map
866: $\xi\mapsto Q_\lambda(\xi,\xi)$ is neither positive definite nor negative definite on
867: $V_\lambda\backslash\{0\}$.
868:
869: Applying this theory to (\ref{flo+}) one sees that $V_{e^{iT}}\equiv V_{e^{-iT}}\cong
870: \R^{2N-2}$, $V_1\cong\R^2$,
871: \beq
872: Q_{e^{iT}}(\xi,\xi)=-(\sin T)\, \xi^{T}\xi,\quad
873: Q_1(\xi,\xi)=(1-\omega^2)\xi_1^2.
874: \eeq
875: Provided $\sin T\neq 0$, i.e. $\omega^{-1}\notin\Z$, $Q_{e^{iT}}$ is definite and hence
876: all the eigenvalues $e^{\pm iT}$ must remain on $\cd D$, at least for sufficiently small
877: $\alpha$. On the other hand, $Q_1$ is only positive {\em semi}-definite, so no such
878: constraint applies to the double eigenvalue $\lambda=1$. Note however that for any
879: nonstatic period $T$ solution $n(t)$ of an autonomous dynamical system, $\dot{n}(0)$ is
880: an eigenvector of $\fl$ with $\fl\, \dot{n}(0)=\dot{n}(0)$ by time translation invariance.
881: It follows that one copy of the eigenvalue $\lambda=1$ is fixed for all $\alpha$. The
882: second is also fixed for sufficiently small $\alpha$ since to move it would have to
883: (at least) double up due to the symmetry properties of $\spec\fl$. Hence, we have proved:
884:
885: \begin{thm} \label{stable}
886: For each $N\in\Z^+$ and
887: for all $\omega\in (-1,1)\backslash\{0\}$ such that
888: $\omega^{-1}\notin\Z$, there exists $\wt{\alpha}(\omega)>0$ such that for all $\alpha\in
889: [0,\wt{\alpha}(\omega))$ the continued $(o+)$-breathers, whose existence on the $N$ site
890: lattice is
891: guaranteed by theorem \ref{exist}, are linearly stable.
892: \end{thm}
893:
894: Unfortunately, this argument cannot be applied to multisite breathers such as $(-o+)$.
895: This is because $V_1$ is now (at least) 4 dimensional and $Q_1$ remains only
896: semidefinite. Time translation symmetry is not sufficient to fix all 4 copies of $\lambda
897: =1$ under perturbation of $\alpha$, so while the $\alpha=0$ $(-o+)$-, $(--o+)$- etc.\,
898: breathers are all stable, instability may develop for arbitrarily small $\alpha$. Note,
899: however, that the Krein criterion is necessary but {\em not sufficient}: it does not
900: guarantee instability of $\alpha>0$ multisite breathers, nor of $(o+)$-breathers with
901: $\omega^{-1}\in\Z$. To investigate these issues, we must resort once again to numerical
902: analysis.
903:
904: One may vastly simplify the task of numerically constructing the Floquet matrix
905: $\fl$ for this problem by transforming to a co-rotating coordinate system. Let
906: $\rot:\R\ra SO(3)$ such that
907: \beq
908: \rot(t)=\diag(R(\omega t),1)
909: \eeq
910: where $R(t)$ is the the $SO(2)$ matrix defined in equation (\ref{Rdef}), and
911: define new variables $\uvec_i(t)$ such that
912: \beq
913: \nvec_i(t)=\rot(t)\uvec_i(t).
914: \eeq
915: Then the equations of motion become
916: \beq
917: \label{eom3}
918: \dot{\uvec_i}=\alpha\uvec_i\times(\uvec_{i+1}+\uvec_{i-1})+(\uvec_i\cdot\vac)(\uvec_i\times\vac)
919: -{\cal A}\uvec_i
920: \eeq
921: where ${\cal A}=(\rot^{-1}\dot{\rot})(t)\equiv(\rot^{-1}\dot{\rot})(0)\in
922: so(3)$ is a constant antisymmetric matrix.
923: The point is that all the breather solutions we have been considering
924: are {\em static} in this coordinate system, namely $\uvec_i(t)=(\sin\theta_i,
925: 0,\cos\theta_i)$, where $\theta_i$ satisfies (\ref{angeom}) as before.
926: Consequently, the linearized flow is defined by an {\em autonomous} linear
927: system of ODEs,
928: \beq
929: \delta\dot{\uvec}_i=\alpha[\uvec_i\times(\delta\uvec_{i+1}+\delta\uvec_{i-1})+\delta\uvec_i
930: \times(\uvec_{i+1}+\uvec_{i-1})]+(\delta\uvec_i\cdot\evec_3)\uvec_i\times\evec_3
931: -{\cal A}\delta\uvec_i.
932: \eeq
933: Using the basis $\{\epsvec_{i},\evec_2\}$ for $T_{\uvec_i}S^2$ and writing
934: \beq
935: \delta\uvec_i=a_i(t)\epsvec_i+b_i(t)\evec_2,
936: \eeq
937: this system reduces to
938: \beq
939: \left(\begin{array}{c}\dot{a} \\ \dot{b}\end{array}\right)=
940: \Lambda\left(\begin{array}{c} a \\ b \end{array}\right)=
941: \left(\begin{array}{cc} 0 & \Lambda^+ \\ \Lambda^- & 0 \end{array}\right)
942: \left(\begin{array}{c} a \\ b \end{array}\right)
943: \eeq
944: where $\Lambda^\pm$ are constant, tridiagonal,
945: symmetric real $N\times N$ matrices with components
946: \bea
947: \Lambda^+_{ij}(\theta)&=& \alpha\{-\delta_{i,j+1}-\delta_{i,j-1}+[\cos(\theta_i-\theta_{i+1})
948: +\cos(\theta_i-\theta_{i-1})]\delta_{i,j}\} \nonumber \\
949: & & \qquad+\cos\theta_i(\cos\theta_i-\omega)\delta_{i,j},
950: \nonumber \\
951: \Lambda^-_{ij}(\theta)&=& \alpha\{\cos(\theta_i-\theta_j)(\delta_{i,j+1}+\delta_{i,j-1})
952: -[\cos(\theta_i-\theta_{i+1})
953: +\cos(\theta_i-\theta_{i-1})]\delta_{i,j}\}
954: \nonumber \\
955: \label{***}
956: & &\qquad + (\omega\cos\theta_i-\cos 2\theta_i)\delta_{i,j}.
957: \eea
958: The Floquet matrix $\fl:\delta\nvec(0)\mapsto\delta\nvec(T)$ is simply
959: \beq
960: \label{**}
961: \fl=\rot(T)\exp(T\Lambda)\rot(0)^{-1}\equiv\exp(T\Lambda).
962: \eeq
963: One may easily check that equations (\ref{***}), (\ref{**}) reproduce the
964: $\alpha=0$ matrices previously obtained. Note that we have {\em not} imposed a monochromatic
965: ansatz for the
966: perturbation: $\delta\uvec_i$ and hence $\delta\nvec_i$ are permitted to have arbitrary
967: time dependence. We have merely made a particularly convenient choice of coordinates.
968:
969: So a period $T$ breather is linearly stable if and only if its corresponding
970: $\Lambda$ matrix has $\spec\Lambda\subset i\R\subset\C$, a criterion which is
971: trivial to test numerically.
972: This should be contrasted with the procedure
973: usually employed in numerical linear stability analyses, where construction of $\fl$
974: requires a coupled system of $2N(N+1)$ linear and nonlinear ODEs to be
975: solved numerically. Of course, the transformation to co-rotating coordinates, and the consequent
976: simplification of the Floquet problem, are only possible because we are considering
977: anisotropic spin
978: chains which retain a $SO(2)$ isotropy group. The technique will not work
979: in chains with fully anisotropic exchange interaction, as are considered in
980: \cite{flazol}.
981:
982: The numerical results described below were obtained by computing
983: $\spec\Lambda$
984: using various standard routines taken from \cite{numrec} on a 41 site
985: lattice (the results do not differ significantly from those obtained with an
986: 81 site lattice). It should be noted that, while the algorithm to construct the breather
987: initial profile $\theta$ employed spatial reflexion symmetry, no such symmetry
988: is imposed on the spectral problem for $\Lambda$.
989: Hence {\em all} possible
990: modes of instability are included in the analysis. Given the spectrum $\{\lambda_i\}$ of
991: $\Lambda$, one constructs
992: \beq
993: \label{mudef}
994: \mu:=\max_i(\re\lambda_i)^2.
995: \eeq
996: The corresponding breather is stable if and only if $\mu=0$.
997:
998: Figure \ref{fig10}
999: presents a graph of the maximal $\wt{\alpha}(\omega)$ of theorem \ref{stable},
1000: that is, the coupling at which instability of the
1001: frequency $\omega$ $(o+)$-breather first occurs. Three features should be noted. The first is
1002: that, contrary to expectation, weakly coupled breathers remain stable even at resonant
1003: frequencies, $\omega^{-1}\in\Z$. In fact, there seems to be nothing special
1004: about these frequencies at all from the standpoint either of breather existence (see sections
1005: \ref{sec-analytic} and \ref{sec-numeric}) or of stability. Second, $\wt{\alpha}(\omega)$ appears
1006: to be finite for all $\omega$: even when $\omega>0.424$ so that the $(o+)$-breather continues
1007: all the way to the
1008: continuum soliton (a linearly stable solution of the continuum system), it loses stability
1009: along the way. Third $\wt{\alpha}(\omega)$ varies very little with $\omega$ except for $\omega$
1010: close to 1. It appears to grow unbounded as $\omega\ra 1$, which seems reasonable since the
1011: $\omega=1$ $(o+)$-``breather'' is nothing but the trivial vacuum $\nvec_i=\evec_3$ which is
1012: stable for all $\alpha$. This limit is numerically inaccessible, however (the breathers tend
1013: to spread out over the whole lattice as $\omega$ grows so that finite size effects dominate), so
1014: one should treat this observation with caution.
1015:
1016: \begin{figure}[ht]
1017: \centerline{
1018: \epsfxsize=7cm
1019: \epsfbox[50 180 570 620]{fig_domain.ps}}
1020: \caption{The coupling of first instability $\wt{\alpha}$ as a function of frequency, $\omega$
1021: for $(o+)$-breathers.}
1022: \label{fig10}
1023:
1024: \end{figure}
1025:
1026: Figure \ref{fig11}
1027: shows plots of $\mu$ against coupling $\alpha$ for $(o+)$-breathers of various
1028: frequencies. For $\omega \ll 0.424$, the breather remains stable until after it has joined the
1029: $(-+)$ branch, only losing stability when this trivial static solution does. After this, $\mu$
1030: grows monotonically with $\alpha$. For $\omega>0.424$
1031: the breather first loses and then regains stability: a hump of compact support
1032: appears in the graph
1033: of $\mu$. This hump is followed by another of roughly the same width but much less tall (note
1034: the change in vertical scale), so that stability is once again lost and regained. This leads us
1035: to conjecture that instability occurs in regularly repeating bands, the
1036: degree of instability (the height of the humps) decaying exponentially as $\alpha$ increases
1037: and the breather approaches the continuum soliton. It is interesting to note that the transition
1038: from stability to instability (or vice versa) generically coincides with a sign change in the
1039: determinant of $D\hat{F}$, where $\hat{F}$ is the {\em unsymmetrized}
1040: finite lattice version of the
1041: continuation function $F$ defined in (\ref{m5}). However no such sign change occurs
1042: in the Jacobian of
1043: the {\em symmetrized} continuation function (\ref{f}), as employed in our continuation scheme,
1044: so the corresponding eigenvector
1045: $\delta\theta\in\ker D\hat{F}$ must have odd spatial parity. It seems likely, therefore, that
1046: stability transitions generically accompany symmetry breaking bifurcations in the full
1047: (i.e.\ unsymmetrized) continuation problem.
1048:
1049: \begin{figure}[ht]
1050: \centerline{
1051: \epsfxsize=7cm
1052: \epsfbox[50 180 550 620]{fig_op.ps}}
1053: \caption{The instability function $\mu(\alpha)$, as defined in equation (\ref{mudef})
1054: for various $(o+)$-breathers (solid: $\omega=0.5$; dashed $\omega=0.1$). The breather is
1055: stable if and only if $\mu=0$. Note that both are stable for small $\alpha$ despite having
1056: resonant frequencies.}
1057: \label{fig11}
1058:
1059: \end{figure}
1060:
1061: Figure \ref{fig12}
1062: shows similar plots for $(-o+)$-breathers. Recall that no result equivalent to
1063: theorem \ref{stable} holds in this context, so there is no reason to expect stability even for
1064: arbitrarily small $\alpha$. Indeed, we found that for all $\omega$, $\mu$ departs from $0$
1065: as soon as $\alpha$ does. For $\omega\notin(0.133,0.424)$ the breather remains unstable until
1066: it joins the appropriate trivial branch ($(--+)$ for $\omega<0.133$, $(-+)$ for $\omega>0.424$),
1067: whereupon it becomes briefly stable before instability irrevocably sets in. For $\omega\in
1068: (0.133,0.424)$, the breathers gain stability, then lose it, rather in the fashion of the
1069: $(o+)$-breathers described above. Again, a pattern of repeating bands of diminishing instability
1070: seems to occur as $\alpha$ grows large and the breather approaches the continuum soliton.
1071: Plots of $\mu(\alpha)$ for $(--o+)$ breathers with $\omega\in(0.045,0.133)$ are very similar:
1072: the breathers start unstable, then gain and lose stability in a repetitive pattern. This leads
1073: us to conjecture that wherever breathers tend to the continuum soliton as $\alpha\ra\infty$,
1074: a similar banded stability pattern occurs.
1075:
1076: \begin{figure}[ht]
1077: \centerline{
1078: \epsfxsize=7cm
1079: \epsfbox[50 180 550 620]{fig_mop.ps}}
1080: \caption{The instability function $\mu(\alpha)$ for various $(-o+)$-breathers
1081: (solid: $\omega=0.3$; dashed: $\omega=0.6$; dotted: $\omega=-0.4$). All are unstable for
1082: small $\alpha$, though they have nonresonant frequency.}
1083: \label{fig12}
1084:
1085: \end{figure}
1086:
1087:
1088:
1089: \section{Conclusion}\news
1090: In this paper we have proved the existence of discrete breathers in ferromagnetic
1091: spin chains with easy-axis anisotropy, and constructed such breathers by means of
1092: numerical analysis. The main novelty of our numerical results is that discrete
1093: breathers exist independent of the inter-spin coupling $\alpha$, right up to the
1094: continuum limit $\alpha\ra\infty$. This should be contrasted with previous studies
1095: of spin chains with easy-plane anisotropy \cite{WMB}, where the breathers are again
1096: ``monochromatic,'' but the spin waves have a
1097: different dispersion relation, and apparently do cause trouble at large $\alpha$.
1098:
1099: The method of proof itself was something of a hybrid: an ansatz was made to reduce
1100: the equation for breathers from a system of ODEs to a purely algebraic system,
1101: but then existence of solutions of the latter was proved by continuing a
1102: decoupled solution in the sequence space $\ell_2$. Consequently, the result shares
1103: nice properties of both the Flach and MacKay-Aubry approaches. Since the reduced
1104: system depends parametrically on $\omega$, spin wave resonances cause no problems,
1105: theoretical or numerical (c.f.\ \cite{aubmac}). On the other hand, the result
1106: generalizes immediately to multi-dimensional spin lattices, as MacKay-Aubry style
1107: theorems usually do. Whether our numerical results generalize similarly, that is,
1108: whether a picture similar to Figure \ref{fig5} emerges for 2 and 3 dimensional lattices is
1109: an interesting open question. It should be noted that the continuum theory in higher dimensions
1110: does support radially symmetric, exponentially localized soliton solutions of the form
1111: analogous to (\ref{ansatzc}), namely
1112: \beq
1113: \bn(\xvec,t)=(\sin\theta(|\xvec|)\cos\omega t,-\sin\theta(|\xvec|)\sin\omega t,
1114: \cos\theta(|\xvec|)),
1115: \eeq
1116: although explicit expressions for $\theta(r)$ are not known \cite{KIK}.
1117:
1118: The ansatz may be interpreted as transforming to co-rotating coordinates, wherein the breather
1119: solutions are static. This viewpoint greatly simplifies the linear stability analysis since
1120: one need only solve an autonomous linear system of ODEs. In this way we have made an extensive
1121: study of the linear stability properties of the breathers. The results suggest that those
1122: breather families which converge as $\alpha\ra\infty$ to the continuum soliton experience
1123: repeating bands of instability of diminishing strength as $\alpha$ increases. We found
1124: that the spin waves have no influence whatsoever on stability issues, even though
1125: the stability analysis includes all possible perturbations, including those {\em outside} the
1126: monochromatic ansatz. Flach et al also perform a numerical linear stability analysis for
1127: breathers in their
1128: spin chains, and obtain results broadly similar to ours \cite{flazol}, namely
1129: 1-site breathers are found to be stable at weak coupling, and certain multisite breathers are
1130: found to be unstable at weak coupling. The subsequent loss and regain of (in)stability for
1131: increasing coupling is not reported - perhaps it is special to chains with isotropic exchange
1132: interaction. Their survey of the breather existence domains is rather less extensive than ours,
1133: however, presumably due to the much higher computational cost in their systems, so it is
1134: possible that similar band structures to those found here exist there also.
1135:
1136: One open question immediately arises: that of breather mobility in this system.
1137: In the continuum limit, there exist travelling solitons which
1138: propagate at constant speed \cite{KIK}.
1139: Can one find analogous propagating breathers at finite coupling? There seems to be
1140: little hope of proving anything rigorously about such objects, but one could still
1141: hope to study breather mobility numerically, focusing on how breather
1142: mobility depends on $\omega$ and $\alpha$. For a survey of what is known
1143: about moving discrete breathers, see the review articles \cite{flarev,laisie}. One
1144: striking observation by Aubry and Cretegny is that breather mobility may be associated
1145: with certain behaviour of the Floquet matrix \cite{aubcre}.
1146: In a one-parameter family of breathers (such as
1147: $\theta^\alpha$), mobility occurs precisely at a value, $\alpha'$ say, where a certain type of
1148: Krein bifurcation occurs:
1149: $\fl$ becomes defective and so-called ``marginal modes'' appear (they span the orthogonal
1150: complement of the sum of the $\fl_{\alpha'}$ eigenspaces).
1151: Perturbing $\theta_{\alpha'}$ in the
1152: direction of such a marginal mode produces a slowly translating breather. So breather mobility
1153: seems to be naturally associated with transitions from linear stability to instability. It
1154: would be interesting to see whether the stability transitions observed in section
1155: \ref{sec-stability} have such significance.
1156:
1157:
1158:
1159: \section*{Acknowledgements}\news
1160: We acknowledge the EPSRC for a Postdoctoral Research Fellowship (JMS) and an
1161: Advanced Fellowship (PMS). One of us (JMS) would like to thank Chris Eilbeck for valuable
1162: conversations.\\
1163:
1164:
1165:
1166: \begin{thebibliography}{99}
1167:
1168: \bibitem{aubstab} S. Aubry,
1169: ``Breathers in nonlinear lattices: Existence, linear stability and quantization''
1170: {\sl Physica} {\bf D103}, 201 (1997).
1171:
1172: \bibitem{aubcre} S. Aubry and T. Cretegny,
1173: ``Mobility and reactivity of discrete breathers''
1174: {\sl Physica} {\bf D119} 34 (1998).
1175:
1176: \bibitem{baemac} C. Baesens and R.S. MacKay,
1177: ``Exponential localization of linear response in networks with
1178: exponentially decaying coupling''
1179: {\sl Nonlinearity} {\bf 10}, 931 (1997).
1180:
1181: \bibitem{ift} Y. Choquet-Bruhat, C. DeWitt-Morette and M. Dillard-Bleick,
1182: {\sl Analysis, Manifolds and Physics, Part I},
1183: (North-Holland, Amsterdam, 1982) p 91.
1184:
1185: \bibitem{eil1} J.C. Eilbeck, P.S. Lomdahl and A.C. Scott,
1186: ``The discrete self-trapping equation''
1187: {\sl Physica} {\bf 16D} 318 (1985).
1188:
1189: \bibitem{eil2} J. Carr and J.C. Eilbeck,
1190: ``Stability of stationary solutions of the discrete self-trapping equation''
1191: {\sl Phys.\ Lett.} {\bf 109A} 201 (1985).
1192:
1193: \bibitem{fla} S. Flach,
1194: ``Existence of localized excitations in nonlinear Hamiltonian lattices'',
1195: {\sl Phys.\ Rev.} {\bf E51}, 1503 (1995).
1196:
1197: \bibitem{flarev} S. Flach and C.R. Willis,
1198: ``Discrete Breathers''
1199: {\sl Phys.\ Rep.} {\bf 295}, 181 (1998).
1200:
1201: \bibitem{flazol} S. Flach, Y. Zolotaryuk and V. Fleurov,
1202: ``Discrete breathers in classical spin lattices''
1203: {\sl Phys.\ Rev.} {\sl B63}, 214422 (2001)
1204:
1205: \bibitem{dnls} M. Johansson and Y.S. Kivshar,
1206: ``Discreteness-induced oscillatory instabilities of dark solitons''
1207: {\sl Phys.\ Rev.\ Lett.} {\bf 82} 85 (1999).
1208:
1209:
1210: \bibitem{KIK} A.M. Kosevich, B.A. Ivanov and A.S. Kovalev,
1211: ``Magnetic solitons'', {\sl Phys.\ Rep.} {\bf 194}, 117 (1990).
1212:
1213: \bibitem{laisie} R. Lai and A.J. Sievers,
1214: ``Nonlinear nanoscale localization of magnetic excitations in atomic lattices''
1215: {\sl Phys.\ Rep.} {\bf 314} 147 (1999).
1216:
1217: \bibitem{aubmac} R.S. MacKay and S. Aubry,
1218: ``Proof of existence of breathers for
1219: time-reversible or Hamiltonian networks of weakly
1220: coupled oscillators'',
1221: {\sl Nonlinearity} {\bf 7}, 1623 (1994).
1222:
1223:
1224: \bibitem{aubseq} J.L. Mar\'{\i}n and S. Aubry,
1225: ``Breathers in nonlinear lattices: numerical
1226: calculation from the anticontinuous limit''
1227: {\sl Nonlinearity} {\bf 9}, 1501 (1996).
1228:
1229: \bibitem{numrec} W.H. Press, B.P. Flannery, S.A. Teukolsky and W.T.Vetterling,
1230: {\sl Numerical Recipes} (Cambridge University Press, Cambridge, UK, 1989) ch.\ 11.
1231:
1232: \bibitem{SM} J.A. Sepulchre and R.S. Mackay,
1233: ``Localized oscillations in conservative or dissipative
1234: networks of coupled autonomous oscillators'' {\sl Nonlinearity} {\bf 10}, 679 (1997).
1235:
1236:
1237: \bibitem{WMB} R.F. Wallis, D.L. Mills and A.D. Boardman,
1238: ``Intrinsic localized spin modes in ferromagnetic chains
1239: with on-site anisotropy'', {\sl Phys.\ Rev.} {\bf B52}, R3828 (1995).
1240:
1241:
1242:
1243:
1244: \end{thebibliography}
1245:
1246:
1247:
1248:
1249: \end{document}
1250:
1251:
1252: