1: \documentclass[aps,pre,twocolumn,groupedaddress,showpacs]{revtex4}
2: \usepackage{graphicx,amsmath,amssymb}
3:
4: \begin{document}
5: \title{On the internal modes in sine-Gordon chain}
6: \author{Jaroslaw E. Prilepsky}
7: \author{Alexander S. Kovalev}\email{kovalev@ilt.kharkov.ua}
8:
9: \affiliation{B.Verkin Institute for Low Temperature Physics and
10: Engineering, NASU, \\ 47 Lenin Ave., Kharkov, 61103, Ukraine}
11:
12: \begin{abstract}
13: We address the issue of internal modes of a kink of a discrete
14: sine-Gordon equation. The main point of the present study is to
15: elucidate how the antisymmetric internal mode frequency dependence
16: enters the quasicontinuum spectrum of nonlocalized waves. We
17: analyze the internal frequency dependencies as functions of both
18: the number of cites and discreteness parameter and explain the
19: origin of spectrum peculiarity which arises after the frequency
20: dependence of antisymmetric mode returns back to the continuous
21: spectrum at some nonzero value of the intersite coupling.
22: \end{abstract}
23: \date{\today} \pacs{ 05.45.-a, 63.20.Pw, 63.20.Ry} \maketitle
24:
25:
26: \thispagestyle{headings} The existence of internal modes is a well
27: known peculiarity of nonintegrable systems \cite{kpc98}: If one
28: linearizes a nonintegrable equation above the static nonlinear
29: solution, then the effective potential of the linear equation
30: obtained appears to be not reflectionless in contrast to the
31: integrable case. The particular example is the existence of
32: internal (shape) modes of kinks in the well known discrete
33: sine-Gordon equation (DSGE). This set of differential-difference
34: equations reads as (in normalized dimensionless units):
35: \begin{equation}\label{dsg}
36: \ddot{u}_n + \sin u_n + \lambda \Big[ (u_n - u_{n-1}) - ( u_{n+1}
37: - u_n ) \Big]= 0 \, ,
38: \end{equation}
39: where $u_n(t)$ is the field variable, which can have a multitude
40: of physical meanings \cite{bk03}, $\lambda$ is a coupling
41: parameter, dot means the time derivative and index $n$ numerates
42: the 1D chain sites. The features and behaviour of kink internal
43: modes for DSGE, as well as for more general types of the so-called
44: Frenkel-Kontorova (FK) model, have been already studied in very
45: details \cite{bk03,bkp97,bkv90,kj00}. The spectrum of linear waves
46: of DSGE around a kink contains either two or one localized mode,
47: depending on the value of parameter $\lambda$. (We consider the
48: stable kink centered between the chain sites.) The frequencies of
49: these modes lie inside the spectrum gap; the linear spectrum
50: itself is given by
51: \begin{equation}\label{spec}
52: \omega^2 = 1 + 4 \lambda \sin^2 \frac{k}{2} \, ,
53: \end{equation}
54: where $\omega$ is the frequency of linear waves and $k$ is the
55: wave number. The lowest (gap edge) frequency is $\omega =1$ (in
56: the renormalized units).
57:
58: However, up to our knowledge, notwithstanding the great quantity
59: of results none so far has concentrated on the way an internal
60: modes ceases to exist: Only the fact of existence or nonexistence
61: of that mode has been the subject of interest
62: \cite{kpc98,bk03,bkp97,bkv90,kj00}. In this paper we focus on the
63: effect entailed by the detachment of the internal mode frequency
64: from the spectrum. So, the question arises: How the internal
65: modes, which split off from the spectrum at some nonzero value of
66: coupling parameter (or some other effective parameter altering the
67: system state), affect the spectrum itself and how they behave
68: while still being inside the spectrum.
69:
70: To begin with, we note that the analytical approximate solution
71: for a static kink of DSGE (\ref{dsg}) can be found either in the
72: so-called anticontinuum limit (extremely small values of
73: $\lambda$) as a series in powers of $\lambda$ \cite{bkv90}, or in
74: the opposite case of strong coupling (large $\lambda$), when the
75: discrete kink solution acquires the form of that in the continuous
76: SGE with small corrections due to discreteness (see e.g.
77: \cite{fk96}). Let us now summarize the results on the existence of
78: internal modes in the DSGE containing a single kink. For small
79: $\lambda$ only the lowest symmetric Pierls-Nabarro (PN) mode,
80: associated with the kink oscillations in the PN relief, has the
81: frequency inside the gap: it corresponds to the translational kink
82: mode of continuous SGE, activated due to discreteness. This mode
83: remains the internal mode for the whole interval of kink
84: stability. The properties of this mode are well understood and
85: studied \cite{bk03,bkv90}, and we shall be mainly concerned with
86: the other internal mode. For larger (but still weak) $\lambda$
87: there exists the "critical" point $\lambda_d$, where one more mode
88: dependence detaches from the continuous spectrum: this mode
89: corresponds to the oscillation of kink width. (More complicated
90: on-site FK-type potentials may have a larger variety of localized
91: internal eigenmodes \cite{bkp97} or, for the case of more than one
92: kink in the chain, the DSGE possesses a larger number of internal
93: modes as well \cite{bk03}.) The existence of the second internal
94: mode for large $\lambda$ obviously matches the criterium given by
95: Kivshar et al. in Ref.\cite{kpc98} for nearly integrable SGE (see
96: also Ref.\cite{kj00} for details).
97:
98: The effects we shall discuss can be readily seen in Fig.\ref{f1}:
99: As the internal mode frequency goes back for small $\lambda$ to
100: the quasicontinuous spectrum, it brings about a conspicuous change
101: in the behaviour of higher modes. However only the dependencies of
102: odd modes (the mode number corresponds to the number of
103: eigenfunction nodes) ''feel'' the return of the localized first
104: antisymmetric mode to the spectrum. The modes with the even number
105: of eigenfunction nodes do not react to this return. So, we can
106: infer that only the modes having the same symmetry as the mode,
107: which detaches from the spectrum, experience pronounced change
108: (which can be also seen in the behaviour of the derivatives,
109: Fig.\ref{f1}b,c). This change ''propagates'' inwards the spectrum
110: becoming less and less abrupt with the increase of mode number. In
111: addition, as can be seen in the inset of Fig.\ref{f1}a, the
112: dependence of the first antisymmetric mode indeed enters the
113: spectrum: it crosses the band edge frequency $\omega=1$ at the
114: point $\lambda_d \approx 0.26$ (of course, it crosses only the
115: band threshold but not any of the frequency dependencies). For
116: smaller $\lambda$ the dependence of former antisymmetric localized
117: mode belongs to the spectrum of usual delocalized waves, tending
118: to coalesce with the dependence of second (symmetric) mode.
119: \begin{figure}\centering
120: \scalebox{0.5}[0.5]{ \includegraphics[bb=63 324 511 795 ]{Fig1}}
121: \caption{\label{f1} Numerical results for DSGE containing 250
122: sites with a kink centered in the middle of the chain between the
123: sites. a) The frequency dependencies for several lowest modes on
124: the value of $\lambda$ (the dependence for the lowest PN-mode is
125: not presented); the numbers correspond to the mode number, i.e. to
126: the number of nodes of corresponding eigenfunction. The curve
127: marked { \em ''1''} is the dependence for localized antisymmetric
128: mode. The inset shows the magnified region in the close vicinity
129: of the detachment value $\lambda_d$ (the bandgap threshold marked
130: by the dashed line) . The bottom panels show the behaviour of b)
131: $\mathrm{d} \omega / \mathrm{d} \lambda$ and c) $\mathrm{d}^2
132: \omega / \mathrm{d} \lambda^2$ for three lowest antisymmetric
133: modes ({\em 1,3} and {\em 5}) . }
134: \end{figure}
135: Also we note another interesting feature of eigenfrequency
136: dependencies. For large $\lambda$, i.e. when we are approaching
137: the continuum limit, the dependencies for odd modes with numbers
138: $2n+1$ tend to coalesce with the dependencies for {\em lower} even
139: modes, i.e. with those having numbers $2n$. However the detachment
140: of one mode results in another tendency: after the splitting, for
141: $\lambda$ close to zero, the dependencies for odd modes tend to
142: coalesce with those of { \em higher} even modes, i.e. the
143: frequencies of modes with numbers $2 n +1$ come closer and closer
144: to those for the modes with numbers $2n+2$. The inflection of the
145: odd modes dependencies and peculiarities of their derivatives in
146: the vicinity of $\lambda_d$ merely indicate this ''change'' in the
147: tendency for coalescence.
148:
149: To delineate and compare the behaviour of internal modes we first
150: study some illustrative examples when we have only a few pendulums
151: with linear bonds.
152:
153: {\em Small number of pendulums with linear bonds.} For the
154: simplest case of two pendulums the dynamical equations read as:
155: \begin{equation}\label{2p}
156: \ddot{u}_{1, \, 2}+\lambda (u_{1, \, 2}-u_{2, \, 1}) + \sin u_{1,
157: \, 2} = 0 \, .
158: \end{equation}
159: Linearizing Eqs.(\ref{2p}) around the ground state solution, $u_1
160: = u_2 = 0$, we arrive at the eigenfrequency dependencies:
161: $\omega_0^2=1$ , $\omega_1^2 = 1+2 \lambda$. The higher frequency
162: corresponds to the antiphase oscillations, the lower one -- to the
163: inphase motion. Now let us study the spectrum of linear
164: oscillations above the inhomogeneous static solution of
165: Eq.(\ref{2p}) (analogue of a kink). First we have to find the
166: static distribution for small values of $\lambda$. We suppose that
167: this distribution has a form: $u_1=u^0$, $u_2 = 2 \pi - u^0$,
168: $u^0=\alpha \lambda + \beta \lambda^2 + \mathcal{O}(\lambda^3)$,
169: and we should determine the corresponding constants, $\alpha$ and
170: $\beta$, in the dependence $u^0(\lambda)$. Expanding the sine in
171: Taylor series and equalizing coefficients at the same powers of
172: $\lambda$ we find:
173: \begin{equation}\label{2u}
174: u^0 = 2 \pi \lambda - 4 \pi \lambda^2 + \ldots \, .
175: \end{equation}
176: Then we linearize the initial set of dynamical equations
177: (\ref{2p}) with respect to small quantities $v_i$: $u_1 = u^0 +
178: v_1 \mathrm{e}^{\mathrm{i} \omega t}$, $u_2= 2 \pi - u^0 + v_2
179: \mathrm{e}^{\mathrm{i} \omega t}$. The consistency condition gives
180: the spectral dependencies: for the lower inphase mode $\omega_0^2
181: = \cos u^0$, that is the analogue of PN mode in a large system,
182: and for the antiphase mode $\omega_1^2 = \cos u^0 + 2 \lambda$ ---
183: the analogue of the second antisymmetric internal mode. The
184: analytical form for the frequencies, which is valid for the small
185: $\lambda$ values, is as follows:
186: \begin{equation}\label{2spec}
187: \omega_0^2 = 1 -2 \pi^2 \lambda^2 + \ldots \, , \qquad \omega^2_1
188: = \omega_0^2 + 2 \lambda \, .
189: \end{equation}
190: The frequency dependence $\omega_1(\lambda)$ grows linearly with
191: $\lambda$ for weak coupling parameter values and stays inside the
192: band, but then, as $\lambda$ becomes larger, this growth slows
193: down showing an ultimate tendency of this mode to get inside the
194: gap below. The value of $\lambda$ for which this dependence
195: crosses the band threshold can be roughly estimated as $\lambda_d
196: \approx 1/ \pi^2$. One also notes that the frequency of PN-mode,
197: $\omega_0$, drops to zero for $u^0 = \pi/2$. This manifests the
198: well known fact that for the finite size chain there occurs an
199: instability when the characteristic spatial scale of a kink
200: overgrows the size of a system (see e.g. Ref.\cite{k78,kp03}). For
201: a large system the PN-mode softens at some critical value
202: $\lambda_c \sim L^2$, where $L\gg1$ is the system size
203: \cite{kp03}. For two pendulums, setting $u^0_1 = \pi/2 - U$,
204: $u^0_2 = 3 \pi/2 + U$, $U\ll1$, one finds the frequency
205: dependencies near the critical point: $\omega_0^2 \approx
206: (\lambda_c - \lambda)/2 \lambda_c^2$, with $\lambda_c = 1/\pi$,
207: and $\omega_1^2 = \omega_0^2 + 2 \lambda$. We see that at the
208: critical point $\omega_1^2(\lambda_c) = 2/ \pi <1$, i.e. the
209: frequency remains inside the gap. Thus we expect that this
210: internal mode dependence for a big system does not return back to
211: the spectrum for large values of $\lambda$ and continues to be the
212: internal mode for the whole interval of kink stability except for
213: the extremely weak coupling. The splitting of this mode dependence
214: from the continuum spectrum is of the order of $\lambda^{-1}$ for
215: large values of coupling parameter \cite{kpc98,kj00}.
216:
217: The systems containing more sites enable us to study the '''degree
218: of localization'' since we now have the outer and inner (core)
219: sites. Let us mark the sites in the kink core (the middle of a
220: chain) by numbers ''1'' and ''-1'', their neighbor sites by ''2''
221: and ''-2'' etc. For the case of four pendulums the ''kink''
222: solution for small $\lambda$ is given by:
223: \begin{equation}\label{4u}
224: u_2^0 \approx 2 \pi \lambda^2 + \ldots\, ,\qquad u_1^0 \approx 2
225: \pi \lambda - 8 \pi \lambda^2 + \ldots \, ,
226: \end{equation}
227: and the obvious solution symmetry, $u_{-i} = 2 \pi - u_{i}$, gives
228: the field distribution for the remaining sites. In this limit the
229: spectral dependencies for two lowest modes
230: are:
231: \begin{subequations}
232: \begin{gather}\label{4pend-om0}
233: \omega_0^2 \approx 1 - \pi^2 \lambda^2 \, , \\ \omega_1^2 \approx
234: 1+(2 - \sqrt 2) \lambda - \frac{\pi^2}{ 2}( 2 - \sqrt 2) \lambda^2
235: \, .\label{4pend-om1}
236: \end{gather}
237: \end{subequations}
238: Putting $\omega_1(\lambda)=1$ in Eq.(\ref{4pend-om1}) one gets the
239: rough estimate for the point where this frequency dependence
240: enters the ''quasicontinuum'' spectrum: $\lambda_d \approx 2/
241: \pi^2$. In the same way one could deal with the systems of bigger
242: size (see below).
243:
244: The spectrums and amplitude ratio dependencies on $\lambda$ for
245: four and six pendulums are given in Fig.\ref{f2}. As the number of
246: sites grows, the spectrum peculiarities, seen in Fig.\ref{f1},
247: become more and more pronounced. In the panels (b) and (d) of
248: Fig.\ref{f2} we present the dependencies for the ratios of the
249: oscillation amplitudes for different pairs of neighbor sites,
250: $v_{i+1}/v_i$, for PN (zeroth) mode and first antisymmetric mode.
251: Both modes display the tendency for localization in the kink core
252: with the growth of $\lambda$, but coming closer to the critical
253: point $\lambda_c$ they become localized worse because of the large
254: kink width.
255: \begin{figure}
256: \centering
257: \scalebox{0.55}[0.55]{\includegraphics[65,462][509,792]{Fig2.eps}}
258: \caption{Numerical results for DSGE containing four and six sites
259: with a ''kink'' centered in the middle between the sites. Left
260: panels: full spectrum of a) 4 pendulums and c) 6 pendulums in the
261: interval of kink stability; the numbers correspond to the mode
262: number. Right panels show the dependencies of oscillations
263: amplitude ratios of different neighbor sites corresponding to two
264: lowest modes: b) for four pendulums $v_2/v_1$; d) for six
265: pendulums $v_{i+1} / v_{i}$ with $i=1$ (solid line) and $i=2$
266: (dashed line). The digits correspond to the mode number.}
267: \label{f2}
268: \end{figure}
269: Compared with the first antisymmetric mode, the PN-mode appears to
270: be localized better. These dependencies have a minima for some
271: intermediate value of coupling, where the inhomogeneity brought by
272: the kink is pronounced well enough to sustain the modes
273: localization and, from the other side, the kink wight is still
274: not large.
275:
276: {\em Large systems.} Now suppose that our system has $2N$ sites.
277: The static distribution for the kink is again given by $u^0_1 = 2
278: \pi \lambda + \mathcal{O}(\lambda^2)$, $u^0_n = \mathcal{O}
279: (\lambda^n)$. Then we substitute $u_n(t) = u^0_n + v_n
280: \mathrm{e}^{\mathrm i \omega t}$, and linearize the system
281: obtained with respect to $v_n$ noting that for the antisymmetric
282: modes the symmetry is $v_n = - v_{-n}$. After that one gains the
283: following linear system:
284: \begin{equation}\label{ant}
285: \begin{split}
286: \begin{cases}
287: &(\omega^2 - 3 \lambda - \cos u^0_1) v_1 + \lambda v_2 =0
288: \, , \\
289: & \; \vdots \\
290: &(\omega^2 - 2 \lambda - \cos u^0_n) v_n + \lambda (v_{n-1}+v_{n+1}) =0 \, , \\
291: & \; \vdots \\
292: &(\omega^2 - \lambda - \cos u^0_N) v_N + \lambda v_{N-1} =0 \, .
293: \end{cases}
294: \end{split}
295: \end{equation}
296: For the symmetric modes ($v_n = v_{-n}$) the first equation of
297: system (\ref{ant}) is to be replaced with
298: \begin{equation}\label{sym}
299: (\omega^2 - \lambda - \cos u^0_1) v_1 + \lambda v_2 =0 \, .
300: \end{equation}
301: For small $\lambda$ we use the approximate expressions for
302: $u_n^0$, and in the leading approximation substitute $\cos u_1^0 \
303: \to 1 - 2 \pi^2 \lambda^2$, $\cos u_n^0 \to 1$ for $n>1$. This
304: means that we effectively replaced the kink with two isotopic
305: impurities located at sites {\em 1} and {\em -1}, the ''strength''
306: of which alters with the change of $\lambda$. Seeking the solution
307: in the form $v_n = A \mathrm{e}^{\kappa n} + B \mathrm{e}^{-
308: \kappa n}$, with constant $A$ and $B$, one obtains the spectral
309: dependence for $\omega(\kappa)$ in form (\ref{spec}), where
310: $\kappa = \mathrm{i} k$. Then using the consistency condition
311: after some straightforward algebra we arrive at the relation which
312: defines the allowed values of $\kappa$ for antisymmetric modes:
313: \begin{eqnarray}\nonumber
314: (1 &-& 2 \pi^2 \lambda) \sinh [\kappa (N-1)] \\ + 2 \pi^2
315: &\lambda& \sinh [\kappa N] - \sinh[\kappa(1+N)] = 0 \, .
316: \label{rel1}
317: \end{eqnarray}
318: Expanding this relation in the vicinity of $\kappa =0$ ($\kappa N
319: \ll 1 $) up to $\kappa^5$ we determine the allowed values of
320: $\kappa$ for the frequency of first antisymmetric mode,
321: $\omega_1$, and for the next, third mode, $\omega_3$, from the
322: biquadratic equation: $a \kappa^4 + b \kappa^2 + c =0$. The
323: expressions for the coefficients are:
324: \begin{gather}\label{coefs} \nonumber
325: a = \frac{1 (\lambda - \lambda_d) + 5(2 N^2 +
326: N^4)(\lambda - \lambda_d) - 5 \lambda ( 2 N^3 + N)}{60} \, , \\
327: \nonumber b = (\lambda - \lambda_d)/3 + N^2 (\lambda - \lambda_d)
328: - N \lambda \,, \\ \nonumber c = 2 (\lambda-\lambda_d) \, ,
329: \end{gather}
330: and the detachment point is $\lambda_d = 1/\pi^2$. The notable
331: fact, which can be extracted from Eq.(\ref{rel1}), is the
332: following: if $\lambda$ is close to zero, $\lambda \ll N^{-1}$, we
333: have $\kappa^2 = - k^2 \sim N^{-2}$, and for this values of
334: coupling parameter one finds the expressions for frequency
335: dependencies as:
336: \begin{subequations}\label{om00}
337: \begin{gather}\label{om0}
338: \omega_1^2 \approx 1 + 2 \lambda (3- \sqrt 3) N^{-2} \, , \\
339: \omega_3^2 \approx 1 + 2 \lambda (3+ \sqrt 3 ) N^{-2} \, .
340: \end{gather}
341: \end{subequations}
342: So, initially, for small $\lambda$, the lowest antisymmetric mode,
343: Eq.(\ref{om0}), goes up being inside the spectrum. However quite a
344: different situation occurs if one moves inside the region where
345: the inequality $|\lambda-\lambda_d| \ll N^{-1}$ holds, i.e. in the
346: close vicinity of $\lambda_d$. In this region one obtains
347: $\kappa^2 \sim N^{-1}$ (here the inequality $\kappa N \ll 1$ is
348: true because of the additional smallness provided by the factor
349: $(\lambda - \lambda_d)$). Then for the frequency dependencies we
350: have:
351: \begin{subequations}\label{om1-3}
352: \begin{gather}\label{om1}
353: \omega_1^2 \approx 1 - 2 (\lambda - \lambda_d)N^{-1} \, , \\
354: \label{om3} \omega_3^2 \approx 1 - 4(\lambda - \lambda_d) N^{-1}+
355: 6 \lambda_d N^{-2} \, .
356: \end{gather}
357: \end{subequations}
358: The second derivative of $\omega_1$ at the point $\lambda =
359: \lambda_d$ involves the term independent on $N$: $\mathrm{d^2}
360: \omega_1 / \mathrm{d} \lambda^2 = - \lambda_d^{-1}$. Therefore we
361: can conclude that in the limit $N \to \infty$ the splitting of
362: this dependence from the lower boundary of the spectrum must have
363: a parabolic form. From the expressions (\ref{om00},\ref{om1-3}) it
364: becomes evident what brings about the peculiarity as $\lambda$
365: approaches $\lambda_d$, i.e. as the first antisymmetric mode
366: detaches from the spectrum. For $\lambda$ close to zero we have
367: $\mathrm{d} \omega_{1,3} / \mathrm{d} \lambda \sim N^{-2}$,
368: whereas in the vicinity of $\lambda_d$ the different dependence
369: takes place: $\mathrm{d} \omega_{1,3} / \mathrm{d} \lambda \sim
370: N^{-1}$. (Note that the sign of the first derivative also
371: changes.) Because of this the initial weak ($\sim N^{-2}$) growth
372: for small values of $\lambda$ changes to more rapid ($\sim
373: N^{-1}$) decrease in the close vicinity of $\lambda_d$. The
374: dependencies for the modulus of first derivatives at $\lambda =
375: \lambda_d$ on the number of sites shown in Fig.\ref{f3}. For
376: $N'=2N \gtrsim 30$ they are in a good agreement with analytical
377: results (\ref{om1-3}). With the increase of $N$ this agreement
378: becomes better inasmuch as we omitted the higher with respect to $
379: N^{-1}$ terms in Eqs.(\ref{om1-3}). In the region $\lambda <
380: \lambda_d$ there must be an extremum point
381: $\lambda_{\mathrm{ext}}$, $\mathrm{d} \omega_i / \mathrm{d}
382: \lambda |_{\lambda_{\mathrm{ext}}} = 0$, for both dependencies
383: $\omega_{1,3}(\lambda)$, at which the monotonic growth changes to
384: decreasing. These extremum points, $\lambda_{\mathrm{ext}}$, tend
385: to $\lambda_d$ as $N$ gets bigger: $(\lambda_d -
386: \lambda_{\mathrm{ext}}) \sim N^{-1}$. The dependencies of the
387: value of difference, $(\lambda_d - \lambda_{\mathrm{ext}})$, on
388: the number of sites $N$ are presented in the inset panel of
389: Fig.\ref{f3}. For $\lambda>\lambda_d$ the first antisymmetric mode
390: gets into the spectrum gap and becomes an internal mode. However
391: the dependence for the next, third antisymmetric mode, in spite of
392: its tendency to drop down, cannot cross the dependence of the
393: preceding second (symmetric) mode and therefore these two
394: dependencies get closer and closer to each other.
395: \begin{figure}
396: \centering
397: \scalebox{0.6}[0.6]{\includegraphics[90,461][469,772]{Fig3.eps}}
398: \caption{The dependencies for the modulus of derivative at the
399: point $\lambda = \lambda_d$ on the number of sites for the first
400: (antisymmetric) mode and for the next, third mode. The digits
401: correspond to the mode number. Dashed curves for each mode show
402: the analytical prediction obtained by virtue of Eqs.(\ref{om1-3}).
403: The inset shows the dependence of deviation $\lambda_d -
404: \lambda_{\mathrm{ext}}$ of the extremal point from the detachment
405: point as a function of number of sites $N' = 2 N$.} \label{f3}
406: \end{figure}
407:
408: Now consider an infinite system. We seek the solution in the form
409: of localized wave, $v_n \sim \mathrm{e}^{-\kappa n}$, $\kappa>0$.
410: Whereupon the only condition defining the allowed values of
411: $\kappa$ becomes as follows:
412: \begin{equation}
413: \mathrm{e}^{- \kappa} + 2 \pi^2 \lambda - 3 - 4 \lambda \sinh^2
414: \frac{\kappa}{2} = 0 \, .
415: \end{equation}
416: Expanding this relation we arrive at the dependence: $\kappa_1 = 2
417: \lambda_d^{-1} (\lambda - \lambda_d)$, and then one finds the
418: expression for the frequency of the antisymmetric localized mode
419: as:
420: \begin{equation}\label{om1inf}
421: \begin{cases}
422: \omega_1^2 = 1 \, , \quad &\text{for} \; \; \lambda < \lambda_d \, ,\\
423: \omega_1^2 \approx 1 - 4 \lambda_d^{-1} (\lambda - \lambda_d)^2 ,
424: \quad &\text{for} \; \; \lambda> \lambda_d \, ,
425: \end{cases}
426: \end{equation}
427: in consistency with the result for finite system. The
428: characteristic spatial scale for the eigenfunction of this
429: internal mode is $l_1 = \kappa_1^{-1}$. Obviously, the smaller
430: $l_1$, the better the eigenfunction is localized. We see that at
431: the outset, as this mode detaches, $l_1 \sim (\lambda -
432: \lambda_d)^{-1}$.
433:
434: Proceed to studying the symmetric modes. The relation which
435: defines the allowed values of $\kappa$ for those reads:
436: \begin{eqnarray}\nonumber
437: (1 &+& 2 \pi^2 \lambda) \sinh [\kappa (N-1)] \\ - 2 (1 + \pi^2
438: &\lambda&) \sinh [\kappa N] + \sinh[\kappa(1+N)] = 0 \, .
439: \label{symk}
440: \end{eqnarray}
441: Then we again expand this relation for $\kappa N \ll 1$ up to the
442: fifth power and determine $\kappa$ from the biquadratic equation:
443: $d \kappa^4 + g \kappa^2 + f =0$, where
444: \begin{gather}\nonumber
445: d = \frac{ 5 N (\lambda + \lambda_d) + 10 N^3 (\lambda + \lambda_d
446: ) -\lambda(5 N^4 + 10 N^2 +1)}{60} \, , \\ \nonumber f = N
447: (\lambda + \lambda_d) - (N^2 + 1/3) \lambda \,, \\ \nonumber g =
448: -2 \lambda \, .
449: \end{gather}
450: In the region $\lambda \ll N^{-1}$ one obtains:
451: \begin{equation}\label{om+0}
452: \omega_0^2 \approx 1 - 2 \lambda_d^{-1} \lambda N^{-1} \, ,
453: \qquad \omega_2^2 \approx 1 + 6 \lambda N^{-2} \, .
454: \end{equation}
455: The dependencies for symmetric modes have neither an inflection
456: point nor any peculiarity at $\lambda = \lambda_d$.
457:
458: For the infinite system the relation for $\kappa$ is written as
459: follows:
460: \begin{equation}
461: \mathrm{e}^{- \kappa} + 2 \pi^2 \lambda - 1 - 4 \lambda \sinh^2
462: \frac{\kappa}{2} = 0 \, .
463: \end{equation}
464: Then we arrive at the dependence for PN-mode for weak coupling in
465: the form:
466: \begin{equation}\label{om0inf}
467: \omega_0^2 \approx 1 - 4 \lambda_d^{-2} \lambda^3 \, ,
468: \end{equation}
469: which is in agreement with this dependence given in
470: Ref.\cite{bkv90}. The localization distance now is: $l_0 =
471: \kappa_0^{-1} \sim \lambda^{-3/2}$. Thus we can infer that the
472: localization for PN-mode with the growth of coupling value
473: $\lambda$ develops faster than for antisymmetric internal mode, as
474: it was pointed above while considering small size systems (cf.
475: Fig.\ref{f2}, right panels).
476:
477: The results for the frequency dependencies
478: (\ref{om1inf},\ref{om0inf}) could also be obtained with the use of
479: Green function (Lifshitz) technique for a linear chain with
480: impurities, (see the monograph \cite{k99}), which was employed in
481: Refs.\cite{bkp97,bkv90}. However, for the sake of getting simple
482: analytical expressions one should replace the kink solution with
483: the finite number of effective impurities. In the case of two
484: impurities model, utilized in the current paper, one would obtain
485: {\em exactly the same results}. In fact, the usage of the two
486: impurities model means that we retain only leading terms in powers
487: of parameter $\lambda$.
488:
489: To sum up, we investigated the features of spectrum and kink
490: internal modes of DSGE focusing on the effect of detachment of an
491: internal mode from the quasicontinuous spectrum. We note that the
492: two-impurities model, used for our analysis, gives a somewhat
493: different value for the detachment point ($\lambda_d = \pi^{-2}$,
494: while the numerically estimated value for DSGE is $\lambda_d
495: \approx 0.26$). This is explained by the fact that the value of
496: $\lambda_d$ is nonzero, whereas we had to use the asymptotical
497: expansions valid for $\lambda \to 0$. This means the higher order
498: terms with respect to $\lambda$ should always result in the
499: corrections to the analytically obtained value of $\lambda_d$.
500: However the two impurities model allowed us to obtain analytical
501: results describing the spectrum peculiarities, which are in good
502: agreement (qualitative and even quantitative for large $N$) with
503: the numerical ones. For the infinite system the dependence of
504: first antisymmetric mode detaches smoothly (with zero first
505: derivative) from the band edge, but for the finite system it does
506: cross the bandgap threshold. For extremely small values of
507: $\lambda$ the first mode belongs to the spectrum. The initial weak
508: growth ($\sim N^{-2}$) of lowest antisymmetric modes frequencies
509: with the value of $\lambda$ then changes to more rapid ($\sim
510: N^{-1}$) decreasing. The extremum point $\lambda_{\mathrm{ext}}$
511: approaches the detachment point $\lambda_d$ as the number of sites
512: gets bigger. At $\lambda = \lambda_d$ the first antisymmetric mode
513: drops into the spectrum gap. The higher antisymmetric mode come
514: closer and closer to the preceding symmetric modes with the
515: increase of coupling but this tendency weakens with the growth of
516: mode number. The symmetric modes do not display any peculiarity at
517: $\lambda_d$.
518: \begin{thebibliography}{99}
519: \bibitem{kpc98} Yu.S. Kivshar, D.E. Pelinovsky, T. Cretegny, M.
520: Peyrard, Phys. Rev. Lett. \textbf{80}, 5032 (1998).
521: \bibitem{bk03} O.M Braun, Yu.S. Kivshar, { \em The Frenkel-Kontorova
522: model} (Srpinger-Verlag, Berlin, 2004); see also O.M Braun, Yu.S.
523: Kivshar, Phys. Rep. \textbf{306}, 1 (1998).
524: \bibitem{bkp97} O.M. Braun, Yu.S. Kivshar, M. Peyrard, Phys. Rev. E
525: \textbf{56}, 6050 (1997).
526: \bibitem{bkv90}
527: M.M. Bogdan, A.M. Kosevich, V.P. Voronov, in { \em Proc. of IV
528: International Workshop "Solitons and applications"}, edt. V.G.
529: Makhankov, V.K. Fedyanin, O.K. Pashaev, World Scientific,
530: Singapore, 231 (1990).
531: \bibitem{kj00} P.G. Kevrekidis, C. K. R. T. Jones, Phys. Rev. E
532: \textbf{61}, 3114 (2000).
533: \bibitem{fk96} S. Flach, K. Kladko, Phys. Rev. E \textbf{54}, 2912 (1996).
534: \bibitem{k78} A.S. Kovalev, Theor. Math. Phys. \textbf{37}, 135 (1978).
535: \bibitem{kp03} A.S. Kovalev, J.E. Prilepsky, Low Temp. Phys.
536: \textbf{29}, 138 (2003) [A.S. Kovalev, J.E. Prilepsky, Fiz. Nizk.
537: Temp. \textbf{29}, 189 (2003)].
538: \bibitem{k99} A.M. Kosevich,{\em The crystal lattice: Phonons,
539: Solitons, Dislocations} (Wiley-Vch, Berlin, 1999).
540:
541: \end{thebibliography}
542:
543: \end{document}
544: