1: %\documentclass[aps,showpacs,pre,twocolumn,superscriptaddress]{revtex4}
2: %\usepackage{graphicx}
3: \documentclass{elsart}
4: %\documentclass[doublespacing]{elsart}
5: \usepackage{epsfig}
6: \newcommand{\be}{\begin{equation}}
7: \newcommand{\ee}{\end{equation}}
8:
9: \begin{document}
10:
11: \begin{frontmatter}
12:
13: \title{Multi-peaked localized states of DNLS in one and two dimensions}
14:
15: \author{G. Kalosakas \corauthref{cor1}}
16: \corauth[cor1]{Fax: +49-351-871-1999. E-mail: georgek@pks.mpg.de}
17: \address{Max Planck Institute for the Physics of Complex
18: Systems, \\ N\"othnitzer Str. 38, Dresden, 01187, Germany}
19:
20:
21: %\date{\today}
22:
23:
24: \begin{abstract}
25:
26: Multi-peaked localized stationary solutions of the discrete
27: nonlinear Schr\"odinger (DNLS) equation are presented in
28: one (1D) and two (2D) dimensions. These are excited states
29: of the discrete spectrum and correspond to multi-breather
30: solutions. A simple, very fast, and efficient numerical method,
31: suggested by Aubry, has been used for their calculation.
32: The method involves no diagonalization, but just iterations of a map,
33: starting from trivial solutions of the anti-continuous limit.
34: Approximate analytical expressions are presented and compared with
35: the numerical results. The linear stability of the calculated stationary
36: states is discussed and the structure of the linear stability spectrum
37: is analytically obtained for relatively large values of nonlinearity.
38:
39: \end{abstract}
40:
41: \begin{keyword}
42: DNLS \sep stationary states \sep multi-breathers \sep discrete spectrum
43: \PACS 05.45.-a \sep 02.70.-c \sep 03.75.Lm \sep 42.65.Tg
44: \end{keyword}
45: \end{frontmatter}
46:
47: %\maketitle
48:
49:
50: \section{Introduction}
51:
52:
53: The DNLS equation, Eq. (\ref{dnls}), has been extensively used
54: as a generic model for studying nonlinear effects (breathers,
55: for example) in a discrete system
56: \cite{ELS,eilb,MA,JA,FKM,HT,RABT,JA2,BBJ,BKRK,VMK,floria}. In addition,
57: specific applications have been proposed for the description of:
58: {\it i)} local intramolecular stretching vibrations in symmetric
59: polyatomic molecules \cite{lm},
60: {\it ii)} arrays of coupled nonlinear optical waveguides \cite{nopt},
61: {\it iii)} interacting electron-lattice models \cite{hol,KAT}
62: (or, equivalently, intramolecular excitation-phonon coupled systems
63: \cite{dav,acn,zolo}) in solid state physics,
64: where localized solutions of DNLS correspond to
65: polarons (vibrational polarons, respectively), and recently
66: {\it iv)} Bose-Einstein condensates \cite{TS,KRB}.
67:
68: In DNLS the evolution of a complex probability amplitude $\Psi_n$
69: at the site $n$ of a $d$-dimensional lattice is given by
70: \be \label{dnls}
71: i \frac{d \Psi_n}{dt} = -V \sum_{\delta} \Psi_{n+\delta}
72: + \chi |\Psi_n|^2 \Psi_n,
73: \ee
74: where $V$ represents nearest neighbor coupling,
75: $\chi$ is the strength of nonlinearity, and the sum over $\delta$
76: contains the nearest neighbors of the lattice site~$n$.
77: For example, in 1D is $\sum_{\delta} \Psi_{n+\delta}= \Psi_{n+1}+\Psi_{n-1}$,
78: while in a 2D square lattice, if each lattice site $n$ is represented
79: by a pair ($n_x,n_y$), is $\sum_{\delta} \Psi_{n+\delta}=
80: \Psi_{n_x+1,n_y}+\Psi_{n_x-1,n_y}+\Psi_{n_x,n_y+1}+\Psi_{n_x,n_y-1}$.
81:
82: The DNLS equation is derived from the Hamiltonian (assuming an
83: infinite lattice, or periodic boundary conditions)
84: \be \label{ham}
85: H = - \; V \; \sum_n \sum_{\delta} \Psi_n^\star \Psi_{n+\delta}
86: \; + \; \frac{\chi}{2} \sum_n |\Psi_n|^4.
87: \ee
88:
89: There exist two conserved quantities during the DNLS dynamics:
90: the Hamiltonian (\ref{ham}) and the norm
91: \be
92: N = \sum_n |\Psi_n|^2.
93: \ee
94:
95: A trivial transformation ($\Psi_n \rightarrow \Psi_n / \sqrt{N}$)
96: connects the solutions of DNLS
97: for an arbitrary norm $N$ with those normalized to unity, through a
98: rescalling of the nonlinearity ($\chi \rightarrow \chi \cdot N$).
99: Therefore, any solution $\Phi_n$ of Eq.(\ref{dnls}) with norm $N$
100: is obtained from a solution $\Psi_n$ with norm 1, through
101: \be \label{metN}
102: \Phi_n(t;V,\chi) = \sqrt{N} \cdot \Psi_n(t;V,\chi N)
103: \ee
104:
105: Furthermore, the magnitude of the hopping integral
106: $V$ can be considered unity by appropriate rescalling of time and $\chi$.
107: Changing sign in $V$ is equivalent to the transformation
108: $\Psi_n \rightarrow (-1)^n\Psi_n $ (or
109: $\Psi_{n_x,n_y} \rightarrow (-1)^{n_x+n_y} \Psi_{n_x,n_y}$ in 2D, etc.).
110: Therefore, in the following we
111: consider $N=1$ and $V=1$ without loss of the generality, while the
112: nonlinearity $\chi$ remains the only free parameter of the system.
113:
114:
115: \subsection{Stationary solutions of DNLS}
116:
117:
118: The stationary states of DNLS are characterized by a simple harmonic
119: evolution with frequency $\omega$:
120: \be \label{statsol}
121: \Psi_n = \psi_n \cdot e^{-i \omega t}.
122: \ee
123:
124: The time independent amplitudes $\psi_n$ satisfy the nonlinear
125: eigenvalue problem
126: \be \label{sdnls}
127: \omega \psi_n = - \sum_{\delta} \psi_{n+\delta} + \chi
128: |\psi_n|^2 \psi_n
129: \ee
130: (since we consider $V=1$).
131: The energy $E_s$ of a stationary state is related to its
132: frequency $\omega$ through
133: \be
134: E_s = \omega - \frac{\chi}{2} \sum_n |\psi_n|^4.
135: \ee
136:
137: There are two kinds of stationary solutions of Eq.~(\ref{sdnls}):
138: extended (like Bloch states and standing waves \cite{MJKA}) and
139: localized states. The Bloch states, $\psi_n^q \sim \exp(iqn)$, form
140: a band of frequencies or energies from $-2dV$ to $2dV$ (for an infinite
141: $d$-dimensional lattice). Localized states have discrete
142: frequencies outside of the Bloch band.
143:
144: The most obvious and well studied localized state is the single-peaked
145: solution, which has its maximum amplitude on one lattice site. This state
146: has extreme (minimal for negative $\chi$ and maximal for positive $\chi$)
147: energy and frequency, compared to the other localized states.
148: However, in general there are infinite (in an infinite lattice)
149: multi-peaked stationary solutions of DNLS (which, for example, can
150: be continued from trivial multi-peaked solutions of the anti-continuous
151: limit), resulting in a very rich discrete spectrum with many
152: quasi-degenerate levels (see below). In spite of this complexity, a part of
153: the spectrum can be understood by classifying the stationary states from the
154: anti-continuous limit. For the general concept of the anti-continuous,
155: or anti-integrable, limit in nonlinear lattices see Ref.~\cite{sergeAC}.
156: Here, multi-peaked and real stationary solutions of high symmetry, which
157: have direct counterparts at the anti-continuous limit, are presented.
158: Their location on the DNLS spectrum, their stability, and approximate
159: analytical expressions are discussed.
160:
161:
162: \section{Aubry's method for the numerical calculation of localized
163: stationary states of DNLS}
164:
165:
166: Localized stationary states can be obtained as attractors of the map:
167: \be \label{map}
168: \overline{\psi} \; \longrightarrow \; \overline{\psi'} =
169: \; sgn\chi \; \cdot \; \frac{ \aleph \{ \overline{\psi} \} }
170: {|| \aleph \{ \overline{\psi} \} ||}.
171: \ee
172: In this equation, $ \overline{\psi} = (\psi_1,\ldots,\psi_L)$,
173: where $L$ is the total number of lattice sites,
174: $sgn\chi$ denotes the sign of $\chi$, $\aleph \{ \overline{\psi} \}$
175: is defined through the right-hand-side of the stationary equation
176: (\ref{sdnls}), with its $n$-th component given by
177: \be
178: \aleph \{ \overline{\psi} \}_n = - \sum_{\delta} \psi_{n+\delta} + \chi
179: |\psi_n|^2 \psi_n,
180: \ee
181: and $|| \aleph \{ \overline{\psi} \} ||$ represents its
182: norm $\sqrt{\sum_{n=1}^{L} |\aleph \{ \overline{\psi} \}_n|^2}$.
183:
184: Starting the iterations from appropriate initial states, obtained through
185: trivial solutions of the anti-continuous limit, i.e.
186: \be
187: \psi_n^{(r=0)} = \delta_{n,n_0}, \hspace{0.5cm} \mbox{or} \hspace{0.5cm}
188: \psi_n^{(r=0)} =\frac{1}{\sqrt{2}}(\delta_{n,n_0} \pm \delta_{n,n_1}),
189: \hspace{0.5cm} \mbox{etc.,}
190: \ee
191: depending on the desired localized stationary solution (single-peaked,
192: double-peaked with interpeak separation $|n_0-n_1|$, etc.),
193: and iteratively applying the map (\ref{map})
194: \be
195: \psi_n^{(r+1)} = \; sgn\chi \; \cdot \; \frac{\aleph
196: \{\overline{\psi}^{(r)}\}_n }{|| \aleph \{\overline{\psi}^{(r)}\} ||},
197: \ee
198: the procedure can rapidly converge to the corresponding localized state.
199:
200: Up to now this method has been successfully applied for calculating
201: single-peaked stationary states of DNLS in 1D, 2D, and 3D \cite{KAT},
202: as well as for the single-peaked ground states
203: in other similar systems \cite{voul,VKBT,dirk,KH,HSAP}.
204: The method has been invented by S. Aubry for finding polarons
205: in the adiabatic Holstein model \cite{AQ,AAR,serge1}, a problem
206: which reduces to the stationary solutions of DNLS.
207:
208: In the following, after briefly recalling some results obtained in
209: Ref~\cite{KAT} regarding the stable single-peaked
210: stationary states of DNLS (section 3), this method is applied for
211: calculating multi-peaked stationary solutions in 1D (section 4) and
212: 2D (section 5). Stationary states are presented in sections 4 and 5
213: for negative values of $\chi$; the obtained solutions in this
214: case can be directly transformed to the corresponding ones at $-\chi$
215: (i.e. at positive nonlinearities) by changing the sign of the frequency
216: ($\omega \rightarrow -\omega$) and energy ($E_s \rightarrow -E_s$),
217: and making the transformation $\psi_n \rightarrow (-1)^n \psi_n$ (or
218: $\Psi_{n_x,n_y} \rightarrow (-1)^{n_x+n_y} \Psi_{n_x,n_y}$ in 2D, etc.)
219: in the wavefunction. However, analytical results and the general discussion
220: concern both signs of $\chi$.
221:
222:
223: \section{Single-peaked (SP) stationary states}
224:
225:
226: For negative (positive) nonlinearity the single-peaked solution of
227: DNLS corresponds to the lowest (highest) frequency stationary state.
228: In 1D there is always a SP state with extreme frequency
229: and energy, for any nonzero value of $\chi$. As $\chi$ is approaching
230: zero from negative (positive) values, the frequency and the energy
231: of the SP solution tends to the bottom (top) of the Bloch band.
232: The branch of SP solutions, as obtained by varying $\chi$, has
233: two well-known limits: for large $|\chi|$ (in the anti-continuous limit,
234: where the first term of the right-hand-side of Eq. (\ref{sdnls})
235: can be neglected) tends to a single-site localized state
236: ($\psi_n = \delta_{n,n_0}$), while for $|\chi| \rightarrow 0$ tends to
237: a solution obtained by the soliton of the continuous nonlinear
238: Schr\"odinger equation (see Eq. ~(\ref{nlsw}) below).
239:
240: The picture is qualitatively different in 2D and 3D. In these cases
241: there is a critical value of nonlinearity $\chi_1 >0$, such that
242: for $|\chi|<\chi_1$ does not exist a single-peaked, or any other
243: localized, stationary state. At $|\chi|=\chi_1$ a pair of SP
244: states appears, through a saddle-node bifurcation; a narrow stable
245: solution of high amplitude and an unstable one of relatively large
246: extent and small amplitude. Due to the simultaneous existence of two
247: SP states and the proximity of the unstable one with the extended
248: Bloch states of the band edges,
249: it is not paid any attention to the unstable solution and from now on
250: we exclusively refer to the stable one wherever a single-peaked state
251: is mentioned in 2D or 3D. A second nonlinearity threshold $\chi_2 >\chi_1$
252: exists, such that for $\chi_1<|\chi|<\chi_2$ the SP state has
253: extreme frequency, but not extreme energy, since its energy $E_s$ lies
254: inside the Bloch band. Only for $|\chi|>\chi_2$ the single-peaked stationary
255: solution provides an extreme of the energy (i.e. it is the ground state
256: for negative $\chi$). In 2D the values of these thresholds are
257: $\chi_1 \approx 5.701$ and $\chi_2 \approx 6.679$, while in 3D are
258: $\chi_1 \approx 7.852$ and $\chi_2 \approx 10.816$ \cite{KAT}\footnote{
259: Note that the nonlinear parameter $\chi$ used in this work
260: corresponds to $-k^2$ of Ref.~\cite{KAT}.}.
261:
262: Analytical approximate expressions have been presented in Ref. \cite{KAT},
263: which accurately describe the SP stationary states.
264: In particular, for the whole branch of SP solutions in 2D and
265: 3D, as well as for values of $|\chi|$ larger than about 3 in 1D,
266: the exponentially decaying function
267: \be \label{vap}
268: \psi_{n_x,n_y,n_z}^{SP} = \left( \frac{1-\zeta^2}{1+\zeta^2} \right) ^{d/2}
269: \cdot \zeta^{|n_x|+|n_y|+|n_z|}, \hspace{1.0cm} \mbox{with} \hspace{0.2cm}
270: \zeta = -\frac{1}{\chi} - \frac{4d-2}{\chi^3}
271: \ee
272: (where one has to disregard the index $n_z$ in 2D and both $n_z$ and
273: $n_y$ in 1D), can be used to describe the exact SP solution.
274: The corresponding frequencies and energies are given by
275: \be
276: \omega = \chi - \frac{2d(4d-3)}{x^3} \hspace{1.0cm} \mbox{and}
277: \hspace{1.0cm} E_s = \frac{\chi}{2}+ \frac{2d}{\chi}+ \frac{d(4d-3)}{\chi^3}.
278: \ee
279: The expression (\ref{vap}) is derived from a variational method, by
280: employing a perturbative expansion of $\zeta$ in powers of $1/\chi$
281: in the condition providing the minima of the variational energy.
282: The next non-zero correction of $\zeta$ in Eq.(\ref{vap}) is of the order
283: of $1/\chi^5$.
284:
285: The above analytical results describe accurately (the larger the $|\chi|$
286: the better the approximation) the SP solutions of DNLS in all
287: cases, except for relatively small values of $|\chi|$ in 1D. Then
288: a smooth transition occurs for $|\chi|$ in the region $2.5-3$, from the
289: above expressions to the static soliton of the continuous nonlinear
290: Schr\"odinger equation. Therefore, for $|\chi|$ smaller than about 2.5
291: the SP solutions in 1D can be approximated by
292: \be \label{nlsw}
293: \psi_n^{SP} = (-sgn \chi )^n \; \sqrt{\frac{|\chi|}{8}}
294: \cdot \frac{1}{\cosh \frac{\chi n}{4} },
295: \ee
296: with corresponding frequency and energy
297: \be
298: \omega = sgn \chi \cdot \left( 2 + \frac{\chi^2}{16} \right)
299: \hspace{1.0cm} \mbox{and} \hspace{1.0cm}
300: E_s = sgn \chi \cdot \left( 2 + \frac{\chi^2}{48} \right).
301: \ee
302: Note that the $(-sgn\chi)^n$ term in (\ref{nlsw}) provides the alteration
303: of signs in successive lattice sites, which characterizes the solution
304: at positive $\chi$. In Eq.~(\ref{vap}) this is obtained through the
305: negative sign of $\zeta$.
306:
307:
308: \section{Multi-peaked solutions in 1D}
309:
310:
311: \subsection{Frequency spectrum}
312:
313: As it has been already mentioned in the introduction, the frequency (or
314: energy) spectrum of DNLS in 1D comprises many discrete levels, corresponding
315: to single-peaked and multi-peaked localized stationary solutions.
316:
317: These discrete levels have well determined limits in the
318: anti-continuous regime. In this limit the discrete spectrum is given by
319: \be \label{acsp}
320: \omega = \frac{\chi}{M}, \hspace{0.5cm} E_s = \frac{\chi}{2M},
321: \hspace{0.5cm} \mbox{where} \; M=1,2,3, \ldots \; \; .
322: \ee
323: Each (highly degenerate) level of Eq.(\ref{acsp}) corresponds to any
324: $M$-peaked stationary state, where all the $M$ peaks (at arbitrary sites)
325: have the same norm $1/\sqrt{M}$ (and arbitrary complex phases).
326: Such stationary solutions can be continued away from the anti-continuous
327: limit \cite{AAR} up to some value of $\chi$, depending on the
328: particular state.
329:
330:
331: \begin{figure}
332: \centerline{\hbox{\psfig{figure=Fig1dfreq.eps,width=10cm,height=7.5cm}}}
333: \caption{Frequencies of single- double- and triple-peaked stationary
334: solutions of DNLS in 1D (points). Dashed lines show analytical
335: expressions obtained for large values of $|\chi|$.
336: The horizontal line at $\omega=-2$ indicates the lower edge of the
337: band of Bloch stationary states, which extends from $-2$ to 2.
338: The spectrum is antisymmetric on $\chi$; $\omega (-\chi)=-\omega (\chi)$.}
339: \label{figFS1}
340: \end{figure}
341:
342:
343: A stationary solution at a given value of $\chi$ is classified as a
344: single- double- or, in general, $M$-peaked state, depending on how
345: many sites (one, two, or in general $M$, respectively) are occupied
346: at the anti-continuous limit of the branch in which this state belongs.
347: Such a classification is always possible for stationary states of high
348: symmetry. Moreover, it can be used for {\it each} stationary solution,
349: providing a complete description of the discrete spectrum, up to $\omega
350: \approx -5.45$ \cite{ABK}.
351:
352: Fig.~\ref{figFS1} shows a part of the frequency spectrum
353: which contains contributions from many double-peaked solutions, as well
354: as few branches (just shown for indication) of triple-peaked states.
355: Analytical expressions, obtained at large values of $|\chi|$ and shown
356: by dashed lines, describe well these branches (the next corrections are of
357: the order of inverse powers of $\chi$, see Appendix).
358: Bifurcations lead to the disappearance of some branches (or
359: merging with other branches) by decreasing the strength of nonlinearity.
360:
361: What appears as a middle branch of the double-peaked (DP)
362: solutions in Fig.~\ref{figFS1} actually consists of
363: many branches of closely spaced levels (corresponding to DP
364: solutions with any interpeak separation larger than one lattice constant),
365: which merge to the level of Eq.(\ref{acsp}) for $M=2$ as $|\chi|$
366: increases. The lower and upper branch of the DP solutions are
367: single branches corresponding to symmetric and antisymmetric, respectively,
368: double-peaked states with their peaks at neighboring lattice sites
369: (examples are shown at the upper left plots of Figs. \ref{fDS1} and
370: \ref{fDA1}, respectively). The interpeak separation of a DP state
371: is determined by the distance $S$ of the sites where the peaks appear.
372: Symmetric and antisymmetric DP solutions in 1D, with various
373: interpeak distances, are presented in the following subsection.
374:
375:
376: \subsection{Double-peaked symmetric and antisymmetric solutions}
377:
378: \subsubsection{Stationary states}
379:
380: \begin{figure}
381: \begin{center}
382: \begin{tabular}{cc}
383: \epsfig{file=Fig1d2S1.eps,height=6.0cm,width=6.8cm} & \hspace{-0.3cm}
384: \epsfig{file=Fig1d2S10.eps,height=6.0cm,width=6.8cm} \\
385: \epsfig{file=Fig1d2S2.eps,height=6.0cm,width=6.8cm} & \hspace{-0.3cm}
386: \epsfig{file=Fig1d2S20.eps,height=6.0cm,width=6.8cm} \\
387: \end{tabular}
388: \end{center}
389: \caption{Double-peaked symmetric stationary solutions of DNLS in 1D
390: (points) for different values of the nonlinearity $\chi$ and various
391: interpeak separations $S$: {\bf (a)} $S=1$ lattice site, {\bf (b)} $S=2$
392: sites, {\bf (c)} $S=10$ sites, and {\bf (d)} $S=20$ sites. Dashed lines
393: show analytical approximations of the solutions using Eq.~(\ref{dpd}) for
394: the discrete cases where $|\chi|>6$, Eq.~(\ref{dpc}) for $\chi=-5.2$ in (c)
395: and $\chi=-4$ in (d), and Eq.~(\ref{nlsw}) centered in the middle between the
396: sites of maximum amplitude for $\chi=-0.5$ up to $-5$ in (a) (see text). }
397: \label{fDS1}
398: \end{figure}
399:
400:
401: In order to find the branches of symmetric and antisymmetric
402: DP solutions, the method described in section 2 is applied
403: using as initial states
404: \be
405: \psi_n^{(r=0)}= \frac{1}{\sqrt{2}}(\delta_{n,n_1} + \delta_{n,n_2})
406: \hspace{1.0cm} \mbox{and} \hspace{1.0cm}
407: \psi_n^{(r=0)}= \frac{1}{\sqrt{2}}(\delta_{n,n_1} - \delta_{n,n_2}),
408: \ee
409: respectively. The distance $|n_2-n_1|$ determines the interpeak separation
410: $S$ of the corresponding solution.
411:
412:
413: \begin{figure}
414: \begin{center}
415: \begin{tabular}{cc}
416: \epsfig{file=Fig1d2A1.eps,height=6.0cm,width=6.8cm} & \hspace{-0.3cm}
417: \epsfig{file=Fig1d2A10.eps,height=6.0cm,width=6.8cm} \\
418: \epsfig{file=Fig1d2A2.eps,height=6.0cm,width=6.8cm} & \hspace{-0.3cm}
419: \epsfig{file=Fig1d2A20.eps,height=6.0cm,width=6.8cm} \\
420: \end{tabular}
421: \end{center}
422: \caption{Double-peaked antisymmetric stationary solutions of DNLS in 1D
423: (points) for different values of the nonlinearity $\chi$ and various
424: interpeak separations $S$: {\bf (a)} $S=1$ lattice site, {\bf (b)} $S=2$
425: sites, {\bf (c)} $S=10$ sites, and {\bf (d)} $S=20$ sites. Dashed lines
426: show analytical approximations of the solutions using Eq.~(\ref{dpd}) in the
427: discrete cases where $|\chi|>6$ and Eq.~(\ref{dpc}) in the two more extended
428: cases of (c) and (d).}
429: \label{fDA1}
430: \end{figure}
431:
432:
433: Stationary solutions with interpeak separations 1, 2, 10, and 20 sites
434: are shown for different values of $\chi$ in Fig.~\ref{fDS1} for symmetric
435: and in Fig.~\ref{fDA1} for antisymmetric states.
436: As it is expected, by decreasing $|\chi|$ the solutions spread
437: more and more, until a value of $\chi$ where a bifurcation occurs
438: resulting in the disappearance of the corresponding branch.
439: The smaller values of $|\chi|$ shown at each plot of Fig.~\ref{fDS1}
440: (apart from the case of $S=1$, displayed in Fig.~\ref{fDS1}a) and
441: Fig.~\ref{fDA1} are close to the bifurcation point and therefore represent
442: one of the most extended solutions of the corresponding branch.
443: The case of symmetric states with interpeak distance equal to one lattice
444: constant (Fig.~\ref{fDS1}a) is exceptional in this respect, since the
445: corresponding branch survives until the limit $|\chi| \rightarrow 0$, as it
446: merges to the SP branch for sufficiently small values of $|\chi|$.
447: As the continuous limit is approached, the DP symmetric state with $S=1$
448: becomes practically indistinguishable from the
449: SP state, i.e. the solution given by Eq.~(\ref{nlsw}).
450:
451: The symmetric (antisymmetric) states with $S=1$ provide the lower (upper)
452: single branch of the DP solutions of Fig.~\ref{figFS1},
453: with frequency equal to $\omega = \frac{\chi}{2}-1$
454: ($\omega = \frac{\chi}{2}+1$), for relatively large $|\chi|$. All the
455: other symmetric and antisymmetric DP solutions with interpeak
456: separations $S>1$ are congested in the middle branch (with
457: frequency equal to $\omega = \frac{\chi}{2}$ for large $|\chi|$).
458: In particular, in Fig.~\ref{figFS1} the middle branch contains
459: solutions for $S=2$, 3, 4, 5, 10, 20, and 200.
460: As the interpeak separation $S$ increases, the two peaks start to
461: not overlap much, even for small values of $|\chi|$, and in this case
462: the corresponding branch survives longer (i.e. persists closer to $\chi=0$).
463: These branches, as $|\chi|$ decreases, eventually deviate from the line
464: $\omega = \frac{\chi}{2}$, and for even smaller values of $|\chi|$ they
465: can be described by double-peaked solutions of the continuous nonlinear
466: Schr\"odinger equation.
467:
468: Approximate analytical expressions, which describe the DP
469: states in the most of the cases, can be obtained by appropriate
470: superpositions of the single-peaked states (\ref{vap}) and (\ref{nlsw}).
471: One has to take into account that when the two peaks do not significantly
472: overlap the norm of each peak is about half of the total norm. Then
473: from Eq.~(\ref{metN}) follows that the two individual wavefunctions
474: superimposed in a double-peaked solution should be provided by Eqs.
475: (\ref{vap}) or (\ref{nlsw}) corresponding to $\frac{\chi}{2}$. As a result
476: a symmetric or antisymmetric DP solution with interpeak separation $S$,
477: where the first peak is located at the site $n_1$ and the second at
478: $n_2=n_1+S$ ($S$ is assumed to be positive), can be approximated by
479: \be \label{dpd}
480: \psi_n^{DP} = \frac{1}{\sqrt{2(1\pm P)}} \sqrt{\frac{1-\zeta^2}{1+\zeta^2}}
481: \left( \zeta^{|n-n_1|} \pm (-sgn \chi)^S \zeta^{|n-n_1-S|} \right),
482: \ee
483: \be \label{ov1}
484: \mbox{where} \hspace{0.3cm}
485: P = (-sgn \chi)^S \; \frac{(1+S) \zeta^S - (S-1) \zeta^{S+2}}{1+\zeta^2}
486: \ee
487: \be
488: \mbox{and} \hspace{0.3cm} \zeta = -\frac{1}{\chi/2} - \frac{2}{(\chi/2)^3}
489: = -\frac{2}{\chi} - \frac{16}{\chi^3},
490: \ee
491: or
492: \be \label{dpc}
493: \psi_n^{DP} = \frac{(-sgn \chi)^{n-n_1}}{\sqrt{2(1\pm P)}}
494: \sqrt{\frac{|\chi|}{16}} \left( \frac{1}{\cosh \frac{\chi (n-n_1)}{8}}
495: \pm \frac{(-sgn \chi)^S}{\cosh \frac{ \chi (n-n_1-S)}{8}} \right),
496: \ee
497: \be \label{ov1c}
498: \mbox{where} \hspace{0.3cm}
499: P= (-sgn \chi)^S \; \frac{ \frac{\chi S}{8}}{\sinh \frac{\chi S}{8}}.
500: \ee
501: The former (latter) solution can be used for a DP state
502: with discrete (rather extended) peaks, valid for relatively large
503: (small) values of $|\chi|$. Roughly speaking the transition
504: from one form to the other occurs for $|\chi|$ in the region $5-6$.
505: The normalization factors $P$ in Eqs. (\ref{ov1}) and (\ref{ov1c}) result
506: from the non-orthogonality of the superimposed wavefunctions, i.e.
507: $P=\sum_n \psi_n^{SP[n_1]}(\frac{\chi}{2}) \psi_n^{SP[n_1+S]}(\frac{\chi}{2})$,
508: where $\psi_n^{SP[m]}(\frac{\chi}{2})$ denotes the SP solution
509: centered at the site $m$ and calculated for the value $\frac{\chi}{2}$
510: of the nonlinearity parameter. The plus (minus) signs in these approximate
511: solutions correspond to symmetric (antisymmetric) states, except for the
512: case of positive $\chi$ and odd $S$, where they give the antisymmetric
513: (symmetric) DP state.
514:
515: Eq.~(\ref{dpc}) is not applicable for small values of $|\chi|$ in the
516: exceptional case of symmetric states with $S=1$, due to the significant
517: overlap of the two non-distinguished peaks (Fig.~\ref{fDS1}a for $|\chi|$
518: approximately less than 5-6). In this case the
519: corresponding DP solution, whose branch is merging to the
520: single-peaked branch, can be well described by Eq.~(\ref{nlsw})
521: centered in the middle between the consecutive sites of the two peaks.
522: The analytical approximations discussed above have been plotted in
523: Figs. \ref{fDS1} and \ref{fDA1} (dashed lines) along with the numerical
524: solutions (points) for comparison. In particular, Eq.~(\ref{nlsw}) has been
525: used, as just explained, in
526: Fig.~\ref{fDS1}a for $\chi=-0.5$, $-1$, $-2$, and $-5$, Eq.~(\ref{dpc})
527: has been used only for the smallest values of $|\chi|$ in Figs.
528: \ref{fDS1}c, \ref{fDS1}d, \ref{fDA1}c, and \ref{fDA1}d (where
529: $|\chi|<6$), and Eq.~(\ref{dpd}) has been plotted in all the other
530: cases. The more distinguished the two peaks, i.e. the higher the
531: $S$, or the higher the $|\chi|$ even for smaller values of $S$, the better
532: the approximations (\ref{dpd}) and (\ref{dpc}) are.
533:
534:
535: \subsubsection{Linear stability}
536:
537:
538: Symmetric and antisymmetric DP states show qualitatively different
539: behavior regarding their stability. A detailed investigation of the
540: stability eigenvalues is presented below for negative values of $\chi$.
541: From this, the behavior at positive $\chi$ can be obtained as follows:
542: for even $S$ the situation, regarding the stability eigenvalues,
543: is exactly the same as that of $\chi<0$, while
544: for odd $S$ the symmetric (antisymmetric) states behave exactly like
545: the antisymmetric (symmetric) states of $\chi<0$. Note here that the
546: transformation $(-1)^n \psi_n$, connecting stationary solutions at opposite
547: values of $\chi$, turns a symmetric DP state to antisymmetric and
548: vis versa when $S$ is odd. Therefore, what is mentioned below for $\chi<0$,
549: also holds as it is for $\chi>0$ when $S$ is even, while it valids after
550: interchanging roles between symmetric and antisymmetric states when $S$
551: is odd.
552:
553: The eigenvalues of the linear stability problem are always obtained as pairs
554: of opposite sign and there always exists a pair of eigenvalues at zero.
555: In our approach (see Appendix), if there is an eigenvalue with non-zero
556: imaginary part, then the stationary solution is unstable. All relevant
557: discussions in this and the following section
558: will refer to those eigenvalues with non-negative real part, without usually
559: mentioning the pinned pair of eigenvalues at zero. The spectrum of
560: eigenvalues is always symmetric with respect to the imaginary axis.
561: The complete structure of the linear stability spectrum close to the
562: anti-continuous limit is analytically derived in the Appendix.
563:
564: For negative $\chi$, the nonlinear term of DNLS has the same sign as the
565: tunneling term. Then it is known that the symmetric DP states are unstable
566: \cite{LKS,KKM}, in contrary to the case of the antisymmetric ones
567: that may be linearly stable \cite{JA,PKF}. In fact, the latter
568: are linearly stable in the larger part of the corresponding branch.
569: When this branch becomes unstable, by decreasing $|\chi|$,
570: soon it disappears. The larger the interpeak distance, the smaller the
571: $|\chi|$'s at which the branch turns unstable and disappears. The scenario
572: for development of instability and disappearance seems to roughly be as
573: follows. For large $|\chi|$, linear stability analysis provides,
574: a discrete eigenvalue outside of the band.
575: For relatively large $|\chi|$ the band extends from $\frac{|\chi|}{2}-2$ to
576: $\frac{|\chi|}{2}+2$, apart from the case of symmetric (antisymmetric) states
577: with interpeak separation $S=1$, where it extends from $\frac{|\chi|}{2}-1$
578: to $\frac{|\chi|}{2}+3$ (from $\frac{|\chi|}{2}-3$ to $\frac{|\chi|}{2}+1$).
579: As $|\chi|$ decreases, the band moves towards zero, while the discrete
580: eigenvalue keeps off zero for $S>2$, remains almost constant for $S=2$, and
581: goes to zero (but slower than the band) for $S=1$. After their collision (or
582: the collision of the discrete eigenvalue with another eigenvalue splitted off
583: the band) the instability develops.
584: As $|\chi|$ decreases more, the band is approaching zero and the
585: branch disappears when an eigenvalue splitted from the band collides
586: with the pinned eigenvalue at zero. Table~1 shows for antisymmetric
587: DP solutions of various interpeak separations (first column)
588: the nonlinearity regime where the corresponding branch becomes unstable
589: (second column; for larger values of $|\chi|$ the solutions are linearly
590: stable) and the nonlinearity regime where the branch disappears
591: (third column).
592:
593: \begin{table}
594: \caption{Nonlinearity strength $\chi$ for the development of instability and
595: disappearance of antisymmetric and the disappearance of symmetric double-peaked
596: stationary solutions of DNLS at different interpeak separations. Values in
597: parentheses in the second column show rough analytical estimates using
598: Eq.~(\ref{antun}).}
599: \centerline{
600: \begin{tabular}{|c|c|c|c|} \hline
601: Interpeak & Regime of $\chi$ at which & Regime of $\chi$ at which
602: & Regime of $\chi$ at which \\ separation & the antisymmetric branch &
603: the antisymmetric & the symmetric \\
604: & becomes unstable & branch disappears & branch disappears \\ \hline
605: $S=1$ & $[-19.73,-19.72]$ \hspace{0.05cm} ($-18$) & $[-9.58,-9.57]$ & $-$
606: \\ $S=2$ & $[-8.8,-8.7]$ \hspace{0.5cm} ($-8$) & $[-6.54,-6.53]$ &
607: $[-9.63,-9.62]$ \\ $S=3$ & $[-7.1,-7.0]$ \hspace{0.5cm} ($-6.3$) &
608: $[-5.9,-5.8]$ & $[-7.9,-7.8]$ \\ $S=4$ & $[-6.5,-6.4]$ \hspace{0.5cm}
609: ($-5.5$) & $[-5.7,-5.6]$ & $[-7.0,-6.9]$ \\ $S=5$ & $[-6.2,-6.1]$
610: \hspace{0.5cm} ($-5.0$) & $[-5.5,-5.4]$ & $[-6.5,-6.4]$ \\ $S=10$ &
611: $[-5.0,-4.9]$ \hspace{0.5cm} ($-4.2$) & $[-4.9,-4.8]$ & $[-5.2,-5.1]$ \\
612: $S=20$ & $[-3.90,-3.89]$ \hspace{0.15cm} ($-4.0$) & $[-3.90,-3.89]$ &
613: $[-4.0,-3.9]$
614: \\ \hline
615: \end{tabular} }
616: \end{table}
617:
618:
619: Regarding the unstable symmetric DP states, linear stability
620: analysis reveals that there is always one pair of purely imaginary unstable
621: eigenvalues. If the magnitude of instability is denoted by $\lambda_u$
622: (i.e. the unstable eigenvalues are $\pm i \lambda_u$),
623: then for fixed value of $\chi$, $\lambda_u$ decreases
624: as the interpeak separation increases. For fixed interpeak distance $S$,
625: but larger than two lattice sites, $\lambda_u$ decreases by increasing
626: $|\chi|$. For $S=2$, $\lambda_u$ tends to a constant value for large $|\chi|$,
627: while for $S=1$, $\lambda_u$ decreases with decreasing $|\chi|$, in accordance
628: with the fact that this branch merges to the stable single-peaked branch for
629: $|\chi| \rightarrow 0$. This behavior is shown in Fig.~\ref{fstab}.
630: In some cases, especially for large interpeak separations, the magnitude
631: of instability is so small that for any practical purpose the corresponding
632: solution can be considered as quasi-stable \cite{LKS}.
633: The log-log plot of $\lambda_u$ as a function of $|\chi|$ in Fig.~\ref{fstab}
634: demonstrates a power-law dependence $\lambda_u \sim |\chi|^a$ at large
635: values of $|\chi|$. Relating the stability eigenvalues with the energy
636: spectrum of a tight-binding problem in the presence of a deep and narrow
637: double-well potential (see Appendix), one obtains
638: \be \label{inst1}
639: \lambda_u = 2^{S/2} \; |\chi|^{1-\frac{S}{2}}, \hspace{1.0cm}
640: \mbox{for} \; \; |\chi| \gg 1.
641: \ee
642: This relation is plotted in Fig.~\ref{fstab} for $S=$ 1, 2, 3, 4, 10, 15, 20,
643: and 30 (solid lines) along with the corresponding numerical results (points).
644: The agreement is very good for $|\chi|$ larger than $10-20$.
645:
646:
647: \begin{figure}
648: \centerline{\hbox{\psfig{figure=Fig1dstab.eps,width=10cm,height=7.5cm}}}
649: \caption{Magnitude of instability $\lambda_u$ of symmetric double-peaked
650: solutions of DNLS for $\chi<0$ in 1D (points) as a function of the strength
651: of nonlinearity
652: $|\chi|$, for various interpeak separations: $S=1$ (filled circles), $S=2$
653: (filled squares), $S=3$ (filled diamonds), $S=4$ (filled triangles), $S=10$
654: (open circles), $S=15$ (open squares), $S=20$ (open triangles), $S=30$
655: (open diamonds), and $S=50$ (crosses). Lines show the power-law relation,
656: Eq.(\ref{inst1}), derived for relatively large values of $|\chi|$.}
657: \label{fstab}
658: \end{figure}
659:
660: The disappearance of the symmetric branches occurs for non-zero $\chi$
661: (apart from the case of $S=1$, where the corresponding branch merges with
662: the SP branch) when, as in the antisymmetric case, a real
663: eigenvalue splitted from the band collides with the pinned eigenvalues
664: at zero. Table~1 shows (fourth column) the nonlinearity regime where
665: different branches of symmetric DP solutions disappear. It seems that for
666: fixed $S$ ($S>1$), the antisymmetric branches survive until smaller
667: values of $|\chi|$.
668:
669: As it is shown in the Appendix, close to the anti-continuous limit the
670: stable (real) discrete eigenvalue of an antisymmetric DP state has the same
671: dependence like in Eq.~(\ref{inst1}) (see Eq.~(\ref{steig})). Numerical
672: simulations confirm that for relative large $|\chi|$ the stable eigenvalues
673: of antisymmetric states have the same magnitude with the unstable of the
674: symmetric ones. This result can be used for a rough estimate of the
675: nonlinearity value, $\chi_{un}$, where an antisymmetric state becomes
676: unstable (through the collision of the discrete eigenvalue with the band), by
677: \be \label{antun}
678: \frac{|\chi|}{2}-2= 2^{S/2} \; |\chi|^{1-\frac{S}{2}} \hspace{0.4cm}
679: \mbox{for} \; \; S>1, \hspace{0.6cm} \mbox{or} \hspace{0.6cm}
680: \frac{|\chi|}{2}-3=\sqrt{2|\chi|} \hspace{0.4cm} \mbox{for} \; \; S=1,
681: \ee
682: where $\frac{|\chi|}{2}-2$ ($\frac{|\chi|}{2}-3$) is the lower band edge.
683: The larger the $|\chi_{un}|$ the better the estimate, since Eq.~(\ref{inst1})
684: is valid for $|\chi| \gg 1$. Estimates of $\chi_{un}$, resulting from the
685: solution of Eq.~(\ref{antun}), are shown in the second column of Table~1
686: inside parentheses, next to the numerical results. The relative error is less
687: than $10\%$ for $S=1$ and $S=2$, but it increases for
688: larger $S$ where $|\chi_{un}|$ is getting smaller.
689:
690:
691: \section{Multi-peaked solutions in 2D}
692:
693:
694: \subsection{Frequency spectrum}
695:
696:
697: Similarly to the 1D case, also in 2D there are many families of multi-peaked
698: localized stationary states corresponding to discrete levels in the
699: frequency spectrum. For large values of $|\chi|$ these levels tend
700: to the anti-continuous limit spectrum.
701: Fig.~\ref{fFS2} shows a part of the frequency spectrum in 2D, which,
702: apart from the single-peaked states of lowest frequency,
703: contains many branches of double-peaked and quadruple-peaked (QP)
704: stationary solutions at various interpeak distances.
705:
706:
707: \begin{figure}
708: \centerline{\hbox{\psfig{figure=Fig2dfreq.eps,width=9.9cm,height=7.5cm}}}
709: \caption{Frequencies of single- double- and quadruple-peaked stationary
710: solutions of DNLS in 2D (points). Dashed lines show analytical expressions
711: obtained for large values of $|\chi|$. The horizontal line at $\omega=-4$
712: indicates the lower edge of the band of Bloch stationary states, which
713: extends from $-4$ to 4. The spectrum is antisymmetric on $\chi$;
714: $\omega (-\chi)=-\omega (\chi)$.} \label{fFS2}
715: \end{figure}
716:
717:
718: All the calculated DP states (symmetric or antisymmetric, along the
719: lattice axes or along the diagonal, and with different interpeak separations)
720: have frequencies around $\omega = \frac{\chi}{2}$, apart from the
721: symmetric and antisymmetric states with their two peaks at neighboring sites
722: (examples are shown at the left columns of Fig.~\ref{fDS2} and
723: Fig.~\ref{fDA2}), which have frequencies around
724: $\omega = \frac{\chi}{2}-1$ and $\omega = \frac{\chi}{2}+1$, respectively.
725: Therefore, from the branches of the DP solutions of
726: Fig.~\ref{fFS2}, the middle one is highly crowded, tending to the
727: level of Eq.(\ref{acsp}) for $M=2$ at large $|\chi|$, while the two
728: external branches are single branches.
729:
730: Similar considerations are valid for the branches of quadruplet stationary
731: states. What appears as a middle branch of the QP states
732: in Fig.~\ref{fFS2} actually contains many stationary solutions with
733: frequencies around $\omega = \frac{\chi}{4}$. Below and above these
734: highly congested branches there are
735: single branches corresponding to the symmetric (see left column of
736: Fig.~\ref{fQS2}) and antisymmetric (see left column of Fig.~\ref{fQA2})
737: solutions, respectively, with their four peaks on the corners of the
738: unit cell of the square lattice.
739:
740: Multi-peaked stationary states, representative of some of the branches
741: shown in Fig.~\ref{fFS2}, are presented in the following two subsections
742: and their linear stability is discussed. Note that there also exist
743: triple-peaked solutions (with their positive or negative peaks along the
744: axes, or along the diagonal, or at random sites), which are not discussed
745: here. Only to mention that their corresponding branches tend to
746: $\omega=\frac{\chi}{3}$ for large values of $|\chi|$, i.e. are in between
747: the double-peaked and quadruple-peaked branches shown in Fig.~\ref{fFS2}.
748:
749:
750: \subsection{Double-peaked solutions}
751:
752:
753: \begin{figure}
754: \begin{center}
755: \begin{tabular}{ccc}
756: \epsfig{file=Fig2d2S1a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
757: \epsfig{file=Fig2d2S2a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
758: \epsfig{file=Fig2d2S3a.eps,height=4.4cm,width=4.4cm} \\
759: \epsfig{file=Fig2d2S1b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
760: \epsfig{file=Fig2d2S2b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
761: \epsfig{file=Fig2d2S3b.eps,height=4.4cm,width=4.4cm} \\
762: \epsfig{file=Fig2d2S1c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
763: \epsfig{file=Fig2d2S2c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
764: \epsfig{file=Fig2d2S3c.eps,height=4.4cm,width=4.4cm} \\
765: \end{tabular}
766: \end{center}
767: \caption{3D plots (first row) and density plots (second row) of
768: double-peaked symmetric solutions (with their peaks along a lattice axis)
769: of DNLS in 2D. {\it Left column:} interpeak separation $S=1$ lattice site,
770: $\chi=-10.5$. {\it Middle column:} interpeak separation $S=2$ sites,
771: $\chi=-15$. {\it Right column:} interpeak separation $S=3$ sites, $\chi=-14$.
772: Cross-sections of the wavefunctions are shown in the third row with points:
773: $\psi_{n_x=n_1,n_y}$ (filled circles), $\psi_{n_x=n_1+1,n_y}$ (filled
774: squares), $\psi_{n_x=n_1+2,n_y}$ (filled diamonds), $\psi_{n_x=n_1+3,n_y}$
775: (filled triangles), where $n_1=20$ is the x$-$coordinate of the two peaks.
776: Dashed lines show analytical approximations of the solutions using
777: Eq.~(\ref{dp2d}). }
778: \label{fDS2}
779: \end{figure}
780:
781:
782: \begin{figure}
783: \begin{center}
784: \begin{tabular}{ccc}
785: \epsfig{file=Fig2d2A1a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
786: \epsfig{file=Fig2d2A2a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
787: \epsfig{file=Fig2d2A3a.eps,height=4.4cm,width=4.4cm} \\
788: \epsfig{file=Fig2d2A1b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
789: \epsfig{file=Fig2d2A2b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
790: \epsfig{file=Fig2d2A3b.eps,height=4.4cm,width=4.4cm} \\
791: \epsfig{file=Fig2d2A1c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
792: \epsfig{file=Fig2d2A2c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
793: \epsfig{file=Fig2d2A3c.eps,height=4.4cm,width=4.4cm} \\
794: \end{tabular}
795: \end{center}
796: \caption{3D plots (first row) and density plots (second row) of
797: double-peaked antisymmetric solutions (with their peaks along a lattice axis)
798: of DNLS in 2D. {\it Left column:} interpeak separation $S=1$ lattice site,
799: $\chi=-14$. {\it Middle column:} interpeak separation $S=2$ sites,
800: $\chi=-12$. {\it Right column:} interpeak separation $S=3$ sites, $\chi=-14$.
801: Cross-sections of the wavefunctions are shown in the third row with points:
802: $\psi_{n_x=n_1,n_y}$ (filled circles), $\psi_{n_x=n_1+1,n_y}$ (filled
803: squares), $\psi_{n_x=n_1+2,n_y}$ (filled diamonds), $\psi_{n_x=n_1+3,n_y}$
804: (filled triangles), where $n_1=20$ is the x$-$coordinate of the two peaks.
805: Dashed lines show analytical approximations of the solutions using
806: Eq.~(\ref{dp2d}). }
807: \label{fDA2}
808: \end{figure}
809:
810:
811: DP solutions with the two peaks along a lattice axis
812: (let say the y-axis) are presented first. The corresponding branches
813: are calculated starting from the initial state
814: \be \label{2d2as}
815: \psi_{n_x,n_y}^{(r=0)}= \frac{1}{\sqrt{2}}(\delta_{n_x,n_1} \delta_{n_y,n_2}
816: + \delta_{n_x,n_1} \delta_{n_y,n_2+S})
817: \ee
818: for the symmetric and
819: \be
820: \psi_{n_x,n_y}^{(r=0)}= \frac{1}{\sqrt{2}}(\delta_{n_x,n_1} \delta_{n_y,n_2}
821: - \delta_{n_x,n_1} \delta_{n_y,n_2+S})
822: \ee
823: for the antisymmetric stationary states, respectively, where
824: $S= 1, 2, \ldots$ determines the interpeak separation along the
825: lattice axis. Figs. \ref{fDS2} and \ref{fDA2} show examples
826: for $S=1$ (left columns), $S=2$ (middle columns), and $S=3$
827: (right columns). Fig.~\ref{fFS2} contains branches of such solutions for
828: $S=1$, 2, 3, 4, 5, and 10. For $S=1$ the upper and lower single branches
829: of the DP states of Fig.~\ref{fFS2} are obtained.
830:
831:
832: \begin{figure}
833: \begin{center}
834: \begin{tabular}{ccc}
835: \epsfig{file=Fig2d2Sd1a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
836: \epsfig{file=Fig2d2Ad1a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
837: \epsfig{file=Fig2d2Sd2a.eps,height=4.4cm,width=4.4cm} \\
838: \epsfig{file=Fig2d2Sd1b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
839: \epsfig{file=Fig2d2Ad1b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
840: \epsfig{file=Fig2d2Sd2b.eps,height=4.4cm,width=4.4cm} \\
841: \epsfig{file=Fig2d2Sd1c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
842: \epsfig{file=Fig2d2Ad1c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
843: \epsfig{file=Fig2d2Sd2c.eps,height=4.4cm,width=4.4cm} \\
844: \end{tabular}
845: \end{center}
846: \caption{3D plots (first row) and density plots (second row) of
847: double-peaked symmetric and antisymmetric solutions (with their peaks along
848: the diagonal of the lattice axes) of DNLS in 2D.
849: {\it Left column:} symmetric state with $l=1$ and $\chi=-14.1$.
850: {\it Middle column:} antisymmetric state with $l=1$ and $\chi=-11$.
851: {\it Right column:} symmetric state with $l=2$ and $\chi=-13.1$.
852: Cross-sections of the wavefunctions are shown in the third row with points:
853: $\psi_{n_x=n_1,n_y}$ (filled circles), $\psi_{n_x=n_1+1,n_y}$ (filled squares),
854: $\psi_{n_x=n_1+2,n_y}$ (filled diamonds), $\psi_{n_x=n_1+3,n_y}$ (filled
855: triangles), $\psi_{n_x=n_1+4,n_y}$ (open circles, in the second and third
856: column), $\psi_{n_x=n_1+5,n_y}$ (open squares, in the third column), where
857: $n_1=20$ is the x$-$coordinate of the first peak. Dashed lines show
858: analytical approximations of the solutions using Eq.~(\ref{dp2d}). }
859: \label{fDd2}
860: \end{figure}
861:
862:
863: Branches of DP solutions with their two peaks along the
864: diagonal of the lattice axes have been also shown in Fig.~\ref{fFS2}.
865: These branches are obtained starting from the initial state
866: \be \label{2d2dsa}
867: \psi_{n_x,n_y}^{(r=0)}= \frac{1}{\sqrt{2}}(\delta_{n_x,n_1} \delta_{n_y,n_2}
868: \pm \delta_{n_x,n_1+l} \delta_{n_y,n_2+l}),
869: \ee
870: where the plus sign gives the symmetric and the minus the antisymmetric,
871: respectively, stationary states. Fig.~\ref{fFS2} contains branches of these
872: solutions for $l=1$,
873: 2, 3, and 10 and their frequencies are around $\omega = \frac{\chi}{2}$.
874: Some examples of such states are shown in Fig.~\ref{fDd2}.
875:
876: Finally, many other DP solutions exist, with their peaks, of the
877: same or opposite sign, at random lattice sites (not aligned along a
878: lattice axis, or the diagonal). These can be obtained starting from
879: the initial state
880: \be
881: \psi_{n_x,n_y}^{(r=0)}= \frac{1}{\sqrt{2}}(\delta_{n_x,n_1} \delta_{n_y,n_2}
882: \pm \delta_{n_x,n_1+S_x} \delta_{n_y,n_2+S_y}),
883: \ee
884: with any combination of non-zero integers $S_x$, $S_y$ and $S_x \neq S_y$.
885: The corresponding branches are close to the middle branch of the
886: DP solutions of Fig.~\ref{fFS2}.
887:
888: As in the 1D case, one can obtain approximate analytical expressions for
889: the DP solutions of DNLS in 2D, by appropriate superpositions
890: of the single-peaked solutions (\ref{vap}). If $(n_1,n_2)$ is the lattice
891: site of the first peak and $(n_1+S_x,n_2+S_y)$ of the second one ($S_x$ and
892: $S_y$ are assumed to be positive), then
893: the corresponding approximate solution is
894: \be \label{dp2d}
895: \psi_{n_x,n_y}^{DP} = \frac{1}{\sqrt{2(1\pm P)}} \frac{1-\zeta^2}{1+\zeta^2}
896: \left( \zeta^{|n_x-n_1|+|n_y-n_2|} \pm (-sgn\chi)^{S_x+S_y}
897: \zeta^{|n_x-n_1-S_x|+|n_y-n_2-S_y|} \right),
898: \ee
899: \be \label{ov2}
900: \mbox{where} \hspace{0.3cm} P=(-sgn \chi)^{S_x+S_y} \; \frac{[(1+S_x)-(S_x-1)
901: \zeta^2] [(1+S_y) - (S_y-1) \zeta^2] \zeta^{S_x+S_y}} {(1+\zeta^2)^2}
902: \ee
903: \be \label{dp2dz}
904: \mbox{and} \hspace{0.5cm} \zeta= -\frac{1}{\chi/2} - \frac{6}{(\chi/2)^3}
905: = -\frac{2}{\chi} - \frac{48}{\chi^3}.
906: \ee
907: Here also, the individual wavefunctions superimposed in this solution
908: should correspond to $\frac{\chi}{2}$, which has been taken into account
909: in the relation (\ref{dp2dz}) providing $\zeta$. The plus and minus signs
910: in Eq.~(\ref{dp2d}) correspond to symmetric and antisymmetric states,
911: respectively, apart from the case of positive $\chi$ and odd $S$ where it
912: is the other way around. The overlap $P$ of the two superimposed single-peaked
913: solutions is $P=\sum_{n_x,n_y} \psi_{n_x,n_y}^{SP[n_1,n_2]}(\frac{\chi}{2})
914: \psi_{n_x,n_y}^{SP[n_1+S_x,n_2+S_y]}(\frac{\chi}{2})$, where
915: $\psi_{n_x,n_y}^{SP[n_1,n_2]}(\frac{\chi}{2})$ is the SP solution
916: in 2D, centered at $(n_1,n_2)$ and corresponding to $\frac{\chi}{2}$.
917: Cross-sections of the approximate solution (\ref{dp2d}) are shown with
918: dashed lines in the third rows of Figs.~\ref{fDS2}-\ref{fDd2}.
919:
920:
921: \begin{figure}
922: \centerline{\hbox{\psfig{figure=Fig2dstab.eps,width=10cm,height=7.5cm}}}
923: \caption{Magnitude of instability $\lambda_u$ of symmetric double-peaked
924: solutions of DNLS for $\chi<0$ in 2D (points) as a function of the strength
925: of nonlinearity $|\chi|$, for various interpeak separations. Results
926: for solutions along a
927: lattice axis with $S_x=1$, $S_y=0$ (filled circles), $S_x=2$, $S_y=0$ (filled
928: diamonds), $S_x=4$, $S_y=0$ (filled squares), $S_x=7$, $S_y=0$ (open circles),
929: and along the diagonal with $S_x=S_y=1$ (open squares), $S_x=S_y=2$ (filled
930: triangles), and $S_x=S_y=3$ (open triangles), are presented.
931: Lines show the power-law relation, Eq.(\ref{inst2}), derived for
932: large values of $|\chi|$.}
933: \label{fstab2}
934: \end{figure}
935:
936:
937: Concerning the stability of the DP stationary states, the picture
938: is similar like in 1D. For $\chi<0$, or $\chi>0$ and even $S$, all the
939: symmetric solutions are unstable, while the antisymmetric ones are in general
940: (at least for relatively large values of $|\chi|$) linearly stable. For
941: $\chi>0$ and odd $S$ the reverse is true. General arguments are
942: presented in the Appendix showing that close to the anti-continuous limit
943: any symmetric (antisymmetric) DP solution should be unstable (linearly
944: stable), except when $\chi>0$ and $S$ is odd, where it is linearly stable
945: (unstable). Further, this calculation allows to determine the
946: variation of the magnitude of instability $\lambda_u$ of the unstable
947: solution with the nonlinearity strength (for relatively large
948: values of $|\chi|$), depending on the interpeak distance. As it is shown in
949: the Appendix, if the separation of the two peaks in the 2D lattice is given
950: by $S_x$ and $S_y$, then
951: \be \label{inst2}
952: \lambda_u = \sqrt{1+S_x S_y} \; \; 2^{(S_x+S_y)/2} \;
953: |\chi|^{1-\frac{S_x+S_y}{2}}, \hspace{1.0cm} \mbox{for} \; \; |\chi| \gg 1.
954: \ee
955: Fig.~\ref{fstab2} presents numerical results in the case of $\chi<0$
956: regarding the magnitude of instability
957: of symmetric DP states with different interpeak separations $S_x$,
958: $S_y$ (points), as well as a comparison with the power-paw (\ref{inst2}).
959:
960: Besides the pair of stable or unstable discrete eigenvalues, there is also
961: the band of eigenvalues extending
962: approximately from $\frac{|\chi|}{2}-4$ to $\frac{|\chi|}{2}+4$, except for
963: the cases where the two peaks are located in first neighboring sites. Then,
964: for the symmetric states the band extends from $\frac{|\chi|}{2}-3$ to
965: $\frac{|\chi|}{2}+5$, while for the antisymmetric states extends from
966: $\frac{|\chi|}{2}-5$ to $\frac{|\chi|}{2}+3$. As $|\chi|$ decreases, the
967: band moves towards zero and the states which are stable at large $|\chi|$
968: become unstable when
969: the band collides with the discrete eigenvalues lying on the real axis.
970:
971:
972: \subsection{ Quadruple-peaked solutions}
973:
974:
975: In this subsection QP solutions of high symmetry are presented, where
976: their four peaks are on lattice sites forming a square of edge equal
977: to $l$ lattice constants. Such solutions may be symmetric, antisymmetric
978: along both lattice axes, or symmetric along the one axis and antisymmetric
979: along the other one, and at various interpeak distances $l$.
980:
981:
982: \begin{figure}
983: \begin{center}
984: \begin{tabular}{ccc}
985: \epsfig{file=Fig2d4S1a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
986: \epsfig{file=Fig2d4S2a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
987: \epsfig{file=Fig2d4S3a.eps,height=4.4cm,width=4.4cm} \\
988: \epsfig{file=Fig2d4S1b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
989: \epsfig{file=Fig2d4S2b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
990: \epsfig{file=Fig2d4S3b.eps,height=4.4cm,width=4.4cm} \\
991: \epsfig{file=Fig2d4S1c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
992: \epsfig{file=Fig2d4S2c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
993: \epsfig{file=Fig2d4S3c.eps,height=4.4cm,width=4.4cm} \\
994: \end{tabular}
995: \end{center}
996: \caption{3D plots (first row) and density plots (second row)
997: of quadruple-peaked symmetric solutions of DNLS in 2D.
998: {\it Left column:} interpeak separation $l=1$ lattice site, $\chi=-22.5$.
999: {\it Middle column:} interpeak separation $l=2$ sites, $\chi=-32$.
1000: {\it Right column:} interpeak separation $l=3$ sites, $\chi=-30$.
1001: Cross-sections of the wavefunction are shown in the third row with points:
1002: $\psi_{n_x=n_1,n_y}$ (filled circles), $\psi_{n_x=n_1+1,n_y}$ (filled squares),
1003: $\psi_{n_x=n_1+2,n_y}$ (filled diamonds), $\psi_{n_x=n_1+3,n_y}$ (filled
1004: triangles), $\psi_{n_x=n_1+4,n_y}$ (open circles), $\psi_{n_x=n_1+5,n_y}$
1005: (open squares, in the third column), where $n_1=20$ is the x$-$coordinate of
1006: the first peak. Dashed lines show analytical approximations of the solutions
1007: using Eq.~(\ref{qp2d}). }
1008: \label{fQS2}
1009: \end{figure}
1010:
1011:
1012: The former are calculated from the initial state
1013: \be \label{qs2}
1014: \psi_{n_x,n_y}^{(r=0)}= \frac{1}{\sqrt{4}}(\delta_{n_x,n_1} \delta_{n_y,n_2}
1015: + \delta_{n_x,n_1} \delta_{n_y,n_2+l} + \delta_{n_x,n_1+l} \delta_{n_y,n_2+l}
1016: + \delta_{n_x,n_1+l} \delta_{n_y,n_2}).
1017: \ee
1018: Fig.~\ref{fQS2} shows such symmetric stationary states for $l=1$, 2, and 3.
1019: The solutions with $l=1$ give the single branch with frequencies around
1020: $\omega = \frac{\chi}{4}-2$ in Fig.~\ref{fFS2}. These solutions are unstable
1021: for negative $\chi$.
1022: Linear stability analysis shows three purely imaginary pairs of opposite
1023: eigenvalues (two of these pairs are degenerate). For the case $l=1$ ($l>2$),
1024: as $|\chi|$ increases the magnitude of the unstable eigenvalues increases
1025: (decreases). For $l=2$ the unstable eigenvalues approach constant values
1026: (around $\pm 2.83$ the most unstable one and around $\pm 2$ the doubly
1027: degenerate eigenvalues) for $|\chi| \gg 1$. However, for relatively large
1028: $\chi>0$ the symmetric solutions corresponding to odd values of $l$ are
1029: linearly stable with three real discrete pairs of eigenvalues. The band of
1030: eigenvalues extends approximately from $\frac{|\chi|}{4}-4$ to
1031: $\frac{|\chi|}{4}+4$, apart from the case of $l=1$, where it extends from
1032: $\frac{|\chi|}{4}-2$ to $\frac{|\chi|}{4}+6$.
1033:
1034:
1035: \begin{figure}
1036: \begin{center}
1037: \begin{tabular}{ccc}
1038: \epsfig{file=Fig2d4A1a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1039: \epsfig{file=Fig2d4A2a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1040: \epsfig{file=Fig2d4A3a.eps,height=4.4cm,width=4.4cm} \\
1041: \epsfig{file=Fig2d4A1b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1042: \epsfig{file=Fig2d4A2b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1043: \epsfig{file=Fig2d4A3b.eps,height=4.4cm,width=4.4cm} \\
1044: \epsfig{file=Fig2d4A1c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1045: \epsfig{file=Fig2d4A2c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1046: \epsfig{file=Fig2d4A3c.eps,height=4.4cm,width=4.4cm} \\
1047: \end{tabular}
1048: \end{center}
1049: \caption{ 3D plots (first row) and density plots (second row)
1050: of quadruple-peaked antisymmetric solutions of DNLS in 2D.
1051: {\it Left column:} interpeak separation $l=1$ lattice site, $\chi=-38$.
1052: {\it Middle column:} interpeak separation $l=2$ sites, $\chi=-21$.
1053: {\it Right column:} interpeak separation $l=3$ sites, $\chi=-26$.
1054: Cross-sections of the wavefunction are shown in the third row with points:
1055: $\psi_{n_x=n_1,n_y}$ (filled circles), $\psi_{n_x=n_1+1,n_y}$ (filled squares),
1056: $\psi_{n_x=n_1+2,n_y}$ (filled diamonds), $\psi_{n_x=n_1+3,n_y}$ (filled
1057: triangles), $\psi_{n_x=n_1+4,n_y}$ (open circles), $\psi_{n_x=n_1+5,n_y}$
1058: (open squares, in the third column), where $n_1=20$ is the x$-$coordinate of
1059: the first peak. Dashed lines show analytical approximations of the solutions
1060: using Eq.~(\ref{qp2d}).}
1061: \label{fQA2}
1062: \end{figure}
1063:
1064:
1065: The antisymmetric along both axes QP solutions are obtained
1066: from alternating signs on neighboring peaks:
1067: \be \label{qa2}
1068: \psi_{n_x,n_y}^{(r=0)}= \frac{1}{\sqrt{4}}(\delta_{n_x,n_1} \delta_{n_y,n_2}
1069: - \delta_{n_x,n_1} \delta_{n_y,n_2+l} + \delta_{n_x,n_1+l} \delta_{n_y,n_2+l}
1070: - \delta_{n_x,n_1+l} \delta_{n_y,n_2}).
1071: \ee
1072: Some examples are shown in Fig.~\ref{fQA2}. The branch of these solutions
1073: with $l=1$ gives the single branch with frequencies around
1074: $\omega = \frac{\chi}{4}+2$ in Fig.~\ref{fFS2}. For $\chi<0$ these
1075: solutions are linearly
1076: stable for large values of $|\chi|$. In this case there exist three discrete
1077: real eigenvalues (two of them are degenerate) and the band extends from
1078: $\frac{|\chi|}{4}-4$ to $\frac{|\chi|}{4}+4$, apart from the case of $l=1$,
1079: where it extends from $\frac{|\chi|}{4}-6$ to $\frac{|\chi|}{4}+2$. As long as
1080: the discrete eigenvalues are outside of the band the solution is linearly
1081: stable. When they collide instabilities develop. The dependence of the
1082: discrete eigenvalues on $|\chi|$ is the usual: for $l=1$ ($l>2$) they
1083: increase (decrease) with $|\chi|$, while for $l=2$ they slightly vary,
1084: approaching constant values at $|\chi| \gg 1$. Regarding the solutions shown
1085: in Fig.~\ref{fQA2} the last one (for $l=3$, in the right column) is linearly
1086: stable and the other two are unstable. As previously, the opposite sign of
1087: $\chi$ ($\chi>0$) does not change anything regarding the stability eigenvalues
1088: of symmetric and fully antisymmetric QP states, when $l$ is even. On the
1089: contrary, for odd $l$ these families of solutions interchange stability
1090: eigenvalues when $\chi \rightarrow -\chi$.
1091:
1092:
1093: \begin{figure}
1094: \begin{center}
1095: \begin{tabular}{ccc}
1096: \epsfig{file=Fig2d4AS1a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1097: \epsfig{file=Fig2d4AS2a.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1098: \epsfig{file=Fig2d4AS3a.eps,height=4.4cm,width=4.4cm} \\
1099: \epsfig{file=Fig2d4AS1b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1100: \epsfig{file=Fig2d4AS2b.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1101: \epsfig{file=Fig2d4AS3b.eps,height=4.4cm,width=4.4cm} \\
1102: \epsfig{file=Fig2d4AS1c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1103: \epsfig{file=Fig2d4AS2c.eps,height=4.4cm,width=4.4cm} & \hspace{-0.2cm}
1104: \epsfig{file=Fig2d4AS3c.eps,height=4.4cm,width=4.4cm} \\
1105: \end{tabular}
1106: \end{center}
1107: \caption{3D plots (first row) and density plots (second row)
1108: of quadruple-peaked symmetric/antisymmetric solutions of DNLS in 2D.
1109: {\it Left column:} interpeak separation $l=1$ lattice site, $\chi=-27$.
1110: {\it Middle column:} interpeak separation $l=2$ sites, $\chi=-27.5$.
1111: {\it Right column:} interpeak separation $l=3$ sites, $\chi=-26$.
1112: Cross-sections of the wavefunction are shown in the third row with points:
1113: $\psi_{n_x=n_1,n_y}$ (filled circles), $\psi_{n_x=n_1+1,n_y}$ (filled squares),
1114: $\psi_{n_x=n_1+2,n_y}$ (filled diamonds), $\psi_{n_x=n_1+3,n_y}$ (filled
1115: triangles), $\psi_{n_x=n_1+4,n_y}$ (open circles), $\psi_{n_x=n_1+5,n_y}$
1116: (open squares, in the third column), where $n_1=20$ is the x$-$coordinate of
1117: the first peak. Dashed lines show analytical approximations of the solutions
1118: using Eq.~(\ref{qp2db}).}
1119: \label{fQSA2}
1120: \end{figure}
1121:
1122:
1123: The last example of QP states, which are symmetric along one
1124: lattice axis and antisymmetric along the other one, are obtained from the
1125: initial state
1126: \be \label{qsa2}
1127: \psi_{n_x,n_y}^{(r=0)}= \frac{1}{\sqrt{4}}(\delta_{n_x,n_1} \delta_{n_y,n_2}
1128: + \delta_{n_x,n_1} \delta_{n_y,n_2+l} - \delta_{n_x,n_1+l} \delta_{n_y,n_2+l}
1129: - \delta_{n_x,n_1+l} \delta_{n_y,n_2}).
1130: \ee
1131: Three cases (for $l=1$, 2, and 3) are presented in Fig.~\ref{fQSA2}.
1132: These solutions are always unstable since there are two pairs of purely
1133: imaginary eigenvalues $\pm i \beta$ and $\pm i \gamma$. There is also a
1134: discrete eigenvalue $\delta$ which is real for large values of $|\chi|$,
1135: but when it collides with the band (which extends approximately from
1136: $\frac{|\chi|}{4}-4$ to $\frac{|\chi|}{4}+4$) one more instability is
1137: developed. For $|\chi| \gg 1$, $\beta$ and $\delta$ tend to the same value
1138: and they show the typical dependence on $|\chi|$: they increase (decrease)
1139: for $l=1$ ($l>2$) and they tend to 2 for $l=2$. This picture does not
1140: change for positive $\chi$, regardless whether $l$ is even or odd.
1141:
1142: In Fig.~\ref{fFS2} branches of solutions of the symmetry obtained from
1143: (\ref{qs2}) and
1144: (\ref{qa2}) are shown for $l=1$, 2, 3, 4, 5, 6, 7, 10, and 11 and from
1145: (\ref{qsa2}) for $l=1$, 2, 3, 4, 5, 10, and 11. Of course there are many
1146: more quadruplets having their four peaks in a rectangle or in random lattice
1147: sites, with any combination of signs, which are congested around the middle
1148: quasi-degenerate branch of the QP solutions of Fig.~\ref{fFS2}.
1149: An example of antisymmetric along {\it both diagonals} QP
1150: solution, forming a square with its edges (of length equal to
1151: $\sqrt{2}$) not along the axes, as the states presented above, but
1152: along the diagonals, is reported as a quasivortex in Ref.~\cite{KMCF}
1153:
1154: Once more, approximate analytical expressions can be derived for the
1155: QP solutions in 2D, by superimposing four single-peaked solutions
1156: of Eq. (\ref{vap}), each one corresponding to $\frac{\chi}{4}$. Regarding
1157: the high symmetry solutions presented above, if $(n_1,n_2)$ is the position
1158: of the first peak and $l$ the interpeak distance along the lattice axes
1159: ($l>0$), then these stationary states can be approximated by
1160: \begin{eqnarray}
1161: \psi_{n_x,n_y}^{QP}=\frac{1}{2\sqrt{1 \pm 2P+P^2}} \frac{1-\zeta^2}{1+\zeta^2}
1162: & \left( \zeta^{|n_x-n_1|+|n_y-n_2|} \pm (-sgn\chi)^l
1163: \zeta^{|n_x-n_1|+|n_y-n_2-l|} \right. \nonumber \\ & \left. +
1164: \zeta^{|n_x-n_1-l|+|n_y-n_2-l|} \pm (-sgn\chi)^l \zeta^{|n_x-n_1-l|+|n_y-n_2|}
1165: \right) \hspace{1.0cm} \label{qp2d} \end{eqnarray}
1166: and
1167: \begin{eqnarray}
1168: \psi_{n_x,n_y}^{QP} = \frac{1}{2\sqrt{1-P^2}} \frac{1-\zeta^2}{1+\zeta^2}
1169: & \left( \zeta^{|n_x-n_1|+|n_y-n_2|} + \zeta^{|n_x-n_1|+|n_y-n_2-l|}
1170: \right. \nonumber \\ & \left.
1171: - \zeta^{|n_x-n_1-l|+|n_y-n_2-l|} - \zeta^{|n_x-n_1-l|+|n_y-n_2|} \right),
1172: \hspace{1.0cm} \label{qp2db} \end{eqnarray}
1173: \be
1174: \mbox{where} \hspace{0.3cm}
1175: P=(-sgn\chi)^l \; \frac{(1+l) \zeta^l - (l-1) \zeta^{l+2}}{1+\zeta^2}
1176: \ee
1177: \be
1178: \mbox{and} \hspace{0.3cm} \zeta = -\frac{1}{\chi/4} - \frac{6}{(\chi/4)^3}
1179: = -\frac{4}{\chi} - \frac{384}{\chi^3}.
1180: \ee
1181: Plus (minus) signs in Eq.~(\ref{qp2d}) provide the symmetric (fully
1182: antisymmetric along both axes) QP solutions, except when $\chi$ is positive
1183: and $l$ odd, where it is the reverse. Eq.~(\ref{qp2db})
1184: gives the solutions which are symmetric along one axis and antisymmetric
1185: along the other one, like those of Fig.~\ref{fQSA2}.
1186: $P=\sum_{n_x,n_y} \psi_{n_x,n_y}^{SP[n_1,n_2]}(\frac{\chi}{4})
1187: \psi_{n_x,n_y}^{SP[n_1,n_2+l]}(\frac{\chi}{4})$ is the overlap of two
1188: neighboring single-peaked states at distance $l$. The overlap of two
1189: single-peaked states across the diagonal of the square,
1190: $\sum_{n_x,n_y} \psi_{n_x,n_y}^{SP[n_1,n_2]}(\frac{\chi}{4})
1191: \psi_{n_x,n_y}^{SP[n_1+l,n_2+l]}(\frac{\chi}{4})$, is equal to $P^2$.
1192: Cross-sections of Eqs. (\ref{qp2d}) and (\ref{qp2db}) are shown in the
1193: third rows of Figs. \ref{fQS2}-\ref{fQSA2} with dashed lines.
1194:
1195:
1196: \section{Conclusions}
1197:
1198:
1199: Multi-peaked localized excited states are discussed for the one- and
1200: two-dimensional discrete nonlinear Schr\"odinger equation. Their numerical
1201: calculation is achieved by iterations of a simple map, where trivial
1202: initial states rapidly converge to the desired solution.
1203: Examples have been presented of symmetric and antisymmetric states
1204: with different interpeak separations for double-peaked solutions in 1D,
1205: as well as for double-peaked states along a lattice axis or along the
1206: diagonal and quadruple-peaked states on a square in 2D. Analytical
1207: approximations and the linear stability of the solutions have been discussed.
1208: For strong nonlinearities, the symmetric double-peaked states are unstable
1209: and the antisymmetric linearly stable, except for the case of positive
1210: nonlinearity and odd interpeak separation, where the situation is reversed.
1211: An interesting application that such multi-peaked solutions may have,
1212: concerns their potential use for information encoding and transfer
1213: in optical lattices \cite{KETC}. Multi-peaked solutions have been also
1214: discussed in different contexts \cite{BE,fuentes,pan}.
1215:
1216: The classification of these stationary states is based
1217: on their origin at the anti-continuous limit, even though sometimes
1218: their name may be misleading. For example, the symmetric double-peaked
1219: states in 1D and 2D and the symmetric quadruplet in 2D with interpeak
1220: separations equal to one lattice constant, can be viewed as solutions
1221: having one peak. However, this classification is very useful for
1222: organizing the discrete frequency and energy spectrum of DNLS.
1223: The concept of anti-continuous limit \cite{sergeAC} facilitates the
1224: interpretation of the structure of this spectrum.
1225:
1226: Some of the stationary states presented in this work have been also
1227: discussed in other studies and different names are attributed to them.
1228: The double-peaked solution in 1D with interpeak separation $S=1$
1229: has been named Page mode, even mode, or centered-between-sites \cite{KC}.
1230: Antisymmetric double-peaked states in 1D with $S=1$ or $S=2$ are
1231: also known as twisted modes \cite{DKL}, while their analogues in 2D
1232: with their peaks along the diagonal have been discussed in \cite{KMB}.
1233: In Ref. \cite{kevr} the symmetric double-peaked state along a lattice
1234: axis with $S=1$ in 2D and the symmetric quadruple-peaked state
1235: with interpeak separation $l=1$ have been named hybrid mode and
1236: Page-like mode, respectively.
1237: The double-peaked symmetric stationary solution along the diagonal with
1238: $l=1$ in 2D has been also known \cite{rodrigo}.
1239:
1240:
1241: {\bf Acknowledgements}
1242:
1243: The author would like to thank S. Aubry for useful and stimulating
1244: suggestions, R. Vicencio and M. Johansson for interesting discussions
1245: and for bringing to his attention relevant references, and the referees
1246: for useful comments.
1247: This work is dedicated to Serge on the occasion of his 60th birthday,
1248: with a strong appreciation for the valuable knowledge and fruitful ideas
1249: that he has tried to communicate to us and many wishes for even more
1250: breakthroughs on nonlinear physics in the future.
1251:
1252:
1253: \section*{ Appendix A}
1254:
1255:
1256: \subsection*{A.1 Linear stability of a stationary state}
1257:
1258: Considering small deviations $\delta \psi_n(t)$ from a stationary solution
1259: of Eq.~(\ref{statsol}), i.e.
1260: $\Psi_n(t) = \left( \psi_n +\delta \psi_n(t) \right) \cdot e^{-i\omega_0 t}$
1261: (where $\psi_n$ is time-independent and real and $\omega_0$ is the
1262: corresponding frequency), substituting in DNLS Eq.~(\ref{dnls}), and
1263: linearizing in respect to the complex small perturbations
1264: $\delta \psi_n(t)$, one obtains
1265: \be \label{ldnls}
1266: i \frac{d \delta \psi_n}{dt} = (-\omega_0 + \chi \psi_n^2) \delta \psi_n
1267: - \sum_{\delta} \delta \psi_{n+\delta} + 2 \chi \psi_n^2 Re(\delta \psi_n),
1268: \ee
1269: where $Re(\delta \psi_n)$ is the real part of $\delta \psi_n(t)$.
1270: Substituting solutions of the form
1271: \be \label{lsol}
1272: \delta \psi_n(t) = a_n \sin(\omega t) + i b_n \cos(\omega t)
1273: \ee
1274: in the linearized equation (\ref{ldnls}), yields the following coupled
1275: system
1276: \begin{eqnarray} \label{ls1}
1277: \omega a_n= & (-\omega_0 +\chi \psi_n^2) b_n & - \sum_\delta b_{n+\delta} \\
1278: \omega b_n= & (-\omega_0 +3\chi \psi_n^2) a_n & - \sum_\delta a_{n+\delta}.
1279: \label{ls2} \end{eqnarray}
1280: This implies that the stability eigenvalues $\omega$ can be obtained
1281: as the eigenvalues of the $2L \times 2L$ (where $L$ is the total number of
1282: lattice sites) matrix M:
1283: \be
1284: M \left(
1285: \begin{array}{c}
1286: A \\ B
1287: \end{array} \right) = \left(
1288: \begin{array}{cc}
1289: {\it 0} & M_1 \\ M_2 & {\it 0}
1290: \end{array}
1291: \right) \left(
1292: \begin{array}{c}
1293: A \\ B
1294: \end{array} \right) = \omega \left(
1295: \begin{array}{c}
1296: A \\ B
1297: \end{array} \right),
1298: \ee
1299: where the $2L \times 1$ column
1300: $(A,B)^T \equiv (a_1, \ldots,a_L,b_1,\ldots,b_L)^T$ corresponds to the
1301: eigenvectors, ${\it 0}$ denotes the $L \times L$ zero matrix and $M_1$,
1302: $M_2$ are the tight-binding $L \times L$ matrices given through the
1303: right-hand-sides
1304: of Eqs. (\ref{ls1}) and (\ref{ls2}), respectively. $M_1$ and $M_2$ differ
1305: only in their diagonal elements; $M_{1_{ii}}=-\omega_0 +\chi \psi_i^2$ and
1306: $M_{2_{ii}}=-\omega_0 +3\chi \psi_i^2$, while the non-diagonal matrix
1307: elements are zero, except when they correspond to first neighboring sites
1308: where they are equal to $-1$.
1309:
1310: From Eq.~(\ref{lsol}) we see that if there is an eigenvalue of $M$ with
1311: non-zero imaginary part, then the stationary solution is unstable. It can be
1312: easily verified that any stationary solution has a stability eigenvalue
1313: $\omega=0$, with eigenvector $a_n=0$, $b_n=\psi_n$, since Eq.~(\ref{ls2})
1314: is trivially satisfied and Eq.~(\ref{sdnls}) provides a solution making
1315: zero the right-hand-side of (\ref{ls1}). The eigenvalue $\omega=0$ is
1316: doubly degenerate possessing a second eigenvector known as growth mode
1317: \cite{JA2}.
1318:
1319:
1320: \subsection*{A.2 Obtaining the linear stability eigenvalues through a
1321: tight-binding Hamiltonian}
1322:
1323:
1324: Here it is shown that the
1325: eigenvalues of the linear stability problem can be calculated through the
1326: energy eigenvalues and eigenvectors of a tight-binding
1327: Hamiltonian\footnote{ a similar procedure has been used in
1328: Ref.~\cite{KAT} for the linear stability of a slightly more complicated
1329: system, viz. the polarons of the adiabatic Holstein model} with an on-site
1330: potential determined by the stationary state $\psi_n$. In particular,
1331: consider the Hamiltonian eigenvalue problem
1332: \be \label{etb}
1333: H_l \phi_n^\nu = -\sum_\delta \phi_{n+\delta}^\nu + U_n \phi_n^\nu
1334: = E_\nu \phi_n^\nu, \hspace{0.5cm} \mbox{with potential} \; \;
1335: U_n = \chi \psi_n^2,
1336: \ee
1337: where $E_\nu$ and $\phi_n^\nu$ are the eigenvalues and the corresponding
1338: eigenvectors. The potential $U_n$
1339: has the shape of the stationary state under discussion, multiplied by
1340: the nonlinearity parameter. Comparing Eqs. (\ref{sdnls}) and (\ref{etb})
1341: one obtains that $\psi_n$ is an eigenvector $\phi_n^{\nu=0}$ of $H_l$
1342: with $E_{\nu=0}=\omega_0$.
1343:
1344: Expressing the eigenvectors $a_n$ and $b_n$ of the linear stability
1345: system (\ref{ls1}), (\ref{ls2}) in the complete basis $\phi_n^\nu$
1346: of the Hamiltonian $H_l$, i.e. $a_n = \sum_\nu a_\nu \phi_n^\nu$ and
1347: $b_n = \sum_\nu b_\nu \phi_n^\nu$,
1348: substituting in the system (\ref{ls1}), (\ref{ls2}), and using
1349: Eq.~(\ref{etb}) and the relation
1350: $\sum_n \phi_n^\nu \phi_n^{\nu^\prime}=\delta_{\nu,\nu^\prime}$,
1351: yields that for $\nu \neq 0$ is $b_\nu=\frac{\omega}{E_\nu-E_0}a_\nu$ and
1352: \be \label{sptb}
1353: \left[ (E_\nu-E_0)-\frac{\omega^2}{E_\nu-E_0}+2\chi
1354: \sum_n(\phi_n^0 \phi_n^\nu)^2 \right] a_\nu + 2\chi
1355: \sum_{\nu^\prime \neq \nu} a_{\nu^\prime} \left(
1356: \sum_n (\phi_n^0)^2 \phi_n^\nu \phi_n^{\nu^\prime} \right) =0.
1357: \ee
1358: These $L-1$ equations (since $\nu \neq 0$) provide the $2L-2$ non-zero
1359: stability eigenvalues $\omega=\pm \sqrt{\omega^2}$. We see that the
1360: eigenvalues of the linear stability analysis appear as pairs of opposite
1361: sign and the $\omega^2$ result from the diagonalization of the
1362: $(L-1) \times (L-1)$ matrix
1363: \begin{eqnarray} \label{dtb}
1364: K_{\nu,\nu^\prime}= & \left[ (E_\nu-E_0)^2 + 2 \chi (E_\nu-E_0)
1365: \sum_n(\phi_n^0 \phi_n^\nu)^2 \right] \delta_{\nu,\nu^\prime} & \\
1366: \nonumber & + (1-\delta_{\nu,\nu^\prime}) 2 \chi (E_\nu-E_0)
1367: \sum_n (\phi_n^0)^2 \phi_n^\nu \phi_n^{\nu^\prime}, & \hspace{0.3cm}
1368: \mbox{where} \; \; \nu,\nu^\prime \neq 0.
1369: \end{eqnarray}
1370: $K_{\nu,\nu^\prime}$ is constructed from the eigenspectrum of $H_l$.
1371: Using this matrix, analytical results are obtained in the following
1372: subsection for the linear stability
1373: eigenvalues of double-peaked solutions of DNLS for large values of
1374: $|\chi|$, and the Eqs. (\ref{inst1}) and (\ref{inst2}) are derived.
1375:
1376:
1377: \subsection*{A.3 Application: double-peaked stationary states close to the
1378: anti-continuous limit}
1379:
1380:
1381: For a DP stationary solution the potential $U_n$ of
1382: Eq.~(\ref{etb}) is a double well for negative $\chi$ and a double barrier
1383: for positive $\chi$. If $|\chi| \gg 1$, then $\psi_n$ is localized at
1384: almost two lattice sites, where the two peaks are located, and each well
1385: (barrier) of $U_n$ is very deep (high) and narrow, with a strength about
1386: $\frac{\chi}{2}$. In this case the Hamiltonian $H_l$, Eq.~(\ref{etb}), is
1387: equivalent to a tight-binding problem with two equal impurities with
1388: large on-site energies
1389: $\frac{\chi}{2}$. Then the energy spectrum has two discrete eigenvalues
1390: (the $E_0=\omega_0$ and a second one, denoted by $E_1$) and, for an
1391: extended system, all the other eigenvalues belong to the continuous band
1392: from $-2d$ to $2d$ \cite{econ}. The eigenvectors of the continuous spectrum
1393: are proportional to $\frac{1}{\sqrt{L}}$ and therefore negligibly small
1394: at any lattice site. Since the sums over the lattice sites $n$ that appear in
1395: $K_{\nu,\nu^\prime}$ in Eq.~(\ref{dtb}) contain the tightly localized
1396: around two sites wavefunction $\phi_n^0 \equiv \psi_n$, they can be
1397: neglected when they involve an eigenstate of the continuous.
1398: The $\phi_n^\nu, \phi_n^{\nu^\prime}$ of $K_{\nu,\nu^\prime}$
1399: do not include the $\nu=0$ and thus there is only one localized eigenstate
1400: among them; the $\phi_n^1$ corresponding to $E_1$. As a result the
1401: non-diagonal matrix elements of $K_{\nu,\nu^\prime}$ can be neglected, since
1402: they contain products of two eigenstates $\phi_n^\nu \phi_n^{\nu^\prime}$
1403: and at least one of them it belongs to the continuum. The diagonal ones,
1404: providing directly the stability eigenvalues $\omega^2$, are
1405: \be \label{fcon}
1406: \omega^2 = K_{\nu,\nu} = (E_\nu-E_0)^2,
1407: \hspace{0.8cm} \mbox{for} \hspace{0.2cm} \nu \neq 1 \hspace{1.2cm} \mbox{and}
1408: \ee
1409: \be \label{fdis}
1410: \omega^2 = K_{1,1} = (E_1-E_0)^2 +2 \chi (E_1-E_0)
1411: \sum_n(\phi_n^0 \phi_n^1)^2, \hspace{0.8cm} \mbox{for} \hspace{0.2cm} \nu=1
1412: \ee
1413: Eq. (\ref{fcon}) gives two bands of real eigenvalues $\omega=\pm |E_\nu-E_0|$,
1414: symmetrically positioned around zero, and Eq.~(\ref{fdis}) a discrete
1415: pair of eigenvalues $\omega=\pm \sqrt{K_{1,1}}$.
1416:
1417: The discrete eigenspectrum of $H_l$ is given by
1418: \be \label{distb}
1419: \phi_n^\pm = \frac{1}{\sqrt{2(1\pm P)}} \left( g_n^{[n_0]} \pm
1420: (-sgn\chi)^S g_n^{[n_0+S]} \right) \longrightarrow E_\pm =
1421: \frac{\epsilon\pm\upsilon}{1\pm P} \approx\epsilon \pm\upsilon\mp\epsilon P,
1422: \ee
1423: where $g_n^{[n_0]}, g_n^{[n_0+S]}$ represent single-peaked wavefunctions,
1424: centered at $n_0$ and $n_0+S$, respectively. Here, $n$, $n_0$, and $S$
1425: ($S$ or its components are assumed positive)
1426: have to be understood as integers in 1D and pairs or sums of integers in 2D,
1427: e.g. $n\rightarrow (n_x,n_y)$, $n_0 \rightarrow(n_1,n_2)$,
1428: $S\rightarrow (S_x,S_y)$, $(-sgn\chi)^S \rightarrow (-sgn\chi)^{S_x+S_y}$,
1429: etc. The other quantities in the energy eigenvalues $E_\pm$ of
1430: Eq.~(\ref{distb}) are $\epsilon= \langle g_n^{[n_0]}
1431: |H_l|g_n^{[n_0]} \rangle=\langle g_n^{[n_0+S]}|H_l|g_n^{[n_0+S]} \rangle$
1432: and the small overlap integrals
1433: $\upsilon= (-sgn\chi)^S \langle g_n^{[n_0]}|H_l|g_n^{[n_0+S]} \rangle$ and
1434: $P= (-sgn\chi)^S \langle g_n^{[n_0]}|g_n^{[n_0+S]} \rangle$.
1435: For $\chi$ negative, i.e. attractive potential $U_n$, the upper (lower) signs
1436: in Eq.~(\ref{distb}) correspond to the ground (first excited) state of
1437: $H_l$, while for positive $\chi$, i.e. repulsive $U_n$, they provide the
1438: highest-energy (second highest-energy) state. Note that for positive $\chi$
1439: and odd $S$ the highest-energy state is the antisymmetric one because of the
1440: $(-sgn\chi)^S$ term in (\ref{distb}), in contrary to the case of even $S$.
1441: It is convenient to distinguish three cases: \\
1442: (i) negative $\chi$, \\ (ii) positive $\chi$ and even $S$ (or even $S_x+S_y$),
1443: and \\ (iii) positive $\chi$ and odd $S$ (or odd $S_x+S_y$). \\
1444: For symmetric DP stationary states $\psi_n$, the $\phi_n^+,E_+$ of
1445: Eq.~(\ref{distb}) correspond to $\phi_n^0 \equiv \psi_n$ and $E_0$ and the
1446: $\phi_n^-,E_-$ to $\phi_n^1$ and $E_1$ in cases (i) and (ii), while it is
1447: the other way around in case (iii). For antisymmetric DP states $\psi_n$,
1448: holds the opposite: the $\phi_n^-,E_-$ correspond to $\phi_n^0$ and $E_0$
1449: and the $\phi_n^+,E_+$ to $\phi_n^1$ and $E_1$ in cases (i) and (ii),
1450: and the reverse is valid in case (iii).
1451:
1452: The discrete eigenvalue (\ref{fdis}) is dominated by the second term,
1453: since $E_1-E_0$ is a small quantity and $\chi \gg 1$.
1454: For a symmetric DP solution $E_1-E_0$ is positive in cases (i) and (iii)
1455: and negative in case (ii). Therefore, $K_{1,1}$, which has the same sign
1456: as $\chi (E_1-E_0)$, is negative (positive) in cases (i) and (ii) (in case
1457: (iii)), meaning that the symmetric DP solution is unstable (linearly stable)
1458: with a pair of purely imaginary (real) discrete eigenvalues. The situation
1459: is reverse for an antisymmetric DP stationary state since $E_1-E_0$ has the
1460: opposite sign now. Thus, when the symmetric
1461: DP solution is linearly stable (unstable) the antisymmetric one is unstable
1462: (linearly stable). These results are confirmed by the numerical simulations
1463: of sections 4 and 5. The above arguments about the linear stability/instability
1464: of a symmetric or antisymmetric DP solution of DNLS close to the
1465: anti-continuous limit, are generally applied at any dimension. Therefore,
1466: one expects that, for large $|\chi|$, also in 3D an antisymmetric DP
1467: stationary state is linearly stable and a symmetric unstable, except when
1468: $\chi$ is positive and the interpeak distance $S_x+S_y+S_z$ is odd, where
1469: they interchange roles regarding their stability.
1470:
1471: The previous discussion can be quantified and using Eqs. (\ref{fdis}) and
1472: (\ref{distb}) analytical expressions are
1473: obtained for the discrete pair of eigenvalues in the limit of $|\chi| \gg 1$.
1474: The wavefunctions $g_n$ in (\ref{distb}) are given by Eq.~(\ref{vap}) for
1475: $\frac{\chi}{2}$. Taking into account that $\zeta=-\frac{2}{\chi}$ one
1476: obtains to leading order: $P=(-sgn\chi)^S (1+S) \zeta^S$ in 1D and
1477: $P=(-sgn\chi)^{S_x+S_y} (1+S_x)(1+S_y) \zeta^{S_x+S_y}$ in 2D [see also Eqs.
1478: (\ref{ov1}) and (\ref{ov2})], the sum $\sum_n(\phi_n^0\phi_n^1)^2=\frac{1}{2}$,
1479: the coupling $\upsilon=(-sgn\chi)^S (-S\zeta^{S-1}+\chi\zeta^S)$ in 1D and
1480: $\upsilon= (-sgn\chi)^{S_x+S_y} [-(S_x+S_y+2S_xS_y)\zeta^{S_x+S_y-1}+\chi
1481: \zeta^{S_x+S_y}]$ in 2D, and the on-site energy $\epsilon=\frac{\chi}{2}$.
1482: Then for symmetric DP states in 1D is
1483: $E_1-E_0=-2(\upsilon-\epsilon P)=\frac{2^S}{|\chi|^{S-1}}$ in case (i),
1484: $E_1-E_0=-2(\upsilon-\epsilon P)=-\frac{2^S}{\chi^{S-1}}$ in case (ii), and
1485: $E_1-E_0=2(\upsilon-\epsilon P)=\frac{2^S}{\chi^{S-1}}$ in case (iii).
1486: The signs are the same in 2D, but
1487: the magnitudes change to $\frac{(1+S_xS_y)2^{S_x+S_y}}{|\chi|^{S_x+S_y-1}}$.
1488: Therefore Eq.~(\ref{fdis}) yields that in cases (i) and (ii)
1489: $\omega^2=-\frac{2^S}{|\chi|^{S-2}}$ in 1D and
1490: $\omega^2=-\frac{(1+S_xS_y)2^{S_x+S_y}}{|\chi|^{S_x+S_y-2}}$ in 2D, while
1491: in case (iii) $\omega^2$ is positive with the same magnitude. For the
1492: antisymmetric DP states the $E_1-E_0$ and consequently the $\omega^2$
1493: are the same like previously, but with opposite sign. As a result the
1494: discrete eigenvalues are
1495: \be \label{unsteig}
1496: \omega=\pm i 2^{S/2} |\chi|^{1-S/2} \hspace{0.2cm} \mbox{in 1D}
1497: \hspace{0.3cm} \mbox{and} \hspace{0.3cm} \omega=\pm i \sqrt{1+S_xS_y} \;
1498: 2^{(S_x+S_y)/2} |\chi|^{1-\frac{S_x+S_y}{2}} \hspace{0.2cm} \mbox{in 2D}
1499: \ee
1500: for the (unstable) symmetric DP states in cases (i) and (ii) and the
1501: antisymmetric in case (iii), while
1502: \be \label{steig}
1503: \omega=\pm 2^{S/2} |\chi|^{1-S/2} \hspace{0.2cm} \mbox{in 1D}
1504: \hspace{0.3cm} \mbox{and} \hspace{0.3cm} \omega=\pm \sqrt{1+S_xS_y} \;
1505: 2^{(S_x+S_y)/2} |\chi|^{1-\frac{S_x+S_y}{2}} \hspace{0.2cm} \mbox{in 2D}
1506: \ee
1507: for the (linearly stable) symmetric states in case (iii) and the
1508: antisymmetric ones in cases (i) and (ii). Eq.~(\ref{unsteig}) provides the
1509: instability magnitudes given in Eqs. (\ref{inst1}) and (\ref{inst2}),
1510: while Eq.~(\ref{steig}) the stable eigenvalue used in Eq.~(\ref{antun}).
1511: By decreasing $|\chi|$, as the stationary state $\psi_n$ and the
1512: potential $U_n$ of (\ref{etb}) become more extended, additional discrete
1513: levels of $H_l$ may appear, resulting in additional discrete eigenvalues
1514: of the linear stability problem.
1515:
1516: For the DP states, $E_0 \equiv \omega_0$ equals, to leading order, to
1517: $\epsilon = \frac{\chi}{2}$. The next corrections are
1518: $\pm (\upsilon-\epsilon P)$, which are equal to
1519: $\pm \frac{E_1-E_0}{2}=\pm \frac{2^{S-1}}{|\chi|^{S-1}}$ in 1D and a similar
1520: expression in 2D (see above). These corrections are inverse powers of $\chi$,
1521: apart from the case of $S=1$ (or $S_x+S_y=1$). Then, if $\chi<0$
1522: for example, the symmetric DP states have
1523: $E_0 \equiv \omega_0=\frac{\chi}{2}+(\upsilon-\epsilon P)=\frac{\chi}{2}-1$
1524: and the antisymmetric have
1525: $E_0 \equiv \omega_0=\frac{\chi}{2}-(\upsilon-\epsilon P)=\frac{\chi}{2}+1$
1526: (see the lower and upper branches of DP states in Figs. \ref{figFS1}
1527: and \ref{fFS2}, while similar is the case for the single branches of
1528: the QP states presented in 2D). The knowledge of $E_0$ and the continuous
1529: band of $H_l$ (from $-2d$ to $2d$) provides the bands of stability
1530: eigenvalues $\omega=\pm |E_\nu-E_0|$, Eq.~(\ref{fcon}). The latter, for
1531: positive $\omega$, extends from $\frac{|\chi|}{2}-2d$ to
1532: $\frac{|\chi|}{2}+2d$, apart from the case of $S=1$ (or $S_x+S_y=1$), where
1533: it extends between $\frac{|\chi|}{2}\pm1-2d$ and $\frac{|\chi|}{2}\pm1+2d$
1534: (the upper signs correspond to symmetric DP states in cases i) and ii) and
1535: antisymmetric in case iii), while the minus signs correspond to the
1536: complementary situations). These results are in accordance with the
1537: numerical observations in sections 4 and 5.
1538:
1539: \vspace{-0.8cm}
1540:
1541: \begin{thebibliography}{99}
1542:
1543:
1544: \bibitem{ELS}
1545: J.C. Eilbeck, P.S. Lomdahl, and A.C. Scott, Physica D {\bf 16}, 318 (1985).
1546:
1547: \bibitem{eilb}
1548: J.C. Eilbeck in {\it Davydov's Soliton Revisited}, edited by
1549: P.L. Christiansen and A.C. Scott (Plenum Press, NY, 1990), p. 473.
1550:
1551: \bibitem{MA}
1552: R.S. MacKay and S. Aubry, Nonlinearity {\bf 7}, 1623 (1994).
1553:
1554: \bibitem{JA}
1555: M. Johansson and S. Aubry, Nonlinearity {\bf 10}, 1151 (1997).
1556:
1557: \bibitem{FKM} S. Flach, K. Kladko, and R.S. MacKay,
1558: Phys. Rev. Lett. {\bf 78}, 1207 (1997).
1559:
1560: \bibitem{HT} D. Hennig and G.P. Tsironis, Phys. Rep. {\bf 307}, 334 (1999).
1561:
1562: \bibitem{RABT}
1563: K.\O. Rasmussen, S. Aubry, A.R. Bishop, and G.P. Tsironis,
1564: Eur. Phys. J. B {\bf 15}, 169 (2000).
1565:
1566: \bibitem{JA2}
1567: M. Johansson and S. Aubry, Phys. Rev. E {\bf 61}, 5864 (2000).
1568:
1569: \bibitem{BBJ} J.M. Bergamin, T. Bountis, and C. Jung,
1570: J. Phys. A: Math. Gen. {\bf 33}, 8059 (2000).
1571:
1572: \bibitem{BKRK}
1573: A.R. Bishop, G. Kalosakas, K.\O. Rasmussen, and P.G. Kevrekidis,
1574: Chaos {\bf 13}, 588 (2003).
1575:
1576: \bibitem{VMK} R.A. Vicencio, M.I. Molina, and Y.S. Kivshar,
1577: Opt. Lett. {\bf 28}, 1942 (2003).
1578:
1579: \bibitem{floria} J. C\'omez-Garde\~nes, F. Falo, and L.M. Flor\'ia,
1580: Phys. Lett. A {\bf 332}, 213 (2004).
1581:
1582: \bibitem{lm}
1583: A.C. Scott and J.C. Eilbeck, Chem. Phys. Lett. {\bf 132}, 23 (1986).
1584:
1585: \bibitem{nopt}
1586: D.N. Christodoulides and R.I. Joseph, Opt. Lett. {\bf 13}, 794 (1988).
1587: %H.S. Eisenberg {\it el al.}, Phys. Rev. Lett. {\bf 81}, 3383 (1998).
1588:
1589: \bibitem{hol}
1590: T. Holstein, Ann. Phys. (N.Y.) {\bf 8}, 325 (1959).
1591:
1592: \bibitem{KAT} G. Kalosakas, S. Aubry, and G.P. Tsironis,
1593: Phys. Rev. B {\bf 58}, 3094 (1998).
1594:
1595: \bibitem{dav} A.S. Davydov and N.I. Kislukha,
1596: Phys. Stat. Sol. (b) {\bf 59}, 465 (1973).
1597:
1598: \bibitem{acn} A. Scott, Phys. Rep. {\bf 217}, 1 (1992).
1599:
1600: \bibitem{zolo} A.V. Zolotaryuk, K.H. Spatschek, and O. Kluth,
1601: Phys. Rev. B {\bf 47}, 7827 (1993).
1602:
1603: \bibitem{TS}
1604: A. Trombettoni and A. Smerzi, Phys. Rev. Lett. {\bf 86}, 2353 (2001).
1605:
1606: \bibitem{KRB} G. Kalosakas, K.\O. Rasmussen, and A.R. Bishop
1607: Phys. Rev. Lett. {\bf 89}, 030402 (2002).
1608:
1609: \bibitem{MJKA} A.M. Morgante, M. Johansson, G. Kopidakis, and
1610: S. Aubry, Physica D {\bf 162}, 53 (2002).
1611:
1612: \bibitem{sergeAC} S. Aubry, Physica D {\bf 71}, 196 (1994).
1613:
1614: \bibitem{voul}
1615: N.K. Voulgarakis and G.P. Tsironis, Phys. Rev. B {\bf 63}, 014302 (2000).
1616:
1617: \bibitem{VKBT} N.K. Voulgarakis, G. Kalosakas, A.R. Bishop,
1618: and G.P. Tsironis, Phys. Rev. B {\bf 64}, 020301 (2001).
1619:
1620: \bibitem{dirk} D. Hennig, Phys. Rev. E {\bf 64}, 041908 (2001);
1621: Eur. Phys. J. B {\bf 24}, 377 (2001);
1622: Phys. Rev. B {\bf 65}, 174302 (2002).
1623:
1624: \bibitem{KH} S. Komarnicki and D. Hennig,
1625: J. Phys.: Cond. Matt. {\bf 15}, 441 (2003).
1626:
1627: \bibitem{HSAP} D. Hennig, E.B. Starikov, J.F.R. Archilla, and
1628: F. Palmero, J. Biol. Phys. {\bf 30}, 227 (2004).
1629:
1630: \bibitem{AQ} S. Aubry and P. Quemerais, in {\it Low-Dimensional
1631: Electronic Properties of Molybdenum Bronzes and Oxides},
1632: edited by C. Schlenker (Kluwer Academic, Dordrecht, 1989), p.~295.
1633:
1634: \bibitem{AAR} S. Aubry, G. Abramovici, and J.-L. Raimbault,
1635: J. Stat. Phys. {\bf 67}, 675 (1992).
1636:
1637: \bibitem{serge1} S. Aubry, in {\it Phase Separation in Cuprate
1638: Superconductors}, edited by K.A. M\"uller and G. Benedek
1639: (World Scientific, Singapore, 1993), p.~304.
1640:
1641: \bibitem{ABK} G.L. Alfimov, V.A. Brazhnyi, and V.V. Konotop,
1642: Physica D {\bf 194}, 127 (2004).
1643:
1644: \bibitem{LKS} E.W. Laedke, O. Kluth, and K.H. Spatschek,
1645: Phys. Rev. E {\bf 54}, 4299 (1996).
1646:
1647: \bibitem{KKM} T. Kapitula, P.G. Kevrekidis, and B.A. Malomed,
1648: Phys. Rev. E {\bf 63}, 036604 (2001).
1649:
1650: \bibitem{PKF} D.E. Pelinovsky, P.G. Kevrekidis, and D.J. Frantzeskakis,
1651: arXiv:nlin.PS/0410005.
1652:
1653: \bibitem{KMCF} P.G. Kevrekidis, B.A. Malomed, Z. Chen, and
1654: D.J. Frantzeskakis, Phys. Rev. E {\bf 70}, 056612 (2004).
1655:
1656: \bibitem{KETC} Y.V. Kartashov, A.A. Egorov, L. Torner, and
1657: D.N. Christodoulides, Opt. Lett. {\bf 29}, 1918 (2004)
1658:
1659: \bibitem{BE} L.S. Brizhik and A.A. Eremko,
1660: Physica D {\bf 81}, 295 (1995).
1661:
1662: \bibitem{fuentes} M.A. Fuentes, P. Maniadis, G. Kalosakas,
1663: K.\O. Rasmussen, A.R. Bishop, V.M. Kenkre, and Y.B. Gaididei,
1664: Phys. Rev. E {\bf 70}, 025601 (2004).
1665:
1666: \bibitem{pan} P. Panayotaros, "Multibreather solitons in the diffraction
1667: managed NLS equation", to appear in Phys. Lett. A.
1668:
1669: \bibitem{KC}
1670: Y.S. Kivshar and D.K. Campbell, Phys. Rev. E {\bf 48}, 3077 (1993).
1671:
1672: \bibitem{DKL} S. Darmanyan, A. Kobyakov, and F. Lederer,
1673: JETP {\bf 86}, 682 (1998).
1674:
1675: \bibitem{KMB} P.G. Kevrekidis, B.A. Malomed, and A.R. Bishop,
1676: J. Phys. A: Math. Gen. {\bf 34}, 9615 (2001).
1677:
1678: \bibitem{kevr} P.G. Kevrekidis, K.\O. Rasmussen, and A.R. Bishop,
1679: Phys. Rev. E {\bf 61}, 2006 (2000).
1680:
1681: \bibitem{rodrigo} R. Vicencio, unpublished.
1682:
1683: \bibitem{econ}
1684: E.N. Economou, {\it Green's functions in quantum physics},
1685: (Springer-Verlag, Berlin, 1990).
1686:
1687:
1688: \end{thebibliography}
1689:
1690:
1691: \end{document}
1692: