1: %\documentstyle[osa,psfig,manuscript]{revtex} %for submission
2: \documentstyle[osa,psfig,multicol]{revtex}
3:
4: \begin{document}
5:
6: \title{Generic features of modulational instability in nonlocal Kerr media.}
7:
8: \author{John Wyller}
9: \address{Department of Mathematical Sciences, Agricultural University
10: of Norway, P.O.Box 5065, N-1432 \AA s, Norway.}
11:
12: \author{Wieslaw Krolikowski}
13: \address{Australian Photonics Cooperative Research Centre,
14: Laser Physics Centre, Research School of Physical
15: Sciences and Engineering, Australian National
16: University, Canberra ACT 0200, Australia.}
17:
18: \author{Ole Bang}
19: \address{Department of Informatics and Mathematical Modelling,
20: Technical University of Denmark, DK-2800 Kongens Lyngby, Denmark.}
21:
22: \author{Jens Juul Rasmussen}
23: \address{Ris\o\ National Laboratory, Optics and Fluid Dynamics Department,
24: OFD - 128, P. O. Box 49, DK-4000 Roskilde, Denmark.}
25:
26: \maketitle
27:
28: \begin{abstract}
29: The modulational instability (MI) of plane waves in nonlocal Kerr media
30: is studied for a general, localized, response function.
31: It is shown that there always exists a finite number of well-separated MI
32: gain bands, with each of them characterised by a unique maximal growth rate.
33: This is a general property and is demonstrated here for the Gaussian,
34: exponential, and rectangular response functions.
35: In case of a focusing nonlinearity it is shown that although the nonlocality
36: tends to suppress MI, it can never remove it completely, irrespectively of
37: the particular shape of the response function.
38: For a defocusing nonlinearity the stability properties depend sensitively
39: on the profile of the response function.
40: It is shown that plane waves are always stable for response functions with
41: a positive-definite spectrum, such as Gaussians and exponentials.
42: On the other hand, response functions whose spectra change sign (e.g.,
43: rectangular) will lead to MI in the high wavenumber regime, provided the
44: typical length scale of the response function exceeds a certain threshold.
45: Finally, we address the case of generalized multi-component response functions
46: consisting of a weighted sum of N response functions with known properties .\\
47:
48: {\em OCIS codes:} 190.5530, 190.4420, 190.5940.
49: \end{abstract}
50:
51: \begin{multicols}{2}
52:
53: \newlength{\figwidthb}
54: \setlength{\figwidthb}{0.95\linewidth}
55:
56: \section{Introduction}
57:
58: The phenomena of modulational instability (MI) of plane waves has been
59: identified and studied in various physical systems, such as fluids
60: \cite{mi-fluid}, plasma \cite{mi-plasma}, nonlinear optics
61: \cite{mi-nlo1,mi-nlo2}, discrete nonlinear systems, such as molecular
62: chains \cite{discrete-mi-molchain} and Fermi-resonant interfaces and
63: waveguide arrays \cite{discrete-mi-array}, dispersive nonlinear
64: directional couplers with the change of refractive index following a
65: exponential relaxation law \cite{trillo} etc.
66: It has been shown that MI is strongly affected by various
67: mechanisms present in nonlinear systems, such as higher order dispersive
68: terms in the case of optical pulses \cite{mi-dispersive}, saturation of the
69: nonlinearity \cite{mi-saturation}, and coherence properties of optical beams
70: \cite{mi-coherence}.
71:
72: In this work we study the MI of plane waves propagating in a nonlinear Kerr
73: type medium with a nonlinearity (the refractive index change, in nonlinear
74: optics) that is a nonlocal function of the incident field. We consider a
75: phenomenological model
76: \begin{equation}
77: i\partial_z\psi + \frac{1}{2}\partial_x^2\psi +
78: s\psi\int_{-\infty}^{\infty} R(x^{\prime}-x)|\psi(x^{\prime})|^2
79: dx^{\prime} = 0,
80: \label{generalNLS}
81: \end{equation}
82: for the wave modulation $\psi $ where $x$ is the transverse spatial
83: coordinate and $s=1$ $(s=-1)$ corresponds to a focusing (defocusing)
84: nonlinearity. The evolution coordinate $z$ can be time, as for Bose-Einstein
85: Condensates (BEC's), or the propagation coordinate, as for optical
86: beams. We consider only symmetric spatial response functions that are
87: positive definite and (without loss of generality) obey the normalization
88: condition
89: \begin{equation}
90: \int_{-\infty }^{\infty }R(x)dx=1. \label{normalization}
91: \end{equation}
92: Thus we exclude asymmetric effects, such as those generated by asymmetric
93: temporal response functions (with $x$ being time), as in the case of the
94: Raman effect on optical pulses \cite{Wyller}.
95:
96: In nonlinear optics (\ref{generalNLS}) represents a general phenomenological
97: model for media in which the nonlinear refractive index change (or
98: polarization) induced by an optical beam is determined by some kind of a
99: transport process. It may include, e.g., heat conduction in materials with a
100: thermal nonlinearity \cite{thermal1,thermal2,thermal3} or diffusion of
101: molecules or atoms accompanying nonlinear light propagation in atomic
102: vapours \cite{suter}. Nonlocality also accompanies the propagation of waves
103: in plasma \cite{cusp,litvak,df,litvak75,juul}, and a nonlocal response in
104: the form (\ref{generalNLS}) appears naturally as a result of many body
105: interaction processes in the description of Bose-Einstein condensates \cite
106: {bose}.
107:
108: The width of the response function $R(x)$ relative to the width of the
109: intensity profile $|\psi|^2(x,z)$ determines
110: the degree of nonlocality. In the limit of a singular response we get the
111: well-known nonlinear Schr\"odinger (NLS) equation, which appears in all
112: areas of physics. Here the
113: focusing case ($s$=1) produces MI of the finite bandwidth type,
114: while the defocusing case ($s$=$-1$) predicts modulational
115: stability \cite{mi-nlo1}. When the width of the response function is finite
116: but small compared to the intensity, the model (\ref{generalNLS}) is
117: approximated by the modified NLS equation
118: \cite{KBJW,Parola98,molecular,nakamura,wk-ob}
119: \begin{equation}
120: i\partial_z\psi + \frac{1}{2}\partial_x^2\psi + s[|\psi|^2+
121: \gamma\partial_x^2|\psi|^2]\psi =0.
122: \label{weaknonlocal}
123: \end{equation}
124: Here $\gamma \ll 1$ is defined as the second virial of $R(x)$ as
125: \begin{equation}
126: \gamma \equiv \frac{1}{2}\int_{-\infty }^{\infty }x^{2}R\left( x\right) dx
127: \label{scaling0}
128: \end{equation}
129: and it scales as $\gamma \sim \sigma ^{2},$ where $\sigma $ is the
130: characteristic length of the response function. In contrast to the local NLS
131: limit ($\gamma$=0), the MI now depends not only on the sign of $s$, but
132: also on the intensity of the plane waves \cite{litvak}.
133: Finally, in the case of strong nonlocality it has been shown that
134: (\ref{generalNLS}) simplifies to a linear model, and hence there
135: is no MI in this limit \cite {Snyder97}.
136:
137: MI has thus been studied in different limits. The general case
138: (\ref{generalNLS}) has recently been investigated with respect to MI and
139: compared
140: with the weakly nonlocal limit \cite{KBJW}. Here we present an analytical
141: study of the full nonlocal case with arbitrary profile $R(x)$ whose spectrum
142: obeys a sufficient degree of smoothness, with particular emphasis on generic
143: features of the MI. The present paper complements and extends the results
144: obtained in \cite{KBJW}.
145:
146: \section{M$\allowbreak $I in the nonlocal NLS equation}
147:
148: The model (\ref{generalNLS}) permits plane wave solutions of the form
149: \begin{equation}
150: \psi (x,z)=\sqrt{\rho _{0}}\exp (ik_{0}x-i\omega _{0}z),\quad \rho _{0}>0,
151: \label{planewave}
152: \end{equation}
153: where $\rho _{0}$, $k_{0}$, and $\omega _{0}$ are linked through the
154: nonlinear dispersion relation
155: \begin{equation}
156: \omega _{0}=\frac{1}{2}k_{0}^{2}-s\rho _{0}, \label{nonlindisp}
157: \end{equation}
158: Following \cite{KBJW}, we perturb the plane wave solutions (\ref{planewave})
159: - (\ref{nonlindisp}) as follows: Assume that
160: \begin{eqnarray}
161: \psi \left( x,z\right) &=&\left[ \sqrt{\rho _{0}}+u\left( \xi ,\tau \right)
162: +iv\left( \xi ,\tau \right) \right] \exp (i\theta _{0}) \\
163: \xi &=&x-k_{0}z,\text{ }\tau =z, \\
164: \theta _{0} &=&k_{0}x-\omega _{0}z,
165: \end{eqnarray}
166: where $u$ and $v$ denote the real- and imaginary part of the perturbation.
167: Inserting this expression into the nonlocal NLS equation (\ref{generalNLS})
168: and linearizing around the solution (\ref{planewave}) - (\ref{nonlindisp})
169: yields the equations
170: \begin{equation}
171: \begin{array}{c}
172: \partial _{\tau }u+\frac{1}{2}\partial _{\xi }^{2}v=0, \\
173: \\
174: \partial _{\tau }v-\frac{1}{2}\partial _{\xi }^{2}u-2s\rho _{0}\int_{-\infty
175: }^{\infty }R(\xi ^{\prime }-\xi )u(\xi ^{\prime },\tau )d\xi ^{\prime }=0,
176: \end{array}
177: \end{equation}
178: for the perturbations $u$ and $v.$ \ By introducing the Fourier transforms
179: \begin{eqnarray}
180: \widehat{u}(k,\tau ) &=&\int_{-\infty }^{\infty }u(\xi ,\tau )\exp [ik\xi
181: ]d\xi , \\
182: \widehat{v}(k,\tau ) &=&\int_{-\infty }^{\infty }v(\xi ,\tau )\exp [ik\xi
183: ]d\xi , \\
184: \widehat{R}(k) &=&\int_{-\infty }^{\infty }R(\xi )\exp [ik\xi ]d\xi ,
185: \end{eqnarray}
186: and exploiting the convolution theorem for Fourier transforms, the
187: linearized system is converted to a set of ordinary differential equations
188: in $k$-space
189: \begin{equation}
190: \begin{array}{c}
191: \partial _{\tau }\widehat{u}-\frac{1}{2}k^{2}\widehat{v}=0, \\
192: \\
193: \partial _{\tau }\widehat{v}+\frac{1}{2}k^{2}\widehat{u}-2s\rho _{0}\widehat{%
194: R}\,\widehat{u}=0,
195: \end{array}
196: \end{equation}
197: which can be written in the compact matrix form
198: \begin{equation}
199: \partial _{\tau }\underline{X}=\underline{\underline{A}}\;\underline{X},
200: \end{equation}
201: where the vector $\underline{X}$ and matrix $\underline{\underline{A}}$ are
202: defined as
203: \begin{equation}
204: \underline{X}=\left[
205: \begin{array}{c}
206: \widehat{u} \\
207: \widehat{v}
208: \end{array}
209: \right] ,\;\underline{\underline{A}}=\left[
210: \begin{array}{cc}
211: 0 & \frac{1}{2}k^{2} \\
212: -\frac{1}{2}k^{2}+2s\rho _{0}\widehat{R}(k) & 0
213: \end{array}
214: \right] .
215: \end{equation}
216: The eigenvalues $\lambda $ of the matrix $\underline{\underline{A}}$ are
217: given by
218: \begin{equation}
219: \lambda ^{2}=k^{2}\rho _{0}\left[ s\widehat{R}(k)-\frac{1}{4\rho _{0}}k^{2}%
220: \right] . \label{generaleigenvalue}
221: \end{equation}
222: The general dispersion relation (\ref{generaleigenvalue}) constitutes the
223: basis of our study of MI.
224:
225: Let us summarize the properties of the spectrum $\widehat{R}(k)$:
226: \begin{enumerate}
227: \item
228: Since $R(x)$ is realvalued and symmetric, then so is $\widehat{R}(k)$,
229: i.e. $\widehat{R}(k)=\widehat{R}(-k)=\widehat{R}^{\ast}(k)$.
230:
231: \item
232: The normalization (\ref{normalization}) implies $\widehat{R}(0)$=1.
233:
234: \item
235: The symmetry condition for the Fourier transforms imposes $\widehat{R}'(0)$=0,
236: i.e. the spectrum has a critical point at $k$=0. Here and in the following
237: prime denotes differentiation with respect to the argument.
238:
239: \item
240: The normalization (\ref{normalization}) for $R(x)$ means by assumption that
241: $R(x)$ is absolute integrable, and hence by Riemann - Lebesque lemma we have
242: $\lim_{k\rightarrow\infty}\widehat{R}(k)=0$ \cite{Folland}.
243:
244: \item
245: The functions $R(x)$, $xR(x)$ and $x^2R(x)$ are assumed to be absolute
246: integrable, and thus by \cite{Folland} $\widehat{R}(k)$, $\widehat{R}'(k)$,
247: and $\widehat{R}''(k)$ are continuous for all $k$.
248:
249: \item
250: The response functions are characterized by typical widths and scaling
251: lengths $\sigma$ and they assume the generic form $R(x)=\sigma^{-1}\Phi
252: (x/\sigma)$, where $\Phi$ is a non-dimensional scaling function.
253: \end{enumerate}
254:
255: The spectrum $\widehat{R}\left( k\right) $ can be expressed in terms of the
256: Fourier - transform $\widehat{\Phi }$ of the scaling function $\Phi $ as
257: \begin{equation}
258: \widehat{R}\left( k\right) =\widehat{\Phi }\left( \sigma k\right) \equiv
259: \int_{-\infty }^{\infty }\Phi \left( \zeta \right) \exp \left[ i\sigma
260: k\zeta \right] d\zeta . \label{scaling1}
261: \end{equation}
262: Notice that the list of general properties 1-6 of the spectrum
263: $\widehat{R}$ carries over to $\widehat{\Phi}$.
264:
265: The dispersion relation (\ref{generaleigenvalue}) is now conveniently
266: rewritten as
267: \begin{equation}
268: \lambda ^{2}/\lambda _{0}^{2}=k^{2}\phi \left( k;s,\sigma \right) ,\text{ }
269: \label{generaldisp1}
270: \end{equation}
271: by means of\ (\ref{scaling1}), the definition
272: \begin{equation}
273: \lambda _{0}\equiv 2\rho _{0} \label{scaling2}
274: \end{equation}
275: the redefinitions
276: \begin{equation}
277: 2\sqrt{\rho _{0}}\sigma \rightarrow \text{ }\sigma ,\text{ }\frac{1}{2\sqrt{%
278: \rho _{0}}}k\rightarrow k \label{scaling3}
279: \end{equation}
280: and
281: \begin{equation}
282: \phi \left( k;s,\sigma \right) \equiv s\widehat{\Phi }\left( \sigma k\right)
283: -k^{2} \label{phi}
284: \end{equation}
285: The parameter $\sigma $ which measures the width of the response function,
286: plays the role as a {\em control parameter}. Observe that $\widehat{\Phi }
287: $ is a continuous differentiable function of $k.$ The crucial point in the
288: stability analysis is the properties of the function $\phi .$ More
289: precisely, by appealing to the list of properties 1-6 we can
290: characterize the set $\Omega $ of $k$ fulfilling the inequality $\phi \left(
291: k;s,\sigma \right) \geq 0$ for a given value of $\sigma $ as follows:
292:
293: \begin{itemize}
294: \item If $\phi \left( k\right) <0$ for all $k\geq 0,$ then $\Omega $ is
295: empty. In this case we have modulational stability.
296:
297: \item If the {\em transversality condition }$\phi ^{\prime }\neq 0$%
298: {\em is satisfied at the zeros of }$\phi ,$ the number of zeros is
299: finite. In addition, these points are distinct and isolated. In this case $%
300: \Omega $ is given as a {\em finite} {\em union of well-separated,
301: closed, bounded subintervals} of the positive $k$ - axis. In this case we
302: have MI of the finite bandwidth type.
303:
304: \item The breakdown of the transversality condition for certain values of $%
305: \sigma $, i.e.,
306: \begin{equation}
307: \phi \left( k\right) =\phi ^{\prime }\left( k\right) =0\text{ for }%
308: k=k_{c},\sigma =\sigma _{c} \label{bifurcational condition}
309: \end{equation}
310: describes bifurcation phenomena like {\em excitation, vanishing, coalescence,
311: and separation} of MI bands. A careful
312: analysis of the behavior of $\phi $ in the neighborhood of the bifurcation
313: point will reveal the type of phenomena which takes place. The number of
314: zeros of $\phi $ will change as $\sigma $ passes the critical value $\sigma
315: _{c}$. The case $\phi ^{\prime \prime }\left( k_{c}\right) >0$ for $\sigma
316: =\sigma _{c}$, can be interpreted as a condition for two MI bands to merge
317: together or that one single MI band separates into two gain bands, while $%
318: \phi ^{\prime \prime }\left( k_{c}\right) <0$ for $\sigma =\sigma _{c}$
319: represents excitation or vanishing of an MI band. The critical value $\sigma
320: _{c}$ and the type of bifurcation can be determined as follows: \ Solve the
321: equation
322: \begin{equation}
323: z\widehat{\Phi }^{\prime }\left( z\right) =2\widehat{\Phi }\left( z\right) ,%
324: \text{ }z>0 \label{bifurcational condition1}
325: \end{equation}
326: Then the widths $\sigma _{c}$ can be expressed as
327: \begin{equation}
328: \sigma _{c}=\frac{z}{\sqrt{s\widehat{\Phi }\left( z\right) }}
329: \label{critical value}
330: \end{equation}
331: provided $s\widehat{\Phi }\left( z\right) >0.$ The second derivative of $%
332: \phi $ evaluated at the bifurcation point can also be expressed in terms of $%
333: z:$%
334: \begin{equation}
335: \phi ^{\prime \prime }=\frac{z^{2}\widehat{\Phi }^{\prime \prime }\left(
336: z\right) }{\widehat{\Phi }\left( z\right) }-2 \label{second derivative}
337: \end{equation}
338: \end{itemize}
339:
340: From now on we will assume that the transversality condition is satisfied at
341: the zeros of $\phi ,$ except for a set of isolated bifurcation points. This
342: means that the behavior of the graph of $\ $spectrum $\widehat{\Phi }\left(
343: \sigma k\right) $ relative to the parabola $k^{2}$ determines the stability
344: properties. We will discuss this aspect in more detail in the coming
345: subsections. Let us first assume that $\phi \left( k;s,\sigma \right) \geq 0$
346: for some positive $k$. For such values of $k$ the normalized growth rate $%
347: \Gamma $ reads
348: \begin{equation}
349: \Gamma =\left| {\cal Re}\left( \lambda \right) \right| /\lambda _{0}=\left|
350: k\right| \sqrt{\phi \left( k;s,\sigma \right) }
351: \label{normalized growth rate}
352: \end{equation}
353: In the present paper we will consider situations where the MI of the
354: nonlocal NLS equation is of finite bandwidth type and that each gain band
355: has a unique maximum growth rate, in the same as we have for the
356: MI of the focusing NLS model and the weakly nonlocal limit, and it emerges
357: as a generic property for a large class of response functions like the
358: Gaussian, the exponentially decay function and the square pulse function. It
359: is straightforward to derive the conditions which must be fulfilled: We find
360:
361: \begin{equation}
362: \Gamma ^{\prime }=\frac{1}{2}\phi ^{-1/2}\left[ 2\phi +k\phi ^{\prime }%
363: \right] \label{derivative}
364: \end{equation}
365: and hence the maximum point $k=$ $k_{\max }$ obeys
366: \begin{equation}
367: \left[ 2\phi +k\phi ^{\prime }\right] _{k=k_{\max }}=0
368: \label{critical point}
369: \end{equation}
370: The second derivative of $\Gamma $ at $k=$ $k_{\max }$ is given as
371: \begin{equation}
372: \Gamma ^{\prime \prime }=\frac{1}{2}\phi ^{-1/2}\left[ 3\phi ^{\prime
373: }+k\phi ^{\prime \prime }\right] _{k=k_{\max }}
374: \end{equation}
375: and in order to have a local maximum at $k=$ $k_{\max }$ the function $\phi $
376: must obey the inequality
377: \begin{equation}
378: \left[ 3\phi ^{\prime }+k\phi ^{\prime \prime }\right] _{k=k_{\max }}<0
379: \label{generic condition}
380: \end{equation}
381: The condition (\ref{generic condition}) is referred to as {\em the
382: generic condition for the existence of a maximum growth rate}. One can prove
383: that if all the critical points of $\Gamma $ obey (\ref{generic condition}),
384: these points coincide, and hence we have an MI pattern which resembles the
385: MI of the local, focusing NLS.
386:
387: \subsection{The focusing case (s=1)}
388:
389: In this case the function $\phi $ defined by (\ref{phi}) appearing in (\ref
390: {generaldisp1}) and (\ref{normalized growth rate}) is given as
391: \begin{equation}
392: \phi \equiv \widehat{\Phi }\left( \sigma k\right) -k^{2} \label{focusingphi}
393: \end{equation}
394: From the list of general properties we find that $\phi \left( 0\right) =1$
395: and we can prove that the number of zeros of $\phi $, denoted by $N$, is
396: {\em odd}, i.e. $N=2m-1$ $\left( m=1,2,..\right) $ and that there are $m$
397: MI bands in this case. Notice that it always exists a closed, bounded
398: interval $\left[ 0,k_{1}\right] $ for which $\phi \left( k_{1}\right) =0$
399: and $\phi \left( k\right) >0$ for $0\leq k<k_{1}.$ We refer to $\left[
400: 0,k_{1}\right] $ as the fundamental gain \ band. This means that we always
401: have MI in the focusing case, due to the existence of the fundamental band$.$
402: In all the gain bands the growth rate is given by (\ref{normalized growth
403: rate}).
404:
405: Now, assume $\widehat{\Phi }$ is a monotonically decreasing \ function for
406: all $k\geq 0.$ In this case there is a unique $k_{1}>0$ such that $\phi
407: \left( k_{1}\right) =0$, $\phi \left( k\right) >0$ $\left( \phi \left(
408: k\right) <0\right) $ for $0\leq k<k_{1}\left( k>k_{1}\right) $ i.e. there is
409: only one band producing MI. Now, since $\phi \left( 0\right) =1,$ $\phi
410: ^{\prime }\left( 0\right) =0$ and $\phi \left( k_{1}\right) =0,$ we find
411: that $\Gamma ^{\prime }$given by (\ref{derivative}) satisfies $\Gamma
412: ^{\prime }\left( 0\right) =1$ and $\lim_{k\rightarrow k_{1}^{-}}\Gamma
413: ^{\prime }\left( k\right) =-\infty .$ Hence, since $\phi $ and $\phi
414: ^{\prime }$ are continuous, we can appeal to the intermediate value theorem
415: and conclude that there is at least one $k_{cr}$ such that $\Gamma ^{\prime
416: }\left( k_{cr}\right) =0.$
417:
418: It is also easy to study the variation of the growth rate curve with the
419: width $\sigma .$ Simple computation yields
420: \begin{equation}
421: \partial _{\sigma }\Gamma =\frac{1}{2}k^{2}\phi ^{-1/2}\widehat{\Phi }%
422: ^{\prime }\left( \sigma k\right) \label{width of growth rate}
423: \end{equation}
424: and since $\widehat{\Phi }$ is a monotonically decreasing \ function, it
425: follows that the growth rate decreases with the width $\sigma $ for a fixed
426: modulation wavenumber $k$. Moreover, the end point $k_{1}$ of the MI band is
427: a function of $\sigma ,$ and the implicit function theorem yields the change
428: of $k_{1}$ with $\sigma :$
429: \begin{equation}
430: \frac{dk_{1}}{d\sigma }=-\frac{\widehat{\Phi }^{\prime }\left( \sigma
431: k_{1}\right) k_{1}}{\widehat{\Phi }^{\prime }\left( \sigma k_{1}\right)
432: \sigma -2k_{1}} \label{velocity1}
433: \end{equation}
434: Since $\widehat{\Phi}^{\prime}(\sigma k)\leq 0$ , we will
435: have $\widehat{\Phi}^{\prime}(\sigma k_1)\sigma-2k_1<0$
436: (which is the transversality condition $\phi^{\prime}\neq 0$ evaluated at
437: $k=k_1$). Thus $k_1$ is a decreasing function of $\sigma $, which means
438: that the width of the MI band decreases in size.
439:
440: Let us use the Gaussian response function
441: \begin{equation}
442: R(x) = \frac{1}{\sigma}\Phi\left(\frac{x}{\sigma}\right); \;\;\;
443: \Phi(\zeta) = \frac{1}{\sqrt{\pi}}\exp(-\zeta^2)
444: \label{gauss}
445: \end{equation}
446: as an example. The Fourier - transform is given by
447: \begin{equation}
448: \widehat{R}(k) = \widehat{\Phi}(\sigma k) =
449: \exp\left[-\frac{1}{4}\sigma^2k^2\right] ,
450: \label{gaussf}
451: \end{equation}
452: which is a monotonically decreasing function. In Fig.~1 it appears as a
453: characteristic feature that the MI is being suppressed as the characteristic
454: widths $\sigma$ of these response functions are increased.
455: Moreover, the MI gain band has a unique maximum growth rate.
456: Notice that exactly the same characteristic features are found for the
457: exponential response \cite{KBJW}.
458:
459: \begin{figure}
460: \centerline{\psfig{figure=Wyller_fig1.eps,width=\figwidthb}}
461: \caption{The growth rate $\Gamma$ as function of the wavenumber $k$ for
462: $\protect\sigma=1,2,5$ and Gaussian response function in the focusing case.}
463: \end{figure}
464:
465: Now, let us consider
466: situations where it is not assumed that the spectrum is strictly decreasing
467: for all $k$. Then it is possible to have additional gain bands in the
468: focusing case. We assume that
469: \begin{equation}
470: \phi \left( k_{2j}\right) =\phi \left( k_{2j+1}\right) =0;
471: \label{a}
472: \end{equation}
473: \begin{equation}
474: \phi \left( k\right) >0\text{ for }k\in \left\langle
475: k_{2j},k_{2j+1}\right\rangle \label{b}
476: \end{equation}
477: for $j=1,2,..,m-1$ where $m$ is the number of MI bands. Since $\Gamma $ $%
478: \equiv k\phi ^{1/2}$ is continuous and realvalued on $\left[ k_{2j},k_{2j+1}%
479: \right] $ and differentiable on $\left\langle k_{2j},k_{2j+1}\right\rangle ,$
480: it follows from Rolles theorem that there is at least one $k_{cr}^{\left(
481: j\right) }\in \left\langle k_{2j},k_{2j+1}\right\rangle $ such that $\Gamma
482: ^{\prime}(k_{cr}^{(j)})=0$. If, in addition
483: \begin{equation}
484: \phi^{\prime}(k_{2j}) >0, \;\;\; \phi^{\prime}(k_{2j+1})<0,
485: \label{c}
486: \end{equation}
487: we find the following limits
488: \[
489: \lim_{\;\;\;\;k\rightarrow k_{2j}^+} \Gamma^{\prime}(k) = +\infty, \;\;\;
490: \lim_{\;\;\;\;\;\;\;\;k\rightarrow k_{2j+1}^-}\Gamma^{\prime}(k) = -\infty.
491: \]
492:
493: Now, if the condition (\ref{generic condition}) is satisfied, then $%
494: k_{cr}^{\left( j\right) }=k_{\max }^{\left( j\right) }$ is a unique. Hence
495: we have a unique maximum growth rate for this wavenumber band as well.
496:
497: As an example we consider the square pulse function
498: \begin{equation}
499: R\left( x\right) =\frac{1}{\sigma }\Phi \left( \frac{x}{\sigma }\right) ;%
500: \text{ }\Phi \left( \zeta \right) =\left\{
501: \begin{array}{c}
502: \frac{1}{2};\text{ }\left| \zeta \right| \leq 1 \\
503: 0;\text{ }\left| \zeta \right| >1
504: \end{array}
505: \right. ; \label{square pulse}
506: \end{equation}
507: whose Fourier - transform is given by
508: \begin{equation}
509: \widehat{R}(k) =\widehat{\Phi}(\sigma k) =\frac{\sin(k\sigma)}{k\sigma}.
510: \label{square pulsef}
511: \end{equation}
512: For small and moderate values of the width $\sigma$ we only have one
513: fundamental MI gain band.
514: The bifurcation equation (\ref{bifurcational condition1}) reads
515: \begin{equation}
516: \tan(z) = z/3
517: \label{bifurcationalsq}
518: \end{equation}
519: in this case, and by means of (\ref{critical value}) with $\widehat{\Phi}
520: (z)>0$ ($\Leftrightarrow \sin(z)>0$), we find the
521: critical values $\sigma _{c}.$ Moreover, one finds that $\phi^{\prime\prime}
522: =-z^2-6$, from which it follows that new MI bands are excited at
523: the bifurcation points. The table below summarizes the findings for the
524: lowest order excitations.
525: \begin{equation}
526: \begin{tabular}{|c|c|}
527: \hline
528: $\sigma _{c,1}=21.03$ & The second MI band is excited \\ \hline
529: $\sigma _{c,2}=52.555$ & The third MI band is excited \\ \hline
530: \end{tabular}
531: \end{equation}
532: A simple computation based on (\ref{generic condition}) now reveals that
533: each MI band has a maximum growth rate. Finally, since $\widehat{\Phi}$
534: of the square response function is a monotonically
535: decreasing function on an interval containing $[0,k_1]$ for
536: all values of $\sigma $, we conclude by appealing to (\ref{width of growth
537: rate}) and (\ref{velocity1}) that an increase in the width $\sigma $
538: decreases both the fundamental band and the maximum growth rate.
539: The expression (\ref{velocity1}) (with $k_1$ replaced with $k_i$,
540: $(i=2j,2j+1))$ and (\ref{a})-(\ref{c}) show that the higher order MI gain
541: bands move towards lower wavenumbers as the width $\sigma$ increases.
542: Fig.~2 summarizes these features graphically, and the results are consistent
543: with the findings in \cite{KBJW}.
544:
545: \begin{figure}
546: \centerline{\psfig{figure=Wyller_fig2.eps,width=\figwidthb}}
547: \caption{The growth rate $\Gamma$ as function of the wavenumber $k$ for
548: $\protect\sigma =20,40,90$ and the square response function in the
549: focusing case.}
550: \end{figure}
551:
552:
553: \subsection{The defocusing case (s=-1)}
554:
555: The function $\phi $ given by
556: \begin{equation}
557: \phi \equiv -\widehat{\Phi }\left( \sigma k\right) -k^{2};
558: \label{defocusingphi}
559: \end{equation}
560: in the defocusing case. Here the number of zeros of $\phi $, denoted by $N$,
561: is {\em even}, i.e. $N=2m$ $\left( m=0,1,2,..\right) $ and there are $m$
562: MI bands. We observe from the list of properties of the spectrum $\widehat{%
563: \Phi }\left( \sigma k\right) $ that MI can only exist in the high wavenumber
564: regime, as opposed to the focusing case. Notice that the case $m=0$
565: corresponds to the situation where $\phi \left( k\right) <0$ for all $k\geq
566: 0 $ which means modulational stability.
567:
568: A typical situation revealing stability occurs when $\widehat{\Phi}(
569: \sigma k)\geq 0$ for all $k\geq 0$, irrespective of the value of $\sigma$.
570: The spectrum of the Gaussian (\ref{gaussf}) satisfies this property.
571: The same holds true for the exponential response function
572: \begin{equation}
573: R(x) = \frac{1}{\sigma}\Phi\left(\frac{x}{\sigma}\right); \;\;\;
574: \Phi(\zeta) = \frac{1}{2}\exp(-|\zeta|)
575: \end{equation}
576: for which the Fourier - transform is given by
577: \begin{equation}
578: \widehat{R}(k) = \widehat{\Phi}(\sigma k) = \frac{1}{1+\sigma^2k^2}.
579: \end{equation}
580:
581: When $\widehat{\Phi }(k)$ changes sign, the picture
582: is somewhat more complicated. First of all, we will still have stability if
583: $\widehat{\Phi}(\sigma k)<0$ for all $k$ provided the control
584: parameter $\sigma $ is below certain threshold $\sigma_c$. The threshold
585: value $\sigma_c$ and the corresponding value of $k$ can be found by
586: requiring the bifurcation condition (\ref{bifurcational condition}) to be
587: fulfilled. Secondly, if $\sigma $ belongs to the complementary regime, i.e.,
588: $\sigma>\sigma_c$ we have $\widehat{\Phi}(\sigma k) \geq 0$ for certain
589: wavenumber intervals, and hence we may have one or more MI gain bands, whose
590: growth rate $\Gamma$ is given by (\ref{normalized growth rate}) and
591: (\ref{defocusingphi}).
592: Even in this case one can prove the existence of a unique maximum growth
593: rate within each unstable band in the same way as in the focusing case.
594:
595: Again we use the square pulse function (\ref{square pulse}) - (\ref{square
596: pulsef}) as an example. The bifurcation problem is solved by means of (\ref
597: {bifurcationalsq}) - (\ref{critical value}) with $s=-1$ and hence the extra
598: constraint that $\sin \left( z\right) <0.$ Here we also find that $\phi
599: ^{\prime \prime }=-z^{2}-6<0$, which means that new MI bands are excited
600: through the bifurcation process. The table below displays the bifurcation
601: values of $\sigma $ for the lowest order excitations.
602:
603: \begin{equation}
604: \begin{tabular}{|c|c|}
605: \hline
606: $\sigma _{c,1}=9.146$ & The f$\text{irst MI band is excited}$ \\ \hline
607: $\sigma _{c,2}=35.78$ & The s$\text{econd MI band is excited}$ \\ \hline
608: $\sigma _{c,3}=71.31$ & $\text{The third MI band isexcited}$ \\ \hline
609: \end{tabular}
610: \end{equation}
611:
612: The number of MI bands increases when the width increases. By (\ref{generic
613: condition}) each MI band has a maximum growth rate. Finally, by
614: using the expression for the velocity of the zeros $k_{2j-1}$ and $k_{2j}$
615: of $\phi $ $(j=1,2,..,m)$ in the defocusing case
616: \begin{equation}
617: \frac{dk_{i}}{d\sigma }=\frac{\widehat{\Phi }^{\prime }\left( \sigma
618: k_{i}\right) k_{i}}{-\widehat{\Phi }^{\prime }\left( \sigma k_{i}\right)
619: \sigma -2k_{i}};\text{ }i=2j-1,2j
620: \end{equation}
621: and (\ref{a}) - (\ref{c}) with $\phi $ given as (\ref{defocusingphi}) and $%
622: k_{2j}$ and $k_{2j+1\text{ }}$replaced with $k_{2j-1}$and $k_{2j},$%
623: respectively, it can be shown in a the same way as in the focusing case that
624: $\frac{dk_{i}}{d\sigma }<0$ and hence the MI bands move to the low
625: wavenumber regimes as the width increases. The latter analytical result is
626: also in accordance with the numerical results obtained in \cite{KBJW}. In
627: Fig.3.\ the MI result for the square pulse (\ref{square pulse}) in the
628: defocusing case is displayed.
629:
630: \begin{figure}
631: \centerline{\psfig{figure=Wyller_fig3.eps,width=\figwidthb}}
632: \caption{The growth rate $\Gamma$ as function of the wavenumber $k$ for
633: $\protect\sigma =20,40,90$ for the square response function in the
634: defocusing case.}
635: \end{figure}
636:
637:
638: \section{Generalized response functions}
639:
640: Now, let us consider a localized response function $R(x)$ which can be
641: written as a convex combination of $N$ response functions $R_{n}\left(
642: x\right) $ whose properties are known i.e. the weighted mean
643: \begin{equation}
644: R(x)=\sum_{n=1}^{M}f_{n}R_{n}\left( x\right) \text{ , }n=1,2,3,...,M;
645: \end{equation}
646: where
647: \begin{equation}
648: \sum_{n=1}^{M}f_{n}=1\text{ , \ }0\leq f_{n}\leq 1\text{ , }\int_{-\infty
649: }^{\infty }R_{n}\left( x\right) dx=1;
650: \end{equation}
651: and $R_{n}\left( x\right) $ is parametrized by a typical lengthscale $\sigma
652: _{n}$, i.e. $R_{n}\left( x\right) =\sigma _{n}^{-1}\Phi _{n}\left( \frac{x}{%
653: \sigma _{n}}\right) $ with $\Phi _{n}$ as the scaling functions. From now on
654: we refer to the functions $R_{n}\left( x\right) $ as ''building blocks''.
655:
656: The corresponding spectrum $\widehat{R}\left( k\right) $ appears as a convex
657: combination of $\widehat{R}_{n}\left( k\right) $
658:
659: \begin{equation}
660: \widehat{R}\left( k\right) =\sum_{n=1}^{M}f_{n}\widehat{R}_{n}\left(
661: k\right) =\sum_{n=1}^{M}f_{n}\widehat{\Phi }_{n}\left( \sigma _{n}k\right) ;
662: \end{equation}
663: where $\widehat{\Phi }_{n}$ is the spectrum of the scaling function $\Phi
664: _{n}.$ Moreover, we find in accordance with (\ref{scaling0}) that
665: \begin{equation}
666: \int_{-\infty }^{\infty }x^{2}R_{n}\left( x\right) dx\sim \sigma _{n}^{2};
667: \end{equation}
668: A typical lengthscale $\sigma $ of the composite response function $R\left(
669: x\right) $ can be defined
670: \begin{equation}
671: \sigma ^{2}=\sum_{n=1}^{N}f_{n}\sigma _{n}^{2}; \label{convex1}
672: \end{equation}
673: It is straightforward to show that the dispersion relation on normalized
674: form can be expressed as
675: \begin{equation}
676: \lambda ^{2}/\lambda _{0}^{2}=k^{2}\phi _{conv};\text{ }
677: \label{convex disprelation1}
678: \end{equation}
679: where
680: \begin{equation}
681: \phi _{conv}\equiv s\sum_{n=1}^{M}f_{n}\widehat{\Phi }_{n}\left( \sigma
682: _{n}k\right) -k^{2}; \label{convex disprelation2}
683: \end{equation}
684: which is a generalization of the dispersion relation (\ref{generaldisp1}).
685: Here $\lambda _{0}$ and $k$ are given as (\ref{scaling2}) and (\ref{scaling3}%
686: ), respectively, while $\sigma _{n}$ is the redefined width of the response
687: function:
688: \begin{equation}
689: 2\sqrt{\rho _{0}}\sigma _{n}\rightarrow \sigma _{n};
690: \end{equation}
691: Now, if the transversality condition $\phi _{conv}^{\prime }\neq 0$ is
692: satisfied on points where \ $\phi _{conv}=0,$ there are a finite union of
693: disjoint, bounded and closed subintervals of the positive $k$ axis, for
694: which $\phi _{conv}\geq 0.$ This means that the MI is of finite bandwidth
695: type in this case as well. In each gain band the normalized growth rate $%
696: \Gamma $ is given as
697: \begin{equation}
698: \Gamma \equiv \left| {\cal Re}\left( \lambda \right) \right| /\lambda
699: _{0}=\left| k\right| \phi _{conv}^{1/2}; \label{convex growth rate}
700: \end{equation}
701: In (\ref{convex disprelation1}) - (\ref{convex growth rate}) there are $2M-1
702: $ {\em control parameters}, namely the $M$ width parameters $\sigma
703: _{1},\sigma _{2},...,\sigma _{M}$ and the $M-1$ weight parameters $%
704: f_{1},f_{2},...,f_{M-1}$. To summarize, the stability properties depend on
705: the weight $f_{n}$ of each Fourier - transform $\widehat{\Phi }_{n}$ and
706: length scale $\sigma _{n}$ .
707:
708: Let us consider some special cases. Assume that the Fourier transforms
709: $\widehat{R}_n(k)$ of all the building block functions $R_n(x)$ are
710: monotonically decreasing positive functions, such as the exponential
711: and Gaussian response functions. The sum $\sum_{n=1}^{M}f_n\widehat{\Phi}_n
712: (\sigma_nkt)$ will now be a decreasing and positive function of $k$.
713: In the defocusing case $(s=-1)$ we will have modulational stability in
714: this case, while we get MI with one gain band about $k=0$ in the focusing
715: case ($s$=1), which has a unique maximum growth rate.
716: In Fig.~4. we display this phenomenon for the case when the total response
717: function is a sum of a Gaussian and an exponential response function.
718:
719: \begin{figure}
720: \centerline{\psfig{figure=Wyller_fig4.eps,width=\figwidthb}}
721: \caption{Growth rate $\Gamma$ versus wavenumber $k$ for a convex
722: combination of a Gaussian response function with weight factor
723: $f_g=f$ and width $\sigma_g$=1 and an exponential response function
724: with weight factor $f_e=1-f$ and width $\sigma_e$=2 for different
725: values of $f$.}
726: \end{figure}
727:
728:
729: In Fig.~5 we show what happens for a focusing nonlinearity when the response
730: function is a 3-component combination of a Gaussian, an exponential, and a
731: square response function. In Fig.~5 two MI gain bands are revealed.
732: A typical feature of the fundamental band is that the growth rate
733: exhibits a more complex behavior with several local extrema, as opposed to
734: the single-maximum generic situation defined by (\ref{generic condition}).
735:
736: \begin{figure}
737: \centerline{\psfig{figure=Wyller_fig5.eps,width=\figwidthb}}
738: \caption{Growth rate $\Gamma$ versus wavenumber $k$ for a convex
739: combination of a Gaussian (width $\sigma_g$=10), an exponential
740: (width $\sigma_e$=20), and a square response function (width
741: $\sigma_{sq}$=90) in the focusing case. The corresponding weight
742: factors are $f_g$=0.1, $f_e$=0.3, and $f_{sq}$=0.6.}
743: \end{figure}
744:
745:
746: \section{Conclusion}
747:
748: The linear stage of the MI for the nonlocal NLS equation is conveniently
749: studied in terms of the spectrum of the response function. From the
750: dispersion relation (\ref{generaleigenvalue}) it appears that the crucial
751: point in this discussion is the location of the spectrum of the
752: response function relative to the parabola in $k$-space. The following
753: features complement and extend the results obtained in \cite{KBJW}:
754:
755: \begin{itemize}
756: \item
757: The MI is of the finite bandwidth type. It consists of a finite number of
758: well-separated gain bands. Moreover, it is possible to predict the
759: occurrence of excitation, vanishing, coalescence and separation of
760: MI bands.
761:
762: \item
763: For a large class of response functions, each MI band has
764: a unique maximum growth rate. This property holds true for the
765: Gaussian, the exponential and the square response functions. It
766: resembles the structure of the MI bands in the focusing local
767: NLS equation.
768:
769: \item
770: In the focusing case we always find at least one MI gain band centered
771: about $k=0$. It is verified analytically that the width of this MI band,
772: as well as the corresponding growth rate, decreases when increasing the
773: width of the response function, provided the spectrum of the response
774: function is decreasing in this MI band.
775: Furthermore, additional MI bands are excited at higher wavenumbers when
776: the width parameter exceeds a certain threshold, i.e. when the nonlinearity
777: becomes sufficiently nonlocal.
778: The latter phenomenon is a unique feature of the nonlocal nonlinearity
779: and has no equivalent in the local case and the weakly nonlocal limit.
780:
781: \item
782: In the defocusing case we can either have stability or MI of the
783: finite bandwidth type. The latter situation can only occur in the high
784: wavenumber regime, and only if the width of the response function exceeds
785: a certain threshold, i.e. when the nonlinearity becomes sufficiently nonlocal.
786:
787: \item
788: In both the focusing and defocusing case the higher order MI bands
789: move towards lower wavenumbers as the width of the response function
790: increases. In the limit of strong nonlocality the MI bands vanish
791: completely. This result agrees with the fact the strongly nonlocal
792: limit of the NLS model (\ref{generalNLS}) is a linear model.
793:
794: \item
795: It is possible to study MI for more complicated multiscale scenarios
796: where the response functions decompose into a weighted mean of ''building
797: blocks'', where each ''building block'' has a characterized width.
798: In this case we get a richer repertory of response functions to deal with.
799: This aims at showing that the results we have obtained represents generic
800: features of the MI of the nonlocal NLS with general symmetric,
801: positive response functions.
802: \end{itemize}
803:
804: This work was supported by the Danish Technical Research Council
805: (STVF - Talent Grant 5600-00-0355), the Danish Natural Sciences
806: Foundation (SNF - grant 9903273), and the Graduate School in
807: Nonlinear Science (The Danish Research Academy).
808:
809: \begin{thebibliography}{99}
810: \bibitem{mi-fluid}
811: T.B. Benjamin and J.E. Feir, J. Fluid. Mech. {\bf 27}, 417 (1967).
812:
813: \bibitem{mi-plasma}
814: A. Hasegawa, {\it Plasma instabilities and Nonlinear Effects}
815: (Springer-Verlag, Heidelberg, 1975).
816:
817: \bibitem{mi-nlo1}
818: L.A. Ostrovskii, Sov. Phys. JETP {\bf 24}, 797 (1967).
819:
820: \bibitem{mi-nlo2}
821: V.I. Bespalov and V.I. Talanov, JETP Lett. {\bf 3}, 307 (1966);
822: V.I. Karpman, JETP Lett. {\bf 6}, 277 (1967).
823:
824: \bibitem{discrete-mi-molchain}
825: Yu.S. Kivshar and M. Peyrard,
826: "Modulational instabilities in discrete latices",
827: \pra {\bf 46}, 3198 (1992).
828:
829: \bibitem{discrete-mi-array}
830: P.D. Miller and O. Bang,
831: "Macroscopic dynamics in quadratic nonlinear lattices",
832: Phys. Rev. E {\bf 57}, 6038-6049 (1998).
833:
834: \bibitem{trillo}
835: S. Trillo, S. Wabnitz, G.I. Stegeman, and E.M. Wright,
836: "Parametric amplification and modulational instabilities in dispersive
837: nonlinear directional couplers with relaxing nonlinearity",
838: \josab {\bf 6}, 889-900 (1989).
839:
840: \bibitem{mi-dispersive}
841: M.J. Potasek, \ol {\bf 12}, 921 (1987)
842:
843: \bibitem{mi-saturation}
844: Yu.S. Kivshar, D. Anderson, and M. Lisak,
845: Phys. Scripta {\bf 48} 679 (1993).
846:
847: \bibitem{mi-coherence}
848: M. Soljacic, M. Segev, T. Coskun, D. Christodoulides, and A. Vishwanath,
849: \prl {\bf 84}, 467 (2000).
850:
851: \bibitem{Wyller}
852: J. Wyller, "Nonlinear wavefields in optical fibres with finite time
853: response and amplification effects", Physica D (to appear).
854:
855: \bibitem{thermal1}
856: J.P. Gordon, R.C. Leite, R.S. Moore, S.P. Porto, and J.R. Whinnery,
857: J. Appl. Phys. {\bf 36}, 3 (1965).
858:
859: \bibitem{thermal2}
860: S. Akhmanov, D.P. Krindach, A.V. Migulin, A.P. Sukhorukov, and
861: R.V. Khokhlov, IEEE J. Quant. Electron. QE-{\bf 4}, 568 (1968).
862:
863: \bibitem{thermal3}
864: M. Horovitz, R. Daisy, O. Werner, and B. Fischer,
865: \ol {\bf 17}, 475 (1992).
866:
867: \bibitem{suter} %nonlocal response in atomic vapors
868: D. Suter and T. Blasberg,
869: "Stabilization of transverse solitary waves by a nonlocal response
870: of the nonlinear medium", \pra {\bf 48}, 4583 (1993).
871:
872: \bibitem{cusp}
873: M.V. Porkolab and M.V. Goldman,
874: "Upper-hybrid solitons and oscillating-two-stream instabilities",
875: Phys. Fluids {\bf 19}, 872 (1976).
876:
877: \bibitem{litvak}
878: A.G. Litvak and A.M. Sergeev, JETP Lett. {\bf 27}, 517 (1978).
879:
880: \bibitem{df}
881: T.A. Davydova and A.I. Fishchuk,
882: "Upper hybrid nonlinear wave structures",
883: Ukr. J. Phys. {\bf 40}, 487 (1995).
884:
885: \bibitem{litvak75}
886: A.G. Litvak, V.A. Mironov, G.M. Fraiman, and A.D. Yunakovskii,
887: "Thermal self-effect of wave beams in a plasma with a nonlocal nonlinearity",
888: Sov. J. Plasma Phys. {\bf 1}, 31 (1975).
889:
890: \bibitem{juul}
891: H.L. Pecseli and J.J. Rasmussen,
892: "Nonlinear electron waves in strongly magnetized plasmas",
893: Plasma Phys. {\bf 22}, 421 (1980).
894:
895: \bibitem{bose} %nonlocal character of many-body interaction
896: F. Dalfovo, S. Giorgini, L.P. Pitaevskii, and S. Stringari, K. Goral,
897: and K. Rzazewski, \pra {\bf 61} 051601R (2000);
898: Rev. Mod. Phys. {\bf 71}, 463 (1999);
899: V.M. Perez-Garcia, V.V. Konotop, and J.J. Garcia-Ripoll,
900: Phys. Rev. E {\bf 62}, 4300 (2000).
901:
902: \bibitem{KBJW}
903: W. Krolikowski, O. Bang, J.J. Rasmussen, and J. Wyller.
904: "Modulational instability in nonlocal nonlinear Kerr media",
905: Phys. Rev. E (to appear).
906:
907: \bibitem{Parola98}
908: A. Parola, L. Salanich, and L. Reatto,
909: "Structure and stability of bosonic clouds: Alkali-metal atoms
910: with negative scattering length",
911: \pra {\bf 57}, R3180 (1998).
912:
913: \bibitem{molecular} %nonlocal solitons in a biological system
914: X. Wang, D. W.Brown, K. Lindenberg, and B.J. West,
915: "Alternative formulation of Davydov's theory of energy transport
916: in biomolecular systems",
917: \pra {\bf 37}, 3557 (1988).
918:
919: \bibitem{nakamura} %collision of bright solitons in a nonlocal case
920: A. Nakamura, J. Phys. Soc. Japan {\bf 42}, 1824 (1977).
921:
922: \bibitem{wk-ob}
923: W. Krolikowski and O. Bang,
924: "Solitons in nonlocal nonlinear media: Exact solutions",
925: Phys. Rev. E {\bf 63}, 016610 (2001).
926:
927: \bibitem{Snyder97}
928: A. Snyder and J. Mitchell, "Accessible solitons",
929: Science {\bf 276}, 1538 (1997).
930:
931: \bibitem{Folland}
932: G.B. Folland, {\it Real analysis. Modern Techniques and Their
933: Applications}. p. 241. John Wiley \& Sons. ISBN 0471-80958-6.
934: 1984.
935: \end{thebibliography}
936:
937: \end{multicols}
938:
939: \end{document}
940: