1: \documentstyle[prl,aps,multicol,subfigure,psfig,subeqnar]{revtex}
2: %\documentstyle[prl,aps,multicol,epsfig]{revtex}
3: %\psdraft
4:
5: \newcommand{\Frac}[2]{\frac{\textstyle #1}
6: {\textstyle #2}}
7: \newcommand{\cL} { {\cal L} }
8: \newcommand{\dn} {\hskip 0.01in {\rm dn} \hskip 0.01in}
9: \newcommand{\cn} {\hskip 0.01in {\rm cn} \hskip 0.01in}
10: \newcommand{\sn} {\hskip 0.01in {\rm sn} \hskip 0.01in}
11: \newcommand{\Rtwo}{{\bf R}^2}
12: \newcommand{\Rn}{{\bf R}^n}
13: \newcommand{\demi}{\frac{1}{2}}
14: \newcommand{\ddt}{\Frac{d}{dt}}
15: \newcommand{\Ltwo}[1] {\| #1 \|_2}
16: \newcommand{\be}{\begin{equation}}
17: \newcommand{\ee}{\end{equation}}
18: \newcommand{\ba}[1] {\begin{array}{ #1 }}
19: \newcommand{\ea}{\end{array}}
20: \newcommand{\Ref}[1] { ( \ref{ #1 } ) }
21:
22: \begin{document}
23:
24:
25: \title{Stability of Repulsive Bose-Einstein Condensates in a Periodic Potential}
26: %
27: \author{J. C. Bronski$^{1}$, L. D. Carr$^{2}$,
28: %R. Carretero-Gonz\'alez$^{3}$
29: B. Deconinck$^{3}$, J. N. Kutz$^{3}$\cite{byline}, and K. Promislow$^{4}$ \\}
30: \address{$^{1}$Department of Mathematics, University of Illinois
31: Urbana-Champaign, Urbana, IL 61801, USA\\}
32: \address{$^{2}$Department of Physics, University of Washington,
33: Seattle, WA 98195-1560, USA\\}
34: \address{$^{3}$Department of Applied Mathematics, University of
35: Washington, Seattle, WA 98195-2420, USA\\}
36: \address{$^{4}$Department of Mathematics, Simon Fraser University,
37: Burnaby, B.C., CANADA V5A 1S6\\}
38: \maketitle
39:
40: \date{\today}
41:
42: \begin{abstract}
43: %
44: The cubic nonlinear Schr\"odinger equation with repulsive nonlinearity and
45: an elliptic function potential models a quasi-one-dimensional repulsive
46: dilute gas Bose--Einstein condensate trapped in a standing light wave. New
47: families of stationary solutions are presented. Some of these solutions have
48: neither an analog in the linear Schr\"odinger equation nor in the
49: integrable nonlinear Schr\"odinger equation. Their stability is
50: examined using analytic and numerical methods.
51: All trivial-phase stable solutions are deformations of the ground state of the
52: linear Schr\"odinger equation. Our results show that a large number of
53: condensed atoms is sufficient to form a stable, periodic condensate.
54: Physically, this implies stability of states near the Thomas--Fermi limit.
55: %
56: \end{abstract}
57:
58:
59: \pacs{}
60: % PACS numbers??
61:
62: \begin{multicols}{2}
63:
64: \section{Introduction}
65:
66: Recent experiments on dilute--gas Bose-Einstein condensates (BECs) have
67: generated great interest in macroscopic quantum
68: phenomena~\cite{ketterle1,dalfovo1} in both the theoretical and experimental
69: physics community. Such BECs are experimentally realized when certain gases
70: are super-cooled below a critical temperature and trapped in electromagnetic
71: fields~\cite{huang}.
72: %
73: Many BEC experiments use harmonic confinement.
74: Recently, however, there has been much interest in sinusoidal
75: confinement of repulsive
76: BECs using standing light waves.
77: %
78: Such BECs have been used to study phase
79: coherence~\cite{anderson3,hagley,crawford}
80: and matter-wave diffraction~\cite{ovchinnikov1}.
81: They have also been predicted to apply to quantum logic~\cite{jaksch1,brennen},
82: matter-wave transport~\cite{choi1}, and matter-wave gratings.
83: In this paper, we consider the dynamics and stability of repulsive BECs trapped
84: in standing light waves.
85:
86: A mean--field description for the macroscopic BEC wavefunction is constructed
87: using the Hartree--Fock approximation~\cite{hartree} and results in the
88: Gross-Pitaevskii equation~\cite{pitaevskii1,gross1}. The dimensions of the
89: BEC play an important role: 1D, 2D, and 3D BECs all behave in a radically
90: different manner~\cite{petrov1,petrov2}. In the quasi-1D regime, the
91: Gross-Pitaevskii equation reduces to the 1D nonlinear Schr\"odinger equation
92: (NLS) with an external potential. This regime holds when the transverse
93: dimensions of the condensate are on the order of its healing length and the
94: longitudinal dimension is much longer than its transverse
95: dimensions~\cite{carr15,carr22}. In this regime the BEC remains phase
96: coherent and the governing equations are one-dimensional.
97: This is in contrast to a truly 1D mean-field theory which requires
98: transverse dimensions on the order of or less than
99: the atomic interaction length.
100:
101: %For quasi--one--dimensional
102: %configurations, the mean--field of the BEC is modeled by the cubic nonlinear
103: %Schr\"odinger equation (NLS) with a potential~\cite{dalfovo1,carr22,key1,dekker1}.
104: %The various traps which are used to confine
105: %the BEC have led to new solutions of the NLS with a
106: %potential~\cite{prl,kunze1,kivshar}. The content of this paper applies to
107: %standing wave confinement of the BEC. BECs trapped in such a way have been
108: %used to study phase coherence~\cite{anderson3,hagley} and matter-wave
109: %diffraction~\cite{ovchinnikov1} and have been predicted to have applications in
110: %quantum logic~\cite{jaksch1,brennen}, matter-wave transport~\cite{choi1}, and for
111: %matter-wave gratings. We present new solutions pertaining to the BEC with a
112: %standing wave potential. Their dynamics, stability, and confinement
113: %under perturbations is studied both analytically and numerically.
114:
115: The recent trapping of a BEC in a hollow blue-detuned laser beam~\cite{bongs1}
116: demonstrates that a quasi-1D BEC is experimentally realizable. A variety of other
117: experiments~\cite{ketterle1,key1,dekker1,bongs1,andrews1,close1,matthews1}
118: are also modeled by the 1D NLS with an external potential.
119: Upon rescaling, the governing evolution is given by
120: %
121: \begin{equation}
122: \label{eqn:NLS}
123: i\psi_t = -\frac{1}{2}\psi_{xx} + |\psi|^2 \psi
124: + V(x) \psi \, ,
125: \end{equation}
126: %
127: where $\psi(x,t)$ is the macroscopic wave function of the condensate and $V(x)$
128: is an experimentally generated macroscopic potential. Confinement in a
129: standing light wave results in $V(x)$ being periodic. In a recent
130: experiment~\cite{anderson3,hagley}, a shallow harmonic potential was applied in
131: addition to a standing light wave. The standing light wave in this case was
132: sufficiently intense so that the condensate was strongly localized in each
133: well. This is referred to as the tight-binding regime. Additionally, the
134: apparatus was tilted vertically so that gravity caused tunneling between
135: wells. Our theoretical findings consider the complimentary experiment in which
136: the condensate is free to move between wells.
137: With the advent of quasi-1D,
138: cylindrical geometries~\cite{bongs1}, additional harmonic confinement is no
139: longer necessary and the BEC dynamics considered here are applicable.
140:
141: To model the quasi-1D confinement produced by a standing light wave,
142: we use the periodic potential
143: %
144: \begin{equation}
145: V(x) = -V_0~ {\rm sn}^2(x,k)
146: \label{eqn:potential}
147: \end{equation}
148: %
149: where ${\rm sn}(x,k)$ denotes the Jacobian elliptic sine
150: function~\cite{abro} with elliptic modulus $0\leq k\leq 1$. In the
151: limit $k=0$ the potential is sinusoidal and thus $V(x)$ is a
152: standing light wave. For intermediate values (e.g. $k<0.9$) the potential
153: closely resembles the sinusoidal behavior and thus provides a good
154: approximation to a standing light wave. Finally, for $k\rightarrow 1^-$,
155: $V(x)$ becomes an array of well-separated hyperbolic secant potential barriers
156: or wells. The potential is plotted in Fig.~\ref{fig:Sn_potential} for values
157: of $k=0, 0.9, 0.999$ and $0.999999$. Only for $k$ very near one (e.g.
158: $k>0.999$) does the solution appear visibly elliptic.
159: %
160: %%%%%%%%% figure 1 %%%%%%%%%%%%%%%%%%%%%%%%
161: %
162: \begin{figure}[htb]
163: \centerline{\psfig{figure=Sn_pot.eps,width=83mm,silent=}}
164: \begin{center}
165: \begin{minipage}{83mm}
166: \caption{ \label{fig:Sn_potential} The ${\rm sn}^2(x,k)$ structure of
167: the potential for varying
168: values of $k$. Note that the $x$-coordinate has been scaled by the period of the
169: elliptic function. This period approaches infinity as $k\rightarrow 1$.
170: Since ${\rm sn}(x,k)$ is periodic in $x$
171: with period $4K(k)=4\int_0^{\pi/2}{d\alpha}/{\sqrt{1-k^2\sin^2\alpha}}$,
172: $V(x)$ is periodic in $x$ with period $2K(k)$.}
173: \end{minipage}
174: \end{center}
175: \end{figure}
176: %
177: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
178: %
179: \noindent
180: The freedom in
181: choosing $k$ allows us to consider much more general potentials
182: than considered previously~\cite{barra1,barra2,steel1,steel2}
183: and allows for great flexibility
184: in considering a wide variety of physically realizable potentials.
185:
186: The paper is outlined as follows: in the next section we derive and
187: consider various properties and limits of
188: two types of explicit solutions of Eq.~(\ref{eqn:NLS}) with
189: (\ref{eqn:potential}). Section~III develops the analytic framework
190: for the linear stability properties of the new solutions of Sec.~II.
191: The stability results are confirmed by numerical computations. In certain
192: cases, the stability analysis is intractable and we rely solely on
193: simulations to determine stability.
194: %
195: %Nonstationary solutions are
196: %considered numerically in Sec.~IV. This includes breathers and
197: %time-- and space--periodic solutions. The possible applications
198: %of these new solutions and their relationship to experiment are
199: %considered in Sec.~V with emphasis give to the possible construction
200: %of such solutions from existing experimental techniques.
201: %
202: We conclude the paper in Sec.~IV with a brief summary and highlights
203: of the primary results of the paper and their consequences for
204: BEC dynamics and confinement.
205:
206:
207:
208:
209: \section{Stationary Solutions}
210:
211: Equation~(\ref{eqn:NLS}) with $V(x)=0$ is an integrable equation and many
212: explicit solutions corresponding to various boundary conditions are known. A
213: comprehensive overview of these solutions is found in~\cite{belokolos}.
214: If $V(x)\neq 0$, the NLS is not
215: integrable. In this case, only small classes of explicit solutions can most
216: likely be obtained. Our choice of potential ~(\ref{eqn:potential}) is
217: motivated by the form of the stationary solution of the NLS with $V(x)=0$. An
218: overview of these stationary solutions and their properties is found
219: in~\cite{carr15}. At present, we restrict our attention to stationary solutions of
220: Eq.~(\ref{eqn:NLS}), $i.e.,$ solutions whose time-dependence is restricted to
221: %
222: \begin{equation}
223: \psi(x,t) = r(x)~\exp(-i \omega t+i \theta(x)) \, .
224: \label{eqn:ansatz}
225: \end{equation}
226: %
227: If $\theta_x\equiv 0$, then the solution is referred to as having trivial
228: phase and we choose $\theta(x)=0$. Substituting the ansatz
229: Eq.~(\ref{eqn:ansatz}) in Eq.~(\ref{eqn:NLS}) and dividing out the exponential factor
230: results in two equations: one from the real part and one from the imaginary
231: part. The second equation can be integrated:
232: %
233: \begin{equation}
234: \theta(x) = c\int_0^x \frac{dx'}{r^2(x')} \, ,
235: \label{eqn:genphase}
236: \end{equation}
237: %
238: where $c$ is a constant of integration. Note that $\theta(x)$ is a
239: monotonous function of $x$. Substitution of this result in the remaining
240: equation gives
241: %
242: \begin{equation}
243: \omega r^4(x)=\frac{c^2}{2}-\frac{r^3(x)r''(x)}{2}+r^6(x)-V_0~{\rm sn}^2(x,k)
244: r^4(x).
245: \label{eqn:ode}
246: \end{equation}
247: %
248: The following subsections describe two classes of solutions of this equation.
249:
250: \subsection*{Type A}
251:
252: \subsubsection{Derivation}
253:
254: For these solutions, $r^2(x)$ is a quadratic function of
255: sn$(x,k)$:
256: %
257: \begin{equation}
258: r^2(x) = A~{\rm sn}^2(x,k)+B.
259: \label{eqn:quadratic}
260: \end{equation}
261: %
262: Substituting this ansatz in Eq.~(\ref{eqn:ode}) and equating the coefficients of
263: equal powers of ${\rm sn}(x,k)$ results in relations among the solution
264: parameters $\omega, c, A$ and $B$ and the equation parameters $V_0$ and $k$.
265: These are
266: %
267: \begin{subeqnarray}\label{eqn:parametersA}
268: \omega &=& \frac{1}{2}\left(1+k^2+3B-\frac{BV_0}{V_0+k^2}\right),\\
269: c^2 &=& B~\left(1+\frac{B}{V_0+k^2}\right)\left(V_0+k^2+B k^2\right),\\
270: A &=& V_0+k^2.
271: \end{subeqnarray}
272: %
273: For a given potential $V(x)$, this solution class has one free parameter
274: $B$ which plays the role of a constant background level or offset. The freedom
275: in choosing the potential gives a total of three free parameters: $V_0$, $k$
276: and $B$.
277:
278: The requirements that both $r^2(x)$ and $c^2$ are positive imposes conditions
279: on the domain of these parameters:
280: %
281: \begin{subeqnarray}\label{eqn:validityA}
282: &V_0\geq -k^2,~~ B\geq0,~~~~\mbox{or}&\\
283: &V_0\leq -k^2,~~ -(V_0+k^2)\leq B \leq
284: -\left(1+\displaystyle{\frac{V_0}{k^2}}\right) \, .&
285: \end{subeqnarray}
286: %
287: %%%%%%%%% figure 2 %%%%%%%%%%%%%%%%%%%%%%%%
288: %
289: \begin{figure}[htb]
290: \centerline{\psfig{figure=validity1.eps,width=83mm,silent=}}
291: \begin{center}
292: \begin{minipage}{83mm}
293: \caption{ \label{fig:validity1} The region of validity of the solutions of Type
294: A is displayed shaded for a fixed value of $k$. The edges of these regions
295: correspond to various trivial phase solutions.}
296: \end{minipage}
297: \end{center}
298: \end{figure}
299: %
300: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
301: %
302: \noindent The region of validity of these solutions is displayed in
303: Fig.~\ref{fig:validity1}.
304:
305:
306: For typical values of $V_0, k$ and $B$, the above equations give rise to
307: solutions of Eq.~(\ref{eqn:NLS}) which are not periodic in $x$: $r(x)$ is
308: periodic with period $2K(k)$, whereas $\exp(i \theta(x))$ is periodic with
309: period $T=\theta^{-1}(2\pi)$. In general these two periods $2K(k)$ and $T$ are
310: not commensurable. Thus, requiring periodic solutions results in another condition,
311: namely $2K(k)/T=p/q$, for two positive integers $p$ and $q$. The most
312: convenient way to express this phase quantization condition is to assume the
313: potential ($i.e.,$ $V_0$ and $k$) is given, and to consider values of $B$ for which
314: the quantization condition is satisfied. Introducing $\beta=B/(V_0+k^2)$, we
315: find
316: %
317: \begin{equation}\label{eqn:genquant}
318: \pm\frac{\sqrt{\beta(1+\beta)(1+k^2\beta)}}{\pi}\int_0^{K(k)}\frac{dx}{{\rm
319: sn}(x,k)^2+ \beta}=\frac{p}{q}.
320: \end{equation}
321: %
322: This equation is solved for $\beta$, after which $B=\beta (V_0+k^2)$. For
323: numerical simulations, the number of periods of the potential is set. This
324: determines $q$, limiting the number of solutions of Eq.~(\ref{eqn:genquant}).
325: Solutions with the same periodicity as the potential require $p/q=1$.
326:
327: Note that solutions of Type A reduce to stationary solutions of Eqs.~(\ref{eqn:NLS})
328: and (\ref{eqn:potential}) with $V_0=0$. Furthermore, all stationary solutions
329: of the integrable equation are obtained as limits of solutions of Type A.
330:
331: \subsubsection{Limits and Properties}
332:
333: The properties of these solutions are best understood by considering their
334: various limit cases.
335:
336: {\bf The trivial phase case:} The solutions of Type A have trivial phase when
337: $c=0$. Since $c^2$ has three factors which are linear in $B$ (see
338: Eq.~(\ref{eqn:parametersA})), there are three choices of $B$ for which this occurs:
339: $B=0$, $B=-(V_0+k^2)$ and $B=-(V_0+k^2)/k^2$. These possibilities are three of
340: the four boundary lines of the region of validity in Fig.~\ref{fig:validity1}.
341: Note that the remaining boundary line ($V_0=-k^2$) corresponds to $r^2(x)=B$,
342: which gives rise to a plane wave solution. Using Jacobian elliptic
343: function identities~\cite{abro}, one finds that the three other boundary lines give rise
344: to simplified solution forms: $B=0$ gives
345: %
346: \begin{equation}\label{eqn:sn}
347: r(x)=\sqrt{V_0+k^2}~{\rm sn}(x,k), \,\, ~\omega=\frac{1+k^2}{2}.
348: \end{equation}
349: %
350: $B=-(V_0+k^2)$ gives
351: %
352: \begin{equation}\label{eqn:cn}
353: r(x)=\sqrt{-(V_0+k^2)}~{\rm cn}(x,k), \,\, ~\omega=\frac{1}{2}-V_0-k^2,
354: \end{equation}
355: %
356: where ${\rm cn}(x,k)$ denotes the Jacobian elliptic cosine function.
357: Lastly, $B=-(V_0+k^2)/k^2$ gives
358: %
359: \begin{equation}\label{eqn:dn}
360: r(x)=\frac{\sqrt{-(V_0+k^2)}}{k}~{\rm dn}(x,k), \,\,
361: ~\omega=-1-\frac{V_0}{k^2}+\frac{k^2}{2},
362: \end{equation}
363: %
364: where ${\rm dn}(x,k)$ denotes the third Jacobian elliptic function.
365: Solution~(\ref{eqn:sn}) is valid for $V_0\geq-k^2$, whereas the other two
366: solutions~(\ref{eqn:cn}) and (\ref{eqn:dn}) are valid for $V_0\leq-k^2$.
367: The amplitude of
368: these solutions as a function of potential strength $V_0$ is shown in
369: Fig.~\ref{fig:bif}.
370:
371: %%%%%%%%% figure 3 %%%%%%%%%%%%%%%%%%%%%%%%
372: %
373: \begin{figure}[htb]
374: \centerline{\psfig{figure=bif3.eps,width=83mm,silent=}}
375: \begin{center}
376: \begin{minipage}{83mm}
377: \caption{ \label{fig:bif} The amplitude of the trivial phase solutions of Type A
378: versus the potential strength $V_0.$}
379: \end{minipage}
380: \end{center}
381: \end{figure}
382: %
383: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
384:
385: %%%%%%%%% figure 4 %%%%%%%%%%%%%%%%%%%%%%%%
386: %
387: \begin{figure}[htb]
388: \centerline{\psfig{figure=trivphase.eps,width=83mm,silent=}}
389: \begin{center}
390: \begin{minipage}{83mm}
391: \caption{\label{fig:triv1} Trivial phase solutions for $k=0.5$. $V(x)$ is
392: indicated with a solid line. For the top figure $V_0=1$. For the bottom figure
393: $V_0=-1$.}
394: \end{minipage}
395: \end{center}
396: \end{figure}
397: %
398: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
399:
400: %%%%%%%%% figure 5 %%%%%%%%%%%%%%%%%%%%%%%%
401: %
402: \begin{figure}[htb]
403: \vspace*{6mm}
404: \centerline{\psfig{figure=trigfigs.eps,width=83mm,silent=}}
405: \begin{center}
406: \begin{minipage}{83mm}
407: \caption{\label{fig:trig} Phase and amplitude of the
408: trigonometric solutions. For all these figures,
409: the solid line denotes $V(x)$, the dashed line is $r(x)$ and the dotted line is
410: $\theta(x)/(2\pi)$. Note that $\theta(x)$ becomes piecewise constant, as $B$
411: approaches the boundary of the region of validity. Far away from this boundary,
412: $\theta(x)$ is essentially linear.}
413: \end{minipage}
414: \end{center}
415: \end{figure}
416: %
417: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
418:
419:
420: Both ${\rm cn}(x,k)$ and ${\rm sn}(x,k)$ have zero average as functions of $x$
421: and lie in $[-1,1]$. On the other hand, ${\rm dn}(x,k)$ has nonzero average.
422: Its range is $[\sqrt{1-k^2},1]$. Furthermore, ${\rm cn}(x,k)$ and ${\rm
423: sn}(x,k)$ are periodic in $x$ with period $4K(k)$, whereas ${\rm dn}(x,k)$ is
424: periodic with period $2K(k)$. These properties matter greatly for the stability
425: analysis, as will be seen in Section 3.
426: Some solutions with trivial phase are shown in Fig.~\ref{fig:triv1}.
427:
428:
429: {\bf The trigonometric limit:} In the limit $k\rightarrow 0$, the elliptic
430: functions reduce to trigonometric functions and
431: $V(x)=-V_0~\sin^2(x)=(V_0/2)~\cos(2x)-V_0/2$. Then
432: %
433: \begin{equation}\label{eqn:trig}
434: r^2(x)=V_0~\sin^2(x)+B,~~~\omega=\frac{1}{2}+B.
435: \end{equation}
436: %
437: In this case, the phase integral Eq.~(\ref{eqn:genphase}) results in
438: %
439: \begin{equation}\label{trig:phase}
440: \tan(\theta(x))=\pm\sqrt{1+V_0/B}~\tan(x).
441: \end{equation}
442: %
443: Note that this formula guarantees that the resulting solution is periodic with
444: the same period as the potential, so no phase quantization is required. In the
445: trigonometric limit, the wedge between the two regions of validity in
446: Fig.~\ref{fig:validity1} disappears. This is no surprise, as in this limit,
447: ${\rm dn}(x,k)\rightarrow 1$, and the third trivial phase solution reduces
448: to a plane wave solution. The cornerpoint of the region of validity also moves
449: to the origin. Some trigonometric solutions are illustrated in Fig.~\ref{fig:trig}.
450:
451:
452: {\bf The solitary wave limit:} $k=1$. In this limit the elliptic functions
453: reduce to hyperbolic functions. Specifically, ${\rm sn}(x,k)=
454: \tanh(x)$. Hence in this limit, the potential has only a
455: single well or a single peak. Then $V_0<0$ gives rise to a repulsive potential,
456: whereas $V_0>0$ gives rise to an attractive potential: $V(x)=-V_0~\tanh^2(x)$.
457: In this case the phase $\theta(x)$ of Eq.~(\ref{eqn:genphase}) can be calculated
458: explicitly:
459: %
460: \begin{subeqnarray}\label{eqn:gensoliton}
461: r^2(x)&=&(V_0+1)~\tanh^2(x)+B,\\
462: \theta(x)\!\!&=&\!\!\sqrt{\frac{B}{V_0+1}}x\!+\!\arctan\!\!\left(\!
463: \sqrt{\frac{V_0+1}{B}}\tanh(x)\!\right)\!,~~~~~
464: \end{subeqnarray}
465: %
466: which is valid for $V_0>-1$ and $B>0$. This solution is a
467: stationary solitary wave of depression on a positive background. It
468: is a deformation of the gray soliton solution of the NLS
469: equation with repulsive nonlinearity. Note that these solutions can exist with
470: an attractive potential provided $-1<V_0<0$. Two solutions with repulsive
471: potential are illustrated in Fig.~\ref{fig:solitons}a-b.
472: Another solution is valid for $B=-(V_0+1)>0$:
473: $r(x)=\sqrt{-(V_0+1)}~{\rm sech}(x)$ and
474: $\theta(x)=0$.
475: This solution represents a stationary elevated solitary wave. It is
476: reminiscent of the bright soliton solution of the NLS equation with attractive
477: nonlinearity. This solution is shown in Fig.~\ref{fig:solitons}c. A surprising
478: consequence of considering Eq.~(\ref{eqn:NLS}) is that the potential strength
479: $V_0$ acts as a switch between the equation with repulsive and attractive
480: nonlinearity, as illustrated by these solitary wave solutions.
481:
482: Understanding the solitary wave limit facilitates the understanding of what
483: occurs for $k\rightarrow 1$. In this case the solutions of Type A
484: reduce to a periodic train of solitons with exponentially small interactions
485: as illustrated in Fig.~\ref{fig:solitons}d.
486:
487:
488: %%%%%%%%% figure 6 %%%%%%%%%%%%%%%%%%%%%%%%
489: %
490: \begin{figure}[htb]
491: \vspace*{6mm}
492: \centerline{\psfig{figure=solitonss.eps,width=83mm,silent=}}
493: \begin{center}
494: \begin{minipage}{83mm}
495: \caption{\label{fig:solitons} Solutions with k=1 (a,b,c) or $k\rightarrow 1$
496: (d). The solid line denotes $V(x)$, the dashed line is
497: $r(x)$ and the dotted line is $\theta(x)/2\pi$. In (d), a value of $k=1-10^{-16}$
498: was used.}
499: \end{minipage}
500: \end{center}
501: \end{figure}
502: %
503: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
504:
505:
506: %{\bf The Kronig--Penney limit:} by scaling the $x$-variable, another interesting
507: %limit is obtained as $k\rightarrow 1$. In (\ref{eqn:NLS}), replace $x$ by
508: %$2K(k)x/\pi$. This gives
509: %
510: %\begin{equation}
511: %\label{eqn:scalednls}
512: %i\hat{\psi}_t = -\frac{\pi^2}{8K^2(k)}\hat{\psi}_{xx} + |\hat{\psi}|^2
513: %\hat{\psi} + \hat{V}(x) \hat{\psi},
514: %\end{equation}
515: %
516: %with $\hat{V}(x)=-V_0~{\rm sn}^2(2K(k)x/\pi,k)$ and
517: %$\hat{\psi}(x,t)=\hat{r}(x)\exp(-i\omega
518: %t+i\hat{\theta}(x))=\psi(2K(k)x/\pi,t)$. This is the same scaling as in
519: %Fig.~\ref{fig:Sn_potential}. For all values of $k$, both the potential and
520: %$\hat{r}(x)$ are periodic in $x$ with period $2\pi$ ($\pi$ suffices if we do
521: %not consider the trivial phase limit.) As $k \rightarrow 1$, the potential
522: %limits to $\hat{V}(x)=-V_0~C(x)$, where $C(x)=1$ if $x\neq n\pi$, and
523: %$C(n\pi)=0$, for any integer $n$. This is a degenerate case of a Kronig-Penney
524: %potential, where the width of the potential wells is zero. In this limit,
525: %$\hat{r}^2(x)=(V_0+1)~C(x)+B$, which is equal to $V_0+1+B$, almost everywhere.
526: %This allows the explicit calculation of the phase (\ref{eqn:genphase})
527: %explicitly: $\hat{\theta}(x)=\pm\sqrt{B/(V_0+1)}~x$. Requiring the solutions to
528: %be periodic results in the phase quantization condition $B/(V_0+1)=p^2/q^2$,
529: %for arbitrary integers $p$ and $q$.
530: %Figure~\ref{fig:solitons}d shows this
531: %scaling limit for $k=1-10^{-16}$. In the limit, the width of the peaks of the
532: %potential and the troughs of the solution is zero.
533:
534: %Note that a more conventional Kronig-Penney potential could be obtained by
535: %scaling the potential strength $V_0$ with $k$ such that $V_0\rightarrow \infty$
536: %as $k \rightarrow 1$.
537:
538: %{\bf The Thomas--Fermi limit:} This limit is achieved when the constant
539: %background is sufficiently high compared to the amplitude of the oscillations
540: %of the solution. In other words $B\gg V_0+k^2$. Then to first order in
541: %$\epsilon^2=(V_0+k^2)/B$, $r(x)=\sqrt{B}(1+\epsilon^2 {\rm sn}^2(x,k))$ and
542: %$\theta(x)=\pm\sqrt{B}kx/\epsilon$, provided $k$ is nonzero or
543: %$\theta(x)=\sqrt{B}x$, if $k=0$. The solution is thus a plane wave with weakly
544: %oscillating envelope. The implications for stability of being near the
545: %Thomas--Fermi limit are discussed extensively in Section 3.
546:
547: %{\bf Potential strength $ V_0 \rightarrow$ 0:} In this limit, (\ref{eqn:NLS})
548: %reduces to the integrable nonlinear Schr\"{o}dinger equation. Simultaneously,
549: %solutions of Type A reduce to stationary solutions of the equation with
550: %$V_0=0$. Furthermore, all stationary solutions of the integrable equation are
551: %obtained as limits of solutions of Type A. An overview of the stationary
552: %solutions of the integrable equation and their properties is found in
553: %\cite{carr15}.
554:
555: \subsection*{Type B}
556:
557: \subsubsection*{1.~Derivation}
558:
559: For these solutions, $r^2(x)$ is linear in ${\rm cn}(x,k)$ or ${\rm dn}(x,k)$.
560: First we discuss the solution with ${\rm cn}(x,k)$. The quantities associated
561: with this solution will be denoted with a subindex 1. The quantities associated
562: with the ${\rm dn}(x,k)$ solution receive a subindex 2.
563:
564: Substituting
565: %
566: \begin{equation}\label{eqn:linear1}
567: r_1^2(x)=a_1~{\rm cn}(x,k)+b_1 \, ,
568: \end{equation}
569: %
570: in Eq.~(\ref{eqn:ode}) and equating different powers of ${\rm cn}(x,k)$
571: gives the relations:
572: %
573: \begin{subeqnarray}\label{eqn:parametersB1}
574: V_0&=&-\frac{3}{8}k^2 \, ,\\
575: \omega_1&=&\frac{1}{8}(1+k^2)+\frac{6a_1^2}{k^2} \, ,\\
576: c^2_1&=&\frac{a_1^2}{4k^6}(16a_1^2-k^4)(16a_1^2+k^2-k^4) \, ,\\
577: b_1&=&\frac{4a_1^2}{k^2} \, .
578: \end{subeqnarray}
579: %
580: The class of potentials Eq.~(\ref{eqn:potential}) is restricted by the first of
581: these relations so that $V_0$ is in the narrow range
582: $-3k^2/8\leq V_0 \leq0$. The solution class now depends on one free amplitude
583: parameter $a_1$ and the free equation parameter $k$.
584:
585: The region of validity of this solution is, as before, determined by the
586: requirements $c_1^2\geq 0$ and $r_1^2(x)\geq 0$:
587: %
588: \begin{equation}\label{eqn:validityB1}
589: |a_1|\geq\frac{k^2}{4}.
590: \end{equation}
591: %
592: The period of $r_1(x)$ is twice the period of the potential. Requiring
593: periodicity in $x$ of this first solution of Type B gives
594: %
595: \begin{equation}\label{eqn:genquant1}
596: \pm\frac{\sqrt{(\beta_1^2 \!- \!k^2)(\beta_1^2\!+\!1\!-\!k^2)}}{4\pi}
597: \!\! \int_0^{2K(k)} \!\!\!\!\!\!\!
598: \frac{dx}{4\beta_1\!+\!k~{\rm cn}(x,k)} \!=\! \frac{p}{q}.
599: \end{equation}
600: %
601: For given $k$ and integers $p$, $q$, this equation is
602: solved for $\beta_1$, from which $a_1=\beta_1 k/4$.
603:
604: The ${\rm dn}(x,k)$ solutions are found by substituting
605: %
606: \begin{equation}\label{eqn:linear2}
607: r_2^2(x)=a_2~{\rm dn}(x,k)+b_2,\\
608: \end{equation}
609: %
610: in Eq.~(\ref{eqn:ode}). Equating different powers of ${\rm dn}(x,k)$
611: imposes the following constraints on the parameters:
612: %
613: \begin{subeqnarray}\label{eqn:parametersB2}
614: V_0&=&-\frac{3}{8}k^2 \, ,\\
615: \omega_2&=&\frac{1}{8}(1+k^2)+6a_2^2 \, ,\\
616: c^2_2&=&\frac{a_2^2}{4}(16a_2^2-1)(16a_2^2+k^2-1) \, ,~\\
617: b_2&=&4a_2^2 \, .
618: \end{subeqnarray}
619: %
620: The class of potentials (\ref{eqn:potential}) is restricted as for the previous
621: solution by the first of
622: these relations. The solution class again depends on one free amplitude
623: parameter $a_2$ and the free equation parameter $k$.
624:
625: The region of validity of this solution is once more determined by the
626: requirements $c_2^2\geq 0$ and $r_2^2(x)\geq 0$:
627: %
628: \begin{equation}\label{eqn:validityB2}
629: |a_2|\geq\frac{1}{4}~~\mbox{or}~~0\leq a_2\leq\frac{\sqrt{1-k^2}}{4}.
630: \end{equation}
631: %
632:
633: The period of $r_2(x)$ is equal to the period of the potential. Requiring
634: periodicity in $x$ of this second solution of Type B gives
635: %
636: \begin{equation}\label{eqn:genquant2}
637: \pm\frac{\sqrt{(16
638: a_2^2\!-\!1)(16a_2^2\!+\!k^2\!-\!1)}}{\pi} \!\! \int_0^{K(k)} \!\!\!\!\!\!\!
639: \frac{dx}{4a_2 \!+\!{\rm
640: dn}(x,k)}\!=\! \frac{p}{q}.
641: \end{equation}
642: %
643: For given $k$ and integers $p$, $q$, this equation needs to
644: be solved to determine $a_2$.
645:
646: In contrast to solutions of Type A, solutions of Type B do not have a
647: nontrivial trigonometric limit. In fact, for solutions of Type B,
648: this limit is identical to the limit in which the potential strength
649: $V_0=-3k^2/8$ approaches zero. Thus it is clear that the solutions of Type B
650: have no analogue in the integrable nonlinear Schr\"{o}dinger equation. However,
651: other interesting limits do exist.
652:
653: \subsubsection*{2.~Limits and Properties}
654:
655: %As for solutions of Type A, we consider various limits of these solutions.
656:
657: {\bf The trivial phase case:} Trivial phase corresponds to $c=0$. This occurs
658: precisely at the boundaries of the regions of validity. For the first solution
659: of Type B, there are two possibilities: $a_1=k^2/4$ or $a_1=-k^2/4$. By replacing
660: $x$ by $x+2K(k)$, one sees that these two possibilities are completely
661: equivalent, so only the first one needs to be considered:
662: %
663: \begin{equation}\label{eqn:trivphaseB1}
664: r_1^2(x)=\frac{k^2}{4}(1+{\rm cn}(x,k))~,~~~\omega_1=\frac{1}{8}+\frac{k^2}{2}.
665: \end{equation}
666: %
667: For the second solution, there are four possibilities: $a_2=1/4$, $a_2=-1/4$,
668: $a_2=0$ and $a_2=\sqrt{1-k^2}/4$. The third one of these results in a zero
669: solution. The others give interesting trivial phase solutions. For $a_2=1/4$,
670: %
671: \begin{equation}\label{eqn:trivphaseB2}
672: r_2^2(x)=\frac{1}{4}(1+{\rm dn}(x,k))~,~~~\omega_2=\frac{1}{2}+\frac{k^2}{8}.
673: \end{equation}
674: %
675: The case $a_2=-1/4$ gives
676: %
677: \begin{equation}\label{eqn:trivphaseB3}
678: r_2^2(x)=\frac{1}{4}(1-{\rm dn}(x,k))~,~~~\omega_2=\frac{1}{2}+\frac{k^2}{8}.
679: \end{equation}
680: %
681: Finally, for $a_2=\sqrt{1-k^2}/4$,
682: %
683: \begin{equation}\label{eqn:trivphaseB4}
684: r_2^2(x)\!=\!\frac{\sqrt{1\!-\!k^2}}{4}({\rm dn}(x,k)\!+\!\sqrt{1\!-\!k^2}),
685: ~~\omega_2\!=\!\frac{1\!-\!k^2}{2}.
686: \end{equation}
687: %
688: %%%%%%%%% figure 7 %%%%%%%%%%%%%%%%%%%%%%%%
689: %
690: \begin{figure}[htb]
691: \centerline{\psfig{figure=trivphase2.eps,width=83mm,silent=}}
692: \begin{center}
693: \begin{minipage}{83mm}
694: \caption{\label{fig:trivphase2} Solutions of Type B with trivial phase. The
695: figures correspond to, from top to bottom, $k=0.5$, $k=0.9$ and $k=0.999$. The
696: potential is indicated with a solid line. The other curves are: (1) $|r_1(x)|$
697: with $a_2=k^2/4$, (2) $r_2(x)$ with $a_2=1/4$, (3) $|r_2(x)|$ with $a_2=-1/4$
698: and (4) $r_2(x)$ with $a_2=\sqrt{1-k^2}/4$.}
699: \end{minipage}
700: \end{center}
701: \end{figure}
702: %
703: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
704: %%%%%%%%% figure 8 %%%%%%%%%%%%%%%%%%%%%%%%
705: %
706: \begin{figure}[htb]
707: \centerline{\psfig{figure=solitonsB.eps,width=83mm,silent=}}
708: \begin{center}
709: \begin{minipage}{83mm}
710: \caption{\label{fig:solitonsB}
711: Solitary wave solutions of Type B. The potential is indicated with a solid
712: line. The dashed line solution corresponds to $a=0.3$, the dotted line to
713: $a=-0.3$.}
714: \end{minipage}
715: \end{center}
716: \end{figure}
717: %
718: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
719: %
720: \noindent
721: These solutions are shown in Fig.~\ref{fig:trivphase2}.
722:
723:
724: {\bf The solitary wave limit:} In this limit $V_0=-3/8$ and the potential is
725: $V(x)=3 \tanh(x)^2/8$. Since both ${\rm cn}(x,k) = {\rm sech}(x)$
726: and ${\rm dn}(x,k)= {\rm sech}(x)$ when $k=1$,
727: $r_1(x)=r_2(x)$ in the solitary wave limit. Their ranges of validity also share
728: the same limit: $|a|\geq 1/4$ with $a=a_1=a_2$. The phase can be calculated
729: explicitly and the solitary wave solution of Type B is
730: %
731: \begin{subeqnarray}\label{eqn:solitonB}
732: r^2(x)&\!\!\!=\!\!\!&4a^2+a~{\rm sech}(x),\\
733: \theta(x)&\!\!=\!\!\!&\pm \! \frac{x\sqrt{16a^2\!-\!1}}{2}\!
734: \mp \!\arctan\left(\!\sqrt{\frac{4a\!-\!1}{4a\!+\!1}}
735: \tanh\frac{x}{2}\!\right) . \,\,\,\,\,\,\,\,\,\,\,\,\,
736: \end{subeqnarray}
737: %
738: The region of validity consists of two separated regions: $a\geq 1/4$ and
739: $a\leq -1/4$. In the first region, the solution is a stationary elevated
740: solitary wave with a constant background $4a^2$. In the second region $a\leq
741: -1/4$ and the solution is a stationary solitary wave of depression with a
742: constant background $4a^2$. These solitary wave solutions are illustrated in
743: Fig.~\ref{fig:solitonsB}
744:
745:
746:
747: As for the solutions of type A, the solitary wave limit gives an idea of the
748: behavior of the solution for values of $k\rightarrow 1$, where the solution
749: behaves as a periodic array of solitons with exponentially small interactions.
750:
751:
752:
753: %{\bf The Kronig--Penney limit:} by applying the same scaling as before,
754: %Eqn.~(\ref{eqn:scalednls}) is obtained. Under this scaling, the potential for
755: %solutions of Type B limits to $\hat{V}(x)=3/8~C(x)$, as $k \rightarrow 1$.
756: %Similarly, $\hat{\psi}(x,t)=\hat{r}(x)\exp(-i\omega t+i\hat{\theta}(x))$,
757: %$\hat{r}(x)=2a$, if $x\neq n \pi$ and $\hat{r}(n\pi)=\sqrt{4a^2+a}$, for any
758: %integer $n$.
759:
760: %Note that since $V_0$ is not a free parameter, it is not possible to scale it
761: %by a function of $k$. Thus it is not possible to obtain infinitely deep, zero
762: %width wells.
763:
764: %{\bf The trigonometric limit or the zero potential strength limit:} For
765: %solutions of Type B, these limits are identical: the potential strength
766: %$V_0=-3k^2/8$, so that the limit $k \rightarrow 0$ and $V_0 \rightarrow 0$ are
767: %the same. From (\ref{eqn:parametersB1}d), $a_1$ is necessarily zero for a
768: %finite solution with $k=0$. But then $r_1^2(x)=0$. Similarly, the solution
769: %(\ref{eqn:linear2}) limits to a plane wave solution. Thus there are no
770: %interesting trigonometric solution of Type B. It is also clear that the
771: %solutions of Type B have no analogue in the integrable nonlinear
772: %Schr\"{o}dinger equation (apart from the plane wave solution).
773:
774: %{\bf Thomas--Fermi limit:} For this limit, the mean value of the solution to
775: %the amplitude of its oscillations is necessarily large. This ratio increases
776: %for $r_1(x)$ with $|a_1|$. The trivial phase case uses the minimal value of
777: %$|a_1|$. Departing from the trivial phase case brings one closer to the
778: %Thomas--Fermi limit. For $r_2(x)$ it is only possible to attain the
779: %Thomas--Fermi limit in the region of validity $|a_2|\geq 1/4$. also in this
780: %case does the trivial phase case use the minimal value of $|a_2|$. Hence for
781: %solutions of Type B, it is only possible to attain the Thomas--Fermi limit if
782: %the solution has a nontrivial phase, as for the solutions of Type A.
783:
784: %\subsection*{A note on the linear limit}
785:
786: %The linear Schr\"{o}dinger equation is
787: %obtained when the potential dominates the nonlinear term.
788: %Unfortunately, none of the solutions of Type A have a
789: %nontrivial ($i.e.,$ nonzero) linear limit. This is seen as follows: the linear
790: %limit requires the potential to dominate the nonlinear term. However, a large
791: %value for the potential strength $V_0$ implies a large value for $A=V_0+k^2$,
792: %which implies a large maximal value for $r(x)$. Hence, the nonlinear term
793: %cannot be ignored.
794: %However, there are solutions of Type B for which the potential strength
795: %dominates the solution strength, as illustrated in the bottom figure of
796: %Fig.~\ref{fig:trivphase2}: if $0\leq a_2\leq \sqrt{1-k^2}/4$ and $k$ is close
797: %to 1, the potential approaches $V(x)=3~{\rm sn}^2(x,k)/8\approx 3\tanh^2(x)/8$,
798: %whereas $r_2^2(x)=\epsilon~{\rm dn}(x,k)+\epsilon^2\approx\epsilon~{\rm
799: %sech}(x)+\epsilon^2$, where $0\leq \epsilon\leq \sqrt{1-k^2}/4$.
800:
801: %The resulting
802: %stationary equation is rewritten as Hill's equation: let
803: %$\phi(x)=r(x)\exp(i\theta(x))$, then $\phi_{xx}=2(-2V_0~{\rm
804: %sn}^2(x,k)-\omega)\phi)$. For specific values of $V_0$ this equation reduces to
805: %Lam\'{e}'s equation and explicit solutions can be written down, using elliptic
806: %functions. In the trigonometric limit $k \rightarrow 0$, the equation reduces to
807: %a Mathieu equation and solutions are written down using Mathieu functions.
808: %However, for general values of $V_0$ with $k\neq 0$, solutions in closed form
809: %cannot be written down.
810:
811:
812: \section{Stability}\label{sec:stability}
813:
814: We have found a large number of new solutions to the
815: governing Eqs.~(\ref{eqn:NLS}) and (\ref{eqn:potential}). However,
816: only solutions that are stable can be observed in experiments. In
817: this section, we consider the stability of the different solutions.
818: Both analytical and numerical results are presented for the solutions
819: with trivial phase. In contrast,
820: only numerical results are presented for the nontrivial phase
821: cases.
822:
823: We first consider the linear stability of the
824: solution~(\ref{eqn:ansatz}). To do so, consider perturbations
825: of the exact solutions of the form
826: %
827: \begin{equation}\label{eqn:perturb}
828: \psi(x,t)=(r(x)+\epsilon\phi(x,t)) \exp [{i(\theta(x)-\omega t)}]
829: \end{equation}
830: %
831: where $\epsilon \ll 1$ is a small parameter.
832: Collecting terms at $O(\epsilon)$ gives the linearized
833: equation. In terms of the real and imaginary parts
834: ${\bf U}=(U_1,U_2)^t=(Re[\phi],Im[\phi])^t$ the linearized evolution
835: is given by:
836: %
837: \begin{equation}
838: \label{eqn:linearized}
839: {\bf U}_t=JL{\bf U}=J\pmatrix{L_+ & -S \cr S & L_-} {\bf U},
840: \end{equation}
841: %
842: where
843: %
844: \begin{subeqnarray}
845: L_+ &= & -\demi \left(\partial_x^2-\frac{c^2}{r^4(x)} \right)+3r^2(x)
846: +V(x)-\omega,~~~~\\
847: L_- & = & -\demi \left(\partial_x^2-\frac{c^2}{r^4(x)} \right)+r^2(x)
848: +V(x)-\omega,\\
849: S &=& \frac{c}{r(x)}\partial_x\frac{1}{r(x)},
850: \end{subeqnarray}
851: %
852: and $J=\pmatrix{0&-1\cr 1&0}$ is a skew--symmetric matrix. The operators
853: $L$, $L_+$ and $L_-$ are Hermitian while $S$ is anti--Hermitian.
854: Considering solutions of the form ${\bf U} (x,t) = {\hat{\bf U}}(x)\exp
855: (\lambda t)$
856: gives the eigenvalue problem
857: %
858: \begin{equation}
859: \label{eqn:keith}
860: \cL \hat{\bf U}=\lambda \hat{\bf U},
861: \end{equation}
862: %
863: where $\cL=JL$ and $\lambda$ is complex. If all $\lambda$ are imaginary,
864: then linear stability is established. In contrast, if there is at
865: least one eigenvalue with a positive real part, then instability results.
866: Using the phase invariance $\psi\mapsto e^{i\gamma}\psi$ of
867: Eq.~(\ref{eqn:NLS}), Noether's theorem~\cite{classical} gives
868: %
869: \begin{equation}
870: \label{eqn:null}
871: \cL \pmatrix{0\cr r(x)}=0,
872: \end{equation}
873: %
874: which implies that $L_- r(x)=0.$ Thus $\lambda=0$ is in
875: the spectrum of $L_-$.
876: For general solutions of the form (\ref{eqn:ansatz}),
877: determining the spectrum of the associated linearized
878: eigenvalue problem (\ref{eqn:linearized}) is beyond the scope of current
879: methods.
880: However, some
881: cases of trivial phase solutions $(c=0)$ are amenable to analysis.
882:
883: The Hermitian operators $L_\pm$ are periodic Schr\"{o}dinger
884: operators and thus the spectra of these operators is real and
885: consists of bands of continuous spectrum contained in
886: $[\lambda_\pm,\infty)$\cite{classical}. Here $\lambda_\pm$ denote
887: the ground state eigenvalues of $L_\pm$ respectively. They are given by
888: %
889: \begin{equation}
890: \lambda_\pm=\inf\limits_{\|\phi\|=1} \left< \phi|L_\pm |\phi
891: \right> \label{Linf} \, ,
892: \end{equation}
893: %
894: where $\|\phi\|^2=\left<\phi|\phi\right>$.
895: %The associated
896: %eigenfunctions are the ground states of the operator $L_\pm$
897: %respectively.
898: From the relation $L_+=L_-+2r^2(x)$ it follows that $\lambda_+>\lambda_-$.
899: Also $\lambda_-\leq 0$ since $\lambda=0$ is an eigenvalue of $L_-$.
900:
901: If $\lambda_+ > 0$, then $L_+$ is positive, so we can
902: define the positive square root, $L_+^\demi$, via the spectral
903: theorem\cite{classical}, and hence the Hermitian operator
904: $H=L_+^\demi L_- L_+^\demi$ can be constructed. The eigenvalue
905: problem for $\cL$ in Eq.~(\ref{eqn:keith}) is then equivalent to
906: %
907: \begin{equation}\label{eqn:H}
908: (H+\lambda^2)\varphi=0,
909: \end{equation}
910: %
911: with $\varphi=L_+^\demi\hat{U_1}$.
912: Denote the left-most point of the spectrum of $H$
913: by $\mu_0$. If $\mu_0\geq 0$ then $\lambda^2<0$ and
914: the eigenvalues of $\cL$ are imaginary and linear stability
915: results. Since $H=L_+^\demi L_-
916: L_+^\demi$ and $L_+^\demi$ is positive, $\mu_0\geq 0$ if and only if $L_-$ is
917: non-negative.
918: In contrast, if $\mu_0<0$ then $\lambda^2>0$ and
919: $\cL$ has at least one pair of real eigenvalues with opposite
920: sign. This shows the existence of a growing mode leading
921: to instability of the solution.
922: %The spectrum of this eigenvalue problem is purely imaginary
923: %iff the operator $H$ is non-negative. Since $H=L_+^\demi L_-
924: %L_+^\demi$ this holds iff $L_-$ is non-negative.
925:
926: Three distinct cases are possible for linear stability
927: %
928: \begin{itemize}
929: %
930: \item If $r(x)>0$ then $r(x)$ is the ground state of $L_-$~\cite{classical},
931: and Eq.~(\ref{eqn:null}) implies $\lambda_-=0$ and hence $\lambda_+>0$.
932: Thus the solution (\ref{eqn:ansatz}) is linearly stable.
933: %
934: \item If $r(x)$ has a zero, it is no longer the ground
935: state~\cite{classical} and $\lambda_-<0$. Thus
936: there exists a $\psi_0$ such that $\left<\psi_0 | L_- |\psi_0 \right> <0$.
937: If in addition $\lambda_+>0$, then we can construct
938: $\phi=L_+^{-\demi}\psi_0/\|L_+^{-\demi}\psi_0\|$ which
939: gives $\left< \phi | H |\phi \right><0$. Hence
940: $\mu_0<0$ and $\cL$ has positive, real eigenvalues so that
941: the solution (\ref{eqn:ansatz}) is linearly
942: unstable.
943: %
944: \item For $\lambda_-$ and $\lambda_+$ both negative the
945: situation is indefinite and our methods are insufficient
946: to determine linear stability or instability.
947: %
948: \end{itemize}
949: %
950: In what follows,
951: these results are applied to the Type A and B trivial phase solutions
952: constructed in the preceding section. ~Specifically,~ we
953: construct~
954: the~ operators~~$L_-$
955: %
956: %%%%%%%%% figure 90 %%%%%%%%%%%%%%%%%%%%%%%%
957: %
958: \begin{figure}[t]
959: \centerline{\psfig{figure=speccn.eps,width=83mm,silent=}}
960: \begin{center}
961: \begin{minipage}{83mm}
962: \caption{The spectrum of $L_-$ for the Type A cn$(x,k)$ trivial phase solution.}
963: \label{fig:speccn}
964: \end{minipage}
965: \end{center}
966: \end{figure}
967: %
968: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
969: %
970: \noindent
971: and $L_+$ for each solution, which allows us to
972: use one of the above criteria. The analytical results are accompanied
973: by direct computations on the nonlinear governing Eqs.~(\ref{eqn:NLS}) and
974: (\ref{eqn:potential}). For all computational simulations, twelve
975: spatial periods are used. However, to better illustrate the dynamics,
976: typically four spatial periods are plotted. Moreover, all computations
977: are performed with white noise included in the initial data.
978:
979:
980:
981: \subsection{Trivial Phase: Type A}
982:
983:
984: \subsubsection{cn$(x,k)$}
985:
986: For the cn$(x,k)$ solution the $L_\pm$ operators are
987: %
988: \begin{subeqnarray}\label{eqn:speccn}
989: L_+&\!=\!&-\demi\partial_x^2-(2V_0+3k^2)\cn^2(x,k)+k^2-\demi \,\,\,\,\, \\
990: L_-&\!=\!&-\demi\partial_x^2-k^2\cn^2(x,k)+k^2-\demi \, ,
991: \end{subeqnarray}
992: %
993: with $V_0<-k^2.$ Note that $L_-$, which is independent of $V_0$, is the
994: classical 1-gap Lam\'e operator
995: %which arises in (among other
996: %places) the algebro-geometric approach to integrable systems
997: ~\cite{belokolos}. The
998: spectrum of $L_-$ can be calculated explicitly. The ground state eigenvalue is
999: $\lambda_-=(k^2-1)/2$ with associated eigenfunction dn$(x,k)$. The elliptic
1000: functions
1001: $\cn(x,k)$ and $\sn(x,k)$ are also eigenfunctions of $L_-$. They are the first
1002: and second excited state and have eigenvalue 0 and $k^2/2$ respectively.
1003: These are the only eigenvalues and the spectrum consists of the bands
1004: $[(k^2-1)/{2},0]\cup[{k^2}/{2},\infty)$. The spectrum is illustrated in
1005: Fig.~\ref{fig:speccn}.
1006:
1007:
1008: Since $\dn(x,k)>0$ and $\lambda_-=({k^2-1})/{2}<0$
1009: the arguments of the previous section imply that
1010: the cn$(x,k)$ wave is unstable whenever the operator
1011: $L_+>0$. It is clear from Eq.~(\ref{eqn:speccn}a) that $L_+$ is positive
1012: if $V_0<-(k^2 + 1/4)$ and $k^2 > 1/2$. Thus, the cn$(x,k)$ wave is unstable
1013: for parameter values in this region.
1014: %$L_+>0$ if $V_0 < -(1+k^2)/2$ and $k^2>1/2$.
1015: %Note that when
1016: %$k\approx 1$, the limit of well-separated wells, this implies that
1017: %all solutions are unstable except for possibly a small interval
1018: %near $V_0=-1$, which translates to $A\approx 0$.
1019: Moreover, this region can be enlarged to $V_0<-(k^2+1)/2$ and $k^2>1/2$ by observing that
1020: the ground state eigenvalue of an operator
1021: ${L_0 + \gamma L_1}$ is a convex function of $\gamma$:
1022: $\lambda=\Lambda(\gamma)$.
1023: This follows from the fact that the ground state eigenvalue
1024: is the minimizer of the Rayleigh quotient. Let $\alpha\in [0,1]$, then
1025: %
1026: %%%%%%%%% figure 9 %%%%%%%%%%%%%%%%%%%%%%%%
1027: %
1028: \begin{figure}[t]
1029: \centerline{\psfig{figure=trivialCNL.eps,width=83mm,silent=}}
1030: \begin{center}
1031: \begin{minipage}{83mm}
1032: \caption{Unstable evolution of a Type A cn$(x,k)$ solution
1033: given by Eq.~(\ref{eqn:cn}) over 40 time units with $k=0.5$ and
1034: $V_0=-1.0$.}
1035: \label{fig:trivialCN}
1036: \end{minipage}
1037: \end{center}
1038: \end{figure}
1039: %
1040: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1041: %
1042: %%%%%%%%% figure 10 %%%%%%%%%%%%%%%%%%%%%%%%
1043: %
1044: \begin{figure}[t]
1045: \centerline{\psfig{figure=spec-trivial3.eps,width=83mm,silent=}}
1046: \begin{center}
1047: \begin{minipage}{83mm}
1048: \caption{Wavenumber spectrum evolution of a Type A cn$(x,k)$
1049: solution given by Eq.~(\ref{eqn:cn}) over 100 time units with
1050: $k=0.5$ and $V_0=-0.55$. The modal evolution
1051: shows the band of unstable modes which result from starting
1052: with the unstable cn$(x,k)$ solution. This shows that the
1053: instability
1054: occurs in a neighborhood of the dominant wavenumber of the stationary
1055: solution.}
1056: \label{fig:spec-trivial}
1057: \end{minipage}
1058: \end{center}
1059: \end{figure}
1060: %
1061: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1062: %
1063: \begin{eqnarray}\nonumber
1064: \Lambda\!\!\!\!\!&&\!\!(\alpha\gamma_1+(1-\alpha)\gamma_2)\\\nonumber
1065: &=&\inf\limits_{\|\phi\|=1} \left<\phi|\alpha(L_0+\gamma_1
1066: L_1)+(1-\alpha)(L_0+\gamma_2 L_1)|\phi\right>\\\nonumber
1067: &\geq&\alpha \!\!\inf\limits_{\|\phi\|=1}\!\!\! \left<\phi|L_0\!+\!\gamma_1
1068: L_1|\phi\right>+(1\!-\!\alpha)\!\!\!\inf\limits_{\|\phi\|=1}\!\!\!
1069: \left<\phi|L_0\!+\!\gamma_2 L_1|\phi\right>\\
1070: &=&\alpha\Lambda(\gamma_1)+(1-\alpha)\Lambda(\gamma_2).
1071: \end{eqnarray}
1072: %
1073: Now consider the ground state eigenvalue $\lambda_+=\Lambda_+(V_0)$
1074: and note that $\Lambda_+(-k^2) = (k^2-1)/2$
1075: and $\Lambda_+(-3k^2/2) = k^2-1/2$. The line through these two
1076: points is given by $\Lambda_+(V_0)=-V_0 - (1+k^2)/2$, so by convexity
1077: $\Lambda_+(V_0)\ge -V_0 - (1+k^2)/2$ for $V_0 \in [-3k^2/2,-k^2]$.
1078: Thus, $\lambda_+=\Lambda(V_0) \ge 0$ if $V_0 \le -(1+k^2)/2$.
1079:
1080: If $k^2<1/2$, less is known. However
1081: the results of Weinstein and Keller~\cite{WK} show that
1082: the ground state eigenvalue grows as $\lambda_+\approx
1083: (2(1-k^2)|V_0|)^{1/2} + k^2- {1}/{2}$ for $-V_0\gg 1$.
1084: Hence for $k^2 \leq 1/2$ instability occurs for sufficiently negative $V_0$.
1085:
1086: The most unstable modes of the cn$(x,k)$ solution
1087: can be determined perturbatively when $\varepsilon=-2(V_0+k^2)\ll 1$. This
1088: corresponds to a solution with small amplitude. Since
1089: $L_+ = L_- + \varepsilon \cn^2(x,k)$, it follows that $L_+$ is not necessarily
1090: positive, disallowing the construction of $H$ in Eq.~(\ref{eqn:H}).
1091: %This allows
1092: %the perturbative calculation of the spectrum of $L_+L_-$ .
1093: %Denote the eigenfunctions of the $L_-$ by $\psi(\eta)$ and the
1094: %corresponding eigenvalues by $\mu(\eta)$ we find that the eigenvalues
1095: %of the composite operator $L_+L_-$ are, to leading order
1096: %
1097: %\begin{eqnarray}
1098: %\lambda^2(\eta) &=& -<\!\!\psi| L_-^2 |\psi\!\!> -
1099: %\epsilon <\!\psi | PL_- | \psi\!\!> \nonumber \\
1100: %&=& -\mu(\eta)^2 - \epsilon \mu(\eta) <\!\!\psi |P| \psi\!\!> \nonumber \\
1101: %&=& -(\mu(\eta)^2 + \epsilon \mu(\eta) p(\eta))
1102: %\end{eqnarray}
1103: %
1104: However, from Eq.~(\ref{eqn:keith}), $L_+L_- \hat{U_2}=-\lambda^2 \hat{U_2}$,
1105: which offers an alternative to Eq.~(\ref{eqn:H}) to calculate the spectrum of
1106: Eq.~(\ref{eqn:keith}). Let $\lambda=i \nu+\varepsilon \lambda_1$ and
1107: $\hat{U_2}=\phi_\nu+\varepsilon \phi_1$, where $\nu$ is an eigenvalue of $L_-$
1108: and $\phi_\nu$ is its associated normalized eigenfunction. Then a first order
1109: calculation using time-independent perturbation theory gives
1110: %
1111: \begin{equation}
1112: \lambda^2=-\nu^2-\varepsilon \nu \left<\phi_\nu|\cn^2(x,k)|\phi_\nu\right>.
1113: \end{equation}
1114: %
1115: Thus, $\lambda^2>0$ only if $-\varepsilon
1116: \left<\phi_\nu|\cn^2(x,k)|\phi_\nu\right> < \nu < 0$. Hence, only modes
1117: $\phi_\nu$ with $\nu$ in this band near zero are unstable.
1118: %
1119: %Since $P$ is a positive operator, $Pv = 2 \cn^2(x,k) v$, the right
1120: %hand side is negative except when $-p(\eta)\le \mu(\eta) \le 0$.
1121: %In other words the instability occurs in a small band near the
1122: %zero eigenvalue. Since the eigenfunction corresponding to the
1123: %zero eigenvalue is given by $\cn^2(x,k)$ the unstable band is
1124: %approximately given by $\lambda\in (-p^*,0)$, with
1125: %
1126: For these unstable modes, the eigenfunction $\phi_\nu$ is approximately the
1127: zero mode $\cn (x,k)$. Thus the onset of instability in the Fourier domain
1128: occurs near the wavenumbers of the $\cn(x,k)$ solution.
1129: %
1130: %To leading order, the band of unstable modes is then
1131: %\begin{equation}
1132: %\lambda\in\left[-\frac{\int_0^{4K(k)}\cn^4(x,k)d\!x}
1133: %{\int_0^{4K(k)}\cn^2(x,k)d\!x},0\right].
1134: %\end{equation}
1135: %
1136: This is characteristic of a modulational instability
1137:
1138: %Since the unstable modes are located
1139: %in a band about the $\cn$ mode we would expect to see growth in the
1140: %modes near the periodicity of $\cn$ - the instability is essentially
1141: %a modulational instability.
1142:
1143: To illustrate this instability, we display in Fig.~\ref{fig:trivialCN} the
1144: evolution of a cn$(x,k)$ solution over the time interval $t \in [0,40]$ for
1145: $V_0=-1.0$ and $k=0.5$. The solution goes quickly unstable with the
1146: instability generated near the first wavenumber. This agrees with the
1147: analytical prediction. It is illustrated in the evolution of the wavenumber
1148: spectrum in Fig.~\ref{fig:spec-trivial}. Here a close-up of the spectrum near
1149: wavenumber one is shown. This shows that the instability indeed
1150: occurs in a neighborhood of the dominant wavenumber of the stationary solution.
1151:
1152:
1153:
1154:
1155:
1156: \subsubsection{sn$(x,k)$}
1157:
1158: For the sn$(x,k)$ solutions the $L_\pm$ operators are given by
1159: %
1160: \begin{subeqnarray}
1161: L_+ &\!=\!& -\frac{1}{2}\partial_{xx} + (3 k^2 + 2 V_0)
1162: \sn^2(x,k)-\frac{1+k^2}{2},~~~\\
1163: L_- &\!=\!& -\frac{1}{2}\partial_{xx} + k^2 \sn^2(x,k) - \frac{1+k^2}{2}.
1164: \end{subeqnarray}
1165: %
1166: Again $L_-$ is a 1-gap Lam\'e operator, differing from $L_-$
1167: for the cn$(x,k)$ solution only by a constant. The spectrum is
1168: given by $[-{k^2}/{2},-{1}/{2}]\cup[0,\infty)$. It again follows from
1169: the work of Weinstein and Keller~\cite{WK} that for sufficiently large values
1170: of $V_0$
1171: the ground state eigenvalue of $L_-$ is approximately given by
1172: %
1173: \begin{equation}
1174: \lambda_+ \approx (2V_0)^\frac{1}{2} - \frac{1+k^2}{2}
1175: \end{equation}
1176: %
1177: %%%%%%%%% figure 11 %%%%%%%%%%%%%%%%%%%%%%%%
1178: %
1179: \begin{figure}[t]
1180: \centerline{\psfig{figure=trivialDN2L.eps,width=83mm,silent=}}
1181: \begin{center}
1182: \begin{minipage}{83mm}
1183: \caption{Stable evolution of the Type A dn$(x,k)$ solutions given
1184: by Eq.~(\ref{eqn:dn}) over 80 time units with $k=0.5$ and $V_0=-1.0$.
1185: \label{fig:trivialDN}}
1186: \end{minipage}
1187: \end{center}
1188: \end{figure}
1189: %
1190: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1191: %
1192: \noindent
1193: and thus $L_+$ is positive definite for sufficiently large $V_0$.
1194: This, in turn, implies instability of the sn$(x,k)$ solution for
1195: sufficiently large $V_0$, which corresponds to large amplitude
1196: solutions. The sn$(x,k)$ solution goes quickly unstable in a
1197: similar fashion to the $\cn(x,k)$ solution (see Fig.~\ref{fig:trivialCN}).
1198:
1199:
1200: %
1201: %Figure~\ref{fig:trivialSN} depicts the evolution
1202: %of the sn$(x,k)$ branch of solution over the time interval
1203: %$t \in [0,40]$ and for $V_0=1.0$ and $k=0.5$. The computational
1204: %run was for twelve spatial periods but only four periods are depicted to
1205: %better illustrate the instability.
1206: %
1207: %Figure~\ref{fig:trivialSN} depicts the evolution of an
1208: %exact sn$(x,k)$ solution under perturbation.
1209: %The sn$(x,k)$ solution quickly goes
1210: %unstable in a similar fashion to the $\cn$ solution.
1211:
1212: %%%%%%%%%% figure 1 %%%%%%%%%%%%%%%%%%%%%%%%
1213: %%
1214: %\begin{figure}[t]
1215: %%\centerline{\psfig{figure=trivialSN.eps,width=83mm,silent=}}
1216: %\begin{center}
1217: %\begin{minipage}{83mm}
1218: %\caption{Unstable evolution of the Type A sn$(x,k)$ branch of solutions
1219: %given by Eq.~(\ref{eqn:sn}) over 40 time units with $k=0.5$ and
1220: %$V_0=1.0$.
1221: %Only four of the twelve spatial periods of the computational run are
1222: %depicted so as to illustrate the resulting instability more clearly.
1223: %\label{fig:trivialSN}}
1224: %\end{minipage}
1225: %\end{center}
1226: %\end{figure}
1227: %%
1228: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1229:
1230:
1231: \subsubsection{dn$(x,k)$}
1232:
1233: From the previously established results, linear stability for the dn$(x,k)$
1234: solutions follows immediately since $r(x)>0$, because dn$(x,k)$ has no zeros.
1235: %
1236: %can be shown
1237: %for all values of $k$. The linearized operators take the form
1238: %
1239: %\begin{equation}
1240: % L_-=-\demi\partial_x^2 +(1+\frac{2V_0}{k^2})\dn^2(x,k)
1241: % +1-k^2/2,
1242: %\end{equation}
1243: %
1244: %and $L_+=L_-+2\dn^2(x,k).$ From the phase invariance of
1245: %Eq.~(\ref{eqn:NLS}) we find that $L_-r(x)=0$. Since $r(x)>0$,
1246: %it must be the ground state. Further, $\lambda_-=0$ and
1247: %$\lambda_+>0$, so that
1248: %we have stability of the entire $\dn$ branch of solutions.
1249: %
1250: Thus in contrast to the cn$(x,k)$ and sn$(x,k)$ solutions,
1251: the dn$(x,k)$ solutions given by
1252: Eq.~(\ref{eqn:dn}) are linearly stable.
1253: Figure~\ref{fig:trivialDN} displays the evolution
1254: of a dn$(x,k)$ solution over the time interval
1255: $t \in [0,80]$ for $V_0=-1.0$ and $k=0.5$.
1256: Although noise was added to the initial data, the solution
1257: shape persists and the solution
1258: is stable, as predicted analytically.
1259: For this case, the wavenumber spectrum
1260: is supported primarily by three modes: the zero mode which determines the
1261: offset, and two other modes which determine the oscillation frequency
1262: of the dn$(x,k)$ solution. Even with large perturbations, this
1263: solution persists. This indicates that the offset of a solution is
1264: important for its stability. This observation is reconfirmed for
1265: other stable solutions below.
1266: %the offset being a key ingredient for the
1267: %stabilization.
1268:
1269:
1270:
1271:
1272:
1273: \subsection{Trivial Phase: Type B}
1274:
1275:
1276:
1277: \subsubsection{cn$(x,k)$}
1278:
1279: The Type B trivial phase solution is obtained for $a_1=\pm {k^2}/{4}$ and
1280: corresponding amplitude $|r(x)|=({k}/{2})\sqrt{1+\cn(x,k)}$. The solution
1281: $r(x)$ is not strictly positive. The operator $L_+$ is
1282: %
1283: \begin{equation}
1284: L_+ \!=\!-\demi\partial_x^2\!+\!\frac{k^2}{4}\!-\!\frac{1}{8}
1285: \!+\!\frac{3}{8}k^2\sn^2 (x,k) \!+\!3a_1\cn (x,k). \!\!\!
1286: \end{equation}
1287: %
1288: Thus we find that the situation
1289: is indeterminate.
1290: %However, as $k$ increases,
1291: %$\lambda_+$ increases and we arrive at the instability case $\lambda_+>0.$
1292:
1293: Numerical simulations for the Type B cn$(x,k)$ solutions given by
1294: Eq.~(\ref{eqn:trivphaseB1}) are illustrated in Fig.~\ref{fig:lin-CN}. This
1295: figure displays the evolution of the cn$(x,k)$ branch of solution for $k=0.5$
1296: (top panel) and $k=0.999$ (bottom panel) over the time interval $t \in
1297: [0,800]$ and $t\in [0,400]$ respectively for $V_0=-3k^2/8$. For both $k=0.5$
1298: and $k=0.999$ the solutions are unstable, but this instability manifests
1299: itself only after several hundred time units. Figure~\ref{fig:spec-lin} shows
1300: the evolution of the wavenumber spectrum for both these cases. For $k=0.5$,
1301: the onset of instability occurs near wavenumber one as is the case of
1302: Type A
1303: solutions. After 800 time units, the wavenumbers have only just begun to
1304: spread, causing the solution~to
1305: %
1306: %%%%%%%%% figure 12 %%%%%%%%%%%%%%%%%%%%%%%%
1307: %
1308: \begin{figure}[t]
1309: \centerline{\psfig{figure=lin-CNL.eps,width=83mm,silent=}}
1310: \begin{center}
1311: \begin{minipage}{83mm}
1312: \caption{Unstable evolution of the Type B cn$(x,k)$ solutions
1313: given by Eq.~(\ref{eqn:trivphaseB1}) for $k=0.5$ (top panel) and
1314: $k=0.999$ (bottom panel) for $a_1=k^2/4$ and $V_0=-3k^3/8$.
1315: \label{fig:lin-CN}}
1316: \end{minipage}
1317: \end{center}
1318: \end{figure}
1319: %
1320: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1321: %
1322: %%%%%%%%% figure 13 %%%%%%%%%%%%%%%%%%%%%%%%
1323: %
1324: \begin{figure}[t]
1325: \centerline{\psfig{figure=spec-lin.eps,width=83mm,silent=}}
1326: \begin{center}
1327: \begin{minipage}{83mm}
1328: \caption{Wavenumber spectrum evolution of a Type B cn$(x,k)$
1329: solution given by Eq.~(\ref{eqn:cn}) for $a_1=k^2/4$ and corresponding to
1330: $k=0.5$ (left panel) and $k=0.999$ (right panel) of Fig~\ref{fig:lin-CN}.
1331: The evolution shows that the unstable band of modes is generated
1332: near wavenumber one and for $k=0.999$ near wavenumber one and its
1333: harmonics.
1334: \label{fig:spec-lin}}
1335: \end{minipage}
1336: \end{center}
1337: \end{figure}
1338: %
1339: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1340: %
1341: %%%%%%%%% figure 14 %%%%%%%%%%%%%%%%%%%%%%%%
1342: %
1343: \begin{figure}[t]
1344: \centerline{\psfig{figure=lin-DN2L.eps,width=83mm,silent=}}
1345: \begin{center}
1346: \begin{minipage}{83mm}
1347: \caption{Stable evolution of a Type B dn$(x,k)$ solution
1348: given by Eq.~(\ref{eqn:trivphaseB2}) for $k=0.999$ and $a_2=1/4$.
1349: %Note that there is sufficient DC offset to stabilize
1350: %the condensate.
1351: \label{fig:lin-DN2}}
1352: \end{minipage}
1353: \end{center}
1354: \end{figure}
1355: %
1356: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1357: %
1358: \noindent
1359: destabilize. For $k=0.999$, the solution is
1360: composed of a much larger number of wavenumbers which destabilize much more
1361: quickly than the $k=0.5$ case. Here the instability is generated near
1362: wavenumber one and its harmonics.
1363:
1364:
1365:
1366:
1367:
1368: \subsubsection{dn$(x,k)$}
1369:
1370: The trivial phase dn$(x,k)$ solution requires $c=0$ which is achieved for
1371: $a_2=\pm {1}/{4}$, $a_2=0$, or $a_2={\sqrt{1-k^2}}/{4}.$
1372: Thus three distinct parameter regimes need to be considered.
1373: The relevant operators in this case are
1374: %
1375: \begin{subeqnarray}
1376: L_+&\!\!\!\!=\!\!\!\!&-\demi\partial_x^2\!-\!\frac{k^2\!+\!1}{8}\!+\!6a_2^2
1377: \!\!+\!\!\frac{3k^2}{8}\sn^2 \! (x,\!k) \!+\!3a_2\dn (x,\!k), ~~~~~~~ \\
1378: L_-&\!\!\!\!=\!\!\!\!&-\demi\partial_x^2\!-\!\frac{k^2\!+\!1}{8}\!-\!2a_2^2
1379: \!\!+\!\!\frac{3k^2}{8}\sn^2 \! (x,\!k) \!+\!a_2\dn (x,\!k).
1380: \end{subeqnarray}
1381: %
1382: The case $a_2=1/4$ gives $L_-r(x)=0$ with $r(x)>0$. Hence from the linear
1383: stability criteria, these waves are stable for all values of $k$. As with the
1384: $a_2=1/4$ case, the regime where $a_2={\sqrt{1-k^2}}/{4}$ gives a solution
1385: $r(x)$ which is strictly positive and is the ground state of $L_-$.
1386: Thus stability follows for all values of $k$. The last parameter regime, for
1387: which $a_2=-1/4$, is indeterminate since both $\lambda_-$ and $\lambda_+$ are
1388: negative and our linear stability analysis is inconclusive.
1389:
1390: These analytic predictions are confirmed in
1391: Figs.~\ref{fig:lin-DN2}--\ref{fig:lin-DN}. In Fig.~\ref{fig:lin-DN2} the
1392: evolution of a dn$(x,k)$ solution is shown for $a_2=1/4$ and $k=0.999$. As
1393: predicted analytically, this parameter regime is stable for all $k$ values.
1394: This simulation once again illustrates the importance of an offset for
1395: stabilizing the condensate~\cite{prl}. In contrast to this stable evolution,
1396: the case $a_2=-1/4$ is unstable as illustrated in Fig.~\ref{fig:lin-DN}. The
1397: linear stability results in this case are indeterminate. ~However, the
1398: numerical~simulations
1399: %
1400: %%%%%%%%% figure 15 %%%%%%%%%%%%%%%%%%%%%%%%
1401: %
1402: \begin{figure}[t]
1403: \centerline{\psfig{figure=lin-DNL.eps,width=83mm,silent=}}
1404: \begin{center}
1405: \begin{minipage}{83mm}
1406: \caption{Unstable evolution of a Type B dn$(x,k)$ solution
1407: given by Eq.~(\ref{eqn:trivphaseB3}) for $k=0.5$ (top panel) and
1408: $k=0.999$ (bottom panel) given $a_2=-1/4$.
1409: In this case, there is no offset to stabilize the condensate.
1410: \label{fig:lin-DN}}
1411: \end{minipage}
1412: \end{center}
1413: \end{figure}
1414: %
1415: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1416: %
1417: %%%%%%%%% figure 16 %%%%%%%%%%%%%%%%%%%%%%%%
1418: %
1419: \begin{figure}[t]
1420: \centerline{\psfig{figure=nontrivialBL.eps,width=83mm,silent=}}
1421: \begin{center}
1422: \begin{minipage}{83mm}
1423: \caption{Evolution of a nontrivial phase Type A solution with
1424: $V_0=1.0$ and $B=1$ (top panel) and $B=1/2$ (bottom panel).
1425: For $B$ sufficiently large, the offset provided is able
1426: to stabilize the condensate whereas for $B$ below a critical
1427: threshold the condensate destabilizes as shown for $B=1/2$.
1428: \label{fig:nontrivialA}}
1429: \end{minipage}
1430: \end{center}
1431: \end{figure}
1432: %
1433: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1434: %
1435: \noindent
1436: conclusively show the evolution to be unstable for all $k$ values.
1437: For this case, the offset of the solution is insufficient to stabilize the
1438: condensate. We note that for small values of $k$, the onset of instability
1439: occurs after a very long time. Higher values of $k$ result in instabilities
1440: on a much faster time scale. Finally, we consider the parameter regime for
1441: which $a_2={\sqrt{1-k^2}}/{4}$. In this case, the analytic predictions once
1442: again suggest stability for all $k$ values. We do not illustrate this case
1443: since it is qualitatively very similar to Fig.~\ref{fig:lin-DN}. However, in
1444: contrast to the $a_2=1/4$ case, for values of $k$ close to one, there is a
1445: negligible amount of offset, distinguishing this stable case from previous
1446: ones. For these values of $k$, the solution has a small amplitude compared to
1447: the potential so that the behavior is essentially linear and stability is
1448: achieved because the condensate is trapped in the wells of the potential, as
1449: in ordinary quantum mechanics.
1450:
1451:
1452:
1453:
1454:
1455: %%%%%%%%%% figure 1 %%%%%%%%%%%%%%%%%%%%%%%%
1456: %%
1457: %\begin{figure}[t]
1458: %%\centerline{\psfig{figure=lin-DN3.eps,width=83mm,silent=}}
1459: %\begin{center}
1460: %\begin{minipage}{83mm}
1461: %\caption{Stable evolution of the Type B dn$(x,k)$ branch of solutions
1462: %given by Eq.~(\ref{eqn:trivphaseB2}) for $k=0.999$ and
1463: %$a_2=\sqrt{1-k^2}/4$.
1464: %Only four of the twelve spatial periods of the computational run are
1465: %%depicted so as to illustrate the resulting stable evolution more
1466: %clearly. Just as with Fig.~\ref{fig:lin-DN2}, there is sufficient DC
1467: %offset
1468: %to stabilize the condensate.
1469: %\label{fig:lin-DN3}}
1470: %\end{minipage}
1471: %\end{center}
1472: %\end{figure}
1473: %%
1474: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1475:
1476:
1477: \subsection{Nontrivial Phase}
1478:
1479:
1480:
1481: As stated at the beginning of Section \ref{sec:stability}, determining the
1482: linear stability for nontrivial phase solutions is not amenable to analysis.
1483: This leads us to consider the stability of nontrivial phase solutions using
1484: numerical computations.
1485:
1486: To begin, consider the trigonometric limit of the nontrivial phase solutions
1487: of Type A. These solutions are given by Eqs.~(\ref{eqn:trig}) and
1488: (\ref{eqn:genphase}). Figure~\ref{fig:nontrivialA} depicts the evolution of a
1489: pair of initial conditions with $V_0=1.0$ and for which $B=1$ (top panel) and
1490: $B=1/2$ (bottom panel). Since $B$ determines the offset of the condensate,
1491: these numerical results show directly the importance of this offset for
1492: stability. In contrast, if the offset is too small, it is unable to stabilize
1493: the condensate.
1494:
1495:
1496:
1497: %\subsection{nontrivial phase: Type B}
1498:
1499: For Type B solutions, qualitatively nothing changes from the
1500: dynamics illustrated for the trivial phase case. In particular, numerical
1501: simulations can be performed using exact solutions which are constructed
1502: subject to the phase quantization condition given by either
1503: Eq.~(\ref{eqn:genquant1}) or (\ref{eqn:genquant2}). A numerical shooting
1504: method is used to find appropriate values of $a_2$ for which a
1505: phase--quantized, periodic solution exists. Once this is achieved, numerical
1506: simulations can easily be performed. Note that any integer value $p$ is allowed
1507: as input for the phase quantization conditions, provided solutions exist for
1508: the parameter values. It turns out this imposes a lower bound on value of $p$.
1509: In the simulations, the actual value of $p$ does not affect the stability of
1510: the solution. Increasing the phase--quantization integer $p$ leads to a
1511: solution with a steeper phase profile, suggesting a more unstable situation.
1512: However, this phase effect is balanced by an increased offset $a_2$ of the
1513: amplitude. Qualitatively, the dynamics are as depicted in
1514: Figs.~\ref{fig:lin-DN2}--\ref{fig:lin-DN}. Thus the nontrivial phase solutions
1515: of Type B are stable for $a_2>1/4$ and for $0<a_2<\sqrt{1-k^2}/4$, whereas the
1516: nontrivial phase solution is unstable for $a_2<-1/4$.
1517:
1518:
1519: %\section{Nonstationary Solutions}
1520:
1521: %\subsection{breathers}
1522:
1523: %\subsection{others}
1524:
1525:
1526: %\section{Applications}
1527:
1528: %\subsection{growing the condensate}
1529:
1530: %\subsection{growing the potential: matter wave grating}
1531:
1532: \section{Summary and Conclusions}
1533:
1534: We considered the repulsive nonlinear Schr\"odinger
1535: equation with an elliptic function potential as a
1536: model for a trapped, quasi-one-dimensional Bose-Einstein
1537: condensate. Two new families of periodic solutions of
1538: this equation were found and their stability was investigated both
1539: analytically and numerically. Using analytical results for
1540: trivial phase solutions, we showed that solutions with sufficient offset
1541: are linearly stable.
1542: %
1543: Moreover, all such stable solutions are
1544: deformations of the ground state of the linear Schr\"odinger equation.
1545: %
1546: This is confirmed with extensive numerical
1547: simulations on the governing nonlinear equation.
1548: %
1549: Likewise, nontrivial phase
1550: solutions are stable if their density is sufficiently offset.
1551: %
1552: %These conclusions also hold for solutions with nontrivial phase.
1553: %
1554: %Even in the case
1555: %of nontrivial phase, only the dn$(x,k)$ solutions are found
1556: %to be stable provided a sufficiently high DC offset is
1557: %present.
1558: %
1559: Since we are modeling a Bose-Einstein
1560: condensate trapped in a standing light wave, our results imply that
1561: a large number of condensed atoms is sufficient
1562: to form a stable, periodic condensate. Physically, this implies
1563: stability of states near the Thomas--Fermi limit.
1564:
1565: To quantify this phenomena, we consider the $k=0$ limit and note
1566: that from Eqs.~(\ref{eqn:NLS}) and~(\ref{eqn:trig}), the number
1567: of particles per well $n$ is given by
1568: $n=(\int_0^\pi |\psi(x,t)|^2 dx)/\pi=V_0/2 + B$.
1569: In the context of the BEC, and for a fixed atomic coupling strength,
1570: this means a large number of condensed atoms
1571: per well $n$ is sufficient to provide an offset on the order of
1572: the potential strength. This ensures stabilization of the condensate.
1573: Alternatively, a condensate with a large enough number of atoms can be
1574: interpreted as a developed condensate for which the nonlinearity
1575: acts as a stabilizing mechanism.
1576:
1577:
1578:
1579: {\bf Acknowledgments:} We benefited greatly from discussions with Ricardo
1580: Carretero-Gonz\'alez and William Reinhardt. The work of J. Bronski,
1581: L. D. Carr, B. Deconinck, and J. N. Kutz
1582: was supported by National Science Foundation Grants DMS--9972869, CHE97--32919,
1583: DMS--0071568, and DMS--9802920 respectively. K. Promislow acknowledges support
1584: from NSERC--611255
1585:
1586: %\bibliographystyle{prsty}
1587:
1588:
1589: %\bibliography{/u/lcarr/research/writing/refs/refs}
1590:
1591: \begin{thebibliography}{10}
1592:
1593: \bibitem[*]{byline} to whom correspondence should be addressed
1594:
1595:
1596: \bibitem{ketterle1}
1597: W. Ketterle, D.~S. Sturfee, and D.~M. Stamper-Kurn, in {\em Proceedings of the
1598: International School of Physics "Enrico Fermi"} (IOS Press, Amsterdam;
1599: Washington, D.C., 1999), pp. 67--176.
1600:
1601: \bibitem{dalfovo1}
1602: F. Dalfovo, S. Giorgini, L.~P. Pitaevskii, and S. Stringari, Rev. Mod. Phys.
1603: {\bf 71}, 463 (1999).
1604:
1605: \bibitem{huang}
1606: K. Huang, {\em Statistical Mechanics}, (John Wiley, New York, 1963).
1607:
1608: \bibitem{anderson3}
1609: B.~P. Anderson and M.~A. Kasevich, Science {\bf 282}, 1686 (1998).
1610:
1611: \bibitem{hagley}
1612: E.~W. Hagley {\it et~al.}, Science {\bf 283}, 1706 (1999).
1613:
1614: \bibitem{crawford}
1615: M.L. Chiofalo and M.P. Tosi, Phys. Lett. A {\bf 268} 406 (2000).
1616:
1617: \bibitem{ovchinnikov1}
1618: Y.~B. Ovchinnikov {\it et~al.}, Phys. Rev. Letts. {\bf 83}, 284 (1999).
1619:
1620: \bibitem{jaksch1}
1621: D. Jaksch {\it et~al.}, Phys. Rev. Letts. {\bf 81}, 3108 (1998);
1622:
1623: \bibitem{brennen}
1624: G.~K. Brennen, C.~M. Caves, P.~S. Jessen, and I.~H. Deutsch, Phys. Rev. Letts.
1625: {\bf 82}, 1060 (1999).
1626:
1627: \bibitem{choi1}
1628: D.-I. Choi and Q. Niu, Phys. Rev. Letts. {\bf 82}, 2022 (1999).
1629:
1630: \bibitem{hartree} G. Baym, {\em Lectures in Quantum Mechanics},
1631: (Addison-Wesley, Redwood City, CA), Ch.~20.
1632:
1633:
1634: \bibitem{pitaevskii1}
1635: L.~P. Pitaevskii, Sov. Phys. JETP {\bf 13}, 451 (1961).
1636:
1637: \bibitem{gross1}
1638: E.~P. Gross, Nuovo Cimento {\bf 20}, 454 (1961).
1639:
1640: \bibitem{petrov1}
1641: D.~S. Petrov, M. Holzmann, and G.~V. Shylapnikov, Phys. Rev. Letts. {\bf 84},
1642: 2551 (2000).
1643:
1644: \bibitem{petrov2}
1645: D.S. Petrov, G.V. Shlyapnikov, and J.T.M. Walraven, e-print
1646: cond-mat/0006339 (2000).
1647:
1648: \bibitem{carr15}
1649: L.D. Carr, C.W. Clark, and W.P. Reinhardt, Phys. Rev. A {\bf 62} 0436XX-1,
1650: e-print cond-mat/9911177 (2000).
1651:
1652: \bibitem{carr22}
1653: L.D. Carr, M.~A. Leung, and W.P. Reinhardt, J. Phys. B {\bf 33} p.xxx,
1654: e-print cond-mat/0004287 (2000).
1655:
1656: \bibitem{key1}
1657: M. Key {\it et~al.}, Phys. Rev. Letts. {\bf 84}, 1371 (2000).
1658:
1659: \bibitem{dekker1}
1660: N.~H. Dekker {\it et~al.}, Phys. Rev. Letts. {\bf 84}, 1124 (2000).
1661:
1662: \bibitem{prl} J. C. Bronski, L. Carr, B. Deconinck, and J. N. Kutz, PRL
1663:
1664: \bibitem{kunze1}
1665: M. Kunze and {\it et. al}, Physica D {\bf 128}, 273 (1999).
1666:
1667: \bibitem{kivshar}
1668: Y.~S. Kivshar, T.~J. Alexander, and S.~K. Turitsyn, e-print cond-mat/9907475 (1999).
1669:
1670: \bibitem{bongs1}
1671: K. Bongs {\it et~al.}, cond-mat/0007381 (unpublished).
1672:
1673: \bibitem{andrews1}
1674: M.~R. Andrews {\it et~al.}, Science {\bf 273}, 84 (1996).
1675:
1676: \bibitem{close1}
1677: J.~D. Close and W. Zhang, J. Opt. B {\bf 1}, 420 (1999).
1678:
1679: \bibitem{matthews1}
1680: M.~R. Matthews {\it et~al.}, Phys. Rev. Letts. {\bf 83}, 2498 (1999).
1681:
1682: \bibitem{abro}
1683: {\em Handbook of Mathematical Functions}, edited by M. Abramowitz and I.~A.
1684: Stegun (National Bureau of Standards, Washington, D. C., 1964).
1685:
1686: \bibitem{barra1}
1687: S. Theodorakis and E. Leontidis, J. Phys. A {\bf 30}, 4835 (1997).
1688:
1689: \bibitem{barra2}
1690: F. Barra, P. Gaspard, and S. Rica, Phys. Rev. E {\bf 61}, 5852 (2000).
1691:
1692: \bibitem{steel1}
1693: K. Berg-S{\o}renson and K. M{\o}lmer, Phys. Rev. A {\bf 58}, 1480 (1998).
1694:
1695: \bibitem{steel2}
1696: M.~J. Steel and W. Zhang, e-print cond-mat/9810284 (1998).
1697:
1698: \bibitem{belokolos}{\it Algebro-geometric approach to nonlinear integrable
1699: equations},
1700: E.~D. Belokolos, A. ~I. Bobenko, V.~Z. Enol'skii, A.~R. Its and V.~B. Matveev
1701: (Springer-Verlag, Berlin, 1994).
1702:
1703: \bibitem{classical} R. Courant and D. Hilbert, {\it Methods of Mathematical
1704: Physics}, (Wiley, New York, 1989).
1705:
1706: \bibitem{WK}
1707: M. I. Weinstein and J. B. Keller, SIAM J. Appl. Math. {\bf 45},
1708: 200 (1985)
1709:
1710: \end{thebibliography}
1711:
1712: \end{multicols}
1713: \end{document}
1714:
1715:
1716:
1717:
1718:
1719:
1720:
1721: