1: \documentclass[prl,twocolumn,showpacs,floatfix]{revtex4}
2: \usepackage{graphicx}
3: \begin{document}
4: \title {Stability of the solutions of the Gross-Pitaevskii equation}
5: \author{A. D. Jackson$^1$, G. M. Kavoulakis$^2$, and E. Lundh$^3$}
6: \affiliation{$^1$Niels Bohr Institute, Blegdamsvej 17, DK-2100,
7: Copenhagen \O, Denmark \\
8: $^2$Mathematical Physics, Lund Institute of
9: Technology, P.O. Box 118, SE-22100 Lund, Sweden \\
10: $^3$Department of Physics, KTH, SE-10691, Stockholm, Sweden}
11: \date{\today}
12:
13: \begin{abstract}
14:
15: We examine the static and dynamic stability of the solutions of
16: the Gross-Pitaevskii equation and demonstrate the intimate
17: connection between them. All salient features related to
18: dynamic stability are reflected systematically in static
19: properties. We find, for example, the obvious result that
20: static stability always implies dynamic stability and present a
21: simple explanation of the fact that dynamic stability can exist
22: even in the presence of static instability.
23:
24: \end{abstract}
25: \pacs{03.75.Kk, 67.40.Vs}
26: \maketitle
27:
28: \section{Introduction}
29:
30: The question of excitations is central to the study of cold
31: atoms. Numerous experimental and theoretical investigations
32: have been devoted to the study of a variety of collective and
33: elementary excitations in these gases including vortex states,
34: solitary waves, and normal modes. When determining the
35: properties of an excited state, it is natural to consider two
36: kinds of possible instabilities -- static and dynamic. In the
37: former case, one wishes to determine whether a state which
38: extremizes the energy is a genuine local minimum of the energy
39: subject to certain physically motivated constraints. Since the
40: wave function extremizes the energy, infinitesimal
41: perturbations will make no first-order change of the energy.
42: The state will be stable provided that an arbitrary
43: infinitesimal variation of the wave function necessarily
44: increases the energy to second order.
45:
46: For the problem of dynamic stability, one conventionally
47: considers the temporal evolution of arbitrary infinitesimal
48: perturbations to the wave function using approximate linearized
49: equations (e.g., the Bogoliubov equations). While the full
50: time-dependent equations conserve probability, the linearized
51: equations do not, and the resulting eigenvalue problem is
52: non-Hermitean. If the corresponding eigenvalues are real, the
53: system is dynamically stable. In this case small-amplitude
54: time-dependent perturbations of the solution lead to bounded
55: motion about the original equilibrium position. Instability is
56: indicated by complex eigenvalues, and the associated
57: exponential growth of small-amplitude perturbations drives the
58: system to a new state beyond the scope of the linearized
59: equations.
60:
61: Here, we will explore the intimate connection between these two
62: seemingly-different criteria for stability. Two main points
63: will emerge. First, static stability implies dynamic
64: stability, but dynamic stability can exist even in the absence
65: of static stability. Second, the transition from dynamic
66: stability to instability is reflected in the features of the
67: corresponding problem of static stability (subject to
68: appropriate constraints). Here, inspired by the experiment of
69: Ref.\,\cite{Ketterle} and by the numerical studies of Refs.
70: \cite{Pu,Mottonen,Ohmi}, we will focus on the specific problem
71: of the stability (static and dynamic) of a doubly-quantized
72: vortex state. Our study is, however, more general, and our
73: results can be applied to the study of the stability of any
74: problem described by the non-linear Gross-Pitaevskii equation.
75: One remarkable observation obtained from numerical simulations
76: of this problem \cite{Pu,Mottonen,Ohmi} is that the system
77: alternates between regions of dynamical stability and
78: instability as the strength of the interatomic interaction
79: increases. Since some (and probably all) regions of dynamical
80: stability coincide with regions of static instability, it is
81: useful to seek a simple description of this surprising
82: phenomenon.
83:
84: In the following we first examine the questions of static and
85: dynamic stability separately. We will then demonstrate the
86: connection between them.
87:
88: \section{Model}
89:
90: We consider $N$ atoms subject to a spherically-symmetric
91: single-particle Hamiltonian, $h_i$, interacting through a
92: short-range effective interaction,
93: \begin{equation}
94: H = \sum_{i=1}^N h_i + U_0 \sum_{i \neq j = 1}^N
95: \delta({\bf r}_{i} - {\bf r}_{j})/2.
96: \label{intlab}
97: \end{equation}
98: Here $U_0 = 4 \pi \hbar^2 a/M$ is the strength of the effective
99: two-body interaction, $a$ is the scattering length for elastic
100: atom-atom collisions, and $M$ is the atomic mass.
101:
102: For simplicity we assume strong confinement along the $z$ axis,
103: which is also chosen as the axis of rotation. This assumption
104: implies that the cloud is in its lowest state of motion in the
105: $z$ direction, and the problem thus becomes two-dimensional
106: with higher degrees of freedom along the $z$ axis frozen out.
107: We also assume that $h_i$ is rotationally symmetric about the
108: $z$ axis (e.g., a harmonic oscillator Hamiltonian) with the
109: result that angular momentum is conserved.
110:
111: \section{Energetic stability}
112:
113: To be concrete, we will consider the problem of a
114: doubly-quantized vortex state. We start by examining the
115: static stability of this state within the subspace of states of
116: the lowest Landau level. We will generalize our arguments
117: below. The corresponding nodeless eigenfunctions of the
118: (two-dimensional) single-particle Hamiltonian, $h_i$, will be
119: denoted as $\phi_m$, with $m$ the angular momentum; their
120: energy eigenvalues are $\epsilon_m$.
121:
122: We now consider the state
123: \begin{eqnarray}
124: \phi = c_0 \phi_0 + c_2 \phi_2 + c_4 \phi_4,
125: \label{defwf}
126: \end{eqnarray}
127: which describes a doubly-quantized vortex state when $c_2=1$
128: and $c_0 = c_4 =0$. Since we wish to consider the effect of
129: small admixtures of states with $m=0$ and $m=4$, we expand the
130: energy to second order in $c_0$ and $c_4$ to obtain the
131: expectation value of the energy per particle
132: \begin{eqnarray}
133: E = E_2 + c_0^* c_0 M_{00} + c_4^* c_4 M_{44} +
134: M_{04} (c_0^* c_4^* + c_0 c_4),
135: \label{enstab}
136: \end{eqnarray}
137: where
138: \begin{eqnarray}
139: E_2 &=& \epsilon_2 + \frac 1 2 g I_{2222}
140: \nonumber \\
141: M_{00} &=& \epsilon_0 - \epsilon_2 + g (2 I_{0202} - I_{2222})
142: \nonumber \\
143: M_{44} &=& \epsilon_4 - \epsilon_2 + g (2 I_{2424} - I_{2222})
144: \nonumber \\
145: M_{04} &=& g I_{0422} = g I_{2204},
146: \label{def}
147: \end{eqnarray}
148: with $I_{nmlk} \equiv \int \phi_n^* \phi_m^* \phi_l \phi_k \, d
149: {\bf r}$ and $g=N U_0$. The elements, $M_{mm'}$, are real due
150: to the Hermiticity of $H$.
151:
152: The energy $E$ is then equal to
153: \begin{equation}
154: E = E_2 + |c_0|^2 M_{00} + |c_4|^2 M_{44} + 2 |c_0| |c_4| M_{04}
155: \cos(\theta_0 + \theta_4),
156: \label{enstaab}
157: \end{equation}
158: where $c_m = |c_m| e^{i \theta_m}$. The absence of terms
159: linear in $c_0$ and $c_4$ is a consequence of the spherical
160: symmetry of $H$ and implies that $c_2 = 1$ is a solution to the
161: variational problem in this restricted space. For fixed
162: $|c_0|$ and $|c_4|$, it is elementary that $E$ has extrema for
163: $\cos(\theta_0 + \theta_4) = \pm 1$. In general, the linearly
164: independent curvatures of this quadratic energy surface are
165: obtained as the eigenvalues of the real symmetric matrix
166: \begin{eqnarray}
167: M_s = \left( \begin{array}{cc}
168: M_{00} & M_{04}\\
169: M_{40} & M_{44}
170: \end{array} \right)
171: \end{eqnarray}
172: with $M_{40}=M_{04}^{\dag}$.
173:
174: In this problem, it is evidently of physical interest to
175: consider the stability of the doubly quantized vortex subject
176: to the constraint that angular momentum is conserved. This
177: constraint will be satisfied if $|c_0| = |c_4| = |c|$, and the
178: energy $E$ then becomes
179: \begin{eqnarray}
180: E = E_2 + 2 g |c|^2 (I_{0202} + I_{2424} - I_{2222} \pm I_{0422}).
181: \label{enstabb}
182: \end{eqnarray}
183: The doubly-quantized vortex will be an energy minimum if the
184: coefficient of $|c|^2$ is positive. In other words, the
185: doubly-quantized vortex will be energetically stable if
186: \begin{equation}
187: I_{0202} + I_{2424} - I_{2222} \ge I_{0422}.
188: \label{cond}
189: \end{equation}
190: This condition is not satisfied when only lowest Landau levels
191: are retained \cite{KMP}, and a doubly-quantized vortex state is
192: energetically unstable within this approximation.
193:
194: Several comments are in order before turning to the dynamical
195: behavior of the same (truncated) problem. We have tested the
196: stability of the state with respect to the admixture of states
197: $\phi_0$ and $\phi_4$. A complete test of stability requires
198: the investigation of the admixture of $\phi_m$ for all $m$.
199: Fortunately, the spherical symmetry of $H$ ensures that the
200: generalized quadratic form of Eq.\,(\ref{enstab}) will only mix
201: states with $m=2 \pm k$. With the inclusion of additional
202: values of $k$, the matrix $M_s$ becomes block diagonal and
203: leads to a sequence of equations for each $k$ completely
204: analogous to those considered above. (The $k=2$ choice
205: considered here is known to be the most unstable case
206: \cite{KMP}.) The problem of finding the eigenvalues of the
207: Hermitean matrix, $M_s$, is subject to familiar variational
208: arguments. Expansion of the dimension of this matrix, either
209: by including more values of $k$ or by relaxing the restriction
210: to the lowest Landau level, can only decrease the smallest
211: eigenvalue of $M_s$. Thus, the fact that a given state is
212: energetically unstable for a given choice of the finite space
213: used to construct $M_s$ is conclusive proof of instability.
214: Finally, we note that changing the state under investigation,
215: e.g., through the inclusion of higher Landau levels, leads to
216: fundamental changes in $M_s$. Under such circumstances, simple
217: variational arguments cannot tell us whether the improved state
218: will be more or less stable.
219:
220: \section{Dynamic stability}
221:
222: As mentioned above, dynamic stability probes the temporal
223: behavior of the system. We again start with the wave function
224: of Eq.\,(\ref{defwf}) and construct the Lagrangian to second
225: order in the small parameters $c_0$ and $c_4$. Variations with
226: respect to the coefficients $c_0^*$ and $c_4$ then lead to the
227: equations
228: \begin{eqnarray}
229: i {\dot c}_0 &=& \epsilon_0 c_0 + 2 g c_0 c_2^* c_2 I_{0202} +
230: g c_4^* c_2 c_2 I_{0422}
231: \nonumber \\
232: -i {\dot c}_4^* &=& \epsilon_4 c_4^* + 2 g c_4^* c_2^* c_2 I_{2424} +
233: g c_0 c_2^* c_2^* I_{0422}
234: \nonumber \\
235: i {\dot c}_2 &=& \epsilon_2 c_2 + g c_2 c_2^* c_2 I_{2222},
236: \label{tm}
237: \end{eqnarray}
238: where the dot denotes a time derivative. We can solve these
239: equations with the ansatz $c_0(t)=c_0(0) {\rm e}^{-i(\mu +
240: \omega) t}$ and $c_4^*(t)=c_4^*(0) {\rm e}^{i(\mu - \omega)
241: t}$. (The time dependence of the solution, $\phi_2$, is given
242: by the factor ${\rm e}^{-i\mu t}$ with $\mu = \epsilon_2 + g
243: I_{2222}$.) The resulting Bogoliubov equations can be written
244: in the form
245: \begin{eqnarray}
246: \left( \begin{array}{cc} \epsilon_0 - \mu + 2 g I_{0202} &
247: gI_{0422}
248: \\
249: -g I_{0422} & - \epsilon_4 + \mu - 2 g I_{2424}
250: \end{array} \right)
251: \left( \begin{array}{c}
252: c_0 \\
253: c_4^*
254: \end{array} \right) =
255: \nonumber \\
256: \omega_d \left( \begin{array}{c}
257: c_0 \\
258: c_4^*
259: \end{array} \right).
260: \label{bog}
261: \end{eqnarray}
262: This equation can also be written as
263: \begin{equation}
264: M_s \, \left( \begin{array}{c}
265: c_0 \\
266: c_4^*
267: \end{array} \right)= \omega_d \, \sigma
268: \left( \begin{array}{c}
269: c_0 \\
270: c_4^*
271: \end{array} \right)\ \ {\rm with}\ \ \sigma \equiv
272: \left( \begin{array}{cr}
273: 1 & 0\\
274: 0 & -1\\ \end{array} \right),
275: \label{bog2}
276: \end{equation}
277: where $M_s$ is the matrix appearing in the static stability
278: problem. The dynamic problem is thus governed by the
279: eigenvalues of the non-Hermitean Eq.\,(\ref{bog}), and these
280: eigenvalues can be complex. Since $M_s$ is real Hermitean, the
281: dynamic eigenvalues are either real or come in conjugate pairs.
282: For each eigenfunction, $\psi_d$, with a complex eigenvalue,
283: $\omega_d$, $\psi_d^*$ will also be an eigenfunction with
284: eigenvalue $\omega_d^*$. As a result, the roots move along the
285: real axis, touch (i.e., become degenerate) and then move off in
286: the complex plane. Evidently, the existence of complex
287: eigenvalues indicates exponential divergence and dynamic
288: instability.
289:
290: From Eqs.\,(\ref{bog}) or (\ref{bog2}) we see that the
291: Bogoliubov eigenvalues are real and the state under
292: investigation is dynamically stable provided that $I_{0202} +
293: I_{2424} - I_{2222} \ge I_{0422}$. This condition is identical
294: to the condition for static stability found in
295: Eq.\,(\ref{cond}). As we show below in greater generality,
296: static stability always implies dynamic stability, but dynamic
297: stability can occur even in the presence of static instability.
298:
299: \section{General results}
300:
301: At this point it is useful to generalize our static and dynamic
302: formalisms to allow for the inclusion of an arbitrary number of
303: Landau levels in the description of both the doubly quantized
304: vortex and the potentially unstable states with $m=0$ and
305: $m=4$. The matrix appropriate for the static problem now has
306: the Hermitean block form
307: \begin{eqnarray}
308: M_s = \left( \begin{array}{cc}
309: M_{00} & M_{04} \\
310: M_{40} & M_{44}
311: \end{array} \right),
312: \end{eqnarray}
313: with $M_s^{\dag} = M_s^T = M_s$. Aside from its dimension, the
314: only change in the construction of $M_s$ lies in the
315: replacement of the lowest Landau state, $\phi_2$, by a
316: superposition of Landau states, $\psi_2$, which satisfies the
317: obvious Euler equation
318: \begin{equation}
319: h_0 \psi_2 + g | \psi_2 |^2 \psi_2 = \mu \psi_2.
320: \label{euler}
321: \end{equation}
322: Here, $\mu$ is a Lagrangian multiplier introduced to ensure
323: that $\psi_2$ is normalized to unity. It is understood that
324: the replacement $\phi_2 \to \psi_2$ is to be made as
325: appropriate in the integrals, $I_{nmlk}$, contributing to
326: $M_s$. As before, positive eigenvalues of $M_s$ imply
327: energetic stability.
328:
329: The dynamic Bogoliubov equations again assume the form of
330: Eq.\,(\ref{bog2}) and can be written as
331: \begin{eqnarray}
332: M_s \left( \begin{array}{c}
333: c_0 \\
334: c_4^*
335: \end{array} \right) =
336: \omega_d \, \sigma \left( \begin{array}{c}
337: c_0 \\
338: c_4^*
339: \end{array} \right),
340: \label{bog3}
341: \end{eqnarray}
342: where $M_s$ is again the real Hermitean matrix governing
343: energetic stability. The quantities $c_0$ and $c_4$ now
344: represent column vectors; their dimensions need not be equal.
345: The reality of all eigenvalues, $\omega_d$, is again an
346: indication of dynamical stability.
347:
348: As we have seen, $H$ is Hermitean and spherically symmetric.
349: The time evolution of an arbitrary wave function, $\psi$, is
350: governed by the full time-dependent Gross-Pitaevskii equation
351: \begin{equation}
352: h_0 \psi + g | \psi |^2 \psi = i \hbar \frac{\partial
353: \psi}{\partial t}.
354: \end{equation}
355: It thus comes as no surprise that energy and angular momentum
356: are, in general, constants of the motion. On the other hand,
357: given that the (linearized) Bogoliubov equations do not even
358: conserve probability, it is not obvious that energy and
359: momentum are conserved when these equations are used to
360: describe the temporal evolution of the system. This point
361: merits some attention. Consider a dynamic eigenvector of the
362: Bogoliubov equations, $\psi_d$, and its (possibly complex)
363: eigenvalue, $\omega_d$. Given the Hermiticity of $M_s$,
364: standard arguments reveal that
365: \begin{equation}
366: (\omega_d - \omega_d^* ) \, \langle \psi_d | \sigma | \psi_d
367: \rangle = 0.
368: \label{conserve}
369: \end{equation}
370: This equation is evidently trivial when $\omega_d$ is real, but
371: it shows that $\langle \psi_d | \sigma | \psi_d \rangle = 0$
372: when $\omega_d$ is complex. This tells us that the
373: probabilities of finding $m=0$ is equal to that for $m=4$ for
374: all times. In this sense, angular momentum is rigorously
375: conserved in spite of the approximations leading to the
376: Bogoliubov equations.
377:
378: The extension of this familiar conservation law presumes that
379: the trial state $\psi_2$ is deformed by the inclusion (with
380: arbitrary amplitude) of a single Bogoliubov eigenvector with
381: complex eigenvalue. Angular momentum is not in general
382: conserved with arbitrary deformations of $\psi_2$ involving
383: either single Bogoliubov eigenvectors with real eigenvalues or
384: superpositions of Bogoliubov eigenvectors. Since the primary
385: rationale for studying the Bogoliubov equations at all is to
386: determine the existence or non-existence of complex
387: eigenvalues, this point is of interest in spite of such
388: caveats. Further, it suggests that we are most likely to reveal
389: relations between the problems of static and dynamic stability
390: if we consider the question of static stability subject to the
391: constraint of constant angular momentum. Indeed, this
392: constraint was imposed trivially in Eq.\,(\ref{enstabb}) above,
393: where it was crucial in establishing the identity of static and
394: dynamic stability criteria in the simple case of single Landau
395: levels.
396:
397: For the more general problem considered here, it is easiest to
398: proceed by introducing real Lagrange multipliers, $\lambda$ and
399: $\eta$, in order to impose the constraints of overall
400: normalization and constant angular momentum, respectively, in
401: the static problem. The constrained static eigenvalue problem
402: then reads
403: \begin{eqnarray}
404: M_s {\bar \psi}_s =
405: \lambda \, \left( \begin{array}{cc}
406: 1 & 0\\
407: 0 & \eta \end{array} \right) {\bar \psi}_s .
408: \label{omegas}
409: \end{eqnarray}
410: In practice, the real parameter $\eta$ is adjusted so that the
411: resulting static eigenfunctions ${\bar \psi}_s$ satisfy
412: $\langle {\bar \psi}_s | \sigma | {\bar \psi}_s \rangle = 0$.
413: The desired extrema of $\langle M_s \rangle$ then follow as
414: ${\bar \omega}_s \equiv \langle {\bar \psi}_s | M_s | {\bar
415: \psi}_s \rangle$. The similarity between the constrained
416: static problem of Eq.\,(\ref{omegas}) and the dynamic
417: eigenvalue problem of Eq.\,(\ref{bog3}) is now obvious. We
418: shall exploit this connection below.
419:
420: \section{Connection between static and dynamic stability}
421:
422: Given the fact that $\langle \psi_d | \sigma | \psi_d \rangle =
423: 0$, we see that the inner product of $\langle \psi_d |$ with
424: Eq.\,(\ref{bog3}) also yields $\langle \psi_d | M_s | \psi_d
425: \rangle = 0$. If $\psi_2$ is deformed by the inclusion (with
426: arbitrary amplitude) of a single Bogoliubov eigenvector with
427: complex eigenvalue, the energy is rigorously conserved and
428: identical to that obtained for the pure state $\psi_2$. We can
429: use any $\psi_d$ with complex eigenvalue as a trial function to
430: provide a variational upper bound of zero for the lowest
431: constrained static eigenvalue, ${\bar \omega}_s$. Said
432: somewhat more simply, it is elementary that the smallest ${\bar
433: \omega}_s$ cannot be positive if there exists a state $\psi_d$
434: such that $\langle \psi_d | M_s | \psi_d \rangle = 0$.
435: (Henceforth, we will consider only the constrained static
436: problem and will refer to it simply as the ``static problem''.)
437: Thus, if the Bogoliubov problem has complex eigenvalues, the
438: lowest eigenvalue of the static problem must be negative (or
439: zero). Conversely, if all ${\bar \omega}_s > 0$, none of the
440: Bogoliubov eigenvalues can be complex. In other words, {\it
441: dynamic instability guarantees static instability, and static
442: stability guarantees dynamic stability}. This establishes the
443: first rigorous connection between the two stability problems.
444: While this result may appear obvious, a similar argument shows
445: that it is equally impossible to find a state $\psi_d$ such
446: that $\langle \psi_d | M_s | \psi_d \rangle = 0$ if all
447: $\bar{\omega}_s < 0$. This yields the more surprising result
448: that {\it complete static instability also leads to dynamic
449: stability}.
450:
451: In order to find additional ties between the static and dynamic
452: stability problems, it is useful to follow the trajectory of
453: two Bogoliubov eigenvalues (as a function of the coupling
454: constant) as they evolve from distinct real values to a
455: conjugate pair. To see this, it is sufficient to consider the
456: two-dimensional space of $(\psi_0, 0)$ and $(0, \psi_4)$.
457: (Here, $\psi_0$ and $\psi_4$ are of arbitrary dimension and
458: normalized.) We thus write
459: \begin{eqnarray}
460: M_s \approx \left( \begin{array}{cc}
461: E_0 & g V \\
462: g V & E_4
463: \end{array} \right),
464: \end{eqnarray}
465: with all quantities real. The Bogoliubov problem of
466: Eq.\,(\ref{bog3}) is readily solved with the following results.
467: For $|g| < g_c = |(E_0 + E_4)/2V|$, the $\omega_d$ are real and
468: distinct. The eigenvectors are real and do not satisfy the
469: constraint of constant angular momentum. For $|g| = g_c$, the
470: eigenvalues are degenerate with $\omega_d = (E_0 - E_4)/2$. The
471: eigenvectors are identical and given as $( \psi_0 , - \psi_4
472: )/\sqrt{2}$. (Recall that the problem is non-Hermitean.) For $g
473: > g_c$, the eigenvalues form a conjugate pair. The
474: eigenvectors also form a conjugate pair with a non-trivial
475: phase, and the magnitudes of their $\psi_0$ and $\psi_4$
476: components are equal (i.e., they conserve angular momentum).
477:
478: We can also solve the static problem using Eq.\,(\ref{omegas})
479: in the same truncated basis. The values of $\eta$ required to
480: impose the constraint of conserved angular momentum are found
481: to be $\eta = (E_4 \pm gV)/(E_0 + gV)$. The corresponding
482: static eigenvalues are ${\bar \omega}_s = (E_0 + E_4)/2 \pm gV$
483: with ${\bar \psi}_s = (\psi_0 , \pm \psi_4) /\sqrt{2}$, as
484: expected. For $|g| < g_c$, the static eigenvalues are positive.
485: For $|g| > g_c$, one of the static eigenvalues is negative. For
486: $|g| = g_c$, one static eigenvalue is precisely zero with
487: ${\bar \psi}_s = \psi_d = (\psi_0 , - \psi_4)/\sqrt{2}$. The
488: condition for a static eigenvalue of zero is the same as the
489: condition for degenerate dynamic eigenvalues.
490:
491: We also see that the static and dynamic eigenvectors are
492: identical at this point. This should come as no surprise. At
493: $|g| = g_c$, the Lagrange multiplier, $\eta$, is $-1$. Equation
494: (\ref{omegas}), which determines ${\bar \psi}_s$, becomes
495: identical to the Bogoliubov equation, Eq.\,(\ref{bog3}), and
496: identical solutions must result. This result is general. While
497: the form of these two equations is similar, the fact that $M_s$
498: is a real symmetric matrix forces ${\bar \psi}_s$ and ${\bar
499: \omega}_s$ to be real (up to a trivial phase). When $\omega_d
500: \ne \omega_d^*$, it is clear that the associated dynamic
501: eigenvectors, $\psi_d$ and $\psi_d^*$ cannot be chosen equal.
502: Thus, the phase of $\psi_d$ is non-trivial, and $\psi_d$ cannot
503: be the desired solution to Eq.\,(\ref{omegas}). (The case of
504: $\omega_d$ real can be dismissed summarily, since the
505: amplitudes of $m=0$ and $4$ states are not equal.) When two
506: Bogoliubov roots are degenerate, however, $\psi_d$ satisfies
507: both the constraints of angular momentum conservation and
508: reality. Under such conditions, ${\bar \psi}_s = \psi_d$, and
509: ${\bar \omega}_s = 0$. In other words, the number of conjugate
510: pairs of complex Bogoliubov eigenvalues changes by one when a
511: static eigenvalue passes through zero.
512:
513: It might be thought that the number of conjugate pairs of
514: dynamic eigenvalues was equal to the number of negative ${\bar
515: \omega}_s$ and that a stronger statement was thus possible. As
516: we will show below, this is not the case. The present
517: statement can nevertheless give us a useful corollary. Start
518: from a manifestly stable choice of $\psi_2$ for which all
519: ${\bar \omega}_s > 0$ and dynamical stability is ensured. Vary
520: some external parameter (e.g., the coupling constant, $g$). As
521: demonstrated above, a single conjugate pair of Bogoliubov
522: eigenvalues will appear when the first ${\bar \omega}_s$ passes
523: through zero, and the system becomes dynamically unstable. With
524: every subsequent passage of an ${\bar \omega}_s$ through zero,
525: the number of conjugate pairs changes (i.e., increases or
526: decreases) by one. It is then clear that an odd number of
527: ${\bar \omega}_s$ must correspond to an odd number of conjugate
528: Bogoliubov pairs and, hence, dynamic instability.
529:
530: \section{Dynamic stability in the absence of static stability}
531:
532: As mentioned earlier, it is possible for the system to be
533: dynamically stable even in the presence of static instability.
534: This effect is more subtle but is also useful in tightening the
535: connection between the problems of static and dynamic
536: stability. A simple calculation will be helpful. Start with
537: two solutions to the constrained static problem of
538: Eq.\,(\ref{omegas}), $| {\bar 1} \rangle$ and $| {\bar 2}
539: \rangle$. The corresponding static eigenvalues are ${\bar
540: \omega}_{s1}$, which is assumed to be negative, and ${\bar
541: \omega}_{s2}$, which can pass through zero. Approximate the
542: Bogoliubov equations by truncating the basis to include these
543: two states and the related states $| \bar{3} \rangle = \sigma |
544: \bar{1} \rangle$ and $| \bar{4} \rangle = \sigma | \bar{2}
545: \rangle$. (This calculation is exact when the dimension of
546: $M_s$ is equal to four.) Since these states are not necessarily
547: orthogonal, the Bogoliubov equations assume the form
548: \begin{equation}
549: M_s \psi_d = \omega_d \Sigma \psi_d,
550: \label{nonorthog}
551: \end{equation}
552: with $(M_s)_{ij} = \langle \bar{i} | M_s | \bar{j} \rangle$ and
553: $(\Sigma)_{ij} = \langle \bar{i} | \sigma | \bar{j} \rangle$.
554: Both matrices are real symmetric. The first two diagonal
555: elements of $M_s$ are the constrained static eigenvalues,
556: $\bar{\omega}_{s1}$ and $\bar{\omega}_{s2}$. The remaining two
557: diagonal elements, $\bar{\omega}_{s3}$ and $\bar{\omega}_{s4}$,
558: will be assumed to be positive. The matrix, $\Sigma$, is
559: intimately related to the overlap matrix with elements $\langle
560: \bar{i} | \bar{j} \rangle$. Since the basis states conserve
561: angular momentum, the diagonal elements of $\Sigma$ vanish.
562: Given the origin of the $| \bar{1} \rangle$ and $| \bar{2}
563: \rangle$ as constrained extrema of $\langle M_s \rangle$, it is
564: clear that the elements of $M_s$ and $\Sigma$ are not
565: independent. We shall ignore this fact in the following
566: qualitative arguments.
567:
568: First, consider the case when these four states are maximally
569: linearly dependent. State $| \bar{1} \rangle$ has been assumed
570: to be distinct from $| \bar{2} \rangle$ and is explicitly
571: orthogonal to $| \bar{3} \rangle$. Let us assume it is
572: identical to $| \bar{4} \rangle$. This immediately implies
573: that $| \bar{3} \rangle$ is identical to $| \bar{2} \rangle$.
574: The approximate Bogoliubov equation is thus reduced to a $2
575: \times 2$ matrix equation in the space of $| \bar{1} \rangle$
576: and $| \bar{2} \rangle$ and assumes the form
577: \begin{eqnarray}
578: \frac{1}{\langle {\bar 1} | \sigma | {\bar 2} \rangle} \left(
579: \begin{array}{cc}
580: \langle {\bar 1} | M_s | {\bar 2} \rangle & {\bar \omega}_{s2}
581: \\
582: {\bar \omega}_{s1} & \langle {\bar 1} | M_s | {\bar 2} \rangle
583: \end{array}
584: \right) \, \psi_d = \omega_d \, \psi_d.
585: \nonumber \\
586: \label{invff}
587: \end{eqnarray}
588: The eigenvalues of this problem are
589: \begin{eqnarray}
590: \omega_d = \frac 1 {\langle {\bar 1} | \sigma | {\bar 2}
591: \rangle}
592: \left( \langle {\bar 1} | M_s | {\bar 2}
593: \rangle \pm \sqrt{{\bar \omega}_{s1}
594: \, {\bar \omega}_{s2}} \right),
595: \end{eqnarray}
596: and the corresponding (unnormalized) eigenvectors are
597: \begin{equation}
598: \psi_d \sim \sqrt{{\bar \omega}_{s2}} \, \, | {\bar 1} \rangle
599: + \sqrt{{\bar \omega}_{s1}} \, \, | {\bar 2} \rangle.
600: \label{evecs}
601: \end{equation}
602: When ${\bar \omega}_{s2} > 0$, the Bogoliubov eigenvalues form
603: a conjugate pair. Further, Eq.\,(\ref{evecs}) shows that
604: $\psi_d$ acquires a non-trivial phase and that $\langle \psi_d
605: | \sigma | \psi_d \rangle \sim {\rm Re}{[({\bar \omega}_{s1}^*
606: {\bar \omega}_{s2})^{1/2}]}$ is zero. The system is
607: dynamically unstable. When ${\bar \omega}_{s2} = 0$, the
608: Bogoliubov eigenvalues are degenerate and real. In this case,
609: $\psi_d$ is equal to $| {\bar 2} \rangle$, which is real and
610: conserves angular momentum. These results are all consistent
611: with those found above. When ${\bar \omega}_{s2} < 0$,
612: however, the Bogoliubov eigenvalues are real and distinct. The
613: corresponding $\psi_d$ can now be chosen real, and angular
614: momentum is no longer conserved. The presence of two negative
615: static minima of $\langle M_s \rangle$ is thus capable of
616: creating dynamic stability in spite of manifest static
617: instability. This argument is actually more detailed than
618: necessary. We have already seen that the number of conjugate
619: pairs of Bogoliubov eigenvalues necessarily changes by one
620: every time a static eigenvalue passes through zero. Linear
621: dependence has reduced the present problem to one spanned by a
622: two-dimensional space. The fact that ${\bar \omega}_{s1} < 0$
623: ensures that there are initially two complex eigenvalues in the
624: space. We have seen in general that the number of complex
625: eigenvalues must change by two when ${\bar \omega}_{s2}$
626: crosses zero. Since all eigenvalues of this $2 \times 2$
627: problem are already complex, the only possibility is thus that
628: the number of complex eigenvalues is reduced to zero with
629: dynamic stability as a consequence, as we have seen. States $|
630: \bar{1} \rangle$ and $| \bar{2} \rangle$ are both essential to
631: this process.
632:
633: Now consider the case where the $| \bar{i} \rangle$ are all
634: mutually orthogonal and maximally linearly independent.
635: Partition the matrix into $2 \times 2$ blocks spanned by the
636: states $| \bar{1} \rangle$ and $| \bar{3} \rangle$ and the
637: states $| \bar{2} \rangle$ and $| \bar{4} \rangle$,
638: respectively. Each of the diagonal sub-blocks of $M_d$ now has
639: a form familiar from Eq.\,(\ref{invff}). The diagonal elements
640: are equal in each case. The products of off-diagonal elements
641: are $\bar{\omega}_{s1} \bar{\omega}_{s3}$ and
642: $\bar{\omega}_{s2} \bar{\omega}_{s4}$, respectively. If the
643: off-diagonal blocks of this matrix are sufficiently small, the
644: Bogoliubov eigenvalues will be given by the eigenvalues of
645: these two $2 \times 2$ matrices. Since $\bar{\omega}_{s1}
646: \bar{\omega}_{s3} < 0$ by assumption, this block produces the
647: original conjugate pair of dynamical eigenvalues independent of
648: the properties of $| \bar{2} \rangle$. As $\bar{\omega}_{s2}$
649: passes through zero, $\bar{\omega}_{s2} \bar{\omega}_{s4}$
650: becomes negative, and an additional conjugate pair of
651: eigenvalues appears independent of the properties of $| \bar{1}
652: \rangle$. This behaviour will persist whenever the off-diagonal
653: blocks of $M_d$ are sufficiently small.
654:
655: We thus see that there are two possible outcomes when a second
656: static extremum becomes negative. If the corresponding states
657: are ``strongly'' coupled with $|\langle \bar{1} | \sigma |
658: \bar{2} \rangle| = | \langle \bar{1} | \bar{4} \rangle |
659: \approx 1$, the initial dynamic instability will be eliminated.
660: If these states are ``weakly'' coupled with $|\langle \bar{1} |
661: \sigma | \bar{2} \rangle| \approx 0$, a second unstable dynamic
662: mode will appear. There is another and potentially fruitful
663: way to distinguish these alternatives. Consider the evolution
664: of the closed surfaces with $\langle M_s \rangle = 0$ which
665: bound domains within which $\langle M_s \rangle < 0$. [It is
666: assumed that $\langle M_s \rangle$ is explored with real trial
667: vectors subject to the constraint of constant angular momentum.
668: Hence, these vectors lie on a (hyper)torus.] Every negative
669: extremum of $\langle M_s \rangle$ is evidently surrounded by
670: such a surface, and every such surface must contain at least
671: one negative extremum. If there is one negative extremum,
672: there is one $\langle M_s \rangle = 0$ surface. There are two
673: possible ways for the topology of these surfaces to change when
674: a second static extremum becomes negative. A second $\langle
675: M_s \rangle = 0$ surface, not connected to the first, can
676: appear. In this case, the new negative extremum must be a local
677: minimum. Alternatively, the original surface can spread,
678: exploiting the fact that the manifold is periodic, and touch
679: itself. The point of contact necessarily represents a new
680: extremum of $\langle M_s \rangle$, and a few moments thought
681: reveals that this extremum is necessarily a saddle point. (When
682: there are more than two negative static modes, independent
683: $\langle M_s \rangle = 0$ surfaces can merge in a similar
684: manner.) Randomly drawn numerical examples suggest that the
685: appearance of a new $\langle M_s \rangle = 0$ surface
686: corresponds to the case of weak coupling discussed above, and a
687: new unstable dynamic mode emerges. The merger of an $\langle
688: M_s \rangle = 0$ surface with itself corresponds to the case of
689: strong coupling, and the original unstable dynamic mode
690: disappears. Although the evidence is admittedly slim, it is
691: tempting to guess more generally that each closed surface with
692: $\langle M_s \rangle = 0$ produces zero unstable dynamic modes
693: when it contains an even number of negative extrema of $\langle
694: M_s \rangle$ and one unstable dynamic mode when it contains an
695: odd number of such extrema.
696:
697: As we have shown, the reality and Hermiticity of $M_s$ and the
698: non-Hermiticity introduced by the specific form of $\sigma$ are
699: not sufficient to provide a unique answer to the question of
700: dynamic stability in the presence of an even number of unstable
701: static modes. (As noted above, an odd number of negative
702: static modes always implies dynamic instability.) The question
703: of what constitutes ``strong'' or ``weak'' coupling would thus
704: seem to require more details regarding the physical system in
705: question. For this reason, we now turn to the question of the
706: stability of vortex solutions to the physically relevant
707: Gross-Pitaevskii equation.
708:
709: \section{Numerical results}
710:
711: We now study the static and dynamic stability of a doubly
712: quantized vortex numerically. The single-particle Hamiltonian,
713: $h_0=-\hbar^2 \nabla^2 / (2 M) +M\omega^2 r^2/2$, represents a
714: harmonic oscillator potential of strength $\omega$. In an
715: anharmonic potential, a doubly quantized vortex is
716: energetically stable for weak couplings, but in a purely
717: harmonic potential this is known not to be the case
718: \cite{lundh,jkl}.
719:
720: In our simulations Eq.\,(\ref{euler}) was first solved to find
721: the doubly quantized vortex. This is equivalent to finding a
722: stationary solution to the Gross-Pitaevskii equation subject to
723: the constraint that the system has a 4$\pi$ phase singularity
724: at the origin. This minimization was carried out within a
725: truncated basis composed of the $N_c$ lowest radially excited
726: states and is valid for small values of $g$ such that $g /(2
727: \hbar \omega) \lesssim N_c$. The static and dynamic eigenvalues
728: were then calculated according to Eqs.\,(\ref{bog3}) and
729: (\ref{omegas}) in a basis consisting of $N_x$ radially excited
730: states for both the $m=0$ and $m=4$ components. The computation
731: of the eigenvalues was carried out in Matlab. The results of
732: this calculation are shown in Fig.\,\ref{fig:frequencies}.
733: %
734: \begin{figure}
735: %\includegraphics[width=\columnwidth]{frequencies.eps}
736: \includegraphics[width=9cm,height=8cm]{frequencies.eps}
737: \caption[]{Eigenvalues of the static and dynamic stability
738: matrices as functions of the coupling constant $g$, computed by
739: means of exact diagonalization in a truncated basis. Solid
740: lines represent the imaginary part of the eigenvalues of the
741: dynamic stability matrix, ${\rm Im} (\omega_d)$. Dots represent
742: the lowest eigenvalues of the static stability matrix,
743: $\bar\omega_s$. \label{fig:frequencies}}
744: \end{figure}
745: %
746: Results are shown for $N_c=N_x=8$, and we have checked
747: convergence with respect to both $N_c$ and $N_x$ for the range
748: of couplings displayed in the figure. The dynamic eigenvalues
749: $\omega_d$ coincide with those previously found numerically in
750: Refs.\,\cite{Pu,Mottonen,Ohmi}.
751:
752: The figure clearly shows the correlation between the ``window''
753: structure of the complex frequencies and the eigenvalues,
754: $\bar\omega_s$, of the static problem in agreement with our
755: general arguments above. For small $g$ such that $g\ll
756: \hbar\omega$, the properties of the system follow from the
757: perturbative analysis of the lowest Landau level as described
758: above: The system possesses one negative static eigenvalue and
759: therefore one pair of complex dynamic eigenvalues. As the
760: coupling increases, a second static eigenvalue crosses zero at
761: $g/\hbar\omega\approx 6$, and the system becomes dynamically
762: stable despite its manifest static instability. As a third
763: static eigenvalue crosses zero, one pair of dynamic eigenvalues
764: again become complex but become real again when a fourth static
765: eigenvalue becomes negative. The numerical findings indicate
766: that as $g$ is further increased, a succession of such
767: alternations occurs with the result that the system is found to
768: be dynamically stable (unstable) when there are an even (odd)
769: number of negative static eigenvalues.
770:
771: For the range of couplings, $g$, studied here, there is never
772: more than one complex pair of dynamic eigenvalues. It is also
773: seen that the system is always statically unstable. In the
774: limit of large $g$, well-known results for vortices in an
775: infinite system apply \cite{textbook}. In that limit the
776: energy of two vortices increases monotonically with decreasing
777: separation, thereby implying that at least one eigenvalue of
778: the static stability matrix is negative. Hence it is
779: reasonable to conclude that the system is statically unstable
780: for all values of $g$. We thus conclude that the ``window''
781: structure of alternating dynamically stable and unstable
782: regions arises from the non-trivial interplay between negative
783: modes of the static stability matrix.
784:
785: \section{Conclusions}
786:
787: The purpose of this paper has been to point out the intimate
788: connections between the static and dynamic stability of
789: solutions to the Gross-Pitaevskii equations. We have shown
790: that (suitably constrained) static stability necessarily
791: implies dynamic stability, that the number of complex conjugate
792: pairs of dynamic eigenvalues changes by one every time a
793: constrained static eigenvalue passes through zero, and that an
794: odd number of negative static eigenvalues thus implies dynamic
795: instability. Numerical investigations revealed that the
796: doubly-quantized vortex solution to the Gross-Pitaevskii
797: equation is statically unstable over the full range of coupling
798: constants explored but displays windows of dynamic stability.
799: The general nature of the arguments presented here suggests
800: that similar connections between the problems of static and
801: dynamic stability are likely to be wide-spread. Further, we
802: believe that additional insight obtained by studying both
803: stability problems is likely to be well worth the minimal
804: additional effort required.
805:
806: \section{Acknowledgments}
807: GMK acknowledges financial support from the European Community
808: project ULTRA-1D (NMP4-CT-2003-505457).
809:
810: \begin{thebibliography}{0}
811:
812: \bibitem{Ketterle} Y.\ Shin, M.\ Saba, M.\ Vengalattore, T.~A.\ Pasquini,
813: C.\ Sanner, A.~E.\ Leanhardt, M.\ Prentiss, D.~E.\ Pritchard,
814: and W.\ Ketterle, Phys.\ Rev.\ Lett.\ {\bf 93}, 160406 (2004).
815:
816: \bibitem{Pu} H.\ Pu, C.~K.\ Law, J.~H.\ Eberly, and N.~P.\
817: Bigelow, Phys.\ Rev.\ A {\bf 59}, 1533 (1999).
818:
819: \bibitem{Mottonen} M.\ M{\"o}tt{\"o}nen, T.\ Mizushima,
820: T.\ Isoshima, M.~M.\ Salomaa, and K.\ Machida, Phys.\ Rev.\ A
821: {\bf 68}, 023611 (2003).
822:
823: \bibitem{Ohmi} Yuki Kawaguchi and Tetsuo Ohmi, Phys. Rev. A
824: {\bf 70}, 043610 (2004).
825:
826: \bibitem{KMP} G.~M.\ Kavoulakis, B.\ Mottelson, and C.~J.\
827: Pethick, Phys. Rev. A {\bf 63}, 063605 (2000).
828:
829: \bibitem{lundh} Emil Lundh, Phys.\ Rev.\ A {\bf 65}, 043604 (2002).
830:
831: \bibitem{jkl} A.~D.\ Jackson, G.~M.\ Kavoulakis, and E.\ Lundh,
832: Phys.\ Rev.\ A {\bf 69}, 053619 (2004).
833:
834: \bibitem{textbook} C.~J.\ Pethick and H.\ Smith, {\it Bose-Einstein
835: Condensation in Dilute Gases} (Cambridge University Press,
836: Cambridge, 2001).
837:
838: \end{thebibliography}
839: \end{document}
840:
841: