1: \section{Zero modes}\label{ZeroModes}
2: \reseteqnos
3:
4: Solitary waves are not stable in the ordinary sense that the
5: solution to the evolution equation with perturbed initial data differs
6: little from the unperturbed solution for all time. This can be seen by
7: considering, for example, two solitary waves with slightly different
8: velocities whose initial data are approximately equal but which later
9: drift apart. Instead, solitary-wave stability is defined in terms of
10: the difference between perturbed and unperturbed solutions modulo
11: symmetries of the evolution equation. For instance, the two solitary
12: waves just discussed do remain close modulo spatial translation.
13:
14: Invariance of an evolution equation under symmetries gives rise to an
15: eigenvalue-zero eigenspace for the generator of time evolution for the
16: equation linearized about a solution. Linear instability due to
17: generalized eigenfunctions in such a space need not correspond to a
18: real instability, but rather to ``drift'' that can be cancelled by a
19: symmetry transformation. On the other hand, an imaginary eigenvalue
20: with a generalized eigenfunction (of index larger than 1) that does
21: not correspond to a symmetry operation would suggest instability.
22:
23: In this section we make use of symmetries to enumerate the zero modes of
24: \eq{1nls}, obtaining generalized eigenfunctions of the operator $A$
25: corresponding to eigenvalue $\lambda=0$. We identify the same
26: number of zero modes that we find numerically.
27:
28: Given a family $u_\tau$ of solutions to \eq{1nls} with smooth dependence on
29: parameter $\tau$, we can differentiate the identity
30: \be
31: -i{\D_t}
32: u_\tau - \Delta u_\tau = g(u_\tau)
33: \ee
34: with respect to $\tau$ to obtain a solution $\tilde u$ to the
35: corresponding equation linearized about $u_0$:
36: \be
37: -i{\D_t} \tilde u - \Delta \tilde u =
38: \alpha(|u_0|)\,\tilde u + \beta(|u_0|)\frac{u_0}{\bar u_0}\bar {\tilde u}
39: \label{zm1}
40: \ee
41: where
42: \be
43: \tilde u\equiv \left.{ \D_\tau u_\tau}\right|_{\tau =0}.
44: \ee
45:
46: We apply this procedure to various one-parameter families of solutions,
47: each having $u_0=e^{i\omega \,t}v_0(x)=e^{i\omega \,t}e^{i\,m\theta }w(r)$.
48: With $\tilde v$ given by $\tilde u = e^{i\omega \,t}\tilde v$,
49: equation (\ref{zm1}) becomes
50: \be
51: \D_t\tilde v = B\tilde v
52: \label{zm2}
53: \ee
54: where $B$ is the real-linear operator given by
55: \be
56: Bv\equiv i\left( (\Delta+\alpha(w)-\sigma^2)\,v +
57: \beta(w)e^{2i\,m\theta}\,\bar{v}\right).
58: \ee
59: We note that (\ref{zm2}) is equivalent to equation (\ref{(9)})
60: and that $B$ is the single-component form of the operator $A$.
61: (Because we are considering the real eigenvalue $\lambda=0$, in this section
62: we prefer not to work with the two-component form and subsequent
63: complexification.)
64:
65: \begin{itemize}
66: \item Invariance of NLS under phase changes:
67:
68: We define $u_\tau\equiv e^{i\tau}u_0$. Because
69: $g(e^{i\tau}u_0)=e^{i\tau}g(u_0)$ and $u_0$ is a solution of \eq{1nls},
70: $u_\tau$ is a solution for all real $\tau$.
71: For this family, $\tilde u = iu_0$ and $\tilde v = iv_0$.
72: Since $\tilde v$ is time-independent, we see from (\ref{zm2}) that
73: $\tilde v$ is an eigenfunction of $B$ with eigenvalue zero.
74: The corresponding eigenfunction
75: ${\bf \Psi}=(\Re\tilde v,\Im\tilde v)$
76: of $A$ lies in the $j=0$ subspace, according to (\ref{(15)}),
77: (\ref{(20)}), and (\ref{(17)}).
78:
79: \item Invariance of NLS under rotation:
80:
81: We set $u_\tau\equiv R_\tau u_0$, where $R_\tau$ is the coordinate
82: transformation that rotates $u_0$ by angle $\tau$. Thus
83: $u_\tau=e^{i\omega \,t}e^{i\,m(\theta-\tau)}w(r)$, whence
84: $\tilde u=-imu_0$ and $\tilde v=-imv_0$. This eigenfunction $\tilde v$
85: of $B$ is a real constant multiple of that induced by
86: invariance under phase change.
87:
88: \item Invariance of NLS under time translation:
89:
90: We set $u_\tau (x,t)\equiv u_0(x,t-\tau)$. Then $\tilde u=-i\omega u_0$
91: and $\tilde v=-i\omega v_0$. This eigenfunction $\tilde v$
92: of $B$ is a real constant multiple of those induced by
93: invariance under phase change and rotations.
94:
95: \item Invariance of NLS under spatial translation:
96:
97: Writing $(x,y)$ for the components of the spatial coordinate vector, we
98: begin by setting $u_\tau (x,y,t)\equiv u_0(x-\tau,y,t)$. Then
99: $\tilde u= \D_x u_0$ and $\tilde v= \D_x v_0$.
100: Since $\tilde v$ is time-independent, we see from (\ref{zm2}) that
101: $\tilde v$ is an eigenfunction of $B$ with eigenvalue zero.
102: By explicit computation, we see that according to (\ref{(15)}),
103: (\ref{(20)}), and (\ref{(17)}),
104: the corresponding eigenfunction ${\bf \Psi}$ of $A$ has components in
105: the $j=1$ and $j=-1$ subspaces.
106:
107: In a similar way, setting $u_\tau (x,y,t)\equiv u_0(x,y-\tau,t)$ yields
108: the eigenfunction $\tilde v=\D_y v_0$
109: of the operator $B$ with eigenvalue zero, and the corresponding
110: eigenfunction of $A$ has components in the $j=1$ and $j=-1$ subspaces.
111: It is straightforward to check that these two eigenfunctions resulting
112: from translations along the $x$ and $y$ axes are linearly independent.
113:
114: \item Invariance of NLS under change of standing-wave frequency:
115:
116: As mentioned in Section \ref{tax}, localized standing-wave solutions
117: $u_0=e^{i\omega\,t}e^{i\,m\theta }w(r)$ of \eq{1nls} exist for a range of
118: frequencies $\omega$. The radial profile $w(r)$ depends on $\omega$
119: through the differential equation (\ref{1weq}), in which
120: $f(w)=g(w)-\omega w$. Consequently, $v_0$ as well as $u_0$ is
121: parametrized by $\omega$. Differentiating with
122: respect to parameter $\omega$, we have
123: \be
124: \tilde u=itu_0 +
125: e^{i\omega\,t} \D_\omega v_0, \quad
126: \tilde v=itv_0 + \D_\omega v_0.
127: \ee
128: Equation (\ref{zm2}) thus gives
129: $ B\tilde v=iv_0$.
130: Since we established earlier that $B(iv_0)=0$, we see that
131: $ B \left(\D_\omega v_0 \right)=iv_0$.
132: Thus (again since $B(iv_0)=0$),
133: \be
134: B^2 \left( \D_\omega v_0\right)=0.
135: \ee
136: That is, $\D_\omega v_0$ is a
137: generalized eigenfunction of B of index 2 corresponding to
138: eigenvalue zero.
139: Because
140: \be
141: \D_\omega v_0 =
142: e^{i\,m\theta}\D_\omega w(r),
143: \ee
144: the angular dependence of the generalized eigenfunction
145: $\D_\omega v_0$
146: is identical to that generated by invariance under phase changes, and as
147: a result the corresponding generalized eigenfunction $\Psi$ of $A$ lies
148: in the $j=0$ subspace.
149:
150: \item Invariance of NLS under velocity boost:
151:
152: Let $c\in\R^2$ be a constant vector. If $u(x,t)$ is a solution to \eq{1nls},
153: then so is the function
154: \be
155: U_c(x,t)\equiv\exp \left\{ {i\left( {{{1 \over 2}}c\cdot x-
156: {{1 \over 4}}\left| c \right|^2t} \right)}\right\}
157: \,\,u(x-c\,t,\,\,t).
158: \ee
159: Again writing $(x,y)$ for the components of the spatial coordinate
160: vector, we first let $u_\tau$ be the standing wave $u_0$ boosted with
161: speed $\tau$ in the $x$ coordinate direction (that is, $c=(\tau,0)$):
162: \be
163: u_\tau(x,y,t)\equiv\exp \left\{ {i\left( {{{1 \over 2}}\tau \,
164: x-{{1 \over 4}}\tau ^2t} \right)} \right\}\,\,
165: u_0(x-\tau \,t,\,\,y,\,\,t).
166: \ee
167: Then
168: $\tilde u={{i\over 2}}x\,u_0-
169: t\D_x u_0$ and
170: $\tilde v={{i\over 2}}x\,v_0-
171: t\D_x v_0$.
172: Equation (\ref{zm2}) thus gives
173: $ B\tilde v=-\D_x v_0$.
174: Since we established earlier that
175: $B\left(\D_x v_0\right)=0$,
176: we see that
177: $ B(ixv_0) = -2\D_x v_0$.
178: Therefore (again since
179: $B\left(\D_x v_0\right)=0$),
180: \be
181: B^2 \left(ixv_0\right)=0.
182: \ee
183: That is, $ixv_0$ is a
184: generalized eigenfunction of B of index 2 corresponding to
185: eigenvalue zero.
186: By explicit computation, we see that
187: the corresponding eigenfunction ${\bf \Psi}$ of $A$ has components in
188: the $j=1$ and $j=-1$ subspaces.
189:
190: In a similar way,
191: letting $u_\tau$ be the standing wave $u_0$ boosted with
192: speed $\tau$ in the $y$ coordinate direction (that is, $c=(0,\tau)$),
193: we find that $iyv_0$ is a
194: generalized eigenfunction of B of index 2 corresponding to
195: eigenvalue zero, and the corresponding
196: eigenfunction of $A$ has components in the $j=1$ and $j=-1$ subspaces.
197: The two generalized eigenfunctions $ixv_0$ and $iyv_0$
198: resulting from velocity boosts in the $x$ and $y$ directions are linearly
199: independent.
200:
201: \end{itemize}
202:
203: We have thus found two $j=0$ generalized eigenfunctions, and four
204: $|j|=1$ generalized eigenfunctions, for eigenvalue
205: $\lambda=0$, as summarized in Table 2.
206:
207: \begin{table}\label{T.zeromodes}
208: \begin{center}
209: \caption{Zero modes}
210: \medskip
211: \begin{tabular}{|c|c|c|c|l|}
212: \hline
213: Transformation & $|j|$ & Index & Generalized eigenfunction \\ \hline
214: {Phase, rotation, time translation} & 0 & 1 & $i v_0$ \\
215: {Standing-wave frequency change} & 0 & 2 &${\D_\omega{v_0}}$ \\
216: {Space translation in } $x$ { direction} & 1 & 1 &${\D_x{v_0}}$\\
217: {Space translation in } $y$ { direction} & 1 & 1 &${\D_y{v_0}}$ \\
218: {Boost in } $x$ { direction} & 1 & 2 & $i x v_0$ \\
219: {Boost in } $y$ { direction} & 1 & 2 & $i y v_0$ \\
220: \hline \end{tabular}\end{center}
221: \end{table}
222:
223: To establish that these Jordan chains of generalized eigenfunctions do in fact
224: terminate with index 2 in the general case, we make use of the following
225: observation.
226:
227: {\lemma
228: Suppose $A=JL$ where $L$ is self-adjoint and $J^*=J^{-1}=-J$.
229: Suppose $\Psi_K$ is a generalized eigenfunction of $A$ with index $K$
230: corresponding to eigenvalue zero.
231: If there exists $\Phi$ in the domain of $A$
232: such that $A\Phi=\Psi_K$, then
233: $\left<\Psi_j \left| J\Psi_{K+1-j}\right.\right>=0$ for $j=1,\dots,K$,
234: where $\Psi_j\equiv A^{K-j}\Psi_K$.
235: \label{endchain}}
236:
237: \proof
238: From the definition of $\Psi_j$ it follows that $A\Psi_j=\Psi_{j-1}$ and
239: $A^j\Psi_j=0$ but $A^{j-1}\Psi_j\ne 0$. Thus $\Psi_j$ is a generalized
240: eigenfunction of $A$ with index $j$.
241:
242: Define $\Psi_j^*\equiv J\Psi_j$. From the assumed form for $A$ we then have
243: \be
244: A^*\Psi_j^* = -JA\Psi_j = -J\Psi_{j-1} = -\Psi_{j-1}^*.
245: \ee
246: It follows that
247: \be
248: (A^*)^j\Psi_j^* = (-1)^j JA\Psi_1 = 0.
249: \ee
250: Thus, for $1\le j\le K$, we have
251: \be
252: \ba{rcl}
253: \left< \Psi_j \left| J\Psi_{K+1-j} \right.\right>&=&
254: -\left< \Psi_j^* \left| \Psi_{K+1-j} \right.\right>
255: =-\left< \Psi_j^* \left| A^{j-1}\Psi_{K} \right.\right>
256: \cr
257: &=&-\left< \Psi_j^* \left| A^{j-1}A\Phi \right.\right>
258: =-\left< \Psi_j^* \left| A^j\Phi \right.\right>
259: \cr
260: &=&-\left< (A^*)^j\Psi_j^* \left| \Phi \right.\right>
261: =0,
262: \ea
263: \ee
264: as asserted.
265: \eproof
266:
267: We now apply this observation to our index-2 generalized eigenfunctions.
268: Consider first the generalized eigenfunction $\psi_2=ixv_0$ of $B$ arising from
269: velocity boost in the $x$ direction. Then
270: $\psi_1\equiv B\psi_2=-2\D_x v_0$ and we have
271: \be
272: \ba{rcl}
273: \left< \Psi_1 \left| J\Psi_2 \right.\right>
274: =\Re\left< \psi_1 \left| i\,\,\psi_2 \right.\right>
275: =\Re \left<-2\D_x v_0
276: \left| i\,\,ixv_0\right.\right>
277: =-\left\|{v_0}\right\|^2\ne 0.
278: \ea
279: \ee
280: Thus there is no generalized eigenfunction in this Jordan chain with index 3
281: (or higher).
282: Similarly, analyzing the index-2 generalized eigenfunction $\psi_2=iyv_0$ of $B$
283: arising from velocity boost in the $y$ direction, we find that there is no
284: generalized eigenfunction in this chain with higher index.
285:
286: We next consider the index-2 generalized eigenfunction
287: $\psi_2=\D_\omega v_0$
288: of $B$ arising from change of standing-wave frequency, which lies in the $j=0$
289: subspace.
290: We set $\psi_1\equiv B\psi_2 = iv_0$, and we find
291: \be
292: \ba{rcl}
293: \left< \Psi_1 \left| J\Psi_2 \right.\right>
294: =\Re\left< \psi_1 \left| i\,\,\psi_2 \right.\right>
295: =\Re \left<iv_0\left|
296: i\,\,\D_\omega v_0\right.\right>
297: =\half\D_\omega \left\|{v_0}\right\|^2.
298: \ea
299: \ee
300: Generically, we expect the quantity
301: $\D_\omega \left\|{v_0}\right\|^2$ to be nonzero,
302: hence that $\psi_2=\D_\omega v_0$
303: is the highest-index generalized eigenfunction in this Jordan chain. Our
304: numerical studies confirm this expectation for
305: nonlinearities other than pure cubic.
306:
307: The six-dimensional eigenspace for eigenvalue zero summarized in the
308: table is in agreement with our numerical observations. For
309: nonlinearities other than pure cubic,
310: we find numerically that the eigenvalue zero has
311: multiplicity two in each of the $j=0$, $j=1$, and $j=-1$ subspaces.
312:
313: In contrast, in the special case of a purely cubic nonlinearity
314: $g(u)=\gamma|u|^2u$ in two spatial dimensions,
315: the norm $\left\|{v_0}\right\|^2$ is independent of $\omega.$
316: To see this, we note that the function ${\tilde v}_0$ defined by
317: $v_0(x)=\sqrt{{\omega/{\gamma}}} \,\,
318: {\tilde v}_0(\sqrt{\omega}\,\,x)$ satisfies the $\omega$-independent equation
319: $\Delta{\tilde v}_0-{\tilde v}_0+|{\tilde v}_0|^2{\tilde v}_0=0$. It is
320: straightforward to compute that, in $N$ spatial dimensions,
321: $\left\|{v_0}\right\|^2={{1\over{\gamma}}}\omega^{(2-N)/2}
322: \left\|{\tilde v}_0\right\|^2$.
323: Thus, for the pure-cubic nonlinearity, in $N=2$ spatial dimensions, the index-2
324: generalized eigenfunction does not necessarily terminate the Jordan chain in
325: the $j=0$ subspace, and in fact generalized eigenfunctions
326: with indices 3 and 4 are present in this special case.
327:
328: The structure of the generalized null spaces of $A$ in the case of \eq{1nls}
329: with pure-power nonlinearities is established in Appendix B of the
330: paper \ccite{W85}.
331: The index-3 generalized eigenfunction
332: $\psi_3=ir^2v_0$ of $B$ arises from the invariance of cubic NLS in two
333: spatial dimensions under the Talanov lens transformation.
334: An index-4 generalized eigenfunction $\psi_4$ has the form
335: $\psi_4=\xi(r)e^{im\theta}$, where $\xi$ is the bounded solution to the
336: ordinary differential equation $L_3\xi=r^2w_0$, where
337: \[
338: L_3={{\D^2\over\D r^2}+{1\over r}{\D\over\D r}-{m^2
339: \over r^2}} -\omega +3\gamma w_0^2.
340: \]
341: There are no generalized eigenfunctions with higher index.
342: Our numerical observations in the pure cubic case show that eigenvalue
343: zero has multiplicity four in the $j=0$ subspace, in agreement with this
344: analysis.
345:
346:
347: \endinput
348:
349: \bigskip
350: $\matrix{
351: & & & \hbox{Generalized} \cr
352: \hbox{Transformation} & |j| & \hbox{Index} &
353: \hbox{Eigenfunction} \cr
354: & & & \cr
355: \hbox{Phase, rotation, time translation} & 0 & 1 & i v_0 \cr
356: \hbox{Standing-wave frequency change} & 0 & 2 &
357: {\partial{v_0}/\partial\omega} \cr
358: \hbox{Space translation in } x \hbox{ direction} & 1 & 1 &
359: {\partial{v_0}/\partial x}
360: \cr \hbox{Space translation in } y \hbox{ direction} & 1 & 1 &
361: {\partial{v_0}/\partial y} \cr
362: \hbox{Boost in } x \hbox{ direction} & 1 & 2 & i x v_0 \cr
363: \hbox{Boost in } y \hbox{ direction} & 1 & 2 & i y v_0 \cr
364: }$
365:
366: