1:
2: \section{WKB methods for nonlinear Schr\"odinger equations}
3: \label{sec:bkw}
4:
5: Consider the initial value problem, for $x\in \R^n$:
6: \begin{equation}
7: \label{eq:nlssemi}
8: ih \d_t u^h +\frac{h^2}{2}\Delta u^h = h^\kappa |u^h|^2
9: u^h\quad ; \quad u^h_{\mid t=0} = a_0^h(x)e^{i\phi_0(x)/h}\, .
10: \end{equation}
11: The aim of WKB methods is to describe $u^h$ in the limit $h \to 0$,
12: when $\phi_0$ does not depend on $h$, and
13: $a_0^h$ has an asymptotic expansion of the form:
14: \begin{equation*}
15: a_0^h (x)\sim a_0(x) +h a_1(x)+h^2 a_2(x) +\ldots
16: \end{equation*}
17: The parameter $\kappa \ge 0$ describes the strength of a coupling
18: constant, which makes nonlinear effects more or less important in the
19: limit $h \to 0$; the larger $\kappa$, the weaker the nonlinear
20: interactions. Note that since we consider an homogeneous nonlinearity,
21: this amounts to considering the case where the coupling constant is
22: $1$, with initial data of order $h^{\kappa /2}$.
23:
24:
25:
26: An interesting feature of \eqref{eq:nlssemi} is that one does not
27: expect the creation of harmonics. The WKB methods consist in seeking
28: an approximate solution to
29: \eqref{eq:nlssemi} of the form:
30: \begin{equation}\label{eq:defBKW}
31: u^h(t,x)\sim \(\a^{(0)}(t,x) + h \a^{(1)}(t,x) + h^2
32: \a^{(2)}(t,x)+\ldots \) e^{i\phi(t,x)/h}\, .
33: \end{equation}
34: For such an expansion to be available with profiles $\a^{(j)}$
35: independent of $h$, it is reasonable to assume that $\kappa$ is an
36: integer, $\kappa \in \N$.
37: One must not expect this approach to be valid when caustics are
38: formed: roughly speaking, when a caustic appears, all the terms
39: $\phi$, $\a^{(0)}$, $\a^{(1)}$, \ldots become singular. In this paper,
40: we always consider times preceding this break-up.
41:
42:
43: \subsection{Notion of criticality}
44: \label{sec:crit}
45:
46: If $\kappa \ge 2$, then nonlinear effects are negligible at
47: leading order in WKB methods. On the other hand, if $\kappa =1$
48: (weakly nonlinear geometric
49: optics), then nonlinear effect are relevant at leading order. The
50: present discussion is
51: formal, its aim being to prepare the study of the case $\kappa =0$.
52:
53: When $\kappa \ge 2$, plugging the asymptotic expansion \eqref{eq:defBKW}
54: into \eqref{eq:nlssemi} yields formally:
55: \begin{equation*}%\label{eq:veryweak}
56: \begin{aligned}
57: \d_t \phi + \frac{1}{2}|\nabla \phi|^2 =0 \quad & ;\quad
58: \phi_{\mid t=0} = \phi_0 \, .\\
59: \d_t \a^{(0)} + \nabla \phi \cdot \nabla \a^{(0)}
60: +\frac{1}{2}\a^{(0)}\Delta \phi =0 \quad & ;\quad
61: \a^{(0)}_{\mid t=0} = a_0\, .
62: \end{aligned}
63: \end{equation*}
64: The first equation is the well known eikonal equation, which describes
65: the geometry of the propagation. If $\phi_0$ is smooth, it has a
66: smooth solution, locally in time. This solution may become singular in
67: finite time, this phenomenon being the formation of a caustic.
68:
69: The second equation is a transport
70: equation, which is simply an ordinary differential equation for the
71: leading order amplitude along the rays of geometrical optics. To see
72: this, introduce a parametrization of these rays:
73: \begin{equation*}
74: \frac{d}{dt}X_t(x) = \nabla \phi\(t,X_t(x)\) \quad ;\quad X_0(x)=x
75: \, ,
76: \end{equation*}
77: and the Jacobi determinant:
78: $\displaystyle J_t(x) =\operatorname{det}\nabla X_t(x)$.
79: It is well defined and smooth so long as no caustic appears.
80: %and solves the
81: %Liouville equation:
82: %\begin{equation*}
83: % \frac{d}{dt}J_t(x) = \Delta \phi \(t,X_t(x)\)J_t(x) \quad ;\quad
84: % J_0(x) =1\, .
85: %\end{equation*}
86: The break-up time $t_c>0$, if any, is such that there exists $x_c$ such
87: that $J_{t_c}(x_c)=0$.
88: The transport equation for $\a^{(0)}$ is the trivial
89: ordinary differential equation:
90: \begin{equation*}
91: \frac{d}{dt} \( \a^{(0)}\( t,X_t(x)\)\sqrt{J_t(x)}\)=0\, .
92: \end{equation*}
93: From this, we easily see that when a caustic appears, not only $\phi$
94: becomes singular, but also $\a^{(0)}$, since $J_t(x)$ goes to zero at
95: the caustic.
96:
97: %The correctors $(\a^{(j)})_{j\ge 2}$ solve similar linear transport
98: %equations. For instance, if $\kappa =2$, we find:
99: %\begin{equation*}
100: % \d_t \a^{(1)} + \nabla \phi \cdot \nabla \a^{(1)}
101: % +\frac{1}{2}\a^{(1)}\Delta \phi =-\frac{1}{2}\Delta a^{(0)} \quad ;\quad
102: % a^{(1)}_{\mid t=0} = a_1 \,.
103: %\end{equation*}
104: %This shows that in general, $\a^{(1)}$ also becomes singular at the
105: %caustic.
106:
107: The value $\kappa =1$ is critical as far as leading order phenomena
108: are concerned: the transport equation for $\a^{(0)}$ is then nonlinear,
109: \begin{equation}\label{eq:weak}
110: \begin{aligned}
111: \d_t \phi + \frac{1}{2}|\nabla \phi|^2 &=0 \quad ;\quad
112: \phi_{\mid t=0} = \phi_0 \, .\\
113: \d_t \a^{(0)} + \nabla \phi \cdot \nabla \a^{(0)}
114: +\frac{1}{2}\a^{(0)}\Delta \phi &=-i \left|\a^{(0)}\right|^2 \a^{(0)}
115: \quad ;\quad
116: \a^{(0)}_{\mid t=0} = a_0\, .
117: \end{aligned}
118: \end{equation}
119: On the other hand, the eikonal equation is still the same as in the
120: linear case, hence the term ``weakly nonlinear'' (see also
121: \cite{RauchUtah} and references therein). The correctors
122: $(\a^{(j)})_{j\ge 2}$ solve linear transport
123: equations. With the above notations, the
124: nonlinear transport equation is again an ordinary differential
125: equation along rays:
126: \begin{equation*}
127: \frac{d}{dt} \( \a^{(0)}\( t,X_t(x)\)\sqrt{J_t(x)}\)=-i
128: \left|\a^{(0)}\( t,X_t(x)\)\right|^2
129: \a^{(0)}\( t,X_t(x)\)\sqrt{J_t(x)}\, .
130: \end{equation*}
131: This ordinary differential
132: equation is of the form $\dot y = iV y$, where the nonlinear potential
133: $V$ is real-valued. In particular, the modulus of $y$ is constant, and
134: we just have to solve a \emph{linear} differential
135: equation. Thus, leading order
136: nonlinear effects are measured by a (nonlinear) phase shift, which may
137: be compared to the phenomenon of \emph{phase self-modulation} in laser
138: physics (see e.g. \cite{ZS,Boyd,Donnat}).
139:
140:
141: \subsection{Super-critical case}
142: \label{sec:super}
143:
144: In the super-critical case $\kappa =0$, the nonlinearity is
145: present in the eikonal equation: the hierarchy of the case $\kappa=1$
146: is shifted, so that the corrector $\a^{(1)}$ is present in the
147: transport equation for $\a^{(0)}$. As noted in
148: \cite{PGX93}, the system for the phase $\phi$ and the
149: amplitudes
150: $\a^{(0)}$, $\a^{(1)},\ldots$ is not closed (see also
151: \cite{CheverryBullSMF,CGM04,CG05}). For instance, we find:
152: \begin{equation}\label{eq:strong}
153: \begin{aligned}
154: \d_t \phi + \frac{1}{2}|\nabla \phi|^2 +\left|\a^{(0)}\right|^2
155: &=0 \, .\\
156: \d_t \a^{(0)} + \nabla \phi \cdot \nabla \a^{(0)}
157: +\frac{1}{2}\a^{(0)}\Delta \phi &=-2i
158: \a^{(0)}\operatorname{Re}\(\a^{(0)}\overline{\a^{(1)}} \)\, .\\
159: \d_t \a^{(1)} + \nabla \phi \cdot \nabla \a^{(1)}
160: +\frac{1}{2}\a^{(1)}\Delta \phi &=\frac{i}{2}\Delta
161: \a^{(0)}-i\left|\a^{(1)}\right|^2 a^{(0)}\\
162: -2i
163: \a^{(1)}&\operatorname{Re}\(\a^{(0)}\overline{\a^{(1)}} \)
164: -2i
165: \a^{(0)}\operatorname{Re}\(\a^{(0)}\overline{\a^{(2)}} \)
166: \, .
167: \end{aligned}
168: \end{equation}
169: However, as pointed out in \cite{PGX93}, the phase $\phi$ can be found when
170: considering:
171: $$(\rho ,v)= \(|\a^{(0)}|^2, \nabla
172: \phi\).$$
173: Indeed, it solves the compressible, isentropic Euler
174: equation:
175: \begin{equation}\label{eq:euler}
176: \begin{aligned}
177: \d_t \rho +\operatorname{div}\(\rho v\) &= 0\quad ;\quad \rho_{\mid
178: t=0}=|a_0|^2\, .\\
179: \d_t v +v\cdot \nabla v + \nabla \rho &= 0\quad ;\quad v_{\mid
180: t=0}=\nabla\phi_0\, .
181: \end{aligned}
182: \end{equation}
183: For smooth initial data decaying to zero at infinity, this system as a
184: smooth solution locally in time \cite{Majda,MUK86,JYC90}. In general,
185: finite time blowup occurs \cite{MUK86,JYC90,Xin}, but not always
186: \cite{Grassin}; the known results depend on the propagation of
187: the initial velocity by the (multi-dimensional) Burgers' equation.
188:
189: Once $(\rho,v)$ is
190: determined, $\d_t \phi$ is given by the eikonal equation; this yields
191: $\phi$. Note that knowing $(\rho,v)$ suffices to compute important
192: quadratic quantities such as Wigner measures.
193: To complete the closure of the system, and provided that the
194: leading order amplitude $a_0$ is nowhere zero, one may consider a
195: generalized Madelung transform (see \cite{PGX93}), which we do not
196: describe here.
197:
198:
199:
200:
201:
202:
203: \subsection{Justification on small time intervals}
204: \label{sec:justifsmall}
205:
206: Justifying geometric optics in the super-critical case is, in general,
207: an open problem. However, as noticed in \cite{PGX93} and exploited in
208: many other works (see
209: e.g. \cite{Kuksin95,MetivierBourbaki,CCT2,BGTENS,BZ}), if one studies
210: this limit on time
211: intervals of the form $[0, c_0 h|\ln h|]$ for some $c_0>0$, then the
212: problem is simpler. Consider the more general nonlinear Schr\"odinger
213: equation in
214: $\R^n$:
215: \begin{equation}
216: \label{eq:NLSgensemi}
217: ih\d_t u^h +\frac{h^2}{2}\Delta u^h = \omega |u^h|^{2\si}u^h \quad ;
218: \quad u^h(0,x)=a_0(x)\, ,
219: \end{equation}
220: where $\omega \in \R\setminus \{0\}$ and $\si\in \N\setminus\{0\}$. We
221: consider the case $\phi_0\equiv 0$ to prove that in this case, one can
222: choose an
223: approximate which is even simpler than the one given by
224: \eqref{eq:euler}.
225: Formally, $u^h$ is formally approximated by $a
226: e^{i\phi /h}$ where:
227: \begin{equation*}
228: \begin{aligned}
229: \d_t \phi +\frac{1}{2}|\nabla \phi|^2 + \omega |a|^{2\si} =0 \quad ;\quad
230: \phi_{\mid t=0} =0\, .\\
231: \d_t a +\nabla \phi\cdot \nabla a +\frac{1}{2}a\Delta \phi =0\quad
232: ; \quad a_{\mid t=0}=a_0\, .
233: \end{aligned}
234: \end{equation*}
235: Looking at Taylor expansions for $\phi$ and $a$ as $t\to 0$, we see that
236: \begin{equation*}
237: a(t,x) = a_0(x) + \O(t^2)\quad ;\quad \phi(t,x) = -t \omega
238: |a_0(x)|^{2\si}+\O(t^3)\, .
239: \end{equation*}
240: We prove that $a(t,x)
241: e^{i\phi(t,x) /h}$ can approximated by $a_0(x) e^{-it \omega
242: |a_0(x)|^{2\si}/h}$ on some time interval of the form $[0, c_0 h|\ln
243: h|]$. Call $\varphi^h$ the latter function. It solves the ordinary
244: differential equation:
245: \begin{equation}\label{eq:varphi}
246: ih\d_t \varphi^h = \omega
247: |\varphi^h|^{2\si}\varphi^h \quad ;
248: \quad \varphi^h_{\mid t=0}=a_0\, .
249: \end{equation}
250: We prove that if $c_0$ is sufficiently small, then $\varphi^h$ is a
251: good approximation of $u^h$ on $[0, c_0 h|\ln
252: h|]$. Let $w^h = u^h-\varphi^h $. It solves:
253: \begin{equation}
254: \label{eq:1435}
255: ih\d_t w^h +\frac{h^2}{2}\Delta w^h= \omega\( F(w^h
256: +\varphi^h) - F(\varphi^h)\) +\O(h^2)+\O(t^2)+\O(h t)\, ,
257: \end{equation}
258: with $ w^h_{\mid t=0}=0$, where we have set $F(z)=|z|^{2\si}z$. The
259: $\O(h^2)$ term corresponds to the fact that we consider only the first
260: two terms of a WKB analysis, the term $\O(t^2)$ stems from the
261: approximation of the phase for small times, and $\O(ht)$ from the
262: approximation of the amplitude for small times. To be more precise, we
263: must say that these source terms are measured in $L^2\cap L^\infty
264: (\R^n)$. When measured in $H^k$, they must be multiplied by a factor
265: of order $ 1 + (t/h)^k$, due to the differentiation of the phase.
266: For $k\ge 0$, we have:
267: \begin{align*}
268: \|w^h\|_{L^\infty([0,t];H^k)} \lesssim & \frac{1}{h}\left\|F(w^h
269: +\varphi^h) - F(\varphi^h) \right\|_{L^1([0,t];H^k)} \\
270: &+
271: \frac{1}{h}\int_0^t \(h^2 +
272: s^2 +h s\)\left\langle\frac{s}{h}\right\rangle^kds .
273: \end{align*}
274: At least for $\si$ integer, we have, when $k>n/2$:
275: \begin{align*}
276: \left\|F(w^h(t)
277: +\varphi^h(t)) - F(\varphi^h(t)) \right\|_{H^k}&\lesssim \(
278: \|w^h(t)\|^{2\si}_{H^k} +
279: \|\varphi^h(t)\|^{2\si}_{H^k}\)\|w^h(t)\|_{H^k}\\
280: &\lesssim \(
281: \|w^h(t)\|^{2\si}_{H^k} +
282: \left\langle\frac{t}{h}\right\rangle^{2\si k}\)\|w^h(t)\|_{H^k}.
283: \end{align*}
284: On any time interval where we have, say, $\|w^h\|_{H^k}\le 1$, we infer:
285: \begin{equation*}
286: \|w^h\|_{L^\infty([0,t];H^k)} \le
287: \frac{C}{h}\int_0^t \left\langle\frac{s}{h}\right\rangle^{2\si k}
288: \| w^h (s)\|_{H^k}ds + C_1\int_0^t \(h +
289: \frac{s^2}{h} +s\)\left\langle\frac{s}{h}\right\rangle^k ds.
290: \end{equation*}
291: Gronwall lemma yields:
292: \begin{align*}
293: \|w^h\|_{L^\infty([0,t];H^k)} &\lesssim \int_0^t \(h +
294: \frac{s^2}{h}
295: +s\)\left\langle\frac{s}{h}\right\rangle^k\exp\(\frac{C}{h}\int_s^t
296: \left\langle\frac{\tau}{h}\right\rangle^{2\si k}
297: d\tau\)ds.
298: \end{align*}
299: Let $t^h = c_0 h |\ln h|^\theta$:
300: \begin{align*}
301: \|w^h\|_{L^\infty([0,t^h];H^k)} &\lesssim \exp\( Cc_0 |\ln h|^\theta
302: \left\langle \ln h \right\rangle^{2\si k\theta} \)\int_0^{t^h} \(h s +
303: \frac{s^3}{h}
304: +s^2\)\left\langle\frac{s}{h}\right\rangle^k ds\\
305: &\lesssim \exp\( Cc_0 |\ln h|^\theta
306: \left\langle c_0\ln h \right\rangle^{2\si k\theta} \)h^2 |\ln h|^{4\theta}.
307: \end{align*}
308: For $\theta =(1+2\si k)^{-1}$ and $c_0$ sufficiently small, this
309: yields:
310: \begin{equation*}
311: \|w^h\|_{L^\infty([0,t^h];H^k)} \lesssim h |\ln h|^{\frac{4}{1+2\si k}}.
312: \end{equation*}
313: We can then conclude with a continuity argument, for $h$
314: sufficiently small:
315: \begin{prop}\label{prop:bkwlog}
316: Let $n\ge 1$, $\omega \in \R\setminus \{0\}$, $\si>0$ an integer,
317: and $a_0\in \S(\R^n)$. Fix $k>n/2$. Then we can find $c_0,c_1,\theta>0$
318: independent of $h\in
319: ]0,1]$ such that $u^h$ and $\varphi^h$, solutions to
320: \eqref{eq:NLSgensemi} and \eqref{eq:varphi} respectively, satisfy:
321: \begin{equation*}
322: \|u^h-\varphi^h\|_{L^\infty([0,c_0 h|\ln h|^\theta];H^k)} \lesssim h
323: |\ln h|^{c_1}\, .
324: \end{equation*}
325: \end{prop}
326: