1: \section{Stability issues}
2: \label{sec:stab}
3:
4: The construction of Section~\ref{sec:formal} and the results of the
5: previous paragraph do not agree. To see this, we come back to
6: Proposition~\ref{prop:small}: in \eqref{eq:bkwlimite}, we have
7: \begin{align}
8: \O\(t^{n-2}\):\quad & \phi_1(x) = \frac{1}{n-1}f'(0)|a_0(x)|^2\,
9: ,\label{eq:phi1} \\
10: \O\(t^{n-1}\):\quad & a_1 + \nabla \phi_1\cdot \nabla a_0
11: +\frac{1}{2}a_0\Delta \phi_1 =0\, ,\label{eq:a1}\\
12: \O\(t^{2n-2}\):\quad & (2n-1)\phi_2 +\frac{1}{2}|\nabla \phi_1|^2
13: +2\operatorname{Re}(\overline{a_0}a_1)f'(0)
14: +\frac{f''(0)}{2}|a_0|^4=0\, . \label{eq:phi2}
15: \end{align}
16: The function $\phi_1$ is the same as the one obtained by the approach
17: of Section~\ref{sec:formal}: the two approximate solutions are close
18: to each other up to the first boundary layer, when the first phase
19: shift appears. On the other hand, we see that to get
20: $\phi_2$, the modulation of the amplitude ($a_1$) must be taken into
21: account; in \eqref{eq:evolg}, $g_2$ is computed without
22: evaluating $\Delta a_0$, unlike $\phi_2$. This means in particular
23: that the two approximate solutions diverge when reaching the second
24: boundary layer: the approach of Section~\ref{sec:formal} is only
25: formal, and does not lead to a good approximation. And yet, the source
26: term in Proposition~\ref{prop:formal} is small: thus, the linearized
27: semi-classical Schr\"odinger operator is not stable, in the
28: semi-classical limit. We will see below that this instability is not
29: due to a spectral instability, but to the fact that the approach
30: followed to construct the formal approximation was too crude.
31:
32: This phenomenon is due to the super-criticality of the problem.
33: Indeed, for fixed $\e$, we
34: deal with a nonlinear Schr\"odinger equation with repulsive
35: nonlinearity ($f'>0$), for which global well-posedness results are
36: available (see Remark~\ref{rema:reg}). When using the transform
37: \eqref{eq:pseudo}, notice that the
38: parameter $\h$ in \eqref{eq:psi} goes to zero as $\e\to 0$ only when
39: $n>k$, that is in the super-critical case (compare with
40: Section~\ref{sec:critical}).
41:
42:
43: To understand better the instability mechanism, let us go back to the
44: comparison between the construction of
45: Section~\ref{sec:formal} and the results of the previous
46: paragraph. Letting $N\to +\infty$ in Proposition~\ref{prop:formal}, we
47: have an approximate solution of the form
48: \begin{align*}
49: v^\e (t,x) &= \frac{e^{i\frac{|x|^2}{2\e(t-1)}}}{(1-t)^{n/2}}
50: a_0\(\frac{x}{1-t}\) \exp\(i\frac{1-t}{\e}
51: g\(\frac{\e^\g}{1-t} \virgp \frac{x}{1-t}\)\)\\
52: &= \frac{e^{-i\frac{|\xi|^2}{2\h \tau}}}{(1-t)^{n/2}}
53: a_0\(\xi\) \exp\(\frac{i}{\h }
54: \frac{g\(\tau,\xi\)}{\tau}\)\Big|_{(\tau,\xi)=\(\frac{\e^\g}{1-t} \virgp
55: \frac{x}{1-t}\) }.
56: %\\
57: %&= \frac{e^{-i\frac{|\xi|^2}{2\h \tau}}}{(1-t)^{n/2}}
58: % a_0\(\xi\)e^{i\tilde \phi\(\tau,\xi\)/\h}
59: % \Big|_{(\tau,\xi)=\(\frac{\e^\g}{1-t} \virgp
60: % \frac{x}{1-t}\) }
61: \end{align*}
62: This formula and the transform \eqref{eq:pseudo} show that the
63: approximation of Section~\ref{sec:formal} is too crude, since it ignores
64: the coupling between phase and amplitude for \eqref{eq:psi}.
65: Proposition~\ref{prop:small} and \eqref{eq:bkwlimite} show that to
66: have a good approximation of the phase, the coupling between phase and
67: amplitude must be taken into account at \emph{every} order.
68:
69: We can go one step further in the understanding of this apparent
70: instability, by applying the
71: transform \eqref{eq:pseudo} to the intermediary approximate solution
72: $v_N^\e$. We show that the formal approximation stops being a good
73: approximation between the first and the second boundary layer. Assume
74: $\si=1$ so that the homogeneous nonlinearity satisfies
75: Assumptions~\ref{hyp:tout}.
76: Like for the exact solution, write
77: \begin{equation*}
78: v_N^\e(t,x) =
79: \frac{1}{(1-t)^{n/2}}\psi_N^\e\(\frac{\e^\g}{1-t}\virgp
80: \frac{x}{1-t} \) e^{i\frac{|x|^2}{2\e(t-1)}}\, .
81: \end{equation*}
82: Using the expression \eqref{eq:r_N}, we check that $\psi_N^\h$
83: solves
84: \begin{align*}
85: i\h \d_t \psi_N^\h +\frac{\h^2}{2}\Delta \psi_N^\h = t^{n-2}
86: |\psi_N^\h|^2
87: \psi_N^\h +\theta_N^\h(t,x)\, ,
88: \end{align*}
89: along with the initial condition
90: $\psi_N^\h\big|_{ t =
91: \h^{\frac{\g}{1-\g}}} = a_0(x)+\O\(\h^{(\alpha -1)(1-\g)}\)$ in
92: $H^s(\R^n)$ for any $s>0$, where:
93: \begin{equation*}
94: \theta_N^\h(t,x) = \(t^{(N+1)n-2}K_0(x) + i\h K_1(t,x)\)
95: \psi_N^\h(t,x) +i\h K_2(t,x) + \h^2 K_3(t,x)\, ,
96: \end{equation*}
97: for some ``nice'' functions $K_j$. Now write $\psi_N^\h(t,x) =
98: a_N^\h(t,x) e^{i\phi_N^\h(t,x)/\h}$. We have:
99: \begin{equation}
100: \label{eq:systhyp2}
101: \begin{aligned}
102: \d_t \bv^\h &+\sum_{j=1}^n A_j(\bv^\h)\d_j \bv^\h
103: = \frac{\h}{2} L
104: \bv^\h + {\tt S}^\h (t,x)\, , \text{ with }\bv^\h(t,x) = \left(
105: \begin{array}[l]{c}
106: \operatorname{Re} a_N^\h \\
107: \operatorname{Im} a_N^\h\\
108: \d_1\phi_N^\h \\
109: \vdots \\
110: \d_n\phi_N^\h
111: \end{array}
112: \right),\\
113: \text{and}&\quad {\tt S}^\h(t,x) = (t+t_0^\h)\left(
114: \begin{array}[l]{c}
115: K_1+ {\rm Re}\( (K_2-i\h K_3)e^{i\phi_N^\h/\h}\) \\
116: K_1+ {\rm Im}\( (K_2-i\h K_3)e^{i\phi_N^\h/\h}\) \\
117: -(t+t_0^\h)^{(N+1)n-3}\d_1K_0 \\
118: \vdots \\
119: -(t+t_0^\h)^{(N+1)n-3}\d_nK_0
120: \end{array}
121: \right)\, ,
122: \end{aligned}
123: \end{equation}
124: where the matrices $A_j$ are the same as in Section~\ref{sec:grenier} and the
125: functions in the definitions of $\bv^\h$ and ${\tt S}^\h$ are evaluated
126: at $(t+t_0^\h,x)$. We can proceed like in Section~\ref{sec:grenier}:
127: the new term is the source ${\tt S}^\h$. Unlike for the exact
128: solution, the oscillatory aspect of the problem has not disappeared:
129: the first two components of ${\tt S}^\h$ contain a highly oscillatory
130: factor. Therefore, we cannot expect $\h$ independent energy
131: estimates here. To measure the effect of this oscillatory term, forget
132: the shift in time, and take $ t_0^\h=0$. Then assuming that for
133: small times, $\d_x^a \phi_N^\h (t,x) = \O(t^{n-1})$ for any
134: multi-index $a$ (like for the exact solution), the $H^s$ norms of
135: the first two components of ${\tt S}^\h$ are controlled by
136: \begin{equation*}
137: \O\( t + \frac{t^{1+s(n-1)}}{\h^s}\).
138: \end{equation*}
139: A source of order $\O(t)$ is not a problem, since we eventually
140: consider the limit $t\to 0$. On the other hand, let us examine the
141: last term. Back to the initial variables, this yields a control by
142: \begin{equation*}
143: \( \frac{\e^\g}{1-t}\)^{1+s(n-1)} \e^{-s(1-\g)} = \frac{\e^{\g
144: +s\alpha -s}}{(1-t)^{1+s(n-1)}}\, .
145: \end{equation*}
146: This is small for $1-t\gg \e^\omega$, with
147: \begin{equation*}
148: \omega = \frac{\g
149: +s\alpha -s}{1+s(n-1)}\, .
150: \end{equation*}
151: We check that for $n >\alpha =k>1$, we have
152: \begin{equation*}
153: \beta = \frac{\alpha -1}{n-1}< \omega = \frac{\g
154: +s(\alpha -1)}{1+s(n-1)} < \frac{2\alpha -1}{2n-1}\, , \text{ for
155: any } s\ge 1\,.
156: \end{equation*}
157: The first inequality means that we can expect the formal approximation
158: to be a good approximation of the exact solution beyond the first
159: boundary layer (and indeed, it is close to the approximate solution of
160: Section~\ref{sec:justif}). The second one explains why the
161: approximation ceases to
162: be relevant before the second boundary layer.
163:
164: A possible way to understand the above computation is that the choice
165: of the variables is crucial: working with the ``usual'' unknown $v^\e$
166: (as in Section~\ref{sec:formal}) is not very efficient. On the other
167: hand, with
168: the variables introduced by E.~Grenier for his generalized WKB
169: methods, a precise and rigorous
170: analysis is possible, \emph{via} the transform
171: \eqref{eq:pseudo}. Thus, adding new variables helps the analysis:
172: this goes in the same direction as the general theory of geometric
173: optics, and the recent approach followed by
174: C.~Cheverry for a refinement of this principle
175: \cite{CheverryCMP,CheverryBullSMF}.
176: