1: \documentclass[11pt]{amsart}
2: \usepackage{standard}
3: %%% \DeclareMathOperator{\WFlag}{{}^{\lag}WF}
4: \newcommand{\Slag}{S_\lag}
5: \newcommand{\CIc}{{\mathcal{C}^\infty_c}}
6: \newcommand{\Psih}{\Psi_h}
7: \newcommand{\lag}{\mathcal{L}}
8: \newcommand{\md}{\mathcal{M}}
9: \newcommand{\Ibar}{\overline{I}}
10: \newcommand{\Tbar}{\overline{T}}
11: \newcommand{\ombar}{\overline{\omega}}
12: \newcommand{\orbit}{\rho}
13: \newcommand{\gammabar}{\bar{\gamma}}
14: \newcommand{\tW}{\tilde{W}}
15:
16: \title{Spreading of Lagrangian regularity on rational invariant tori}
17: \author{Jared Wunsch}
18: \date{\today}
19:
20: \begin{document}
21: \begin{abstract}
22: Let $P_h$ be a self-adjoint semiclassical pseudodifferential operator on a
23: manifold $M$ such that the bicharacteristic flow of the principal symbol on
24: $T^*M$ is completely integrable and the subprincipal symbol of $P_h$
25: vanishes. Consider a semiclassical family of eigenfunctions, or, more
26: generally, quasimodes $u_h$ of $P_h.$ We show that on a nondegenerate
27: rational invariant torus, Lagrangian regularity of $u_h$ (regularity under
28: test operators characteristic on the torus) propagates both along
29: bicharacteristics, and also in an additional ``diffractive'' manner. In
30: particular, in addition to propagating along null bicharacteristics,
31: regularity fills in the interiors of small annular tubes of
32: bicharacteristics.
33: \end{abstract}
34:
35: \maketitle
36:
37: \section{Introduction}
38: It is a well-known fact of semiclassical microlocal analysis, that the
39: analogue of H\"ormander's theorem on propagation of singularities for
40: operators of real principal type \cite{Hormander:Propagation} holds for the
41: semiclassical wavefront set (also known as ``frequency set''): it
42: propagates along null bicharacteristics of operators with real principal
43: symbol \cite{Evans-Zworski,Martinez}. Given a Lagrangian submanifold
44: $\lag$ of $T^*M,$ we may introduce a finer notion of regularity, the local
45: \emph{Lagrangian} regularity along $\lag.$ We show here that on rational
46: invariant tori in integrable systems, local Lagrangian regularity not only
47: propagates along bicharacteristics, but spreads in additional ways as well.
48:
49: Let $P_h$ be a semiclassical pseudodifferential operator on a manifold $M,$
50: with real principal symbol $p$ (this is automatic if $P$ is self-adjoint).
51: Assume that the bicharacteristic flow of $p$ is completely integrable. (In
52: fact we only need to assume integrability \emph{locally,} near one
53: invariant torus.) Let $u_h$ be a family of quasimodes of $P_h,$ i.e.\
54: assume that $\norm{(P_h-\lambda)u_h}_{L^2} =O(h^N)$ for some $N\in \NN,$ as
55: $h\downarrow 0$ either through a discrete sequence or continuously. (Note
56: that this certainly includes the possibility of letting $u_h$ be a sequence
57: of actual eigenfunctions). Let $\lag$ be an invariant torus in the
58: characteristic set $\{p=\lambda\}.$ Then the bicharacteristic flow is by
59: definition tangent to $\lag,$ and we show (even in the absence of the
60: integrability hypothesis) that Lagrangian regularity propagates along
61: bicharacteristics---this is Theorem~A below. If a single trajectory is
62: dense in $\lag,$ then this is the whole story for propagation, as the set
63: on which Lagrangian regularity holds is open, hence the whole torus either
64: enjoys Lagrangian regularity or none of it does. At the opposite extreme,
65: if $\lag$ is a torus on which all frequencis of the motion are rationally
66: related, we may ask the finer question: what subsets of the space of all
67: orbits may carry Lagrangian regularity? The answer (assuming a
68: nondegeneracy condition holds) turns out to be somewhat constrained: given
69: a single orbit, Lagrangian regularity along a small tube around it implies
70: Lagrangian regularity along the orbit itself. This is our Theorem~B. (In
71: the special case of two-dimensional tori, we can go further: again, either
72: the whole torus enjoys Lagrangian regularity or no points on it do.) The
73: order of regularity up to which our result holds is constrained by the
74: order of the quasimode. We speculate that a finer theorem may be
75: obtainable by more authentically ``second-microlocal'' methods.
76:
77: \begin{example}
78: As a simple example of our main result, Theorem~B, we consider the case
79: $M=S_x^1 \times S_y^1,$ $P_h = h^2 \Lap =-h^2(\pa^2/\pa x^2+ \pa^2/ \pa y^2);$
80: we consider Lagrangian regularity on the Lagrangian torus $\lag =
81: \{\xi=0,\eta =1\}$ for quasimodes satisfying
82: $$
83: (h^2\Lap -1) u_h \in h^{k+1} L^2(S^1 \times S^1).
84: $$ Lagrangian regularity on this particular $\lag$ is special in that we
85: may test for it using powers of the \emph{differential} operator
86: $D_x=i^{-1} (\pa/\pa x).$ The theorem tells us the following in this case:
87: let $\Upsilon(x)$ be a smooth cutoff function supported on $\{\abs{x} \in
88: [\ep,3\ep]\}$ and nonzero at $\pm 2 \ep.$ Let $\phi$ be another cutoff,
89: nonzero at the origin and supported in $[-2\ep,2\ep].$ If, for all $k'\leq
90: k,$ we have
91: $$
92: \norm{D_x^{k'}(\Upsilon(x) u_h)}\leq C<\infty,
93: $$
94: then for all $k'\leq k,$
95: $$
96: \norm{D_x^{k'}(\phi(x) u_h)}\leq \tilde{C}<\infty,
97: $$ i.e.\ the $D_x^k$ regularity fills in the ``hole'' in the support
98: of $\Upsilon.$ In this special case, the result can be proved directly by
99: employing a positive commutator argument using only differential operators;
100: the positive commutator will arise from the usual commutant $h^{-1} xD_x.$
101: \end{example}
102:
103: A less trivial example, that of the spherical pendulum, is discussed in
104: \S\ref{section:flow} below.
105:
106: The methods of proof (and the idea of the paper) arose from work of
107: Burq-Zworski \cite{Burq-Zworski1,Burq-Zworski2} and a subsequent refinement
108: by Burq-Hassell-Wunsch \cite{BHW1} on the spreading of $L^2$ mass for
109: quasimodes on the Bunimovich stadium. The central argument here is a
110: generalization of the methods used to prove that a quasimode cannot
111: concentrate too heavily in the interior of the rectangular part of the
112: stadium (which is essentially the example discussed above on $M=S^1 \times
113: S^1$).
114:
115: We remark that our hypotheses in this paper are quite far from those in the
116: study of ``quantum integrable systems'' where one examines eigenfunctions
117: of a system of $n$ commuting operators on an $n$-manifold. For instance,
118: if we take $P_h = h^2 \Lap +h^2 V$ on the torus, with $V$ a real valued,
119: smooth bump-function, then the operator $P_h$ satisfies the hypotheses of
120: our Theorems A and B, and yet there does not exist a system of $n-1$ other
121: operators commuting with $P_h,$ with independent symbols. Moreover, even
122: in the completely integrable case, given that we study eigenfunctions of a
123: single operator, it may be possible to use the degeneracy of the system to
124: construct non- or partially-Lagrangian quasimodes. Little seems to be
125: known in this direction.
126:
127: The author is grateful to Andr\'as Vasy for helpful discussions on
128: Lagrangian regularity, and to Clark Robinson for introducing him to
129: isoenergetic nondegeneracy. He has also benefitted greatly from comments
130: on an earlier version of the manuscript by Maciej Zworski, and by an
131: anonymous referee. This work was supported in part by NSF grant
132: DMS-0401323.
133:
134: \section{Lagrangian Regularity}
135: We begin by setting some notation and recalling some concepts of
136: semiclassical analysis. For detailed background on this subject, we refer
137: the reader to \cite{Evans-Zworski,Martinez}.
138:
139: Let $M^n$ be a smooth manifold and fix $\lag \subset T^*M$ a Lagrangian
140: submanifold. Throughout the rest of the paper, we assume\footnote{We may
141: just as well assume that $h \downarrow 0$ through a discrete sequence; this
142: will make no difference in what follows.} $u_h \in L^2(M;\Omega^{1/2}),$
143: with $h \in (0, h_0);$ here $\Omega^{1/2}$ denotes the bundle of
144: \emph{half-densities} on $M,$ i.e.\ the square root of the
145: density bundle $\abs{\Lambda^n M}.$
146: We will in future, however, suppress the
147: half-density nature of $u_h$ as well as its $h$-dependence, writing simply
148: $u \in L^2(M);$ similarly, all operators will tacitly be semiclassical
149: families of operators, operating on half-densities. The hypothesis that
150: our operators act on half-densities ensures that if $A =\Op_h(a)$ with
151: $a(x,\xi;h) \sim a_0(x,\xi) + h a_1(x,\xi) +\dots,$ the terms $a_0$
152: (principal symbol) and $a_1$ (subprincipal symbol) are both invariantly
153: defined as functions on $T^*M$ (see \cite{Evans-Zworski}).
154:
155: Furthermore, we will deal with an operator $P$ rather than $P-\lambda,$
156: absorbing the constant term into the definition of the operator.
157:
158: We begin by defining a notion of Lagrangian regularity of a family of
159: functions along $\lag,$ following the treatment of the ``homogeneous'' case
160: in \cite{Hormander:book4}.
161: \begin{definition}
162: Let $\md$ denote the module (over
163: $\Psih(M)$) of semiclassical pseudodifferential operators with symbols
164: vanishing on $\lag.$
165:
166: Let $q \in \lag,$ $k \in \NN,$ and $u \in L^2(M).$ We say that $u$ has
167: Lagrangian regularity of order $k$ at $q,$ and write $q \in \Slag^k(u),$ if
168: and only if there is a neighborhood $U$ of $q$ in $T^*M$ such that for all
169: $k'=0,1,\dots,k$ and all $A_1,\dots, A_{k'} \in \md$ with $\WF' A_j \subset
170: U,$ $h^{-k'} A_1\cdots A_{k'} u \in L^2(M).$
171: \end{definition}
172:
173: \begin{proposition}\label{prop:basis}
174: Fix $q \in \lag,$ and let $A_i$ ($i=1,\dots,n$) be a
175: collection of elements of $\md$ with $d\sigma(A_i)$ spanning $N^*_q\lag.$
176: We have
177: $$ q \in \Slag(u) \Longleftrightarrow h^{-k'} A_{i_1}\cdots A_{i_{k'}} u
178: \in L^2\quad \forall (i_1,\dots,i_{k'}) \in \{1,\dots,n\}^{k'},\ k'\leq k.$$
179: \end{proposition}
180: \begin{proof}
181: We begin with the case $k=1.$
182: Given any $B$ characteristic on $\lag$ and microsupported sufficiently
183: close to $q,$ we may factor $\sigma(B)=\sum c_i \sigma(A_i)$ by Taylor's
184: theorem. Thus, letting $C_i$ be operators with symbol $c_i,$ we obtain
185: $$
186: h^{-1} B u= \sum h^{-1} C_i A_i u + R u
187: $$
188: for some semiclassical operator $R,$ hence we obtain the desired estimate
189: on $h^{-1} Bu$ since $R$ is uniformly (in $h$) $L^2$-bounded.
190:
191: More generally, if $B_{\alpha_1},\dots, B_{i_k}$ is a $k$-tuple of operators
192: characteristic on $\lag,$ we have
193: $$
194: h^{-k} B_{i_1}\cdots B_{i_k} u= h^{-k}\prod_{j=1}^k (C_{i_j}
195: A_{i_j}+hR_{i_j}) u;
196: $$
197: We then obtain the desired estimate inductively, using the fact that each
198: commutator of the form $[C,A]$ or $[R,A]$ produces a further factor of $h.$
199: \end{proof}
200:
201: We note that it follows from the work of Alexandrova \cite{Alexandrova}
202: that $S_\lag=\lag$ if and only if we can actually write $u$ in the form of
203: an oscillatory integral
204: $$
205: \int a(x,\theta,h) e^{i \phi(x,\theta)/h} \, d\theta
206: $$ plus a term with semiclassical wavefront set away from $\lag;$ here
207: $\phi$ is a phase function parametrizing the Lagrangian $\lag$ in the sense
208: introduced by H\"ormander. This is the semiclassical analog of a central
209: result in the H\"ormander-Melrose theory of conic Lagrangian distributions
210: \cite[Chapter 25]{Hormander:book4}.
211:
212: We now observe that the analogue of H\"ormander's theorem on propagation of
213: singularities for operators of real principal type is easy to prove in our
214: setting.
215: \begin{theorema}\label{thm:propagation}
216: Let $P\in \Psi_h(M)$ have real principal symbol $p.$ Let $\lag\subset
217: \{p=0\}$ be a Lagrangian submanifold of $T^*M.$ Then $P u\in h^{k+1}
218: L^2(M)$ implies that $\Slag^k(u)$ is invariant under the Hamilton flow of
219: $p.$
220: \end{theorema}
221:
222: The author is grateful to M.~Zworski for suggesting the following brief proof.
223: \begin{proof}[Sketch]
224: By \cite[Theorem~21.1.6]{Hormander:book3}, there is a local
225: symplectomorphism taking $p$ to $\xi_1$ and $\lag$ to $\lag_0 \equiv
226: \{\xi=0\}.$ Following the development in \cite{Alexandrova}, we may
227: quantize this to a semiclassical FIO that conjugates $P$ to $h D_{x^1}$
228: modulo $O(h^\infty)$ (cf.\ \cite[Theorem~26.1.3]{Hormander:book4} in the
229: non-semiclassical setting). Lagrangian regularity along $\lag_0$ is
230: iterated regularity under $h^{-1} (hD_{x^i}),$ i.e.\ is just classical
231: Sobolev regularity, uniform in $h.$ The theorem thus reduces to the
232: statement that Sobolev regularity for solutions to $D_{x^1} u \in
233: h^{k}L^2(M)$ propagates along the lines $(x^1\in \RR,x' =\text{const}),$ which
234: is easily verified.
235: \end{proof}
236:
237: \section{Integrable flow}\label{section:flow}
238:
239: We continue to assume that $P \in \Psi_h(M)$ has real principal symbol. We
240: now further assume that $p=\sigma(P)$ has \emph{completely integrable}
241: bicharacteristic flow, i.e.\ that there exist functions $f_2,\dots,f_n$ on
242: $T^*M,$ Poisson commuting with $p$ and with each other, and with
243: $dp,df_2,\dots,df_n$ pointwise linearly independent. We again emphasize
244: that we in fact only require the $f_i$'s to exist in some open subset of
245: interest in $T^*M.$ Let $\Sigma$ denote the characteristic set in $T^*M.$
246: Let $(I_1,\dots, I_n,\theta_1,\dots, \theta_n)$ be action-angle variables
247: and let $\omega_i = \partial p/\partial I_i$ be the frequencies. We also
248: let $\omega_{ij} = \pa^2 I/\pa I_i \pa I_j.$ (We refer the reader to
249: \cite{Arnold} for an account of the theory of integrable systems, and in
250: particular for a treatment of action-angle variables.)
251:
252: Let $\lag \subset \Sigma$ be a \emph{rational} invariant torus,
253: i.e.\ one on which $\omega_i/\omega_j \in \QQ$ for all $i,j=1,\dots,n.$ We
254: further assume that $\lag$ is \emph{nondegenerate} in the following sense:
255: we assume that the matrix
256: \begin{equation}\label{bigmatrix}
257: \begin{pmatrix}
258: \omega_{11}& \dots & \omega_{1n} & \omega_1 \\
259: \vdots & \ddots & \vdots & \vdots\\
260: \omega_{n1} & \dots & \omega_{nn} & \omega_n\\
261: \omega_{1} & \dots & \omega_{n} & 0\\
262: \end{pmatrix}
263: \end{equation}
264: is invertible on $\lag.$ This is precisely the condition of
265: \emph{isoenergetic nondegeneracy} often used in KAM theory (see
266: \cite{Arnold}, Appendix 8D). It is easy to verify that the condition is
267: equivalent to the condition that the map from the energy surface to the
268: projectivization of the frequencies
269: $$
270: \{p=0\} \ni I \mapsto [\omega_1(I):\dots: \omega_n(I)] \in \mathbb{RP}^n
271: $$
272: be a local diffeomorphism.
273:
274: For later convenience, we introduce special notation for the frequencies
275: and their derivatives on $\lag:$ we let
276: $$\ombar_i =\omega|_{\lag},\quad \ombar_{ij}
277: =\omega_{ij}|_{\lag}.$$
278:
279: On $\lag,$ we of course only know from Theorem~A that $\Slag^k(u)$ is a union
280: of orbits of $H_p,$ which, being rational, are not dense in $\lag.$
281: There are, however, further constraints on $\Slag^k(u).$
282:
283: \begin{definition}
284: An \emph{annular neighborhood} of a closed orbit $\orbit$ is an open set
285: $U=V\backslash K\subset \lag$ such that $\orbit \subset K \subset V$ with
286: $K$ compact and $V$ open in $\lag.$
287: \end{definition}
288:
289: We can now state our main result.
290: \begin{theoremb}
291: Suppose $P u=f \in h^{k+1} L^2.$ Let $\orbit$ be a null bicharacteristic
292: for $p$ on the rational invariant torus $\lag.$ If a small enough annular
293: neighborhood of $\orbit$ is in $\Slag^k (u),$ then so is $\orbit.$
294:
295: The meaning of ``small enough'' depends only on the $\ombar_i$'s.
296:
297: If $n=2$ then either $\Slag (u) = \lag$ or $\Slag(u) = \emptyset.$
298: \end{theoremb}
299: Thus, conormal regularity propagates ``diffusively'' to fill in annular
300: neighborhoods.
301:
302:
303: \begin{example}
304: Horozov \cite{Horozov1,Horozov2} has studied the \emph{spherical pendulum},
305: i.e.\ the system on $T^*S^2$ with Hamiltonian $h=(1/2)\abs{\xi}^2+x_3$ on
306: $T^*S^2$ (with $x_3$ one of the Euclidean coordinates on $S^2
307: \subset\RR^3$). Integrals of motion are $h$ and $p_\theta,$ the angular
308: momentum. Horozov showed that when $h\in (-1,1] \cup
309: [7/\sqrt{17},\infty),$ all values of $p_\theta$ lead to isoenergetically
310: nondegenerate invariant tori, while for $h \in (1,7/\sqrt{17}),$ there are
311: exactly two values of $p_\theta$ for which isoenergetic nondegeneracy
312: fails. Thus our results show that if we consider quasimodes for the
313: operator
314: $$
315: P_h = (1/2) h^2\Lap_{S^2} +x_3
316: $$ then for any torus $\lag$ not associated to the one of the exceptional
317: pairs of $(h,p_\theta)$ identified by Horozov, either $S_\lag=\lag$ or
318: $S_\lag=\emptyset.$
319: \end{example}
320:
321: \begin{example}
322: We now illustrate with an example the necessity of the isoenergetic
323: nondegeneracy condition. As in the introduction, let $M=S^1 \times S^1,$
324: but now let $P=hD_x;$ it is easy to verify that \emph{no} Lagrangian torus
325: is isoenergetically nondegenerate in this case. Let $\lag =
326: \{\xi=\eta=0\},$ the zero-section of $T^*M.$ Lagrangian regularity in this
327: setting is, as noted above, just Sobolev regularity, uniform in $h.$
328:
329: Let $\psi(y)$ be a bump function supported near $y=0.$ Then $$u(x,y) =
330: e^{i\psi(y)/\sqrt{h}}$$ has wavefront set only in $\lag.$ It is manifestly
331: Lagrangian on the complement of $\supp\psi,$ which forms an annular
332: neighborhood of the orbit $\{x\in S^1,y =0, \xi=\eta=0\}\subset \supp
333: \psi.$ It is not Lagrangian, however, on $\supp \psi,$ as it lacks iterated
334: regularity under $h^{-1}(h D_y).$
335:
336: \end{example}
337: \section{Symbol Construction}
338: By shifting coordinates, we may assume that $\orbit$ is the orbit passing
339: through $\{\theta=0\}.$
340:
341: For each $i,j$ let
342: $$ \gamma_{ij}= \widetilde{\min}_{k,l \in \ZZ} ((\theta_i+2\pi k)
343: \ombar_j-(\theta_j+2\pi l) \ombar_i),
344: $$ where $\widetilde{\min}$ denotes the value with the smallest norm, i.e.\
345: may be positive or negative. Each $\gamma_{ij}$ then takes values in an
346: interval determined by $\ombar_i,$ $\ombar_j,$ and is smooth where it takes
347: on values in the interior of the interval. (If $\ombar_i=p/q$ and
348: $\ombar_j=p'/q'$ then $\gamma_{ij}$ takes values in $[-\pi a, \pi a]$ where
349: $a=\gcd(qp',pq')/q q'.$) The ``small enough'' condition in the statement of
350: Theorem~B is just the following: each $\gamma_{ij}$ should be smooth on the
351: annular neighborhood of $\orbit$ where we assume Lagrangian regularity.
352:
353: Note that $\gamma_{ij}(\theta)=0$ for all $i,j$ exactly when there exists
354: $\tilde{\theta} \in \RR^n,$ equivalent to $\theta$ modulo $2\pi\ZZ^n,$ such
355: that $[\tilde\theta_1: \dots \tilde\theta_n] = [\ombar_1:\dots:\ombar_n].$
356: Thus the functions $\gamma_{ij}$ define $\orbit$ on $\lag:$ we have
357: $\{I=\Ibar,\ \gamma_{ij} = 0\ \forall i,j\} = \orbit;$ indeed, the
358: vanishing of each $\gamma_{i,i+1}$ and of $\gamma_{n,1}$ suffices to define
359: $\orbit,$ and these $n$ functions may be taken as coordinates on $\lag$ in
360: a neighborhood of $\orbit.$ The central point of our argument will be that
361: the $\gamma_{ij}$ are ``propagating variables'' with derivatives along the
362: flow that, taken together, will suffice to give Lagrangian regularity.
363:
364: Since the $\gamma_{kl}$ define $\orbit$ and are smooth on the annular
365: neighborhood $U$ where we have assumed regularity, there is a smooth cutoff
366: function $$\psi
367: :=\psi(\gamma_{12,},\gamma_{23},\dots,\gamma_{n-1,n},\gamma_{n,1})$$ with
368: $\psi=1$ on $\orbit$ and $\nabla \psi$ having its support on $\lag$
369: contained in $U.$ We may also arrange for $\psi$ to be the square of a
370: smooth function. Let $\phi_\ep$ be a cutoff supported in $[-\ep, \ep],$
371: with smooth square root.
372:
373: Let
374: \begin{equation}\label{adef}
375: a_{ij}(x) = \psi \cdot \phi_\ep(\abs{I-\Ibar})\cdot \gamma_{ij}\cdot (\omega_i \ombar_j -\omega_j \ombar_i)
376: \end{equation}
377:
378: We compute first that, where $\gamma_{ij} \in \CI,$
379: \begin{equation}\label{pgamma}
380: \{p,\gamma_{ij}\} = (\omega_i\ombar_j-\omega_j\ombar_i)
381: \end{equation}
382: (since $\gamma_{ij}$ is locally given by expressions of the form
383: $((\theta_i+2\pi k) \ombar_j-(\theta_j+2\pi l) \ombar_i)$ with $k,l$
384: \emph{fixed}) and hence that
385: $$
386: \{p,a_{ij}\}= \{p,\psi\} \phi_\ep (\abs{I-\Ibar}) \gamma_{ij} (\omega_i \ombar_j -\omega_j \ombar_i)+
387: \psi \phi_\ep (\abs{I-\Ibar}) (\omega_i\ombar_j-\omega_j\ombar_i)^2.
388: $$
389: We further note that as $\psi$ is a function of the $\gamma_{ij}$'s, by
390: \eqref{pgamma} the first term in this expression is a sum of
391: terms divisible by $$(\omega_{k_1} \ombar_{l_1} -\omega_{l_1} \ombar_{k_1})(\omega_{k_2} \ombar_{l_2} -\omega_{l_2} \ombar_{k_2})$$ for various
392: $k_i,l_i.$ Thus we may write
393: \begin{equation}\label{bracket}
394: \{p,a_{ij}\}= \sum e_k f_l+
395: \psi \phi_\ep (\abs{I-\Ibar}) (\omega_i\ombar_j-\omega_j\ombar_i)^2,
396: \end{equation}
397: where each $e_k$ and $f_l$ vanishes on $\lag$ and with support
398: intersecting $\lag$ only in $U.$
399:
400: We will also employ a symbol that is \emph{invariant} under the flow:
401: for each $j=1,\dots, n,$ set
402: $$
403: w_j = \phi_\ep(\abs{I-\Ibar}) I_j.
404: $$
405:
406: \section{Nondegeneracy}
407: Using a positive commutator argument, we will find that we can control
408: operators whose symbols are multiples of
409: $(\omega_i\ombar_j-\omega_j\ombar_i).$ These quantities vanish on $\lag,$
410: but our nondegeneracy hypothesis permits us to use them to control
411: Lagrangian regularity on $\lag.$ To see this, rewrite
412: $$
413: (\omega_i\ombar_j-\omega_j\ombar_i) = (\omega_i-\ombar_i)\ombar_j-(\omega_j
414: -\ombar_j)\ombar_i$$
415: and expand about $\lag$ in the $I$ variables, to rewrite this as
416: $$
417: \sum_k (\ombar_{ik}\ombar_j-\ombar_{jk}
418: \ombar_i) (I_k-\Ibar_k)+ O((I-\Ibar)^2)
419: $$
420:
421: We now prove a key algebraic lemma:
422: \begin{lemma}\label{lemma:nondeg}
423: Let $v_1,\dots, v_n,$ and $v_{ij},$ $i,j=1,\dots,n$ be real numbers, with
424: $v_{ij}=v_{ji}.$ The functionals $\alpha_{ij}(x) = \sum_k(v_{ik}v_j-v_{jk}
425: v_i)x_k$ (for $i,j=1,\dots,n$) together with the covector $(v_1,\dots,v_n)$
426: span $(R^n)^*$ iff the matrix
427: \begin{equation}\label{thematrix}
428: \begin{pmatrix}
429: v_{11}& \dots & v_{1n} & v_1 \\
430: \vdots & \ddots & \vdots & \vdots\\
431: v_{n1} & \dots & v_{nn} & v_n\\
432: v_{1} & \dots & v_{n} & 0\\
433: \end{pmatrix}
434: \end{equation}
435: is nondegenerate.
436: \end{lemma}
437: \begin{proof}[Proof of Lemma]
438: We may assume that not all of the $v_i$'s are zero, as the result is
439: trivial in that case.
440:
441: Let
442: $$
443: \vec\zeta_{ij} = \begin{pmatrix}
444: v_{1i}v_j-v_{1j}v_i\\
445: \vdots\\
446: v_{ni}v_j-v_{nj}v_i
447: \end{pmatrix}.
448: $$
449: Letting $A$ be the matrix with entries $v_{ij}$ and
450: $$
451: \vec u_{ij} = v_j e_i-v_i e_j
452: $$
453: where $e_i$ is the standard basis for $\RR^n,$
454: we have
455: $$
456: \vec\zeta_{ij} = A \vec u_{ij}.
457: $$
458: Let $U$ denote the span of the $\vec u_{ij}$'s. Thus,
459: $$
460: U^\perp = \bigcap_{i,j} \vec u_{ij}^\perp = \bigcap_{i,j} \RR \cdot \{\vec{w}
461: \in \RR^n| [w_i:w_j]=[v_i:v_j]\ \forall i,j\} = \RR \vec v
462: $$ where $\vec v = (v_1,\dots v_n)^t.$ Thus, $U = \vec v^\perp.$ Hence the
463: span of the $\vec \zeta_{ij}$ is of $A (\vec v^\perp).$ The assertion of
464: the lemma is then that $A(\vec v^\perp)$ and $\vec v$ are complementary iff
465: the matrix \eqref{thematrix} is nondegenerate. This equivalence follows
466: from the observation that
467: $$
468: \begin{pmatrix}
469: A & \vec v \\
470: \vec v^t & 0 \\
471: \end{pmatrix} \cdot
472: \begin{pmatrix}
473: \vec w \\ z
474: \end{pmatrix}
475: =
476: \begin{pmatrix}
477: A \vec w +z \vec v \\ \ang{\vec v,\vec w},
478: \end{pmatrix}
479: $$ hence \eqref{thematrix} has nontrivial nullspace iff there exists a
480: nonzero $\vec w \in \vec v^\perp$ with $A \vec w \in \RR \vec v.$
481: \end{proof}
482:
483: \section{Proof of Theorem B}
484:
485: We note, first of all, that in the special case when $n=2,$ a neighborhood
486: of any closed orbit $\orbit'\neq \orbit$ is itself an annular neighborhood
487: of $\orbit.$ Hence the special result for $n=2$ follows directly from the
488: general one.
489:
490: We now prove Theorem B by induction on $k;$ we suppose it true for $k\leq
491: K-1$ (and note that for $k=0$ it is vacuous).
492:
493: Let $A_{ij} \in \Psih(M)$ be self-adjoint, with symbol $a_{ij}$ constructed
494: above and vanishing subprincipal symbol. Then we have by \eqref{bracket},
495: \begin{equation}\label{comm1} ih^{-3} [P, A_{ij}] = h^{-2} B_{ij}^2 + \sum_{k,l}
496: h^{-2}E_k F_l + R
497: \end{equation} with $B_{ij}$ self-adjoint with vanishing
498: subprincipal symbols, and
499: \begin{equation}\label{bsymbol}
500: \sigma(B_{ij})=b_{ij} =({\psi}{\phi_\ep} (\abs{I-\Ibar}))^{1/2}\cdot
501: (\omega_i\ombar_j-\omega_j\ombar_i),
502: \end{equation}
503: and with $E_k,$ $F_l$ characteristic on $\lag$ with the supports of
504: $\sigma(E_k),$ $\sigma(F_l)$ intersecting $\lag$ only in $U.$ ($R,$ $E_k,$
505: and $F_l$ of course depend on $i,j$ but we suppress these extra indices.)
506:
507: Let $W_j$ have symbol $w_j$ constructed above, and be self-adjoint with
508: vanishing subprincipal symbol. Then
509: $$
510: i h^{-3}[P, W_j] \in \Psi_h(M).
511: $$
512: For a multi-index $\alpha$ with $\abs{\alpha} = K-1,$
513: set
514: $$
515: Q_{ij}=A_{ij} W_1^{2\alpha_1}\dots W_n^{2\alpha_n},
516: $$
517: We will also need the operator denoted in multi-index notation
518: $$
519: W^\alpha = W_1^{\alpha_1}\dots W_n^{\alpha_{K-1}}
520: $$
521:
522: Now we examine
523: \begin{equation}
524: i h^{-2K-1} \ang{ (P^* Q_{ij}-Q_{ij}P u, u} = i
525: h^{-2K-1} (\ang{Q_{ij}u,f}-\ang{f,Q_{ij}^* u}).
526: \end{equation}
527: For any $\delta>0,$ we may estimate the RHS by
528: $$
529: C_\delta\norm{h^{-K-1} f}^2+\delta(\norm{h^{-K} Q_{ij}u}^2+\norm{h^{-K} Q_{ij}^*u}^2).
530: $$ Note that both $Q_{ij}$ and $Q_{ij}^*$ are $(2K+1)$-fold products of
531: operators vanishing on $M,$ and that each contains the factors $A_{ij}$ and
532: $W^\alpha.$ By \eqref{adef} and \eqref{bsymbol}, $\sigma(A_{ij})$ is
533: divisible by $\sigma(B_{ij});$ thus, by elliptic regularity we may estimate
534: the RHS by
535: $$
536: C_\delta\norm{h^{-K-1} f}^2+C \delta\sum_{\alpha} \norm{h^{-K} B_{ij} W^\alpha u}^2+
537: \sum_{j=0}^{2K-2} h^{-j} \ang{D_j u,u}
538: $$ where $C$ is independent of $\delta,$ and each $D_j$ is a sum of
539: products of $j$ elements of $\md,$ all microsupported on $\supp A_{ij};$
540: these arise from commutator terms in which we have reordered products of
541: elements of $\md.$
542:
543: Now we recall that $P^*-P \in h^2 \Psi_h(M),$ hence, by \eqref{comm1} we
544: may write
545: \begin{multline}\label{LHS}
546: \ang{ih^{-2K-1} (P^* Q_{ij}-Q_{ij}P)u,u} \\= \norm{h^{-K} B_{ij} W^\alpha
547: u}^2
548: + \sum_{k,l}\ang{h^{-K} E_k W^\alpha u, h^{-K} F_l W^\alpha u}
549: +\sum_{j=0}^{2K-2} h^{-j} \ang{\tilde{D}_j u,u}
550: \end{multline}
551: with the $\tilde{D}_j$ sharing the properties of the $D_j$ above.
552:
553: Putting together the information from our commutator, we now have, for all $\delta>0,$
554: \begin{multline}\label{ijestimate}
555: (1-C\delta)\norm{h^{-2K} B_{ij} W^\alpha u}^2 \\ \leq C_\delta\norm{h^{-K-1} f}^2+
556: \sum_{k,l}(\norm{h^{-K} E_k W^\alpha u}^2+\norm{h^{-K} F_l W^\alpha u}^2)
557: +\sum_{j=0}^{2K-2} h^{-j} \abs{\ang{\tilde{\tilde{D}}_j u,u}},
558: \end{multline}
559: with the $\tilde{\tilde{D}}_j$'s satisfying the same properties as $D_j$
560: above. Each of the $E_k$ and $F_l$ terms is controlled by our hypothesis
561: on $\Slag^k u,$ while the $\tilde{\tilde{D}}_j$ terms are bounded by the
562: inductive assumption.
563:
564: Now we use our nondegeneracy hypothesis as reflected in
565: Lemma~\ref{lemma:nondeg}. Recall that $\lag \subset \Sigma,$ hence the
566: operator $P$ is characteristic on $\lag;$ moreover, we have $dp|_{\lag} =
567: \sum \ombar_k dI_k,$ hence Lemma~\ref{lemma:nondeg} tells us that $P$ and
568: $B_{ij},$ for $i,j=1,\dots,n,$ are a collection of operators fulfilling the
569: hypotheses of Proposition~\ref{prop:basis}. Thus, adding together
570: equations \eqref{ijestimate} for all possible values of $i,j,$ and
571: multi-index $\alpha,$ together with terms involving $P$ rather than
572: $B_{ij}$ (which vanish up to commutators of $P$ with $W$'s), we obtain the
573: desired estimate, by Proposition~\ref{prop:basis}.\qed
574:
575: \begin{thebibliography}{X}
576: \bibitem{Alexandrova} I. Alexandrova, \emph{Semi-Classical Wavefront Set
577: and Fourier Integral Operators,} Can. J. Math, to appear.
578:
579: \bibitem{Arnold} V. I. Arnol'd, \emph{Mathematical methods of classical
580: mechanics,} second edition, Graduate Texts in Mathematics, 60
581: Springer-Verlag, New York, 1989.
582:
583: \bibitem{BHW1} N. Burq, A. Hassell, and J. Wunsch, \emph{Spreading of
584: quasimodes in the Bunimovich stadium,} Proc. AMS, to appear.
585:
586: \bibitem{Burq-Zworski1} N. Burq, M. Zworski, \emph{Control in the presence of a black box}, J. Amer. Math. Soc. 17 (2004), 443--471.
587:
588: \bibitem{Burq-Zworski2} N. Burq, M. Zworski, \emph{Bouncing Ball Modes and Quantum Chaos}, SIAM Review \textbf{47} (2005), no.~1, 43--49.
589:
590: \bibitem{Evans-Zworski}
591: L.C. Evans and M. Zworski, \emph{Lectures on semiclassical analysis,}
592: preprint, available at \begin{verbatim}www.math.berkeley.edu/~zworski.\end{verbatim}
593:
594: \bibitem{Hormander:Propagation} L. H{\"o}rmander,
595: \emph{On the existence and the regularity of solutions of linear
596: pseudo-differential equations,} Enseignement Math. (2),
597: \textbf{17}, 1971, 99--163.
598:
599: \bibitem{Hormander:book3} L. H\"ormander, \emph{The analysis of linear
600: partial differential operators. III. Pseudodifferential operators,}
601: Grundlehren der Mathematischen Wissenschaften, 274. Springer-Verlag,
602: Berlin, 1985.
603:
604: \bibitem{Hormander:book4} L. H\"ormander, \emph{The analysis of linear
605: partial differential operators. IV. Fourier integral operators,}
606: Grundlehren der Mathematischen Wissenschaften, 275. Springer-Verlag,
607: Berlin, 1985.
608:
609:
610: \bibitem{Horozov1} E. Horozov, \emph{Perturbations of the spherical
611: pendulum and abelian integrals} J. Reine Angew. Math. 408 (1990), 114--135.
612:
613: \bibitem{Horozov2} E. Horozov,
614: \emph{On the isoenergetical nondegeneracy of the spherical pendulum},
615: Phys. Lett. A 173 (1993), no. 3, 279--283.
616:
617: \bibitem{Martinez}
618: A. Martinez,
619: \emph{An introduction to semiclassical and microlocal analysis,}
620: Springer-Verlag, New York, 2002.
621: \end{thebibliography}
622:
623: \end{document}
624: