math0501101/sec2.tex
1: \section{The thin-slab propagator. Regularity properties.}
2: \label{sec:2}
3: We follow the terminology introduced in \cite[Sections
4: 25.4-5]{Hoermander:V4} for FIOs with complex phase.  Let $z',z \in
5: [0,Z]$ with $z'\geq z$ and let $\Delta:= z'-z$. Define $\phi_{(z',z)}
6: \in \Cinf (X'\times X \times \R^n)$ as
7: \begin{multline}
8:   \label{eq: phase function}
9:   \phi_{(z',z)}(x',x,\xi):= \inp{x'-x}{\xi}
10:   +i \Delta a_1(z,x',\xi)\\
11:   = \inp{x'-x}{\xi}  + \Delta b_1(z,x',\xi) +i
12:   \Delta c_1(z,x',\xi).
13: \end{multline}
14: \begin{remark}
15:   The function $\phi_{(z',z)}$ is assumed to be homogeneous of degree
16:   1 only when $|\xi| \geq 1$. This however is not an obstacle to the
17:   subsequent analysis, e.g., FIO properties, since to define such
18:   operators the phase function need not be homogeneous of degree 1 for
19:   small $|\xi|$. In the subsequent results concerning the phase
20:   function and FIOs one will then assume that $|\xi|$ is large enough,
21:   i.e., $|\xi| \geq 1$.
22: \end{remark}
23: %%%%%%%%%%%%%%%%%%%%%%%%
24: % lemma                %
25: %%%%%%%%%%%%%%%%%%%%%%%%
26: \begin{lemma}
27:   $\phi_{(z',z)}$ is a non-degenerate complex phase function of
28:   positive type (at any point $(x_0',x_0,\xi_0)$ where $\d_\xi
29:   \phi_{(z',z)} =0$).
30: \end{lemma}
31: \begin{proof} 
32:   Note that, by Assumption~\ref{assumpt:general assumption},
33:   $\Im(\phi_{(z',z)}) \geq 0$ and $\phi$ is homogeneous of degree 1;
34:   $\d_x \phi=0$ implies $\xi=0$. Thus, $\phi$ is a phase function of
35:   positive type. Inspecting the partial derivatives of $\d_\xi \phi$
36:   \wrt $x$ we conclude that the differentials
37:   $d(\d_{\xi_1}\phi), \dots, d(\d_{\xi_{n}}\phi)$ are linearly
38:   independent.
39:  \end{proof}
40:  
41: 
42:  With $a_0(z,.) \in S^0(X\times\R^{n})$ we have
43:  $\exp[-\Delta a_0(z,.)] \in S^0(X \times\R^{n})$ by Lemma
44:  18.1.10 in \cite{Hoermander:V3}. We define
45:  \begin{align}
46:    \label{eq: amplitude g}
47:    g_{(z',z)}(x,\xi) := \exp[-\Delta a_0(z,x,\xi)].
48:  \end{align}
49:  We shall keep this notation (for this symbol and others in the
50:  sequel) but it will be useful however to consider this symbol to
51:  depend on the parameters $z$ and $\Delta$ instead of $z$ and $z'$ in
52:  the following analysis. Note that $g_{(z',z)}$ is bounded \wrt $z$
53:  and $\Cinf$ \wrt $\Delta$ with values in $S^0(X \times\R^{n})$.
54:  Hence, we may define a distribution kernel $G_{(z',z)}(x',x) \in
55:  \D'(X'\times X)$
56: \begin{multline*}
57:   G_{(z',z)}(x',x) = \int \exp[i \inp{x'-x}{\xi} ]
58:   \exp[ - \Delta a(z,x',\xi)]\, \dslash \xi\\
59:   = \int \exp[i \phi_{(z',z)}(x',x,\xi)]\: 
60:   g_{(z',z)}(x',\xi) \, \dslash \xi
61: \end{multline*}
62: as an oscillatory integral. We denote the associated operator by
63: $\G_{(z',z)}$.  This operator is often referred to as the {\em thin-slab
64: propagator} (see e.g.~\cite{dHlRW:00,dHlRB:03}). We
65: show that $\G_{(z',z)}$ is a global FIO in $\R^n$.
66: 
67: Define $\alpha:=(x',x,\xi',\xi)$ and
68: \begin{align*}
69:   u_{\theta_j}(\alpha, \theta)
70:   &= \d_{x_j} \phi_{(z',z)}(x',x,\theta) + \xi_j = - \theta_j + \xi_j,\\
71:   u_{\xi_j}(\alpha, \theta)
72:   &= \d_{x_j'} \phi_{(z',z)}(x',x,\theta) - \xi_j'
73:   = \theta_j - \xi_j' + i \Delta \d_{x_j} a_1(z,x',\theta), \\
74:   u_{x_j}(\alpha, \theta)
75:   &= \d_{\theta_j}\phi_{(z',z)}(x',x,\theta)
76:   = x_j' - x_j +  i \Delta \d_{\xi_j} a_1(z,x',\theta),
77: \end{align*}
78: where $j=1,\dots,n$. We denote by $\hat{J}_{(z',z)}$ the ideal in
79: $\Cinf(\R^{5n})$ generated by the functions $u_{\theta_j},u_{\xi_j}$, and
80: $u_{x_j}$, and we let $J_{(z',z)}$ be the subset of the functions in
81: $\hat{J}_{(z',z)}$ that are independent of $\theta$.
82: %%%%%%%%%%%%%%%%%%%%%%%%
83: % lemma                %
84: %%%%%%%%%%%%%%%%%%%%%%%%
85: \addtocounter{delta}{1}
86: \begin{lemma}
87:   \label{lem:generators of  J_(z'z)R}
88:   There exists $\Delta_\thedelta > 0$, such that, for all $z',z \in
89:   [0,Z]$, with $z'>z$ and $\Delta=z'-z\leq
90:   \Delta_\thedelta$, the ideal $J_{(z',z)}$ is generated by the
91:   functions
92:   \begin{align}
93:     v_{\xi_j}(\alpha)
94:     &= \d_{x_j'} \phi_{(z',z)}(x',x,\xi) - \xi_j'
95:     \label{eq: generators of J_z'z},\\
96:     &= \xi_j - \xi_j' + i \Delta \d_{x_j} a_1(z,x',\xi)
97:     = u_{\xi_j} |_{\theta = \xi}\nonumber,\\
98:     v_{x_j}(\alpha) &=\d_{\xi_j}\phi_{(z',z)}(x',x,\xi) = x_j' - x_j +
99:     i \Delta \d_{\xi_j} a_1(z,x',\xi) = u_{x_j} |_{\theta = \xi}
100:     \nonumber
101:   \end{align}
102:   $j=1,\dots,n$.
103: \end{lemma}
104: Some of the key arguments of the proof are close to that in the proof
105: of Theorem 25.4.4 in \cite{Hoermander:V4}.
106: \begin{proof}
107:   The ideal $\hat{J}_{(z',z)}$ is also generated by the functions
108:   \[
109:   u_{\theta_j}, \ \ \tilde{u}_{\xi_j}:=u_{\theta_j} + u_{\xi_j}= \xi_j - \xi_j' + i
110:   \Delta \d_{x_j} a_1(z,x',\theta), \ \ u_{x_j},
111:   \]
112:   $j=1,\dots,n$. We define $\nu:=(x',\xi',\theta)$, $\mu:=(x,\xi)$. We
113:   set a point $(\nu_0, \mu_0) = (x_0',\xi_0',\theta_0, x_0,\xi_0)$
114:   where these generators vanish and we work in a neighborhood of this
115:   point.  (Note that $\theta_0=\xi_0$.)  Since $z \mapsto a_1(z,.) \in
116:   S^1(X\times \R^n)$ is bounded we have that $\exists \Delta_\thedelta
117:   > 0$ such that for $0\leq \Delta \leq \Delta_\thedelta$, and all $z
118:   \in [0,Z]$,
119:   \begin{align*}
120:   \det \d \left(u_{\theta_1},\dots ,u_{\theta_{n}},\tilde{u}_{\xi_1},
121:     \dots, \tilde{u}_{\xi_{n}},
122:     u_{x_1},\dots,u_{x_{n}}\right)/\d \nu \neq
123:   0.
124:   \end{align*}
125:   By Theorem 7.5.7 in \cite{Hoermander:V1} we have
126:   \begin{eqnarray*}
127:     \left(
128:       \begin{array}{c}
129:         x'-x\\
130:         \xi' -\xi\\
131:         \theta
132:       \end{array}
133:     \right)
134:     =
135:     \left(
136:       \begin{array}{cc}
137:         Q(\nu,\mu) & P(\nu, \mu)\\
138:         0 & I_n
139:       \end{array}
140:     \right)
141:     \left(
142:       \begin{array}{c}
143:         u_{x}\\
144:         \tilde{u}_\xi\\
145:         -u_\theta
146:       \end{array}
147:     \right)
148:     +\left(
149:       \begin{array}{c}
150:         \tilde{x}(\mu)\\
151:         \tilde{\xi}(\mu)\\
152:         \xi
153:       \end{array}
154:     \right)
155:     \end{eqnarray*}
156:     where $P$ is a $\Con^\infty$ $2n\times n$ matrix and $Q$ is a
157:     $\Con^\infty$ $2n\times 2n$ matrix and the functions $\tilde{x}$,
158:     $\tilde{\xi}$ are also $\Con^\infty$ in a neighborhood of
159:     $(\nu_0,\mu_0)$. As the functions $w_x(\nu,\mu)
160:     :=x'-x-\tilde{x}(\mu)$, $w_\xi(\nu,\mu)
161:     :=\xi'-\xi-\tilde{\xi}(\mu)$, $w_\theta(\nu,\mu) := \theta - \xi$
162:     have linearly independent differentials, Lemma 7.5.8 in
163:     \cite{Hoermander:V1} proves that they generate $\hat{J}_{(z',z)}$
164:     and the proof of that lemma shows that $Q$ is invertible in a
165:     neighborhood of $(\nu_0,\mu_0)$.  Letting $\theta=\xi$ we have
166:     \begin{eqnarray*}
167:       \hspace*{1cm}
168:       Q(x',x,\xi,\mu)^{-1}\ \left(
169:       \begin{array}{c}
170:         w_x(\nu,\mu)\\
171:         w_\xi(\nu,\mu)
172:         \end{array}
173:     \right)
174:     =
175:     \left(
176:       \begin{array}{c}
177:         u_{x}(x',x,\xi)\\
178:        \tilde{u}_{\xi}(x',x,\xi)
179:       \end{array}
180:     \right)
181:     =\left(
182:       \begin{array}{c}
183:         v_x(\alpha)\\
184:         v_\xi(\alpha)
185:       \end{array}
186:     \right).
187:     \end{eqnarray*}
188:     We thus obtained that $\hat{J}_{(z',z)}$ is generated by the functions
189:     $u_{\theta_j}$, $v_{x_j}$, $v_{\xi_j}$,
190:     $j=1,\dots,n$.  We then see that $J_{(z',z)}$ is generated by
191:     $v_{x_j}$, $v_{\xi_j}$, $j=1,\dots,n$.
192:     In fact, using Theorem 7.5.7 in
193:     \cite{Hoermander:V1} again, any $\Con^\infty$ function $h(\alpha)$ can
194:     be locally written in the form
195:     \begin{align*}
196:     h(\alpha) = \sum_{1\leq i \leq n} (h_{x_j}(\alpha',\mu) v_{x_j}(\alpha',\mu)
197:      + h_{\xi_j}(\alpha',\mu) v_{\xi_j}(\alpha',\mu))+ r(\mu),
198:     \end{align*}
199:     with $\alpha'=(x',\xi')$ provided that $0\leq \Delta\leq \Delta_\thedelta$.  If $h
200:     \in J_{(z',z)}$ then $r\in J_{(z',z)}$ and Lemma 7.5.10 in
201:     \cite{Hoermander:V1} implies that $\forall N \in \N$, $\exists C_N>0$:
202:     \begin{align*}
203:     r(\mu) \leq C_N \max (|\Im\ \tilde{x}(\mu)|,
204:     |\Im\ \tilde{\xi}(\mu)|)^N,
205:     \end{align*}
206:     locally. Therefore, Theorem 7.5.12 in \cite{Hoermander:V1} yields $r \in
207:     I(w_{x},w_{\xi}) = I(v_{x},
208:     v_{\xi})$; which in turn implies $g \in I(v_{x},
209:     v_{\xi})$ and thereby completes the proof.
210: \end{proof}
211: 
212: 
213: As the Poisson brackets (for the symplectic $2$-form $\sigma' -
214: \sigma$ on $T^\ast (X'\times X)$, where $\sigma'$ and $\sigma$ are the
215: symplectic $2$-forms on $T^\ast(X')$ and $T^\ast(X)$ respectively) of
216: any two of the functions in (\ref{eq: generators of J_z'z}) vanish
217: identically we obtain that the ideal generated by these functions is
218: globally a conic canonical ideal in the sense of \cite[Definition
219: 25.4.1. and Section 25.5]{Hoermander:V4}. The phase function
220: $\phi_{(z',z)}$ thus defines $J_{(z',z)}$ in the neighborhood of any
221: point of $J_{(z',z)\R}$: it thus {\em globally} defines $J_{(z',z)}$,
222: which is then of positive type.  Therefore the operator $\G_{(z',z)}$
223: is a global FIO with complex phase (see Definitions 25.4.9. and
224: 25.5.1. in \cite{Hoermander:V4}).
225: %%%%%%%%%%%%%%%%%%%%%%%%
226: % proposition          %
227: %%%%%%%%%%%%%%%%%%%%%%%%
228: \begin{proposition}
229:   There exists $\Delta_\thedelta>0$ such that if $0\leq
230:   \Delta=z'-z\leq \Delta_\thedelta$ then the operator $\G_{(z',z)}$ is a
231:   global Fourier integral operator with complex phase and $G_{(z',z)}
232:   \in I^0(X'\times X, (J_{(z',z)})', \Omega^{1/2}_{X'\times X})$.
233: \end{proposition}
234: We denote the half density bundle on $X'\times X$ by
235: $\Omega^{1/2}_{X'\times X}$.  Note that $(J_{(z',z)})'$ stands for the
236: twisted canonical ideal, i.e.  a Lagrangian ideal (see Section 25.5 in
237: \cite{Hoermander:V4}).
238: 
239: Note that, with the following analysis, we could also consider
240: $\G_{(z',z)}$ as a global FIO with real phase with amplitude in
241: $S^0_\hf (X'\times X \times \R^n)$ (see e.g. \cite{Ruzhansky:00}). However
242: such consideration would be rather technical as one usually restricts
243: oneself to the type $S^m_\rho$ with $\rho > \hf$ for FIOs (see the
244: remark at the end of Section 25.1 in \cite{Hoermander:V4}; see Also
245: \cite[pages 391-392]{MeSj:76}). Viewing the thin-slab propagator
246: $\G_{(z',z)}$ as a FIO with complex phase is also a good framework to
247: understand the propagation of singularities in Part II. We shall
248: however make this interpretation for $\G_{(z',z)}$ in
249: Proposition~\ref{prop: Hs cont any symbol}, below, to apply a result
250: of Kumano-go~\cite[Theorem 2.5]{Kumano-go:76}.
251: 
252: 
253: We now establish some global continuity properties of the operator
254: $\G_{(z',z)}$ stated in a slightly more general form (for similar
255: results with global symbols see for instance \cite{Kumano-go:76},
256: where phase functions are real and other conditions are imposed on the
257: phase function).
258: %%%%%%%%%%%%%%%%%%%%%%%%
259: % lemma                %
260: %%%%%%%%%%%%%%%%%%%%%%%%
261: % S -> S continuous.
262: \begin{lemma}
263: \label{lemma:S cont}
264: Let $A$ be an FIO with a kernel of the form
265: \begin{align*}
266: K_A(x,y) = \int \exp[i \varphi (x, \xi) - i \inp{y}{\xi}]\, \sigma_A(x,\xi)
267: \dslash \xi \in \D'(\R^n\times\R^n),
268: \end{align*}
269: where $\sigma_A \in S^m(\R^n\times \R^n)$ and $\varphi \in \Cinf(\R^n\times
270: \R^n)$ is such that $\Im (\varphi(x,\xi)) \geq 0$ and $\varphi$ is
271: homogeneous of degree 1 in $\xi$, for $|\xi|$ large enough, and $\d_{x_i}
272: \varphi \in S^1(\R^n\times \R^n)$. Furthermore, for all $i=1,\dots,n$ we assume
273: $\d_{\xi_i}\varphi(x,\xi) = x_i + f_i(x,\xi)$ where $f_i \in S^0(\R^n\times
274: \R^n)$. Then A maps $\S$ into $\S$ continuously.
275: \end{lemma}
276: \begin{proof}
277: Let $u \in \S$. We then have
278: \begin{align*}
279: |A u(x)| &\leq \int | \sigma_A(x,\xi) (1+ |\xi|)^{-m}| | (1+ |\xi|)^{m} \hat{u}(\xi)|
280: \,\dslash(\xi)\\
281: & \leq C \sup_{\xi \in \R^n} | \sigma_A(x,\xi) (1+ |\xi|)^{-m}| \sup_{\xi \in
282: \R^n} | (1+ |\xi|)^{m+n+1} \hat{u}(\xi)|,
283: \end{align*}
284: where $ C = \int (1+|\xi|)^{-n-1} \dslash \xi$. The operator $A$ is hence well
285: defined from $\S$ into $\Con(\R^n)$. If we differentiate we obtain
286: \begin{align*}
287: D_{x_i} A u(x) = \int \exp[i \varphi(x,\xi)] \left( \d_{x_i} \varphi (x,\xi)
288: \sigma_A(x,\xi) - i \d_{x_i} \sigma_A(x,\xi) \right) \hat{u}(\xi)\, \dslash
289: \xi\, .
290: \end{align*}
291: Noting that $\d_{x_i} \varphi (x,\xi) \sigma_A(x,\xi) - i \d_{x_i} \sigma_A(x,\xi)
292: \in S^{m+1}(\R^n\times \R^n)$ we similarly have
293: \begin{align*}
294: |D_{x_i}A u(x)| &\leq C \sup_{\xi \in \R^n} | (1+ |\xi|)^{m+n+2} \hat{u}(\xi)|\\
295: & \leq C' \sup_{x \in \R^n} |x^\alpha D^\beta_x u(x)| \, \text{ for some
296: }\alpha, \beta \geq 0.
297: \end{align*}
298: Iterating we find that $Au \in \Cinf(\R^n)$. Integrating by parts we also have
299: \begin{multline*}
300:  A(x_j u)(x) = \int \exp[i \varphi(x,\xi)] \left(
301: \d_{\xi_i} \varphi (x,\xi) \sigma_A(x,\xi) - i \d_{\xi_i} \sigma_A(x,\xi)
302: \right)
303: \hat{u}(\xi)\, \dslash(\xi)\\
304: = x_j A u(x) + \int \exp[i \varphi(x,\xi)] \left( f_i(x,\xi) \sigma_A(x,\xi) - i \d_{\xi_i} \sigma_A(x,\xi)
305: \right) \hat{u}(\xi)\, \dslash(\xi).
306: \end{multline*}
307: Since $f_i(x,\xi) \sigma_A(x,\xi) - i \d_{\xi_i} \sigma_A(x,\xi) \in
308: S^m(\R^n\times \R^n)$ we obtain
309: \begin{align*}
310:   |x_j A u(x)| \leq C \sup_{x \in \R^n} |x^\alpha D^\beta_x u(x)| +
311:   C\sup_{x \in \R^n} |x^{\alpha'} D^{\beta'}_x u(x)|,
312: \end{align*}
313: for some $\alpha, \alpha', \beta, \beta' \geq 0$. Similar estimates
314: hold for $|x^\alpha D_x^\beta A u(x)|$ because of the hypothesis made
315: on $f_i$, $i=1,\dots,n$. The operator $A$ thus maps $\S$ into $\S$
316: continuously.
317: \end{proof}
318: %%%%%%%%%%%%%%%%%%%%%%%%
319: % lemma                %
320: %%%%%%%%%%%%%%%%%%%%%%%%
321: % S' -> S' continuous.
322: To show continuity from $\S'$ into $\S'$ we shall need the following
323: lemma.
324: \begin{lemma}
325:   \label{lemma:7.7.1}
326:   Let $j,k$ non-negative integers, $u \in \S(\R^n)$, $f \in
327:   C^{k+1}(\R^n)$ such that
328:   \begin{align*}
329:     0\leq \Im\: f (x)\leq C_0,\ x\in \R^n, \  \ \  
330:     |f^{(r)}(x)| \leq C_r, \ x\in \R^n,  \ 1\leq r\leq k+1.
331:   \end{align*}
332: Then we have 
333: \begin{multline}
334:   \label{eq:7.7.1}
335:   \omega^{j+k} \left| \int u(x) (\Im\: f(x))^j \exp[i \omega f(x)]\ d x \right| \\
336:   \leq C \sum_{|\alpha| \leq k} \sup_{x \in \R^n} |D^\alpha u(x)|
337:   (|f'(x)|^2 + \Im\: f(x))^{|\alpha|/2 -k},\ \ \omega>0,
338: \end{multline}
339: where the constant $C$ is bounded when the function $f$ stays in a
340: domain of $\Con^{k+1}(\R^n)$ where $C_0$, $C_1, \dots, C_{k+1}$ can be
341: chosen bounded.
342: \end{lemma}
343: \begin{proof}
344:   The proof is the same as that of Theorem 7.7.1 in
345:   \cite{Hoermander:V1} where $u \in \Con^k_0(\R^n)$. In fact the
346:   further assumptions on $f$ made here allow to give global bounds
347:   that are needed since $u \in \S$ in the present case.
348: \end{proof}
349: \begin{lemma}
350: Let $A$ be an FIO with a kernel of the form:
351: \begin{align*}
352: K_A(x,y) = \int \exp[i \inp{x-y}{\xi}+i \gamma (x,\xi)]\, \sigma_A(x,\xi) \dslash(\xi)
353: \in \D'(\R^n\times\R^n),
354: \end{align*}
355: where $\sigma_A \in S^m(\R^n\times \R^n)$ and $\gamma \in S^1(\R^n\times
356: \R^n)$ is such that $\Im (\gamma(x,\xi)) \geq 0$, and $\gamma$ is homogeneous
357: of degree 1 in $\xi$, for $|\xi|$ large enough.  Furthermore, we assume that
358:  there exists $d\geq 0$ such that
359: \begin{eqnarray}
360: |\Re\left(\d_x \gamma(x,\xi)\right)| \leq d<1, \ x \in \R^n, \ 
361: \xi \in \R^n, \ |\xi|=1.
362: \label{eq:hyp phase}
363: \end{eqnarray}
364: Then $A$ maps $\S'$ into $\S'$ continuously. \label{lemma:S' cont}
365: \end{lemma}
366: Observe that the differential of $\phi(x,\xi):= \inp{x-y}{\xi}+ \gamma
367: (x,\xi)$ does not vanish in $\R^{2n}\times \R^n\backslash 0$. The
368: function $\phi$ is thus a complex phase function. The differentials
369: $d(\d_{\xi_1}\phi),\dots,d(\d_{\xi_n}\phi)$ are linearly independent.
370: Hence $\phi$ is a non degenerate complex phase function of positive
371: type.  Note that by~(\ref{eq:hyp phase}) the function $\inp{x-y}{\xi}
372: + \gamma(x,\xi)$ is an operator phase function in the sense of
373: \cite[Definition 1.4.4.]{Hoermander:71}.
374: \begin{proof}
375:   Without loss of generality we may assume that $\gamma$ is
376:   homogeneous of degree 1 for $|\xi|\geq 1$.  Let $A^t$ be the transpose
377:   of $A$ and let $u \in \S$, then.
378:   \begin{align*}
379:     A^t u (x) & = \int \exp[-i\inp{x}{\xi}]
380:     \int\exp[i \inp{y}{\xi} + i \gamma(y,\xi)]\,
381:     \sigma_A (y,\xi)\, u(y)\, d y \dslash \xi    \nonumber
382:   \end{align*}
383:   Define 
384:   \begin{align*}
385:     v(\xi,\eta) = \int \exp[i \inp{y}{\xi} + i \gamma(y,\xi)]\,
386:     \sigma_A (y,\eta)\, u(y)\, d y\, ,
387:     \nonumber
388:   \end{align*}
389:   and put $w(\xi) = v(\xi,\xi)$. As $u \in \S$ then $v$ and $w$ are
390:   both $\Cinf$. Then $A^t u$ is the Fourier transform of $w$. The
391:   lemma is proved if we show that $u \mapsto w(\xi)$ is continuous
392:   from $\S$ to $\S$.
393:   
394:   Let $\omega = |\xi|\geq 1$ and $\xi_0 = \xi/|\xi| \in \sph^{n-1}$.
395:   We then have $\inp{y}{\xi} + \gamma(y,\xi) = \omega f(y,\xi)$ with
396:   $f$ homogeneous of degree 0 in $\xi$, for $|\xi|\geq 1$. Note that $\d_y
397:   f(y,\xi) = \xi_0 + \d_y \gamma (y,\xi_0)$.  With the assumption made
398:   on $\d_y \gamma$ we have $|\d_y f(y,\xi)| \geq c >0$.  Applying
399:   Lemma~\ref{lemma:7.7.1} and estimate (\ref{eq:7.7.1}) we obtain
400:   \begin{align*}
401:     \omega^k |v(\xi,\eta)|
402:     &\leq K_k \sum_{|\alpha|\leq k}
403:     \sup | D^\alpha_y (\sigma_A(y,\eta) u(y))| \\
404:     & \leq K_k'(1 + |\eta|)^m
405:     \sup_{{|\alpha| \leq k} \atop {y \in \R^n}} | D^\alpha u (y)|, \ 
406:     \omega\geq 1
407:   \end{align*}
408:   where the constants $K_k$, $K_k'$ can be chosen uniformly \wrt
409:   $\xi$, $|\xi| \geq 1$ since the constants $C_0$, $C_1, \dots,
410:   C_{k+1}$ of Lemma~\ref{lemma:7.7.1} can be chosen bounded (as $\xi_0
411:   \in \sph^{n-1}$). Now setting $\eta=\xi$ we obtain that for all $k
412:   \in \N$, $\exists K_k''>0$
413:   \begin{align}
414:     (1+|\xi|)^{k-m} | w(\xi) |
415:     \leq K_k'' \!\sup_{{|\alpha| \leq k} \atop {y \in \R^n}}
416:     |D^\alpha u (y)|\, , \ \xi\in \R^n, \ |\xi|\geq 1.
417:     \label{eq: S estimate}
418:   \end{align}
419:   We now consider
420:   \begin{multline*}
421:     D_{\xi_i} w(\xi) = \int \exp[i \inp{y}{\xi} + i \gamma(y,\xi)]\\
422:     \big( (y_i + \d_{\xi_i} \gamma(y,\xi)) \sigma_A(y,\xi) - i \d_{\xi_i} \sigma_A(y,\xi)\big)
423:     u(y) d y\, .
424:   \end{multline*}
425:   As $y_i u(y) \in \S$ and $ \d_{\xi_i} \gamma(y,\xi)$ is homogeneous of
426:   degree 0 for $|\xi|\geq 1$ estimates similar to those in (\ref{eq: S
427:     estimate}) are valid.
428: \end{proof}
429: 
430: It is immediate from the structure of $\phi_{(z',z)}$ in (\ref{eq:
431:   phase function}) that Lemma \ref{lemma:S cont} applies to
432: $\G_{(z',z)}$. If $\Delta=z'-z$ is small enough we have $|\Delta
433: \partial_{x_i} b_1(z,x',\xi)|\leq d < 1$, due to
434: Assumption~\ref{assumpt:general assumption}, and then Lemma
435: \ref{lemma:S' cont} applies. We thus have
436: %%%%%%%%%%%%%%%%%%%%%%%%
437: % proposition          %
438: %%%%%%%%%%%%%%%%%%%%%%%%
439: \addtocounter{delta}{1}
440: \begin{proposition}
441:   There exists $\Delta_\thedelta>0$ such that if $z',z \in [0,Z]$ with
442:   $0\leq \Delta:= z'-z \leq \Delta_\thedelta$ then $\G_{(z',z)}$ maps
443:   $\S$ into $\S$ and $\S'$ into $\S'$ continuously.
444: \end{proposition}
445: \begin{remark}
446:   By the above result,  composition of the two FIOs $\G_{(z'',z')}$ and
447:   $\G_{(z',z)}$ is thus natural without further requirement such as
448:   having the operators properly supported.
449: \end{remark}
450: 
451: 
452: We now turn to {\em global} $L^2$ and Sobolev space continuity for the
453: operator $\G_{(z',z)}$.  We shall use the following lemma.
454: %%%%%%%%%%%%%%%%%%%%%%%%
455: % lemma                %
456: %%%%%%%%%%%%%%%%%%%%%%%%
457: \addtocounter{delta}{1}
458: \begin{lemma}
459:   \label{lemma: change of variable - symbol}
460:   Let $p_s(x,\xi)$ be bounded \wrt the parameter $s$
461:   with values in $S^m_\rho(X\times \R^n)$ and
462:   define
463:   \begin{align*}
464:   \tilde{\xi}(\Delta,x,\xi):= \xi - \Delta f(x,\xi)
465:   \end{align*}
466:   where $f$ is in $S^1 (X\times\R^n,\R^n)$ and homogeneous of
467:   degree $1$ in $\xi$, for  $|\xi|\geq 1$.  Then
468:   \begin{align*}
469:   \tilde{p}_s(\Delta,x,\xi):=
470:   p_s (x,\tilde{\xi}(\Delta,x,\xi))
471:   \end{align*}
472:   is bounded \wrt $s$ with values in
473:   $\Cinf([0,\Delta_\thedelta], S^m_\rho(X\times
474:   \R^n))$ for $\Delta_\thedelta$ small enough.
475: \end{lemma}
476: \begin{proof}
477:   Let $\Delta_\thedelta$ be small enough such that $|\xi - \Delta
478:   f(x,\xi)| \geq C_0>0$ if $|\xi|=1$ and
479:   $\Delta\in [0,\Delta_\thedelta]$. We then have
480:   \begin{align*}
481:     1+ C_0 |\xi| \leq 1+ |\xi - \Delta f(x,\xi)|\leq 1+ C_1|\xi|, \ 
482:     \xi \in \R^n, \ |\xi|\geq 1, \ \Delta\in [0,\Delta_\thedelta].
483:   \end{align*}
484:   This inequality yields the proper estimates for $\d_{x}^\alpha
485:   \d_{\xi}^\beta \tilde{p}_s$ to prove that $\tilde{p}_s \in
486:   S^m_\rho(X\times\R^n)$. As derivatives \wrt $\Delta$ do not
487:   affect the symbol order and type, the proof is finished. Bounds
488:   \wrt the parameter $s$ come naturally. The proof is complete.
489: \end{proof}
490: 
491: 
492: Following \cite{Stolk:04} we introduce
493: %%%%%%%%%%%%%%%%%%%%%%%%
494: % definition           %
495: %%%%%%%%%%%%%%%%%%%%%%%%
496: \begin{definition}
497:   \label{definition:special damping}
498:   Let $L\geq 2$. A symbol $q(z,.)$ bounded \wrt $z$ with
499:   values in $S^1(\R^{p}\times \R^{r})$ is said to
500:   satisfy Property (\ref{eq:P_L}) if it is non-negative and satisfies
501:   \begin{multline}
502:     \label{eq:P_L}
503:     \tag{$P_L$}
504:     | \d^\alpha_{y} \d_{\eta}^\beta q(z,y,\eta)|
505:     \leq
506:     C (1+|\eta|)^{-|\beta| + (|\alpha| + |\beta|)/L}\\
507:     (1+ q(z,y,\eta))^{1 - (|\alpha| + |\beta|)/L},\ \
508:     z \in [0,Z],\ y \in \R^{p},\ \eta \in \R^r.
509:   \end{multline}
510:   We then set $\rho=1-1/L$ and $\delta=1/L$. 
511: \end{definition}
512: %%%%%%%%%%%%%%%%%%%%%%%%
513: % remark               %
514: %%%%%%%%%%%%%%%%%%%%%%%%
515: \begin{remark}
516: \label{rem: P_L for alpha + beta >= L}
517: Suppose $q(z,.)$ as in Definition~\ref{definition:special damping} and
518: $|\alpha| + |\beta| \geq L$ then 
519: \begin{align*}
520:   (1+ |\eta|)^{1 - (|\alpha| + |\beta|)/L} \leq C(1+ q(z,y,\eta))^{1 -
521:   (|\alpha| + |\beta|)/L}, z \in [0,Z],\ y \in \R^{p},\ \eta \in \R^r.
522: \end{align*}
523: Estimate~(\ref{eq:P_L}) is thus clear in this case.
524: \end{remark}
525: Examples of symbols with such a property with $L>2$ are given in
526: \cite{Stolk:04}. In fact we prove that $c_1$ satisfies Property
527: (\ref{eq:P_L}) for $L=2$.
528: %%%%%%%%%%%%%%%%%%%%%%%%
529: % lemma                %
530: %%%%%%%%%%%%%%%%%%%%%%%%
531: \begin{lemma}
532:   \label{lemma: PL with L=2 for c_1}
533:   Let $q(z,y,\eta)$ be bounded \wrt $z$ with values in
534:   $S^1(\R^{p}\times \R^{r})$.  If $q\geq 0$ then $q$ satisfies
535:   Property (\ref{eq:P_L}) for $L=2$.
536: \end{lemma}
537: \begin{proof}
538:   Bounds \wrt $z$ are natural; we shall omit the dependence on $z$ in
539:   the proof for concision.  We have to prove that
540:   \begin{align*}
541:     |\d_y^\alpha \d_{\eta}^\beta q|
542:     \leq C\: (1+ |\eta|)^{\hf(|\alpha|-|\beta|)}\
543:     (1+q)^{1 -\hf(|\alpha|+|\beta|) }
544:   \end{align*}
545:   The property is clearly true for $|\alpha|+|\beta|=0$ and for
546:   $|\alpha|+|\beta|\geq 2$ by the remark above. Let us now treat the
547:   case $|\alpha|+|\beta|= 1$.  For this we use Landau's inequality:
548:   let $f\in C^2(\R)$ with $f\geq 0$ and $f''$ is bounded then (see
549:   \cite[page 40]{Duistermaat:96} and \cite[Lemma
550:   7.7.2]{Hoermander:V1})
551:   \begin{align*}
552:     |f'(t)|\leq 2 \: (f(t))^{\hf}\ \left(\sup_{t \in \R} |f''(t)|\right)^{\hf}.
553:   \end{align*}
554:   We first treat the case $|\alpha|=1$.
555:   Define $p(y,\eta) = (1+ |\eta|^2)^{-\hf}\ q(y,\eta)$. Then $p
556:   \in S^0(\R^p\times \R^r)$ and $\d_y^{2\alpha} p(y,\eta)$ is in
557:   $S^0(\R^p\times \R^r)$ and is thus bounded. We thus have
558:   \begin{align*}
559:     (1+|\eta|^2)^{-\hf}\ |\d_y^\alpha q (y,\eta)| 
560:     \leq C\: ((1+ |\eta|^2)^{-\hf}\ q (y,\eta ))^{\hf},
561:   \end{align*}
562:   which yields 
563:   \begin{align*}
564:      |\d_y^\alpha q (y,\eta)| 
565:     \leq C\:  (1+ |\eta|)^{\hf}\ (1+ q(y,\eta))^{\hf},
566:   \end{align*}
567:   which is the expected estimate. Let us now treat the case
568:   $|\beta|=1$, with for instance, $\beta=(1,0,\dots,0)$ and
569:   $\alpha=(0,\dots,0)$.  Define $p(y,\eta) = (1+
570:   |\eta|^2)^{\hf}\ q(y,\eta)$. Then $p \in S^2(\R^{p}\times
571:   \R^{r})$ and thus $\d_\eta^{2\beta} p(y,\eta)$ is bounded. We hence have
572:   \begin{align*}
573:     |\d_\eta^\beta p (y,\eta)|\leq C\:  (p(y,\eta))^{\hf}.
574:     \end{align*}
575:     With
576:     \begin{align*}
577:       \d_\eta^\beta p (y,\eta) = (1+
578:       |\eta|^2)^{\hf}\ \d_\eta^\beta q(y,\eta)  + \eta_1 (1+
579:       |\eta|^2)^{-\hf}\ q(y,\eta),
580:     \end{align*}
581:     the triangular inequality yields
582:     \begin{multline*}
583:       (1+
584:       |\eta|^2)^{\hf}\ |\d_\eta^\beta q(y,\eta)|
585:       \leq C\: (p(y,\eta))^{\hf} + |\eta_1| (1+
586:       |\eta|^2)^{-\hf}\ q(y,\eta)\\ 
587:       \leq C\: (q(y,\eta))^{\hf}
588:       ((1+
589:       |\eta|^2)^{\frac{1}{4}} + (q(y,\eta))^{\hf})\\
590:       \leq C\: (q(y,\eta))^{\hf}\ (1+ |\eta|^2)^{\frac{1}{4}}.
591:     \end{multline*}
592:     We finally obtain 
593:     \begin{align*}
594:       |\d_\eta^\beta q(y,\eta)| \leq C\: (q(y,\eta))^{\hf} (1+|\eta|)^{-\hf},
595:     \end{align*}
596:     which is the expected estimate.
597: \end{proof}
598: %%%%%%%%%%%%%%%%%%%%%%%%
599: % remark               %
600: %%%%%%%%%%%%%%%%%%%%%%%%
601: \begin{remark}
602:   \label{remark:special damping}
603:   If the symbol $q(z,y,\eta)$ satisfies Property (\ref{eq:P_L}) then the
604:   amplitude $q(z,y',\eta) +q(z,y,\eta)$ also satisfies Property (\ref{eq:P_L})
605:   (with derivatives \wrt $y$, $y'$ and $\eta$).
606: \end{remark}
607: %%%%%%%%%%%%%%%%%%%%%%%%
608: % proposition          %
609: %%%%%%%%%%%%%%%%%%%%%%%%
610: \begin{proposition}
611: \label{Prop:exp(-Dq)}
612: Let $q(z,.)$ be bounded \wrt $z$ with values in $S^1(\R^p\times \R^r)$
613: with $q(z,.) \geq 0$. Let $q(z,.)$ satisfy Property (\ref{eq:P_L}) and
614: define $\rho_\Delta(z,y,\eta) = \exp[-\Delta q(z,y,\eta)]$.  Let $m\in
615: \N$. Then $q^m \rho_\Delta$ is smooth \wrt $\Delta$, bounded \wrt $z$,
616: with values in $S^0_\rho(\R^{p}\times \R^r)$ for $\Delta$ in any
617: interval $[\Delta_{min},\Delta_{max}]$ with $ \Delta_{min}> 0$.
618: \end{proposition}
619: \begin{proof}
620:   $\d_y^\alpha \d_\eta^\beta (q^m \rho_\Delta) $ is a linear combination
621:   of terms of the form
622:   \begin{align*}
623:     \Delta^k (\d_y^{a_1} \d_\eta^{b_1} q) \dots (\d_y^{a_l} \d_\eta^{b_l} q)
624:     (\d_y^{\alpha_1} \d_\eta^{\beta_1}q) \dots 
625:     (\d_y^{\alpha_k} \d_\eta^{\beta_k} q) 
626:     q^{m-l} \rho_\Delta 
627:   \end{align*}
628:   with $0\leq l\leq m$ and $a_1 + \dots +a_l + \alpha_1 + \dots
629:   \alpha_k = \alpha$ and $b_1 + \dots +b_l + \beta_1 + \dots \beta_k =
630:   \beta$. We can estimate the absolute value of each of these terms, 
631:   using Property (\ref{eq:P_L}), by
632:   \begin{multline*}
633:     C \Delta^k (1+ |\eta|)^{- |\beta| + \frac{|\alpha|+|\beta|}{L}}
634:     (1+q)^{l+k - \frac{|\alpha|+|\beta|}{L}} q^{m-l} \rho_\Delta\\
635:     \leq C (1+ |\eta|)^{- |\beta| + \frac{|\alpha|+|\beta|}{L}} \Delta_{min}^{-m}
636:   \end{multline*}
637:   as $(1+q)^{l+k - \frac{|\alpha|+|\beta|}{L}} q^{m-l} \Delta^{k+m}
638:   \rho_\Delta \leq C$.
639: \end{proof}
640: 
641: While the symbol $\exp[-\Delta q(z,y,\eta)]$ is bounded \wrt $z$
642: and smooth \wrt $\Delta$ with $\Delta \geq \Delta_{min}>0$ with
643: values in $S^0_\rho(\R^p\times\R^r)$, this fails to
644: be true at $\Delta=0$:
645: \begin{align*}
646:  \d_\Delta \exp[-\Delta q]|_{\Delta=0} 
647:  = -q \notin S^0_\rho(\R^p\times\R^r).
648: \end{align*}
649: In fact when we want to control the behavior of $\exp[-\Delta q]$ close
650: to $\Delta=0$ we shall use the following definition and lemmas.
651: %%%%%%%%%%%%%%%%%%%%%%%%
652: % definition           %
653: %%%%%%%%%%%%%%%%%%%%%%%%
654: \begin{definition}
655:   Let $L\geq 2$, $\rho =1-1/L$ and $\delta=1/L$.
656:   Let $\rho_\Delta(z,y,\eta)$ be a function in
657:   $\Cinf(\R^p\times\R^r)$ depending on the parameters
658:   $\Delta\geq 0$ and $z \in [0,Z]$. We say that $\rho_\Delta$
659:   satisfies Property (\ref{eq:Q_L}) if the following holds
660:   \begin{multline}
661:     \label{eq:Q_L}
662:     \tag{$Q_L$}
663:     \d_y^\alpha \d_\eta^\beta (\rho_\Delta -\rho_\Delta|_{\Delta=0})
664:     (z,y,\eta) =
665:     \Delta^{m+\delta(|\alpha|+|\beta|)} \rho^{m \alpha
666:       \beta}_\Delta(z,y,\eta), \\
667:     \mbox{for} \ |\alpha| + |\beta| \leq L, \ \ \
668:     0\leq m \leq 1 - \delta (|\alpha|+|\beta|),
669:   \end{multline}
670:   where $\rho^{m \alpha \beta}_\Delta(z,y,\eta)$ is
671:   bounded \wrt $\Delta$ and $z$ with values in $S^{m -
672:     \rho|\beta| + \delta|\alpha|}_\rho
673:   (\R^p\times\R^r)$.  It follows that
674:   $\rho_\Delta(z,y,\eta) - \rho_\Delta|_{\Delta=0}(z,y,\eta)$ is itself
675:   bounded \wrt $\Delta$ and $z$ with values in $S^0_\rho (\R^p\times \R^r)$.
676: \end{definition}
677: %%%%%%%%%%%%%%%%%%%%%%%%
678: % lemma                %
679: %%%%%%%%%%%%%%%%%%%%%%%%
680: \begin{lemma}
681:   \label{lemma:exp(-Dq)- QL}
682:   Let $q(z,.)$ be bounded \wrt $z$ with values in $S^1(\R^p\times
683:   \R^r)$ and satisfy Property (\ref{eq:P_L}). Define
684:   $\rho_\Delta(z,y,\eta) = \exp[-\Delta q(z,y,\eta)]$.  Then
685:   $\rho_\Delta$ satisfies Property (\ref{eq:Q_L}) for $\Delta \in
686:   [0,\Delta_{max}]$ for any $\Delta_{max}>0$.  As
687:   $\rho_\Delta|_{\Delta=0} =1$, $\rho_\Delta$ is itself bounded \wrt
688:   $\Delta$ and $z$ with values in $S^0_\rho (\R^{p}\times \R^r)$.
689: \end{lemma}
690: \begin{proof}
691:   In the proof all the functions and symbols will naturally be bounded
692:   \wrt $z$. We thus drop the variable $z$ here for concision.
693: 
694:   We define 
695:   \begin{align*}
696:     \rho_\Delta^{m\alpha\beta}:= \Delta^{-m-\delta(|\alpha|+|\beta|)}
697:     \d_y^\alpha \d_\eta^\beta (\rho_\Delta - \rho_\Delta|_{\Delta=0}).
698:   \end{align*}
699:   We first consider the case $|\alpha|+|\beta|=0$ with $0\leq m \leq
700:   1$.  We need to estimate $|\d_y^a \d_\eta^b \rho_\Delta^{m00}|$.  The
701:   case $m=0$, $|a+b|=0$ has to be treated independently but is
702:   trivial: we clearly have $|\rho_\Delta^{000}|= |\rho_\Delta -1| \leq
703:   C$.  We shall now estimate $|\d_y^a \d_\eta^b \rho_\Delta^{m00}| = |
704:   \Delta^{-m} \d_y^a \d_\eta^b ( \rho_\Delta-1 )|$ in the case where
705:   $m>0$ or $|a|+|b|>0$.  For this we write 
706:   \begin{align}
707:     \label{eq:Taylor rho_Delta}
708:     \rho_\Delta(y,\eta) -1 = -\Delta \int_0^1 q(y,\eta)
709:     \exp[-s\Delta q(y,\eta)] d s.
710:   \end{align}
711:   We then have
712:   $\rho_\Delta^{m00}(y,\eta) = -\int_0^1 q_\Delta^m(s,y,\eta) ds$ with
713:   \begin{align*}
714:   q_\Delta^m(s,y,\eta) = \Delta^{1-m} q(y,\eta) \exp[-s\Delta q(y,\eta)].
715:   \end{align*}
716:   We prove that
717:   \begin{align*}
718:     |\d_y^a \d_\eta^b q_\Delta^m(s,y,\eta)| \leq C(s)
719:     (1+|\eta|)^{m-\rho |b|+ \delta |a|}
720:   \end{align*}
721:   with $C(s)$ bounded \wrt $\Delta$ and $L^1$ \wrt $s \in [0,1]$.
722:   The result then follows for $\rho_\Delta^{m00}$.
723: 
724:   When computing $\d_y^a\d_\eta^b q_\Delta^m$ we obtain
725:   a linear combination of terms of the form
726:   \begin{multline*}
727:     \Delta^{1-m} (\d^{a_0}_y\d^{b_0}_\eta q)
728:     (-s \Delta)^k (\d^{a_1}_y\d^{b_1}_\eta q)\dots
729:     (\d^{a_k}_y\d^{b_k}_\eta q)
730:     \exp[-s\Delta q],\\
731:     \mbox{with}\ a_0 + a_1 + \dots + a_k = a, \ \ 
732:     b_0 + b_1 + \dots + b_k = b,
733:   \end{multline*}
734:   (where $k$ can be $0$). Using Property (\ref{eq:P_L}) we find that
735:   the absolute value of such a term is bounded by
736:   \begin{multline*}
737:     C \Delta^{1-m}(s \Delta)^k (1+|\eta|)^{-|b| + \delta(|a+b|)}
738:     (1+q)^{k+1-\delta(|a+b|)} \exp[-s\Delta q]\\
739:     \leq C s^{m +\delta(|a+b|) -1}(1+|\eta|)^{m -\rho|b|+\delta|a|}
740:     \Delta^{\delta(|a+b|)} \\
741:     (s \Delta (1+q))^{-m + k+1-\delta(|a+b|)}\exp[-s\Delta q],
742:   \end{multline*}
743:   as $1\leq C(1+|\eta|)^m (1+q)^{-m}$ if $m\geq 0$.  If $l:=-m +
744:   k+1-\delta(|a+b|) \geq 0$ we use that $(s\Delta (1+q))^l
745:   \exp[-s\Delta q]\leq C$ if $0\leq s \leq 1$, $0 \leq \Delta \leq
746:   \Delta_{max}$ and $q\geq 0$ and we obtain the following estimate
747:   \begin{align*}
748:     C s^{m +\delta(|a+b|) -1}(1+|\eta|)^{m -\rho|b| + \delta|a|}
749:     \Delta^{\delta(|a+b|)}.
750:   \end{align*}
751:   If $l<0$, $(1+q)^l$ is simply bounded ($q\geq 0$) and we obtain the
752:   following estimate:
753:   \begin{align*}
754:     C \Delta^{k+1-m} s^k (1+|\eta|)^{m -\rho|b| + \delta |a|}.
755:   \end{align*}
756:   As $m +\delta(|a+b|) -1>-1$ in the considered case, both estimates
757:   exhibit bounds that are in $L^1([0,1])$ \wrt $s$. We also
758:   have  uniform bounds \wrt $\Delta$ as we have assumed $m\leq 1$.
759:   
760:   % Let us treat the case $|\alpha| + |\beta|=0$ and $m=1$.  We have
761: %   proved that $\rho_\Delta$ is bounded \wrt $\Delta$ with values in
762: %   $S^0_\rho(\R^n\times \R^n)$ (case $m=0$,
763: %   $|\alpha+\beta|=0$). Then similarly $\exp [-s\Delta q]$ is bounded
764: %   \wrt $s \in [0,1]$ and $\Delta$ with values in
765: %   $S^0_\rho$. As $q \in S^1(\R^n\times \R^n)$ we thus obtain
766: %   that $q \exp [-s\Delta q]$ is bounded \wrt $s \in [0,1]$ and
767: %   $\Delta$ with values in $S^1_\rho(\R^n\times \R^n)$.
768: %   Equation~(\ref{eq:Taylor rho_Delta}) then yields the result in this
769: %   case.
770: 
771: 
772:   We now treat the case $1\leq |\alpha| + |\beta| \leq L$, $0\leq m
773:   \leq 1 - \delta (|\alpha|+|\beta|)$.  We estimate the absolute value
774:   of
775:   \begin{align*}
776:     \d_y^a \d_\eta^b( \rho_\Delta^{m\alpha\beta}) =
777:     \Delta^{-m-\delta(|\alpha|+|\beta|)}\
778:     \d_y^{a+\alpha} \d_\eta^{b+\beta}\rho_\Delta 
779:   \end{align*}
780:   which is a linear combination of terms of the form
781:   \begin{multline*}
782:     \Delta^{k-m-\delta(|\alpha|+|\beta|)}\ (\d^{a_1}_y\d^{b_1}_\eta
783:     q)\dots (\d^{a_k}_y\d^{b_k}_\eta q)
784:     \exp[-\Delta q],\\
785:     \mbox{with}\ a_1 + \dots + a_k = a+\alpha, \ \ b_1 + \dots + b_k =
786:     b+\beta,
787:   \end{multline*}
788:   where $k\geq 1$. Using Property (\ref{eq:P_L}) we
789:   find that the absolute value of such a term is bounded by
790:   \begin{multline*}
791:     C\Delta^{k-m-\delta(|\alpha|+|\beta|)}
792:     (1+|\eta|)^{-|\beta|-|b| + \delta(|\alpha|+|a|+|\beta|+|b|)} \\
793:     (1+q)^{k -\delta(|\alpha|+|a|+|\beta|+|b|)} \exp[-\Delta q]\\
794:     \leq C (1+|\eta|)^{m - \rho (|\beta|+|b| ) + \delta (|\alpha|+|a|)}
795:     (1+q)^{-\delta (|a|+|b|)}
796:     \\
797:     (\Delta (1+q))^{k -m - \delta(|\alpha|+|\beta|)}\exp[-\Delta q]\\
798:     \leq C (1+|\eta|)^{m - \rho (|\beta|+|b| ) + \delta (|\alpha|+|a|)},
799:   \end{multline*}
800:   as $k -m - \delta (|\alpha|+|\beta|) \geq 1- m -
801:   \delta(|\alpha|+|\beta|)\geq 0$ and $0\leq \Delta \leq
802:   \Delta_{max}$. This completes the proof.
803: \end{proof}
804: %%%%%%%%%%%%%%%%%%%%%%%%
805: % lemma                %
806: %%%%%%%%%%%%%%%%%%%%%%%%
807: \begin{lemma}
808:   \label{lemma:QL stable by f}
809:   Let $f \in \Cinf(\R)$ and $q_\Delta(z,y,\eta)$ in $\Cinf(\R^p\times
810:   \R^r)$ that satisfies Property~(\ref{eq:Q_L}) and such that
811:   $q_\Delta(z,.)|_{\Delta=0}$ is independent of $y$ and $\eta$. Then
812:   $f(q_\Delta)(z,y,\eta)$ satisfies Property (\ref{eq:Q_L}).
813: \end{lemma}
814: \begin{proof}
815:   Again bounds \wrt $z$ are clear.
816:   We first treat the case $|\alpha|+|\beta|=0$. We write
817:   \begin{align*}
818:     f(q_\Delta) - f(q_\Delta|_{\Delta=0}) = (q_\Delta -q_\Delta|_{\Delta=0})
819:     \int_0^1 f'((1-s)q_\Delta|_{\Delta=0} + s q_\Delta) d s.
820:   \end{align*}
821:   As $q_\Delta|_{\Delta=0}$ is independent of $y$ and $\eta$, then
822:   $q_\Delta$ is bounded \wrt $\Delta$ with values in
823:   $S^0_\rho(\R^p\times \R^r)$ by Property~(\ref{eq:Q_L})  and so are
824:   $(1-s)q_\Delta|_{\Delta=0} + s q_\Delta$ and
825:   $f'((1-s)q_\Delta|_{\Delta=0} + s q_\Delta)$ by Lemma 18.1.10 in
826:   \cite{Hoermander:V3} with bounds in $S^0_\rho(\R^p\times
827:   \R^r)$ uniform with respect to $s$. We thus obtain that $\int_0^1
828:   f'((1-s)q_\Delta|_{\Delta=0} + s q_\Delta) d s$ is bounded \wrt
829:   $\Delta$ with values in $S^0_\rho(\R^p\times \R^r)$. We
830:   conclude using Property (\ref{eq:Q_L}) for $q_\Delta
831:   -q_\Delta|_{\Delta=0}$.  Let us now treat the case $1 \leq
832:   |\alpha|+|\beta|\leq L$ and choose $0\leq m \leq 1 -
833:   \delta(|\alpha|+|\beta|)$.  We see that $\d_y^\alpha \d_\eta^\beta
834:   f(q_\Delta) $ is a linear combination of terms of the form
835:   \begin{align*}
836:     (\d_y^{\alpha_1} \d_\eta^{\beta_1} q_\Delta)\dots 
837:     (\d_y^{\alpha_k} \d_\eta^{\beta_k} q_\Delta) f^{(k)}(q_\Delta),
838:   \end{align*} 
839:   where $k \geq 1$, $\alpha_1 + \dots + \alpha_k = \alpha$, $\beta_1+
840:   \dots + \beta_k = \beta$.  Now choose $0\leq m_i \leq 1 -
841:   \delta(|\alpha_i|+|\beta_i|)$, $i=1,\dots, k$, such that $m=m_1+
842:   \dots +m_k$. Then Property (\ref{eq:Q_L}) yields terms of the form
843:   \begin{align*}
844:     \Delta^{m_1 + \delta(|\alpha_1|+|\beta_1|)}\dots \Delta^{m_k +
845:       \delta(|\alpha_k|+|\beta_k|)} q_\Delta^{m_1 \alpha_1 \beta_1 }
846:     \dots q_\Delta^{m_k \alpha_k \beta_k} = \Delta^{m + \delta(|\alpha|+|\beta|)}
847:     q_\Delta^{m\alpha \beta}
848:   \end{align*}
849:   with $q_\Delta^{m_i \alpha_i \beta_i}$, $i=1,\dots, k$, bounded
850:   \wrt $\Delta$ with values in $S^{m_i - \rho|\alpha_i| + \delta
851:     |\beta_i| }_\rho(\R^p\times \R^r)$ and $q_\Delta^{m\alpha \beta} :=
852:   q_\Delta^{m_1 \alpha_1 \beta_1 } \dots q_\Delta^{m_k \alpha_k
853:     \beta_k}$. We note that $f^{(k)}(q_\Delta)$ is bounded \wrt
854:   $\Delta$ with values in $S^0_\rho(\R^p\times \R^r)$.  The symbol
855:   $q_\Delta^{m\alpha \beta}$ is bounded \wrt  $\Delta$ with values
856:   in $S^{m - \rho|\alpha| + \delta |\beta| }_\rho(\R^p\times \R^r)$, which
857:   yields the result.
858: % For instance
859: %   $\d_{y_i} f(q_\Delta) = \d_{y_i} q_\Delta f'(q_\Delta)$,
860: %   $i=1,\dots,p$. Again $f'(q_\Delta)$ is bounded in
861: %   $S^0_\rho$ and we use Property (\ref{eq:Q_L}) for $\d_{y_i}
862: %   q_\Delta$. We proceed the same for $\d_{\eta_j} f(q_\Delta)$,
863: %   $j=1,\dots,r$.  Consider now
864: %   \begin{align*}
865: %     \d_{y_i} \d_{\eta_j} f(q_\Delta) = \d_{y_i} \d_{\eta_j} q_\Delta\
866: %     f'(q_\Delta) + \d_{y_i}q_\Delta\ \d_{\eta_j}q_\Delta\ f''(q_\Delta).
867: %   \end{align*}
868: %   The first term is treated as before. For the second term we write
869: %   \begin{align*}
870: %   \d_{y_i}q_\Delta\ \d_{\eta_j}q_\Delta = \Delta^{m_1+\delta}
871: %   \Delta^{m_2+\delta}\
872: %   q_\Delta^{m_1\alpha^1\beta^1}(z,y,\eta)\
873: %   q_\Delta^{m_2\alpha^2\beta^2}(z,y,\eta),
874: %   \end{align*}
875: %   using Property (\ref{eq:Q_L}) for both terms with $\alpha^1_k=\delta_{ki}$,
876: %   $k=1,\dots,p$, $\beta^1=(0,\dots,0)$, $\alpha^2=(0,\dots,0)$,
877: %   $\beta^2_l=\delta_{jl}$, $l=1,\dots,r$, and $0 \leq m_1,m_2 \leq 1-
878: %   \delta$. Note that $q_\Delta^{m_1\alpha^1\beta^1} \in S^{m_1 +
879: %     \delta}_\rho$ while $q_\Delta^{m_2\alpha^2\beta^2} \in
880: %   S^{m_2 - \rho}_\rho$. Define $m=m_1+m_2$ we then obtain
881: %   \begin{align*}
882: %     \d_{y_i}q_\Delta\ \d_{\eta_j}q_\Delta = \Delta^{m+2\delta}
883: %     \tilde{q}_\Delta^{m_1 m_2 \alpha^1\beta^2},
884: %   \end{align*}
885: %   where $\tilde{q}_\Delta^{m_1 m_2 \alpha^1\beta^2} \in S^{m -
886: %     \rho+\delta}_\rho$ and $0 \leq m \leq 2 - 2 \delta$,
887: %   which gives the conclusion. The other cases where $1 \leq
888: %   |\alpha|+|\beta|\leq 2$ can be treated in a similar fashion.
889: \end{proof}
890: %%%%%%%%%%%%%%%%%%%%%%%%
891: % corollary            %
892: %%%%%%%%%%%%%%%%%%%%%%%%
893: With Remark~\ref{remark:special damping}, Lemma~\ref{lemma:QL stable
894: by f} and the previous lemma we obtain
895: \begin{corollary}
896:   \label{corollary: property QL for p_Delta}
897:   Let $f \in \Cinf(\R)$ and let $q(z,.)$ bounded \wrt $z$
898:   with values in $S^1(\R^{p}\times \R^r)$ satisfy Property
899:   (\ref{eq:P_L}). Define
900:   \begin{align*}
901:   p_\Delta(z,y',y,\eta) =\exp[-\Delta (q(z,y',\eta)+ q(z,y,\eta))].
902:   \end{align*}
903:   Then $f(p_\Delta)$ satisfies Property (\ref{eq:Q_L}). As
904:   $f(p_\Delta)|_{\Delta=0}=1$, $f(p_\Delta)$ is itself bounded \wrt
905:   $\Delta$ and $z$ with values in $S^0_\rho(\R^{2p}\times
906:   \R^r)$.
907: \end{corollary}
908: %%%%%%%%%%%%%%%%%%%%%%%%
909: % proposition          %
910: %%%%%%%%%%%%%%%%%%%%%%%%
911: Note that the property (\ref{eq:Q_L}) is stable when we go from
912: amplitudes to symbols:
913: \begin{proposition}
914:   Let $q_\Delta(z,x,y,\xi)$ be an amplitude in
915:   $S^0_\rho(\R^{2p}\times \R^p)$ depending on the
916:   parameters $\Delta\geq 0$ and $z \in [0,Z]$ that satisfies Property
917:   (\ref{eq:Q_L}). Then $\symb{q_\Delta} (z,x,\xi)$ satisfies property
918:   (\ref{eq:Q_L}).
919: \end{proposition}
920: \begin{proof}
921:   We use the oscillatory integral representation for the symbol:
922:   \begin{align*}
923:     \symb{q_\Delta}(z,x,\xi) := \iint \exp[-i \inp{y}{\eta}]\
924:     q_\Delta(z,x,x-y,\xi-\eta) \ \dslash \eta \ d y.
925:   \end{align*}
926:   Let $0\leq |\alpha|+|\beta|\leq L$ and $0\leq m \leq 1 - \delta
927:   (|\alpha|+|\beta|)$.  Computing $\d_x^\alpha \d_\xi^\beta (
928:   \symb{q_\Delta} - \symb{q_\Delta}|_{\Delta=0})$, we obtain a linear
929:   combination of terms of the form, with $\alpha_1+\alpha_2 = \alpha$,
930:   \begin{multline*}
931:     \iint \exp[-i \inp{y}{\eta}]\ \d_2^{\alpha_1} \d_3^{\alpha_2}
932:     \d_4^\beta (q_\Delta-q_\Delta|_{\Delta=0}) (z,x,x-y,\xi-\eta)\
933:     \dslash \eta \ d y\\ = \iint \exp[-i \inp{y}{\eta}]\
934:     \Delta^{m+\delta(|\alpha|+|\beta|)} q_\Delta^{m
935:     (\alpha_1,\alpha_2) \beta}(z,x,x-y,\xi-\eta) \ \dslash \eta \ d
936:     y\\ = \Delta^{m+\delta(|\alpha|+|\beta|)} \symb{q_\Delta^{m
937:     (\alpha_1,\alpha_2) \beta}},
938:   \end{multline*}
939:   where $q_\Delta^{m (\alpha_1,\alpha_2) \beta}$ is bounded \wrt
940:     $\Delta$ and $z$ with values in the symbol space $S^{m -
941:     \rho|\beta| + \delta|\alpha|}_\rho (\R^{2p}\times \R^p)$.  As the
942:     map $a \mapsto \symb{a}$ maps bounded sets into bounded sets the
943:     result follows.
944: \end{proof}
945: We shall also need the following lemma.
946: %%%%%%%%%%%%%%%%%%%%%%%%
947: % lemma                %
948: %%%%%%%%%%%%%%%%%%%%%%%%
949: \begin{lemma}
950:   \label{lemma:symb - amp}
951:   Let $q_\Delta(z,x,y,\xi)$ be an amplitude in
952:   $S^0_\rho(\R^{2p}\times \R^p)$ depending on
953:   the parameters $\Delta\geq 0$ and $z \in [0,Z]$ that satisfies
954:   Property (\ref{eq:Q_L}) for $1\leq |\alpha|+|\beta|\leq 2$ and such that
955:   $q_\Delta(z,.)|_{\Delta=0}$ is independent of $(x,y,\xi)$. Let
956:   $r(x,\xi) \in S^s(\R^p\times \R^p)$ for some $s \in \R$.  Then
957:   \begin{align*}
958:     \symb{q_\Delta\ r } (z,x,\xi) - q_\Delta(z,x,x,\xi)\ r(x,\xi) =
959:     \Delta^{m + 2 \delta} \lambda_{\Delta }^{m}(z,x,\xi), \ 0\leq m \leq \rho
960:     - \delta,
961:   \end{align*}
962:   where the function $\lambda_\Delta^{m }(z,x,\xi)$ is bounded with
963:   respect to $\Delta$ and $z$ with values in $S^{m +s - (\rho -
964:   \delta)}_\rho(\R^p\times \R^p)$.
965: \end{lemma}
966: \begin{proof}
967:   For the sake of concision we take $p=1$ in the proof but it
968:   naturally extends to $p\geq 1$. We write $\lambda_\Delta = q_\Delta
969:   r$. Using the oscillatory integral representation of
970:   $\symb{q_\Delta}$ we obtain
971:   \begin{multline*}
972:     \symb{q_\Delta r } (z,x,\xi) - q_\Delta(z,x,x,\xi) r(x,\xi) \\
973:     =\iint \exp[-i \inp{y}{\xi-\eta}]
974:     (\lambda_\Delta(z,x,x-y,\eta)  - \lambda_\Delta(z,x,x,\eta)\
975:     \dslash \eta \ d y.
976:   \end{multline*}
977:   Taylor's formula yields
978:   \begin{multline*}
979:     \symb{q_\Delta r } (z,x,\xi) - q_\Delta(z,x,x,\xi)r(x,\xi) \\
980:     =\int_0^1 \iint -y \exp[-i \inp{y}{\xi-\eta}] \
981:    \d_3 \lambda_\Delta(z,x,x-s y,\eta) \
982:     \dslash \eta \ d y\ d s.
983:   \end{multline*}
984:   With an integration by parts we obtain
985:   \begin{multline*}
986:     \symb{q_\Delta r} (z,x,\xi) - q_\Delta(z,x,x,\xi)r(x,\xi) \\
987:     =\int_0^1 \iint i  \exp[-i \inp{y}{\xi-\eta}] \ \d_3 \d_4
988:     \lambda_\Delta(z,x,x-s y,\eta) \ \dslash \eta \ d y\ d s\\
989:     = \symb{i \int_0^1 \d_3 \d_4 \lambda_\Delta(z,x,(1-s)x+s y,\xi)\ ds},
990:   \end{multline*}
991:   where $\d_3 \d_4 \lambda_\Delta(z,x,y,\xi) = (\d_y \d_\xi
992:   q_\Delta)(z,x,y,\xi) \ r(x,\xi) + \d_y q_\Delta(z,x,y,\xi) \d_\xi
993:   r(x,\xi)$, as $r$ does not  depend on $y$.  The first term is treated
994:   using Property (\ref{eq:Q_L}) while for the second one we write
995:   \begin{align*}
996:     \d_y q_\Delta\  \d_\xi r 
997:     = \Delta^{m'+\delta}\  q_\Delta^{m'(0,1)0}\  \d_\xi r,
998:   \end{align*}
999:   where $0\leq m' \leq 1-\delta$ and $q_\Delta^{m'(0,1)0}\ r \in
1000:   S^{m'+s-1+\delta}_\rho(\R^{2p}\times \R^p)$ by Property
1001:   (\ref{eq:Q_L}). We actually take $\delta \leq m' \leq 1-\delta$ and
1002:   write $m=m'-\delta$.  We obtain
1003:   \begin{align*}
1004:     \d_y q_\Delta \d_\xi r = \Delta^{m+2\delta}\  \tilde{q}_\Delta^{m},
1005:   \end{align*}
1006:   where $\tilde{q}_\Delta^{m}$ is bounded \wrt $\Delta$ with values
1007:   in $S^{m+s-\rho+\delta}_\rho(\R^{2p}\times \R^p)$ and
1008:   $0\leq m\leq 1-2\delta= \rho -\delta$.  We conclude since the map
1009:   $\sigma\{.\}$ maps bounded sets into bounded sets.
1010: \end{proof}
1011: We are now ready to give an estimate of the operator norm of the
1012: thin-slab propagator, $\G_{(z',z)}$, in $L(H^{(s)}(X),H^{(s)}(X'))$
1013: for any $s \in \R$.
1014: %%%%%%%%%%%%%%%%%%%%%%%%
1015: % theorem              %
1016: %%%%%%%%%%%%%%%%%%%%%%%%
1017: \addtocounter{delta}{1}
1018: \begin{theorem}
1019:   \label{theorem:H^s estimate}
1020:   Let $s \in \R$. There exists $M>0$, $\Delta_\thedelta
1021:   >0$ such that
1022:   \begin{align*}
1023:   \| \G_{(z',z)}\|_{(H^{(s)},H^{(s)})}\leq 1+ \Delta M,
1024:   \end{align*}
1025:   for all $z',z \in [0,Z]$ such that $0 \leq\Delta=z'-z\leq
1026:   \Delta_\thedelta$.
1027: \end{theorem}
1028: In the proof we assume that $c_1$ satisfies property $P_L$ for some $L
1029: \geq 2$. We know that it is always true for $L=2$ by Lemma~\ref{lemma:
1030: PL with L=2 for c_1} but special choices for $c_1$ can be made. As
1031: before we use $\rho=1-1/L$ and $\delta=1/L$ with $\rho > \delta$ for
1032: $L>2$ and $\rho=\delta=\hf$ for $L=2$. In the proof we proceed
1033: classically by computing $\G_{(z',z)} \G_{(z',z)}^\ast$ and use the
1034: classical results on $\psi$DOs (see e.g. \cite[Section 5]{MeSj:76} and
1035: also \cite{Hoermander:83}). Here we however do not content ourself
1036: with the continuity of $\G_{(z',z)}$ but we want to obtain a precise
1037: estimate of the operator norm in $L(H^{(s)}(X),H^{(s)}(X'))$, which
1038: will be required in the sequel. Here we exploit the fact that
1039: $\Delta$ can be taken arbitrarily small  which allows to carry out
1040: some explicit computations.
1041: \begin{proof}
1042:   Let $s \in \R$, then the kernel of $\A_{(z',z)}:=\G_{(z',z)}\circ
1043:   E^{(-s)}$ is given by
1044:   \begin{align*}
1045:     \A_{(z',z)}(x',x)
1046:     = \int \exp[i
1047:     \phi_{(z',z)}(x',x,\xi)]\ g_{(z',z)}(x',\xi)\
1048:     \e{\xi}{-s}\, \dslash \xi.
1049:   \end{align*}
1050:   Computing the kernel $D_{(z',z)}$ of ${\cal
1051:     D}_{(z',z)}:=\A_{(z',z)}\circ\A_{(z',z)}^\ast$ we obtain
1052:   \begin{multline*}
1053:     D_{(z',z)}(x',x) \\
1054:     = \int \exp\left[ i\inp{x'-x}{\xi}
1055:     + i\Delta \left(b_1(z,x',\xi)-b_1(z,x,\xi)\right)\right]
1056:     d_{(z',z)}(x',x,\xi)\
1057:     \dslash \xi
1058:   \end{multline*}
1059:   where
1060:   \begin{multline*}
1061:     d_{(z',z)}(x',x,\xi) \\
1062:     = \exp[-\Delta (c_1(z,x',\xi)+c_1(z,x,\xi) ] g_{(z',z)}(x',\xi)\ 
1063:     \overline{g_{(z',z)}}(x,\xi)\ \e{\xi}{-2s}.
1064:   \end{multline*}
1065:   We write $b_1(z,x',\xi)-b_1(z,x,\xi) = \inp{x'-x}{ h(z,x',x,\xi)}$
1066:   where $h$ is smooth and homogeneous of degree one in $\xi$,
1067:   $|\xi|\geq 1$.  The function $h$ and continuous \wrt $z$ with
1068:   values in $S^1(X\times \R^n)$ by Assumption~\ref{assumpt:general
1069:     assumption} and estimate (1.1.9) in \cite{Hoermander:V1}. We thus
1070:   obtain that the change of variables $\xi \to \xi+\Delta
1071:   h(z,x',x,\xi)=H_{(\Delta,z,x',x)}(\xi)$ is a global diffeomorphism
1072:   for $\Delta$ small enough (uniformly in $z \in [0,Z]$).  We denote
1073:   $\tilde{\xi} (\Delta,z,x',x,\xi) = H^{-1}_{(\Delta,z,x',x)}(\xi)$.
1074:   We thus have
1075:   \begin{align*}
1076:     D_{(z',z)}(x',x) =\!
1077:     \int \exp\left[ i\inp{x'-x}{\xi} \right]
1078:     d_{(z',z)}(x',x,\tilde{\xi}(\Delta,z,x',x,\xi))\
1079:     \J_\Delta(z,x',x,\xi)\  \dslash \xi
1080:   \end{align*}
1081:   where $\J_\Delta(z,x',x,\xi)$ is the Jacobian.
1082:   \begin{lemma}
1083: %%%%%%%%%%%%%%%%%%%%%%%%
1084: % sub-lemma            %
1085: %%%%%%%%%%%%%%%%%%%%%%%%
1086:     \label{lemma: change of variables}
1087:     The function $\tilde{\xi}(\Delta,z,x',x,\xi)$ is homogeneous of degree 1 in
1088:     $\xi$, for $|\xi|\geq 1$,  continuous \wrt $z$, $\Cinf$ \wrt $\Delta$ with values in
1089:     $S^1(\R^{2n}\times(\R^{n}))$ if
1090:     $\Delta$ is small enough, i.e.,
1091:     \begin{align*}
1092:     \exists \Delta_\thedelta >0, \ \
1093:     \tilde{\xi} \in \Con^0([0,Z],\Cinf([0,\Delta_\thedelta],
1094:     S^1(\R^{2n}\times(\R^{n})))).
1095:     \end{align*}
1096:   \end{lemma}
1097:   \begin{proof}
1098:     Homogeneity is clear. We
1099:     have
1100:     \begin{multline*}
1101:       |\tilde{\xi}(\Delta,z,x',x,\xi)| = |\xi - 
1102:       \Delta h (z,x',x,\tilde{\xi}(\Delta,z,x',x,\xi))| \\
1103:       \leq 1 + \Delta C (1+|\tilde{\xi}(\Delta,z,x',x,\xi)|),\ \ |\xi|=1,
1104:     \end{multline*}
1105:     which yields, because of homogeneity,
1106:     \begin{align*}
1107:       |\tilde{\xi}(\Delta,z,x',x,\xi)| \leq \frac{1+ \Delta C}{1-
1108:         \Delta C}(1+ |\xi|), \ |\xi|\geq 1, 
1109:     \end{align*}
1110:     for $\Delta$ small enough, uniformly chosen \wrt $z \in
1111:     [0,Z]$, $x',x \in \R^n$.  Differentiating the $j^{th}$ coordinate
1112:     of $\xi$,
1113:     \begin{align*}
1114:     \xi_j = \tilde{\xi}_j(\Delta,z,x',x,\xi) + \Delta
1115:     h_j (z,x',x,\tilde{\xi}(\Delta,z,x',x,\xi)),\ j=1,\dots,n,
1116:     \end{align*}
1117:     \wrt $x_i$ yields
1118:     \begin{multline}
1119:       \label{eq:dx eta}
1120:       \d_{x_i} \tilde{\xi}_j(\Delta,z,x',x,\xi) +
1121:       \Delta \partial_{x_i} h_j (z,x',x,\tilde{\xi}(\Delta,z,x',x,\xi))\\
1122:       +\Delta \sum_l \partial_{\tilde{\xi}_l}
1123:       h_j (z,x',x,\tilde{\xi}(\Delta,z,x',x,\xi)) \ 
1124:       \d_{x_i} \tilde{\xi}_l(\Delta,z,x',x,\xi) = 0, \\ j=1,\dots,n.
1125:     \end{multline}
1126:     The partial derivatives of $h$ are bounded for $|\xi|=1$.
1127:     We can hence solve for $\d_{x_i} \tilde{\xi}(\Delta,z,x',x,\xi)$
1128:     when $\Delta$ is small enough and find the expected estimate from
1129:     that obtained for $\tilde{\xi}(\Delta,z,x',x,\xi)$:
1130:     \begin{align*}
1131:     \exists C> 0, \  |\d_{x_i} \tilde{\xi}(\Delta,z,x',x,\xi)|
1132:     \leq C (1+ | \xi|),
1133:     \ x',x\in\R^{n}, \ \xi \in \R^n.
1134:     \end{align*}
1135:     Differentiating \wrt $x_i'$, $\xi_i$, and
1136:     $\Delta$ yields similar structures and the proper symbol
1137:     estimates. The proof carries on by induction. Note that the
1138:     required size for $\Delta$ to solve the systems of the form
1139:     (\ref{eq:dx eta}) remains fixed along the induction process.
1140:   \end{proof}
1141: %%%%%%%%%%%%%%%%%%%%%%%%
1142: % back to theorem      %
1143: %%%%%%%%%%%%%%%%%%%%%%%%
1144:   {\em Continuation of the proof of Theorem~\ref{theorem:H^s
1145:       estimate}.}  From (the proof of) Lemma~\ref{lemma: change of
1146:       variables} we also obtain that the Jacobian
1147:       $\J_\Delta(z,x',x,\xi)$ is homogeneous of degree zero in $\xi$,
1148:       $|\xi|\geq 1$, and is continuous \wrt $z$ and $\Cinf$ \wrt
1149:       $\Delta$ with values in $S^0(\R^{2n}\times \R^{n}))$.
1150:   
1151:   We write $\tilde{p}_\Delta(z,x',x,\xi):= \exp[-\Delta
1152:   (c_1(z,x',\xi)+c_1(z,x,\xi) ]$.  As $c_1$ satisfies Property
1153:   (\ref{eq:P_L}) we then have $\tilde{p}_\Delta$ satisfying property
1154:   (\ref{eq:Q_L}) by Corollary~\ref{corollary: property QL for
1155:   p_Delta}.  Define $p_\Delta(z,x',x,\xi) :=
1156:   \tilde{p}_\Delta(z,x',x,\tilde{\xi}(\Delta,z,x',x,\xi))$.
1157:   Lemma~\ref{lemma: change of variable - symbol} and Lemma~\ref{lemma:
1158:   change of variables} yield that $p_\Delta$ satisfy property
1159:   (\ref{eq:Q_L}). We then have
1160:   \begin{align*}
1161:     d_{(z',z)}(x',x,\tilde{\xi}(\Delta,z,x',x,\xi))\
1162:     \J_\Delta(z,x',x,\xi)\
1163:     =:  p_\Delta(z,x',x,\xi)\ k_\Delta(z,x',x,\xi)
1164:   \end{align*}
1165:   where $k_\Delta(z,.)$ is bounded \wrt $z$ and $\Cinf$
1166:   \wrt $\Delta$ with values in $S^{-2s} (\R^{2n} \times
1167:   \R^n)$ and $k_\Delta(z,.)|_{\Delta=0}= \e{.}{-2s}$ by
1168:   Lemma~\ref{lemma: change of variable - symbol} and Lemma~\ref{lemma:
1169:     change of variables}. By Theorem 1.1.9 and formula (1.1.9) in
1170:   \cite{Hoermander:V1} we obtain
1171:   \begin{align*}
1172:     k_\Delta(z,x',x,\xi) = \e{\xi}{-2s} + \Delta
1173:     \tilde{k}_\Delta(z,x',x,\xi),
1174:   \end{align*}
1175:   where $\tilde{k}_\Delta$ is bounded \wrt $z$ and $\Cinf$
1176:   \wrt $\Delta$ with values in $S^{-2s} (\R^{2n} \times
1177:   \R^n)$.
1178: 
1179:   Call ${\cal F}_{(z',z)}:= E^{(s)} \circ {\cal D}_{(z',z)}\circ
1180:   E^{(s)}$.  Its symbol is in $S^0_\rho(\R^{n} \times \R^n)$
1181:   and is given by
1182:   \begin{multline*}
1183:     f_{(z',z)}(x',\xi):= (\e{\xi}{s}\ \#\
1184:     \symb{ p_\Delta(z,x',x,\xi) \
1185:       k_\Delta(z,x',x,\xi) }\
1186:     \#\ \e{\xi}{s}) (x',\xi)\\
1187:     = (\e{\xi}{s}\ \#\ \symb{ p_\Delta(z,x',x,\xi)
1188:       \e{\xi}{-2s}}\ \#\ \e{\xi}{s}) (x',\xi)\\
1189:     + \Delta (\e{\xi}{s}\ \#\ \symb{ p_\Delta(z,x',x,\xi)
1190:        \tilde{k}_\Delta(z,x',x,\xi) \e{\xi}{-2s} }\ \#\ \e{\xi}{s} ) (x',\xi)
1191:   \end{multline*}
1192:   As $p_\Delta$ bounded \wrt $z$ and $\Delta$, $\Delta$
1193:   small enough, with values in $S^0_\rho(\R^{2n} \times \R^n)$
1194:   (Property (\ref{eq:Q_L})) we obtain that the second term in the equation
1195:   above satisfies the same property and thus we can write
1196:   \begin{align*}
1197:     {\cal F}_{(z',z)}={\cal F}^a_{(z',z)}+ \Delta {\cal F}^1_{(z',z)}
1198:   \end{align*}
1199:   where ${\cal F}^a_{(z',z)}$ has for symbol
1200:   \begin{align*}
1201:     (\e{\xi}{s}\  \#\ \symb{ p_\Delta(z,x',x,\xi)
1202:       \e{\xi}{-2s} }\ \#\ \e{\xi}{s} ) (x',\xi)
1203:   \end{align*}
1204:   and $\|{\cal F}^1_{(z',z)}\|_{(L^2,L^2)} \leq K^1$, uniformly in $z
1205:   \in [0,Z]$ and $\Delta$, $\Delta$ small enough, by the
1206:   Calder\'{o}n-Vaillancourt theorem (see \cite[Chapter 7, Sections
1207:   1,2]{Kumano-go:81} or \cite[Section XIII-2]{Taylor:81}) in the case
1208:   $L=2$ and by Theorem 18.1.11 in \cite{Hoermander:V3} in the case
1209:   $L>2$.  With Lemma~\ref{lemma:symb - amp} we see that
1210:   \begin{align*}
1211:     \symb{ p_\Delta(z,x',x,\xi)
1212:      \e{\xi}{-2s} } - p_\Delta(z,x',x',\xi)
1213:     \e{\xi}{-2s} = \Delta \lambda_\Delta (z,x',\xi)
1214:   \end{align*}
1215:   where $\lambda_\Delta$ is bounded \wrt $\Delta$ and $z$
1216:   with values in $S^{-2s}_\rho(\R^{n} \times \R^n)$.
1217:   We thus obtain
1218:   \begin{align*}
1219:     {\cal F}^a_{(z',z)}={\cal F}^b_{(z',z)}+ \Delta {\cal F}^2_{(z',z)}
1220:   \end{align*}
1221:   where ${\cal F}^b_{(z',z)}$ has for symbol
1222:   \begin{multline*}
1223:     f_\Delta^b(z,x',\xi) :=
1224:     (\e{\xi}{s}\ \#\ p_\Delta(z,x',x',\xi) \e{\xi}{-2s}\ \#\ \e{\xi}{s})
1225:     (z,x',\xi) \\
1226:     = (\e{\xi}{s}\ \#\ p_\Delta(z,x',x',\xi)  \e{\xi}{-s})(z,x',\xi)
1227:   \end{multline*}
1228:   and $\|{\cal F}^2_{(z',z)}\|_{(L^2,L^2)} \leq K^2$ uniformly in $z
1229:   \in [0,Z]$ and $\Delta$, $\Delta$ small enough.
1230: 
1231:   For the rest of the proof, if we don't write it explicitly, by
1232:   $p_\Delta$ and $p_\Delta(z,x,\xi)$ we shall actually mean
1233:   $p_\Delta(z,x,x,\xi)$.
1234:   %%%%%%%%%%%%%%%%%%%%%%%%  
1235:   % sub-lemma            %
1236:   %%%%%%%%%%%%%%%%%%%%%%%%
1237:   \begin{lemma}
1238:     \begin{align*}
1239:       (\e{.}{s}\ \#\ p_\Delta(z,.)  \e{.}{-s})(z,x,\xi)-
1240:       p_\Delta(z,x,\xi) = \Delta \mu_\Delta(z,x,\xi),
1241:     \end{align*}
1242:     where $\mu_\Delta(z,x,\xi)$ is bounded \wrt $z$
1243:     and $\Delta$ with values in $S^{0}_\rho(X\times \R^n)$.
1244:   \end{lemma}
1245:   \begin{proof}
1246:     We write
1247:     \begin{align*}
1248:       p_\Delta(z,x,\xi) = \e{\xi}{-s}
1249:       \iint\exp[-i\inp{y}{\xi-\eta}]\
1250:       \e{\eta}{s} \ p_\Delta(z,x,\xi)\ \dslash
1251:       \eta\ d y
1252:     \end{align*}
1253:     and thus obtain, with the oscillatory integral representation for
1254:     the composition formula,
1255:     \begin{multline*}(\e{.}{s}\ \#\ p_\Delta(z,.)  \e{.}{-s})(z,x,\xi)-
1256:       p_\Delta(z,x,\xi) =\\
1257:       \e{\xi}{-s}
1258:       \iint\exp[-i\inp{y}{\xi-\eta}]\
1259:       \e{\eta}{s}\
1260:       (p_\Delta(z,x-y,\xi) - p_\Delta(z,x,\xi))\
1261:       \dslash \eta \ d y.
1262:     \end{multline*}
1263:     With Taylor's formula and applying an integration by part, we find
1264:     (we have supposed $n=1$ for the sake of simplicity but it
1265:     naturally extends to $p\geq1$)
1266:     \begin{multline*}(\e{.}{s}\ \#\ p_\Delta(z,.)  \e{.}{-s})(z,x,\xi)-
1267:       p_\Delta(z,x,\xi) =\\
1268:       \e{\xi}{-s}
1269:       \int_0^1 \iint i \exp[-i\inp{y}{\xi-\eta}]\
1270:       \d_\eta \e{\eta}{s}\
1271:       \d_x p_\Delta(z,x- r y,\xi) \
1272:       \dslash \eta\ d y\ d r .
1273:     \end{multline*}
1274:     Using Property (\ref{eq:Q_L}) with $m=1-\delta$ we find
1275:     \begin{multline*}(\e{.}{s}\ \#\ p_\Delta(z,.)  \e{.}{-s})(z,x,\xi)-
1276:       p_\Delta(z,x,\xi) = \Delta \e{\xi}{-s}\\
1277:       \int_0^1 \iint i \exp[-i\inp{y}{\xi-\eta}]\
1278:       \d_\eta \e{\eta}{s}\
1279:       q_\Delta^{m 1 0}(z,(1-r)x+r (x-y),\xi) \
1280:       \dslash \eta\ d y\ d r \\
1281:       = \Delta \e{\xi}{-s}
1282:       (\d_\xi \e{\xi}{s}\ \#\ \tilde{q}_{\Delta}^{m 1 0}(z,u,x,\xi))|_{u=x}
1283:     \end{multline*}
1284:     where
1285:     \begin{align*}
1286:       \tilde{q}_{\Delta}^{m 1 0}(z,u,x,\xi) =
1287:       \int_0^1 q_\Delta^{m 1 0}(z,(1-r)u+ r x,\xi) \ d r.
1288:     \end{align*}
1289:     As $\tilde{q}_{\Delta}^{m 1 0}$ is bounded \wrt
1290:     $\Delta$ and $z$ with values in $S^1_\rho(\R^n\times \R^n)$
1291:     we obtain the result.
1292:   \end{proof}
1293: %%%%%%%%%%%%%%%%%%%%%%%%
1294: % back to theorem      %
1295: %%%%%%%%%%%%%%%%%%%%%%%%
1296:   {\em End of the proof of Theorem~\ref{theorem:H^s
1297:       estimate}.}
1298:   With the previous lemma we see that
1299:   \begin{align*}
1300:     {\cal F}^b_{(z',z)}={\cal F}^c_{(z',z)}+ \Delta {\cal F}^3_{(z',z)}
1301:   \end{align*}
1302:   where ${\cal F}^c_{(z',z)}$ has for symbol
1303:   $p_\Delta(z,x,x',\xi)$ and $\|{\cal
1304:     F}^3_{(z',z)}\|_{(L^2,L^2)} \leq K^3$ uniformly in $z \in [0,Z]$
1305:   and $\Delta$, $\Delta$ small enough.
1306: 
1307:   To estimate $\|{\cal F}^c_{(z',z)}\|_{(L^2,L^2)}$ we follow the
1308:   procedure at the end of the proof of Theorem 18.1.11 in
1309:   \cite{Hoermander:V3}. Let $A:=1+\Delta$. Define
1310:   \begin{align*}
1311:     \nu_\Delta(z,x',\xi) =
1312:     \sqrt{A^2 - |p_\Delta(z,x',x',\xi)|^2},
1313:   \end{align*}
1314:   which satisfies Property (\ref{eq:Q_L}) by Lemma~\ref{lemma:QL stable by
1315:     f}. Then define $r_\Delta$ by
1316:   \begin{align*}
1317:     \nu_\Delta\ \#\ \nu_\Delta^\ast = A^2 - p_\Delta\ \#\ p_\Delta^\ast - r_\Delta.
1318:   \end{align*}
1319:   Note that
1320:   \begin{align*}
1321:     \nu_\Delta\ \#\ \nu_\Delta^\ast(z,x',\xi) =
1322:     \symb{\nu_\Delta (z,x',\xi) \overline{\nu_\Delta}(z,x,\xi)}
1323:     (z,x,\xi)
1324:   \end{align*}
1325:   It is easy to check that $\nu_\Delta (z,x',\xi)
1326:   \overline{\nu_\Delta}(z,x,\xi)$ satisfies Property (\ref{eq:Q_L})
1327:   for $|\alpha| + |\beta|\geq 1$. The same applies to $p_\Delta
1328:   (z,x',\xi) \overline{p_\Delta}(z,x,\xi)$.  Lemma~\ref{lemma:symb -
1329:   amp} applies and with $m=\rho-\delta$ we thus obtain that $r_\Delta
1330:   = \Delta \tilde{r}_\Delta$ with $\tilde{r}_\Delta$ bounded \wrt $z$
1331:   and $\Delta$ with values $S^0_\rho(X\times \R^n)$.  Thus
1332:   \begin{align*}
1333:     \|{\cal F}^c_{(z',z)}\|_{(L^2,L^2)} =
1334:     \|({\cal F}^c_{(z',z)})^\ast\|_{(L^2,L^2)}
1335:     \leq \sqrt{(1+\Delta)^2 + \Delta C}\leq 1+ \Delta K^4,
1336:   \end{align*}
1337:   for some $K^4>0$ large enough.  We thus obtain that $\|{\cal
1338:     F}_{(z',z)}\|_{(L^2,L^2)} \leq 1 + \Delta K$ where $K=K^1 + K^2 +
1339:   K^3 + K^4$. With the definition of ${\cal F}_{(z',z)}$ it follows
1340:   that
1341:   \begin{align*}
1342:     \|\G_{(z',z)}\|_{(H^{(s)},H^{(s)})} =
1343:     \|(\G_{(z',z)})^\ast\|_{(H^{(s)},H^{(s)})} \leq \sqrt{1 + \Delta K}
1344:   \end{align*}
1345:   which concludes the proof of Theorem~\ref{theorem:H^s estimate}.
1346: \end{proof}
1347: 
1348: We observe that for $\Delta$ small enough, the function $\inp{x'}{\xi}
1349: + \Delta b_1(z,x',\xi)$ satisfies the conditions $(P)$-$(i)$,
1350: $(P)$-$(ii)$, and $(P)$-$(iii)$ in \cite[page 2]{Kumano-go:76}.  With
1351: Lemmas~\ref{lemma: PL with L=2 for c_1} and ~\ref{lemma:exp(-Dq)- QL},
1352: we observe that an FIO with phase function $\phi_{(z',z)}(x',x,\xi)$
1353: and amplitude in $\sigma_A(z,x',\xi)$ in $S^m(X\times \R)$ may
1354: actually be understood as an FIO with real phase $\inp{x'-x}{\xi} +
1355: \Delta b_1(z,x',\xi)$ and amplitude $\sigma_A(z,x',\xi) \exp[-\Delta
1356: c_1(z,x',\xi)]$ in $S^m_\rho(X\times \R)$. Applying Theorem
1357: 2.5 and the following remark in \cite{Kumano-go:76} we obtain
1358: \addtocounter{delta}{1}
1359: \begin{proposition}
1360:   \label{prop: Hs cont any symbol}
1361:   Let $\A_{(z',z)}$ be the global FIO with kernel
1362:   \begin{align*}
1363:     A_{(z',z)}(x',x) = \int \exp[i
1364:     \phi_{(z',z)}(x',x,\xi)]\  \sigma_A(z,x',\xi)\,
1365:     \dslash \xi
1366:   \end{align*}
1367:   with $\sigma_A(z,.)$ bounded \wrt $z$ with values in
1368:   $S^m(X\times \R^{n})$, $m\in \R$.  Then for all $s
1369:   \in \R$ there exists $M=M(s,m)\geq 0$, $\Delta_\thedelta >0$ such that 
1370:   \begin{align*}
1371:     \|\A_{(z',z)}\|_{(H^{(s)},H^{(s-m)})}
1372:     \leq M\  p(\sigma_A(z,.))
1373:   \end{align*}
1374:   for all $z \in
1375:   [0,Z]$, and $0 \leq\Delta\leq \Delta_\thedelta$, where $p(.)$ is
1376:   some appropriately chosen semi-norm in $S^m(X\times \R^{n})$.
1377: \end{proposition}
1378: This proposition could also be proved by adapting the proof of
1379:   Theorem~\ref{theorem:H^s estimate} to this case. Note that in the
1380:   case $\sigma_A = g_{(z',z)}$ we were able, in the proof of
1381:   Theorem~\ref{theorem:H^s estimate}, to achieve a finer estimate. The
1382:   proof heavily relies on the particular structure of the phase
1383:   function and the amplitude that can be taken as ``close'' as we want
1384:   to those of the identity operator by taking $\Delta$ small enough.
1385: %%% Local Variables:
1386: %%% mode: latex
1387: %%% TeX-master: "FIO1"
1388: %%% End:
1389: