1: \documentclass[11pt,twoside,leqno]{aomamlt2e}
2: \pageno{943}
3: \received{April 13, 2005}
4: \allowdisplaybreaks
5:
6: \renewcommand{\thesection}{\arabic{section}}
7: \makeatletter
8: \renewcommand{\@seccntformat}[1]{\csname
9: the#1\endcsname.\hspace{0.5em}\setcounter{Subsec}{0}\setcounter{Subsubsec}{0}}\makeatother
10:
11:
12: \newtheorem{lemma}{Lemma}
13:
14: \newtheorem{prop}{Proposition}
15: \theorembodyfont{\upshape}
16: \newtheorem{defin}{{\it Definition}}
17: \newcommand\sdemo[1]{\demo{\scshape #1}}
18: \begin{document}
19: \currannalsline{162}{2005}
20:
21: \title{Hypoellipticity and loss of derivatives}
22:
23:
24: \acknowledgements{Research was partially supported by
25: NSF Grant DMS-9801626.}
26: \author{J.\ J.\ Kohn}
27:
28: \institution{Princeton University, Princeton, NJ
29: \\
30: \email{kohn@math.princeton.edu}}
31:
32:
33:
34: \shorttitle{Hypoellipticity and loss of derivatives}
35:
36: \centerline{(with an Appendix by Makhlouf Derridj and David S. Tartakoff)}
37: \vskip12pt
38:
39: \centerline{\it Dedicated to Yum-Tong Siu for his $60^{\rm th}$ birthday.}
40:
41: \vglue15pt
42: \centerline{\bf Abstract}
43: \vglue12pt
44:
45: Let $\{X_1,\dots,X_p\}$ be complex-valued vector fields
46: in $\mathbb R^n$ and assume that they satisfy the bracket
47: condition (i.e. that their Lie algebra spans all vector fields).
48: Our object is to study the operator $E=\sum X_i^*X_i$, where $X_i^*$ is
49: the
50: $L_2$ adjoint of $X_i$. A result of H\"ormander is that when the $X_i$
51: are real then $E$ is hypoelliptic and furthemore it is subelliptic (the
52: restriction of a destribution $u$ to an open set $U$ is ``smoother''
53: then the restriction of $Eu$ to $U$). When the $X_i$ are complex-valued
54: if
55: the bracket condition of order one is satisfied (i.e. if the
56: $\{X_i,[X_i,X_j]\}$ span), then we prove that the operator $E$ is still
57: subelliptic. This is no longer true if brackets of higher order are
58: needed to span. For each $k\ge1$ we give an example of two complex-valued
59: vector fields,
60: $X_1$ and $X_2$, such that the bracket condition of order $k+1$ is
61: satisfied and we prove that the operator $E=X_1^*X_1+X_2^*X_2$ is
62: hypoelliptic but that it is not subelliptic. In fact it ``loses'' $k$
63: derivatives in the sense that, for each $m$, there exists a distribution
64: $u$ whose restriction to an open set $U$ has the property that the
65: $D^\alpha Eu$ are bounded on $U$ whenever $|\alpha|\le m$ and for some
66: $\beta$, with
67: $|\beta|=m-k+1$, the restriction of $D^\beta u$ to $U$ is not locally
68: bounded.
69:
70: \section{Introduction}
71:
72:
73:
74: We will be concerned with local $C^\infty$ hypoellipticity in
75: the following sense. A linear differential operator
76: operator $E$ on $\mathbb R^n$ is hypoelliptic if, whenever $u$ is a
77: distribution such that the restriction of $Eu$ to an open
78: set $U\subset\mathbb R^n$ is in $C^\infty(U)$, then the restriction of
79: $u$ to $U$ is also in $C^\infty(U)$. If
80: $E$ is hypoelliptic then it satisfies the following {\it a priori\/}
81: estimates.
82: \begin{enumerate}
83: \item Given open sets $U, U'$ in $\mathbb R^n$ such that
84: $U\subset\bar U\subset U'\subset\mathbb R^n$, a nonnegative
85: integer $p$, and a real number $s_o$, there exist an integer $q$
86: and a constant $C=C(U,p,q,s_o)$ such that
87: $$
88: \sum_{|\alpha|\le p} sup\,_{x\in U}|D^\alpha u(x)|\le
89: C(\sum_{|\beta|\le q}sup\,_{x\in U'}|D^\beta Eu(x)|+\|u\|_{-s_o}),
90: $$
91: for all $u\in C_0^\infty(\mathbb R^n)$.
92: \item Given $\varrho,\varrho'\in C^\infty_0(\mathbb R^n)$ such that
93: $\varrho'=1$ in a neighborhood of ${\rm supp}(\varrho)$, and
94: $s_o, s_1\in\mathbb R$, there exist $s_2\in\mathbb R$ and a
95: constant $C=C(\varrho,\varrho',s_1,s_2,s_0)$ such that
96: $$
97: \|\varrho u\|_{s_1}\le C(\|\varrho'Eu\|_{s_2}+\|u\|_{-s_o}),
98: $$
99: for all $u\in C_0^\infty(\mathbb R^n)$.
100: \end{enumerate}
101: \noindent Assuming that $E$ is hypoelliptic and that $q$ is the
102: smallest integer so that the first inequality above holds (for
103: large
104: $s_o$) then, if $q\le p$, we say that $E$ {\it gains} $p-q$
105: derivatives in the sup norms and if $q\ge p$, we say that $E$
106: {\it loses} $q-p$ derivatives in the sup norms. Similarly, assuming
107: that $s_2$ is the smallest real number so that the
108: second inequality holds (for large $s_o$) then, if $s_2\le s_1$, we say
109: that $E$ {\it gains} $s_1-s_2$ derivatives in the
110: Sobolev norms and if $s_2\ge s_1$, we say that $E$ {\it loses}
111: $s_2-s_1$ derivatives in the Sobolev norms. In particular if
112: $E$ is of order $m$ and if $E$ is elliptic then $E$ gains exactly $m$
113: derivatives in the Sobolev norms and gains exactly
114: $m-1$ derivatives in the sup norms. Here we will present hypoelliptic
115: operators $E_k$ of order 2 which lose exactly $k-1$
116: derivatives in the Sobolev norms and lose at least $k$ derivatives in
117: the sup norms.
118:
119: Loss of derivatives presents a very major difficulty: namely,
120: how to derive the
121: {\it a priori\/} estimates? Such estimates depend on localizing the right-hand side and (because of the loss of derivatives)
122: the errors that arise are apparently always larger then the terms one
123: wishes to estimate. This difficulty is overcome here
124: by the use of subelliptic multipliers in a microlocal setting. In this
125: introduction I would like to indicate the ideas behind
126: these methods, which were originally devised to study hypoellipticity
127: with gain of derivatives. It should be remarked that
128: that for global hypoellipticity the situation is entirely different; in
129: that case loss of derivatives can occur and is well
130: understood but, of course, the localization problems do not arise.
131:
132: We will restrict ourselves to operators $E$ of second order
133: of the form
134: $$
135: Eu=-\sum_{i,j}\frac{\partial}{\partial x_i}a_{ij}\frac{\partial
136: u}{\partial x_j},
137: $$
138: where $(a_{ij})$ is a hermitian form with $C^\infty$ complex-valued
139: components. If at some point $P\in\mathbb R^n$ the form
140: $(a_{ij}(P))$ has two nonzero eigenvalues of different signs then $E$
141: is not hypoelliptic so that, without loss of generality,
142: we will assume that $(a_{ij})\ge0$.
143: \begin{defin} The operator $E$ is subelliptic at $P\in\mathbb R^n$ if
144: there exists a neighborhood $U$ of $P$, a real
145: number $\varepsilon>0$, and a constant $C=(U,\varepsilon)$, such that
146: $$
147: \|u\|^2_\varepsilon\le C(|(Eu,u)|+\|u\|^2),
148: $$
149: for all $u\in C_0^\infty(U)$.
150: \end{defin}
151:
152: Here the Sobolev norm $\|u\|_s$ is defined by
153: $$
154: \|u\|_s=\|\Lambda^su\|,
155: $$
156: and $\Lambda^su$ is defined by its Fourier transform, which is
157: $$
158: \widehat{\Lambda^su}(\xi)=(1+|\xi|^2)^{\frac{s}{2}}\hat u(\xi).
159: $$
160: We will denote by $H^s(\mathbb R^n)$ the completion of
161: $C_0^\infty(\mathbb R^n)$ in the norm $\|\,\|_s$. If
162: $U\subset\mathbb R^n$ is open, we denote by $H^s_{\rm loc}(U)$ the set of
163: all distributions on $U$ such that $\zeta
164: u\in H^s(\mathbb R^n)$ for all $\zeta\in C_0^\infty(U)$. The following
165: result, which shows that subellipticity implies
166: hypoellipticity with a gain of $2\varepsilon$ derivatives in Sobolev
167: norms, is proved in [KN].
168:
169: \sdemo{Theorem} {\it Suppose that $E$ is subelliptic at each
170: $P\in U\subset\mathbb R^n$. Then $E$ is hypoelliptic on
171: $U$. More precisely{\rm ,} if $u\in H^{-s_o}\cap H^s_{\rm loc}(U)$ and if $Eu\in
172: H^s_{\rm loc}(U)${\rm ,} then} $u\in H^{s+2\varepsilon}_{\rm loc}(U)$.
173: \Enddemo
174:
175: In [K1] and [K2] I introduced subelliptic multipliers in
176: order to establish subelliptic estimates for the
177: $\bar\partial$-Neumann problem. In the case of
178: $E$, subelliptic multipliers are defined as follows.
179:
180: \begin{defin} A {\it subelliptic multiplier} for $E$ at $P\in\mathbb
181: R^n$ is a pseudodifferential operator $A$ of order zero,
182: defined on $C_0^\infty(U)$, where $U$ is a neighborhood of $P$, such
183: that there exist $\varepsilon>0$, and a constant
184: $C=C(\varepsilon,P,A)$, such that
185: $$
186: \|Au\|^2_\varepsilon\le C(|(Eu,u)|+\|u\|^2),
187: $$
188: for all $u\in C_0^\infty(U)$.
189: \end{defin}
190:
191: If $A$ is a subelliptic multiplier and if $A'$ is a pseudodifferential
192: operator whose principal symbol equals the principal
193: symbol of $A$ then $A'$ is also a subelliptic multiplier. The existence
194: of subelliptic estimates can be deduced from the
195: properties of the set symbols of subelliptic multipliers. In the case of
196: the $\bar\partial$-Neumann problem this leads to the
197: analysis of the condition of ``D'Angelo finite type.'' Catlin and
198: D'Angelo, in [C] and [D'A], showed that D'Angelo
199: finite type is a necessary and sufficient condition for the
200: subellipticity of the $\bar\partial$-Neumann problem. To
201: illustrate some of these ideas, in the case of an operator~$E$, we will
202: recall H\"ormander's theorem on the sum of squares of
203: vector fields.
204:
205: Let $\{X_1,\dots,X_m\}$ be vector fields on a neighborhood of
206: the origin in $\mathbb R^n$.
207: \begin{defin} The vectorfields $\{X_1,\dots,X_m\}$ satisfy the {\it
208: bracket condition} at the origin if the Lie algebra
209: generated by these vector fields evaluated at the origin is the tangent
210: space.
211: \end{defin}
212:
213: In [Ho], H\"ormander proved the following
214:
215: \sdemo{Theorem} {\it If the vectorfields $\{X_1,\dots,X_m\}$ are
216: real and if they satisfy the {\rm bracket condition}
217: at the origin then the operator $E=\sum X_j^2$ is hypoelliptic in a
218: neighborhood of the origin.}
219: \Enddemo
220:
221: The key point of the proof is to establish that for some
222: neighborhoods of the origin $U$ there exist
223: $\varepsilon>0$ and $C=C(\varepsilon, U)$ such that
224: \begin{equation} \label{S}
225: \|u\|_\varepsilon^2\le C\big(\sum\|X_ju\|^2+\|u\|^2\big),
226: \end{equation}
227: for all $u\in C^\infty_0(U)$.
228: \medskip
229: Here is a brief outline of the proof of estimate (\ref{S}) using
230: subelliptic multipliers. Note that
231: \begin{enumerate}
232: \item[1.] The operators $A_j=\Lambda^{-1}X_j$ are subelliptic multipliers
233: with $\varepsilon=1$, that is
234: $$
235: \|A_ju\|_1^2\le C\big(\sum\|X_ju\|^2+\|u\|^2\big),
236: $$
237: for all $u\in C^\infty_0(U)$.
238: \item[2.] If $A$ is a subelliptic multiplier then $[X_j,A]$ is a
239: subelliptic multiplier. (This is easily seen: we have
240: $X_j^*=-X_j+a_j$ since $X_j$ is real and
241: \begin{align*}
242: \|[X_j,A]u\|^2_{\frac{\varepsilon}{2}}&\le |(X_jAu,R^\varepsilon
243: u)|+|(AX_ju,R^\varepsilon u)|\\
244: &\le |(Au,\tilde R^\varepsilon u)|+O(\|u\|^2)+|(Au,R^\varepsilon
245: X_ju)|+|(AX_ju,R^\varepsilon u)|\\
246: &\le C\left(\|Au\|^2_\varepsilon+\sum\|X_ju\|^2+\|u\|^2\right),
247: \end{align*}
248: where $R^\varepsilon=\Lambda^\varepsilon[X_j,A]$ and $\tilde
249: R^\varepsilon=[X^*_j,R^\varepsilon]$ are pseudodifferential
250: operators of order $\varepsilon$.)
251: \end{enumerate}
252: Now using the bracket condition and the above we see that 1 is a
253: subelliptic multiplier and hence the estimate (\ref{S}) holds.
254:
255: The more general case, where the $a_{ij}$ are real but $E$
256: cannot be expressed as a sum of squares
257: (modulo $L_2$) has been analyzed by Oleinik and Radkevic (see [OR]).
258: Their result can also be obtained by use
259: of subelliptic multipliers and can then be connected to the geometric
260: interpretation given by Fefferman and Phong in
261: [FP]. The next question, which has been studied fairly extensively, is
262: what happens when subellipticity fails and yet there
263: is no loss. A striking example is the operator on $\mathbb R^2$ given by
264: $$
265: E=-\frac{\partial^2}{\partial x^2}-a^2(x)\frac{\partial^2}{\partial
266: y^2},
267: $$
268: where $a(x)\ge0$ when $x\neq0$. This operator was studied by Fedii in
269: [F], who showed that $E$ is always hypoelliptic, no matter
270: how fast $a(x)$ goes to zero as $x\to0$. Kusuoka and Stroock (see [KS])
271: have shown that the operator on $\mathbb R^3$ given by
272: $$
273: E=-\frac{\partial^2}{\partial x^2}-a^2(x)\frac{\partial^2}{\partial
274: y^2}-\frac{\partial^2}{\partial z^2},
275: $$
276: where $a(x)\ge0$ when $x\neq0$, is hypoelliptic if and only if
277: $\lim_{x\to0}\log a(x)=0$.
278: Hypoellipticity when there is no loss but when the gain is smaller than
279: in the subelliptic case has also been studied by Bell and
280: Mohamed [BM], Christ [Ch1], and Morimoto [M]. Using subelliptic
281: multipliers has provided new insights into these results (see
282: [K4]); for example Fedii's result is proved when $a^2$ is replaced by
283: $a$ with the requirement that $a(x)>0$ when $x\neq0$. In
284: the case of the
285: $\bar\partial$-Neumann problem and of the operator $\Box_b$ on CR
286: manifolds, subelliptic multipliers are used to established
287: hypoellipticity in certain situations where there is no loss of
288: derivatives in Sobolev norms but in which the gain is weaker
289: than in the subelliptic case (see [K5]). Stein in [St] shows that the
290: operator $\square_b+\mu$ on the Heisenberg group
291: $\mathcal H\subset\mathbb C^2$, with $\mu\neq0$, is analytic
292: hypoelliptic but does not gain or lose any derivatives. In his thesis
293: Heller (see [He]), using the methods developed by Stein in [St], shows
294: that the fourth order operator $\square_b^2+X$ is
295: analytic hypoelliptic and that it loses derivatives (here $X$ denotes a
296: ``good'' direction). In a recent work, C. Parenti and A. Parmeggiani studied classes of
297: pseudodifferential operators with large losses of derivatives (see [PP1]).
298:
299:
300: The study of subelliptic multipliers has led to the concept of
301: multiplier ideal sheaves (see [K2]). These have had
302: many applications notably Nadel's work on K\"ahler-Einstein metrics
303: (see [N]) and numerous applications to algebraic
304: geometry. In algebraic geometry there are three areas in which
305: multiplier ideals have made a decisive contribution: the
306: Fujita conjecture, the effective Matsusaka big theorem, and invariance
307: of plurigenera; see, for example, Siu's article [S].
308: Up to now the use of subelliptic multipliers to study the
309: $\bar\partial$-Neumann problem and the laplacian $\square_b$ has been
310: limited to dealing with Sobolev norms, Siu has developed a program to
311: use multipliers for the $\bar\partial$-Neumann problem to
312: study H\"older estimates and to give an explicit construction of the
313: critical varieties that control the D'Angelo type. His
314: program leads to the study of the operator
315: $$
316: E=\sum_1^m X^*_jX_j,
317: $$
318: where the $\{X_1,\dots,X_m\}$ are {\it complex} vector fields
319: satisfying the bracket condition. Thus Siu's program
320: gives rise to the question of whether the above operator $E$ is
321: hypoelliptic and whether it satisfies the subelliptic estimate
322: (\ref{S}). These problems raised by Siu have motivated my
323: work on this paper. At first I found that if the bracket
324: condition involves only one bracket then (\ref{S}) holds with
325: $\varepsilon=\frac{1}{4}$ (if the $X_j$ span without taking
326: brackets then $E$ is elliptic). Then I found a series of examples for
327: which the bracket condition is satisfied with $k$
328: brackets, $k>1$, for which (\ref{S}) does not hold. Surprisingly I
329: found that the operators in these examples
330: are hypoelliptic with a loss of $k-1$ derivatives in the Sobolev norms.
331: The method of proof involves calculations with
332: subelliptic multipliers and it seems very likely that it will be
333: possible to treat the more general cases, that is when $E$
334: given by complex vectorfields and, more generally, when $(a_{ij})$ is
335: nonnegative hermitian, along the same lines.
336:
337: The main results proved here are the following:
338:
339: \sdemo{Theorem A} I{\it f $\{X_i,[X_i,X_j]\}$ span the
340: complex
341: tangent space at the origin then a subelliptic estimate is satisfied{\it ,}
342: with
343: $\varepsilon =\frac{1}{2}$.}
344:
345: \sdemo{Theorem B} {\it For $k\ge0$ there exist complex
346: vector fields $X_{1k}$ and
347: $X_2$ on a neighborhood of the origin in $\mathbb R^3$ such that the
348: two vectorfields $\{X_{1k}, X_2\}$ and their
349: commutators of order $k+1$ span the complexified tangent space at the
350: origin{\rm ,} and when $k>0$ the subelliptic estimate
351: {\rm (\ref{S})} does not hold. Moreover{\rm ,} when $k>1${\rm ,} the operator
352: $E_k=X_{1k}^*X_{1k}+X_2^*X_2$ loses $k$ derivatives in the sup
353: norms and $k-1$ derivatives in the Sobolev norms.}\Enddemo
354:
355: Recently Christ (see [Ch2]) has shown that the operators
356: $-\frac{\partial^2}{\partial s^2}+E_k$ on $\mathbb R^4$ are not
357: hypoelliptic when $k>0$.
358:
359: \sdemo{Theorem C} {\it If $X_{1k}$ and $X_2$ are the
360: vectorfields given in Theorem {\rm B} then the operator
361: $E_k=X_{1k}^*X_{1k}+X_2^*X_2$ is hypoelliptic. More precisely{\rm ,} if $u$
362: is a distribution solution of $Eu=f$ with $u\in
363: H^{-s_0}(\mathbb R^3)$ and if
364: $U\subset\mathbb R^3$ is an open set such that $f\in H^{s_2}_{\rm loc}(U)${\rm ,}
365: then $u\in H^{s_2-k+1}_{\rm loc}(U)$.}
366: \Enddemo
367:
368: This paper originated with a problem posed by Yum-Tong Siu. The author
369: wishes to thank Yum-Tong Siu and Michael Christ for
370: fruitful discussions of the material presented here.
371:
372:
373: \demo{Remarks}
374: In March 2005, after this paper had been accepted for publication, I circulated a preprint.
375: Then M. Derridj and D. Tartakoff proved analytic hypoellipticity for the operators constructed here (see [DT]). The work
376: of\break Derridj and Tartakoff used ``balanced'' cutoff functions to estimate the size of
377: derivatives starting with the
378: $C^\infty$ local hypoellipticity proved here; then Bove, Derridj, Tartakoff, and I (see [BDKT])
379: proved
380: $C^\infty$ local hypoellipticity using the balanced cutoff functions, starting from the estimates for functions with
381: compact support proved here. Also at this time, in [PP2], Parenti and Parmeggiani, following their work in
382: [PP1],
383: gave a different proof of hypoellipticity of the operators discussed here and in [Ch2].
384:
385:
386: \section{Proof of Theorem A}
387:
388: The proof of Theorem A proceeds in the same way as given above in the
389: outline of H\"ormander's theorem. It works only when one
390: bracket is involved because (unlike the real case) $\bar X_j$ is not
391: in the span of the $\{X_1,\dots,X_m\}$. The constant $\varepsilon=\frac{1}{2}$ is the largest possible,
392: since (as proved in [Ho]) this is already so when the $X_i$ are real.
393:
394:
395: First note that $\|X^*_iu\|_{-\frac{1}{2}}^2\le
396: \|X_iu\|^2+C\|u\|^2$, since
397: \begin{eqnarray*}
398: \|X^*_iu\|_{-\frac{1}{2}}^2 &=&
399: (X^*_iu,\Lambda^{-1}X^*_iu)=(X^*_iu,P^0u)\\
400: &=&(u,X_iP^0 u)= -(u,P^0X_iu)+O(\|u\|^2);
401: \end{eqnarray*}
402: hence,
403: $$
404: \|X^*_iu\|_{-\frac{1}{2}}^2\le C\big(\sum\|X_ku\|^2+\|u\|^2\big),
405: $$
406: where $P^0=\Lambda^{-1}\bar{X}_i$ is a pseudodifferential operator of
407: order zero. Then we have
408: \begin{eqnarray*}
409: \|X^*_iu\|^2 &=&
410: (u,X_iX^*_iu)=\|X_iu\|^2+(u,[X_i,X^*_i]u)\\
411: &=&\|X_iu\|^2+(\Lambda^{\frac{1}{2}}u,\Lambda^{-\frac{1}{2}}[X_i,X_i^*]u)\\
412: &\le& \|X_iu\|^2+C\|u\|^2_{\frac{1}{2}}.
413: \end{eqnarray*}
414:
415:
416: To estimate $\|u\|^2_{\frac{1}{2}}$ by
417: $C(\sum\|X_ku\|^2+\|u\|^2)$ we will estimate $\|Du\|^2_{-\frac{1}{2}}$ by\break
418: \vglue-10pt\noindent
419: $C(\sum\|X_ku\|^2+\|u\|^2)$ for all first order operators $D$. Thus it
420: suffices to estimate $Du$ when $D=X_i$ and when $D=[X_i,X_j]$. The estimate
421: is clearly satisfied if $D=X_i$, if $D=[X_i,X_j]$ we have
422: \begin{eqnarray*}
423: \|[X_i,X_j]u\|_{-\frac{1}{2}}^2
424: &=&(X_iX_ju,\Lambda^{-1}[X_i,X_j]u)-(X_jX_iu,\Lambda^{-1}[X_i,X_j]u)\\
425: &=&(X_iX_ju,P^0u)-(X_jX_iu,P^0u);
426: \end{eqnarray*}
427: the first term on the right is estimated by
428: \begin{eqnarray*}
429: (X_iX_ju,P^0u) &=&
430: (X_ju,X_i^*P^0u)=-(X_ju,P^0X^*_iu)+O(\|u\|^2+\|X_ju\|^2)\\
431: &\le & C(\|X_ju\|\|X^*_iu\|+\|u\|^2+\|X_ju\|^2)\\ &\le &
432: {\rm l.c.}\sum(\|X_ku\|^2+{\rm s.c.}\|X^*_iu\|^2+C\|u\|^2
433: \end{eqnarray*}
434: and the second term on the right is estimated similarly.
435: Combining these we have
436: $$
437: \|u\|^2_{\frac{1}{2}}\le
438: C(\sum\|\frac{\partial u}{\partial x_i}\|^2_{-\frac{1}{2}}+\|u\|^2)
439: \le C(\sum\|X_ku\|^2+\|u\|^2)+{\rm s.c.}\|u\|^2_{\frac{1}{2}};
440: $$
441: hence
442: $$
443: \|u\|^2_{\frac{1}{2}}\le C\big(\sum\|X_ku\|^2+\|u\|^2\big)
444: $$
445: which concludes the proof of theorem A.
446:
447:
448:
449: \section{The operators $E_k$}
450:
451: In this section we define the operators: $L,\bar L, X_{1k},
452: X_2,$ and $E_k$.
453:
454: Let $\mathfrak H$ be the hypersurface in $\mathbb C^2$
455: given by:
456: $$
457: \Re(z_2)=-|z_1|^2.
458: $$
459: We identify $\mathbb R^3$ with the Heisenberg group represented by
460: $\mathfrak H$ using the mapping
461: $\mathfrak H\to\mathbb R^3$ given by $x=\Re
462: z_1,\, y=\Im z_1,\, t=\Im z_2$. Let $z=x+\sqrt{-1}\,y$. Let
463: $$
464: L=\frac{\partial}{\partial z_1}-2\bar z_1\frac{\partial}{\partial z_2}=
465: \frac{\partial}{\partial z}+\sqrt{-1}\bar z\frac{\partial}{\partial t}
466: $$
467: and
468: $$
469: \bar L=\frac{\partial}{\partial\bar
470: z_1}-2z_1\frac{\partial}{\partial\bar z_2}=
471: \frac{\partial}{\partial\bar z}-\sqrt{-1}\,z\frac{\partial}{\partial t}.
472: $$
473: Let $X_{1k}$ and $X_2$ be the restrictions to $\mathfrak H$ of the
474: operators
475: $$
476: X_{1k}=\bar z_1^kL=\bar z^k\frac{\partial}{\partial z}+\sqrt{-1}\,\bar
477: z^{k+1}\frac{\partial}{\partial t}.
478: $$
479: We set
480: $$
481: X_2=\bar L= \frac{\partial}{\partial\bar
482: z}-\sqrt{-1}\,z\frac{\partial}{\partial t}
483: $$
484: and
485: $$
486: E_k=X^*_{1k}X_{1k}+X^*_2X_2=-\bar L|z|^{2k}L-L\bar L.
487: $$
488:
489: By induction on $j$ we define the commutators $A^j_k$ setting
490: $A^1_k=[X_{1k},X_2]$ and
491: $A^j_k=[A^{j-1}_k,X_2]$. Note that
492: $X_2, A^k_k$ and $A^{k+1}_k$ span the tangent space of~$\mathbb R^3$.
493:
494:
495: \section{Loss of derivatives (part I)}
496:
497:
498: In this section we prove that the subelliptic
499: estimate (\ref{S}) does not hold when $k\ge1$. We also prove a
500: proposition which gives the loss of derivatives in the sup
501: norms which is part of Theorem B. To complete the proof of Theorem B,
502: by establishing loss in the Sobolev norms, we will
503: use additional microlocal analysis of $E_k$, the proof of Theorem B is
504: completed in Section 6.
505:
506:
507: \begin{defin} If $U$ is a neighborhood of the origin then $\varrho\in
508: C_0^\infty(U)$ is real-valued and is defined as follows
509: $\varrho(z,t)=\eta(z)\tau(t)$, where $\eta\in C_0^\infty(\{z\in\mathbb
510: C\mid|z|<2\})$ with $\eta(z)=1$ when $|z|\le 1$ and
511: $\tau\in C_0^\infty(\{t\in\mathbb R\mid|t|<2a\})$ with $\tau(t)=1$ when
512: $|t|\le a$.
513: \end{defin}
514:
515: The following proposition shows that the subelliptic estimate
516: (\ref{S}) does not hold when $k>0$.
517:
518:
519: \begin{prop} If $k\ge1$ and if there exist a neighborhood $U$ of the
520: origin and constants $s$ and $C$ such that
521: $$
522: \|u\|^2_s\le C(\|\bar z^kLu\|^2+\|\bar Lu\|^2),
523: $$
524: for all $u\in C_0^\infty(U)${\rm ,} then $s\le0$.
525: \end{prop}
526:
527: {\it Proof.} Let $\lambda_0$ and $a$ be sufficiently large so
528: that the support of $\varrho(\lambda z,t)$ lies in
529: $U$ when $\lambda\ge\lambda_0$. We define $g_\lambda$ by
530: $$
531: g_\lambda(z,t)=\varrho(\lambda
532: z,t)\exp(-\lambda^{\frac{5}{2}}(|z|^2-it)).
533: $$
534: Note that $L\eta(z)=\bar L\eta(z)=0$ when $|z|\le1$, that
535: $L(\tau)=i\bar z\tau'$, and that $\bar L(\tau)=-iz\tau'$.
536: Setting $R^\lambda v(z,t)=v(\lambda z,t)$, we have:
537: \begin{align} \label{Lg}
538: \bar z^kL(g_\lambda)&=(\lambda\bar z^k(R^\lambda L\eta)\tau+i\bar
539: z(R^\lambda\eta)\tau' +\lambda^\frac{5}{2}\bar
540: zR^\lambda\varrho)\exp(-\lambda^\frac{5}{2}(|z|^2+it))
541: \end{align}
542: and
543: \begin{equation} \label{barLg}
544: \bar L(g_\lambda)=(\lambda (R^\lambda\bar
545: L\eta)\tau-iz(R^\lambda\eta)\tau')\exp(-\lambda^\frac{5}{2}(|z|^2+it)).
546: \end{equation}
547: Note that the restriction of $|g_\lambda|$ to $\mathfrak H$ is
548: $$
549: |g_\lambda(z,t)|=\varrho(\lambda z,t)\exp(-\lambda^\frac{5}{2}|z|^2).
550: $$
551: Now we have, using the changes of variables: first $(z,t)\mapsto
552: (\lambda^{-1}z,t)$ and then
553: $z\mapsto \lambda^{-\frac{1}{4}}z$
554: \begin{eqnarray*}
555: \|g_\lambda\|^2&=&\frac{C}{\lambda^2}\int_{\mathbb
556: R^2}\eta(z)^2\exp(-2\lambda^\frac{1}{2}|z|^2) dxdy\\
557: &\ge&\frac{C}{\lambda^2}\int_{\mathbb
558: R^2}\exp(-2\lambda^\frac{1}{2}|z|^2)dxdy
559: -\frac{C}{\lambda^2}\int_{|z|\ge1}\exp(-2\lambda^\frac{1}{2}|z|^2)dxdy\\
560: &\ge&\frac{C}{\lambda^\frac{5}{2}}-
561: \frac{C}{\lambda^2}\exp(-\lambda^{\frac{1}{2}})\int_{\mathbb
562: R^2}\exp(-\lambda^{\frac{1}{2}}|z|^2)dxdy\\
563: &\ge&\frac{C}{\lambda^{\frac{5}{2}}}-
564: \frac{C}{\lambda^{\frac{5}{2}}}\exp(-\lambda^{\frac{1}{2}}).
565: \end{eqnarray*}
566: Then we have
567: $$
568: \|g_\lambda\|^2\ge
569: \frac{{\rm const}.}{\lambda^{\frac{5}{2}}}
570: $$
571: for sufficiently large $\lambda$. Further, using the above coordinate
572: changes to estimate the individual terms in (\ref{Lg}) and
573: in (\ref{barLg}), we have
574: \begin{align*}
575: &\|\bar z^k\lambda (R^\lambda
576: L\eta)\tau\exp(-\lambda^{\frac{5}{2}}(|z|^2-it)\|^2+
577: \|\lambda (R^\lambda\bar
578: L\eta)\tau\exp(-\lambda^{\frac{5}{2}}(|z|^2-it)\|^2\\
579: &\qquad\le
580: C\exp(-\lambda^{\frac{1}{2}})\int_{|z|\ge1}\exp(-
581: \lambda^{\frac{1}{2}}|z|^2)dxdy\le
582: \frac{C}{\lambda^{\frac{1}{2}}}\exp(-\lambda^{\frac{1}{2}}),\end{align*}
583: \begin{eqnarray*}
584: &&\hskip-36pt \||z|(R^\lambda\eta)\tau')\exp(-\lambda^\frac{5}{2}(|z|^2+it))\|^2\\
585: &&\qquad\le
586: \frac{C}{\lambda^2}\int|z|^2\exp(-2\lambda^{\frac{1}{2}}|z|^2)dxdy
587: \le \frac{C}{\lambda^4},\\
588: \noalign{\noindent \rm and}
589: &&\hskip-36pt
590: \|\lambda^\frac{5}{2}\bar
591: z^{k+1}R^\lambda\varrho)\exp(-\lambda^\frac{5}{2}(|z|^2+it))\|^2\\
592: &&\qquad\le
593: C\lambda^{1-2k}\int|z|^{2k+2}\exp(-2\lambda^{\frac{1}{2}}|z|^2)dxdy\le
594: \frac{C}{\lambda^{\frac{5k}{2}}}.
595: \end{eqnarray*}
596: Hence, if $k\ge1$, we have
597: $$
598: \|\bar z^kLg_\lambda\|^2+\|\bar Lg_\lambda\|^2\le
599: \frac{C}{\lambda^{\frac{5}{2}}}.
600: $$
601: Since $|x|\le\frac{2}{\lambda}$ on the support of $g_\lambda$ then we
602: conclude, from the lemma proved below, that given
603: $\varepsilon$ there there exists $C$ such that
604: $$
605: \lambda^\varepsilon\|g_\lambda\|\le C\|g_\lambda\|_\varepsilon,
606: $$
607: for sufficiently large $\lambda$. It the follows that, if $k\ge1$ then
608: the subelliptic estimate
609: $$
610: \|g_\lambda\|_\varepsilon^2\le C(\|\bar z^kLg_\lambda\|^2+\|\bar
611: Lg_\lambda\|^2)
612: $$
613: implies that $\lambda^{2\varepsilon-\frac{5}{2}}\le
614: C\lambda^{-\frac{5}{2}}$ which is a contradiction and thus the
615: proposition
616: follows. The following lemma then completes the proof. For completeness
617: we include a proof which is along the lines given in [ChK].
618: \bigskip
619: \begin{lemma}
620: Let $Q_\delta$ denote a bounded open set contained in the \/{\rm ``}\/slab\/{\rm ''}\break
621: $\{x\in\mathbb R^n\mid|x_1|\le\delta\}$. Then{\rm ,}
622: for each
623: $\varepsilon>0${\rm ,} there exists
624: $C=C(\varepsilon)>0$ such that
625: \begin{equation} \label{CS}
626: \|u\|\le C\delta^\varepsilon\|u\|_\varepsilon,
627: \end{equation}
628: for all $u\in C^\infty_0(Q_\delta)$ and $\delta>0$.
629: \end{lemma}
630:
631: \Proof Note that the general case follows from the case
632: of $n=1$. Since, writing $x=(x_1,x')$, if for each fixed
633: $x'$ we have $\|u(\cdot,x')\|\le
634: C\delta^\varepsilon\|u(\cdot,x')\|_\varepsilon$ then, after integrating
635: with respect to
636: $x'$ and noting that\\
637: $(1+\xi_1^2)^\varepsilon\le(1+|\xi|^2)^\varepsilon$, we obtain the
638: desired estimate. So we will assume
639: that $n=1$ and set $x=x_1$ and $\xi=\xi_1$. We define $\||u\||_s$ by
640: $$
641: \||u\||_s^2=\int|\xi|^{2s}|\hat u(\xi)|^2d\xi.
642: $$
643: We will show that, if $s\ge0$, there exists a constant $C$ such that
644: $$
645: \|u\|_s\le C\||u\||_s,
646: $$
647: for all $u\in C_0^\infty((-1,1))$.
648: First we have
649: $$
650: |\hat u(\xi)|=|\int e^{-ix\xi}u(x)dx|\le \sqrt{2}\|u\|.
651: $$
652: Next, if $|\xi|\le a\le1$,
653: $$
654: (1+\xi^2)^s|\hat u(\xi)|^2\le2^{s+1}\|u\|^2
655: $$
656: and
657: \begin{multline*}
658: \int_{-\infty}^{\infty}(1+\xi^2)^s|\hat u(\xi)|^2d\xi\\=\int_{|\xi|\le
659: a}\dots
660: +\int_{|\xi|>a}\dots
661: \le2^{s+2}a\|u\|^2+\left(\frac{1}{a^2}+1\right)^s\||u\||_s^2.
662: \end{multline*}
663: Hence if $a$ is small we obtain $\|u\|_s\le C\||u\||_s$, as required.
664: If ${\rm supp}(u)\subset(-\delta,\delta)$ then
665: set $u_\delta(x)=u(\delta x)$ so that ${\rm supp}( u_\delta)\subset(-1,1)$.
666: Now
667: $$
668: \hat{u_\delta}(\xi)=\frac{1}{\delta}\hat u\left(\frac{\xi}{\delta}\right)
669: $$
670: so that $\|u_\delta\|^2=\frac{1}{\delta}\|u\|^2$ and
671: $\||u_\delta\||^2_s=\delta^{2s-1}\||u\||^2_s$ which concludes the proof.
672: \Enddemo
673:
674: Next we prove that $E_k$ loses at least $k$ derivatives in
675: the sup norms.
676:
677:
678: \begin{prop} If for some open sets $U$ and $U'${\rm ,} with $\bar U\subset
679: U'${\rm ,} and for each $s_0$ there exists a constant $C=C(s_0)$
680: such that
681: \begin{equation} \label{sup}
682: \sum_{|\alpha|\le p}\sup_{x\in U}|D^\alpha u(x)|\le C\Big(\sum_{|\beta|\le
683: q}\sup_{x\in U'}|D^\beta E_ku(x)|
684: +\|u\|_{-s_0}\Big),
685: \end{equation}
686: for all $u\in C_0^\infty(\mathbb R^3)${\rm ,} then $q\ge p+k$.
687: \end{prop}
688:
689: \Proof If $\delta>0$ define $u_\delta$ by
690: $$
691: u_\delta=(|z|^2-\sqrt{-1}t)^p\log(|z|^2+\delta-\sqrt{-1}t),
692: $$
693: where $\log$ denotes the branch of the logarithm that takes reals into
694: reals. Since $u_\delta$ is the restriction of
695: $(-z_2)^p\log(-z_2+\delta)$ to $\mathfrak H$ we have $\bar
696: Lu_\delta=0$. Then we have
697: $$
698: {\rm lim}_{\delta\to0}|D^p_tu_\delta(0)|=\infty.
699: $$
700: Further
701: \begin{align*}
702: E_ku_\delta&=-\bar
703: L|z|^{2k}Lu_\delta\\
704: &=2k|z_1|^{2k}\!\!\left(-p(-z_2)^{p-1}\log(-z_2+\delta)+(-
705: z_2)^p\log(-z_2+\delta)+
706: \frac{(-z_2)^p}{(-z_2+\delta)}\right)\\
707: &=2k|z|^{2k}\!\!\left(p(|z|^2-\sqrt{-1}t)^{p-1}\log(|z|^2+\delta-\sqrt{-
708: 1}t)+
709: \frac{(|z|^2-\sqrt{-1}t)^p}{|z|^2+\delta-\sqrt{-1}t}\right).
710: \end{align*} \pagebreak
711:
712: \noindent
713: Note that $\|u_\delta\|_{-s_0}$ is bounded independently of $\delta$
714: when $s_0\ge3$.
715: Thus, when $q\le p+k-1$, we have
716: $$
717: \sum_{|\beta|\le q}\sup_{x\in U'}|D^\beta E_ku_\delta(x)|\le {\rm const}.
718: $$
719: with the constant independent of $\delta$. Hence, applying (\ref{sup})
720: to $u_\delta$ we obtain
721: $q\ge p+k$. This concludes the proof of the proposition.
722:
723:
724: \section{Notation}
725:
726: In this section we set down some notation which will be used throughout
727: the rest of the paper.
728: \begin{enumerate}
729: \item[1.] Associated to the cutoff function $\varrho$ defined in Definition
730: 1, is a $C^\infty$ function $\mu$ such that
731: $L\varrho=\bar z\mu$ and $\bar L\varrho=z\bar\mu$ (Such a $\mu$ exists
732: since
733: $$
734: L\varrho(z,t)=D_z\eta(z)\tau(t)+i\bar z\eta(z)D_t\tau(t).
735: $$
736: Since $D_z\eta(z)=0$ in a neighborhood of $z=0$ we can set
737: $\mu(z,t)=\frac{D_z\eta(z)}{\bar z}\tau(t)+i\eta(z)D_t\tau(t)$.)
738: \item[2.] Given cutoff functions $\varrho, \varrho'$, as in Definition 1,
739: with $\varrho'=1$ in a neighborhood of the support of
740: $\varrho$, then we denote by $\{\varrho_i\}$ a special sequence of
741: cutoff functions, each of which satisfies Definition 1 and
742: such that: $\varrho_1=\varrho$, $\varrho'=1$ in a neighborhood of
743: $\bigcup\varrho_i$, and $\varrho_{i+1}=1$ in a neighborhood of
744: the support of $\varrho_i$.
745: \item[3.] The abbreviations ``${\rm s.c.}$'' and ``${\rm l.c.}$'' will be used for
746: ``small constant'' and ``large constant'', respectively in the
747: following sense. $\mathcal Au\le {\rm s.c.}\mathcal Bu+{\rm l.c.}\mathcal Cu$ means
748: that given any constant ${\rm s.c.}$ there exists a constant
749: ${\rm l.c.}$ such that the inequality holds for all $u$ in some specified
750: class.
751: \item[4.] We will use $\|u\|_{-\infty}$ to denote the following. Given
752: $\mathcal Au$, the expression $\mathcal Au\le
753: \|u\|_{-\infty}$ means that: if for any $s_o$ there exist a constant
754: $C=C(s_o)$ such that $\mathcal Au\le C\|u\|_{-s_o}$
755: holds for all $u$ in some specified class.
756: \end{enumerate}
757:
758:
759: \section{Microlocalization on the Heisenberg group}
760:
761:
762: Denote by $T$ the vector field defined by
763: $$
764: T=\frac{1}{\sqrt{-1}}\frac{\partial}{\partial t}.
765: $$
766: Then
767: $$
768: [L,\bar L]=[\frac{\partial}{\partial z}+\sqrt{-1}\,\bar
769: z\frac{\partial}{\partial t},\,\frac{\partial}{\partial\bar
770: z}-\sqrt{-1}\,z\frac{\partial}{\partial t}]=2T.
771: $$
772: The following simple formula, which is obtained by integration by
773: parts, is the starting point of all the estimates connected
774: with the operators $E_k$.
775:
776:
777: \begin{lemma} For $u\in C^\infty_0(\mathbb R^3)$ we have
778: \begin{equation} \label{basic}
779: \|Lu\|^2=2(Tu,u)+\|\bar Lu\|^2.
780: \end{equation}
781: \end{lemma}
782: \vglue8pt
783:
784: \Proof Since $L^*=-\bar L$ and $\bar L^*=-L$, we have
785: $$
786: \|Lu\|^2=(Lu,Lu)=-(\bar LLu,u)=-([\bar L,L]u,u)-(L\bar
787: Lu,u)=2(Tu,u)+\|\bar Lu\|^2.
788: $$
789: We set $x_1=\Re z, x_2=\Im z,$ and $x_3=t$ and denote the dual
790: coordinates by $\xi_1,\xi_2,$ and $\xi_3$.
791: For $(\alpha,t_0)\in\mathbb C\times\mathbb R$ we define
792: $$
793: z^\alpha=z-\alpha \ \ {\rm and}\ \
794: x_3^\alpha=-2\alpha_2x_1+2\alpha_1x_2+x_3-t_0,
795: $$
796: where $\alpha_1=\Re\alpha$ and $\alpha_2=\Im\alpha$. Then
797: $$
798: L=\frac{\partial}{\partial z^\alpha}+i\bar
799: z^\alpha\frac{\partial}{\partial x_3^\alpha}
800: $$
801: and
802: $$
803: \bar L=\frac{\partial}{\partial\bar
804: z^\alpha}-iz^\alpha\frac{\partial}{\partial x_3^\alpha}.
805: $$
806: We set $x_1^\alpha=x_1-\alpha_1$, $x_2^\alpha=x_2-\alpha_2$, and
807: $x^\alpha=(x^\alpha_1,x^\alpha_2,x^\alpha_3)$. Let $\mathcal
808: F_\alpha$ denote the the Fourier transform in the
809: $x^\alpha_j$ coordinates; that is
810: $$
811: \mathcal F_\alpha u(\xi)=\int
812: e^{-ix^\alpha\cdot\xi}u(x^\alpha)dx^\alpha_1x^\alpha_2x^\alpha_3.
813: $$
814:
815:
816: \begin{defin} Let $S^2=\{\xi\in\mathbb R^3\mid |\xi|=1\}$ be the unit
817: sphere.
818: Suppose that $\mathcal U, \mathcal U_1$ are open subsets of $S^2$ with
819: $\bar{\mathcal U}_1\subset\mathcal U$.
820: For each such pair of open sets we define a set of $\gamma\in
821: C^\infty(\mathbb R^3)$, with $\gamma\ge0$, such that
822: \begin{enumerate}
823: \item[1.] $\gamma(\frac{\xi}{|\xi|})=\gamma(\xi)$ when $|\xi|\ge1$.
824: \item[2.] $\gamma(\xi)=1$ when $\xi\in\mathcal U_1$.
825: \item[3.] $\gamma(\xi)=0$ when $\xi\in S^2-\mathcal U$.
826: \end{enumerate}
827: To such a $\gamma$ and $\alpha\in\mathbb C$ we associate the operator
828: $\Gamma_\alpha$ defined by
829: $$
830: \mathcal F_\alpha\Gamma_\alpha u(\xi)=\gamma(\xi)\mathcal F_\alpha
831: u(\xi).
832: $$
833: Let $\mathcal U^+,\mathcal U_1^+,
834: \mathcal U^0,\mathcal U^0_1,\mathcal U^-$, and $\mathcal U^-_1$ be open
835: subsets of $S^2$ defined as follows.
836: \begin{align*}
837: &\mathcal U^+=\Big\{\xi\in S^2\mid \xi_3>\frac{5}{9}\Big\},\ \ \ \mathcal
838: U^+_1=\Big\{\xi\in S^2\mid \xi_3>\frac{4}{9}\Big\},\\
839: &\mathcal U^0=\Big\{\xi\in S^2\mid |\xi_3|<\frac{5}{6}\Big\},\ \ \ \mathcal
840: U^0_1=\Big\{\xi\in S^2\mid |\xi_3|<\frac{2}{3}\Big\},\\*
841: &\mathcal U^-=\{\xi\in S^2\mid -\xi\in\mathcal U^+\},\ {\rm and}\ \
842: \mathcal U^-_1= \{\xi\in S^2\mid -\xi\in\mathcal U^+_1 \}.
843: \end{align*}
844: We denote by $\gamma^+$, $\gamma^0$, and $\gamma^-$ the corresponding
845: functions and require further that
846: $\gamma^+(\xi)=\gamma^-(\xi)=0$ when $|\xi|\le\frac{1}{2}$ and
847: $\gamma^0(\xi)=1$ when $\frac{\xi}{|\xi|}\in\mathcal U^0_1$. The
848: sets of these functions will be denoted by $\mathcal G^+$, $\mathcal
849: G^0$, and $\mathcal G^-$, respectively. The corresponding
850: operators are denoted by $\Gamma^+_\alpha$, $\Gamma^0_\alpha$, and
851: $\Gamma^-_\alpha$. The sets of these operators will be
852: denoted by $\mathfrak G^+_\alpha$, $\mathfrak G^0_\alpha$, and
853: $\mathfrak G^-_\alpha$, respectively. Given
854: $(\alpha,t_0)\in\mathbb C\times\mathbb R$ the functions
855: $\Gamma^+_\alpha u$, $\Gamma^0_\alpha u$, and $\Gamma^-_\alpha u$ will
856: be referred to as microlocalizations of $u$ at $(\alpha,t_0)$ in the
857: regions $+$, $0$, and $-$, respectively.
858: \end{defin}
859:
860: The following lemma shows that the $0$ microlocalization is elliptic
861: for the operators $L$ and $\bar L$.
862: In our estimates we will often encounter error terms which can be
863: bounded by $C_{s_0}\|u\|_{-s_0}$ for every
864: $s_0$; abusing notation we will bound such terms by
865: $``\|u\|_{-\infty}"$.
866:
867: \begin{lemma} If $U$ is a neighborhood of $(\alpha,t_0)$ and if
868: $\gamma^0,\tilde\gamma^0\in\mathcal G^0$ with
869: $\tilde\gamma^0=1$ in a neighborhood of the support of $\gamma^0$ then
870: there exist constants $a>0$ and
871: $C>0$ such that{\rm ,} if $|z-\alpha|<a$ on $U${\rm ,} then
872: $$
873: \|\Gamma^0_\alpha u\|_1\le C(\|\Gamma^0_\alpha
874: Lu\|+\|\tilde\Gamma^0_\alpha u\|+\|u\|_{-\infty})
875: $$
876: and
877: $$
878: \|\Gamma^0_\alpha u\|_1\le C(\|\Gamma^0_\alpha\bar
879: Lu\|+\|\tilde\Gamma^0_\alpha u\|+\|u\|_{-\infty}),
880: $$
881: for all $u\in C^\infty_0(U)$.
882: \end{lemma}
883:
884: \Proof If $\xi\in\mathcal U^0$ and if $|\xi|\ge1$ then
885: $|\xi_3|\le\frac{5}{6}|\xi|$. Thus, if $\xi\in\mathcal U^0$,
886: then $|\xi|\le6(|\xi_1|+|\xi_2|)+1$. Now,
887: $$
888: \|\Gamma^0_\alpha u\|^2_1\le C\Big(\sum_1^2\|\frac{\partial}{\partial
889: x^\alpha_j}\Gamma^0_\alpha u\|^2+
890: \|\tilde\Gamma^0_\alpha u\|^2+\|u\|_{-\infty}^2\Big).
891: $$
892: Let $U'\supset\bar U$ be an open set such that $|z-\alpha|>2a$ on $U'$
893: and let $\varphi\in C_0^\infty(U')$ satisfying
894: $\varphi=1$ in a neighborhood of $\bar U$. Then
895: \begin{align*}
896: \|\Gamma^0_\alpha u\|^2_1&\le C\Big(\sum_1^2\|\frac{\partial}{\partial
897: x^\alpha_j}\Gamma^0_\alpha\varphi u\|^2+
898: \|\tilde\Gamma^0_\alpha u\|^2+\|u\|_{-\infty}^2\Big)\\
899: &\le C'\Big(\sum_1^2\|\frac{\partial}{\partial
900: x^\alpha_j}\varphi\Gamma^0_\alpha u\|^2+
901: \|\tilde\Gamma^0_\alpha u\|^2+\|u\|_{-\infty}^2\Big)\\
902: &\le C''(\|L\varphi\Gamma^0_\alpha u\|^2+\|\bar L\varphi\Gamma^0_\alpha
903: u\|^2\\
904: & \phantom{\le C''(} +
905: \max_{U'}|z-\alpha|^2\|\frac{\partial}{\partial
906: x^\alpha_3}\Gamma^0_\alpha u\|^2+
907: \|\tilde\Gamma^0_\alpha u\|^2+\|u\|_{-\infty}^2)\\
908: &\le C''(\|\Gamma^0_\alpha Lu\|^2+\|\Gamma^0_\alpha\bar Lu\|^2+
909: 4a^2\|\Gamma^0_\alpha u\|_1^2+\|\tilde\Gamma^0_\alpha
910: u\|^2+\|u\|_{-\infty}^2).
911: \end{align*}
912: Hence, taking $a$ suitably small we obtain
913: $$
914: \|\Gamma^0_\alpha u\|^2_1\le C(\|\Gamma^0_\alpha
915: Lu\|^2+\|\Gamma^0_\alpha\bar L u\|^2+
916: \|\tilde\Gamma^0_\alpha u\|^2+\|u\|_{-\infty}^2).
917: $$
918: Furthermore, substituting $\varphi\Gamma^0_\alpha u$ for $u$ in
919: (\ref{basic}), we have
920: \begin{align*}
921: \|L\varphi\Gamma^0_\alpha u\|^2&=2(T\varphi\Gamma^0_\alpha
922: u,\varphi\Gamma^0_\alpha u)+\|\bar L\varphi\Gamma^0_\alpha u\|^2\\
923: &\le {\rm s.c.}\|\frac{\partial}{\partial x^\alpha_3}\Gamma^0_\alpha u\|^2+
924: {\rm l.c.}(\|\tilde\Gamma^0_\alpha
925: u\|^2+\|u\|_{-\infty}^2)+C\|\Gamma^0_\alpha\bar Lu\|^2\\
926: &\le {\rm s.c.}\|\Gamma^0_\alpha u\|^2_1+{\rm l.c.}(\|\tilde\Gamma^0_\alpha
927: u\|^2+\|u\|_{-\infty}^2)+C\|\Gamma^0_\alpha\bar Lu\|^2,
928: \end{align*}
929: and since
930: $$
931: \|L\varphi\Gamma^0_\alpha u\|^2\le C(\|\Gamma^0_\alpha
932: Lu\|^2+\|\tilde\Gamma^0_\alpha u\|^2+\|u\|_{-\infty}^2)
933: $$
934: we get
935: $$
936: \|\Gamma^0_\alpha u\|_1\le C(\|\Gamma^0_\alpha\bar
937: Lu\|+\|\tilde\Gamma^0_\alpha u\|^2+\|u\|_{-\infty}^2).
938: $$
939: Similarly we obtain
940: $$
941: \|\Gamma^0_\alpha u\|_1\le C(\|\Gamma^0_\alpha
942: Lu\|+\|\tilde\Gamma^0_\alpha u\|^2+\|u\|_{-\infty}^2).
943: $$
944: This completes the proof of the lemma.
945:
946:
947: \begin{lemma}If $R^s$ is a pseudodifferential operator of order $s$
948: then there exists $C$ such that
949: \begin{eqnarray*}
950: \|[R^s,\Gamma^+_\alpha]u\|&\le &C(\|\Gamma^0_\alpha
951: u\|_{s-1}+\|u\|_{-\infty})\\
952: \noalign{\noindent and}
953: |[R^s,\Gamma^-_\alpha]u\|&\le &C(\|\Gamma^0_\alpha
954: u\|_{s-1}+\|u\|_{-\infty}).
955: \end{eqnarray*}
956: \end{lemma}
957: \vskip8pt
958:
959: \Proof Since $\gamma^0=1$ on a neighborhood of the support of the
960: derivatives of $\gamma^+$ it also equals one on a
961: neighborhood of the support of the symbol of $[R^s,\Gamma^+_\alpha]$.
962: Hence
963: $[R^s,\Gamma^+_\alpha]=[R^s,\Gamma^+_\alpha]\Gamma^0_\alpha+R^{-
964: \infty}$, where
965: $R^{-\infty}$ is a pseudodifferential operator whose symbol is
966: identically zero. The same argument works for the term
967: $[R^s,\Gamma^-_\alpha]$ and the lemma follows.
968:
969: \begin{defin} For each $s\in\mathbb R$ we define the operator
970: $\Psi^s_\alpha$ as follows. Let $\mathcal U^*$ and
971: $\mathcal U^*_1$ be open sets in $S^2$ such that $\mathcal U^*=\{\xi\in
972: S^2\mid |\xi_3|>\frac{1}{6}$ and
973: $\mathcal U_1^*=\break\{\xi\in S^2\mid |\xi_3|>\frac{1}{3}$. Let $\gamma^*$
974: be the function on $R^3$ associated with
975: $\mathcal U^*,\ \mathcal U^*$ such that $\gamma^*(\xi)=0$ when
976: $|\xi|\le\frac{1}{3}$ and $\gamma^*(\xi)=1$ in the region
977: $\{\xi\in\mathbb R^3 \mid \frac{\xi}{|\xi|}\in\mathcal U^*_1\ {\rm
978: and}\ |\xi|\ge\frac{1}{2}\}$. Then we set
979: $\psi^s(\xi)=(1+|\xi_3|^2)^{\frac{s}{2}}\gamma^*(\xi)$ and define
980: $\Psi^s_\alpha$ by
981: $$
982: \mathcal F_\alpha\Psi^s_\alpha u(\xi)=\psi^s(\xi)\mathcal F_\alpha
983: u(\xi).
984: $$
985: \end{defin}
986:
987: Note that there exist positive constants $c$ and $C$ such that
988: $$
989: c(1+|\xi|^2)^{\frac{s}{2}}\gamma^*(\xi)\le\psi^s(\xi)\le
990: C(1+|\xi|^2)^{\frac{s}{2}}\gamma^*(\xi).
991: $$
992: Hence $\|\Psi^s_\alpha\Gamma^+_\alpha u\|\sim\|\Gamma^+_\alpha u\|_s$
993: and $\|\Psi^s_\alpha\Gamma^-_\alpha u\|\sim\|\Gamma^-_\alpha u\|_s$; by
994: $\sim$ we mean that they differ by an operator of
995: order $-\infty$. Also, since $\gamma^*=1$ on the supports of $\gamma^+$
996: and $\gamma^-$, we have
997: $$
998: \Psi^s\Psi^{s'}\Gamma^+_\alpha\sim\Psi^{s+s'}\Gamma^+_\alpha\ {\rm
999: and}\ \Psi^s\Psi^{s'}\Gamma^-_\alpha\sim
1000: \Psi^{s+s'}\Gamma^-_\alpha.
1001: $$
1002:
1003: \begin{lemma}There exists $C$ such that
1004: $$
1005: \|\Gamma^+_\alpha\bar Lu\|^2+\|\Gamma^+_\alpha u\|_{\frac{1}{2}}^2\le
1006: C(\|\Gamma^+_\alpha Lu\|^2+\tilde\Gamma^+_\alpha
1007: u\|^2+\|u\|_{-\infty}^2),
1008: $$
1009: and
1010: $$
1011: \|\Gamma^-_\alpha Lu\|^2+\|\Gamma^-_\alpha u\|_{\frac{1}{2}}^2\le
1012: C(\|\Gamma^-_\alpha\bar Lu\|^2+\tilde\Gamma^-_\alpha
1013: u\|^2+\|u\|_{-\infty}^2),
1014: $$
1015: for all $u\in C_0^\infty(U)$.
1016: \end{lemma}
1017: \Proof Taking $\varphi\in C_0^\infty$ with $\varphi=1$ in a
1018: neighborhood of $\bar U$ we substitute
1019: $\varphi\Gamma^+_\alpha u$ for $u$ in (\ref{basic}) and obtain
1020: $$
1021: \|L\varphi\Gamma^+_\alpha u\|^2=2(T\varphi\Gamma^+_\alpha
1022: u,\varphi\Gamma^+_\alpha u)+
1023: \|\bar L\varphi\Gamma^+_\alpha u\|^2.
1024: $$
1025: Now, we have
1026: $$
1027: (T\varphi\Gamma^+_\alpha u,\varphi\Gamma^+_\alpha
1028: u)=(T\Gamma^+_\alpha\varphi u,\varphi\Gamma^+_\alpha u)+
1029: O(\tilde\Gamma^+_\alpha u\|^2+\|u\|_{-\infty}^2).
1030: $$
1031: Since $\mathcal F_\alpha(Tu)=\xi_3\mathcal F_\alpha(u)$ we have
1032: $T\Gamma^+_\alpha\sim\Psi^1\Gamma^+_\alpha\sim
1033: \Psi^{\frac{1}{2}}\Psi^{\frac{1}{2}}\Gamma^+_\alpha$ and
1034: $$
1035: (T\varphi\Gamma^+_\alpha u,\varphi\Gamma^+_\alpha u)=
1036: \|\Psi^{\frac{1}{2}}_\alpha\Gamma^+_\alpha
1037: u\|^2+O(\tilde\Gamma^+_\alpha u\|^2+\|u\|_{-\infty}^2).
1038: $$
1039: This proves the first part of the lemma, the second follows from the
1040: fact that $|\xi_3|\gamma^-(\xi)=-\xi_3\gamma^+(-\xi)$.
1041: Then
1042: $\Psi^1\Gamma^-_\alpha\sim\Psi^{\frac{1}{2}}\Psi^{\frac{1}{2}}\Gamma^-
1043: _\alpha$, thus concluding the proof.
1044:
1045:
1046: \section{Loss of derivatives (part II) \\ Conclusion of the proof of
1047: Theorem B}
1048:
1049: In this section we conclude the proof of Theorem B by showing
1050: that if $k\ge2$ then $E_k$ loses at least $k-1$
1051: derivatives in the Sobolev norms.
1052:
1053:
1054: \begin{prop} Suppose that there exist two neighborhoods of the origin
1055: $U$ and $U'${\rm ,} with $\bar U\subset U'${\rm ,} and
1056: real numbers $s_1$ and $s_2$ such that if $\varrho,\varrho'\in
1057: C_0^\infty(U')$ with $\varrho=1$ on $U$ and
1058: $\varrho'=1$ in a neighborhood of the support of $\varrho${\rm ,} and if for
1059: any real number $s_0$ there exists a constant
1060: $C=C(\varrho,\varrho',s_0)$ such that
1061: \begin{equation} \label{hypest}
1062: \|\varrho u\|_{s_1}\le C(\|\varrho' E_ku\|_{s_2}+\|u\|_{-s_0}),
1063: \end{equation}
1064: for all $u\in \mathcal S$, then $s_2\ge s_1+k-1$. Here $\mathcal S$
1065: denotes the Schwartz space of rapidly decreasing functions.
1066: \end{prop}
1067:
1068: \Proof Let $\{\varrho_i\}$ and $\{\varrho'_i\}$ be sequences of
1069: cutoff functions in $C^\infty_0(U)$ and $C^\infty_0(U')$,
1070: respectively. We assume that $\varrho_(z,t)=\eta_i(|z|)\tau_i(t)$ and
1071: $\varrho'_i(z,t)=\eta'_i(|z|)\tau'_i(t)$ as in Definition
1072: 1. We further assume that $\varrho_0=\varrho$, $\varrho'_0=\varrho'$,
1073: $\varrho_{i+1}=1$ in a neighborhood of the support of
1074: $\varrho_i$, and $\varrho'_{i+1}=1$ in a neighborhood of the support of
1075: $\varrho'_i$ and that the $\eta_i(|z|)$ are monotone
1076: decreasing in $|z|$. We also choose
1077: $\{\gamma^+_i\}$ and $\{\gamma^0_i\}$ such that $\gamma_i^+\in\mathcal
1078: G^+$, $\gamma^+_{i+1}=1$, and $\gamma_i^0\in\mathcal
1079: G^0$ and $\gamma^0_{i+1}=1$ in neighborhoods of the supports of
1080: $\gamma^+_i$ and $\gamma^0_i$, respectively. Further we
1081: require that $\gamma^0_i=1$ in a neighborhood of the support of
1082: derivatives of $\gamma^+_i$. Substituting
1083: $\Psi^{-s_1}\Gamma^+_0u$ for $u$ in (\ref{hypest}), replacing
1084: $s_0+s_1$ by $s_0$, we have
1085: $$
1086: \|\varrho\Psi^{-s_1}\Gamma^+_0u\|_{s_1}\le C(\|\varrho'
1087: E_k\Psi^{-s_1}\Gamma^+_0u\|_{s_2}+\|u\|_{-s_0}).
1088: $$
1089: Since $\gamma^+_1\varrho\gamma^+_0=\varrho\gamma^+_0$,
1090: \begin{align*}
1091: \|\varrho\Psi^{-
1092: s_1}\Gamma^+_0u\|_{s_1}&=\|\Psi^{s_1}\Gamma^+_1\varrho\Psi^{-
1093: s_1}\Gamma^+_0u\|+
1094: O(\|u\|_{-1})\\
1095: &=\|\varrho\Gamma^+_0u\|+\|\Psi^{s_1}\Gamma^+_1[\varrho,\Psi^{s_1}]
1096: \Gamma^+_0u\|+O(\|u\|_{-s_0}).
1097: \end{align*}
1098: Furthermore, $\Psi^{s_1}\Gamma^+_1[\varrho,\Psi^{s_1}]\Gamma^+_0$ is an
1099: operator of order $-1$; hence we get
1100: $$
1101: \|\Psi^{s_1}\Gamma^+_1[\varrho,\Gamma^+_0]\Psi^{-s_1}u\|\le
1102: C(\|u\|_{-s_0})
1103: $$
1104: and
1105: $$
1106: \|\varrho\Gamma^+_0u\|\le
1107: C(\|\varrho'E_k\Psi^{-s_1}\Gamma^+_0u\|_{s_2}+\|u\|_{-s_0}).
1108: $$
1109: Next we have
1110: $$
1111: \|\varrho'
1112: E_k\Psi^{-s_1}\Gamma^+_0u\|_{s_2}\le\|\Psi^{s_2-
1113: s_1}\Gamma^+_0\varrho'E_ku\|+
1114: \|[\varrho'E_k,\Psi^{-s_1}\Gamma^+_0]u\|_{s_2}.
1115: $$
1116: Since the symbol of $\gamma^0_1\gamma^+_1\varrho'_1=1$ in a
1117: neighborhood of the symbol of\break
1118: $[\varrho'E_k,\Psi^{-s_1}\Gamma^+_0]$ and since the order of
1119: $[\varrho'E_k,\Psi^{-s_1}\Gamma^+_0]$ is $-s_1+1$, we have
1120: $$
1121: \|[\varrho'E_k,\Psi^{-s_1}\Gamma^+_0]u\|_{s_2}\le
1122: C(\|\varrho'_1\Gamma^0_1\Gamma^+_1u\|_{s_2-s_1+1}+\|u\|_{-s_0}).
1123: $$
1124: Applying Proposition 3, we have
1125: $$
1126: \|\varrho'_1\Gamma^0_1\Gamma^+_1u\|_{s_2-s_1+1}\le
1127: C(\|\varrho'_1E_k\Gamma^+_1u\|_{s_2-s_1-1}+
1128: \|\varrho'_2\Gamma^0_2\Gamma^+_1u\|_{s_2-s_1}+\|u\|_{-\infty})
1129: $$
1130: so that
1131: \begin{eqnarray*}
1132: &&\hskip-6pt\|\varrho' E_k\Psi^{-s_1}\Gamma^+_0u\|_{s_2}\\
1133: &&\quad\le
1134: C(\|\Psi^{s_2-s_1}\Gamma^+_0\varrho'E_ku\|+
1135: \|\varrho'_1\Gamma^0_2\Gamma^+_1u\|_{s_2-
1136: s_1}+\|\varrho'_1E_k\Gamma^+_1u\|_{s_2-s_1-1}+\|u\|_{-s_0}).
1137: \end{eqnarray*}
1138: Therefore
1139: \begin{eqnarray*}
1140: \|\varrho\Gamma^+_0u\| &\le& C(\|\Psi^{s_2-s_1}\Gamma^+_0\varrho'E_ku\|\\
1141: &&\phantom{ C(} +
1142: \|\varrho'_1E_k\Gamma^+_1u\|_{s_2-s_1-
1143: 1}+\|\varrho'_1E_k\Gamma^+_1u\|_{s_2-s_1-1}+\|u\|_{-s_0}).
1144: \end{eqnarray*}
1145: Now we have
1146: $$
1147: \|\varrho'_1E_k\Gamma^+_1u\|_{s_2-s_1-1}\le\|\Psi^{s_2-s_1-
1148: 1}\Gamma^+_1\varrho'_1E_ku\|+
1149: \|[\varrho'_1E_k,\Gamma^+_1]u\|_{s_2-s_1-1},
1150: $$
1151: again since $[\varrho'_1E_k,\Gamma^+_1]$ is an operator of order one
1152: and since $\varrho'_22\gamma^0_2\gamma^+_2=1$ in a
1153: neighborhood of its symbol, we get
1154: $$
1155: \|[\varrho'_2E_k,\Gamma^+_1]u\|_{s_2-s_1-1}\le
1156: C(\|\varrho'_2\Gamma^0_2\Gamma^+_2u\|_{s_2-s_1}+\|u\|_{-s_0}).
1157: $$
1158: Then, again applying Proposition 3, we have
1159: $$
1160: \|\varrho'_2\Gamma^0_2\Gamma^+_2u\|_{s_2-s_1}\le
1161: C(\|\varrho'_3\Gamma^0_3E_k\Gamma^+_2u\|_{s_2-s_1-2}+\|u\|_{-\infty})
1162: $$
1163: so that
1164: $$
1165: \|\varrho'_2E_k\Gamma^+_1u\|_{s_2-s_1-1}\le
1166: C(\|\Psi^{s_2-s_1-1}\Gamma^+_1\varrho'_2E_ku\|+
1167: \|\varrho'_3\Gamma^0_3E_k\Gamma^+_2u\|_{s_2-s_1-2}+\|u\|_{-s_0}).
1168: $$
1169: Hence
1170: \begin{eqnarray*}
1171: \|\varrho\Gamma^+u\|&\le& C(\|\Psi^{s_2-s_1}\Gamma^+_0\varrho'E_ku\|\\
1172: &&\phantom{C(} +
1173: \|\Psi^{s_2-s_1-
1174: 1}\Gamma^+_1\varrho'_2E_ku\|+\|\varrho'_4\Gamma^0_3E_k\Gamma^+_2u\|_{s_2
1175: -s_1-2}+\|u\|_{-s_0}).
1176: \end{eqnarray*}
1177: Proceeding inductively we obtain
1178: \begin{eqnarray*}
1179: \|\varrho\Gamma^+u\|&\le &
1180: C\Big(\sum_{i=0}^N\|\Psi^{s_2-s_1-i}\Gamma^+_i\varrho_i'E_ku\|\\
1181: &&\phantom{C\Big(}+
1182: \|\varrho'_{N+3}\Gamma^0_{N+2}E_k\Gamma^+_{N+1}u\|_{s_2-s_1-N-
1183: 1}+\|u\|_{-s_0}\Big).
1184: \end{eqnarray*}
1185: Since $\|[\Psi^{s_2-s_1-i}\Gamma^+_i,\eta_i']E_ku\|$ can be
1186: incorporated in the successive terms, we get, by choosing
1187: $N\ge s_2-s_1+1-s_0$
1188: $$
1189: \|\varrho\Gamma^+u\|\le
1190: C\Big(\sum_{i=0}^N\|\Psi^{s_2-s_1-i}\Gamma^+_i\tau'_iE_ku\|+\|u\|_{-s_0}\Big).
1191: $$
1192: Let $\tilde\tau\in C^\infty_0$ with $\tilde\tau=1$ on the support of
1193: $\tau'_N$; then $\tau'_iE_ku=\tau'_iE_k\tilde\tau u$
1194: when $i\le N$ so that replacing $u$ by $\tilde\tau u$ we obtain
1195: $$
1196: \|\varrho\Gamma^+\tilde\tau u\|\le
1197: C\Big(\sum_{i=0}^N\|\Psi^{s_2-s_1-i}\Gamma^+_i\tau'_iE_ku\|+
1198: \|\tilde\tau u\|_{-s_o}\Big).
1199: $$
1200: Hence, since $\gamma^0=1$ in a neighborhood of the support of
1201: the symbol of $[\Gamma^+,\tilde\tau]$ and thus can be incorporated in
1202: the estimate as above, we have
1203: \begin{align*}
1204: \|\varrho\Gamma^+u\|^2\le
1205: C\Big(\sum_{i=0}^N\|\Psi^{s_2-s_1-i}\Gamma^+_i\tau'_iE_ku\|^2+\|\tilde\tau
1206: u\|^2_{-s_o}\Big).
1207: \end{align*}
1208: Choosing $\tilde\gamma^+$ so that $\tilde\gamma^+=1$ in a neighborhood
1209: of the supports of the $\gamma_i^+$, we have
1210: $\tilde\tau\tilde\gamma^+=1$ in a neighborhood of the support of the
1211: symbol of $\Psi^{s_2-s_1-i}\Gamma^+_i\tau'_iE_k$.
1212: Then we obtain
1213: \begin{align*}
1214: \|\varrho\Gamma^+u\|^2&\le
1215: C\Big(\sum_{i=0}^N\|\Psi^{s_2-s_1-i}\tilde\Gamma^+\tilde\tau
1216: E_ku\|^2+\|\tilde\tau u\|^2_{-s_o}\Big)\\
1217: &\le C(\|\Psi^{s_2-s_1}\tilde\Gamma^+\tilde\tau E_ku\|^2+\|\tilde\tau
1218: u\|^2_{-s_o}).
1219: \end{align*}
1220: We define $h_{\lambda}$ by
1221: $$
1222: h_{\lambda}(z,t)={\rm exp}\left(-\lambda^2(|z|^2-it)\right),
1223: $$
1224: since $\tilde\tau h_{\lambda}\in\mathcal S$ and obtain
1225: $$
1226: \|\varrho\Gamma^+h_\lambda\|\le
1227: C\left(\|\Psi^{s_2-s_1}\tilde\Gamma^+_i\tilde\tau E_kh_{\lambda}\|+
1228: \|\tilde\tau h_{\lambda}\|_{-s_o})\right).
1229: $$
1230: Assuming that $\eta(|z|)$ is monotone decreasing we have
1231: $\eta(|z|)\ge\eta(\lambda|z|)$; hence, setting
1232: $\eta_\lambda(z)=\eta(\lambda|z|)$, we obtain
1233:
1234: $$
1235: \|\varrho\Gamma^+h_{\lambda}\|\ge\|\eta_\lambda\tau\Gamma^+h_{\lambda}\|.
1236: $$
1237: Then, setting $x'=(x_1,x_2)$, $y'=(y_1,y_2)$, and $\xi'=(\xi_1,\xi_2)$
1238: and changing variables $\lambda y'\to y'$,
1239: $\xi'\to\lambda\xi'$, and $\xi'_3\to\xi'_3+\lambda^2$, we get
1240: \begin{align*}
1241: &\eta_\lambda\tau\Gamma^+h_{\lambda}(x)=
1242: \int\exp(i(x-y)\cdot\xi)\tau(y_3)\gamma^+(\xi)\exp(-\lambda^2(|y'|^2-
1243: iy_3))dyd\xi\\
1244: &=\int\exp(i(x'-y')\cdot\xi'+x_3\xi_3-y_3(\xi_3-
1245: \lambda^2))\tau(y_3)\gamma^+(\xi)\exp(-\lambda^2|y'|^2)dyd\xi\\
1246: &=\lambda^{-2}\int\exp(i(\lambda
1247: x'-y')\cdot\xi')\\
1248: &\quad+(x_3-
1249: y_3)\xi_3)\tau(y_3)\gamma^+(\lambda\xi',\lambda^2+\xi_3)
1250: \exp(-|y'|^2)dyd\xi.
1251: \end{align*}
1252: Making the change of variables $\lambda x'\to x'$ we have
1253: \begin{align*}
1254: &\|\eta_\lambda\tau\Gamma^+h_{\lambda,\delta}\|^2\\
1255: &\quad=\frac{1}{\lambda^6}\int
1256: |
1257: \int\exp(i(x-
1258: y)\cdot\xi)\tau(y_3)\gamma^+(\lambda\xi',\lambda^2+\xi_3)\exp(-
1259: |y'|^2)dyd\xi|^2dx.
1260: \end{align*}
1261: Given $(\xi',\xi_3)\in {\rm supp}(\gamma^+)$ we have
1262: $$
1263: \lim_{\lambda\to\infty}|\frac{\lambda\xi'}{\xi_3+\lambda^2}|=0,
1264: $$
1265: and there exists $\tilde\lambda$ such that
1266: $\gamma^+(\lambda\xi',\lambda^2+\xi_3)=1$ when
1267: $\lambda\ge\tilde\lambda$.
1268: Hence we have
1269: $\lim_{\lambda\to\infty}\gamma^+(\lambda\xi',\lambda^2+\xi_3)=1$; thus
1270: there exist $\lambda_0$ such that
1271: $$
1272: \|\eta_\lambda\tau\Gamma^+h_{\lambda}\|^2\ge\frac{1}{2\lambda^6}
1273: \int|\int\exp(i(x-y)\cdot\xi)\tau(y_3)\exp(-|y'|^2)dyd\xi|^2dx,
1274: $$
1275: when $\lambda\ge\lambda_0$, therefore there exists $C$ independent of
1276: $\lambda$ such that
1277: $$
1278: \|\varrho\Gamma^+h_{\lambda}\|\ge\frac{C}{\lambda^3},
1279: $$
1280: when $\lambda\ge\lambda_0$.
1281:
1282: Next, we will estimate the term $\|\tilde\tau
1283: h_{\lambda}\|_{-s_o}$. We will use the facts that
1284: $\frac{1}{m!}\bar L^m(\bar z^m)\break=1$ and that $\bar L(h_\lambda)=0$.
1285: Taking $m\le s_o$, we
1286: have
1287: \begin{align*}
1288: &\mathcal F(\Lambda^{-s_o}\tilde\tau h_{\lambda})(\xi)=
1289: \int(1+|\xi|^2)^{\frac{-s_o}{2}}\tilde\tau(x_3)\exp(-i(x\cdot\xi-
1290: \lambda^2x_3)-\lambda^2|z|^2)dx\\
1291: &\quad=\frac{1}{m!}\int\bar L^m(\bar z^m)(1+|\xi|^2)^{\frac{-s_o}{2}}
1292: \tilde\tau)\exp(-i(x\cdot\xi-\lambda^2x_3)-\lambda^2|z|^2)dx\\
1293: &\quad=-\frac{1}{m!}\int\bar z^m(1+|\xi|^2)^{\frac{-s_o}{2}}
1294: \bar
1295: L^m(\tilde\tau(x_3)\exp(-i(x\cdot\xi-\lambda^2x_3)-\lambda^2|z|^2)dx\\
1296: &\quad=-\frac{1}{m!}\int\bar z^m(1+|\xi|^2)^{\frac{-s_o}{2}}
1297: \exp(-\lambda^2(|z|^2-ix_3))\bar
1298: L^m(\tilde\tau(x_3)\exp(-ix\cdot\xi))dx\\
1299: &\quad=-\frac{1}{m!}\int\bar z^m(1+|\xi|^2)^{\frac{-s_o}{2}}
1300: \exp(-\lambda^2(|z|^2))\bar
1301: L^m(\tilde\tau(x_3)\\
1302: &\hskip.4in\cdot\exp(-ix'\cdot\xi'-ix_3(\xi_3-\lambda^2))dx
1303: \end{align*}
1304: and
1305: \begin{align*}
1306: &\bar
1307: L^m\left(\tilde\tau(x_3)\exp(-ix'\cdot\xi'+i-ix_3(\xi_3-
1308: \lambda^2))\right)\\
1309: &=\sum_{j=0}^ma_j(x_3)z^j(i\xi_1+\xi_2-2z\xi_3)^{m-j}\exp(-
1310: ix'\cdot\xi'-ix_3(\xi_3-\lambda^2)).
1311: \end{align*}
1312: Thus, setting
1313: $w^{(m)}(x,\xi)=\sum_{j=0}^ma_j(x_3)z^j(i\xi_1+\xi_2-2z\xi_3)^{m-j}$
1314: and denoting the corresponding
1315: pseudodifferential operator by $W^{(m)}$, we have
1316: $$
1317: \|\tilde\tau h_{\lambda}\|_{-s_o}=C\|W^{(m)}\bar
1318: z^mh_{\lambda}\|_{-s_o}\le C\|z^m\tilde\tau'h_{\lambda}\|_{m-s_o}
1319: \le C\|z^m\tilde\tau'h_{\lambda}\|,
1320: $$
1321: where $\tilde\tau'\in C^\infty_0(\mathbb R)$ and $\tilde\tau'=1$ in a
1322: neighborhood of the support of $\tilde\tau$. Now, changing
1323: coordinates $\lambda z\to z$, we get
1324: $$
1325: \|z^m\tilde\tau'h_{\lambda}\|^2=\int|z|^{2m}\tilde\tau'(x_3)^2\exp(-
1326: 2\lambda^2|z|^2)dx\le\frac{C}{\lambda^{2m+2}}.
1327: $$
1328: To estimate the remaining terms we have
1329: $$
1330: E_kh_\lambda(z,t)=-2(k+1)\lambda^2|z|^{2k}h_\lambda(z,t).
1331: $$
1332: Therefore, with the coordinate change $\lambda x'\to x'$, we get
1333: \begin{align*}
1334: &
1335: \mathcal F\left(\Psi^{s}\Gamma^+\tau E_kh_\lambda\right)(\xi)\\
1336: &\qquad=
1337: C\mathcal
1338: F\left(\Psi^{s}\Gamma^+\tau\lambda^2|z|^{2k}h_\lambda\right)(\xi)\\
1339: &\qquad=C\lambda^{-2}(1+\xi_3^2)^{\frac{s}{2}}\gamma^+(\xi)\mathcal
1340: F\left(\tau(x_3)|z|^{2k}\exp(-\lambda^2|z|^2)\right)(\xi',\xi_3-
1341: \lambda^2)\\
1342: &\qquad=C\lambda^{-2k-2}(1+\xi_3^2)^{\frac{s}{2}}\gamma^+(\xi)\hat\tau(\xi_3-
1343: \lambda^2)\mathcal
1344: F\left(|z|^{2k}\exp(-2|z|^2)\right)(\lambda^{-1}\xi').
1345: \end{align*}
1346: Then, integrating and making the changes of coordinates
1347: $\xi'\to\lambda\xi'$, $\xi_3\to\xi_3+\lambda^2$, we get
1348: \begin{align*}
1349: &\hskip-9pt\|\Psi^{s}\Gamma^+\tau E_kh_\lambda\|^2\\
1350: &\le
1351: C\lambda^{-4k-4}\int(1+\xi_3^2)^{s}
1352: \gamma^+(\xi)\hat\tau(\xi_3-\lambda^2)|^2|\mathcal
1353: F\left(|z|^{2k}\exp(-|z|^2)\right)(\lambda^{-1}\xi')|^2d\xi\\
1354: &\le
1355: C\lambda^{-4k-
1356: 2}\\&\enspace
1357: \cdot\int(1+(\xi_3+\lambda^2)^2)^{s}\gamma^+(\lambda\xi',\xi_3+\lambda^2)
1358: \hat\tau(\xi_3)|^2|\mathcal
1359: F\left(|z|^{2k}\exp(-|z|^2)\right)(\xi')|^2d\xi.
1360: \end{align*}
1361: Then if $s\ge0$ and if $\lambda$ is sufficiently large we have
1362: $$
1363: \|\Psi^{s}\Gamma^+\tau E_kh_\lambda\|^2\le C\lambda^{4s-4k-2}.
1364: $$
1365: We assume $k\ge 1$; if $s_2-s_1<0$ then
1366: $$
1367: \|\Psi^{s_1-s_2}\varrho\Gamma^+h_\lambda\|^2\le
1368: C\left(\|\tilde\Gamma^+_i\tilde\tau E_kh_{\lambda}\|^2+
1369: \|\tilde\tau h_{\lambda}\|^2_{-s_o})\right)\le C\lambda^{-4k-2}
1370: $$
1371: and, by Lemma 1,
1372: \begin{align*}
1373: \|\Psi^{s_1-s_2}\varrho\Gamma^+h_\lambda\|^2&=
1374: \|\Psi^{s_1-
1375: s_2}\eta_\lambda\tau\Gamma^+h_{\lambda}\|^2+O(\|\tau\Gamma^+h_{\lambda}\
1376: |^2_{-s_o})\\
1377: &\ge
1378: C\lambda^{2s_2-2s_1}\|\eta_\lambda\tau\Gamma^+h_{\lambda}\|^2-
1379: C'(\|\tau\Gamma^+h_{\lambda}\|^2_{-s_o})\\
1380: &\ge C(\lambda^{2s_2-2s_1-2}+\lambda^{-2m-2}).
1381: \end{align*}
1382: This implies that for large $\lambda$ we have $\lambda^{2s_2-2s_1-2}\le
1383: C(\lambda^{-4k-2}+\lambda^{-2m-2})$,
1384: which is a contradiction, so that $s_2-s_1\ge0$ and
1385: \begin{align*}
1386: C_1\lambda^{-6}&\le C_2\|\varrho\Gamma^+h_\lambda\|^2\le
1387: C\left(\|\Psi^{s_2-s_1}\tilde\Gamma^+_i\tilde\tau E_kh_{\lambda}\|^2+
1388: \|\tilde\tau h_{\lambda}\|^2_{-s_o}\right)\\*
1389: &\le C_3(\lambda^{4s_2-4s_1-4k-2}+\lambda^{-2m-2}).
1390: \end{align*}
1391: Therefore, if $m$ large we get $C_1\le
1392: 2C_3\lambda^{4(s_2-s_1-k+1)}$ for large $\lambda$. Hence
1393: $s_2-s_1-k+1\ge0$, which concludes the proof of the proposition and
1394: also of Theorem~B.
1395:
1396:
1397:
1398: \section{Elliptic and subelliptic microlocalizations}
1399:
1400:
1401: In this section we will show that the {\it a priori\/} estimates for the
1402: operator $E_k$ gain two derivatives in the $0$ microlocalization
1403: and gains one derivative in the $-$ microlocalization, these gains are
1404: in the Sobolev norms. Without loss of generality we will
1405: deal only with microlocalizations near the origin, taking
1406: $\alpha=0$ and setting $\mathfrak G^0=\mathfrak G^0_0$ and
1407: $\mathfrak G^-=\mathfrak G^-_0$. The subscript $\alpha$ will be
1408: dropped from the corresponding operators.
1409:
1410: \begin{prop} Let $U$ and $U'$ be neighborhoods of the origin with $\bar
1411: U\subset U'$ and $|z|\le a$ on $U'${\rm ,} where $a$ is
1412: sufficiently small as in Lemma {\rm 3}. Suppose that $\varrho\in
1413: C_0^\infty(U)$ and $\varrho'\in C_0^\infty(U')$ with
1414: $\varrho'=1$ on a neighborhood of $\bar U$. Further suppose that
1415: $\gamma^0,\tilde\gamma^0\in\mathcal G^0$ with $\tilde\gamma^0=1$
1416: on a neighborhood of the support of $\gamma^0$. Then{\rm ,} given
1417: $s,s_0\in\mathbb R${\rm ,} there exists
1418: $C=C(\varrho,\varrho',\gamma^0,\tilde\gamma^0,s,s_0)$ such that
1419: $$
1420: \|\varrho\Gamma^0u\|^2_{s+2}+\|\varrho\Gamma^0\bar
1421: z^kLu\|^2_{s+1}+\|\varrho\Gamma^0\bar Lu\|^2_{s+1}\le
1422: C(\|\varrho'\tilde\Gamma^0E_ku\|^2_s+\|u\|^2_{-s_0}),
1423: $$
1424: for all $u\in\mathcal S${\rm ,} where $\mathcal S$ denotes the Schwartz class
1425: of rapidly decreasing functions.
1426: \end{prop}
1427:
1428: \Proof Let $\{\varrho_i\}$ be a sequence of functions
1429: such that
1430: $\varrho_i\in C_0^\infty(U)$, $\varrho_0=\varrho$, $\varrho_{i+1}=1$ in
1431: a neighborhood of
1432: the support of $\varrho_i$, and such that $\varrho'=1$ in a
1433: neighborhood of the supports of all the $\varrho_i$. Let
1434: $\{\gamma^0_i\}$ be a sequence in $\mathcal G^0$ such that
1435: $\gamma^0_0=\gamma^0$, $\gamma^0_{i+1}=1$ in a neighborhood of
1436: the support of $\gamma_i$, and $\tilde\gamma^0=1$ in a neighborhood of
1437: the supports of all the $\gamma^0_i$. Then
1438: substituting $\varrho\Lambda^{s+1}\Gamma^0_1u$ for $u$ in Lemma 3 we
1439: have
1440: $$
1441: \|\Gamma^0\varrho\Lambda^{s+1}\Gamma^0_1u\|^2_1\le C(\|\Gamma^0\bar
1442: L\varrho\Lambda^{s+1}\Gamma^0_1u\|^2+
1443: \|\varrho\Lambda^{s+1}\Gamma^0_1u\|^2).
1444: $$
1445: Hence
1446: $$
1447: \|\Gamma^0\varrho\Lambda^{s+1}\Gamma^0_1u\|^2_1\le C(\|\Gamma^0\bar
1448: z^kL\varrho\Lambda^{s+1}\Gamma^0_1u\|^2+
1449: \|\Gamma^0\bar
1450: L\varrho\Lambda^{s+1}\Gamma^0_1u\|^2+\|\varrho\Lambda^{s+1}\Gamma^0_1u\|
1451: ^2).
1452: $$
1453: Then
1454: \begin{eqnarray*}
1455: \|\Gamma^0\varrho\Lambda^{s+1}\Gamma^0_1u\|^2_1&=&\|\varrho\Gamma^0u\|_{s
1456: +2}^2+
1457: O(\|\Lambda^1[\Gamma^0\varrho,\Lambda^{s+1}]\Gamma^0_1u\|^2+\|\Lambda^{s
1458: +2}[\Gamma^0,\varrho]\Gamma^0_1u\|^2
1459: \\
1460: &&\phantom{\|\varrho\Gamma^0u\|_{s
1461: +2}^2+
1462: O(}+\|\Lambda^{s+2}\varrho(\Gamma^0\Gamma^0_1-\Gamma^0)u\|^2).
1463: \end{eqnarray*}
1464: Since $[\Gamma^0\varrho,\Lambda^{s+1}]$ is a pseudodifferential
1465: operator of order $s+1$ and since $\varrho_1=1$ on the
1466: support of its symbol, we have
1467: $$
1468: \|\Lambda^1[\Gamma^0\varrho,\Lambda^{s+1}]\Gamma^0_1u\|^2\le
1469: C(\|\varrho_1\Gamma^0_1u\|^2_{s+1}+\|\Gamma^0_1u\|^2_{-\infty}).
1470: $$
1471: The operator $[\Gamma^0,\varrho]$ is of order $-1$ and $\varrho_1=1$ on
1472: the support of its symbol, so that
1473: $$
1474: \|\Lambda^{s+2}[\Gamma^0,\varrho]\Gamma^0_1u\|^2\le
1475: C(\|\varrho_1\Gamma^0_1u\|^2_{s+1}+\|\Gamma^0_1u\|^2_{-\infty}).
1476: $$
1477: The symbol of the operator
1478: $\Lambda^{s+2}\varrho(\Gamma^0\Gamma^0_1-\Gamma^0)$ is zero so that
1479: $$
1480: \|\Lambda^{s+2}\varrho(\Gamma^0\Gamma^0_1-\Gamma^0)u\|^2\le
1481: C\|u\|^2_{-\infty}.
1482: $$
1483: Then we obtain
1484: $$
1485: \|\varrho\Gamma^0u\|_{s+2}^2\le
1486: C(\|\Gamma^0\varrho\Lambda^{s+1}\Gamma^0_1u\|^2_1+\|\varrho_1\Gamma^0_1u
1487: \|^2_{s+1}+\|u\|^2_{-\infty}),
1488: $$
1489: so that
1490: $$
1491: \|\varrho\Gamma^0u\|_{s+2}^2\le C(\|\Gamma^0\bar
1492: z^kL\varrho\Lambda^{s+1}\Gamma^0_1u\|^2+
1493: \|\Gamma^0\bar
1494: L\varrho\Lambda^{s+1}\Gamma^0_1u\|^2+\|\varrho_1\Gamma^0_1u\|^2_{s+1}+\|
1495: u\|^2_{-\infty}).
1496: $$
1497: The following lemma which involves a vector field $X$ will be applied
1498: with $X=\bar z^kL$ and $X=\bar L$.
1499:
1500:
1501: \begin{lemma}If $X$ is a complex vector field on $\mathbb R^3$ then
1502: $$
1503: \|\Gamma^0X\varrho\Lambda^{s+1}\Gamma^0_1u\|^2\le
1504: C(\|\varrho\Gamma^0Xu\|^2_{s+1}
1505: +\|\varrho_1\Gamma^0_1u\|^2_{s+1}+\|u\|^2_{-\infty})
1506: $$
1507: and
1508: \begin{eqnarray*}
1509: \|\varrho\Gamma^0Xu\|^2_{s+1}&=&(\Lambda^s\varrho\Gamma^0X^*Xu,\Lambda^{s
1510: +2}\varrho\Gamma^0u)\\
1511: &&+
1512: O(\|\varrho_1\Gamma^0_1u\|^2_{s+1}+\|\varrho\Gamma^0u\|^2_{s+2}\|\varrho
1513: _1\Gamma^0_1u\|^2_{s+1}\\
1514: &&\phantom{+O(}+\|\varrho_1\Gamma^0_1Xu\|^2_s+\|\varrho_1\Gamma^0_1u\|_{s+1}\|
1515: \varrho\Gamma^0Xu\|_{s+1}+\|u\|^2_{-\infty}),
1516: \end{eqnarray*}
1517: for all $u\in\mathcal S$.
1518: \end{lemma}
1519:
1520: \Proof We have
1521: $$
1522: \|\Gamma^0X\varrho\Lambda^{s+1}\Gamma^0_1u\|\le\|\Gamma^0\varrho\Lambda^
1523: {s+1}\Gamma^0_1Xu\|+
1524: \|\Gamma^0[X,\varrho\Lambda^{s+1}\Gamma^0_1]u\|.
1525: $$
1526: The operator
1527: $P=\Gamma^0\varrho\Lambda^{s+1}\Gamma^0_1X-
1528: \Lambda^{s+1}\Gamma^0_1\varrho\Gamma^0X$ is of order
1529: $s+1$ and $\varrho_1\gamma^0_1=1$ in a neighborhood of the symbol of
1530: $P$; hence
1531: $$
1532: \|Pu\|\le C(\|P\varrho_1\Gamma^0_1u\|+\|u\|_{-\infty})\le
1533: C(\|\varrho_1\Gamma^0_1u\|_{s+1}+\|u\|_{-\infty}).
1534: $$
1535: Since $\gamma^0_1=1$ in a neighborhood of the support of the symbol of
1536: $\varrho\Gamma^0$, we get
1537: $$
1538: \|\Gamma^0\varrho\Lambda^{s+1}\Gamma^0_1Xu\|\le
1539: C(\|\varrho\Gamma^0Xu\|_{s+1}+\|u\|_{-\infty}).
1540: $$
1541: Furthermore, $\Gamma^0[X,\varrho\Lambda^{s+1}\Gamma^0_1]$ is of order
1542: $s+1$ and $\varrho_1\gamma^0_1=1$ in a neighborhood of
1543: the support of its symbol so that
1544: $$
1545: \|\Gamma^0[X,\varrho\Lambda^{s+1}\Gamma^0_1]u\|\le
1546: C(\|\varrho_1\Gamma^0_1u\|_{s+1}+\|u\|_{-\infty}),
1547: $$
1548: which proves the first part of the lemma.
1549:
1550: For the second part of the lemma we write
1551: \begin{align*}
1552: \|\varrho\Gamma^0Xu\|^2_{s+1}&=(\Lambda^{s+1}\varrho\Gamma^0Xu,\Lambda^{
1553: s+1}\varrho\Gamma^0Xu)\\
1554: &=(\Lambda^{s+1}\varrho\Gamma^0Xu,[\Lambda^{s+1}\varrho\Gamma^0,X]u)
1555: +([X^*,\Lambda^{s+1}\varrho\Gamma^0]Xu,\Lambda^{s+1}\varrho\Gamma^0u)\\
1556: &\quad+(\Lambda^s\varrho\Gamma^0X^*Xu,\Lambda^{s+2}\varrho\Gamma^0u).
1557: \end{align*}
1558: Then, since $[\Lambda^{s+1}\varrho\Gamma^0,X]$ is of order $s+1$ and
1559: $\varrho_1\gamma^0_1=1$ in a neighborhood of its symbol,
1560: $$
1561: \|[\Lambda^{s+1}\varrho\Gamma^0,X]u\|^2\le
1562: C(\|\varrho_1\Gamma^0_1u\|_{s+1}^2+\|u\|_{-\infty}).
1563: $$
1564: Then
1565: \begin{eqnarray*}
1566: &&\hskip-7pt
1567: ([X^*,\Lambda^{s+1}\varrho\Gamma^0]Xu,\Lambda^{s+1}\varrho\Gamma^0u)\\
1568: &&=
1569: ([(\Lambda^{s+1}\varrho\Gamma^0)^*,[X^*,\Lambda^{s+1}\varrho\Gamma^0]]Xu
1570: ,u) +((\Lambda^{s+1}\varrho\Gamma^0)^*Xu,[X^*,\Lambda^{s+1}\varrho
1571: \Gamma^0]^*u).
1572: \end{eqnarray*}
1573: Let
1574: $Q=[(\Lambda^{s+1}\varrho\Gamma^0)^*,[X^*,\Lambda^{s+1}\varrho\Gamma^0]]
1575: $; then $Q$ has order $2s+1$ and
1576: $\varrho_1\gamma^0_1=1$ in a neighborhood of its symbol. Thus
1577: \begin{align*}
1578: |(QXu,u)|&\le
1579: C(|(Q\varrho_1\Gamma^0_1Xu,\varrho_1\Gamma^0_1u)|+\|u\|_{-\infty}^2)\\
1580: &\le
1581: C(\|\varrho_1\Gamma^0_1Xu\|^2_{s}+\|\varrho_1\Gamma^0_1u\|^2_{s+1}+\|u\|
1582: _{-\infty}^2).
1583: \end{align*}
1584: The symbol of the operator
1585: $(\Lambda^{s+1}\varrho\Gamma^0)^*-\Lambda^{s+1}\varrho\Gamma^0$ is
1586: zero, the order of\break
1587: $[X^*,\Lambda^{s+1}\varrho\Gamma^0]^*$ is $s+1$ and
1588: $\varrho_1\gamma^0_1=1$ on a neighborhood of its support. Hence
1589: $$
1590: |((\Lambda^{s+1}\varrho\Gamma^0)^*Xu,[X^*,\Lambda^{s+1}\varrho\Gamma^0]^
1591: *u)|\le
1592: C(\|\varrho\Gamma^0Xu\|_{s+2}\|\varrho_1\Gamma_1^0u\|_{s+1}+\|u\|_{-
1593: \infty}^2).
1594: $$
1595: Combining these we conclude the proof of the lemma.
1596:
1597: Returning to the proof of the proposition, by using the above
1598: lemma, when $X=\bar z^kL$ and when
1599: $X=\bar L$, we obtain
1600: \begin{multline*}
1601: \|\varrho\Gamma^0u\|^2_{s+2}+\|\varrho\Gamma^0\bar
1602: z^kLu\|^2_{s+1}+\|\varrho\Gamma^0\bar Lu\|^2_{s+1}\\
1603: \le
1604: C(\|\varrho\Gamma^0 E_ku\|^2_s+\|\varrho_1\Gamma_1^0\bar z^kLu\|^2_{s}
1605: +\|\varrho_1\Gamma_1^0\bar Lu\|^2_{s}+\|u\|_{-\infty}^2).
1606: \end{multline*}
1607: Replacing $\varrho$ by $\varrho_i$, $\varrho_1$ by $\varrho_{i+1}$,
1608: $\Gamma^0$ by $\Gamma^0_i$,
1609: $\Gamma_1^0$ by $\Gamma_{i+1}^0$, and $s$ by $s-i$ we obtain
1610: \begin{multline*}
1611: \|\varrho_i\Gamma_i^0u\|^2_{s+2-i}+\|\varrho_i\Gamma_i^0\bar
1612: z^kLu\|^2_{s+1-i}+\|\varrho_i\Gamma_i^0\bar Lu\|^2_{s+1-i}\\
1613: \le
1614: C(\|\varrho_i\Gamma_i E_ku\|^2_{s-i}+\|\varrho_{i+1}\Gamma_{i+1}^0\bar
1615: z^kLu\|^2_{s-i} +\|\varrho^0_{i+1}\Gamma_{i+1}^0\bar
1616: Lu\|^2_{s-i}+\|u\|_{-\infty}^2).
1617: \end{multline*}
1618: Proceeding inductively, we obtain
1619: \begin{multline*}
1620: \|\varrho\Gamma^0u\|^2_{s+2}+\|\varrho\Gamma^0\bar
1621: z^kLu\|^2_{s+1}+\|\varrho\Gamma^0\bar Lu\|^2_{s+1}\\
1622: \le
1623: C\Big(\sum_{i=0}^N\|\varrho_i\Gamma^0_i E_ku\|^2_{s-i}
1624: +\|\varrho_{N+1}\Gamma_{N+1}^0\bar
1625: z^kLu\|^2_{s-N}+\|\varrho^0_{i+N}\Gamma_{i+N}^0\bar Lu\|^2_{s-N}+
1626: \|u\|_{-\infty}^2\Big).
1627: \end{multline*}
1628: Setting $N\ge s_0+s+1$ we conclude the proof of the proposition since
1629: $$\|\varrho_i\Gamma^0_i E_ku\|^2_{s-i}\le
1630: C(\|\varrho'\tilde\Gamma^0 E_ku\|^2_{s}+\|u\|^2_{-\infty}).
1631: $$
1632:
1633:
1634: \begin{prop}\hskip-2pt Given neighborhoods of the origin $U$ and $U'$ with $\bar
1635: U\!\subset\! U'${\rm ;} suppose that $\varrho\in C_0^\infty(U)$
1636: and $\varrho'\in C_0^\infty(U')$ with $\varrho'=1$ on a neighborhood of~$\bar U$. Further
1637: suppose that
1638: $\gamma^-,\tilde\gamma^-\in\mathcal G^-$ with $\tilde\gamma^-=1$ on a
1639: neighborhood of the support of $\gamma^-$. Then{\rm ,} given
1640: $s,s_0\in\mathbb R${\rm ,} there exists
1641: $C=C(\varrho,\varrho',\gamma^-,\tilde\gamma^-,s,s_0)$ such that
1642: $$
1643: \|\varrho\Gamma^-u\|^2_{s+1}+\|\varrho\Gamma^-\bar
1644: z^kLu\|^2_{s+\frac{1}{2}}+
1645: \|\varrho\Gamma^-\bar Lu\|^2_{s+\frac{1}{2}}\le
1646: C(\|\varrho'\tilde\Gamma^-E_ku\|^2_s+\|u\|^2_{-s_0}),
1647: $$
1648: for all $u\in\mathcal S$.
1649: \end{prop}
1650:
1651: \Proof The proof is entirely analogous to that of the
1652: above proposition. We use Lemma 5 in place of Lemma 3
1653: and substitute $\varrho\Lambda^{s+\frac{1}{2}}\Gamma^-_1u$ for $u$ we
1654: obtain
1655: $$
1656: \|\Gamma^-\varrho\Lambda^{s+\frac{1}{2}}\Gamma^-
1657: _1u\|^2_{\frac{1}{2}}\le C(\|\Gamma^0\bar
1658: L\varrho\Lambda^{s+\frac{1}{2}}\Gamma^-_1u\|^2+
1659: \|\varrho\Lambda^{s+\frac{1}{2}}\Gamma^0_1u\|^2).
1660: $$
1661: Then one proceeds exactly as above to obtain the proof.
1662:
1663: In the case $k=0$ the vectorfields $L$ and $\bar L$ play
1664: exactly the same role and so we obtain the following.
1665: \begin{prop} Given neighborhoods of the origin $U$ and $U'$ with $\bar
1666: U\!\subset\! U'$. Suppose that $\varrho\in C_0^\infty(U)$
1667: and $\varrho'\in C_0^\infty(U')$ with $\varrho'=1$ on a neighborhood of~$\bar U$. Further
1668: suppose that
1669: $\gamma^+,\tilde\gamma^+\in\mathcal G^+$ with $\tilde\gamma^+=1$ on a
1670: neighborhood of the support of $\gamma^+$. Then{\rm ,} given
1671: $s,s_0\in\mathbb R${\rm ,} there exists
1672: $C=C(\varrho,\varrho',\gamma^+,\tilde\gamma^+,s,s_0)$ such that
1673: $$
1674: \|\varrho\Gamma^+u\|^2_{s+1}+\|\varrho\Gamma^+Lu\|^2_{s+\frac{1}{2}}+
1675: \|\varrho\Gamma^+\bar Lu\|^2_{s+\frac{1}{2}}\le
1676: C(\|\varrho'\tilde\Gamma^+E_0u\|^2_s+\|u\|^2_{-s_0}),
1677: $$
1678: for all $u\in\mathcal S$.
1679: \end{prop}
1680:
1681:
1682:
1683:
1684: \section{The operator $E_0$ and gain of derivatives}
1685:
1686: Since $E_0$ is a real operator, it can be written as $E_0=-X^2-Y^2$, where
1687: $X=\frac{1}{\sqrt2}\Re L$ and $Y=\frac{1}{\sqrt2}\Im L$.
1688: Thus it is one of the simplest operators that\break\vskip-12pt\noindent satisfy H\"ormander's
1689: condition and it is well understood. Nevertheless, it is
1690: instructive to write it in terms of $L$ and $\bar L$ and analyze it
1691: microlocally in the framework of the previous section. The
1692: operator $E_0$ gains one derivative. As we have seen the operators
1693: $E_k$ do not gain derivatives when $k>0$ and $z=0$; in a
1694: neighborhood on which $z\neq0$ they do gain derivatives and they also
1695: gain in the $0$ and $-$ microlocalizations.
1696:
1697: In the analysis of $E_0$ we can assume, without loss of
1698: generality, that $\alpha=0$ and we set
1699: $\gamma=\gamma_0$, and $\Gamma=\Gamma_0$. The basic observation is
1700: that the gain of derivatives in the $+$ and $-$
1701: microlocalizations is controlled by the operators $\bar LL$ and $L\bar
1702: L$, respectively. In the $0$ microlocalization the
1703: gain of derivatives is controlled by both $\bar LL$ and $L\bar L$
1704: independently. Propositions 4 and 5 give {\it a priori\/} estimates
1705: for $E_k$ in the $0$ and $-$ microlocalizations, respectively.
1706: Proposition 6 gives these estimates for the $+$
1707: microlocalization. Here we show how to go from the {\it a priori\/}
1708: estimates to
1709: hypoellipticity. In particular we prove that $E_0$ is
1710: hypoelliptic and that $E_k$ is hypoelliptic on open sets on which
1711: $z\neq0$ and that the $0$ and $-$ microlocalizations of the
1712: operators $E_k$ are hypoelliptic.
1713:
1714: \begin{prop} If $u$ is a distribution such that for some open set
1715: $V\subset\mathbb R^3$ the restriction of $E_0u$ to $V$ is
1716: in $C^\infty(V)$ then the restriction of $u$ to $V$ is also in
1717: $C^\infty(U)$. More precisely{\rm ,} if $E_0u\in H_{\rm loc}^s(V)$ then
1718: $u\in H_{\rm loc}^{s+1}(V)$.
1719: \end{prop}
1720:
1721: \Proof Assuming that $E_0u\in H_{\rm loc}^s(V)$, it suffices to show
1722: that any $P\in V$ has a neighborhood $U\subset V$ such
1723: that for any $\varrho\in C_0^\infty(U)$ we have $\varrho u\in
1724: H^{s+1}(\mathbb R^3)$. Without loss of generality we may
1725: assume that $P=0$. Now choose neighborhoods $U$ and $U'$ of $P$ such
1726: that $\bar U\subset U'$ and $|z|\le a$ on $U'$, as in
1727: Proposition 4. Let $\varrho\in C_0^\infty(U)$, let $\varrho'\in
1728: C_0^\infty(U')$ with $\varrho'=1$ in a neighborhood of the
1729: support of $\varrho$, and let $\theta\in C_0^\infty(\mathbb R^3)$ such
1730: that $\theta=1$ on a neighborhood of~$\bar U'$. Since
1731: $u$ is a distribution there exists an $s_0\in\mathbb R$ such that
1732: $\theta u\in H^{-s_0}(\mathbb R^3)$. Then, choosing $\gamma^+$,
1733: $\gamma^0$, and $\gamma^-$ such that $\gamma^++\gamma^0+\gamma^-\ge
1734: {\rm const.}>0$ and combining Propositions 4, 5, and 6 we obtain
1735: the {\it a priori\/} estimate
1736: $$
1737: \|\varrho u\|^2_{s+1}+\|\varrho Lu\|^2_{s+\frac{1}{2}}+\|\varrho \bar
1738: Lu\|^2_{s+\frac{1}{2}}\le
1739: C(\|\varrho'E_0u\|^2_s+\|u\|^2_{-s_0}),
1740: $$
1741: for all $u\in C^\infty(\mathbb R^3)$. Let $\chi\in C_0^\infty(\mathbb
1742: R^3)$ with $\chi(0)=1$. For $\delta>0$ we define the
1743: smoothing operator $S_\delta$ by $\mathcal F(S_\delta
1744: u)(\xi)=\chi(\delta\xi)\hat u(\xi)$. The important facts are that:
1745: \begin{enumerate}
1746: \item[1.] If $\delta>0$ then for any distribution $v$ the function
1747: $S_\delta v\in C^\infty(\mathbb R)$.
1748: \item[2.] If $v$ is a distribution and if $\|S_\delta v\|_s$ is bounded
1749: independently of $\delta$ then $v\in H^s(\mathbb R^3)$.
1750: \item[3.] If $v\in H^s(\mathbb R^3)$ then $\lim_{\delta\to0}\|S_\delta
1751: v-v\|_s=0$.
1752: \item[4.] For $\delta\ge0$ the operator $S_\delta$ is a pseudodifferential
1753: operator which is uniformly of order zero.
1754: \end{enumerate}
1755: Replacing $u$ by $S_\delta\theta u$ in Lemma 6 and in the proofs of
1756: Propositions 4, 5, and 6 and using item 4 above we obtain
1757: $$
1758: \|S_\delta\varrho u\|^2_{s+1}\le C(\|S_\delta\varrho'E_0u\|^2_s+
1759: \|\tilde S_\delta\varrho'u\|^2_{s+\frac{1}{2}}+\|\tilde
1760: S_\delta\tilde\theta u\|^2_{-s_0}),
1761: $$
1762: where $\tilde S_\delta$ has the symbol $\tilde\chi(\delta\xi)$ with
1763: $\tilde\chi=1$ in a neighborhood of the support of
1764: $\chi$. Choose $m$ so that $-s_0\ge s+1-m$, then substituting $s+1-m+j$
1765: for $s$ above we obtain, by induction on $j$, that
1766: $\|S_\delta\varrho u\|^2_{s+1}$ is bounded independently of $\delta$.
1767: Hence $\varrho u\in H^{s+1}(\mathbb R^3)$ thus
1768: concluding the proof.
1769:
1770: Next we will show that in any region in which $z\neq0$ the
1771: operator $E_k$ is hypoelliptic with a gain of one
1772: derivative.
1773: \begin{prop}If $V\subset\mathbb R^3$ is an open set{\rm ,} with the property
1774: that $z\neq0$ on
1775: $V${\rm ,} and if $u$ is a distribution such that the restriction of $E_ku$
1776: to $V$ is in $C^\infty(V)${\rm ,} then the
1777: restriction of $u$ to $V$ is also in $C^\infty(U)$. More precisely{\rm ,} if
1778: $E_ku\in H_{\rm loc}^s(V)$ then $u\in H_{\rm loc}^{s+1}(V)$.
1779: \end{prop}
1780:
1781: \Proof Let $P\in V$ then $P=(\alpha,t_0)$ with $\alpha\neq0$. Let
1782: $U$ be a neighborhood of $P$ such that on $U$ we
1783: have $|z-\alpha|<a$, where $a$ is chosen as in Lemma 3, and also such
1784: that on $U$ we have $|z|\ge b>0$. Then
1785: $$
1786: \|Lu\|^2\le b^{-2k}\|\bar z^kLu\|^2,
1787: $$
1788: for all $u\in C^\infty_0(U)$. Hence Propositions 4, 5, and 6 hold with
1789: $\gamma$ replaced by $\gamma_\alpha$. The proof is then
1790: concluded using the same argument as above, replacing $S_\delta$ with
1791: $S_{\alpha,\delta}$, which is defined by $\mathcal
1792: F_\alpha(S_{\alpha,\delta}u)(\xi)=\chi(\delta\xi)\mathcal F_\alpha
1793: u(\xi)$.
1794:
1795: Now we prove microlocal hypoellipticity in the $0$ and $-$
1796: microlocalizations.
1797:
1798: \begin{prop} Given neighborhoods of the origin $U$ and $U'$ with $\bar
1799: U\subset U'$ and $|z|\le a$ on $U'${\rm ,} where $a$ is
1800: sufficiently small as in Lemma {\rm 3,} suppose that $\varrho\in
1801: C_0^\infty(U)$ and $\varrho'\in C_0^\infty(U')$ with
1802: $\varrho'=1$ on a neighborhood of $\bar U$. Further suppose that
1803: $\gamma^0\in\mathcal G^0$. Then{\rm ,} given $s \in\mathbb R${\rm ,} if $u$
1804: is a distribution such that
1805: $\varrho'E_ku\in H^s(\mathbb R^3)$ then $\varrho\Gamma^0 u\in
1806: H^{s+2}(\mathbb R^3)$.
1807: \end{prop}
1808:
1809: \Proof The proof consists of proving the following estimate
1810: $$
1811: \|S_\delta\varrho\Gamma^0u\|^2_{s+2}\le
1812: C(\varrho'E_ku\|^2_s+\|u\|^2_{-s_0}).
1813: $$
1814: Its proof is exactly analogous to the proof of Proposition 4. Replacing
1815: $u$ by $S_\delta u$ the same proof as of Lemma 6 using
1816: $XS_\delta$ instead of $X$ gives
1817: \begin{align*}
1818: \|S_\delta\varrho\Gamma^0Xu\|^2_{s+1}&=(\Lambda^sS_\delta\varrho\Gamma^0
1819: X^*Xu,\Lambda^{s+2}S_\delta\varrho\Gamma^0u)\\[4pt]
1820: &\quad+
1821: O(\|\tilde S_\delta\varrho_1\Gamma^0_1u\|^2_{s+1}+
1822: \|S_\delta\varrho\Gamma^0u\|^2_{s+2}\|\tilde
1823: S_\delta\varrho_1\Gamma^0_1u\|^2_{s+1}\\[4pt]
1824: &\quad+\|\tilde S_\delta\varrho_1\Gamma^0_1Xu\|^2_s+
1825: \|\tilde
1826: S_\delta\varrho_1\Gamma^0_1u\|_{s+1}\|\varrho\Gamma^0Xu\|_{s+1}+\|u\|^2_
1827: {-\infty}).
1828: \end{align*}
1829: The argument then proceeds exactly as in Proposition 4 and shows that\break
1830: $\|S_\delta\varrho\Gamma^0u\|^2_{s+2}$ is bounded
1831: independently of $\delta$ completing the proof.
1832:
1833: For the $-$ microlocalization we the following result follows
1834: from an argument entirely analogous to \pagebreak the above
1835: proposition.
1836: \begin{prop} \hskip-4pt Given neighborhoods of the origin $U$ and $U'$ with $\bar
1837: U\!\subset\! U'$ and $|z|\le a$ on $U'${\rm ,} where $a$ is
1838: sufficiently small as in Lemma {\rm 3}. Suppose that $\varrho\in
1839: C_0^\infty(U)$ and $\varrho'\in C_0^\infty(U')$ with
1840: $\varrho'=1$ on a neighborhood of $\bar U$. Further suppose that
1841: $\gamma^-\in\mathcal G^0$. Then{\rm ,} given $s \in\mathbb R${\rm ,} if $u$
1842: is a distribution such that $\varrho'E_ku\in H^s(\mathbb R^3)$ then
1843: $\varrho\Gamma^- u\in H^{s+1}(\mathbb R^3)$.
1844: \end{prop}
1845:
1846:
1847: \section{The operator $E_1$: no loss, no gain}
1848:
1849:
1850: As was shown in Section 5 the operator $E_1$ does not gain any
1851: derivatives. Here we will give a proof of an {\it a priori\/} estimate
1852: which shows that it does not lose any derivatives. More precisely, the
1853: estimate will show that $E_1$ does not lose any
1854: derivatives after it is proved that $E_1$ is hypoelliptic. This will be
1855: done using the same estimate with an appropriate
1856: smoothing operator in Section 14. As we have seen all the operators
1857: $E_k$ gain a derivative in regions where $z\neq0$ and in the
1858: $0$ and $-$ microlocalizations. Thus the remaining case is the $+$
1859: microlocalization when $z=0$. Since the operators $E_k$
1860: are invariant under translation in the $t$ direction it will suffice to
1861: consider neighborhoods of the origin. In this section
1862: we will present a direct proof of the {\it a priori\/}
1863: estimates for $E_1$ which
1864: will rely on the following lemma. This proof however
1865: cannot be adopted to prove the corresponding {\it a priori\/} estimate for the
1866: operator $F_1=E_1+c$ unless $c\ge0$. In fact the same
1867: estimates will be proved when we treat the general case of $E_k$ with
1868: $k\ge1$. However that treatment is much more complicated
1869: so it might be worthwhile to note this simpler proof.
1870:
1871: In the previous section we showed that the elliptic
1872: microlocalization $\Gamma^0u$ is smooth whenever $E_ku$ is
1873: smooth. Thus we do not have to keep track of just which microlocalizing
1874: operator in $\mathfrak G^0$ is used; in
1875: order to simplify the calculations we will write $u^0$ instead of
1876: $\Gamma^0u$. Similarly, since all the commutators with
1877: $\Gamma^+$ that arise are dominated as follows $\|[\Gamma^+,R^s]u\|\le
1878: C(\|\Gamma^0u\|_{s-1}+\|u\|_{-\infty}$,
1879: we will write $u^+$ instead of $\Gamma^+$.
1880: \begin{lemma} Given a bounded open set $U\subset\mathbb R^3$ there
1881: exists $C>0$ such that
1882: $$
1883: \|u\|^2\le C(\|\bar zLu\|^2+\|\bar Lu\|^2),
1884: $$
1885: for all $u\in C_0^\infty(U)$.
1886: \end{lemma}
1887:
1888: \Proof If $u\in C_0^\infty(U)$ we have
1889: \begin{align*}
1890: \|u\|^2&=(L(z)u,u)=-(zLu,u)-(zu,\bar Lu)\le \|\bar
1891: zLu\|\|u\|+\|zu\|\|\bar Lu\|\\
1892: &\le {\rm s.c.}\|u\|^2+{\rm l.c.}(\|\bar zLu\|^2+\|\bar Lu\|^2).
1893: \end{align*}
1894: Absorbing the first term on the right into the left-hand side completes
1895: the proof.
1896:
1897: The other estimate we will use here is given in Lemma 5 with
1898: $\alpha=0$, namely
1899: \begin{equation} \label{plus}
1900: \|\bar Lu^+\|^2+\|u^+\|_{\frac{1}{2}}^2\le
1901: C(\|Lu^+\|^2+\|u^+\|^2+\|u\|_{-\infty}^2),
1902: \end{equation}
1903: for all $u\in C_0^\infty(U)$.
1904: \begin{prop}Let $U$ be a bounded neighborhood of the origin such that
1905: $|z|\le a$ on $U${\rm ,} let
1906: $\varrho,\varrho'\in C_0^\infty(U)$ with $\varrho'=1$ in a neighborhood
1907: of the support of $\varrho$. Then{\rm ,} given
1908: $s,s_0\in\mathbb R$ there exists $C=C(\varrho,\varrho',s,s_0)$ such that
1909: $$
1910: \|\Psi^{s+\frac{1}{2}}\varrho u^+\|\le
1911: C(\|\Psi^{s+\frac{1}{2}}\varrho'E_1u\|+\|\Psi^s\varrho'u\|+\|u\|_{-
1912: s_o}),
1913: $$
1914: for all $u\in C_0^\infty(\mathbb R^3)$.
1915: \end{prop}
1916:
1917: \Proof We assume that $u\in C_0^\infty(\mathbb R^3)$ and replace
1918: $u$ in (\ref{plus}) by $\varrho'\bar z\Psi^su$. Then,
1919: following the method of Proposition 4, we get
1920: \begin{align*}
1921: \|\varrho'\bar z\Psi^su^+\|_{\frac{1}{2}}^2&\le
1922: C(\|\bar zL\varrho'\Psi^su^+\|^2+\|\bar L\varrho'\Psi^s
1923: u^+\|^2+\|\varrho''\Psi^su^+\|^2+\|u\|_{-\infty}^2)\\
1924: &\le
1925: C(|(\varrho'\Psi^s(E_ku)^+,\varrho\Psi^su^+)|+\|\varrho''u^0\|_s^2+\|
1926: \varrho''\Psi^su^+\|^2+\|u\|_{-\infty}^2)\\
1927: &\le C(\|\varrho'E_ku\|^2_s+\|\varrho''u\|_s^2+\|u\|_{-\infty}^2).
1928: \end{align*}
1929: Next, we replace $u$ by $\varrho\Psi^{s+\frac{1}{2}}u^+$ in Lemma 8
1930: and, with the use of Lemma 1 and the fact that
1931: $$
1932: \|\varrho'\bar
1933: z\Psi^su^+\|_{\frac{1}{2}}^2=\|z\varrho'\Psi^{s+\frac{1}{2}}u^+\|^2+O(\|
1934: u^0\|^2_{s-\frac{1}{2}}
1935: +\|u\|_{-\infty}^2),
1936: $$
1937: we obtain
1938: \begin{align*}
1939: \|\varrho\Psi^{s+\frac{1}{2}}u^+\|^2&\le C(\|\bar
1940: zL\varrho\Psi^{s+\frac{1}{2}}u^+\|^2+
1941: \|\bar
1942: L\varrho\Psi^{s+\frac{1}{2}}u^+\|^2+\|u^0\|^2_s+\|u\|^2_{-\infty})\\
1943: &\le C(\|\varrho
1944: E_ku\|^2_{s+\frac{1}{2}}\|^2+\|L(\varrho)\Psi^{s+\frac{1}{2}}u^+\|^2\\
1945: &\phantom{\le C(}+
1946: \|\bar L(\varrho)\Psi^{s+\frac{1}{2}}u^+\|^2
1947: +\|\varrho'u^0\|^2_{s+\frac{1}{2}}+\|u\|_{-\infty}^2)\\
1948: &\le C(\|\varrho
1949: E_ku\|^2_{s+\frac{1}{2}}\|^2+\|z\varrho'\Psi^{s+\frac{1}{2}}u^+\|^2+
1950: \|\varrho'u^0\|^2_{s-\frac{1}{2}}+\|u\|_{-\infty}^2)\\
1951: &\le C(\|\varrho' E_ku\|^2_{s+\frac{1}{2}}\|^2
1952: +\|\varrho''u\|_s^2+\|u\|_{-\infty}^2).
1953: \end{align*}
1954: Then, redefining $\varrho'$ and $\varrho''$, we conclude the proof.
1955:
1956:
1957: \section{Estimates of $\varrho\bar Lu^+$ and of $\varrho L\bar Lu^+$}
1958:
1959: \medskip In this section we begin to prove the {\it a priori\/} estimates for
1960: the operators $E_k$ with $k\ge1$. These will
1961: be derived from the estimate (\ref{plus}) and the estimates in the $0$
1962: microlocalization. The main difficulty is the
1963: localization in space; one cannot have a term with the cutoff function
1964: $\varrho$ between $u$ and $L$, or $\bar L$, unless the
1965: term also contains suitable powers of $z$ and $\bar z$. Substituting
1966: $\varrho\Psi^s\bar Lu$ for $u$ in (\ref{plus}) we have
1967: $$
1968: \|\bar L\varrho\Psi^s\bar Lu^+\|^2+\|\varrho\Psi^s\bar
1969: Lu^+\|_{\frac{1}{2}}^2\le
1970: C(\|L\varrho\Psi^s\bar Lu^+\|^2+\|\varrho\Psi^s\bar
1971: Lu^+\|^2+\|u\|_{-\infty}^2),
1972: $$
1973: so that,
1974: \begin{align*}
1975: \|\varrho&\Psi^sL\bar Lu^+\|^2+\|\varrho\Psi^s\bar
1976: L^2u^+\|^2+\|\varrho\Psi^{s+\frac{1}{2}}\bar Lu^+\|\\
1977: &\le C(\|\varrho\Psi^sL\bar Lu^+\|^2+\|\varrho'\Psi^s\bar
1978: Lu^+\|^2+\|\varrho'u^0\|^2_{s+1}+\|u\|_{-\infty}^2)\\
1979: &\le C(|(\varrho\Psi^s\bar LL\bar Lu^+,\varrho\Psi^s\bar
1980: Lu^+)|+\|\varrho'\Psi^s\bar Lu^+\|^2+
1981: \|\varrho''E_ku\|^2_{s-1}+\|u\|_{-\infty}^2).
1982: \end{align*}
1983: Since $\bar LL\bar L=-\bar LE_k-\bar L^2|z|^{2k}L$, we have
1984: \begin{eqnarray*}
1985: &&\hskip-12pt |(\varrho\Psi^s\bar LL\bar Lu^+,\varrho\Psi^s\bar Lu^+)|\\
1986: &&\le
1987: C(|(\varrho\Psi^s\bar LE_ku^+,\varrho\Psi^s\bar Lu^+)|+
1988: |(\varrho\Psi^s\bar L^2|z|^{2k}Lu^+,\varrho\Psi^s\bar Lu^+)|\\
1989: &&\le {\rm l.c.}\|\varrho'E_ku\|^2_s+{\rm s.c.}\|\varrho\Psi^sL\bar Lu^+\|^2+
1990: C|(\varrho\Psi^s\bar L^2|z|^{2k}Lu^+,\varrho\Psi^s\bar Lu^+)|.
1991: \end{eqnarray*}
1992: Then, to estimate $|(\varrho\Psi^s\bar
1993: L^2|z|^{2k}Lu^+,\varrho\Psi^s\bar Lu^+)|$, we have
1994: \begin{align*}
1995: \bar L^2|z|^{2k}L&=-k\bar Lz^k\bar z^{k-1}L+\bar L|z|^{2k}\bar LL\\
1996: &=-k^2\bar L|z|^{2(k-1)}+\bar LLz^k\bar z^{k-1}-2kz^k\bar z^{k-1}T+\bar
1997: L|z|^{2k}L\bar L-2|z|^{2k}T\bar L\\
1998: &=-k^2\bar L|z|^{2(k-1)}-4kz^k\bar z^{k-1}T+k(k-1)Lz^k\bar
1999: z^{k-2}+kLz^k\bar z^{k-1}\bar L\\*
2000: &\quad+\bar L|z|^{2k}L\bar L-2|z|^{2k}T\bar L,
2001: \end{align*}
2002: and, using integration by parts, we get
2003: \begin{align*}
2004: |(\varrho\Psi^s\bar L|z|^{2(k-1)}u^+,\varrho\Psi^s\bar Lu^+)|&\le
2005: {\rm l.c.}\|z^{2(k-1)}\varrho\Psi^su^+\|^2+\mathcal E_1,\\[4pt]
2006: |(\varrho\Psi^sz^k\bar z^{k-1}Tu^+,\varrho\Psi^s\bar Lu^+)|&\le
2007: {\rm l.c.}\|z^{2k-1}\varrho\Psi^{s+\frac{1}{2}}u^+\|^2+\mathcal E_2,\\[4pt]
2008: (k-1)|(\varrho\Psi^sLz^k\bar z^{k-2}u^+,\varrho\Psi^s\bar Lu^+)|&\le
2009: (k-1)({\rm l.c.}\|z^{2(k-1)}\varrho\Psi^su^+\|^2+\mathcal E_3),\\[4pt]
2010: |(\varrho\Psi^sLz^k\bar z^{k-1}\bar Lu^+,\varrho\Psi^s\bar Lu^+)|&\le
2011: {\rm l.c.}\|z^{2k-1}\varrho\Psi^su^+\|^2+\mathcal E_4,\\[4pt]
2012: |(\varrho\Psi^s\bar L|z|^{2k}L\bar Lu^+,\varrho\Psi^s\bar Lu^+)|&\le
2013: \mathcal E_4,\\[4pt]
2014: \noalign{\noindent \rm and}
2015: |(\varrho\Psi^s|z|^{2k}T\bar Lu^+,\varrho\Psi^s\bar Lu^+)|&\le \mathcal
2016: E_2,
2017: \end{align*}
2018: where
2019: \begin{align*}
2020: \mathcal E_1&\sim\|\varrho'u^0\|^2_s+\|\varrho'\Psi^s\bar
2021: Lu^+\|^2+\|u\|^2_{-\infty},\\[4pt]
2022: \mathcal E_2&\sim {\rm s.c.}\|\varrho\Psi^{s+\frac{1}{2}}\bar
2023: Lu^+\|^2+\mathcal E_1,\\[4pt]
2024: \mathcal E_3&\sim {\rm s.c.}\|\varrho\Psi^s\bar L^2u^+\|^2+\mathcal E_1,\\[4pt]
2025: \noalign{\noindent\rm and}
2026: \mathcal E_4&\sim {\rm s.c.}\|\varrho\Psi^sL\bar Lu^+\|^2+\mathcal E_1.
2027: \end{align*}
2028:
2029: \vglue-23pt
2030: \begin{align*}
2031: |(\Psi^s\varrho\bar L|z|^{2(k-1)}u^+,\Psi^s\varrho \bar Lu^+)|&\le
2032: C(\|z^{2k-2}\Psi^s\varrho u^+\|^2
2033: +\mathcal E_2),\\[4pt]
2034: |(\Psi^s\varrho\bar Lz^{k-1}\bar z^k\bar Lu^+,\Psi^s\varrho \bar
2035: Lu^+)|&\le C\mathcal E_1\\[4pt]
2036: |(\Psi^sz^k\bar z^{k-1}\varrho Tu^+,\Psi^s\varrho \bar Lu^+)|&\le
2037: C(\|z^{2k-1}\Psi^{s+\frac{1}{2}}\varrho u^+\|^2+
2038: \mathcal E_1+\mathcal E_3)\\[4pt]
2039: |(\Psi^s|z|^{2k}T\varrho\bar Lu^+,\Psi^s\varrho \bar Lu^+)|&\le
2040: C(\mathcal E_1+\mathcal E_2)\\[4pt]
2041: |(\Psi^sLz^k\bar z^{k-2}\varrho u^+,\Psi^s\varrho \bar Lu^+)|&\le
2042: C(\|z^{2k-2}\Psi^s\varrho u^+\|^2+
2043: \mathcal E_1+\mathcal E_4)\\[4pt]
2044: |(\Psi^sLz^k\bar z^{k-1}\varrho\bar Lu^+,\Psi^s\varrho \bar Lu^+)|&\le
2045: C(\mathcal E_1+\mathcal E_4),\\[4pt]
2046: \noalign{\noindent \rm and}
2047: |(\Psi^sL|z|^{2k}\varrho\bar L^2u^+,\Psi^s\varrho \bar Lu^+)|&\le
2048: C(\mathcal E_1+\mathcal E_4).
2049: \end{align*}
2050: Again, let $\{\varrho_i\}$ be a sequence of cutoff functions as defined
2051: in Section 2. Then substituting $\varrho_{i}$ for
2052: $\varrho$, $s-\frac{i-1}{2}$ for $s$, and $\varrho_{i+1}$ for
2053: $\varrho'$, we get
2054: \begin{align*}
2055: &\hskip-12pt\|\varrho_i\Psi^{s-\frac{i-1}{2}}L\bar
2056: Lu^+\|^2+\|\varrho_i\Psi^{s+1-\frac{i}{2}}\bar Lu^+\|^2\\[4pt]
2057: &\quad\le
2058: C(\|\varrho'E_ku\|^2_{s-\frac{i-1}{2}}+\|z^{2k-2}\Psi^{s-\frac{i-
2059: 1}{2}}\varrho_i u^+\|^2\\[4pt]
2060: &\qquad+\|z^{2k-1}\Psi^{s+1-\frac{i}{2}}\varrho_i u^+\|^2
2061: +\|\varrho_{i+1}\Psi^{s-\frac{i-1}{2}}\bar Lu^+\|^2
2062: +\|u\|^2_{-\infty}).
2063: \end{align*}
2064: Then we obtain the following, by substituting these inequalities into
2065: each other for successive $i$
2066: \begin{eqnarray*}
2067: &&\|\varrho\Psi^{s}L\bar Lu^+\|^2+\|\varrho\Psi^{s+\frac{1}{2}}\bar
2068: Lu^+\|^2\\[4pt]
2069: && \le
2070: C\Big(\sum_{i=1}^N\big(\|\varrho_i\Psi^{s-\frac{i-1}{2}}\bar
2071: LE_ku^+\|^2\!+\|z^{2k-2}\Psi^{s-\frac{i-1}{2}}\varrho_i
2072: u^+\|^2+\|z^{2k-1}\Psi^{s+1-\frac{i}{2}}\varrho_i u^+\|^2\big)
2073: \\[4pt]
2074: &&\qquad+\|\varrho_{N+1}\Psi^{s-\frac{N-1}{2}}\bar Lu^+\|^2
2075: +\|\varrho_NE_ku\|_{s-1}^2+\|u\|^2_{-\infty}\Big).
2076: \end{eqnarray*}
2077: Given $s_o$ we choose $N>2(s-s_o)+1$ then we obtain the following
2078: estimate which will be repeatedly used in establishing
2079: the {\it a priori\/} estimates for $E_k$
2080: \begin{align}\label{central}
2081: \|\varrho\Psi^sL\bar Lu^+\|^2&+\|\varrho\Psi^{s+\frac{1}{2}}\bar
2082: Lu^+\|^2\le
2083: C(\|\varrho'E_ku^+\|^2_s+\|z^{2k-2}\Psi^s\varrho' u^+\|^2\\[4pt]
2084: &\notag+\|z^{2k-1}\Psi^{s+\frac{1}{2}}\varrho' u^+\|^2 +\|u\|^2_{-s_o}).
2085: \end{align}
2086:
2087:
2088:
2089: \section{Estimates of $\|z^j\Psi^{s+ja}\varrho u^+\|$}
2090:
2091:
2092: \begin{lemma} If $a>0$ then for $m\in\mathbb Z^+$ and a small
2093: constant\/ ${\rm s.c.}$ there exists a constant\/ ${\rm l.c.}$ such that
2094: \begin{eqnarray*}
2095: \sum_{j=1}^{m-1}\|z^j\Psi^{s+ja}\varrho u^+\|^2&\le &
2096: {\rm l.c.}\|z^m\Psi^{s+ma}\varrho u^+\|^2+{\rm s.c.}\|\Psi^s\varrho u\|^2
2097: \\*
2098: &&+C(\|\varrho'u^0\|^2_{s+(m-1)a-1}+\|u\|^2_{-\infty}),\\*
2099: \noalign{\noindent for all $u\in C^\infty(U)$.}
2100: \end{eqnarray*}
2101: \end{lemma}
2102:
2103: \vskip-12pt
2104: {\it Proof}. For $m=2$ we have
2105: \begin{align*}
2106: \|z\Psi^{s+a}\varrho u^+\|^2&=(|z|^2\Psi^{s+2a}\varrho
2107: u^+,\Psi^{s}\varrho u^+)+O(\|\varrho'u^0\|^2_{s+a-1}
2108: +\|u\|^2_{-\infty})\\
2109: &\le {\rm l.c.}\|z^2\Psi^{s+2a}\varrho u^+\|^2+{\rm s.c.}\|\Psi^{s}\varrho
2110: u^+\|^2+C(\|\varrho'u^0\|^2_{s+a-1}+\|u\|^2_{-\infty}).
2111: \end{align*}
2112: For $m>2$ we assume
2113: \begin{eqnarray*}
2114: \sum_{j=1}^{m-2}\|z^j\Psi^{s+ja}\varrho u^+\|^2&\le&
2115: {\rm l.c.}\|z^{m-1}\Psi^{s+(m-1)a}\varrho u^+\|^2+{\rm s.c.}\|\Psi^s\varrho
2116: u\|^2\\
2117: &&+C(\|\varrho u^0\|^2_{s+(m-2)a-1}+\|u\|^2_{-\infty}),
2118: \end{eqnarray*}
2119: and we have
2120: \begin{eqnarray*}
2121: \|z^{m-1}\Psi^{s+(m-1)a}\varrho u^+\|^2&=&(z^{m}\bar z\Psi^{s+ma}\varrho
2122: u^+,z^{m-2}\Psi^{s+(m-2)a}\varrho u^+)\\
2123: &&+O(\|\varrho
2124: u^0\|^2_{s+(m-1)a-1}+\|u\|^2_{-\infty})\\
2125: & \le& {\rm l.c.}\|z^{m}\Psi^{s+ma}\varrho
2126: u^+\|^2+{\rm s.c.}\|z^{m-2}\Psi^{s+(m-2)a}\varrho u^+\|^2\\
2127: &&+
2128: C(\|\varrho u^0\|^2_{s+(m-1)a-1}+\|u\|^2_{-\infty}).
2129: \end{eqnarray*}
2130: Adding this to the above and absorbing the term multiplied by ${\rm s.c.}$ in
2131: the right-hand side we conclude the proof.
2132:
2133: \begin{lemma} If $0<j<m$ and if $\frac{mA}{j}<B$ then for any
2134: ${\rm s.c.}$ and any $N$ there exists $C_N$ such that
2135: \begin{eqnarray*}
2136: \|z^j\Psi^{s+A}\varrho u^+\|^2&\le& {\rm s.c.}(\|z^m\Psi^{s+B}\varrho
2137: u^+\|^2+\|\Psi^{s}\varrho u^+\|^2)\\
2138: &&+C(\|\varrho u^0\|^2_{s+B-1}+C_N\|u^+\|^2_{-N}+C(\|\varrho
2139: u^0\|^2_{s+B-1}+\|u\|^2_{-\infty}),
2140: \end{eqnarray*}
2141: for all $u\in C^\infty_0(U)$.
2142: \end{lemma}
2143:
2144: \Proof With $a=\frac{A}{j}$ we have
2145: \begin{eqnarray*}
2146: \|z^j\Psi^{s+A}\varrho u^+\|^2&\le& {\rm l.c.}\|z^m\Psi^{s+ma}\varrho
2147: u^+\|^2+s.c\|\Psi^{s}\varrho u^+\|^2\\
2148: &&+
2149: C(\|\varrho u^0\|^2_{s+ma-1}+\|u\|^2_{-\infty}).
2150: \end{eqnarray*}
2151: Since $ma=\frac{mA}{j}<B$,
2152: $$
2153: \psi^{s+ma}(\xi)\le {\rm s.c.}\psi^{s+B}(\xi)+{\rm l.c.}(1+|\xi|^2)^{-\frac{N}{2}}.
2154: $$
2155: Then
2156: \begin{eqnarray*}
2157: \|z^m \Psi^{s+ma}\varrho u^+\|^2&=&\|\Psi^{s+ma}z^m\varrho
2158: u^+\|^2+O(\|u^0\|^2_{s+ma-1}+\|u\|^2_{-\infty})\\
2159: &\le& {\rm s.c.}\|\Psi^{s+B}z^m\varrho u^+\|^2+C_N\|u^+\|^2_{-N}\\
2160: &&+O(\|\varrho
2161: u^0\|^2_{s+ma-1}+\|u\|^2_{-\infty})\\
2162: &\le &{\rm s.c.}\|z^m\Psi^{s+B}\varrho u^+\|^2+C_N\|u^+\|^2_{-N}\\
2163: &&+O(\|\varrho
2164: u^0\|^2_{s+B-1}+\|u\|^2_{-\infty}).
2165: \end{eqnarray*}
2166: Combining with the above we conclude the proof of the lemma.
2167:
2168:
2169: \begin{lemma} If $\sigma=\frac{1}{2k}$ and if $1\le j\le k$ then
2170: $$
2171: \|z^j\Psi^{s+j\sigma}\varrho u^+\|^2\le
2172: C(\|\varrho'E_ku\|^2_s+\|\varrho'u\|^2_s+\|u\|^2_{-\infty}),
2173: $$
2174: for all $u\in C^\infty_0(U)$.
2175: \end{lemma}
2176:
2177: \Proof First note that
2178: $$
2179: \|\varrho z^k\Psi^su^+\|^2\le
2180: C(\|\varrho'E_ku\|^2_{s-\frac{1}{2}}+\|\varrho'\Psi^{s-
2181: \frac{1}{2}}u^+\|^2+\|u\|^2_{-\infty}).
2182: $$
2183: Then, replacing $s$ by $s+k\sigma$, since $k\sigma-\frac{1}{2}=0$, we
2184: have
2185: \begin{align*}
2186: \|z^j\Psi^{s+j\sigma}\varrho u^+\|^2&\le C(\|z^k\Psi^{s+k\sigma}\varrho
2187: u^+\|^2+\|\Psi^s\varrho
2188: u\|^2\\
2189: &\quad+C\|\varrho'u^0\|^2_{s+(k-1)a-1} +\|u\|^2_{-\infty})\\
2190: &\le C(\|\bar z^kL\Psi^s\varrho u^+\|^2+\|\bar L\Psi^s\varrho
2191: u^+\|^2+\|\varrho'u^+\|^2_s+\|u\|^2_{-\infty})\\
2192: &\le C(\|\varrho'E_ku\|^2_s+\|\varrho'u\|^2_s+\|u\|^2_{-\infty}).
2193: \end{align*}
2194:
2195:
2196: \section{Estimate of $\|\varrho\Psi^{s+\sigma}u^+\|$}
2197:
2198:
2199: \begin{lemma} There exists a $C>0$ such that
2200: $$
2201: \|\varrho\Psi^{s+\sigma}u^+\|\le
2202: C(\|\varrho'E_ku\|^2_{s+\sigma+k-1}+\|\varrho'u\|^2_s+\|u\|^2_{-
2203: \infty}),
2204: $$
2205: for all $u\in C^\infty_0(U)$.
2206: \end{lemma}
2207:
2208: \Proof
2209: \begin{align*}
2210: \|\Psi^{s+\sigma}\varrho u^+\|^2&=(L(z)\Psi^{s+\sigma}\varrho
2211: u^+,\Psi^{s+\sigma}\varrho u^+)\\
2212: &=-(zL\Psi^{s+\sigma}\varrho u^+,\Psi^{s+\sigma}\varrho
2213: u^+)-(z\Psi^{s+\sigma}\varrho u^+,\bar L\Psi^{s+\sigma}\varrho u^+)\\
2214: &\le {\rm l.c.}\|z\Psi^{s+\sigma}\varrho
2215: Lu^+\|^2+C\|\Psi^{s+\sigma}\varrho\bar L u^+\|^2+``error",
2216: \end{align*}
2217: where,
2218: $$
2219: ``{\rm error}"\le {\rm s.c.}\|\Psi^{s+\sigma}\varrho
2220: u^+\|^2+C(\|z\Psi^{s+\sigma}\varrho' u^+\|^2+\|\varrho
2221: u^0\|^2_{s+\sigma}+\|u\|^2_{-\infty}).
2222: $$
2223: In the estimate of the ``error'' the first term on the right gets
2224: absorbed and the other terms are estimated as follows.
2225: $$
2226: \|z\Psi^{s+\sigma}\varrho' u^+\|^2\le C(\|\varrho'
2227: Eu\|^2_s+\|\varrho'u\|^2_s+\|u\|^2_{-\infty}).
2228: $$
2229: The third term, which is microlocalized in the elliptic region, is
2230: estimated by
2231: $$
2232: \|\varrho u^0\|^2_{s+\sigma}\le C(\|\varrho
2233: Eu\|^2_{s+\sigma-2}+\|\varrho'u\|^2_s).
2234: $$
2235: Hence we get
2236: \begin{eqnarray*}
2237: \|\Psi^{s+\sigma}\varrho u^+\|^2&\le&C\Big(\|z\Psi^{s+\sigma}\varrho
2238: Lu^+\|^2+\|\Psi^{s+\sigma}\varrho\bar Lu^+\|^2\\
2239: &&\phantom{C\big(}+\|\varrho Eu\|^2_s
2240: +\|\varrho'u\|^2_s+\|u\|^2_{-\infty}\Big).
2241: \end{eqnarray*}
2242: From (\ref{central}) we have
2243: $$
2244: \|\Psi^{s+\sigma}\varrho\bar Lu^+\|^2\le C(\|\varrho
2245: Eu\|^2_{s+\sigma-\frac{1}{2}}+\|\varrho'u\|^2_s+\|u\|^2_{-\infty}).
2246: $$
2247: So the term that remains to be estimated is $\|z\Psi^{s+\sigma}\varrho
2248: Lu^+\|^2$, and we have
2249: \begin{eqnarray*}
2250: \|z\Psi^{s+\sigma}\varrho
2251: Lu^+\|^2&=&(|z|^2\Psi^{s+\sigma+\frac{1}{2}}\varrho
2252: Lu^+,\Psi^{s+\sigma-\frac{1}{2}}\varrho Lu^+)
2253: \\
2254: &&+O(\|u^0\|^2_{s+\sigma-2}+\|u\|^2_{-\infty})\\
2255: &\le& {\rm l.c.}\|z^2\Psi^{s+\sigma+\frac{1}{2}}\varrho
2256: Lu^+\|^2+{\rm s.c.}\|\Psi^{s+\sigma-\frac{1}{2}}\varrho Lu^+\|^2
2257: \\
2258: &&+O(\|u^0\|^2_{s+\sigma-2}+\|u\|^2_{-\infty})
2259: \end{eqnarray*}
2260: and
2261: \begin{eqnarray*}
2262: \|\Psi^{s+\sigma-\frac{1}{2}}\varrho Lu^+\|^2&=&(\Psi^{s+\sigma}\varrho
2263: Lu^+,\Psi^{s+\sigma-1}\varrho Lu^+)
2264: \\
2265: &=&-(\Psi^{s+\sigma}\varrho u^+,\bar L\Psi^{s+\sigma-1}\varrho
2266: Lu^+)+\mathcal E_1\\
2267: &=&-(\Psi^{s+\sigma}\varrho u^+,[\bar L,\Psi^{s+\sigma-1}\varrho
2268: L]u^+)\\
2269: &&-(\Psi^{s+\sigma}\varrho u^+,\Psi^{s+\sigma-1}\varrho L\bar
2270: Lu^+)+\mathcal E_1\\
2271: &\le& C(\|\Psi^{s+\sigma}\varrho u^+\|^2+\|[\bar
2272: L,\Psi^{s+\sigma-1}\varrho L]u^+\|^2 \\
2273: &&+\|\Psi^{s+\sigma-1}\varrho L\bar
2274: Lu^+\|^2+\mathcal E_1.
2275: \end{eqnarray*}
2276: The second term is estimated as follows
2277: \begin{eqnarray*}
2278: [\bar L,\Psi^{s+\sigma-1}\varrho L]u^+&=&[\bar
2279: L,\Psi^{s+\sigma-1}]\varrho Lu^++\Psi^{s+\sigma-1}\bar L(\varrho)Lu^+
2280: \\
2281: &&-2\Psi^{s+\sigma-1}\varrho Tu^++\Psi^{s+\sigma-1}\varrho L\bar Lu^+
2282: \end{eqnarray*}
2283: so that
2284: \begin{eqnarray*}
2285: \|[\bar L,\Psi^{s+\sigma-1}]\varrho Lu^+\|^2&\le &
2286: C(\|\Psi^{s+\sigma-1}\Gamma^0\varrho u^+\|^2+\|u\|^2_{-\infty})\\
2287: &\le &
2288: C(\|\varrho'u^0\|^2_{s+\sigma}
2289: +\|u\|^2_{-\infty}),
2290: \end{eqnarray*}
2291: and
2292: \begin{multline*}
2293: \|\Psi^{s+\sigma-1}\bar L(\varrho)Lu^+\|^2+\|\Psi^{s+\sigma-1}\varrho
2294: Tu^+\|^2\\
2295: \le C(\|z\Psi^{s+\sigma}\varrho'u^+\|^2
2296: +\|\Psi^{s+\sigma}\varrho u\|^2)+\mathcal E_2.
2297: \end{multline*}
2298: Furthermore we have
2299: $$
2300: \|\Psi^{s+\sigma-1}\varrho L\bar Lu^+\|^2\le C\|\varrho'E_k
2301: u\|^2_{s+\sigma-\frac{1}{2}}+\mathcal E_3.
2302: $$
2303: The terms $\mathcal E$ are bounded as follows
2304: $$
2305: \mathcal E_1\le
2306: C(\|u^0\|^2_{s+\sigma}+\|z\Psi^{s+\sigma}\varrho'u^+\|^2+\|u\|^2_{-
2307: \infty}).
2308: $$
2309: By Lemma 10 we get
2310: \begin{eqnarray*}
2311: \mathcal E_1&\le&
2312: C(\|\varrho'E_ku\|^2_s+\|\varrho'u\|^2_s+\|u\|^2_{-\infty}),
2313: \\[4pt]
2314: \mathcal E_2&\le& C(\varrho'u\|^2_{s+\sigma-1}+\mathcal E_1)\le
2315: C'\mathcal E_1,
2316: \end{eqnarray*}
2317: and
2318: $$
2319: \mathcal E_3\le C(\|z^{2k-1}\varrho u^+\|^2_s+\|z^{2k-2}\varrho
2320: u^+\|^2_{s-\frac{1}{2}}+\mathcal E_2)\le C'\mathcal E_2.
2321: $$
2322: Hence we have
2323: \begin{multline*}
2324: \|\Psi^{s+\sigma}\varrho u^+\|^2+\|z\Psi^{s+\sigma}\varrho Lu^+\|^2\\
2325: \le
2326: C(\|z^2\Psi^{s+\sigma+\frac{1}{2}}\varrho Lu^+\|^2+
2327: \|\varrho'E_ku\|^2_s+\|\varrho'u\|_s^2+\|u\|^2_{-\infty}).
2328: \end{multline*}
2329: To estimate the first term on the right we will use Lemma 8 as follows.
2330: $$
2331: \|z^2\Psi^{s+\sigma+\frac{1}{2}}\varrho Lu^+\|^2\le
2332: C(\|z\Psi^{s+\sigma+\frac{1}{2}}\varrho zLu^+\|^2+\|\varrho
2333: u^0\|^2_{s+\sigma}
2334: +\|u\|^2_{-\infty}).
2335: $$
2336: We apply Lemma 8 with $a=\frac{1}{2}$, $m=k-1$, $s$ replaced by
2337: $s+\sigma$, and $u$ replaced by $zLu$ to obtain
2338: \begin{align*}
2339: \|z^2\Psi^{s+\sigma+\frac{1}{2}}\varrho Lu^+\|^2&\le
2340: {\rm l.c.}\|z^{k-1}\Psi^{s+\sigma+\frac{k-1}{2}}\varrho zLu^+\|^2+
2341: {\rm s.c.}\|z\Psi^{s+\sigma}\varrho Lu^+\|^2\\[4pt]
2342: &\quad+\|\varrho u^0\|^2_{s+\sigma+\frac{k-1}{2}}+\|u\|^2_{-\infty})\\[4pt]
2343: &\le {\rm l.c.}\|z^k\Psi^{s+\sigma+\frac{k-1}{2}}\varrho Lu^+\|^2+
2344: {\rm s.c.}\|z\Psi^{s+\sigma}\varrho Lu^+\|^2\\[4pt]
2345: &\quad+C(\|\varrho u^0\|^2_{s+\sigma+\frac{k-1}{2}}+\|u\|^2_{-\infty}).
2346: \end{align*}
2347: Therefore we have
2348: \begin{align*}
2349: &\hskip-16pt\|\Psi^{s+\sigma}\varrho u^+\|^2\\[4pt]
2350: & \le
2351: C(\|z^k\Psi^{s+\sigma+\frac{k-1}{2}}\varrho
2352: Lu^+\|^2+\|\varrho'E_ku\|^2_{s+\sigma+\frac{k-1}{2}-2}+
2353: \|\varrho'u\|_s^2+\|u\|^2_{-\infty})\\[4pt]
2354: & \le C(\|\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar
2355: z^kLu^+\|^2+\|\varrho'E_ku\|^2_{s+\sigma+\frac{k-1}{2}-2}+
2356: \|\varrho'u\|_s^2+\|u\|^2_{-\infty}).
2357: \end{align*}
2358: Next, from Lemma 8 with $m=k$, $a=\frac{1}{2}$ and $s$ replaced by
2359: $s+\sigma$, we have
2360: \begin{eqnarray*}
2361: \|z\Psi^{s+\sigma+\frac{1}{2}}u\|^2&\le &
2362: {\rm l.c.}\|z^k\Psi^{s+\sigma+\frac{k}{2}}\varrho
2363: u^+\|^2\\[4pt]
2364: &&+{\rm s.c.}\|\Psi^{s+\sigma}\varrho u^+\|^2+
2365: C(\|\varrho'u^0\|^2_{s+\sigma+\frac{k-1}{2}-1}+\|u\|^2_{-\infty})\\[4pt]
2366: &\le &C\|\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar z^kLu^+\|^2+\mathcal
2367: E_1
2368: \end{eqnarray*}
2369: \pagebreak
2370:
2371: \noindent
2372: and
2373: \begin{align*}
2374: \|z^k\Psi^{s+\sigma+\frac{k}{2}}\varrho u^+\|^2&\le
2375: C\|\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar z^kLu^+\|^2+\mathcal E_1\\
2376: &=-C(\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar
2377: L|z|^{2k}Lu^+,\Psi^{s+\sigma+\frac{k-1}{2}}\varrho u^+)\\
2378: &\quad-2C(\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar
2379: z^{k}Lu^+,\Psi^{s+\sigma+\frac{k-1}{2}}\bar z^{k+1}\mu u^+)+\mathcal E_2
2380: \end{align*}
2381: since
2382: \begin{align*} \hskip-8pt
2383: |(\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar
2384: z^{k}Lu^+,\Psi^{s+\sigma+\frac{k-1}{2}}\bar z^{k+1}\mu u^+)|&\le
2385: {\rm s.c.}\|z^{k}\Psi^{s+\sigma+\frac{k}{2}}\varrho Lu^+\|^2\\
2386: &\quad+{\rm l.c.}\|z^{k+1}\Psi^{s+\sigma+\frac{k-
2387: 2}{2}}\varrho'u^+\|^2+\mathcal E_2.
2388: \end{align*}
2389: Hence we obtain
2390: \begin{align*}
2391: \|z^k\Psi^{s+\sigma+\frac{k}{2}}\varrho u^+\|^2 &\le
2392: C\|\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar z^kLu^+\|^2+\mathcal E_2\\
2393: &\le
2394: C|(\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar
2395: L|z|^{2k}Lu^+,\Psi^{s+\sigma+\frac{k-1}{2}}\varrho u^+)|+\mathcal E_3\\
2396: &\le C|(\Psi^{s+\sigma+\frac{k-1}{2}}\varrho
2397: E_ku^+,\Psi^{s+\sigma+\frac{k-1}{2}}\varrho u^+)|\\
2398: &\quad+
2399: |(\Psi^{s+\sigma+\frac{k-1}{2}}\varrho L\bar
2400: Lu^+,\Psi^{s+\sigma+\frac{k-1}{2}}\varrho u^+)|+\mathcal E_3\\
2401: &\le C(\|\Psi^{s+\sigma+k-1}\varrho E_ku\|^2\\
2402: &\quad+
2403: |(\Psi^{s+\sigma+\frac{k-1}{2}}\varrho \bar
2404: Lu^+,\Psi^{s+\sigma+\frac{k-1}{2}}\bar z\mu u^+)|)+\mathcal E_4\\
2405: &\le C(\|\Psi^{s+\sigma+k-1}\varrho
2406: E_ku\|^2+\|\Psi^{s+\sigma+k-1}\varrho \bar Lu^+\|^2)+\mathcal E_5\\
2407: &\le C(\|\varrho
2408: E_ku\|_{s+\sigma+k-1}^2+\|z^{2k-1}\Psi^{s+\sigma+k-1}\varrho' u^+\|^2\\
2409: &\quad +\|z^{2k-2}\Psi^{s+\sigma+k-\frac{3}{2}}\varrho'
2410: u^+\|^2)+\mathcal E_6.
2411: \end{align*}
2412: Thus, applying Lemma 9 with $m=2k-1$, $j=2k-2$, $A=k-\frac{3}{2}$,
2413: $B=k-1$, and $s$ replaced by $s+\sigma$, we have
2414: $$
2415: \frac{mA}{j}=\frac{2k-1}{2k-2}(k-\frac{3}{2})<k-1=B.
2416: $$
2417: Now,
2418: $$
2419: \|z^{2k-2}\Psi^{s+\sigma+k-\frac{3}{2}}\varrho' u^+\|^2\le
2420: {\rm s.c.}\|z^{2k-1}\Psi^{s+\sigma+k-1}\varrho' u^+\|^2+\mathcal E_7.
2421: $$
2422: Replacing $\varrho u^+$ by $\bar z^{k-1}\varrho'u^+$ and $s$ by
2423: $s+\frac{k-2}{2}$ we obtain
2424: \begin{align*} &\hskip-3pt
2425: \|z^{2k-1}\Psi^{s+\sigma+k-1}\varrho' u^+\|^2\\&\le
2426: C(\|\Psi^{s+\sigma+k-\frac{3}{2}}\varrho'\bar z^{2k-1}Lu^+\|^2+\mathcal
2427: E_8\\
2428: &\le
2429: C(|(\Psi^{s+\sigma+k-\frac{3}{2}}\bar
2430: L\varrho'|z|^{2k}Lu^+,|z|^{2k-2}\Psi^{s+\sigma+k-\frac{3}{2}}\varrho'
2431: u)|\\
2432: &\quad+{\rm s.c.}\|\Psi^{s+\sigma+k-1}\varrho'\bar z^{2k-1}Lu^+\|^2+
2433: {\rm l.c.}\|z^{2k-1}\Psi^{s+\sigma+k-2}\varrho''
2434: u\|^2) +\mathcal E_8\\
2435: &\le C(\|\varrho' E_ku\|^2_{s+\sigma+2k-3}+
2436: |(\Psi^{s+\sigma+k-\frac{3}{2}}\varrho' L\bar
2437: Lu^+,|z|^{2k-2}\Psi^{s+\sigma+k-\frac{3}{2}}\varrho' u)|)+\mathcal E_9\\
2438: &\le C(\|\varrho'
2439: E_ku\|^2_{s+\sigma+2k-3}+|(\Psi^{s+\sigma+k-2}\varrho''\bar
2440: Lu^+,z|z|^{2k-2}\Psi^{s+\sigma+k-1}\varrho' u)|\\
2441: &\quad+|(\Psi^{s+\sigma+k-1}\varrho'\bar Lu^+,z^{k-1}\bar
2442: z^{k-2}\Psi^{s+\sigma+k-2}\varrho' u)|)+\mathcal E_9\\
2443: &\le C(\|\varrho'
2444: E_ku\|^2_{s+\sigma+2k-3}+{\rm l.c.}\|\Psi^{s+\sigma+k-2}\varrho''\bar
2445: Lu^+\|^2+
2446: {\rm s.c.}\|z^{2k-1}\Psi^{s+\sigma+k-1}\varrho' u\|^2\\
2447: &\quad+{\rm s.c.}\|\Psi^{s+\sigma+k-1}\varrho'\bar
2448: Lu^+\|^2+{\rm l.c.}\|z^{2k-3}\Psi^{s+\sigma+k-2}\varrho' u^+\|^2)+\mathcal
2449: E_9.
2450: \end{align*}
2451: Now applying Lemma 9 as above but with $j=2k-3$ and $A=k-2$, we get
2452: $$
2453: \frac{mA}{j}=\frac{2k-1}{2k-3}(k-2)<k-1=B.
2454: $$
2455: Hence
2456: $$
2457: \|z^{2k-3}\Psi^{s+\sigma+k-2}\varrho' u^+\|^2\le
2458: {\rm s.c.}\|z^{2k-1}\Psi^{s+\sigma+k-1}\varrho' u^+\|^2+\mathcal E_9.
2459: $$
2460: Combining the above we obtain
2461: $$
2462: \|\Psi^{s+\sigma}\varrho u^+\|^2\le C\|\varrho'
2463: E_ku\|_{s+\sigma+k-1}^2+\mathcal E_{10}.
2464: $$
2465: To complete the proof of the {\it a~priori\/} estimate we will analyze the
2466: error terms:
2467: \begin{align*}
2468: &\mathcal E_1\sim\|z^k\Psi^{s+\sigma+\frac{k-1}{2}}\varrho
2469: u^+\|^2+\|\varrho'u^0\|^2_{s+\sigma+\frac{k-1}{2}}
2470: +{\rm s.c.}\|\Psi^{s+\sigma}\varrho u^+\|^2+\|u\|^2_{-\infty},\\[4pt]
2471: &\mathcal E_2\sim\mathcal
2472: E_1+\|z^{k+1}\Psi^{s+\sigma+\frac{k-1}{2}}\varrho'u^+\|^2,\\[4pt]
2473: &\mathcal E_3\sim\mathcal
2474: E_2+\|\varrho'u^0\|^2_{s+\sigma+\frac{k}{2}}+{\rm s.c.}\|\Psi^{s+\sigma+\frac{
2475: k-1}{2}}\varrho\bar z^kLu^+\|^2,\\[4pt]
2476: &\mathcal E_4\sim\mathcal
2477: E_3+\|\Psi^{s+\sigma+\frac{k-1}{2}}\varrho\bar Lu^+\|^2,\\[4pt]
2478: &\mathcal E_5\sim\mathcal E_4+\|z\Psi^{s+\sigma}\varrho'u^+\|^2\\[4pt]
2479: &\mathcal E_6\sim\mathcal
2480: E_5+\|z^{2k-1}\Psi^{s+\sigma+k-\frac{3}{2}}\varrho'u^+\|^2
2481: +\|z^{2k-2}\Psi^{s+\sigma+k-2}\varrho'u^+\|^2\\[4pt]
2482: &\mathcal E_7\sim\mathcal E_6+\|\varrho u^0\|^2_{s+k-2}+\|u^+\|^2_{-N}\\[4pt]
2483: &\mathcal E_8\sim\mathcal
2484: E_7+\|\Psi^{s+\sigma+k-\frac{3}{2}}\varrho'\bar z^{2k}Lu^+\|^2\\[4pt]
2485: &\mathcal E_9\sim\mathcal E_8+{\rm s.c.}\|\Psi^{s+\sigma+k-1}\varrho\bar
2486: z^{2k-1}Lu^+\|^2+\|z^{2k-1}\Psi^{s+\sigma+k-2}\varrho'u\|^2,\\[4pt]
2487: \noalign{\noindent {\rm and}}
2488: &\mathcal E_{10}\sim\mathcal E_9+{\rm s.c.}\|\Psi^{s+\sigma+k-1}\varrho\bar
2489: Lu^+\|^2+\|\Psi^{s+\sigma+k-2}\varrho'\bar Lu^+\|^2.
2490: \end{align*}
2491: The ``admissible" errors are $\|\varrho'u\|^2_s+\|u\|^2_{-\infty}$. The
2492: terms involving $u^0$ are all bounded by
2493: ${\rm const}.\|\varrho'E_ku\|^2_{s+\sigma+k-2}$ modulo admissible errors. The
2494: terms involving a small constant
2495: ${\rm s.c.}$ are absorbed in the left. The term
2496: $\|z\Psi^{s+\sigma}\varrho'u^+\|$ is bounded by
2497: ${\rm const.}\|\varrho'E_ku\|^2_s$, and the remaining terms can
2498: be bounded by a constant times $\mathcal A(s,\varrho')$, where
2499: $\mathcal A(s,\varrho')$ is defined by
2500: $$
2501: \mathcal A(s,\varrho')=\|z^k\Psi^{s+\sigma+\frac{k-1}{2}}\varrho'u^+\|^2+\|
2502: z^{2k-1}\Psi^{s+\sigma+k-\frac{3}{2}}\varrho'u^+\|^2
2503: +\|\Psi^{s+\sigma+k-2}\varrho'\bar Lu^+\|^2.
2504: $$
2505: Repeating the same estimates with $s$ replaced by $s-\frac{1}{2}$ we
2506: replace the error $\mathcal A(s,\varrho')$ by
2507: $\mathcal A(s-\frac{1}{2},\varrho'')$. Repeating this process $2k-2$
2508: times (and redefining $\varrho'$) we obtain the desired
2509: {\it a~priori\/} estimate, namely:
2510: \begin{equation}\label{final}
2511: \|\Psi^{s+\sigma}\varrho u^+\|^2\le C(\|\varrho'
2512: E_ku\|_{s+\sigma+k-1}^2+\|\varrho'u\|^2_s+\|u\|^2_{-\infty}).
2513: \end{equation}
2514:
2515:
2516:
2517:
2518: \section{Smoothing}
2519:
2520: To conclude the proof of Theorem C we will apply the above
2521: estimate to the smoothing of a solution. Given a
2522: distribution solution $u$ of $E_ku=f$ with $f$ whose restriction to $U$
2523: is in $C^\infty(U)$, we wish to show that the restriction
2524: of $u$ to $U$ is in $C^\infty$. Without loss of generality we assume
2525: that the distribution $u$ has compact support and lies in
2526: $H^{-s_0}(\mathbb R^3)$. For $\delta>0$ we will define a smoothing
2527: operator $K_\delta$ such that $K_\delta u\in C^\infty$ and
2528: ${\rm lim}_{\delta\to 0}K_\delta(\varrho u^+)\sim\varrho u^+$.
2529:
2530:
2531: \begin{defin} Let $\omega\in\ C_0^\infty(\mathbb R)$, with
2532: $\omega(0)=1$ and let
2533: $\kappa_\delta(\xi)=\omega(\delta\xi_3)\gamma^+(\xi)$ and
2534: $$
2535: \widehat{K_\delta u}(\xi)=\kappa_\delta(\xi)\hat u(\xi),
2536: $$
2537: where $\gamma^+(\xi)=1$ in a neighborhood of the support of $\hat{u}^+$.
2538: \end{defin}
2539:
2540:
2541: \begin{lemma} If $\|K_\delta(\varrho u^+)\|_s\le C$ and if
2542: $\varrho'u^0\in H^s$ then $\varrho u^+\in H^s$.
2543: \end{lemma}
2544: Proof: We have
2545: $$
2546: \|K_\delta(\varrho u^+)-\varrho u^+\|_s\le\|K_\delta((\varrho
2547: u)^+)-(\varrho u)^+)\|_s+C\|\varrho'u^0\|_s
2548: $$
2549: and
2550: $$
2551: \lim_{\delta\to
2552: 0}(1+|\xi|^2)^{\frac{s}{2}}\omega(\delta\xi_3)\widehat{(\varrho
2553: u)^+}(\xi))=
2554: (1+|\xi|^2)^{\frac{s}{2}}\widehat{(\varrho u)^+}(\xi)).
2555: $$
2556: Then $(\varrho u)^+\in H^s$ and since $(\widehat{\varrho
2557: u)^+}-\widehat{\varrho u^+}$ is supported in the elliptic region
2558: $\mathfrak U^0$ we
2559: have
2560: $$
2561: \|\varrho u^+\|_s\le\|(\varrho u)^+\|_s+C\|\varrho'u^0\|_s ,
2562: $$
2563: thus concluding the proof.
2564:
2565: \begin{lemma} For $\delta>0${\rm ,} $K_\delta$ is a pseudodifferential
2566: operator of order $-\infty$ which is of order zero uniformly in
2567: $\delta$.
2568: $K_\delta$ has the following commutation properties.
2569: \begin{enumerate}
2570: \item[{\rm 1.}] $[E,K_\delta](I-\Gamma^0)$ is a pseudodifferential operator of
2571: order $-\infty$ uniformly in $\delta$.
2572: \item[{\rm 2.}] If $R^s$ is a pseudodifferential operator of order $s$ then
2573: $$
2574: [R^s,K_\delta]=\Gamma^0R^{s-1}_\delta+\Psi^{s-1}R^{0}_\delta+R^{-
2575: \infty}_\delta,
2576: $$
2577: where $R^{s-1}_\delta${\rm ,} $R^{0}_\delta${\rm ,} and $R^{-\infty}_\delta$ are
2578: pseudodifferential operators of orders $-\infty$ for $\delta>0$ and
2579: of orders $s-1$ and $0$ uniformly in $\delta$.
2580: \end{enumerate}
2581: \end{lemma}
2582:
2583: \Proof Number 1 follows from the fact that when
2584: $|\xi|\ge 1$ then $\gamma^0(\xi)=1$ on the support of these symbols. To
2585: deal
2586: with number 2 we write the principal symbol of $[R^s,K_\delta]$.
2587: Setting $x_1=x,\ x_2=y$ and $x_3=t$, we have
2588: $$
2589: \sum_j\frac{\partial\kappa_\delta}{\partial\xi_j}\frac{\partial
2590: r^s}{\partial x_j}=
2591: \delta\omega'(\delta\xi_3)\tilde\gamma^+\frac{\partial r^s}{\partial
2592: x_3}+
2593: \sum_j\omega(\delta\xi_3)\frac{\partial\tilde\gamma^+}{\partial\xi_j}
2594: \frac{\partial r^s}{\partial x_j}.
2595: $$
2596: The lemma then follows, since
2597: $$
2598: \delta\omega'(\delta\xi_3)\tilde\gamma^+\frac{\partial r^s}{\partial
2599: x_3}
2600: =\xi^{s-1}_3\gamma^+\left\{{\tilde\gamma}^+\xi_3^{-
2601: s}\delta\xi_3\omega'(\delta\xi_3)\frac{\partial r^s}{\partial x_3}\right\},
2602: $$
2603: where ${\tilde\gamma}^+=1$ in a neighborhood of the support of
2604: $\gamma^+$ and equals zero in a neighborhood of the origin. The
2605: expression in braces is the symbol of an operator of order zero
2606: uniformly in $\delta$.
2607:
2608: \demo{Conclusion of proof of Theorem {\rm C}} Substituting $K_\delta
2609: u$ for $u$ in (\ref{final}) we obtain
2610: $$
2611: \|\Psi^{s+\sigma}\varrho K_\delta u^+\|^2\le C(\|\varrho' E_kK_\delta
2612: u\|_{s+\sigma+k-1}^2+\|\varrho'K_\delta u\|^2_s+
2613: \|K_\delta u\|^2_{-\infty}).
2614: $$
2615: Then we have
2616: \begin{eqnarray*}
2617: \|K_\delta(\varrho u^+)\|_{s+\sigma}^2&\le &C(\|\Psi^{s+\sigma}\varrho
2618: K_\delta u^+\|^2+\|\varrho'u\|^2_{s+\sigma-1}),\\
2619: \|[\varrho' E_k,K_\delta] u\|_{s+\sigma+k-1}^2&\le &
2620: C(\|\varrho''u^0\|^2_{s+\sigma+k-1}+\|u\|^2_{-\infty}),\\
2621: \|\varrho'K_\delta u\|^2_s&\le& C\|\varrho'u\|^2_s,\\
2622: \noalign{\noindent{\rm and}}
2623: \|K_\delta u\|^2_{-\infty}&\le &C\|u\|^2_{-\infty}.
2624: \end{eqnarray*}
2625: Further
2626: $$
2627: \|\varrho''u^0\|^2_{s+\sigma+k-1}\le
2628: C(\|\varrho'''E_ku\|^2_{s+\sigma+k-3}+\|u\|^2_{-\infty}).
2629: $$
2630: Therefore, changing notation for the cutoff functions, we get
2631: $$
2632: \|K_\delta(\varrho u^+)\|_{s+\sigma}^2\le C(\|\varrho'
2633: E_ku\|_{s+\sigma+k-1}^2+\|\varrho'u\|^2_s+\|u\|^2_{-s_0}).
2634: $$
2635: Therefore, if $u\in H^{-s_0}$, if $u^+\in H^s_{\rm loc}(U)$, and if
2636: $E_ku\in H^{s+\sigma+k-1}_{\rm loc}(U)$ then $u^+\in H^{s+\sigma}_{\rm loc}(U)$.
2637: It then follows that if $u\in H^{-s_0}$ and if $E_ku\in
2638: H^{s_1}_{\rm loc}(U)$ then $u^+\in H^{s_1-k+1}_{\rm loc}(U)$. Since, under
2639: the same assumptions, we have $u^0\in H^{s_1+2}_{\rm loc}(U)$ and $u^-\in
2640: H^{s_1+1}_{\rm loc}(U)$ we conclude that
2641: $u\in H^{s_1-k+1}_{\rm loc}(U)$, thus proving Theorem~C.
2642:
2643:
2644: \section{Local existence in $L^2$}
2645:
2646:
2647: The {\it a priori\/} estimates for $E_k$ imply the following local existence result.
2648:
2649: \sdemo{Theorem} {\it If $P\in U\subset\mathbb R^3$ with $U$ an open set{\rm ,} then
2650: there exists a neighborhood $U_1\subset\bar U_1\subset U${\rm ,} with $P\in U_1${\rm ,}
2651: such that if $f\in H^{k-1}_{\rm loc}(U)$
2652: then there exists $u\in L^2(U_1)$ and $E_ku=f$ in $U_1$.}
2653:
2654: \Proof Let $U_1$ be a small neighborhood of $P$. In Lemma 11 set
2655: $\varrho=1$ in a neighborhood
2656: of $\bar U_1$ and set $u=v\in C_0^\infty(U_1)$ so that $\varrho u=v$ and
2657: $[\Psi^{s+\sigma},\Gamma^+]$ is an
2658: operator of order $-\infty$ on $C_0^\infty(U_1)$. Hence we obtain
2659: $$
2660: \|\Psi^{s+\sigma}v^+\|^2\le C(\|E_kv\|^2_{s+\sigma+k}+\|v\|^2_s),
2661: $$
2662: for all $v\in C_0^\infty(U_1)$. Setting $s+\sigma+k=0$ and combinig with the estimates
2663: for $v^0$ and $v^-$, we
2664: obtain
2665: $$
2666: \|v\|^2_{-k+1}\le C(\|E_kv\|^2+\|v\|^2_{-k+1-\sigma}).
2667: $$
2668: Then, if the diameter of $U_1$ is sufficiently small, we have
2669: $$\|v\|^2_{-k+1-\sigma}\le \hbox{small const. } \|v\|^2_{-k+1}.
2670: $$
2671: Hence
2672: $$
2673: \|v\|_{-k+1}\le \hbox{const. }\|E_kv\|,
2674: $$
2675: for all $v\in C_0^\infty(U_1)$.
2676:
2677: Let $\mathcal W=C_0^\infty(U_1)$ and let $K:\mathcal W\to\mathbb C$
2678: be the linear functional defined by
2679: $Kw=(v,f)$ with $w=E_kv$. Then
2680: $$
2681: |Kw|=|(v,f)|\le\|v\|_{-k+1}\|f\|_{k-1}\le C\|w\|.
2682: $$
2683: So $K$ is bounded on $\mathcal W$; hence it can be extended to a bounded linear
2684: functional on $L^2(U_1)$. Therefore there
2685: exists $u\in L^2(U_1)$ such that $Kw=(w,u)$, that is $(v,f)=(E_kv,u)=(v,E_ku)$.
2686: Thus $E_ku=f$ in $L^2(U_1)$, which completes the proof.
2687: \vglue-12pt
2688:
2689: {\references {MMNM}
2690:
2691: \bibitem[BM]{BM} \name{D.\ Bell} and \name{S.\ Mohammed}, An extension of H\"ormander's theorem
2692: for infinitely degenerate second-order operators,
2693: {\it Duke Math.\ J.} {\bf 78} (1995), 453--475.
2694:
2695: \bibitem[BDKT]{BDKT} \name{A. Bove, M. Derridj, J. J. Kohn}, and
2696: \name{D. S. Tartakoff},
2697: Hypoellipticity for a sum of squares of complex vector fields with large loss of derivatives,
2698: preprint.
2699:
2700: \bibitem[C]{C} \name{D.\ Catlin},
2701: Necessary conditions for the
2702: subellipticity of the $\bar\partial$-Neumann problem,
2703: {\it Ann.\ of Math.\/} {\bf 117} (1983), 147--171; Subelliptic estimates
2704: for
2705: the $\bar\partial$-Neumann problem on pseudoconvex domains,
2706: {\it Ann.\ of Math\/}.\ {\bf 126} (1987), 131--191.
2707:
2708:
2709: \bibitem[Ch1]{Ch1} \name{M.\ Christ}, Hypoellipticity in the infinitely degenerate regime,
2710: in {\it Complex Analysis and Geometry\/} (Columbus, OH, 1999), 59--84,
2711: {\it Ohio State Univ.\ Math.\ Res.\ Inst.\ Publ\/}.\ {\bf 9}, de
2712: Gruyter, Berlin, 2001.
2713:
2714:
2715: \bibitem[Ch2]{Ch2} \bibline, A counterexample for sums of squares of complex
2716: vector fields, preprint, 2004.
2717:
2718:
2719: \bibitem[ChK]{ChK} \name{M.\ Christ} and \name{G.\ E.\ Karadjov}, Local solvability for a class of
2720: partial differential operators with double
2721: characteristics, preprint.
2722:
2723:
2724: \bibitem[D'A]{D'A} \name{J.\ P.\ D'Angelo}, Real hypersurfaces, orders of contact, and
2725: applications, {\it Ann.\ of Math.\/} {\bf 115} (1982), 615--637.
2726:
2727:
2728: \bibitem[DT]{DT} \name{M. Derridj} and \name{D. Tartakoff}, Local analytic hypoellipticity for a sum of
2729: squares of coplex vector fields with large loss of derivatives, {\it preprint}.
2730:
2731:
2732: \bibitem[F]{F} \name{V.\ S.\ Fedii}, A certain criterion for hypoellipticity, {\it Mat.\ Sb.\/}
2733: {\bf 14} (1971), 15--45.
2734:
2735:
2736: \bibitem[FP]{FP} \name{C.\ Fefferman} and \name{D.\ H.\ Phong}, The uncertainty principle and sharp
2737: G\aa rding inequalities, {\it Comm.\ Pure Appl.\
2738: Math.} {\bf 34} (1981), 285--331.
2739:
2740:
2741: \bibitem[He]{He} \name{P.\ Heller}, Analyticity and regularity for nonhomogeneous
2742: operators on the Heisenberg group, Princeton University
2743: dissertation, 1986.
2744:
2745:
2746: \bibitem[Ho]{Ho} \name{L.\ H\"ormander}, Hypoelliptic second order differential
2747: equations, {\it Acta Math.\/} {\bf 119} (1967), 147--171.
2748:
2749:
2750: \bibitem[K1]{K1} \name{J.\ J.\ Kohn}, Subellipticity on pseudo-convex domains with isolated
2751: degeneracies, {\it Proc.\
2752: Natl.\ Acad.\ Sci.\ U.S.A\/}.\ {\bf 71} (1974), 2912--2914.
2753:
2754:
2755: \bibitem[K2]{K2} \bibline, Subellipticity of the $\bar\partial$-Neumann problem
2756: on pseudo-convex domains: sufficient
2757: conditions, {\it Acta Math\/}.\ {\bf 142} (1979), 79--122.
2758:
2759:
2760: \bibitem[K3]{K3} \bibline, Pseudo-differential operators and non-elliptic
2761: problems (1969 {\it Pseudo-Diff.\ Operators\/} (C.I.M.E., Stresa,
2762: 1968), 157--165,
2763: Edizioni Cremonese, Rome (1969).
2764:
2765:
2766: \bibitem[K4]{K4} \name{J.\ J.\ Kohn}, Hypoellipticity of some degenerate subelliptic
2767: operators, {\it J.\ Funct.\ Anal.\/} {\bf 159} (1998), 203--216.
2768:
2769:
2770: \bibitem[K5]{K5} \bibline, Superlogarithmic estimates on pseudoconvex domains
2771: and CR manifolds, {\it Ann.\ of Math.\/} {\bf 156} (2002), 213--248.
2772:
2773:
2774: \bibitem[KN]{KN} \name{J.\ J.\ Kohn} and \name{L.\ Nirenberg}, Non-coercive boundary value
2775: problems, {\it Comm.\ Pure Appl.\ Math.\/} {\bf 18} (1965), 443--492.
2776:
2777:
2778: \bibitem[KS]{KS} \name{S.\ Kusuoka} and \name{D.\ Stroock}, Applications of the Mallavain
2779: calculus. II, {\it J.\ Fac.\ Sci.\ Univ.\ Tokyo Sec.\ IA Math.}
2780: {\bf 32} (1985), 1--76.
2781:
2782:
2783: \bibitem[M]{M} \name{Y.\ Morimoto}, Hypoellipticity for infinitely degenerate elliptic
2784: operators, {\it Osaka J. Math.\/} {\bf 24} (1987), 13--35.
2785:
2786:
2787: \bibitem[N]{N}Ê \name{A.\ M.\ Nadel}, Multiplier ideal sheaves and K\"ahler-Einstein
2788: metrics of positive scalar curvature, {\it Ann.\ of
2789: Math.\/} {\bf 132} (1990), 549--596.
2790:
2791:
2792: \bibitem[OR]{OR} \name{O.\ A.\ Oleinik} and \name{E.\ V.\ Radkevic}, {\it Second Order Equations with
2793: Nonnegative Characteristic Form\/}, Plenum Press, New York, 1973.
2794:
2795:
2796: \bibitem[PP1]{PP1} \name{C. Parenti} and \name{A. Parmeggiani}, On the hypoellipticity with a big loss of
2797: derivatives, {\it Kyushu J. Math.\/} {\bf 59} (2005), 155-230.
2798:
2799:
2800: \bibitem[PP2]{PP2} \bibline, A note on Kohn's and Christ's examples,
2801: {preprint}.
2802:
2803:
2804: \bibitem[S]{S} \name{Y.-T.\ Siu}, Extension of twisted pluricanonical sections with
2805: plurisubharmonic weight and invariance of semipositively
2806: twisted plurigenera for manifolds not necessarily of general type, in {\it
2807: Complex Geometry\/}: {\it Collection
2808: of Papers Dedicated to Professor Hans Grauert} (G\"ottingen, 2000), 223--277, Springer-Verlag,
2809: New York, 2002.
2810:
2811:
2812: \bibitem[St]{St} \name{E.\ M.\ Stein}, An example on the Heisenberg group related to the
2813: Lewy operator, {\it Invent.\ Math.\/} {\bf 69} (1982), 209--216.
2814:
2815:
2816: \end{thebibliography}
2817: }
2818:
2819: \centerline{\small (Received August 3, 2003)}
2820:
2821:
2822: %%%%%%%Theorem info here
2823: %\newtheorem{theorem}{Theorem}[section]
2824: %\newtheorem{proclaim}[theorem]{Proclaim}
2825:
2826: %%%%%%%%%%%%%% equation numbering style
2827:
2828:
2829: %%
2830: %%%%%%%%%%%%%%%%%%%%
2831: %
2832:
2833: \def\A {{\mathcal{A}}}
2834: \def\D {{\mathcal{D}}}
2835: \def\R {{\mathbb{R}}}
2836: \def\N {{\mathbb{N}}}
2837: \def\C {{\mathbb{C}}}
2838: \def\Z {{\mathbb{Z}}}
2839: \def\l {\ell}
2840: \def\ml {multline}
2841: \def\multiline {\multline}
2842: \def\lessim {\lesssim}
2843:
2844: \def\phi{\varphi}
2845: \def\epsilon{\varepsilon}
2846: %\renewcommand{\theequation}{\thesection.\arabic{equation}}
2847:
2848: \vglue18pt
2849: \begin{center}
2850: {\bf \Large Appendix:}\\
2851: {\bf \Large Analyticity and loss of
2852: derivatives}
2853: \end{center}
2854: \vglue5pt \centerline{\small By {\scshape Makhlouf Derridj} and {\scshape David S.\
2855: Tartakoff}}
2856: \renewcommand{\institution}[1]
2857: {\renewcommand{\theinstitutions}{\vskip16pt\baselineskip10pt\begin{quote}
2858: \scriptsize\scshape #1\end{quote}}}
2859:
2860: \institution{5 rue de la Juviniere, 78350 Les Loges en
2861: Josas, France\\
2862: \email{derridj@club-internet.fr}
2863: \\
2864: \vglue-9pt
2865: University
2866: of Illinois at Chicago, Chicago IL\\
2867: \email{dst@uic.edu\\}
2868: \vglue-30pt \phantom{oh}}
2869:
2870: \shortname{Makhlouf Derridj and David S.\ Tartakoff}
2871:
2872: \shorttitle{Appendix}
2873: \setcounter{section}{0}
2874: \vglue18pt
2875: \centerline{\bf Abstract}
2876: \vglue12pt
2877: In \cite{K2005}, J.\ J.\ Kohn
2878: proves $C^\infty$ hypoellipticity for a sum of squares
2879: of complex vector fields which exhibit a large loss of
2880: derivatives. Here, we prove analytic hypoellipticity for
2881: this operator.
2882:
2883: \section{Introduction and outline}
2884: \renewcommand{\theequation}{\thesection.\arabic{equation}}
2885: \setcounter{equation}{0}
2886:
2887: In \cite{K2005}, J.\ J.\ Kohn proves hypoellipticity
2888: for the operator
2889: $$P = LL^*+ (\overline{z}^kL)^*(\overline{z}^kL),
2890: \qquad L = {\partial \over \partial z} + i\overline{z}
2891: {\partial \over \partial t},$$
2892: for which there is a large loss of
2893: derivatives --- indeed in the {\it a priori} estimate
2894: one bounds only the Sobolev norm of order $-(k-1)/2$,
2895: and thus there is a loss of $k-1$ derivatives: $Pu\in
2896: H^s_{\rm loc}\implies u\in H^{s-(k-1)}_{\rm loc}$.
2897:
2898: We show in this note that solutions of
2899: $Pu=f$ with $f$ real analytic are themselves real
2900: analytic in any open set where $f$ is. In so doing we
2901: use an {\it a priori} estimate which follows easily
2902: from that established by Kohn for this operator,
2903: namely for test functions $v$ of small support near
2904: the origin:
2905: \begin{equation}\label{apeDT}\|\overline{L}v\|_0^2 +
2906: \|\overline{z}^k{L}v\|_0^2 + \|v\|^2_{-{k-1\over 2}}
2907: \lesssim |(Pv, v)_{L^2}|.\end{equation}
2908: In fact, in \cite{T3} (see also \cite{BDKT}),
2909: we give a rapid and direct derivation
2910: of (\ref{apeDT}) for this operator and similar
2911: estimates for more degenerate operators.
2912:
2913: The first two
2914: terms on the left of this estimate exhibit maximal
2915: control in
2916: $\overline{L}$ and $\overline{z}^kL$, but only these
2917: complex directions. Hence in obtaining recursive
2918: bounds for derivatives it is essential to keep one of
2919: these vector fields available for as long as possible.
2920: For this, we will construct a carefully balanced
2921: localization of high powers of $T= -2i
2922: \partial/\partial t$ and use the estimate repeatedly,
2923: reducing the order of powers of $T$ but accumulating
2924: derivatives on the localizing functions. These
2925: Ehrenpreis type localizing functions work `as
2926: if analytic' up to a prescribed order, with
2927: all constants independent of that order, as in
2928: \cite{DT1978}, \cite{DT1980}, but eventually the good
2929: derivatives ($\overline{L}$ or
2930: $\overline{z}^kL$) are lost and we must use the third
2931: term on the left of the estimate, absorb the
2932: loss of $\frac{k-1}{2}$ derivatives, introduce a new
2933: localizing function of larger support and start the
2934: whole process again, but with only a (fixed) fraction
2935: of the original power of $T$.
2936:
2937: \section{Observations and simplifications}
2938:
2939:
2940: Our first observation is that we know the analyticity
2941: of the solution for $z$ different from $0$ from the
2942: earlier work of the second author \cite{DT1978},
2943: \cite{DT1980} and Tr\`eves \cite{Tr1978}. Thus, modulo
2944: brackets with localizing functions whose derivatives
2945: are supported in the known analytic hypoelliptic
2946: region, we take all localizing functions independent of
2947: $z$.
2948:
2949: Our second observation is that it suffices to bound
2950: derivatives measured in terms of high powers of the
2951: vector fields $L$ and $\overline{L}$ in $L^2$ norm, by
2952: standard arguments, and indeed estimating high powers
2953: of ${L}$ can be reduced to bounding high
2954: powers of
2955: $\overline L$ and powers of $T$ of half the order, by
2956: repeated integration by parts. Thus our overall scheme
2957: will be to start with high powers (order $2p$) of $L$ or
2958: $\overline{L},$ use integration by parts and
2959: the {\it a priori} estimate repeatedly to reduce to
2960: treating $T^pu$ in a slightly larger set.
2961:
2962: And to do this, we introduce a new special
2963: localization of
2964: $T^p$ adapted to this problem.
2965:
2966: \section{The localization of high powers of $T$}
2967:
2968:
2969:
2970: The new localization of $T^p$ may be written in the
2971: form:
2972: $$(T^{p_1,p_2})_\phi = \sum_{a\leq p_1 \atop b\leq
2973: p_2}{L^a\circ
2974: z^a\circ
2975: T^{p_1-a}\circ
2976: \phi^{(a+b)}\circ T^{p_2-b}\circ
2977: \overline{z}^b\circ \overline{L}^b
2978: \over a!b!}.$$
2979: %
2980: Here by $\phi^{(r)}$ we mean $(-i\partial/\partial
2981: t)^r\phi(t)$ since near $z=0$ we have seen that we may
2982: take the localizing function independent of $z.$ Note
2983: that the leading term (with $a+b=0)$ is merely
2984: $T^{p_1}\phi T^{p_2}$ which equals $T^{p_1+p_2}$
2985: on the initial open set $\Omega_0$ where $\phi \equiv
2986: 1$.
2987:
2988: We have the commutation relations:
2989: $$[L, (T^{p_1,p_2})_\phi] \equiv L\circ
2990: (T^{p_1-1,p_2})_{\phi'},$$
2991: $$ [\overline{L},
2992: (T^{p_1,p_2})_\phi] \equiv
2993: (T^{p_1,p_2-1})_{\phi'}\circ \overline{L},
2994: $$
2995: $$[(T^{p_1,p_2})_\phi,z] = (T^{p_1-1,p_2})_{\phi'}\circ
2996: z,
2997: $$
2998: and
2999: $$
3000: [(T^{p_1,p_2})_\phi,\overline{z}] =
3001: \overline{z}\circ (T^{p_1,p_2-1})_{\phi'},
3002: $$
3003: %
3004: \vskip.1in\noindent
3005: where the $\equiv$ denotes modulo
3006: $C^{p_1-p_1'+p_2-p_2'}$ terms of the form
3007: \begin{equation}\label{pureL}{L^{p_1-p_1'}\circ
3008: z^{p_1-p_1'}\circ
3009: T^{p_1'}\circ\phi^{(p_1-p_1'+p_2-p_2'+1)}\circ T^{p_2'}
3010: \circ\overline{z}^{p_2-p_2'}\circ
3011: \overline{L}^{p_2-p_2'}\over (p_1-p_1')!(p_2-p_2')!}
3012: \end{equation}
3013: with either $p_1'=0$ or
3014: $p_2'=0$, i.e., terms where all free $T$
3015: derivatives have been eliminated on one side of
3016: $\phi$ or the other. Thus if we start with
3017: $p_1=p_2={p/ 2}$, and iteratively apply these
3018: commutation relations, the number of $T$ derivatives not
3019: necessarily applied to
3020: $\phi$ is eventually at most ${p/ 2}$.
3021:
3022: \section{The recursion}
3023:
3024:
3025: We insert first $v=(T^{{p\over 2},{p\over 2}})_\phi u$
3026: in the {\it a priori} inequality, then bring
3027: $(T^{{p\over 2},{p\over 2}})_\phi$ to the left of
3028: $P=-L\overline{L}-
3029: \overline{L}z^k\overline{z}^kL$ since $Pu$ is known
3030: and analytic. We have, omitting for now the
3031: `subelliptic' term,
3032: $$\|\overline{L}(T^{{p\over 2},{p\over 2}})_\phi
3033: u\|_0^2 +
3034: \|\overline{z}^k{L}(T^{{p\over 2},{p\over 2}})_\phi
3035: u\|_0^2 \lesssim |(P(T^{{p\over 2},{p\over 2}})_\phi u,
3036: (T^{{p\over 2},{p\over 2}})_\phi u)_{L^2}|$$
3037: %
3038: $$\lesssim
3039: |((T^{{p\over 2},{p\over 2}})_\phi Pu, (T^{{p\over 2},{p\over 2}})_\phi
3040: u)_{L^2}| + |([P,(T^{{p\over 2},{p\over 2}})_\phi] u,
3041: (T^{{p\over 2},{p\over 2}})_\phi u)_{L^2}|
3042: $$
3043: and, by the above bracket relations,
3044: \begin{eqnarray*}
3045: &&([P,(T^{{p\over 2},{p\over 2}})_\phi] u,
3046: (T^{{p\over 2},{p\over 2}})_\phi u)\\
3047: &&\qquad = -([L\overline{L},(T^{{p\over 2},{p\over 2}})_\phi]
3048: u, (T^{{p\over 2},{p\over 2}})_\phi u) - ([
3049: \overline L z^k\overline{z}^k{L},(T^{{p\over 2},{p\over
3050: 2}})_\phi] u, (T^{{p\over 2},{p\over 2}})_\phi u)\\
3051: &&\qquad \equiv -(L(T^{{{p\over 2},{p\over 2}}-1})_{\phi'}
3052: \overline{L}u, (T^{{p\over 2},{p\over 2}})_\phi u)
3053: -(L (T^{{p\over 2}-1,{p\over 2}})_{\phi'}
3054: \overline{L}u, (T^{{p\over 2},{p\over 2}})_\phi u)\\
3055: &&\qquad \quad-
3056: ((T^{{p\over 2}-1,{p\over
3057: 2}})_{\phi'}\overline{L} z^k\overline{z}^k {L}u,
3058: (T^{{p\over 2},{p\over 2}})_\phi u)
3059: \\
3060: &&\qquad \quad-\sum_{k'=1}^k
3061: (\overline L{z}^{k'}(T^{{p\over 2},{p\over
3062: 2}-1})_{\phi'}
3063: {z}^{k-k'}\overline z^k{L}u,
3064: (T^{{p\over 2},{p\over 2}})_\phi u) \\
3065: &&\qquad \quad-\sum_{k'=0}^{k-1}(\overline L{z}^k
3066: \overline z^{k'} (T^{{p\over 2}-1,{p\over 2}})_{\phi'}
3067: \overline z^{k-k'}{L}u,
3068: (T^{{p\over 2},{p\over 2}})_\phi u)\\
3069: &&\qquad \quad- (\overline L{z}^k
3070: \overline z^k{L}(T^{{p\over 2},{p\over
3071: 2}-1})_{\phi'}u, (T^{{p\over 2},{p\over
3072: 2}})_\phi u),
3073: \end{eqnarray*}
3074: with the same meaning for $\equiv$ as above. In every
3075: term, no powers of $z$ or $\overline{z}$ have been
3076: lost, though some may need to be brought to the left
3077: of the $(T^{q_1,q_2})_{\tilde{\phi}}$
3078: with again no loss
3079: of powers of $z$ or $\overline{z}$ and a further
3080: reduction in order, every bracket reduces the order of
3081: the sum of the two indices $p_1$ and
3082: $p_2$ by one (here we started with $p_1=p_2=p/2$),
3083: picks up one derivative on $\phi,$ and leave the vector
3084: fields over which we have maximal control in the
3085: estimate intact and in the correct order. Thus we may
3086: bring either
3087: $\overline{L}z^k$ or $L$ to the right as
3088: $\overline{z}^kL$ or
3089: $\overline{L},$ and use a weighted Schwarz inequality on
3090: the result to take maximal advantage of the {\it a
3091: priori} inequality. Iterations of all of this
3092: continue until there remain at most $p/2$
3093: free $T$ derivatives (i.e., the $T$ derivatives on
3094: at least one side of $\phi$ are all `corrected' by
3095: good vector fields) and perhaps as many as
3096: ${p/2} \;L$ or $\overline{L}$ derivatives, and we
3097: may continue further until, at worst, these remaining
3098: $L$ or $\overline{L}$ derivatives bracket two at a
3099: time to produce more $T$'s, with corresponding
3100: combinatorial factors. After all of this, there will be
3101: at most
3102: $T^{3p\over 4}$ remaining, and a factor of
3103: ${p\over 2}!!\sim {p\over 4}!$
3104:
3105: It is here that the final term on the left of the
3106: {\it a priori} inequality is used, in order to bring
3107: the localizing function out of the norm after
3108: creating another balanced localization of $T^{3p/4}$
3109: with a new localizing
3110: function of Ehrenpreis type with slightly larger
3111: support, geared, roughly, to
3112: $3p/4$ instead of to
3113: $p$.
3114:
3115: Recall that such such localizing functions
3116: $\psi$ may be constructed for any $N$ and satisfy
3117: $$\left|\psi^{(r)}\right| \leq \left({C\over
3118: e}\right)^{r+1}N^r,
3119: \quad r\leq 2N$$
3120: where $C$ is independent of $N$ and
3121: $e={\rm dist}(\{\psi\equiv 1\}, ({\rm supp}\,\psi)^c)$.
3122:
3123:
3124: \section{Conclusion of the proof}
3125:
3126:
3127: Finally, this entire process, which reduced the
3128: order from $p$ to at most $3p/4,$ (or more precisely to at most $3p/4 +
3129: (k-1)/2$), is repeated, over and over, each time
3130: essentially reducing the order by a factor of $3/4.$
3131: After at most $\log_{4/3} p$ such iterations
3132: we are reduced to a bounded number of derivatives,
3133: and, as in \cite{DT1978} and \cite{DT1980}, all of these
3134: nested open sets may be chosen to fit in the one open
3135: set $\Omega'$ where $Pu$ is known to be analytic, and
3136: all constants chosen independent of $p$ (but
3137: depending on $Pu$). The fact that in those
3138: works one full iteration reduced the order by half
3139: played no essential role --- a factor of
3140: $3/4$ works just as well.
3141:
3142: To be precise, the sequence of open sets,
3143: $\{\Omega_j\}$, each compactly contained in the next,
3144: with
3145: $\Omega_{\log_{4/3}p}=\Omega'$, have separations
3146: $d_j={\rm dist}(\Omega_j, \Omega_{j+1}^c),$ with $\sum
3147: d_j={\rm dist}(\Omega_0,{\Omega'}^c)=d$, which need to be
3148: picked carefully. The localizing functions
3149: $\{\phi_j\}$ with $\phi_j\in C_0^\infty(\Omega_{j+1})
3150: \equiv 1$ on
3151: $\Omega_j$
3152: satisfy
3153: \begin{equation}\label{j}\left|\phi_j^{(r)}\right|\leq
3154: (C/d_j)^{r+1}((3/4)^jp)^r, \qquad r\leq 2(3/4)^jp.
3155: \end{equation}
3156: We shall take the $d_j={1\over {(j+1)^2}}/d\sum {1\over
3157: {(j+1)^2}}$.
3158: \smallbreak
3159: Now at most $(3/4)^jp$ derivatives will fall on
3160: $\phi_j$, and most of the effect of the derivatives
3161: will be balanced by corresponding factorials in the
3162: denominator, as in (\ref{pureL}), roughly the powers of
3163: $(3/4)^jp$ in (\ref{j}) in view of Stirling's formula.
3164: In addition, as noted immediately before the last
3165: paragraph in Section 4, there will be factorials
3166: corresponding to the diminution of powers of $T$. What
3167: will {\it not} be balanced are the powers of
3168: $d_j^{-1}$, but the product of these factors will
3169: contribute
3170: $$\Pi_{j=1}^{\log_{4/3}p}
3171: \left(j^2\right)^{(3/4)^jp}=
3172: \left(\Pi_{j=1}^{\log_{4/3}p}
3173: j^{(3/4)^j}\right)^{2p}= C^p,$$
3174: %
3175: which, together with the factorials just mentioned,
3176: proves the analyticity of the solution in
3177: $\Omega_0$.
3178:
3179: \vglue-6pt
3180: \references {100}
3181:
3182: \vglue-24pt
3183: \phantom{hi}
3184: \bibitem[1]{BDKT} \name{A. Bove, M. Derridj, J. J. Kohn}, and \name{D. S. Tartakoff},
3185: Hypoellipticity for a sum of squares of complex vector fields with
3186: large loss of derivatives, preprint.
3187:
3188:
3189: \bibitem[2]{K2005} {\sc J.\ J.\ Kohn}, Hypoellipticity and loss
3190: of derivatives, {\it Ann. of Math\/}.\ {\bf 162} (2005), 943--982.
3191:
3192:
3193: \bibitem[3]{DT1978} {\sc D.\ S.\ Tartakoff},
3194: Local analytic
3195: hypoellipticity for $\Box_{b}$ on nondegenerate
3196: Cauchy Riemann manifolds, {\it Proc.\ Nat.\ Acad.\ Sci.\ U.S.A\/}.\
3197: {\bf 75} (1978),
3198: 3027--3028.
3199: %
3200: \bibitem[4]{DT1980} \bibline,
3201: The local real analyticity of solutions to $\Box_{b}$ and
3202: the $\bar\partial$-Neumann problem, {\it Acta
3203: Math\/}.\ {\bf 145} (1980), 117--204.
3204: %
3205: \bibitem[5]{T3} \bibline, Analyticity for singular sums of squares of
3206: degenerate vector fields, {\it Proc.\ Amer.\ Math.\ Soc}., to appear.
3207:
3208: \bibitem[6]{Tr1978} {\sc F.\ Tr\`eves},
3209: Analytic hypo-ellipticity of a class of
3210: pseudodifferential operators with
3211: double characteristics and applications to the
3212: $\bar\partial$-Neumann problem,{\it \ Comm.\ Partial Differential Equations}
3213: {\bf 3} (1978), 475--642.
3214: %
3215:
3216: \Endrefs
3217:
3218:
3219:
3220: \end{document}