1: \documentclass[12pt, dvips]{amsart}
2:
3: \swapnumbers\newtheorem{theorem}{Theorem}[section]
4:
5: \newtheorem{theore}{Theorem}[section]
6: \renewcommand{\thetheore}{} %{}optional, z.B. {A} gibt ein A
7:
8: \newtheorem{corollary}[theorem]{Corollary}
9:
10: \newtheorem{proposition}[theorem]{Proposition}
11: \newtheorem{propositio}{Proposition}
12: \renewcommand{\thepropositio}{}
13:
14: \newtheorem{lemma}[theorem]{Lemma}
15: \newtheorem{lemm}{Lemma}
16: \renewcommand{\thelemm}{}
17:
18: \newtheorem{conjecture}[theorem]{Conjecture}
19: \newtheorem{conjectur}{Conjecture}
20: \renewcommand{\theconjectur}{}
21:
22: \newtheorem{definition}[theorem]{Definition}
23: \newtheorem{definitio}{Definition}
24: \renewcommand{\thedefinitio}{}
25:
26: \newtheorem{remark}[theorem]{Remark}
27: \newtheorem{remarks}[theorem]{Remarks}
28: \newtheorem{remar}{Remark}
29: \renewcommand{\theremar}{}
30:
31: \newtheorem{example}[theorem]{Example}
32: \newtheorem{exampl}{Example}
33: \renewcommand{\theexampl}{}
34:
35: \newtheorem{notation}[theorem]{Notation}
36: \newtheorem{claim}[theorem]{Claim}
37: \newtheorem{recollection}[theorem]{Recollection}
38:
39: \numberwithin{equation}{section}
40:
41: \usepackage{amssymb,latexsym, amsmath, amscd, bbm, graphicx, epsfig, psfrag}
42:
43:
44: \def\s{\smallskip}
45: \def\m{\medskip}
46: \def\ni{\noindent}
47: \def\ra{\rightarrow}
48: \def\ha{\hookrightarrow}
49:
50: \def\RR{\mathbbm{R}}
51: \def\NN{\mathbbm{N}}
52: \def\CC{\mathbbm{C}}
53: \def\ZZ{\mathbbm{Z}}
54:
55: \def\aa{\alpha}
56: \def\bb{\beta}
57: \def\gg{\gamma}
58: \def\dd{\delta}
59: \def\ee{\epsilon}
60: \def\eee{\varepsilon}
61: \def\ff{\varphi}
62: \def\ll{\lambda}
63: \def\oo{\omega}
64: \def\ss{\sigma}
65: \def\gs{\varsigma}
66: \def\gt{\vartheta}
67: \def\yy{\upsilon}
68: \def\vv{\varrho}
69: \def\zz{\zeta}
70:
71: \def\gD{\Delta}
72: \def\gF{\Phi}
73: \def\gG{\Gamma}
74: \def\gL{\Lambda}
75: \def\gO{\Omega}
76: \def\gS{\Sigma}
77:
78:
79: \def\ca{{\mathcal A}}
80: \def\cb{{\mathcal B}}
81: \def\cc{{\mathcal C}}
82: \def\cd{{\mathcal D}}
83: \def\ce{{\mathcal E}}
84: \def\cf{{\mathcal F}}
85: \def\ch{{\mathcal H}}
86: \def\ci{{\mathcal I}}
87: \def\cj{{\mathcal J}}
88: \def\ck{{\mathcal K}}
89: \def\cl{{\mathcal L}}
90: \def\cn{{\mathcal N}}
91: \def\co{{\mathcal O}}
92: \def\cp{{\mathcal P}}
93: \def\cs{{\mathcal S}}
94: \def\cv{{\mathcal V}}
95: \def\cz{{\mathcal Z}}
96:
97: \def\Vol{\operatorname{Vol}\;\!}
98: \def\supp{\operatorname{supp}\:\!}
99: \def\length{\operatorname{length}\:\!}
100:
101:
102: \newcommand{\proofend}{\hspace*{\fill} $\Box$\\}
103: \newcommand{\diam}{\hspace*{\fill} $\Diamond$}
104:
105:
106: \begin{document}
107:
108: \title[]{An extension theorem in symplectic geometry}
109:
110: \date{\today}
111: \thanks{2000 {\it Mathematics Subject Classification.}
112: Primary 53D35, Secondary 54C20.
113: }
114:
115: \author{Felix Schlenk}
116: \address{(F.\ Schlenk) ETH Z\"urich, CH-8092 Z\"urich, Switzerland}
117: \email{felix@math.ethz.ch}
118:
119:
120: \begin{abstract}
121: We extend the ``Extension after Restriction Principle'' for symplectic
122: embeddings of bounded starlike domains to a large class of symplectic
123: embeddings of unbounded starlike domains.
124: \end{abstract}
125:
126:
127: \maketitle
128:
129: \markboth{{\rm An extension theorem in symplectic geometry}}{{}}
130:
131: \section{Introduction}
132: \ni
133: We endow each open subset $U$ of Euclidean space $\RR^{2n}$
134: with the standard symplectic form
135: \[
136: \oo_0 \,=\, \sum_{i=1}^n dx_i \wedge dy_i .
137: \]
138: A smooth embedding $\ff \colon U \ha \RR^{2n}$ is called symplectic if
139: $\ff^* \oo_0 \,=\, \oo_0$.
140: In particular, every symplectic embedding preserves the volume form
141: $\tfrac{1}{n!} \omega_0^n$ and hence the Lebesgue measure on $\RR^{2n}$.
142: Recall that a domain in $\RR^{2n}$ is by definition a non-empty open
143: connected subset of $\RR^{2n}$.
144:
145: \begin{definition}
146: {\rm
147: Consider a symplectic embedding $\ff \colon U \ha \RR^{2n}$
148: of a domain $U$ in $\RR^{2n}$.
149: We say that the pair $(U, \ff)$ has the {\it extension property}\,
150: if for each subset $A \subset U$ whose closure in $\RR^{2n}$ is
151: contained in $U$ there exists a symplectomorphism $\Phi_A$ of $\RR^{2n}$
152: such that $\Phi_A |_A = \ff |_A$.
153: }
154: \end{definition}
155:
156:
157: Not every pair $(U, \ff)$ as above has the extension property as the following
158: example shows.
159:
160: \begin{example} \label{exa:0}
161: {\rm
162: For $0 \le r_0<r_1<\infty$ we define the open annulus
163: \[
164: A \left( r_0,r_1 \right) \,=\,
165: \left\{ (x,y) \in \RR^2 \mid r_0^2 < x^2+y^2 < r_1^2 \right\} .
166: \]
167: Let $\ff \colon A(0,3) \ra A(4,5) \subset \RR^2$ be the symplectic embedding which
168: in polar coordinates is given by
169: \[
170: \ff (r, \gt) \,=\, \left( \sqrt{r^2+16}, \gt \right) .
171: \]
172: Any smooth extension of $\ff |_{A(1,2)}$ to $\RR^2$ maps the disc of
173: area $\pi$ to the disc of area $17\pi$ and hence cannot be symplectic.
174: \diam
175: }
176: \end{example}
177:
178: On the other hand, the well-known
179: ``Extension after Restriction Principle'' \cite{EH1},
180: which is reproved below, states that a pair $(U, \ff)$ has the
181: extension property whenever the geometry of $U$ is simple enough.
182: %
183: Recall that a subset $U$ of $\RR^{2n}$ is said to be {\it starlike}\;\! if
184: $U$ contains a point $p$ such that for every point $x \in U$ the
185: straight line between $p$ and $x$ is contained in $U$.
186:
187: \begin{proposition}\ \label{pa2:ear}
188: {\rm (Extension after Restriction Principle)}
189: Assume that $\ff \colon U \ha \RR^{2n}$ is a symplectic embedding of a
190: bounded starlike domain $U \subset \RR^{2n}$.
191: Then the pair $(U,\ff)$ has the extension property. In fact, for any subset $A \subset U$ whose closure in $\RR^{2n}$ is
192: contained in $U$ there exists a compactly
193: supported symplectomorphism $\Phi_A$ of $\RR^{2n}$ such that
194: $\Phi_A |_A = \ff |_A$.
195: \end{proposition}
196:
197: The purpose of this paper is to prove the extension property for a
198: large class of symplectic embeddings of unbounded starlike domains.
199: %
200: The following example shows that it is not enough to assume that $U$
201: is starlike.
202:
203: \begin{example} \label{exa:1}
204: {\rm
205: We let $U \subset \RR^2$ be the strip
206: $]1, \infty[ \,\times\, ]\:\!\!-1,1[$.
207: Combining the methods used in Step 1 and Step 4 of \text{Section 2.2}
208: in \cite{Diss}
209: we find a symplectic embedding $\ff \colon U \ha \RR^2$ such that
210: $\ff (k,0) = \left( \tfrac{1}{k}, 0 \right)$, $k = 2,3, \dots$.
211: %
212: Then there does not exist any subset $A$ of $U$ containing the set
213: $\{ (k,0) \mid k = 2,3, \dots \}$
214: for which $\ff |_A$ extends to a diffeomorphism of $\RR^2$.
215: \diam
216: }
217: \end{example}
218:
219: Observe that if $(U, \ff)$ has the extension property, then $\ff$ is
220: proper in the sense that each subset $A \subset U$
221: whose closure in $\RR^{2n}$ is contained in $U$ and whose image $\ff
222: (A)$ is bounded is bounded.
223: The map $\ff$ in Example \ref{exa:1} is not proper in this sense.
224: However, the map $\ff$ in the following example is proper in this sense,
225: and still $(U, \ff)$ does not have the extension property.
226:
227: \begin{example} \label{exa:2}
228: {\rm
229: Let $U \subset \RR^2$ be the strip $\RR \,\times\, ]\;\!\!-1,0\:\![$,
230: and let
231: \[
232: A \,=\, \left\{ (x,y) \in U \;\big|\:\! \left| y+\tfrac{1}{2} \right| \le f(x)
233: \right\}
234: \]
235: where $f \colon \RR \ra \left] 0,\tfrac{1}{2} \right[$ is a smooth
236: function such that
237: %
238: \begin{equation} \label{ea:intf}
239: \int_\RR \left( \tfrac{1}{2} - f(x) \right) dx \,<\, \infty ,
240: \end{equation}
241: %
242:
243: %
244: %
245: %
246: \begin{figure}[h]
247: \begin{center}
248: \psfrag{0}{$0$}
249: \psfrag{1}{$1$}
250: \psfrag{-1}{$-1$}
251: \psfrag{x}{$x$}
252: \psfrag{y}{$y$}
253: \psfrag{U}{$U$}
254: \psfrag{A}{$A$}
255: \psfrag{ff}{$\ff$}
256: \leavevmode\epsfbox{figurea1.eps}
257: \end{center}
258: \caption{A pair $(U, \ff)$ which does not have the extension property.}
259: \label{figurea1.fig}
260: \end{figure}
261: %
262: %magnification: 38
263: %
264:
265: \noindent
266: cf.\ Figure \ref{figurea1.fig}.
267: Using the method used in Step 4 of \text{Section 2.2} in \cite{Diss}
268: we find a symplectic embedding $\ff \colon U \ha \RR^2$ such that
269: \[
270: \ff (x,y) = (x,y) \,\text{ if } x \ge 1
271: \quad \text{ and } \quad
272: \ff (x,y) = (-x,-y) \,\text{ if } x \le -1 ,
273: \]
274: cf.\ \text{Figure \ref{figurea1.fig}}.
275: In view of the estimate \eqref{ea:intf} the component $C$ of $\RR^2
276: \setminus \ff(A)$ which contains the point $(1,0)$ has finite volume.
277: Any symplectomorphism $\Phi_A$ of $\RR^2$ such that $\Phi_A |_A = \ff
278: |_A$ would map the ``upper'' component of $\RR^2 \setminus A$,
279: which has infinite volume, to $C$.
280: This is impossible.
281: \diam
282: }
283: \end{example}
284:
285: Example \ref{exa:2} shows that the assumption \eqref{ea2:2} on $\ff$
286: in Theorem \ref{ta2:ear} below cannot be omitted.
287: %
288: For technical reasons in the proof of Theorem \ref{ta2:ear} we shall
289: also impose a mild convexity condition on the starlike domain $U$.
290: The length of a smooth curve $\gg \colon [0,1] \ra \RR^n$ is defined
291: by
292: \[
293: \length (\gg) \,:=\, \int_0^1 \left| \gg' (s) \right| ds .
294: \]
295: On any domain $U \subset \RR^n$ we define a distance function $d_U
296: \colon U \times U \ra \RR$ by
297: \[
298: d_U (z,z') \,:=\, \inf \left\{ \length (\gg) \right\}
299: \]
300: where the infimum is taken over all smooth curves $\gg \colon [0,1] \ra
301: U$ with $\gg(0) = z$ and $\gg(1) = z'$.
302: Then $\left| z-z' \right| \le d_U (z,z')$ for all $z,z' \in U$.
303:
304: \begin{definition} \label{defa:lip}
305: {
306: \rm
307: We say that a domain $U \subset \RR^{n}$ is a {\it Lipschitz domain}\,
308: if there exists a constant $\ll >0$ such that
309: \[
310: d_U (z,z') \,\le\, \ll \left| z-z' \right|
311: \quad \text{for all }\, z, z' \in U.
312: \]
313: }
314: \end{definition}
315:
316: Each {\it convex}\, domain $U \subset \RR^{n}$ is a Lipschitz domain
317: with Lipschitz constant $\ll=1$.
318: It is not hard to see that there do exist starlike domains
319: which are not Lipschitz domains.
320: But we do not know of a starlike domain with smooth boundary which is
321: not a Lipschitz domain.
322:
323: \begin{theorem} \label{ta2:ear}
324: Assume that $\ff \colon U \hookrightarrow \RR^{2n}$ is a symplectic
325: embedding of a
326: starlike Lipschitz domain $U \subset \RR^{2n}$ such that
327: there exists a constant $L>0$ satisfying
328: %
329: \begin{equation} \label{ea2:2}
330: \left| \ff (z) - \ff (z') \right| \,\ge\, L \left| z-z' \right| \quad \text{for all }\, z, z' \in U.
331: \end{equation}
332: %
333: Then the pair $(U,\ff)$ has the extension property.
334: \end{theorem}
335:
336: Theorem \ref{ta2:ear} is applied in \cite{Diss}
337: to extend a symplectic vanishing theorem for bounded domains to
338: certain unbounded domains.
339:
340: \subsection*{Acknowledgements}
341: I cordially thank Urs Lang, Fran\c cois Laudenbach and Edi Zehnder
342: for useful discussions.
343:
344:
345: \section{Proofs}
346:
347: \ni
348: We shall first proceed along the lines of \cite{EH1} and then verify
349: that our assumptions on $U$ and $\ff$ are
350: sufficient to push the arguments through.
351:
352: \m
353: \noindent
354: {\bf Step 1. Reduction to a simpler case}
355:
356: \s
357: \noindent
358: We start with observing that we may assume that $U$ is starlike with
359: respect to the origin and that $\ff (0) = 0$ and $d \ff (0) = id$.
360: Indeed, suppose that \text{Theorem \ref{ta2:ear}} holds in this
361: situation, that $A$ is a subset of $U$ whose closure in $\RR^{2n}$ is
362: contained in $U$,
363: and that $U$ is starlike with respect to $p \neq 0$ or that
364: $\ff (p) \neq 0$ or that $D := d \ff (p) \neq id$.
365: For $w \in \RR^{2n}$ we denote by $\tau_w$
366: the translation $z \mapsto z + w$.
367: We define the symplectic embedding
368: $\psi \colon \left( D \circ \tau_{-p} \right)(U) \ha \RR^{2n}$ by
369: \[
370: \psi \,:=\, \tau_{-\ff(p)} \circ \ff \circ \tau_p \circ D^{-1} .
371: \]
372: Then $\psi (p) =0$ and $d \psi (p) = id$.
373: Since $U$ is starlike with respect to $p$ and $D$ is linear,
374: the domain $\left( D \circ \tau_{-p} \right)(U)$ is starlike with
375: respect to the origin, and $\left( D \circ \tau_{-p} \right)(A)$
376: is a subset of $\left( D \circ \tau_{-p} \right)(U)$ whose closure in
377: $\RR^{2n}$ is contained in $\left( D \circ \tau_{-p} \right)(U)$.
378: %
379: Assume next that $U$ is a $\ll$-Lipschitz domain.
380: We fix $w,w' \in \left( D \circ \tau_{-p} \right)(U)$ and set
381: $z = D^{-1}(w)+p$, $z' = D^{-1}(w')+p$.
382: Given any smooth path $\gg \colon [0,1] \ra U$ with $\gg(0) =z$, $\gg(1)
383: = z'$, the smooth path
384: \[
385: D \circ \tau_{-p} \circ \gg \colon [0,1] \,\ra\, \left( D \circ \tau_{-p}
386: \right)(U)
387: \]
388: runs from $w$ to $w'$,
389: and so
390: \[
391: d_{\left( D \circ \tau_{-p} \right)(U)}(w,w') \,\le\, \int_0^1 \left| D\;\!
392: \gg'(s) \right| ds
393: \,\le\, \left\| D \right\| \int_0^1 \left| \gg'(s) \right| ds.
394: \]
395: It follows that
396: %
397: \begin{eqnarray*}
398: d_{\left( D \circ \tau_{-p} \right)(U)}(w,w')
399: &\le& \left\| D \right\| d_U (z,z') \\
400: &\le& \left\| D \right\| \ll \left| z-z' \right| \\
401: &\le& \left\| D \right\| \ll \left\| D^{-1} \right\| \left| w-w' \right|.
402: \end{eqnarray*}
403: %
404: Since $w, w' \in \left( D \circ \tau_{-p} \right)(U)$ were arbitrary, we
405: conclude that $\left( D \circ \tau_{-p} \right)(U)$ is a
406: $\left\| D \right\| \left\| D^{-1} \right\| \ll$-Lipschitz domain.
407: %
408: Finally, the assumption \eqref{ea2:2} on $\ff$ yields
409: %
410: \[
411: \left| \psi (z) - \psi (z') \right| \,\ge\,
412: L \left| D^{-1}(z-z') \right| \,\ge\, \frac{L}{\| D \|} |z-z'|
413: \]
414: %
415: for all $z,z' \in \left( D \circ \tau_{-p} \right)(U)$.
416: By assumption we therefore find a symplectomorphism
417: $\Psi_{\left( D \circ \tau_{-p} \right)(A)}$ of $\RR^{2n}$
418: such that $\Psi_{\left( D \circ \tau_{-p} \right)(A)} |_{\left( D \circ \tau_{-p} \right)(A)} = \psi |_{\left( D \circ \tau_{-p} \right)(A)}$.
419: Define the symplectomorphism $\Phi_A$ of $\RR^{2n}$ by
420: \[
421: \Phi_A \,:=\, \tau_{\ff(p)} \circ \Psi_{\left( D \circ \tau_{-p}
422: \right)(A)} \circ D \circ \tau_{-p} .
423: \]
424: Then $\Phi |_A = \ff |_A$, as required.
425:
426:
427: \m
428: \noindent
429: {\bf Step 2. The classical approach}
430:
431: \s
432: \noindent
433: So assume that $U$ is starlike with respect to the origin and that $\ff
434: (0) = 0$ and $d \ff (0) = id$.
435: We denote the set of symplectic embeddings of $U$ into $\RR^{2n}$ by
436: $\mbox{Symp}\, (U, \RR^{2n})$.
437: Since $U$ is starlike with respect to the origin we can define a continuous
438: path $\ff_t \subset \mbox{Symp}\, (U, \RR^{2n})$ by setting
439: %
440: \begin{equation} \label{de:phi}
441: \ff_t(z) \,:=\,
442: \left\{ \begin{array}{ll}
443: z & \text{ if }\; t =0 , \\ [0.1em]
444: \frac{1}{t} \, \ff (tz) & \text{ if }\; t \in\;]0,1] .
445: \end{array}
446: \right.
447: \end{equation}
448: %
449: The path $\ff_t$ is smooth except possibly at $t=0$.
450: In order to smoothen $\ff_t$, we define the diffeomorphism $\eta$ of $[0,1]$
451: by
452: %
453: \begin{equation} \label{de:eta}
454: \eta (t) \,:=\,
455: \left\{ \begin{array}{ll}
456: 0 & \text{ if }\; t = 0 , \\ [0.1em]
457: e^2 \;\!e^{-2/t} & \text{ if }\; t \in\;]0,1] ,
458: \end{array}
459: \right.
460: \end{equation}
461: %
462: where $e$ denotes the Euler number,
463: and for $t \in [0,1]$ and $z \in U$ we set
464: %
465: \begin{equation} \label{de:phie}
466: \phi_t(z) \,:=\, \ff_{\eta(t)} (z) .
467: \end{equation}
468: %
469: Then $\phi_t$ is a smooth path in $\mbox{Symp}\, (U, \RR^{2n})$.
470: We have $\phi_0 = id_U$ and $\phi_1 = \ff$.
471:
472: Since $U$ is starlike, it is contractible, and so the same holds true for
473: all the open sets $\phi_t (U)$, $t \in [0,1]$.
474: We therefore find a smooth time-dependent Hamiltonian function
475: %
476: \begin{equation} \label{dr:H}
477: H \colon \bigcup_{t \,\in\, [0,1]} \{ t \} \times \phi_t (U) \,\ra\, \RR
478: \end{equation}
479: %
480: generating the path $\phi_t$,
481: i.e., $\phi_t$ is the solution of the Hamiltonian system
482: %
483: \begin{equation} \label{ee:Ham}
484: \left. \begin{array}{llll}
485: \frac{d}{dt} \phi_t(z) &=& J \nabla H_t (\phi_t(z)),
486: &\; z \in U,\; t \in [0,1] , \\ [0.2em]
487: \phi_0 (z) &=& z, &\; z \in U .
488: \end{array}
489: \right\}
490: \end{equation}
491: %
492: Here, $J$ denotes the standard complex structure defined by
493: \[
494: \omega_0 (z,w) = \langle Jz, w \rangle, \quad z,w \in \RR^{2n} .
495: \]
496: The function $H(z,t) = H_t(z)$ is determined by the first equation in
497: \eqref{ee:Ham} up to a smooth function $h(t) \colon [0,1] \ra \RR$.
498: Notice that $0 \in \phi_t(U)$ for all $t$.
499: We choose $h(t)$ such that
500: %
501: \begin{equation} \label{ea:H0}
502: H_t(0)=0 \quad \text{ for all } t \in [0,1] .
503: \end{equation}
504: %
505:
506:
507: \m
508: \noindent
509: {\bf Step 3. Intermezzo: End of the proof of Proposition \ref{pa2:ear}}
510:
511: \s
512: \noindent
513: Before proceeding with the proof of Theorem \ref{ta2:ear} we shall prove
514: \text{Proposition \ref{pa2:ear}}.
515: Fix a subset $A$ of $U$ whose closure $\overline{A}$ in $\RR^{2n}$ is
516: contained in $U$. Since $U$ is bounded, the set $\overline{A}$ is
517: compact, and so the set
518: \[
519: K = \bigcup_{t \,\in\, [0,1]} \{ t \} \times \phi_t (\overline{A})
520: \,\subset\, [0,1] \times \RR^{2n}
521: \]
522: is also compact and hence bounded.
523: We therefore find a bounded neighbourhood $V$ of $K$ which is open in
524: $[0,1] \times \RR^{2n}$ and is contained in the set
525: $\bigcup_{t \,\in\, [0,1]} \{ t \} \times \phi_t (U)$.
526: By Whitney's Theorem, there exists a smooth function
527: $f$ on $[0,1] \times \RR^{2n}$ which is equal to $1$ on $K$ and
528: vanishes outside $V$. Since $V$ is bounded,
529: the function $fH \colon [0,1] \times \RR^{2n} \ra \RR$ has compact
530: support, and so the Hamiltonian system associated with $fH$ can be
531: solved for all $t \in [0,1]$. We define
532: $\Phi_A$ to be the resulting time-$1$-map.
533: Then $\Phi_A$ is a globally defined symplectomorphism of $\RR^{2n}$ with
534: compact support, and $\Phi_A |_A = \ff |_A$.
535: The proof of \text{Proposition \ref{pa2:ear}} is thus complete.
536:
537:
538: \m
539: \noindent
540: {\bf Step 4. End of the proof of Theorem \ref{ta2:ear}}
541:
542: \s
543: \noindent
544: If the set $U$ is not bounded, the subset $A \subset U$ does not need to
545: be relatively compact, and so there might be no cut off $fH$ of $H$
546: whose Hamiltonian flow exists for all $t \in [0,1]$.
547: We therefore need to extend the Hamiltonian $H$ more carefully.
548: We shall first verify that our assumption \eqref{ea2:2} on $\ff$ implies
549: that $\nabla H$ is linearly bounded.
550: Since we do not know a direct way to extend a linearly bounded gradient
551: field to a linearly bounded gradient field, we shall then pass to the function
552: \[
553: G(t,w) \,=\, \frac{H(t,w)}{g \left( \left| w \right| \right)}
554: \]
555: where $g \left( \left| w \right| \right) = \left| w \right|$ for $\left|
556: w \right|$ large.
557: Our assumption that $U$ is a Lip\-schitz domain will imply that $G$ is
558: Lipschitz continuous in $w$ and can hence be extended to a Lipschitz
559: continuous function $\widehat{G}$ on $[0,1] \times \RR^{2n}$.
560: After smoothing $\widehat{G}$ in $w$ to $\widetilde{G}$ we shall obtain
561: an extension $\widetilde{H} (t,w) = g \left( \left| w \right| \right)
562: \widetilde{G}(t,w)$ whose gradient is linearly bounded.
563:
564: \begin{lemma} \label{la:0}
565: Let $L>0$ be the constant guaranteed by assumption \eqref{ea2:2}.
566: %
567: \begin{itemize}
568: \item[(i)]
569: $\left| \phi_t(z) - \phi_t(z') \right| \,\ge\,
570: L \left| z-z' \right|
571: \quad
572: \text{ for all }\, t \in\;]0,1] \text{ and } z, z' \in U .$
573: \item[(ii)]
574: $\left\| d \phi_t (z) \right\| \,\le\, \displaystyle\frac{1}{L}
575: \quad
576: \text{ for all }\, t \in \;]0,1] \text{ and } z \in U .$
577: \end{itemize}
578: \end{lemma}
579:
580:
581:
582:
583: \proof
584: (i)
585: In view of definitions \eqref{de:phie} and \eqref{de:phi} we have
586: %
587: \begin{equation} \label{da:idae}
588: \phi_t(z) \,=\, \frac{1}{\eta(t)} \ff \left( \eta(t) z \right)
589: \end{equation}
590: %
591: for all $t \in\;]0,1]$ and $z \in U$.
592: Together with assumption \eqref{ea2:2} we find
593: %
594: \begin{eqnarray*}
595: \left| \phi_t(z) - \phi_t(z') \right|
596: &=& \frac{1}{\eta(t)} \left| \ff (\eta(t)z) - \ff (\eta(t)z') \right| \\
597: &\ge& \frac{1}{\eta(t)} L \left| \eta(t)z - \eta(t)z' \right| \\
598: &=& L \left| z-z' \right| .
599: \end{eqnarray*}
600: %
601: Assertion (i) thus follows.
602:
603: \s
604: (ii)
605: We fix $t \in\;]0,1]$ and $z \in U$.
606: Following the proof of Proposition 2.20 in \cite{MS} we decompose the
607: linear symplectomorphism $d \phi_t(z)$ as
608: \[
609: d \phi_t(z) \,=\, PQ
610: \]
611: where both $P$ and $Q$ are symplectic
612: and $P$ is symmetric and positive definite and $Q$ is orthogonal.
613: According to \cite[Lemma 2.18]{MS} the eigenvalues of $P$ are of the
614: form
615: \[
616: 0 \,<\, \ll_1 \,\le\, \ll_2 \,\le\, \dots \,\le\, \ll_n
617: \,\le\, \ll_n^{-1} \,\le\, \dots \,\le\, \ll_2^{-1} \,\le\, \ll_1^{-1} .
618: \]
619: Since $Q$ is orthogonal, we find
620: %
621: \begin{equation} \label{ida:APl}
622: \left\| d \phi_t(z) \right\| \,=\, \left\| P \right\| \,=\, \ll_1^{-1} .
623: \end{equation}
624: %
625: Let $v_1$ be an eigenvector of $\ll_1$. In view of assertion (i) we have
626: \[
627: \ll_1 \left| v_1 \right| \,=\, \left| d \phi_t(z) v_1 \right| \,\ge\, L \left|
628: v_1 \right|
629: \]
630: and so $\ll_1^{-1} \le L^{-1}$.
631: This and the identity \eqref{ida:APl} yield
632: $\left\| d \phi_t(z) \right\| \le L^{-1}$,
633: and so assertion (ii) follows.
634: \proofend
635:
636: For $r >0$ we denote by $B(r)$ the closed $r$-ball around $0 \in \RR^{2n}$.
637: We choose $\ee >0$ so small that $B(\ee) \subset U$.
638: Finally, we abbreviate
639: %
640: \begin{equation} \label{da:Ut}
641: U_t \,=\, U \cap B \left( \frac{\ee}{e}\, e^{1/t} \right), \quad\; t \in\;]0,1].
642: \end{equation}
643: %
644:
645:
646: \begin{lemma} \label{la:1}
647: (i)
648: There exists a constant $C_1>0$ such that
649: \[
650: \left| \nabla H_t(w) \right| \,\le\, \frac{C_1}{t^2} \left| w \right|
651: \quad
652: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \phi_t(U) .
653: \]
654: (ii) There exists a constant $c_1>0$ such that
655: \[
656: \left| \nabla H_t(w) \right| \,\le\,
657: \frac{c_1}{t^2} e^{-1/t} \left| w \right|
658: \quad
659: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \phi_t(U_t) .
660: \]
661: \end{lemma}
662:
663: \proof
664: (i)
665: Fix $t \in\;]0,1]$ and $w = \phi_t(z)$.
666: Using the first line in \eqref{ee:Ham} and the definitions
667: \eqref{da:idae} and \eqref{de:eta} we compute
668: %
669: \begin{eqnarray}
670: J \nabla H_t(w) &=& \frac{d}{dt} \phi_t (z) \notag \\
671: &=& \frac{d}{dt} \left( \frac{1}{\eta(t)} \ff
672: (\eta(t)z) \right) \notag \\
673: &=& \frac{\eta'(t)}{\eta(t)} \left( - \frac{1}{\eta(t)} \ff
674: (\eta(t)z) + d \ff (\eta(t) z) z \right) \label{ide:diff} \\
675: &=& \frac{2}{t^2}
676: \,\big( -w + d \ff ( \eta(t) z ) z \big) . \label{ide:1}
677: \end{eqnarray}
678: %
679: Lemma \ref{la:0}\,(ii) with $t=1$ yields
680: %
681: \begin{equation} \label{re:dp}
682: \| d \ff (z) \| \,\le\, \frac{1}{L} \quad \text{ for all }\, z \in U
683: \end{equation}
684: %
685: and the identity $\phi_t(0) = \frac{1}{\eta(t)} \ff(0) =0$ and Lemma
686: \ref{la:0}\,(i) with $z'=0$ yield
687: %
688: \begin{equation} \label{ide:wLz}
689: |w| = \left| \phi_t(z) \right| \,\ge\, L |z| .
690: \end{equation}
691: %
692: In view of the identity \eqref{ide:1} and the estimates
693: \eqref{re:dp} and \eqref{ide:wLz} we conclude
694: %
695: \begin{eqnarray*}
696: \left| \nabla H_t(w) \right| \,=\, \left| J \nabla H_t(w) \right|
697: &\le& \frac{2}{t^2}
698: \left( |w|+ \| d \ff ( \eta(t) z) \| |z| \right) \\
699: &\le& \frac{2}{t^2}
700: \left( |w|+ \frac{1}{L^2} |w| \right) \\
701: &=& \frac{2}{t^2} \left( 1+\frac{1}{L^2} \right) |w| .
702: \end{eqnarray*}
703: %
704: The constant $C_1 := 2\left(1+\frac{1}{L^2} \right)$ is as required.
705:
706: \s
707: (ii)
708: By the choice of $\ee$, the smooth map
709: $\ff$ is $C^2$-bounded on $B(\ee)$, and so
710: Taylor's Theorem applied to $\ff \colon B(\ee) \ra \RR^{2n}$ and
711: $d \ff \colon B(\ee) \ra \cl (\RR^{2n})$
712: guarantees constants $M_1$ and $M_2$ such that for
713: each $x \in B(\ee)$,
714: \[
715: \begin{array}{rcll}
716: \ff(x) &=& \ff (0) + d \ff (0)x + r(x) &\text{ with }\, \left| r(x)
717: \right| \le M_1 \left| x \right|^2, \\ [0.2em]
718: d \ff(x) &=& d \ff (0) + R(x) &\text{ with }\, \left\| R(x)
719: \right\| \le M_2 \left| x \right| ,
720: \end{array}
721: \]
722: where $\| R(x) \|$ denotes the operator norm of the linear operator
723: $R(x) \in \cl (\RR^{2n})$.
724: Since $\ff (0) = 0$ and $d \ff (0) = id_{\RR^{2n}}$ we conclude that
725: \[
726: \left| \ff (x) - d \ff (x) x \right| \,=\, \left| r(x) - R(x)x \right|
727: \,\le\, (M_1+M_2) \left| x \right|^2 \quad \text{if } \left| x \right| \le
728: \ee
729: \]
730: and so, with $x = \eta(t)z$,
731: %
732: \begin{equation} \label{e:M12}
733: \left| \frac{1}{\eta(t)} \ff \left( \eta(t)z \right) - d \ff \left(
734: \eta(t)z \right) z \right|
735: \,\le\, (M_1+M_2) \eta(t) \left| z \right|^2 \quad \text{if } \eta(t)
736: \left| z \right| \le \ee . \;
737: \end{equation}
738: %
739: Assume now $z \in U_t$.
740: In view of the definition \eqref{da:Ut} of $U_t$ we then have
741: \[
742: \eta(t) \left| z \right| \,\le\, e^2 e^{-2/t} \,\frac{\ee}{e}\, e^{1/t} \,=\,
743: e e^{-1/t} \ee \,\le\, \ee .
744: \]
745: Inserting the estimate \eqref{e:M12} into \eqref{ide:diff} and using
746: \eqref{ide:wLz} we conclude that
747: %
748: \begin{eqnarray*}
749: \left| \nabla H_t(w) \right|
750: &\le& \frac{2}{t^2} (M_1 + M_2) \eta(t) \left|z\right|^2 \\
751: &\le& \frac{2}{t^2} (M_1 + M_2) e e^{-1/t} \ee \left|z\right| \\
752: &\le& \frac{2}{t^2} e^{-1/t} (M_1 + M_2) e \ee \frac{1}{L} \left|w\right| .
753: \end{eqnarray*}
754: %
755: The constant $c_1 := 2(M_1 + M_2) e\:\!\ee \tfrac{1}{L}$ is as required.
756: \proofend
757:
758: \begin{lemma} \label{la:2}
759: (i)
760: There exists a constant $C_2>0$ such that
761: \[
762: \left| H_t(w) \right| \,\le\, \frac{C_2}{t^2} |w|^2 \quad
763: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \phi_t(U) .
764: \]
765: (ii)
766: There exists a constant $c_2>0$ such that
767: \[
768: \left| H_t(w) \right| \,\le\, \frac{c_2}{t^2} e^{-1/t}|w|^2 \quad
769: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \phi_t(U_t) .
770: \]
771: \end{lemma}
772:
773: \proof
774: (i)
775: Fix $t \in\;]0,1]$ and $w = \phi_t(z)$.
776: The smooth path
777: \[
778: \gg \colon [0,1] \ra \phi_t(U), \qquad \gg(s) = \phi_t(sz)
779: \]
780: joins $0$ with $w$.
781: Since $H_t(0) =0$ we find that
782: %
783: \begin{eqnarray} \label{ide2:H}
784: H_t(w) &=& H_t(0) +
785: \int_0^1 \langle \nabla H_t(\gg(s)), \gg'(s) \rangle \,ds
786: \notag \\
787: &=& \int_0^1 \langle \nabla H_t(\phi_t(sz)), d \phi_t(sz)z \rangle \,ds.
788: \end{eqnarray}
789: %
790: The identity $\phi_t(0)=0$, the mean value theorem and Lemma \ref{la:0}\,(ii) yield
791: %
792: \begin{equation} \label{este:Ms}
793: \left| \phi_t(sz) \right| \;=\; \left| \phi_t(sz) - \phi_t(0) \right|
794: \;\le\; \frac{1}{L} s |z| .
795: \end{equation}
796: %
797: Using the identity \eqref{ide2:H}, Lemma \ref{la:1}\,(i),
798: Lemma \ref{la:0}\,(ii)
799: and the estimates \eqref{este:Ms} and \eqref{ide:wLz} we
800: can estimate
801: %
802: \begin{eqnarray*}
803: \left| H_t(w) \right|
804: &\le& \int_0^1 \left| \nabla H_t(\phi_t(sz)) \right| \left| d \phi_t(sz) z
805: \right| \,ds \\
806: &\le& \frac{C_1}{t^2} \frac{1}{L} \,|z| \int_0^1 \left| \phi_t(sz) \right|
807: \, ds \\
808: &\le& \frac{C_1}{t^2} \frac{1}{L^2} \,|z|^2\, \frac{1}{2} \\
809: &\le& \frac{1}{2} \,C_1 \frac{1}{L^4} \frac{1}{t^2} \,|w|^2.
810: \end{eqnarray*}
811: %
812: The constant $C_2 := \tfrac{1}{2} \;\!C_1 \frac{1}{L^4}$ is
813: as required.
814:
815: \s
816: (ii)
817: Assume now $z \in U_t$.
818: Using Lemma \ref{la:1}\,(ii) and estimating as above we obtain
819: \[
820: \left|H_t(w)\right| \,\le\, \frac{1}{2} \,c_1 \frac{1}{L^4}
821: \frac{1}{t^2} \,e^{-1/t} \left| w \right|^2 .
822: \]
823: The constant $c_2 := \tfrac{1}{2}\, c_1 \frac{1}{L^4}$ is as required.
824: \proofend
825:
826: Choose a smooth function $g \colon [0,\infty[ \; \ra [1,\infty[$ such
827: that
828: %
829: \begin{equation} \label{da:gr}
830: g (r) \,=\,
831: \left\{ \begin{array}{ll}
832: 1 & \text{ if }\; r \le \frac{1}{2} , \\ [0.2em]
833: r & \text{ if }\; r \ge 2
834: \end{array}
835: \right.
836: \end{equation}
837: %
838: and $0 \le g'(r) \le 1$ for all $r$.
839:
840: %
841: %
842: %
843: \begin{figure}[h]
844: \begin{center}
845: \psfrag{1}{$1$}
846: \psfrag{2}{$2$}
847: \psfrag{r}{$r$}
848: \psfrag{g}{$g(r)$}
849: \leavevmode\epsfbox{figurea2.eps}
850: \end{center}
851: \caption{The function $g(r)$.} \label{figurea2.fig}
852: \end{figure}
853: %
854: %magnification: 50
855: %
856:
857: \ni
858: We define the smooth function $G \colon \bigcup_{t \,\in\, [0,1]} \{t\}
859: \times \phi_t(U) \ra \RR$ by
860: %
861: \begin{eqnarray} \label{d:G}
862: G(t,w) \,\equiv\, G_t(w) \,:=\, \frac{H_t(w)}{g(|w|)} .
863: \end{eqnarray}
864: %
865: \begin{lemma} \label{la:3}
866: (i)
867: There exists a constant $C_3>0$ such that
868: \[
869: \left| \nabla G_t(w) \right| \,\le\, \frac{C_3}{t^2} \quad
870: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \phi_t(U) .
871: \]
872: (ii)
873: There exists a constant $c_3>0$ such that
874: \[
875: \left| \nabla G_t(w) \right| \,\le\, \frac{c_3}{t^2} e^{-1/t} \quad
876: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \phi_t(U_t) .
877: \]
878: \end{lemma}
879:
880: \proof
881: (i)
882: We have that
883: \[
884: \nabla \left( \frac{1}{g(|w|)} \right) \,=\, - \frac{g'(|w|)}{g(|w|)^2}
885: \, \frac{w}{|w|}
886: \]
887: and so
888: \[
889: \nabla G_t(w) \,=\, - \frac{g'(|w|)}{g(|w|)^2} \, \frac{w}{|w|} \, H_t(w) +
890: \frac{1}{g(|w|)} \, \nabla H_t (w) .
891: \]
892: %
893: Using $g'(t) \in [0,1]$, Lemma \ref{la:2}\,(i) and Lemma \ref{la:1}\,(i) and $|w|
894: \le g(|w|)$ we therefore find that
895: %
896: \begin{eqnarray*}
897: \left| \nabla G_t(w) \right|
898: &\le& \frac{1}{g(|w|)^2} \left| H_t(w) \right|
899: + \frac{1}{g(|w|)} \left| \nabla H_t(w) \right| \\
900: &\le& \frac{1}{g(|w|)^2} \, \frac{C_2}{t^2} \, |w|^2
901: + \frac{1}{g(|w|)} \, \frac{C_1}{t^2} \, |w| \\
902: &\le& (C_1+C_2) \frac{1}{t^2} .
903: \end{eqnarray*}
904: %
905: The constant $C_3 := C_1+C_2$ is as required.
906:
907: \s
908: (ii)
909: Assume now $z \in U_t$.
910: Using Lemma \ref{la:2}\,(ii) and Lemma \ref{la:1}\,(ii) and estimating
911: as above we obtain
912: \[
913: \left| \nabla G_t(w) \right| \,\le\, (c_1+c_2) \frac{1}{t^2} e^{-1/t}.
914: \]
915: The constant $c_3 := c_1+c_2$ is as required.
916: \proofend
917:
918: \begin{lemma} \label{la:4}
919: (i)
920: There exists a constant $C_4>0$ such that
921: \[
922: \left| G_t(w)- G_t(w') \right| \,\le\, \frac{C_4}{t^2} \left| w-w'
923: \right|
924: \quad
925: \text{ for all }\, t \in\;]0,1] \text{ and } w,w' \in \phi_t(U) .
926: \]
927: (ii)
928: There exists a constant $c_4>0$ such that
929: \[
930: \left| G_t(w)- G_t(w') \right| \,\le\, \frac{c_4}{t^2} e^{-1/t}
931: \left| w-w' \right|
932: \quad
933: \text{ for all }\, t \in\;]0,1] \text{ and } w,w' \in \phi_t(U_t) .
934: \]
935: \end{lemma}
936:
937: \proof
938: (i)
939: Fix $t \in\;]0,1]$ and $w = \phi_t (z)$, $w' = \phi_t (z')$,
940: and assume that $U$ is a Lipschitz domain with Lipschitz constant $\ll$.
941: We then find a smooth path $\gg \colon [0,1] \ra U$ such that $\gg (0) =
942: z$, $\gg (1) = z'$ and such that
943: %
944: \begin{eqnarray} \label{ea:Llamda}
945: \length (\gg) \,=\, \int_0^1 \left| \gg'(s) \right| ds
946: \,\le\, 2 \ll \left| z'-z \right| .
947: \end{eqnarray}
948: %
949: Using Lemma \ref{la:3}\,(i), Lemma \ref{la:0}\,(ii), the estimate
950: \eqref{ea:Llamda} and Lemma \ref{la:0}\,(i) we can estimate
951: %
952: \begin{eqnarray*}
953: \left| G_t(w')-G_t(w) \right| &=&
954: \left| \int_0^1 \langle \nabla G_t \left( \phi_t (\gg(s)) \right), d
955: \phi_t (\gg (s)) \gg' (s) \rangle ds \right| \\
956: &\le& \frac{C_3}{t^2}\, \frac{1}{L} \int_0^1 \left| \gg' (s) \right| ds \\
957: &\le& \frac{C_3}{t^2}\, \frac{1}{L} \,2 \ll \left| z'-z \right| \\
958: &\le& \frac{C_3}{t^2}\, \frac{1}{L^2} \, 2 \ll \left| w'-w \right| .
959: \end{eqnarray*}
960: %
961: The constant $C_4 := 2 C_3 \frac{1}{L^2} \ll$ is as required.
962:
963: \s
964: (ii)
965: Assume now $z, z' \in U_t$.
966: Since $U$ is starlike, we can assume that the path $\gg$ chosen above
967: is contained in $U_t$.
968: Using Lemma \ref{la:3}\,(ii) and estimating as above we obtain
969: \[
970: \left| G_t(w')-G_t(w) \right| \,\le\, \frac{c_3}{t^2} e^{-1/t}
971: \frac{1}{L^2} \:\!2 \ll \left|w'-w \right| .
972: \]
973: The constant $c_4 := 2 c_3 \tfrac{1}{L^2} \ll$ is as required.
974: \proofend
975:
976:
977: Our next goal is to extend the function $G$ on $\bigcup_{t \,\in\,
978: [0,1]} \{ t \} \times \phi_t (U)$ to a continuous function
979: $\widehat{G}$ on $[0,1] \times \RR^{2n}$ having similar properties.
980: We shall need two auxiliary lemmata.
981:
982: \begin{lemma} \label{la:McShane}
983: {\rm (Mc\,Shane \cite{Mc}\footnote{I'm grateful to Urs Lang for
984: pointing out to me this reference.})}
985: %
986: Consider a subset $W$ of the metric space $(X, d)$ and a function $f
987: \colon W \ra \RR$ which is $\ll$-Lipschitz continuous.
988: Then the function $\overline{f} \colon X \ra \RR$ defined by
989: \[
990: \overline{f} (x) \,:=\, \inf \left\{ f(w) + \ll\, d (x,w) \mid w \in W \right\}
991: \]
992: is a $\ll$-Lipschitz continuous extension of $f$.
993: \end{lemma}
994:
995: \begin{lemma} \label{la:lip}
996: Assume that $V$ is a subset of $\RR^{2n}$ which contains the origin and
997: that the function
998: $h \colon V \cup B(2r) \ra \RR$ is $\ll_V$-Lipschitz continuous on $V$
999: and $\ll_B$-Lipschitz continuous on $B(2r)$.
1000: Then $h$ is $(2\ll_V + \ll_B$)-Lipschitz continuous on $V \cup B(r)$.
1001: \end{lemma}
1002:
1003: \proof
1004: Fix $w,w' \in V \cup B(r)$.
1005: If $w,w' \in V$ or $w,w' \in B(2r)$ then by assumption
1006: \[
1007: \left| h(w) - h(w') \right| \,\le\, \max (\ll_V, \ll_B) \left| w-w'
1008: \right| .
1009: \]
1010: So assume that $w \in V \setminus B(2r)$ and $w' \in B(r)$.
1011: Then
1012: $\left| w' \right| \le r \le \tfrac{\left| w \right|}{2}$ and so
1013: \[
1014: \frac{\left| w \right|}{2} \,\le\, \left| w \right| - \left| w' \right| \,\le\,
1015: \left| w-w' \right| .
1016: \]
1017: Since $0 \in V$ and $0 \in B(2r)$ we can now estimate
1018: %
1019: \begin{eqnarray*}
1020: \left| h(w) - h(w') \right| &\le& \left| h(w) - h(0) \right| +
1021: \left| h(w') - h(0) \right| \\
1022: &\le& \ll_V \left| w \right| + \ll_B \left| w' \right| \\
1023: &\le& (2 \ll_V + \ll_B) \frac{\left| w \right|}{2} \\
1024: &\le& (2 \ll_V + \ll_B) \left| w-w' \right|,
1025: \end{eqnarray*}
1026: %
1027: and so $h$ is $(2 \ll_V + \ll_B)$-Lipschitz continuous on $V \cup B(r)$.
1028: \proofend
1029:
1030:
1031: \begin{lemma} \label{la:ext}
1032: There exists a continuous function
1033: $\widehat{G} \colon [0,1] \times \RR^{2n} \ra \RR$
1034: with the following properties.
1035: %
1036: \begin{itemize}
1037: \item[(i)]
1038: $\widehat{G} (t,w) = G (t,w) \quad
1039: \text{ for all }\, t \in [0,1] \text{ and } w \in \phi_t(U) .$
1040: %\item[(ii)]
1041: %$\widehat{G} (0,w) = 0 \quad
1042: % \text{ for all }\, w \in \RR^{2n} .$
1043: \item[(ii)]
1044: There exists a constant $C_5>0$ such that
1045: \[
1046: \left| \widehat{G}_t(w)- \widehat{G}_t(w') \right| \,\le\,
1047: \frac{C_5}{t^2} \left| w-w' \right|
1048: \quad
1049: \text{ for all }\, t \in\;]0,1] \text{ and } w,w' \in \RR^{2n} .
1050: \]
1051: \end{itemize}
1052: \end{lemma}
1053:
1054:
1055: \proof
1056: We shall first construct a function $\widehat{G} \colon [0,1] \times
1057: \RR^{2n} \ra \RR$ meeting assertions (i) and (ii), and shall then
1058: verify that $\widehat{G}$ is continuous.
1059:
1060: \s
1061: We set $\widehat{G} (0,w) =0$ for all $w \in \RR^{2n}$.
1062: Since
1063: $H \colon \bigcup_{t \,\in\, [0,1]} \{ t \} \times \phi_t (U) \ra
1064: \RR$ is continuous, Lemma \ref{la:2}\,(ii) and the definition \eqref{da:Ut}
1065: of $U_t$ imply that $H(0,w) =0$ for all $w \in U$.
1066: In view of definition \eqref{d:G} we therefore have $G(0,w)=0$ for all
1067: $w \in U$, and so assertion (i) holds for $t=0$.
1068:
1069: We now fix $t \in\;]0,1]$.
1070: We define the number $R_t$ by
1071: %
1072: \begin{equation} \label{da:Rt}
1073: R_t \,=\, \frac{L}{2} \frac{\ee}{e}\, e^{1/t} .
1074: \end{equation}
1075: %
1076: Fix $w = \phi_t (z) \in B(2R_t)$.
1077: In view of the estimate \eqref{ide:wLz} and the definition \eqref{da:Rt}
1078: we have
1079: \[
1080: \left| z \right| \,\le\, \frac{\left| w \right|}{L} \,\le\,
1081: \frac{2}{L} R_t \,=\, \frac{\ee}{e}\, e^{1/t} ,
1082: \]
1083: and so $z \in U_t$ in view of definition \eqref{da:Ut}.
1084: Lemma \ref{la:4}\,(ii) therefore implies that the function $G_t$ is
1085: $\tfrac{c_4}{t^2} e^{-1/t}$-Lipschitz continuous on $\phi_t(U) \cap B(2
1086: R_t)$.
1087: According to Lemma \ref{la:McShane} the function $\overline{G}_t \colon
1088: B(2R_t) \ra \RR$ defined by
1089: %
1090: \begin{equation} \label{da:Gbar}
1091: \overline{G}_t (x) \,:=\, \inf \left\{ G_t(w) + \frac{c_4}{t^2}e^{-1/t}
1092: \left|x-w \right| \,\big|\, w \in \phi_t(U) \cap B(2R_t) \right\} \;\;
1093: \end{equation}
1094: %
1095: is a $\tfrac{c_4}{t^2} e^{-1/t}$-Lipschitz extension of $G_t$ to
1096: $B(2R_t)$.
1097: In particular, the function
1098: $\overline{\overline{G}}_t \colon \phi_t(U) \cup B(2R_t) \ra \RR$,
1099: %
1100: \begin{equation} \label{da:Gbbar}
1101: \overline{\overline{G}}_t (x) \,:=\,
1102: \left\{ \begin{array}{ll}
1103: G_t(x) & \text{ if }\; x \in \phi_t(U) , \\ [0.1em]
1104: \overline{G}_t(x) & \text{ if }\; x \in B (2R_t) ,
1105: \end{array}
1106: \right.
1107: \end{equation}
1108: %
1109: is well-defined.
1110: According to Lemma \ref{la:4}\,(i), $\overline{\overline{G}}_t$ is
1111: $\tfrac{C_4}{t^2}$-Lipschitz continuous on $\phi_t(U)$, and according to
1112: the above, $\overline{\overline{G}}_t$ is $\tfrac{c_4}{t^2}$-Lipschitz
1113: continuous on $B(2R_t)$.
1114: According to Lemma \ref{la:lip}, the restriction of
1115: $\overline{\overline{G}}_t$
1116: to $\phi_t(U) \cup B(R_t)$ is therefore $\tfrac{C_5}{t^2}$-Lipschitz
1117: continuous where we abbreviated
1118: \[
1119: C_5 \,:=\, 2 C_4 + c_4 .
1120: \]
1121: Applying Lemma \ref{la:McShane} once more, we find that the function
1122: $
1123: \widehat{G}_t \colon \RR^{2n} \ra \RR
1124: $
1125: defined by
1126: %
1127: \begin{equation} \label{da:Ghat}
1128: \widehat{G}_t (x) \,:=\, \inf \left\{ \overline{\overline{G}}_t(w) +
1129: \frac{C_5}{t^2} \left|x-w \right| \,\big|\,
1130: w \in \phi_t(U) \cup B(R_t) \right\} \;\;
1131: \end{equation}
1132: %
1133: is a $\tfrac{C_5}{t^2}$-Lipschitz extension of the restriction of
1134: $\overline{\overline{G}}_t$ to $\phi_t(U) \cup B(R_t)$.
1135: In particular,
1136: \[
1137: \widehat{G}(t,w) \,=\, G(t,w) \quad
1138: \text{ for all }\, w \in \phi_t(U) .
1139: \]
1140: The function $\widehat{G} \colon ]0,1] \times \RR^{2n} \ra \RR$ thus
1141: defined therefore meets assertion (i) for $t \in \;]0,1]$ and assertion
1142: (ii).
1143:
1144: %
1145: %
1146: %
1147: \begin{figure}[h]
1148: \begin{center}
1149: \psfrag{p}{$\phi_t(U)$}
1150: \leavevmode\epsfbox{figurea3.eps}
1151: \end{center}
1152: \caption{The domain $\phi_t(U)$ and its intersections with
1153: $B(R_t)$ and $B(2R_t)$.} \label{figurea3.fig}
1154: \end{figure}
1155: %
1156: %magnification: 35
1157: %
1158:
1159: We are left with showing that the function
1160: $\widehat{G} \colon [0,1] \times \RR^{2n} \ra \RR$ constructed in the
1161: previous two steps is continuous.
1162: %
1163: The definitions \eqref{da:Rt}, \eqref{da:Gbar}, \eqref{da:Gbbar} and
1164: \eqref{da:Ghat} show that the functions
1165: $\widehat{G}(\;\!\cdot\;\!, x) \colon ]0,1] \ra \RR$ and
1166: $\widehat{G}(t,\:\!\cdot\:\!) \colon \RR^{2n} \ra \RR$
1167: are continuous.
1168: This and the fact that the functions $\widehat{G}(t,\:\!\cdot\:\!)$ are
1169: $\tfrac{C_5}{t^2}$-Lipschitz continuous imply that $\widehat{G} \colon
1170: ]0,1] \times \RR^{2n} \ra \RR$ is continuous.
1171: %
1172: In order to show that $\widehat{G}$ is also continuous at $(0,w)$ for
1173: each $w \in \RR^{2n}$ we fix $w \in \RR^{2n}$.
1174: We choose an open ball $B_w \subset \RR^{2n}$ centered at $w$.
1175: %
1176: In view of definition \eqref{da:Rt} we have $R_t \ra \infty$ as $t \ra
1177: 0^+$.
1178: We therefore find $t_0>0$ such that $B_w \subset B(R_t)$ for all $t \in
1179: \;]0,t_0]$. We fix $t \in \;]0,t_0]$ and $w' \in B_w$.
1180: Recalling the definition of $\widehat{G}(t,w') \,\equiv\, \widehat{G}_t(w')$
1181: we see that
1182: %
1183: \begin{eqnarray*}
1184: \widehat{G}_t (w') \;=\; \overline{\overline{G}}_t (w') &\!=\!&
1185: \overline{G}_t(w') \\
1186: &\!=\!& \inf \left\{ G_t(v) + \frac{c_4}{t^2} e^{-1/t} \left|w'-v \right|
1187: \,\big|\, v \in \phi_t(U) \cap B(2R_t) \right\} .
1188: \end{eqnarray*}
1189: %
1190: Since $0 = \phi_t(0) \in \phi_t(U) \cap B(2R_t)$ and $G_t(0) = H_t(0)
1191: =0$ we conclude that
1192: %
1193: \begin{equation} \label{ea:Gt0}
1194: \widehat{G}_t(w') \,\le\, \frac{c_4}{t^2} e^{-1/t} \left| w' \right| .
1195: \end{equation}
1196: %
1197: Moreover, we recall from the beginning of the proof of Lemma
1198: \ref{la:ext} that $G_t$ is $\tfrac{c_4}{t^2} e^{-1/t}$-Lipschitz
1199: continuous on $\phi_t(U) \cap B(2R_t)$.
1200: This and $G_t(0) = 0$ yield
1201: \[
1202: \left| G_t(v) \right| \,\le\, \frac{c_4}{t^2} e^{-1/t} \left| v \right|
1203: \quad \text{ for all }\, v \in \phi_t(U) \cap B(2R_t) .
1204: \]
1205: Therefore,
1206: %
1207: \begin{eqnarray*}
1208: G_t(v) + \frac{c_4}{t^2} e^{-1/t} \left| w'-v \right| &\ge&
1209: - \left| G_t(v) \right| + \frac{c_4}{t^2} e^{-1/t}
1210: \left|w'-v \right| \\
1211: &\ge& \frac{c_4}{t^2} e^{-1/t} ( - \left|v\right| +
1212: \left|w'-v\right|) \\
1213: &\ge& - \frac{c_4}{t^2} e^{-1/t} \left| w' \right|
1214: \end{eqnarray*}
1215: %
1216: for all $v \in \phi_t(U) \cap B(2R_t)$.
1217: We conclude that
1218: %
1219: \begin{equation} \label{ea:Gt0'}
1220: \widehat{G}_t(w') \,\ge\, - \frac{c_4}{t^2} e^{-1/t} \left| w' \right| .
1221: \end{equation}
1222: %
1223: The estimates \eqref{ea:Gt0} and \eqref{ea:Gt0'}, which hold for all $t
1224: \in \;]0,t_0]$ and $w' \in B_w$, now imply that
1225: \[
1226: \left| \widehat{G}_t(w') \right| \,\le\, \frac{c_4}{t^2} e^{-1/t} \left|
1227: w' \right|
1228: \quad
1229: \text{ for all }\, t \in \;]0,t_0] \text{ and } w' \in B_w
1230: \]
1231: and so $\widehat{G}$ is continuous at $(0,w)$.
1232: This completes the proof of Lemma \ref{la:ext}.
1233: \proofend
1234:
1235:
1236: Let now $A$ be a subset of $U$ whose closure in $\RR^{2n}$ is contained
1237: in $U$.
1238: Since also the origin is contained in $U$, we can assume that $A$ is
1239: closed and $0 \in A$.
1240: We abbreviate
1241: \[
1242: \ca \,:=\, \bigcup_{t \in [0,1]} \{t\} \times \phi_t(A) .
1243: \]
1244: %
1245: The next step is to smoothen $\widehat{G}$ in the variable $w$ in such a
1246: way that the smoothened function $\widetilde{G}$ coincides with
1247: $\widehat{G}$ on $\ca$.
1248: We shall first construct a smooth function $G^*$ which approximates
1249: $\widehat{G}$ very well and shall then obtain $\widetilde{G}$ by
1250: interpolating between $\widehat{G}$ and $G^*$.
1251:
1252: \s
1253: Since $\RR^{2n}$ is a normal space, we find an open set $V$ in $\RR^{2n}$
1254: such that
1255: $A \subset V \subset \overline{V} \subset U$. Then
1256: %
1257: \begin{equation} \label{ra:AVU}
1258: \phi_t(A) \,\subset\, \phi_t(V) \,\subset\, \phi_t ( \overline{V} )
1259: \,=\, \overline{\phi_t(V)} \,\subset\, \phi_t(U)
1260: \quad
1261: \text{ for all }\, t \in [0,1] .
1262: \end{equation}
1263: %
1264: We abbreviate
1265: \[
1266: \cv \,:=\, \bigcup_{t \in [0,1]} \{t\} \times \phi_t(V) .
1267: \]
1268: Since $\ca$ is closed and $\cv$ is open in $[0,1] \times \RR^{2n}$,
1269: we find a smooth function $f \colon [0,1] \times \RR^{2n} \ra [0,1]$ such
1270: that
1271: %
1272: \begin{equation} \label{ia:ft01}
1273: f |_\ca = 1
1274: \quad \text{ and } \quad
1275: f |_{[0,1] \times \RR^{2n} \setminus \cv} = 0.
1276: \end{equation}
1277: %
1278:
1279: We say that a continuous function $F \colon [0,1] \times \RR^{2n} \ra
1280: \RR$ is
1281: {\it smooth in the variable $w \in \RR^{2n}$} if all derivatives $D^kF_t(w)$
1282: of $F$ with respect to $w$ exist and are continuous on $[0,1] \times \RR^{2n}$.
1283:
1284: \begin{lemma} \label{la:star}
1285: There exists a continuous function
1286: $G^* \colon [0,1] \times \RR^{2n} \ra \RR$
1287: which is smooth in the variable $w \in \RR^{2n}$
1288: and has the following properties.
1289: %
1290: \begin{itemize}
1291: \item[(i)]
1292: $\left| \nabla f_t (w) \right| \left| G^*_t (w) - \widehat{G}_t (w)
1293: \right| \,\le\, {\displaystyle\frac{C_5}{t^2}} \quad
1294: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \RR^{2n} .$
1295: %\item[(ii)]
1296: %$G^* (0,w) = 0 \quad
1297: % \text{ for all }\, w \in \RR^{2n} .$
1298: \item[(ii)]
1299: $
1300: \left| \nabla G_t^*(w) \right| \le {\displaystyle\frac{2C_5}{t^2}}
1301: \quad
1302: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \RR^{2n} .$
1303: \end{itemize}
1304: \end{lemma}
1305:
1306:
1307: \proof
1308: For each $l \in \NN$ we define the open subset $V_l$ of $[0,1] \times
1309: \RR^{2n}$ by
1310: %
1311: \begin{equation} \label{da:Vl}
1312: V_l \,:=\, \left\{\:\! (t,w) \in [0,1] \times \RR^{2n} \mid \left| \nabla
1313: f_t(w) \right| < l \:\!\right\} .
1314: \end{equation}
1315: %
1316: Then there exists a smooth partition of unity
1317: $\left\{ \theta_i \right\}_{i \in \NN}$
1318: on $[0,1] \times \RR^{2n}$ such that for each $i$ the support $\supp \theta_i$
1319: is compact and contained in some $V_l$. We let $l_i$ be a number such
1320: that $\supp \theta_i \subset V_{l_i}$.
1321: Since $\left\{ \supp \theta_i \right\}$ form a locally finite covering of
1322: $[0,1] \times \RR^{2n}$, the set
1323: \[
1324: \Theta_i \,:=\,
1325: \left\{\:\! j \in \NN \mid \supp \theta_i \cap \supp \theta_j \neq
1326: \emptyset \:\!\right\}
1327: \]
1328: is finite; let its cardinality be $m_i$.
1329: We set
1330: %
1331: \begin{equation} \label{da:Mi}
1332: M_i \,:=\, \max \left\{\:\! m_j \mid j \in \Theta_i \:\!\right\} .
1333: \end{equation}
1334: %
1335: Since the functions $\theta_i$ have compact support, the numbers
1336: %
1337: \begin{equation} \label{da:muit}
1338: \mu_i \,:=\, \max \left\{ \left| \nabla \theta_i^t(w) \right| \,\big|\, (t,w) \in
1339: [0,1] \times \RR^{2n} \right\} +1
1340: \end{equation}
1341: %
1342: are finite.
1343: We define positive numbers $r_i$ by
1344: %
1345: \begin{equation} \label{da:rit}
1346: r_i \,:=\, \frac{1}{l_i M_i \mu_i} .
1347: \end{equation}
1348: %
1349: We next choose a smooth bump function $K \colon \RR^{2n} \ra [0,\infty[$
1350: such that $\supp K \subset B(1)$ and $\int_{\RR^{2n}} K(v) \,dv =1$.
1351: We abbreviate
1352: \[
1353: \kappa \,:=\, \max \left\{ \left| \nabla K (v) \right| \mid v \in
1354: \RR^{2n} \right\} .
1355: \]
1356: For each $i$ we define a smooth function $K_i \colon \RR^{2n}
1357: \ra [0, \infty[$ by
1358: \[
1359: K_i (w) \,:=\,
1360: \frac{1}{r_i^{2n}} \,K \left( \frac{w}{r_i}\right) .
1361: \]
1362: Then $\supp K_i \subset B(r_i)$ and $\int_{\RR^{2n}} K_i (v) \,dv
1363: =1$, and
1364: %
1365: \begin{equation} \label{ea:gKti}
1366: \left| \nabla K_i (w) \right| \,\le\, \frac{1}{r_i^{2n+1}}\, \kappa
1367: \quad
1368: \text{ for all }\, w \in \RR^{2n} .
1369: \end{equation}
1370: %
1371: Let $\widehat{G}$ be the function guaranteed by Lemma \ref{la:ext}.
1372: For each $i$ we define the function $G_i^* \colon [0,1] \times \RR^{2n}
1373: \ra \RR$ as the convolution
1374: %
1375: \begin{equation} \label{da:Gistar}
1376: G_i^* (t,w) \,:=\, \left( \widehat{G}_t * K_i \right) (w) \,\equiv\,
1377: \int_{\RR^{2n}} \widehat{G}_t (v) K_i(w-v)\,dv .
1378: \end{equation}
1379: %
1380: Since for each $t$ the function $\widehat{G}_t$ is continuous and since
1381: $K_i$ is smooth, the function $w \mapsto G_i^*(t,w)$ is smooth and
1382: %
1383: \begin{equation} \label{ia:Dk}
1384: D^k G_i^*(t,w) \,=\, \int_{\RR^{2n}} \widehat{G}_t(v) D^k K_i(w-v) \,dv,
1385: \quad\; k=0,1,2, \dots
1386: \end{equation}
1387: %
1388: (see, e.g., \cite[Chapter 2, Theorem 2.3]{Hi}).
1389: The function $\widehat{G}$ is continuous, and
1390: $D^kK_i$ is continuous and has compact support and is thus uniformly continuous.
1391: Formula \eqref{ia:Dk} therefore shows that $D^kG_i^*$
1392: is continuous, $k=0,1,2, \dots$,
1393: and so $G_i^*$ is continuous and smooth in $w$.
1394: %
1395: It follows that the function $G^* \colon [0,1] \times \RR^{2n} \ra \RR$
1396: defined by
1397: %
1398: \begin{equation} \label{da:Gstar}
1399: G^* (t,w) \,:=\, \sum_i \theta_i (t,w) G_i^* (t,w)
1400: \end{equation}
1401: %
1402: is continuous and smooth in $w$.
1403: %Moreover, the identities $\widehat{G} (0,v) = 0$ holding for all
1404: %$v \in \RR^{2n}$ imply assertion (ii).
1405: In order to prove assertions (i) and (ii) we fix $t \in\;]0,1]$ and abbreviate
1406: %
1407: \[
1408: \theta_i(w) = \theta_i(t,w), \quad \:\!\!
1409: \widehat{G}(w) = \widehat{G}(t,w), \quad \:\!\!
1410: G_i^*(w) = G_i^*(t,w), \quad \:\!\!
1411: G^*(w) = G^*(t,w).
1412: \]
1413: %
1414: %\noindent
1415: {\it Proof of (i)}.\;
1416: Using the definition \eqref{da:Gistar} of the function $G_j^*$ and the identity $\int_{\RR^{2n}} K_j
1417: (v) \,dv =1$ we find
1418: %
1419: \begin{eqnarray*}
1420: G_j^* (w) - \widehat{G}(w)
1421: &=& \int_{\RR^{2n}} \left( \widehat{G}(v) - \widehat{G}(w) \right) K_j (w-v)\, dv \\
1422: &=& \int_{\RR^{2n}} \left( \widehat{G}(w-v) - \widehat{G}(w) \right) K_j
1423: (v)\, dv
1424: \end{eqnarray*}
1425: %
1426: and so, together with Lemma \ref{la:ext}\,(ii),
1427: %
1428: \begin{eqnarray} \label{ea:2}
1429: \left| G_j^* (w) - \widehat{G}(w) \right|
1430: &\le&
1431: \int_{\RR^{2n}} \left| \widehat{G}(w-v) - \widehat{G}(w) \right| K_j
1432: (v)\, dv \notag\\
1433: &\le&
1434: \int_{B(r_j)} \frac{C_5}{t^2} \left| v \right| K_j(v)\, dv \notag\\
1435: &\le&
1436: \frac{C_5}{t^2} \, r_j \int_{B(r_j)} K_j(v)\, dv \notag\\
1437: &=&
1438: \frac{C_5}{t^2} \, r_j .
1439: \end{eqnarray}
1440: %
1441: If $\nabla f_t(w) = 0$, assertion (i) is obvious. So assume $\left|
1442: \nabla f_t (w) \right| > 0$.
1443: Recall from the definitions \eqref{da:Mi} and \eqref{da:muit}
1444: that $M_j \ge 1$ and $\mu_j \ge 1$.
1445: This, the definition \eqref{da:rit} of $r_j$, the inclusion $\supp
1446: \theta_j \subset V_{l_j}$ and the definition \eqref{da:Vl} of $V_{l_j}$ yield
1447: %
1448: \begin{equation} \label{ea:rjL}
1449: r_j\,=\, \frac{1}{l_j M_j \mu_j} \,\le\, \frac{1}{l_j} \,\le\,
1450: \frac{1}{\left| \nabla f_t(w) \right|}
1451: \quad
1452: \text{ for all }\, w \in \supp \theta_j .
1453: \end{equation}
1454: %
1455: The definition \eqref{da:Gstar} of $G^*$ and the estimates \eqref{ea:2} and
1456: \eqref{ea:rjL} now yield
1457: %
1458: \begin{eqnarray*}
1459: \left| G^*(w) - \widehat{G}(w) \right|
1460: &=&
1461: \left| \sum_j \theta_j(w) \left( G_j^*(w) - \widehat{G}(w) \right)
1462: \right| \\
1463: &\le&
1464: \sum_j \theta_j(w) \frac{C_5}{t^2} \frac{1}{\left| \nabla f_t(w) \right|} \\
1465: &=&
1466: \frac{C_5}{t^2} \frac{1}{\left| \nabla f_t(w) \right|}
1467: \end{eqnarray*}
1468: %
1469: and so assertion (i) follows.
1470:
1471: \s
1472: \noindent
1473: {\it Proof of (ii)}.\;
1474: Using the definition \eqref{da:Gstar} of $G^*$ and the identities
1475: $\sum_j \theta_j (w) = \sum_j \theta_j (w') = 1$ we compute that for all
1476: $w, w' \in \RR^{2n}$,
1477: %
1478: \begin{eqnarray} \label{ia:long}
1479: G^* (w') - G^*(w)
1480: &=&
1481: \sum_j \theta_j (w') G_j^* (w') - \sum_j \theta_j (w) G_j^*(w) \notag\\
1482: &=&
1483: \sum_j \left( \theta_j (w') - \theta_j (w) \right) \big( G_j^* (w') -
1484: \widehat{G} (w') \big) \notag \\
1485: & & \qquad \qquad \!\! + \sum_j \theta_j (w) \left( G_j^* (w') - G_j^*(w) \right) . \qquad
1486: \end{eqnarray}
1487: %
1488: Fix now $w$. We choose $i$ such that $\theta_i(w) >0$, and we choose an
1489: open ball $B_w \subset \RR^{2n}$ centered at $w$ such that $B_w \subset
1490: \supp \theta_i$.
1491: Fix $w' \in B_w$.
1492: In view of the mean value theorem and the definition \eqref{da:muit} of
1493: $\mu_j$ we find that
1494: %
1495: \begin{equation} \label{ea:thmj}
1496: \left| \theta_j (w') - \theta_j (w) \right|
1497: \,\le\,
1498: \max_{v \in B_w} \left| \nabla \theta_j (v) \right| \left|
1499: w'-w \right|
1500: \,\le\,
1501: \mu_j \left| w'-w \right|
1502: \end{equation}
1503: %
1504: and the estimate \eqref{ea:2} with $w$ replaced by $w'$ yields
1505: %
1506: \begin{equation} \label{ea:ggj}
1507: \left| G_j^* (w') - \widehat{G} (w') \right|
1508: \,\le\, \frac{C_5}{t^2} r_j .
1509: \end{equation}
1510: %
1511: The definition \eqref{da:Mi} of $M_j$ implies that
1512: $M_j \ge m_i$ whenever $j \in \Theta_i$, and so
1513: %
1514: \begin{equation} \label{ea:mi1}
1515: \sum_{j \in \Theta_i} \frac{1}{M_j} \,\le\, \sum_{j \in \Theta_i}
1516: \frac{1}{m_i} \,=\, 1
1517: \end{equation}
1518: %
1519: in view of the definition of $m_i$.
1520: The definition \eqref{da:rit} of $r_j$ and the inequalities
1521: $l_j \ge 1$ and \eqref{ea:mi1} yield
1522: %
1523: \begin{equation} \label{ea:mujr}
1524: \sum_{j \in \Theta_i} \mu_j r_j \,=\,
1525: \sum_{j \in \Theta_i} \mu_j \frac{1}{l_jM_j\mu_j} \,\le\,
1526: \sum_{j \in \Theta_i} \frac{1}{M_j} \,\le\,1 .
1527: \end{equation}
1528: %
1529: Since $w,w' \in B_w \subset \supp \theta_i$ we have
1530: $\theta_j(w') - \theta_j(w) =0$ if $j \notin \Theta_i$.
1531: This and the estimates \eqref{ea:thmj}, \eqref{ea:ggj} and
1532: \eqref{ea:mujr} now show that
1533: %
1534: \begin{eqnarray} \label{ea:first}
1535: \left| \sum_j \left( \theta_j (w') - \theta_j (w) \right) \left( G_j^* (w') -
1536: \widehat{G} (w') \right) \right|
1537: &\le& \sum_{j \in \Theta_i} \mu_j \left| w'-w \right| \frac{C_5}{t^2}\,
1538: r_j \notag\\
1539: &\le& \frac{C_5}{t^2} \left| w'-w \right| .
1540: \end{eqnarray}
1541: %
1542: Next, the definition \eqref{da:Gistar} of $G_j^*$ and the identity
1543: $\int_{\RR^{2n}} K_j(v)\,dv =1$ yield
1544: %
1545: \begin{eqnarray*}
1546: G_j^*(w') - G_j^*(w)
1547: &=& \int_{\RR^{2n}} \widehat{G}(v) \left( K_j (w'-v) - K_j(w-v) \right) dv \\
1548: &=& \int_{\RR^{2n}} \left( \widehat{G}(w'-v) - \widehat{G} (w-v) \right) K_j (v)\,dv.
1549: \end{eqnarray*}
1550: %
1551: Together with Lemma \ref{la:ext}\,(ii) we obtain
1552: \[
1553: \left| G_j^*(w') - G_j^*(w) \right| \,\le\,
1554: \frac{C_5}{t^2} \int_{\RR^{2n}} \left|w'-w \right| K_j(v)\,dv \,=\,
1555: \frac{C_5}{t^2} \left|w'-w \right|
1556: \]
1557: and so
1558: %
1559: \begin{equation} \label{ea:mid}
1560: \left| \sum_j \theta_j(w) \left( G_j^*(w') - G_j^*(w) \right) \right|
1561: \,\le\,
1562: \frac{C_5}{t^2} \left|w'-w \right| .
1563: \end{equation}
1564: %
1565: The identity \eqref{ia:long} and the estimates \eqref{ea:first} and
1566: \eqref{ea:mid} now imply
1567: \[
1568: \left| G^*(w') - G^*(w) \right| \,\le\, \frac{2C_5}{t^2} \left| w'-w
1569: \right| .
1570: \]
1571: Since $w' \in B_w$ was arbitrary, we conclude that
1572: \[
1573: \left| \nabla G^* (w) \right| \,\le\, \frac{2C_5}{t^2}
1574: \]
1575: and so assertion (ii) follows.
1576: The proof of Lemma \ref{la:star} is complete.
1577: \proofend
1578:
1579:
1580: \begin{lemma} \label{la:smooth}
1581: There exists a continuous function
1582: $\widetilde{G} \colon [0,1] \times \RR^{2n} \ra \RR$
1583: which is smooth in the variable $w \in \RR^{2n}$
1584: and has the following properties.
1585: %
1586: \begin{itemize}
1587: \item[(i)]
1588: $
1589: \widetilde{G} (t,w) = G (t,w) \quad
1590: \text{ for all }\, t \in [0,1] \text{ and } w \in \phi_t(A) .$
1591: %\item[(ii)]
1592: %$\widetilde{G} (t,w) = 0 \quad
1593: % \text{ for all }\, t \le 0 \text{ and } w \in \RR^{2n} .$
1594: \item[(ii)]
1595: There exists a constant $C_6>0$ such that
1596: \[
1597: \left| \nabla \widetilde{G}_t(w) \right| \,\le\,
1598: \frac{C_6}{t^2}
1599: \quad
1600: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \RR^{2n} .
1601: \]
1602: \end{itemize}
1603: \end{lemma}
1604:
1605:
1606: \proof
1607: Let $f \colon [0,1] \times \RR^{2n} \ra [0,1]$ be the smooth function
1608: chosen before Lemma \ref{la:star},
1609: and let $\widehat{G}$ and $G^*$ be the continuous functions on $[0,1]
1610: \times \RR^{2n}$ guaranteed by Lemma \ref{la:ext} and Lemma
1611: \ref{la:star}.
1612: We define a continuous function $\widetilde{G} \colon [0,1] \times
1613: \RR^{2n} \ra \RR$ by
1614: \[
1615: \widetilde{G} (t,w) \,:=\,
1616: f(t,w) \widehat{G} (t,w) + (1-f(t,w)) G^* (t,w).
1617: \]
1618: The inclusions \eqref{ra:AVU} and the identities \eqref{ia:ft01}, Lemma
1619: \ref{la:ext}\,(i) and the fact that $G^*$ is smooth in $w$ imply that
1620: $\widetilde{G}$ is smooth in $w$ and that assertion (i) holds true.
1621: %Assertion (ii) follows from Lemma \ref{la:ext}\,(ii) and Lemma
1622: %\ref{la:star}\,(ii).
1623: In order to verify assertion (ii) we fix $t \in\;]0,1]$. We first assume $w \in
1624: \phi_t(U)$.
1625: On $\phi_t(U)$ we have $\widehat{G}_t = G_t$, and so
1626: \[
1627: \nabla \widetilde{G}_t(w) \,=\,
1628: \nabla f_t(w) \left( \widehat{G}_t(w)-G_t^*(w) \right)
1629: +
1630: f_t(w) \nabla G_t(w) + \left( 1-f_t(w) \right) \nabla G_t^* (w) .
1631: \]
1632: In view of Lemma \ref{la:star}\,(i), Lemma \ref{la:3}\,(i) and Lemma
1633: \ref{la:star}\,(ii) we can therefore estimate
1634: %
1635: \begin{eqnarray*}
1636: \left| \nabla \widetilde{G}_t(w) \right|
1637: &\le&
1638: \left| \nabla f_t(w) \right| \left| G_t^*(w) - \widehat{G}_t(w) \right|
1639: +
1640: \left| \nabla G_t(w) \right| + \left| \nabla G_t^* (w) \right| \\
1641: &\le&
1642: \frac{C_5}{t^2} + \frac{C_3}{t^2} + \frac{2C_5}{t^2} .
1643: \end{eqnarray*}
1644: %
1645: We next assume $w \in \RR^{2n} \setminus \overline{\phi_t(V)}$.
1646: On $\RR^{2n} \setminus \overline{\phi_t(V)}$ we have $f_t \equiv 0$, and
1647: so
1648: \[
1649: \left| \nabla \widetilde{G}_t(w) \right| \,=\,
1650: \left| \nabla G_t^*(w) \right|
1651: \,\le\, \frac{2C_5}{t^2} .
1652: \]
1653: Setting $C_6 := C_3 + 3 C_5$ assertion (ii) follows.
1654: The proof of Lemma \ref{la:smooth} is complete.
1655: \proofend
1656:
1657: We are now in a position to define the desired extension $\widetilde{H}$
1658: of $H$.
1659: Let $g$ be the function chosen in \eqref{da:gr} and let $\widetilde{G}$
1660: be the function guaranteed by Lemma \ref{la:smooth}.
1661:
1662: %
1663: \begin{lemma} \label{la:Hti}
1664: The continuous function
1665: $\widetilde{H} \colon [0,1] \times \RR^{2n} \ra \RR$ defined by
1666: %
1667: \begin{equation} \label{d:Ht}
1668: \widetilde{H}(t,w) \equiv \widetilde{H}_t(w) \,:=\, g(\left|w\right|)
1669: \widetilde{G}_t (w) .
1670: \end{equation}
1671: %
1672: is smooth in the variable $w \in \RR^{2n}$ and has the following properties.
1673: %
1674: \begin{itemize}
1675: \item[(i)]
1676: $
1677: \widetilde{H} (t,w) = H (t,w) \quad
1678: \text{ for all }\, t \in [0,1] \text{ and } w \in \phi_t(A) .$
1679: %\item[(ii)]
1680: %$\widetilde{H} (t,w) = 0 \quad
1681: % \text{ for all }\, t \le 0 \text{ and } w \in \RR^{2n} .$
1682: \item[(ii)]
1683: There exists a constant $C>0$ such that
1684: %
1685: \begin{equation} \label{ea:Htti2}
1686: \left| \nabla \widetilde{H}_t(w) \right| \,\le\, \frac{C}{t^2} \left(
1687: \left|w\right|+1 \right) \quad
1688: \text{ for all }\, t \in\;]0,1] \text{ and } w \in \RR^{2n} .
1689: \end{equation}
1690: %
1691: \end{itemize}
1692: \end{lemma}
1693:
1694: \proof
1695: Since $g \colon [0, \infty[ \;\ra [1,\infty[$ is smooth and
1696: $\widetilde{G} \colon [0,1] \times \RR^{2n} \ra \RR$ is continuous, the
1697: function $\widetilde{H}$ is indeed continuous, and since \text{$g(r) = 1$ if
1698: $r \le \tfrac{1}{2}$} and $\widetilde{G}$ is smooth in $w$, the function
1699: $\widetilde{H}$ is smooth in $w$.
1700: Assertion (i) follows from the
1701: definition \eqref{d:Ht} of $\widetilde{H}$, from Lemma
1702: \ref{la:smooth}\,(i) and from the definition \eqref{d:G} of $G$.
1703: %and assertion (ii) follows from definition \eqref{d:Ht} and Lemma
1704: %\ref{la:smooth}\,(ii).
1705: In order to verify assertion (ii) we
1706: fix $t \in\;]0,1]$ and $w \in \RR^{2n}$.
1707: Using definition \eqref{d:Ht} we compute
1708: %
1709: \begin{equation} \label{ea:grHti}
1710: \nabla \widetilde{H}_t(w) \,=\, g'(\left| w \right|) \frac{w}{\left| w
1711: \right|} \widetilde{G}_t(w) + g(\left| w \right|) \nabla
1712: \widetilde{G}_t(w) .
1713: \end{equation}
1714: %
1715: Since $0 \in A$ and $\phi_t(0) = 0$ we have, together with equation
1716: \eqref{ea:H0},
1717: \[
1718: \widetilde{G}_t(0) \,=\, G_t(0) \,=\, H_t(0) \,=\, 0 .
1719: \]
1720: This, the mean value theorem and Lemma \ref{la:smooth}\,(ii) yield
1721: %
1722: \begin{equation} \label{ea:G2x}
1723: \left| \widetilde{G}_t(w) \right| \,\le\, \frac{C_6}{t^2} \left| w
1724: \right| \quad \text{ and } \quad \left| \nabla \widetilde{G}_t(w)
1725: \right| \,\le\, \frac{C_6}{t^2} .
1726: \end{equation}
1727: %
1728: Using the identity \eqref{ea:grHti}, the estimates \eqref{ea:G2x} and
1729: the estimates $\left| g'(r) \right| \le 1$ and $g(r) \le r+2$ holding
1730: for all $r \ge 0$ we can estimate
1731: \[
1732: \left| \nabla \widetilde{H}_t (w) \right| \,\le\, \frac{C_6}{t^2}
1733: \left| w \right| + ( \left| w \right| +2) \frac{C_6}{t^2} \,=\,
1734: \frac{2C_6}{t^2} ( \left| w \right| +1 ) .
1735: \]
1736: Setting $C := 2C_6$ assertion (ii) follows.
1737: The proof of Lemma \ref{la:Hti} is complete.
1738: \proofend
1739:
1740: Theorem \ref{ta2:ear} is a consequence of Lemma \ref{la:Hti}:
1741: The time-dependent vector field $\nabla \widetilde{H}_t(w)$ on $[0,1]
1742: \times \RR^{2n}$ is continuous, and since it is smooth in $w$, it is
1743: locally Lipschitz continuous in $w$.
1744: This and assertion (ii) of Lemma \ref{la:Hti} imply that
1745: the Hamiltonian system associated with $\widetilde{H}$ can be
1746: solved for all $t \in [0,1]$.
1747: We define $\Phi_A$ to be the resulting time-$1$-map.
1748: Since $\nabla \widetilde{H}_t(w)$ is continuous and smooth in $w$, the
1749: map $\Phi_A$ is smooth (see \cite[Proposition 9.4]{A}), and so
1750: $\Phi_A$ is a globally defined symplectomorphism of $\RR^{2n}$.
1751: %
1752: Moreover, Lemma \ref{la:Hti}\,(i) shows that $\Phi_A |_A = \ff |_A$.
1753: The proof of \text{Theorem \ref{ta2:ear}} is finally complete.
1754: \proofend
1755:
1756:
1757: \begin{remark}
1758: {\rm
1759: Proceeding as in Step 2 we obtain a {\it smooth}\, Hamiltonian
1760: \[
1761: H_A \colon [0,1] \times \RR^{2n} \,\ra\, \RR
1762: \]
1763: which generates the symplectomorphism $\Phi_A$ and is such that $H_A
1764: |_\ca = \widetilde{H} |_\ca$.
1765: However, $H_A$ might not be $C^0$-close to $\widetilde{H}$, and $\nabla
1766: H_A$ might not be linearly bounded.
1767: }
1768: \end{remark}
1769:
1770:
1771:
1772:
1773: \begin{thebibliography}{99}
1774:
1775:
1776: \bibitem{A} H.\ Amann.
1777: {\it Ordinary differential equations.}.
1778: An introduction to nonlinear analysis.
1779: de Gruyter Studies in Mathematics {\bf 13}.
1780: Walter de Gruyter \& Co., Berlin, 1990.
1781:
1782: \bibitem{EH1} I.\ Ekeland and H.\ Hofer. Symplectic topology and
1783: Hamiltonian dynamics. {\it Math. Z.} {\bf 200} (1990) 355--378.
1784:
1785: \bibitem{Hi} M.\ Hirsch. {\it Differential topology}.
1786: Graduate Texts in Mathematics {\bf 33}.
1787: Springer-Verlag, New York--Heidelberg, 1976.
1788:
1789: \bibitem{MS} D.\ Mc\;\!Duff and D.\ Salamon.
1790: {\it Introduction to Symplectic Topology}.
1791: Oxford Mathematical Monographs, Clarendon Press, 1995.
1792:
1793: \bibitem{Mc} E.\ J.\ Mc\,Shane.
1794: Extension of range of functions.
1795: {\it Bull. Amer. Math. Soc.} {\bf 40} (1934) 837--842.
1796:
1797: \bibitem{Diss} F.\ Schlenk.
1798: {\it Embedding problems in symplectic geometry}.
1799: Diss.\ ETH No.\ 14254.
1800: Z\"urich 2001.
1801:
1802:
1803: \end{thebibliography}
1804:
1805:
1806:
1807: \enddocument
1808: