1: \LaTeXparent{nga4.tex}
2:
3: \section{Test functions on a linear space with measure \label{testfunctions}}
4:
5: In this section we will construct the test function space $({\cal N})^1$ and
6: study its properties. On the space ${\cal P}({\cal N}^{\prime })$ we can
7: define a system of norms using the representation from Lemma \ref{PrepLemma}%
8: . Let
9: $$
10: \varphi =\sum_{n=0}^N\langle P_n^\mu ,\;\varphi ^{(n)}\rangle \in {\cal P}(%
11: {\cal N}^{\prime })
12: $$
13: be given, then $\varphi ^{(n)}\in {\cal H}_{p,\Ckl }^{\hat \otimes n}$ for
14: each $p\geq 0\ \ (n\in \N )$. Thus we may define for any $p,q\in \N $ a
15: Hilbertian norm on ${\cal P}({\cal N}^{\prime })$ by
16: $$
17: \left\| \varphi \right\| _{p,q,\mu }^2=\sum_{n=0}^N(n!)^2\;2^{nq}\;|\varphi
18: ^{(n)}|_p^2
19: $$
20: The completion of ${\cal P}({\cal N}^{\prime })$ w.r.t. $\left\| \cdot
21: \right\| _{p,q,\mu }$ is called $({\cal H}_p)_{q,\mu }^1$ .
22:
23: \begin{definition}
24: We define
25: $$
26: ({\cal N})_\mu ^1:=\ \stackunder{p,q\in \N }{\rm pr\ lim}({\cal H}_p)_{q,\mu
27: }^1\ .
28: $$
29: \end{definition}
30:
31: This space has the following properties
32:
33: \begin{theorem}
34: $({\cal N})_\mu ^1$ is a nuclear space. The topology $({\cal N})_\mu ^1$ is
35: uniquely defined by the topology on ${\cal N}$: It does not depend on the
36: choice of the family of norms $\{|\cdot |_p\}$.
37: \end{theorem}
38:
39: \TeXButton{Proof}{\proof}Nuclearity of $({\cal N})_\mu ^1$ follows
40: essentially from that of ${\cal N}.$ For fixed $p,q$ consider the embedding
41: $$
42: I_{p^{\prime },q^{\prime },p,q}:\left( {\cal H}_{p^{\prime }}\right)
43: _{q^{\prime },\mu }^1\rightarrow \left( {\cal H}_p\right) _{q,\mu }^1
44: $$
45: where $p^{\prime }$ is chosen such that the embedding%
46: $$
47: i_{p^{\prime },p}:\,{\cal H}_{p^{\prime }}\rightarrow {\cal H}_p
48: $$
49: is Hilbert--Schmidt. Then $I_{p^{\prime },q^{\prime },p,q}$ is induced by%
50: $$
51: I_{p^{\prime },q^{\prime },p,q}\varphi =\sum_{n=0}^\infty \langle P_n^\mu
52: ,i_{p^{\prime },p}^{\otimes n}\varphi ^{(n)}\rangle \quad \text{ for \quad }%
53: \varphi =\sum_{n=0}^\infty \langle P_n^\mu ,\varphi ^{(n)}\rangle \in \left(
54: {\cal H}_{p^{\prime }}\right) _{q^{\prime },\mu }^1.
55: $$
56: Its Hilbert--Schmidt norm is easily estimated by using an orthonormal basis
57: of $\left( {\cal H}_{p^{\prime }}\right) _{q^{\prime },\mu }^1$. The result
58: is the bound
59: $$
60: \left\| I_{p^{\prime },q^{\prime },p,q}\right\| _{HS}^2\le \sum_{n=0}^\infty
61: 2^{n(q-q^{\prime })}\left\| i_{p^{\prime },p}\right\| _{HS}^{2n}
62: $$
63: which is finite for suitably chosen $q^{\prime }$.
64:
65: Let us assume that we are given two different systems of Hilbertian norms $%
66: \left| \,\cdot \,\right| _p$ and $\left| \,\cdot \,\right| _k^{\prime }$,
67: such that they induce the same topology on ${\cal N}$ . For fixed $k$ and $l$
68: we have to estimate $\left\| \,\cdot \,\right\| _{k,l,\mu }^{\prime }$ by $%
69: \left\| \,\cdot \,\right\| _{p,q,\mu }$ for some $p,q$ (and vice versa which
70: is completely analogous). Since $\left| \,\cdot \,\right| _k^{\prime }$ has
71: to be continuous with respect to the projective limit topology on ${\cal N}$%
72: , there exists $p$ and a constant $C$ such that $\left| f\right| _k^{\prime
73: }\leq C\left| f\right| _p$, for all $f\in {\cal N}$, i.e., the injection $i$
74: from ${\cal H}_p$ into the completion ${\cal K}_k$ of ${\cal N}$ with
75: respect to $|\,\cdot \,|_k^{\prime }$ is a mapping bounded by $C$. We denote
76: by $i$ also its linear extension from ${\cal H}_{p,{\,}\Ckl }$ into ${\cal K}%
77: _{{\,}k,\Ckl }$. It follows that $i^{\otimes n}$ is bounded by $C^n$ from $%
78: {\cal H}_{{\,}p,\Ckl }^{\otimes n}$ into ${\cal K}_{{\,}k,\Ckl }^{\otimes
79: n}$. Now we choose $q$ such that $2^{{\frac{q-l}2}}\geq C$. Then
80:
81: \begin{eqnarray*}
82: \left\| \,\cdot \,\right\| _{k,l,\mu }^{\prime 2}&=&\sum_{n=0}^\infty %
83: (n!)^2\,2^{nl}\left| \,\cdot \,\right| _k^{\prime 2} %
84: \\&\leq & \sum_{n=0}^\infty (n!)^2\,2^{nl}C^{2n}\left| \,\cdot \,%
85: \right| _p^2 %
86: \\&\leq &\left\| \,\cdot \,\right\| _{p,q,\mu }^2\ ,
87: \end{eqnarray*}
88: which had to be proved.\TeXButton{End Proof}{\endproof}\bigskip\
89:
90: \begin{lemma}
91: \label{L2NormPn}There exist $p,C,K>0$ such that for all $n$
92: \begin{equation}
93: \label{Pnx2norm}\int |P_n^\mu (x)|_{-p}^2\;{\rm d}\mu (x)\leq (n!)^2\,C^n\,K
94: \end{equation}
95: \end{lemma}
96:
97: \TeXButton{Proof}{\proof} The estimate (\ref{Pnxnorm}) may be used for $\rho
98: \leq 2^{-q_0}$ and $\rho \leq 2\varepsilon _\mu $ ($\varepsilon _\mu $ from
99: Lemma \ref{equiLemma}).\\This gives
100: $$
101: \int |P_n^\mu (x)|_{-p}^2\;{\rm d}\mu (x)\leq (n!)^2\left( \frac e\rho
102: \left\| i_{p,p_0}\right\| _{HS}\right) ^{2n}\int e^{2\rho |x|_{-p_0}}{\rm d}%
103: \mu (x)
104: $$
105: which is finite because of Lemma \ref{equiLemma}.\TeXButton{End Proof}
106: {\endproof}
107:
108: \begin{theorem}
109: There exist $p^{\prime },q^{\prime }>0$ such that for all $p\geq p^{\prime
110: },\ q\geq q^{\prime }$ the topological embedding $({\cal H}_p)_{q,\mu
111: }^1\subset L^2(\mu )$ holds.
112: \end{theorem}
113:
114: \TeXButton{Proof}{\proof}Elements of the space $({\cal N})_\mu ^1$ are
115: defined as series convergent in the given topology. Now we need to study the
116: convergence of these series in $L^2(\mu )$. Choose $q^{\prime }$ such that $%
117: C>2^{q^{\prime }\text{ }}$ ($C$ from estimate (\ref{Pnx2norm})). Let us take
118: an arbitrary
119: $$
120: \varphi =\sum_{n=0}^\infty \langle P_n^\mu ,\varphi ^{(n)}\rangle \in {\cal P%
121: }({\cal N}^{\prime })
122: $$
123: For $p>p^{\prime }$ ($p^{\prime }$ as in Lemma \ref{L2NormPn} ) and $%
124: q>q^{\prime }$ the following estimates hold%
125: \begin{eqnarray*}
126: \left\| \varphi \right\| _{L^2(\mu )}%
127: & \leq & \sum_{n=0}^\infty \left\| \langle%
128: P_n^\mu ,\varphi ^{(n)}\rangle \right\| _{L^2(\mu )}%
129: \\&\leq & \sum_{n=0}^\infty |\varphi ^{(n)}|_{-p}\left\| \,|P_n^\mu%
130: |_{-p}\right\| _{L^2(\mu )} %
131: \\&\leq & K\sum_{n=0}^\infty n!\,2^{nq/2}\left| \varphi ^{(n)}\right|%
132: _{-p}(C2^{-q})^{n/2} %
133: \\&\leq & K\left( \sum_{n=0}^\infty (C\,2^{-q})^n\right) ^{\frac 12}\left(%
134: \sum_{n=0}^\infty (n!)^2\,2^{qn}\left| \varphi ^{(n)}\right| _{-p}^2\right)%
135: ^{\frac 12} %
136: \\&=& K\left( 1-C\,2^{-q}\right) ^{-1/2}\left\| \varphi \right\| _{p,q,\mu }%
137: \text{.}%
138: \end{eqnarray*}
139: Taking the closure the inequality extends to the whole space $({\cal H}%
140: _p)_q^1$.\TeXButton{End Proof}{\endproof}
141:
142: \begin{corollary}
143: \label{N1inL2}$({\cal N})_\mu ^1$ is continuously and densely embedded in $%
144: L^2(\mu )$.
145: \end{corollary}
146:
147: \example \newcounter{myexponent} \setcounter{myexponent}{\value{example}}
148: {\it ($\mu $-exponentials as test functions)} \smallskip
149: \\The $\mu $-exponential given in (\ref{Pgenerator}) has the following norm%
150: $$
151: ||e_\mu (\theta ;\cdot )||_{p,q,\mu }^2=\sum_{n=0}^\infty 2^{nq}\,|\theta
152: |_p^{2n}\ ,\qquad \theta \in {\cal N}_{\Ckl }
153: $$
154: This expression is finite if and only if $2^q|\theta |_p^2<1$. Thus we have $%
155: e_\mu (\theta ;\cdot )\notin ({\cal N})_\mu ^1$ if $\theta \neq 0$. But we
156: have that $e_\mu (\theta ;\cdot )$ is a test function of finite order i.e., $%
157: e_\mu (\theta ;\cdot )\in ({\cal H}_p)_q^1$ if $2^q|\theta |_p^2<1$. This is
158: in contrast to some useful spaces of test functions in Gaussian Analysis,
159: see e.g., \cite{BeKo88,HKPS93}.
160:
161: The set of all $\mu $--exponentials $\{e_\mu (\theta ;\cdot )\;|\;2^q|\theta
162: |_p^2<1,\ \theta \in {\cal N}_{\Ckl }\}$ is a total set in $({\cal H}_p)_q^1$%
163: . This can been shown using the relation ${\rm d}^ne_\mu (0;\cdot )(\theta
164: _1,...,\theta _n)=\langle P_n^\mu ,\theta _1\hat \otimes \cdots \hat \otimes
165: \theta _n\rangle .$
166:
167: \begin{proposition}
168: \label{N1inEmin}Any test function $\varphi $ in $({\cal N})_\mu ^1$ has a
169: uniquely defined extension to ${\cal N}_{\Ckl }^{\prime }$ as an element of $%
170: {\cal E}_{\min }^1\left( {\cal N}_{\Ckl }^{\prime }\right) $
171: \end{proposition}
172:
173: \TeXButton{Proof}{\proof}Any element $\varphi $ in $({\cal N})_\mu ^1$ is
174: defined as a series of the following type
175: $$
176: \varphi =\sum_{n=0}^\infty \langle P_n^\mu ,\varphi ^{(n)}\rangle \ ,\qquad
177: \varphi ^{(n)}\in {\cal N}_{\Ckl }^{\hat \otimes n}
178: $$
179: such that%
180: $$
181: \left\| \varphi \right\| _{p,q,\mu }^2=\sum_{n=0}^\infty
182: (n!)^2\,2^{nq}\,|\varphi ^{(n)}|_p^2
183: $$
184: is finite for each $p,q\in \N $ . In this proof we will show the convergence
185: of the series%
186: $$
187: \sum_{n=0}^\infty \langle P_n^\mu (z),\varphi ^{(n)}\rangle ,\quad z\in
188: {\cal H}_{-p,\Ckl }
189: $$
190: to an entire function in $z$.
191:
192: Let $p>p_0$ such that the embedding $i_{p,p_0}:{\cal H}_p\hookrightarrow
193: {\cal H}_{p_0}$ is Hilbert-Schmidt. Then for all $0<\varepsilon \leq
194: 2^{-q_0}/e\left\| i_{p,p_0}\right\| _{HS}$ we can use (\ref{(P5)}) and
195: estimate as follows%
196: \begin{eqnarray*}
197: \sum_{n=0}^\infty |\langle P_n^\mu (z),\varphi ^{(n)}\rangle |%
198: & \leq & \sum_{n=0}^\infty |P_n^\mu (z)|_{-p}|\varphi ^{(n)}|_p %
199: \\& \leq & C_{p,\varepsilon }\,e^{\varepsilon |z|_{-p}}\sum_{n=0}^\infty%
200: n!\,|\varphi ^{(n)}|_p\,\varepsilon ^{-n} %
201: \\& \leq & C_{p,\varepsilon }\,\,e^{\varepsilon |z|_{-p}}\,\left(%
202: \sum_{n=0}^\infty (n!)^22^{nq}|\varphi ^{(n)}|_p^2\right) ^{1/2}\left(%
203: \sum_{n=0}^\infty 2^{-nq}\varepsilon ^{-2n}\right) ^{1/2} %
204: \\&=& C_{p,\varepsilon }\,\,\left( 1-2^{-q}\varepsilon ^{-2}\right)%
205: ^{-1/2}\,\left\| \varphi \right\| _{p,q,\mu }\,\;e^{\varepsilon |z|_{-p}} %
206: \end{eqnarray*}
207: if $2^q>\varepsilon ^{-2}$. That means the series $\sum_{n=0}^\infty \langle
208: P_n^\mu (z),\varphi ^{(n)}\rangle $ converges uniformly and absolutely in
209: any neighborhood of zero of any space ${\cal H}_{-p,\Ckl }$ . Since each
210: term $\langle P_n^\mu (z),\varphi ^{(n)}\rangle $ is entire in $z$ the
211: uniform convergence implies that $z\mapsto \sum_{n=0}^\infty \langle P_n^\mu
212: (z),\varphi ^{(n)}\rangle $ is entire on each ${\cal H}_{-p,\Ckl }$ and
213: hence on ${\cal N}_{\Ckl }^{\prime }$. This completes the proof.%
214: \TeXButton{End Proof}{\endproof}\bigskip\
215:
216: The following corollary is an immediate consequence of the above proof and
217: gives an explicit estimate on the growth of the test functions.
218:
219: \begin{corollary}
220: \label{phi(z)Betrag}For all $p>p_0$ such that the norm $\left\|
221: i_{p,p_0}\right\| _{HS}$ of the embedding is finite and for all $%
222: 0<\varepsilon \leq 2^{-q_0}/e\left\| i_{p,p_0}\right\| _{HS}$ we can choose $%
223: q\in \N $ such that $2^q>\varepsilon ^{-2}$ to obtain the following bound.%
224: $$
225: \left| \varphi (z)\right| \leq C\,\left\| \varphi \right\| _{p,q,\mu
226: }\,e^{\varepsilon |z|_{-p}}\ ,\qquad \varphi \in ({\cal N})_\mu ^1,\ z\in
227: {\cal H}_{-p,\Ckl }\text{ ,}
228: $$
229: where%
230: $$
231: C=C_{p,\varepsilon }\,\left( 1-2^{-q}\varepsilon ^{-2}\right) ^{-1/2}.
232: $$
233: \end{corollary}
234:
235: $\ $Let us look at Proposition \ref{N1inEmin} again. On one hand any
236: function $\varphi \in ({\cal N})_\mu ^1$ can be written in the form
237: \begin{equation}
238: \label{phiPn}\varphi (z)=\sum_{n=0}^\infty \langle P_n^\mu (x),\varphi
239: ^{(n)}\rangle \ ,\qquad \varphi ^{(n)}\in {\cal N}_{\Ckl }^{\hat \otimes n}\
240: ,
241: \end{equation}
242: on the other hand it is entire, i.e., it has the representation
243: \begin{equation}
244: \label{phizn}\varphi (z)=\sum_{n=0}^\infty \langle z^{\otimes n},\tilde
245: \varphi ^{(n)}\rangle \ ,\qquad \tilde \varphi ^{(n)}\in {\cal N}_{\Ckl %
246: }^{\hat \otimes n}\ ,
247: \end{equation}
248: To proceed further we need the explicit correspondence $\left\{ \varphi
249: ^{(n)},n\in \N \right\} \longleftrightarrow \left\{ \tilde \varphi
250: ^{(n)},n\in \N \right\} $ which is given in the next lemma.
251:
252: \begin{lemma}
253: \label{Reordering}{\bf (Reordering)} \smallskip
254: \\Equations (\ref{phiPn}) and (\ref{phizn}) hold iff%
255: $$
256: \tilde \varphi ^{(k)}=\sum_{n=0}^\infty \binom{n+k}k\left( P_n^\mu
257: (0),\varphi ^{(n+k)}\right) _{{\cal H}^{\hat \otimes n}}
258: $$
259: or equivalently%
260: $$
261: \varphi ^{(k)}=\sum_{n=0}^\infty \binom{n+k}k\left( {\rm M}_n^\mu ,\tilde
262: \varphi ^{(n+k)}\right) _{{\cal H}^{\hat \otimes n}}
263: $$
264: where $\left( P_n^\mu (0),\varphi ^{(n+k)}\right) _{{\cal H}^{\hat \otimes
265: n}}$ and $\left( {\rm M}_n^\mu ,\tilde \varphi ^{(n+k)}\right) _{{\cal H}%
266: ^{\hat \otimes n}}$ denote contractions defined by (\ref{contraction}).
267: \end{lemma}
268:
269: \noindent This is a consequence of (\ref{(P1)}) and (\ref{(P2)}). We omit
270: the simple proof.\bigskip\ \
271:
272: Proposition \ref{N1inEmin} states%
273: $$
274: ({\cal N})_\mu ^1\subseteq {\cal E}_{\min }^1({\cal N}^{\prime })
275: $$
276: as sets, where%
277: $$
278: {\cal E}_{\min }^1({\cal N}^{\prime })=\left\{ \varphi |_{{\cal N}^{\prime
279: }}\;\Big| \;\varphi \in {\cal E}_{\min }^1({\cal N}_{\Ckl }^{\prime
280: })\right\} \ .
281: $$
282: Corollary \ref{phi(z)Betrag} then implies that the embedding is also
283: continuous. Now we are going to show that the converse also holds.
284:
285: \begin{theorem}
286: \label{N1E1min}For all measures $\mu \in {\cal M}_a({\cal N}^{\prime })$ we
287: have the topological identity%
288: $$
289: ({\cal N})_\mu ^1={\cal E}_{\min }^1({\cal N}^{\prime })\ .
290: $$
291: \end{theorem}
292:
293: \noindent To prove the missing topological inclusion it is convenient to use
294: the nuclear topology on ${\cal E}_{\min }^1({\cal N}_{\Ckl }^{\prime })$
295: (given by the norms $\lnorm \cdot \rnorm _{{p,q,1}}$) introduced in section
296: \ref{Preliminaries}. Theorem \ref{Ekminprlim} ensures that this topology is
297: equivalent to the projective topology induced by the norms ${\rm n}_{p,l,k}$%
298: . Then the above theorem is an immediate consequence of the following norm
299: estimate.
300:
301: \begin{proposition}
302: Let $p>p_\mu $ ($p_\mu $ as in Lemma \ref{equiLemma}) such that $\left\|
303: i_{p,p_\mu }\right\| _{HS}$ is finite and $q\in \N $ such that $2^{q/2}>K_p$
304: ($K_p:=eC\left\| i_{p,p_\mu }\right\| _{HS}$ as in (\ref{MnmuNorm})). For
305: any $\varphi \in {\rm E}_{p,q}^1$ the restriction $\varphi |_{{\cal N}%
306: ^{\prime }}$ is a function from $({\cal H}_p)_{q^{\prime },\mu }^1\ ,\
307: q^{\prime }<q$. Moreover the following estimate holds%
308: $$
309: ||\varphi ||_{p,q^{\prime },\mu }\leq \lnorm \varphi \rnorm
310: _{p,q,1}(1-2^{-q/2}K_p)^{-1}(1-2^{q^{\prime }-q})^{-1/2}\ .
311: $$
312: \end{proposition}
313:
314: \TeXButton{Proof}{\proof}Let $p,q\in \N $, $K_p$ be defined as above. A
315: function $\varphi \in {\rm E}_{p,q}^1$ has the representation (\ref{phizn}).
316: Using the Reordering lemma combined with (\ref{MnmuNorm}) and
317: $$
318: \left| \tilde \varphi ^{(n)}\right| _p\leq \frac 1{n!}\,2^{-nq/2}\lnorm
319: \varphi \rnorm _{p,q,1}
320: $$
321: we obtain a representation of the form (\ref{phiPn}) where
322: \begin{eqnarray*}
323: \left| \varphi ^{(n)}\right| _p & \leq & \sum_{k=0}^\infty \binom{n+k}k\left|%
324: {\rm M}_k^\mu \right| _{-p}\left| \tilde \varphi ^{(n+k)}\right| _p %
325: \\&\leq & \lnorm \varphi \rnorm _{p,q,1}\sum_{k=0}^\infty %
326: \binom{n+k}k\frac{k!}{(n+k)!}K_p^k\,2^{-(n+k)q/2} %
327: \\& \leq & \lnorm \varphi \rnorm _{p,q,1}%
328: \frac1{n!}2^{-nq/2}\sum_{k=0}^\infty (2^{-q/2}K_p)^k %
329: \\& \leq & \lnorm \varphi \rnorm _{p,q,1}%
330: \frac1{n!}2^{-nq/2}(1-2^{-q/2}K_p)^{-1} .%
331: \end{eqnarray*}
332: For $q^{\prime }<q$ this allows the following estimate%
333: \begin{eqnarray*}
334: ||\varphi ||_{p,q^{\prime },\mu }^2%
335: &=&\sum_{n=0}^\infty (n!)^2\,2^{q^{\prime}n}\,|\varphi ^{(n)}|_p^2 %
336: \\& \leq & \lnorm \varphi \rnorm ^2%
337: _{p,q,1}(1-2^{-q/2}K_p)^{-2}\sum_{k=0}^\infty 2^{n(q^{\prime }-q)}<\infty%
338: \end{eqnarray*}
339: This completes the proof.\TeXButton{End Proof}{\endproof}\bigskip\
340:
341: Since we now have proved that the space of test functions $({\cal N})_\mu ^1$
342: is isomorphic to ${\cal E}_{\min }^1({\cal N}^{\prime })$ for all measures $%
343: \mu \in {\cal M}_a({\cal N}^{\prime })$, we will now drop the subscript $\mu
344: $. The test function space $({\cal N})^1$ is the same for all measures $\mu
345: \in {\cal M}_a({\cal N}^{\prime })$.
346:
347: \begin{corollary}
348: $({\cal N})^1$ is an algebra under pointwise multiplication.
349: \end{corollary}
350:
351: \begin{corollary}
352: $({\cal N})^1$ admits `scaling' i.e., for $\lambda \in \C $ the scaling
353: operator $\sigma _\lambda :({\cal N})^1\rightarrow ({\cal N})^1$ defined by $%
354: \sigma _\lambda \varphi (x):=\varphi (\lambda x)$, $\varphi \in ({\cal N})^1$%
355: , $x\in {\cal N}^{\prime }$ is well--defined.
356: \end{corollary}
357:
358: \begin{corollary}
359: For all $z\in {\cal N}_{\Ckl }^{\prime }$ the space $({\cal N})^1$ is
360: invariant under the shift operator $\tau _z:\varphi \mapsto \varphi (\cdot
361: +z)$.
362: \end{corollary}
363:
364: \section{Distributions\label{Distributions}}
365:
366: In this section we will introduce and study the space $({\cal N})_\mu ^{-1}$
367: of distributions corresponding to the space of test functions $({\cal N})^1$%
368: . Since ${\cal P}({\cal N}^{\prime })\subset ({\cal N})^1$ the space $({\cal %
369: N})_\mu ^{-1}$ can be viewed as a subspace of ${\cal P}_\mu ^{\prime }({\cal %
370: N}^{\prime })$%
371: $$
372: ({\cal N})_\mu ^{-1}\subset {\cal P}_\mu ^{\prime }({\cal N}^{\prime })
373: $$
374: Let us now introduce the Hilbertian subspace $({\cal H}_{-p})_{-q,\mu }^{-1}$
375: of ${\cal P}_\mu ^{\prime }({\cal N}^{\prime })$ for which the norm
376: $$
377: \left\| \Phi \right\| _{-p,-q,\mu }^2:=\sum_{n=0}^\infty 2^{-qn}\left| \Phi
378: ^{(n)}\right| _{-p}^2\text{ }
379: $$
380: is finite. Here we used the canonical representation%
381: $$
382: \Phi =\sum_{n=0}^\infty Q_n^\mu (\Phi ^{(n)})\in {\cal P}_\mu ^{\prime }(%
383: {\cal N}^{\prime })\text{ }
384: $$
385: from Theorem \ref{PStrichRep}. The space $({\cal H}_{-p})_{-q,\mu }^{-1}$ is
386: the dual space of $({\cal H}_p)_q^1$ with respect to $L^2(\mu )$ (because of
387: the biorthogonality of $\p -$and $\Q -$systems). By general duality theory
388: $$
389: ({\cal N})_\mu ^{-1}:=\bigcup_{p,q\in \N }({\cal H}_{-p})_{-q,\mu }^{-1}
390: $$
391: is the dual space of $({\cal N})^1$ with respect to $L^2(\mu )$. As we noted
392: in section \ref{Preliminaries} there exists a natural topology on co-nuclear
393: spaces (which coincides with the inductive limit topology). We will consider
394: $({\cal N})_\mu ^{-1}$ as a topological vector space with this topology. So
395: we have the nuclear triple%
396: $$
397: ({\cal N})^1\subset L^2(\mu )\subset ({\cal N})_\mu ^{-1}\ .
398: $$
399: The action of $\Phi =\sum_{n=0}^\infty Q_n^\mu (\Phi ^{(n)})\in ({\cal N}%
400: )_\mu ^{-1}$ on a test function $\varphi =\sum_{n=0}^\infty \langle P_n^\mu
401: ,\varphi ^{(n)}\rangle \in ({\cal N})^1$ is given by
402: $$
403: \langle \!\langle \Phi ,\varphi \rangle \!\rangle _\mu =\sum_{n=0}^\infty
404: n!\langle \Phi ^{(n)},\varphi ^{(n)}\rangle \ .
405: $$
406: \bigskip\
407:
408: For a more detailed characterization of the singularity of distributions in $%
409: ({\cal N})_\mu ^{-1}$ we will introduce some subspaces in this distribution
410: space. For $\beta \in [0,1]$ we define%
411: $$
412: ({\cal H}_{-p})_{-q,\mu }^{-\beta }=\left\{ \Phi \in {\cal P}_\mu ^{\prime }(%
413: {\cal N}^{\prime })\;\bigg|\;\sum_{n=0}^\infty (n!)^{1-\beta }2^{-qn}\left|
414: \Phi ^{(n)}\right| _{-p}^2<\infty \text{ for }\Phi =\sum_{n=0}^\infty
415: Q_n^\mu (\Phi ^{(n)})\right\}
416: $$
417: and
418: $$
419: ({\cal N})_\mu ^{-\beta }=\stackunder{p,q\in \N }{\bigcup }({\cal H}%
420: _{-p})_{-q,\mu }^{-\beta }\ ,
421: $$
422: It is clear that the singularity increases with increasing $\beta $:%
423: $$
424: ({\cal N})^{-0}\subset ({\cal N})^{-\beta _1}\subset ({\cal N})^{-\beta
425: _2}\subset ({\cal N})^{-1}
426: $$
427: if $\beta _1\leq \beta _2$.We will also consider $({\cal N})_\mu ^\beta $ as
428: equipped with the natural topology.\bigskip\
429:
430: \example
431: \newcounter{RadonNy} \setcounter{RadonNy}{\value{example}}{\it (Generalized
432: Radon--Nikodym derivative)} \smallskip
433: \\We want to define a generalized function $\rho _\mu (z,\cdot )\in ({\cal N}%
434: )_\mu ^{-1}\ $, $z\in {\cal N}_{\Ckl }^{\prime }$ with the following property%
435: $$
436: \langle \!\langle \rho _\mu (z,\cdot ),\varphi \rangle \!\rangle _\mu =\int_{%
437: {\cal N}^{\prime }}\varphi (x-z)\;{\rm d}\mu (x)\ ,\qquad \varphi \in ({\cal %
438: N})^1\ .
439: $$
440: That means we have to establish the continuity of $\rho _\mu (z,\cdot )$.
441: Let $z\in {\cal H}_{-p,\Ckl }$.\ If $p^{\prime }\geq p$ is sufficiently
442: large and $\varepsilon >0$ small enough, Corollary \ref{phi(z)Betrag}
443: applies i.e., $\exists q\in \N $ and $C>0$ such that%
444: \begin{eqnarray*}
445: \left| \int_{{\cal N}^{\prime }}\varphi (x-z){\rm d}\mu (x)\ \right| %
446: & \leq & C\left\| \varphi \right\| _{p^{\prime },q,\mu }%
447: \int_{{\cal N}^{\prime}}e^{\varepsilon |x-z|_{-p^{\prime }}}{\rm d}\mu (x) %
448: \\& \leq & C\left\| \varphi \right\| _{p^{\prime },q,\mu }e^{\varepsilon |z|_{-p^{\prime }}}\int_{{\cal N}^{\prime }}e^{\varepsilon |x|_{-p^{\prime }}}{\rm d}\mu (x) %
449: \end{eqnarray*}
450: If $\varepsilon $ is chosen sufficiently small the last integral exists.
451: Thus we have in fact $\rho (z,\cdot )\in ({\cal N})_\mu ^{-1}$. It is clear
452: that whenever the Radon--Nikodym derivative $\frac{{\rm d}\mu (x+\xi )}{{\rm %
453: d}\mu (x)}$ exists (e.g., $\xi \in {\cal N}$ in case $\mu $ is ${\cal N}$%
454: -quasi-invariant) it coincides with $\rho _\mu (\xi ,\cdot )$ defined above.
455: We will now show that in $({\cal N})_\mu ^{-1}$ we have the canonical
456: expansion%
457: $$
458: \rho _\mu (z,\cdot )=\sum_{n=0}^\infty \frac 1{n!}(-1)^nQ_n^\mu (z^{\otimes
459: n}).
460: $$
461: It is easy to see that the r.h.s. defines an element in $({\cal N})_\mu
462: ^{-1} $. Since both sides are in $({\cal N})_\mu ^{-1}$ it is sufficient to
463: compare their action on a total set from $({\cal N})^1$. For $\varphi
464: ^{(n)}\in {\cal N}_{\Ckl }^{\hat \otimes n}$ we have
465: \begin{eqnarray*}
466: \left\langle \!\!\left\langle \rho _\mu (z,\cdot ),\langle P_n^\mu ,\varphi%
467: ^{(n)}\rangle \right\rangle \!\!\right\rangle _\mu %
468: &=& \int_{{\cal N}^{\prime }}\langle P_n^\mu (x-z),\varphi ^{(n)}\rangle%
469: \;{\rm d}\mu (x) %
470: \\&=& \sum_{k=0}^\infty \binom nk(-1)^{n-k}\int_{{\cal N}^{\prime }}\langle %
471: P_k^\mu (x)\hat \otimes z^{\otimes n-k},\varphi ^{(n)}\rangle \;{\rm d}\mu %
472: (x) \\&=& (-1)^n\langle z^{\otimes n},\varphi ^{(n)}\rangle %
473: \\&=& \left\langle \!\!\left\langle \sum_{k=0}^\infty \frac 1{k!}(-1)^kQ_k^\mu %
474: (z^{\otimes k}),\langle P_n^\mu ,\varphi ^{(n)}\rangle \right\rangle%
475: \!\!\right\rangle _\mu \ ,%
476: \end{eqnarray*}
477: where we have used (\ref{(P3)}), (\ref{(P4)}) and the biorthogonality of $%
478: \p
479: $- and $\Q $-systems. This had to be shown. In other words, we have proven
480: that $\rho _\mu (-z,\cdot )$ is the generating function of the $\Q $%
481: -functions
482: \begin{equation}
483: \label{rhomyQn}\rho _\mu (-z,\cdot )=\sum_{n=0}^\infty \frac 1{n!}Q_n^\mu
484: (z^{\otimes n})\ .
485: \end{equation}
486: Let use finally remark that the above expansion allows for more detailed
487: estimates. It is easy to see that $\rho _\mu \in ({\cal N})_\mu ^{-0}$.%
488: \bigskip\
489:
490: \example {\it (Delta distribution)} \smallskip \\For $z\in {\cal N}_{\Ckl %
491: }^{\prime }$ we define a distribution by the following $\Q $-decomposition:%
492: $$
493: \delta _z=\sum_{n=0}^\infty \frac 1{n!}Q_n^\mu (P_n^\mu (z))
494: $$
495: If $p\in \N $ is large enough and $\varepsilon >0$ sufficiently small there
496: exists $C_{p,\varepsilon }>0$ according to (\ref{(P5)}) such that
497: \begin{eqnarray*}
498: \left\| \delta _z\right\| _{-p,-q,\mu }^2 &=&\sum_{n=0}^\infty %
499: (n!)^{-2}2^{-nq}\left| P_n^\mu (z)\right| _{-p}^2 %
500: \\& \leq & C_{p,\varepsilon }^2\,e^{2\varepsilon |z|_{-p}}\sum_{n=0}^\infty%
501: 2^{-nq}\varepsilon ^{-2n}\ ,\qquad z\in {\cal H}_{-p,\Ckl }\;, %
502: \end{eqnarray*}
503: which is finite for sufficiently large $q\in \N $. Thus $\delta _z\in ({\cal %
504: N})_\mu ^{-1}$.
505:
506: For $\varphi =\sum_{n=0}^\infty \langle P_n^\mu ,\varphi ^{(n)}\rangle \in (%
507: {\cal N})^1$ the action of $\delta _z$ is given by
508: $$
509: \langle \!\langle \delta _z,\varphi \rangle \!\rangle _\mu
510: =\sum_{n=0}^\infty \langle P_n^\mu (z),\varphi ^{(n)}\rangle =\varphi (z)
511: $$
512: because of (\ref{QnPnPair}). This means that $\delta _z$ (in particular for $%
513: z$ real) plays the role of a ``$\delta $-function" (evaluation map) in the
514: calculus we discuss.
515:
516: \section{Integral transformations}
517:
518: \begin{sloppypar}
519: We will first introduce the Laplace transform of a function $\varphi \in
520: L^2(\mu )$. The global assumption $\mu \in {\cal M}_a({\cal N}^{\prime })$
521: guarantees the existence of $p_\mu ^{\prime }\in \N \ $, $\varepsilon _\mu
522: >0 $ such that $\int_{{\cal N}^{\prime }}\exp (\varepsilon _\mu |x|_{-p_\mu
523: ^{\prime }})\,{\rm d}\mu (x)<\infty $ by Lemma \ref{equiLemma}. Thus $\exp
524: (\langle x,\theta \rangle )\in L^2(\mu )$ if $2|\theta |_{p_\mu ^{\prime
525: }}\leq \varepsilon _\mu \ ,\theta \in {\cal H}_{p_\mu ^{\prime },\Ckl }$.
526: Then by Cauchy--Schwarz inequality the Laplace transform defined by
527: $$
528: L_\mu \varphi (\theta ):=\int_{{\cal N}^{\prime }}\varphi (x)\exp \langle
529: x,\theta \rangle \,{\rm d}\mu (x)
530: $$
531: is well defined for $\varphi \in L^2(\mu )\ ,\theta \in {\cal H}_{p_\mu
532: ^{\prime },\Ckl }$ with $2|\theta |_{p_\mu ^{\prime }}\leq \varepsilon _\mu $%
533: . Now we are interested to extend this integral transform from $L^2(\mu )$
534: to the space of distributions $({\cal N})_\mu ^{-1}$.
535: \end{sloppypar}
536:
537: Since our construction of test function and distribution spaces is closely
538: related to $\p $- and $\Q $-systems it is useful to introduce the so called $%
539: S_\mu $-transform
540: $$
541: S_\mu \varphi (\theta ):=\frac{L_\mu \varphi (\theta )}{l_\mu (\theta )}\ .
542: $$
543: Since $e_\mu (\theta ;x)=e^{\langle x,\theta \rangle }/l_\mu (\theta )$ we
544: may also write%
545: $$
546: S_\mu \varphi (\theta )=\int_{{\cal N}^{\prime }}\varphi (x)\,e_\mu (\theta
547: ;x)\,{\rm d}\mu (x)\ .
548: $$
549: The $\mu $-exponential $e_\mu (\theta ,\cdot )$ is not a test function in $(%
550: {\cal N})^1$, see Example \arabic{myexponent} . So the definition of the $%
551: S_\mu $-transform of a distribution $\Phi \in ({\cal N})_\mu ^{-1}$ must be
552: more careful. Every such $\Phi $ is of finite order i.e., $\exists p,q\in
553: \N
554: $ such that $\Phi \in ({\cal H}_{-p})_{-q,\mu .}^{-1}$ As shown in Example
555: \arabic{myexponent} $e_\mu (\theta ,\cdot )$ is in the corresponding dual
556: space $({\cal H}_p)_{q,\mu }^1$ if $\theta \in {\cal H}_{p,\Ckl }$ is such
557: that $2^q|\theta |_p^2<1$. Then we can define a consistent extension of $%
558: S_\mu $-transform.%
559: $$
560: S_\mu \Phi (\theta ):=\langle \!\langle \Phi ,\;e_\mu (\theta ,\cdot
561: )\rangle \!\rangle _\mu
562: $$
563: if $\theta $ is chosen in the above way. The biorthogonality of $\p $- and $%
564: \Q $-system implies
565: $$
566: S_\mu \Phi (\theta )=\sum_{n=0}^\infty \langle \Phi ^{(n)},\theta ^{\otimes
567: n}\rangle \ .
568: $$
569: It is easy to see that the series converges uniformly and absolutely on any
570: closed ball $\left\{ \left. \theta \in {\cal H}_{p,\Ckl }\right| \;|\theta
571: |_p^2\leq r,\ r<2^{-q}\right\} $, see the proof of Theorem \ref{CharTh}.
572: Thus $S_\mu \Phi $ is holomorphic a neighborhood of zero, i.e., $S_\mu \Phi
573: \in {\rm Hol}_0({\cal N}_{\Ckl })$. In the next section we will discuss this
574: relation to the theory of holomorphic functions in more detail.
575:
576: The third integral transform we are going to introduce is more appropriate
577: for the test function space $({\cal N})^1$. We introduce the convolution of
578: a function $\varphi \in ({\cal N})^1$ with the measure $\mu $ by%
579: $$
580: C_\mu \varphi (y):=\int_{{\cal N}^{\prime }}\varphi (x+y)\,{\rm d}\mu (x) ,
581: \quad y\in {\cal N}^{\prime } .%
582: $$
583: From Example \arabic{RadonNy} the existence of a generalized Radon--Nikodym
584: derivative $\rho _\mu (z,\cdot )$, $z\in {\cal N}_{\Ckl }^{\prime }$ in $(%
585: {\cal N})_\mu ^{-1}$ is guaranteed. So for any $\varphi \in ({\cal N})^1$, $%
586: z\in {\cal N}_{\Ckl }^{\prime }$ the convolution has the representation
587: $$
588: C_\mu \varphi (z)=\langle \!\langle \rho _\mu (-z,\cdot ),\;\varphi \rangle
589: \!\rangle _\mu \;.
590: $$
591: If $\varphi \in ({\cal N})^1$ has the canonical representation%
592: $$
593: \varphi =\sum_{n=0}^\infty \langle P_n^\mu ,\;\varphi ^{(n)}\rangle
594: $$
595: we have by equation (\ref{rhomyQn})
596: $$
597: C_\mu \varphi (z)=\sum_{n=0}^\infty \langle z^{\otimes n},\varphi
598: ^{(n)}\rangle \ .
599: $$
600:
601: In Gaussian Analysis $C_\mu $- and $S_\mu$-transform coincide. It is a
602: typical non-Gaussian effect that these two transformations differ from each
603: other.
604:
605: \section{Characterization theorems \label{Characterization}}
606:
607: Gaussian Analysis has shown that for applications it is very useful to
608: characterize test and distribution spaces by the integral transforms
609: introduced in the previous section. In the non-Gaussian setting first
610: results in this direction have been obtained by \cite{AKS93,ADKS94}.%
611: \bigskip
612:
613: We will start to characterize the space $({\cal N})^1$ in terms of the
614: convolution $C_\mu $.
615:
616: \begin{theorem}
617: \label{CmuChar} The convolution $C_\mu $ is a topological isomorphism from $(%
618: {\cal N})^1$ on ${\cal E}_{\min }^1({\cal N}_{\Ckl }^{\prime })$.
619: \end{theorem}
620:
621: \TeXButton{Remark }{\remark } Since we have identified $({\cal N})^1$ and $%
622: {\cal E}_{\min }^1({\cal N}^{\prime })$ by Theorem \ref{N1E1min} the above
623: assertion can be restated as follows. We have
624: $$
625: C_\mu :{\cal E}_{\min }^1({\cal N}^{\prime })\rightarrow {\cal E}_{\min }^1(%
626: {\cal N}_{\Ckl }^{\prime })
627: $$
628: as a topological isomorphism.
629:
630: \TeXButton{Proof}{\proof}The proof has been well prepared by Theorem \ref
631: {Ekminprlim}, because the nuclear topology on ${\cal E}_{\min }^1({\cal N}_{%
632: \Ckl }^{\prime })$ is the most natural one from the point of view of the
633: above theorem. Let $\varphi \in ({\cal N})^1$ with the representation%
634: $$
635: \varphi =\sum_{n=0}^\infty \langle P_n^\mu ,\varphi ^{(n)}\rangle \ .
636: $$
637: From the previous section it follows%
638: $$
639: C_\mu \varphi (z)=\sum_{n=0}^\infty \langle z^{\otimes n},\varphi
640: ^{(n)}\rangle \
641: $$
642: It is obvious from (\ref{3StrichNorm}) that%
643: $$
644: \lnorm C_\mu \varphi \rnorm _{p,q,1}=\left\| \varphi \right\| _{p,q,\mu }\
645: $$
646: for all $p,q\in \N _0$, which proves the continuity of
647: $$
648: C_\mu :({\cal N})^1\rightarrow {\cal E}_{\min }^1({\cal N}_{\Ckl }^{\prime
649: })\ .
650: $$
651:
652: Conversely let $F\in {\cal E}_{\min }^1({\cal N}_{\Ckl }^{\prime })$. Then
653: Theorem \ref{Ekminprlim} ensures the existence of a sequence of generalized
654: kernels $\left\{ \varphi ^{(n)}\in {\cal N}_{\Ckl }^{\prime }\;|\;n\in \N %
655: _0\right\} $ such that
656: $$
657: F(z)=\sum_{n=0}^\infty \langle z^{\otimes n},\varphi ^{(n)}\rangle \ .
658: $$
659: Moreover for all $p,q\in \N _0$%
660: $$
661: \lnorm F\rnorm _{p,q,1}^2=\sum_{n=0}^\infty (n!)^2\,2^{nq}\left| \varphi
662: ^{(n)}\right| _p^2
663: $$
664: is finite. Choosing%
665: $$
666: \varphi =\sum_{n=0}^\infty \langle P_n^\mu ,\varphi ^{(n)}\rangle
667: $$
668: we have $\left\| \varphi \right\| _{p,q,\mu }=\lnorm F\rnorm
669: _{p,q,1}$. Thus $\varphi \in ({\cal N})^1$. Since $C_\mu \varphi =F$ we have
670: shown the existence and continuity of the inverse of $C_\mu $.%
671: \TeXButton{End Proof}{\endproof}\bigskip\
672:
673: To illustrate the above theorem in terms of the natural topology on ${\cal E}%
674: _{\min }^1({\cal N}_{\Ckl }^{\prime })$ we will reformulate the above
675: theorem and add some useful estimates which relate growth in ${\cal E}_{\min
676: }^1({\cal N}_{\Ckl }^{\prime })$ to norms on $({\cal N})^1$.
677:
678: \begin{corollary}
679: \hfill \\1) Let $\varphi \in ({\cal N})^1$ then for all $p,l\in \N _0$ and $%
680: z\in {\cal H}_{-p,\Ckl }$ the following estimate holds%
681: $$
682: \left| C_\mu \varphi (z)\right| \leq \left\| \varphi \right\| _{p,2l,\mu
683: }\exp (2^{-l}|z|_{-p})
684: $$
685: i.e., C$_\mu \varphi \in {\cal E}_{\min }^1({\cal N}_{\Ckl }^{\prime })$.%
686: \medskip\ \\2) Let $F\in {\cal E}_{\min }^1({\cal N}_{\Ckl }^{\prime })$.
687: Then there exists $\varphi \in ({\cal N})^1$ with $C_\mu \varphi =F$. The
688: estimate
689: $$
690: \left| F(z)\right| \leq C\exp (2^{-l}|z|_{-p})
691: $$
692: for $C>0,\ p,q\in \N _0$ implies%
693: $$
694: \left\| \varphi \right\| _{p^{\prime },q,\mu }\leq C\left(
695: 1-2^{q-2l}e^2\left\| i_{p^{\prime },p}\right\| _{HS}^2\right) ^{-1/2}
696: $$
697: if the embedding $i_{p^{\prime },p}:{\cal H}_{p^{\prime }}\hookrightarrow
698: {\cal H}_p$ is Hilbert-Schmidt and $2^{l-q/2}>e\left\| i_{p^{\prime
699: },p}\right\| _{HS}$.
700: \end{corollary}
701:
702: \TeXButton{Proof}{\proof}The first statement follows from
703: $$
704: \left| C_\mu \varphi (z)\right| \leq {\rm n}_{p,l,1}(C_\mu \varphi )\cdot
705: \exp (2^{-l}|z|_{-p})
706: $$
707: which follows from the definition of {\rm n}$_{p,l,1}$ and estimate (\ref
708: {nplk3StrichNorm}). The second statement is an immediate consequence of
709: Lemma \ref{3Strichnplk}. \TeXButton{End Proof}{\endproof}\bigskip\
710:
711: The next theorem characterizes distributions from $({\cal N})_\mu ^{-1}$ in
712: terms of $S_\mu$-transform.
713:
714: \begin{theorem}
715: \label{CharTh}The $S_\mu $-transform is a topological isomorphism from $(%
716: {\cal N})_\mu ^{-1}$ on ${\rm Hol}_0({\cal N}_{\Ckl })$.
717: \end{theorem}
718:
719: \TeXButton{Remark }{\remark } The above theorem is closely related to the
720: second part of Theorem \ref{indlimEkmax}. Since we left the proof open we
721: will give a detailed proof here.
722:
723: \TeXButton{Proof}{\proof}Let $\Phi \in ({\cal N})_\mu ^{-1}$ . Then there
724: exists $p,q\in \N $ such that
725: $$
726: \left\| \Phi \right\| _{-p,-q,\mu }^2=\sum_{n=0}^\infty 2^{-nq}|\Phi
727: ^{(n)}|_{-p}^2
728: $$
729: is finite. From the previous section we have
730: \begin{equation}
731: \label{SPhitheta}S_\mu \Phi (\theta )=\sum_{n=0}^\infty \langle \Phi
732: ^{(n)},\theta ^{\otimes n}\rangle \;.
733: \end{equation}
734: For $\theta \in {\cal N}_{\Ckl }$ such that $2^q|\theta |_p^2<1$ we have by
735: definition (Formula (\ref{3StrichNorm}))%
736: $$
737: \lnorm S_\mu \Phi \rnorm _{-p,-q,-1}=\left\| \Phi \right\| _{-p,-q,\mu \
738: }.
739: $$
740: By Cauchy--Schwarz inequality
741: \begin{eqnarray*}
742: \left| S_\mu \Phi (\theta )\right| & \leq & \sum_{n=0}^\infty |\Phi %
743: ^{(n)}|_{-p}|\theta |_p^n %
744: \\&\leq & \left( \sum_{n=0}^\infty 2^{-nq}|\Phi ^{(n)}|_{-p}^2\right)%
745: ^{1/2}\left( \sum_{n=0}^\infty 2^{nq}|\theta |_p^{2n}\right) ^{1/2} %
746: \\&=& \left\| \Phi \right\| _{-p,-q,\mu }\left( 1-2^q|\theta |_p^2\right)%
747: ^{-1/2}\ .
748: \end{eqnarray*}
749: Thus the series (\ref{SPhitheta}) converges uniformly on any closed ball $%
750: \left\{ \left. \theta \in {\cal H}_{p,\Ckl }\right| \;|\theta |_p^2\leq r,\
751: r<2^{-q}\right\} $. Hence $S_\mu \Phi \in {\rm Hol}_0({\cal N}_{\Ckl })$ and
752: $$
753: {\rm n}_{p,l,\infty }(S_\mu \Phi )\leq \left\| \Phi \right\| _{-p,-q,\mu
754: }(1-2^{q-2l})^{-1/2}
755: $$
756: if $2l>q$. This proves that $S_\mu $ is a continuous mapping from $({\cal N}%
757: )_\mu ^{-1}$ to ${\rm Hol}_0({\cal N}_{\Ckl })$. In the language of section
758: \ref{Holomorphy} this reads%
759: $$
760: \stackunder{p,q\in \N }{\rm ind\ lim}\,{\rm E}_{-p,-q}^{-1}\subset {\rm Hol}%
761: _0({\cal N}_{\Ckl })
762: $$
763: topologically.\medskip\
764:
765: Conversely, let $F\in {\rm Hol}_0({\cal N}_{\Ckl })$ be given, i.e., there
766: exist $p,l\in \N $ such that {\rm n}$_{p,l,\infty }(F)<\infty $. The first
767: step is to show that there exists $p^{\prime },q\in \N $ such that
768: $$
769: \lnorm F\rnorm _{-p^{\prime },-q,-1}<{\rm n}_{p,l,\infty }(F)\cdot C\ ,
770: $$
771: for sufficiently large $C>0$. This implies immediately
772: $$
773: {\rm Hol}_0({\cal N}_{\Ckl })\subset \ \stackunder{p,q\in \N }{\rm ind\ lim}%
774: \,{\rm E}_{-p,-q}^{-1}
775: $$
776: topologically, which is the missing part in the proof of the second
777: statement in Theorem \ref{indlimEkmax}.
778:
779: By assumption the Taylor expansion%
780: $$
781: F(\theta )=\sum_{n=0}^\infty \frac 1{n!}\widehat{{\rm d}^nF(0)}(\theta )
782: $$
783: converges uniformly on any closed ball $\left\{ \left. \theta \in {\cal H}%
784: _{p,\Ckl }\right| \;|\theta |_p^2\leq r,\ r<2^{-l}\right\} $ and
785: $$
786: \left| F(\theta )\right| \leq {\rm n}_{p,l,\infty }(F)\ .
787: $$
788: Proceeding analogously to Lemma \ref{nplk3Strich}, an application of
789: Cauchy's inequality gives%
790: \begin{eqnarray*}
791: \frac 1{n!}\widehat{{\rm d}^nF(0)}(\theta ) & \leq & 2^l|\theta |_p^n\sup %
792: _{|\theta |_p\leq 2^{-l}}|F(\theta )| %
793: \\& \leq & {\rm n}_{p,l,\infty }(F)\ \cdot 2^{nl}\cdot |\theta |_p^n
794: \end{eqnarray*}
795: The polarization identity gives
796: $$
797: \left| \frac 1{n!}{\rm d}^nF(0)(\theta _1,\ldots ,\theta _n)\right| \leq
798: {\rm n}_{p,l,\infty }(F)\ \cdot e^n\cdot 2^{nl}\prod_{j=1}^n|\theta _j|_p
799: $$
800: Then by kernel theorem (Theorem \ref{KernelTh}) there exist kernels $\Phi
801: ^{(n)}\in {\cal H}_{-p^{\prime },\Ckl }^{\hat \otimes n}$ for $p^{\prime }>p$
802: with $\left\| i_{p^{\prime },p}\right\| _{HS}<\infty $ such that%
803: $$
804: F(\theta )=\sum_{n=0}^\infty \langle \Phi ^{(n)},\theta ^{\otimes n}\rangle
805: \ .
806: $$
807: Moreover we have the following norm estimate
808: $$
809: \left| \Phi ^{(n)}\right| _{-p^{\prime }}\leq {\rm n}_{p,l,\infty }(F)\
810: \left( 2^le\left\| i_{p^{\prime },p}\right\| _{HS}\right) ^n
811: $$
812: Thus%
813: \begin{eqnarray*}
814: \lnorm F\rnorm _{-p^{\prime },-q,-1}^2%
815: &=& \sum_{n=0}^\infty%
816: 2^{-nq}\left| \Phi ^{(n)}\right| _{-p^{\prime }}^2 %
817: \\& \leq & {\rm n}_{p,l,\infty }^2(F)\sum_{n=0}^\infty %
818: \left( 2^{2l-q}e^2\left\|%
819: i_{p^{\prime },p}\right\| _{HS}^2\right) ^n %
820: \\&=& {\rm n}_{p,l,\infty }^2(F)\left( 1-2^{2l-q}e^2%
821: \left\| i_{p^{\prime },p}\right\| _{HS}^2\right) ^{-1}\
822: \end{eqnarray*}
823: if $q\in \N $ is such that $\rho :=2^{2l-q}e^2\left\| i_{p^{\prime
824: },p}\right\| _{HS}^2\ <1$. So we have in fact
825: $$
826: \lnorm F\rnorm _{-p^{\prime },-q,-1}\leq {\rm n}_{p,l,\infty }(F)(1-\rho
827: )^{-1/2}.
828: $$
829: Now the rest is simple. Define $\Phi \in ({\cal N})_\mu ^{-1}$ by
830: $$
831: \Phi =\sum_{n=0}^\infty Q_n^\mu (\Phi ^{(n)})
832: $$
833: then $S_\mu \Phi =F$ and
834: $$
835: \left\| \Phi \right\| _{-p^{\prime },-q,\mu }=\lnorm F\rnorm
836: _{-p^{\prime },-q,-1}
837: $$
838: This proves the existence of a continuous inverse of the $S_\mu $%
839: --transform. Uniqueness of $\Phi $ follows from the fact that $\mu $%
840: -exponentials are total in any $({\cal H}_p)_q^1$. \TeXButton{End Proof}
841: {\endproof}\bigskip\
842:
843: We can extract some useful estimates from the above proof which describe the
844: degree of singularity of a distribution.
845:
846: \begin{corollary}
847: Let $F\in {\rm Hol}_0({\cal N}_{\Ckl })$ be holomorphic for all $\theta \in
848: {\cal N}_{\Ckl }$ with $|\theta |_p\leq 2^{-l}$. If $p^{\prime }>p$ with $%
849: \left\| i_{p^{\prime },p}\right\| _{HS}<\infty $ and $q\in \N $ is such that
850: $\rho :=2^{2l-q}e^2\left\| i_{p^{\prime },p}\right\| _{HS}^2<1$. Then $\Phi
851: \in ({\cal H}_{-p^{\prime }})_{-q}^{-1}$ and
852: $$
853: \left\| \Phi \right\| _{-p^{\prime },-q,\mu }\leq {\rm n}_{p,l,\infty
854: }(F)\cdot (1-\rho )^{-1/2}.
855: $$
856: \end{corollary}
857:
858: For a more detailed discussion of the degree of singularity the spaces $(%
859: {\cal N})^{-\beta },\ \beta \in [0,1)$ are useful. In the following theorem
860: we will characterize these spaces by means of $S_\mu$-transform.
861:
862: \begin{theorem}
863: \sloppy The $S_\mu $-transform is a topological isomorphism from $({\cal N}%
864: )_\mu ^{-\beta }$, $\beta \in [0,1)$ on ${\cal E}_{\max }^{2/(1-\beta )}(%
865: {\cal N}_{\Ckl })$.
866: \end{theorem}
867:
868: \fussy \TeXButton{Remark }{\remark } The proof will also complete the proof
869: of Theorem \ref{indlimEkmax}.
870:
871: \TeXButton{Proof}{\proof}Let $\Phi \in ({\cal H}_{-p})_{-q,\mu }^{-\beta }$
872: with the canonical representation $\Phi =\sum_{n=0}^\infty Q_n^\mu (\Phi
873: ^{(n)})$ be given. The $S_\mu $-transform of $\Phi $ is given by
874: $$
875: S_\mu \Phi (\theta ) =\sum_{n=0}^\infty \langle \Phi
876: ^{(n)},\theta ^{\otimes n}\rangle .
877: $$
878: Hence
879: $$
880: \lnorm S_\mu \Phi \rnorm _{-p,-q,-\beta }^2=\sum_{n=0}^\infty
881: (n!)^{1-\beta }\,2^{-nq}|\Phi ^{(n)}|_{-p}^2
882: $$
883: is finite. We will show that there exist $l\in \N $ and $C<0$ such that%
884: $$
885: {\rm n}_{-p,-l,2/(1-\beta )}(S_\mu \Phi )\leq C\,\lnorm S_\mu \Phi \rnorm %
886: _{-p,-q,-\beta }\ .
887: $$
888:
889: We can estimate as follows%
890: \begin{eqnarray*}
891: |S_\mu \Phi (\theta )| & \leq & \sum_{n=0}^\infty \left| \Phi ^{(n)}\right| %
892: _{-p}\left| \theta \right| _p^n %
893: \\& \leq & \left( \sum_{n=0}^\infty (n!)^{1-\beta }2^{-nq}|\Phi %
894: ^{(n)}|_{-p}^2\right) ^{1/2}\left( \sum_{n=0}^\infty %
895: \frac 1{(n!)^{1-\beta }}2^{nq}\left| \theta \right| _p^{2n}\right) ^{1/2} %
896: \\&=& \lnorm S_\mu \Phi \rnorm _{-p,-q,-\beta }\left( %
897: \sum_{n=0}^\infty \rho ^{n\beta }\cdot \frac 1{(n!)^{1-\beta }}2^{nq}\rho %
898: ^{-n\beta }\left| \theta \right| _p^{2n}\cdot \right) ^{1/2},
899: \end{eqnarray*}
900: where we have introduced a parameter $\rho \in (0,1)$. An application of
901: H\"older's inequality for the conjugate indices $\frac 1\beta $ and $\frac
902: 1{1-\beta }$ gives%
903: \begin{eqnarray*}
904: \left| S_\mu \Phi (\theta )\right| & \leq & \lnorm S_\mu \Phi \rnorm%
905: _{-p,-q,-\beta }\left( \sum_{n=0}^\infty \rho ^n\right) %
906: ^{\beta /2}\cdot \left( \sum_{n=0}^\infty \frac 1{n!}\left( 2^q\rho %
907: ^{-\beta }|\theta |_p^2\right) ^{\frac n{1-\beta }}\right) %
908: ^{\frac{1-\beta }2} %
909: \\&=& \lnorm S_\mu \Phi \rnorm %
910: _{-p,-q,-\beta }\left( 1-\rho %
911: \right) ^{-\beta /2}\exp \left( \tfrac{1-\beta }2 \,2^{\frac q{1-\beta }}\,%
912: \rho ^{-\frac \beta {1-\beta }}\, |\theta |_p^{\frac 2{1-\beta }}\right) %
913: \end{eqnarray*}
914: If $l\in \N $ is such that%
915: $$
916: 2^{l-\frac q{1-\beta }}>\tfrac{1-\beta }2\rho ^{-\frac \beta {1-\beta }}
917: $$
918: we have%
919: \begin{eqnarray*}
920: {\rm n}_{-p,-l,2/(1-\beta )}(S_\mu \Phi )%
921: &=& \sup _{\theta \in {\cal H}_{p,\Ckkl }}\left| S_\mu \Phi (\theta )%
922: \right| \, \exp \left( -2^l|\theta %
923: |_p^{2/(1-\beta )}\right) %
924: \\& \leq & \left( 1-\rho \right) ^{-\beta /2}\lnorm S_\mu \Phi \rnorm %
925: _{-p,-q,-\beta }%
926: \end{eqnarray*}
927: This shows that $S_\mu $ is continuous from $({\cal N})_\mu ^{-\beta }$ to $%
928: {\cal E}_{\max }^{2/(1-\beta )}({\cal N}_{\Ckl }).$ Or in the language of
929: Theorem \ref{indlimEkmax}%
930: $$
931: \stackunder{p,q\in \N }{\rm ind\ lim}\,{\rm E}_{-p,-q}^{-\beta }\subset
932: {\cal E}_{\max }^{2/(1-\beta )}({\cal N}_{\Ckl })
933: $$
934: topologically.\medskip\
935:
936: The proof of the inverse direction is closely related to the proof of Lemma
937: \ref{3Strichnplk}. So we will be more sketchy in the following.
938:
939: \noindent Let $F\in {\cal E}_{\max }^k({\cal N}_{\Ckl }),\ k=\frac 2{1-\beta
940: }$. Hence there exist $p,l\in \N _0$ such that%
941: $$
942: \left| F(\theta )\right| \leq {\rm n}_{-p,-l,k}(F)\exp (2^l|\theta |_p^k)\
943: ,\qquad \theta \in {\cal N}_{\Ckl }
944: $$
945: From this we have completely analogous to the proof of Lemma \ref
946: {3Strichnplk} by Cauchy inequality and kernel theorem the representation%
947: $$
948: F(\theta )=\sum_{n=0}^\infty \langle \Phi ^{(n)},\theta ^{\otimes n}\rangle
949: $$
950: and the bound%
951: $$
952: \left| \Phi ^{(n)}\right| _{-p^{\prime }}\leq {\rm n}_{-p,-l,k}(F)%
953: \;(n!)^{-1/k}\left\{ (k2^l)^{1/k}e\left\| i_{p^{\prime },p}\right\|
954: _{HS}\right\} ^n\ ,
955: $$
956: where $p^{\prime }>p$ is such that $i_{p^{\prime },p}:{\cal H}_{p^{\prime
957: }}\hookrightarrow {\cal H}_p$ is Hilbert--Schmidt. Using this we have
958: \begin{eqnarray*}
959: \lnorm F\rnorm _{-p^{\prime },-q,-\beta}^2 %
960: &=& \sum_{n=0}^\infty (n!)^{1-\beta }2^{-qn}\left| \Phi ^{(n)}\right| %
961: _{-p^{\prime }}^2 %
962: \\& \leq & {\rm n}_{-p,-l,k}^2(F)\sum_{n=0}^\infty (n!)^{1-\beta -2/k}2^{-qn}%
963: \left\{ (k2^l)^{1/k}e\left\| i_{p^{\prime },p}\right\| %
964: _{HS}\right\} ^{2n} %
965: \\& \leq & {\rm n}_{-p,-l,k}^2(F)\sum_{n=0}^\infty \rho ^n
966: \end{eqnarray*}
967: where we have set $\rho :=2^{-q+2l/k}k^{2/k}e^2\left\| i_{p^{\prime
968: },p}\right\| _{HS}^2$ . If $q\in \N $ is chosen large enough such that $\rho
969: <1$ the sum on the right hand side is convergent and we have
970: \begin{equation}
971: \label{F3Strich}\lnorm F\rnorm _{-p^{\prime },-q,-\beta }\leq {\rm n}%
972: _{-p,-l,2/(1-\beta )}(F)\cdot (1-\rho )^{-1/2}\ .
973: \end{equation}
974: That means%
975: $$
976: {\cal E}_{\max }^{2/(1-\beta )}({\cal N}_{\Ckl })\subset \ \stackunder{%
977: p,q\in \N _0}{\rm ind\ lim}\,{\rm E}_{-p,-q}^{-\beta }
978: $$
979: topologically.
980:
981: If we set
982: $$
983: \Phi :=\sum_{n=0}^\infty Q_n^\mu (\Phi ^{(n)})
984: $$
985: then $S_\mu\Phi =F$ and $\Phi \in ({\cal H}_{-p^{\prime }})_{-q}^{-\beta }$
986: since%
987: $$
988: \sum_{n=0}^\infty (n!)^{1-\beta }2^{-qn}|\Phi ^{(n)}|_{-p^{\prime }}^2
989: $$
990: is finite. Hence%
991: $$
992: S_\mu:({\cal N})_\mu ^{-\beta }\rightarrow {\cal E}_{\max }^{2/(1-\beta )}(%
993: {\cal N}_{\Ckl })
994: $$
995: is one to one. The continuity of the inverse mapping follows from the norm
996: estimate (\ref{F3Strich}).\TeXButton{End Proof}{\endproof}
997: