1: \input liemacs.tex
2:
3:
4:
5: \def\b{\mathop{\bf b}\nolimits}
6:
7:
8:
9: \def\c{\mathop{\bf c}\nolimits}
10:
11:
12:
13: \def\cH{\mathop{{\bf c}_{\hbox{\fiverm G/H}}}\nolimits}
14:
15:
16:
17: \def\e{\mathop{\bf e}\nolimits}
18:
19:
20:
21: \def\hfH{\hbox{\fiverm H}}
22:
23:
24:
25: \def\hfK{\hbox{\fiverm K}}
26:
27:
28:
29: \def\ssmapright#1{\smash{\mathop{\ssarr}\limits^{#1}}}
30:
31: \input epsf.tex
32:
33:
34:
35:
36:
37:
38:
39: \def\bs{\backslash}
40:
41: \def\addots{\mathinner{\mkern1mu\raise1pt\vbox{\kern7pt\hbox{.}}\mkern2mu
42: \raise4pt\hbox{.}\mkern2mu\raise7pt\hbox{.}\mkern1mu}}
43:
44: \pageno=1
45: \def\up#1{\leavevmode \raise.16ex\hbox{#1}}
46: \font\smallheadfont=cmr8 at 8truept
47: \font\smallbfheadfont=cmbx8 at 8truept
48: %\font\largeheadfont=cmdunh10 at 14.4truept
49: \font\headfont=cmdunh10 at 12truept
50: \chardef\ss="19
51: \def\3{\ss}
52:
53: \def\firstpage{\nin
54: {\obeylines \parindent 0pt }
55: \vskip2cm
56: \centerline {\bfone \title}
57: \ssk
58: \centerline {\bfone \titletwo}
59: \gsk
60: \centerline{\bf\author}
61: \vskip1.5cm \rm}
62:
63:
64: \def\title{Holomorphic $H$-spherical distribution vectors}
65: \def\titletwo{in principial series representations}
66:
67: \def\author{Simon Gindikin, Bernhard Kr\"otz and
68: Gestur \'Olafsson}
69:
70: \footnote{}{SG was supported in part by NSF-grant DMS-0070816}
71: \footnote{}{BK was supported in part by NSF-grant DMS-0097314}
72: \footnote{}{G\'O was supported in part by NSF-grant DMS-0070607 and DMS-0139783}
73:
74:
75: \def\address
76: {Simon Gindikin
77:
78: Department of Mathematics
79:
80: Rutgers University
81:
82: New Brunswick, NJ 08903
83:
84: USA
85:
86: {\tt gindikin@math.rutgers.edu}
87:
88: \bsk
89: \bsk
90:
91: Gestur \'Olafsson
92:
93: Louisiana State University
94:
95: Department of Mathematics
96:
97: Baton Rouge, LA 70803
98:
99: USA
100:
101: {\tt olafsson@math.lsu.edu}
102:
103: }
104:
105: \def\addresstwo
106: {Bernhard Kr\"otz
107:
108: Department of Mathematics
109: University of Oregon
110: Eugene Or 97403-1221
111: USA
112:
113: {\tt kroetz@math.uoregon.edu}
114: }
115:
116:
117: \firstpage
118:
119:
120: \subheadline{Abstract}
121: \noindent
122: Let $G/H$ be a semisimple symmetric space.
123: The main tool to embed a principal series
124: representation of $G$ into $L^2(G/H)$ are
125: the $H$-invariant distribution vectors.
126: If $G/H$ is a non-compactly
127: causal symmetric space, then $G/H$ can be realized as a
128: boundary component of the complex crown $\Xi$.
129: In this article we construct a minimal $G$-invariant
130: subdomain $\Xi_H$ of $\Xi$ with
131: $G/H$ as Shilov boundary. Let $\pi$ be
132: a spherical principal series representation
133: of $G$. We show
134: that the space of $H$-invariant
135: distribution vectors of $\pi$, which admit a
136: holomorphic extension to $\Xi_H$,
137: is one dimensional. Furthermore we give a spectral definition of
138: a Hardy space corresponding to those distribution
139: vectors. In particular we achieve a geometric
140: realization of a multiplicity free subspace of $L^2(G/H)_{\rm mc}$
141: in a space of holomorphic functions.
142:
143: \sectionheadline{Introduction}
144:
145: \noindent
146: Holomorphic extensions and boundary value maps have been
147: valuable tools to solve problems in representation theory
148: and harmonic analysis on {\it real} symmetric
149: spaces. Two of the best known constructions
150: are Hardy spaces with their boundary value maps and
151: Cauchy-Szeg\"o-kernels,
152: and Fock space constructions with their corresponding
153: Segal-Barmann transform. It is in this flavour that we establish a
154: correspondence between eigenfunctions on a Riemannian symmetric spaces $X=G/K$
155: and a non-compactly causal (NCC) symmetric spaces $Y=G/H$ in this paper.
156: In particular
157: we, via analytic continuation, relate a {\it spherical function} $\phi_\lambda$ on
158: $G/K$ to a {\it holomorphic} $H$-invariant distribution on $G/H$.
159:
160:
161: \par Let us explain our results in more detail. On the geometric level we construct a certain minimal
162: $G$-invariant Stein domain $\Xi_H\subeq X_\C=G_\C/K_\C$ with
163: the following properties: The Riemannian symmetric space $X$ is embedded into $\Xi_H$ as
164: a totally real submanifold and the affine non-compactly causal space $Y$ is isomorphic to the
165: distinguished (Shilov) boundary of $\Xi_H$. The details
166: of this construction are carried out in Section 1.
167:
168: \par The minimal tube $\Xi_H$ is a subdomain of the complex
169: crown $\Xi\subeq X_\C$ of $X$ -- an object first introduced
170: in [AG90] which became subject of intense study over the last
171: few years. A consequence is that all $\D(X)$-eigenfunctions
172: on $X$ extend holomorphically to $\Xi_H$ [KS01b]. Another key fact
173: is that $\D(X)\simeq \D (Y)$. Thus by taking limits on the boundary $Y$ we obtain a realization
174: of the $\D(X)$-eigenfunctions on $X$ as $\D(Y)$-eigenfunctions on $Y$.
175: Conversely, eigenfunctions on $Y$ which holomorphically extend to $\Xi_H$
176: yield by restriction eigenfunctions on $X$.
177:
178:
179: \par It seems to us that the above mentioned transition between eigenfunctions
180: on $X$ and $Y$ is most efficiently described using the
181: techniques from representation theory. To fix the notation let $(\pi,{\cal H})$
182: denote an admissible Hilbert representation of $G$ with
183: finite length. We write ${\cal H}^K$ for the space of $K$-fixed
184: vectors and $({\cal H}^{-\infty})^H$ for the space of $H$-fixed
185: distribution vectors of $\pi$. Using the method of analytic
186: continuation of representations as developed in [KS01a] we
187: establish a bijection
188: $${\cal H}^K\ssmapright{\simeq} ({\cal H}^{-\infty})_{\rm hol}^H , \ \
189: v_{\hbox{\fiverm K}}\mapsto v_{\hbox{\fiverm H}}$$
190: where $({\cal H}^{-\infty})_{\rm hol}^H\subeq ({\cal H}^{-\infty})^H$
191: denotes the subspace characterized through the property that
192: associated matrix coefficients on $Y$ extend holomorphically
193: to $\Xi_H$ (cf.\ Theorem 2.1.3, Theorem 2.2.4). This bijection and various
194: ramifications are the subject proper of Section 2.
195:
196: \par In Section 3 we give an application
197: of our theory towards the geometric realization of the most-continuous spectrum
198: $L^2(Y)_{\rm mc}$ of $L^2(Y)$. First progress in this direction was achieved in [GK\'O01].
199: There, for the cases where $\Xi=\Xi_H$, we defined a Hardy space
200: ${\cal H}^2(\Xi)$ on $\Xi$ and showed that there is an isometric boundary value mapping
201: realizing ${\cal H}^2(\Xi)$ as a
202: multiplicity one subspace of $L^2(Y)_{\rm mc}$ of full spectrum.
203: It was an open problem how to define
204: Hardy spaces for general NCC symmetric spaces $Y$ and to determine the Plancherel
205: measure explicitely. We solve this problem by giving a spectral definition of the
206: Hardy space, i.e., we take the conjectured Plancherel measure and define a Hilbert
207: space of holomorphic functions ${\cal H}^2(\Xi_H)$ on $\Xi_H$. The identification
208: of ${\cal H}^2(\Xi_H)$ as a Hardy space then follows by establishing
209: an isometric boundary value mapping $b\: {\cal H}^2(\Xi_H)\into L^2(G/H)_{\rm mc}$.
210: In particular we achieve a geometric realization of a multiplicity free subspace of $L^2(Y)_{\rm mc}$
211: in holomorphic functions.
212:
213: \msk It is our pleasure to thank the referee for his very careful work. He pointed out many
214: inaccuracies and made useful remarks on the presentation of the paper.
215:
216:
217:
218: \sectionheadline{1. Complex crowns and the domains $\Xi_H$}
219:
220: \noindent
221: The purpose of this section is to give the geometric preliminaries
222: of the analytical constructions to come. Our two main
223: players are a Riemannian symmetric space $G/K$ on the one hand side
224: and on the other hand a non-compactly causal symmetric space $G/H$.
225: The two symmetric spaces $G/K$ and $G/H$ are ``connected'' through
226: a complex $G$-invariant domain $\Xi_H\subeq G_\C/K_\C$ in the
227: following way: $G/K$ is a totally real submanifold and
228: $G/H$ constitutes the distinguished (Shilov) boundary of $\Xi_H$.
229: The domain $\Xi_H$ constructed in this section is an appropriate
230: subdomain of the complex crown $\Xi$ of the Riemannian symmetric space
231: $G/K$.
232:
233: \par This section is organized as follows. We start by briefly recalling
234: the defintion and some standard features of non-compactly causal
235: symmetric spaces . Then we switch to complex crowns $\Xi$
236: and summarize the main results of [GK02a] on how
237: to realize $G/H$ in the distinguished boundary of $\Xi$. Finally
238: we give the construction of the domain $\Xi_H$.
239:
240:
241:
242: \subheadline{1.1. Non-compactly causal symmetric spaces (NCC)}
243:
244: \noindent
245: In this subsection we recall some facts on non-compactly causal symmetric spaces.
246: The material is standard and can be found in the monograph
247: [H\'O96].
248:
249: \ssk
250: Let $G$ be a connected semisimple Lie group and $\g$ be its Lie algebra.
251: Denote by $\g_\C=\g\otimes_\R\C$ the complexification of $\g$.
252: If $\h$ is a subalgebra of $\g$, then we denote by $\h_\C$ the complex subalgebra
253: of $\g_\C$ generated by $\h$.
254: We assume that $G$ is contained
255: in a complex group $G_\C$ with Lie algebra $\g_\C$.
256:
257: \par If $\sigma\:G \to G$ is
258: an involution, then, by abuse of notation, we use the same letter
259: for the derived involution on the Lie algebra $\g$ and its complex linear
260: extension to $\g_\C$.
261: \par Let $\theta\:G\to G$ be a Cartan involution and denote by $K<G$ the corresponding
262: maximal compact subgroup. Let $\k=\{X\in\g\: \theta(X)=X\}$
263: and $\p =\{X\in \g\: \theta (X)=-X\}$. Then
264: $\k$ is the Lie algebra of $K$.
265:
266: \par In the sequel we let $\tau$ denote an involution on $G$ which we may assume to commute
267: with $\theta$. Let $G^\tau :=\{g\in G\: \tau (g)=g\}$ and
268: let $H$ be an open subgroup of $G^\tau$. Then $G/H$ is called a {\it symmetric
269: spaces}. On the Lie algebra level
270: $\tau$ induces a splitting $\g=\h+\q$ with $\h$ the $+1$ and $\q$ the $-1$-eigenspace of
271: $\tau$. Notice that $\h$ is the Lie
272: algebra of $H$. The pair $(\g,\h)$ is called a {\it symmetric pair}. We have, as $\theta$ and $\tau$ commute:
273: $$\eqalign{ \g &=\k+ \p\cr
274: &=\h+ \q\cr
275: &=\k\cap \h +\k\cap \q + \p\cap \h + \p\cap \q
276: }
277: $$
278: \par
279: The symmetric pair
280: $(\g ,\h)$ is called {\it irreducible} if the only
281: $\tau$-invariant ideals in $\g$ are the trivial ones,
282: $\{0\}$ and $\g$. In this case either $\g$ is
283: simple or $\g \simeq \h\oplus \h$, with $\h$ simple, and $\tau (X,Y)=
284: (Y,X)$ the flip. We say that the symmetric space $G/H$ is {\it irreducible}
285: if the corresponding symmetric pair $(\g ,\h)$ is irreducible.
286:
287: \par Let $\emptyset \not= C\subseteq \g$ be an open subset of $\g$. Then
288: $C$ is said to be {\it hyperbolic} if for all $X\in C$ the map ${\rm ad}(X):\g \to \g$ is
289: semisimple with real eigenvalues.
290:
291: \Definition 1.1.1. {\bf (NCC)} Assume that $G/H$ is an irreducible symmetric space. Then
292: the following two conditions are equivalent:
293:
294: \item{(a)} There exists a non-empty $H$-invariant open hyperbolic convex cone $C
295: \subseteq \q$ which contains no affine lines;
296:
297: \item{(b)} There exists an element $T_0\in \q\cap \p$, $T_0\not= 0$,
298: which is fixed by $H\cap K$.
299:
300: If one of those equivalent conditions are satisfied, then $G/H$ is called
301: {\it non-compactly causal}, or {\it NCC} for short.
302: \qed
303:
304: \Remark 1.1.2. (a) The element $T_0$ in Definition 1.1.1 is
305: unique up to multiplication by scalar. We can normalize
306: $T_0$ such that ${\rm ad}(T_0)$ has spectrum $\{0,1,-1\}$. The eigenspace
307: corresponding to $0$ is exactly $\g^{\theta\tau}=\k\cap\h +\p\cap \q$.
308: \par\nin (b) If $G/H$ is NCC and $\a\subset \p\cap \q$ is maximal
309: abelian, then $T_0\in \a$ by (a). Hence, again by (a), it follows
310: that $\a$ is also maximal abelian in $\p$ and in $\q$.
311: \par\nin (c) Let $T_0$ be as above. Then the interior of the convex hull of $\R^+\Ad (H)T_0$
312: is a minimal, $H$-invariant open hyperbolic convex cone in $\q$.
313: \par\nin (d) All the NCC pairs $(\g, \h)$ are classified and we refer
314: to [H\'O96, Th.\ 3.2.8] for the complete list.\qed
315:
316:
317: \subheadline{1.2. The complex crown of a Riemannian symmetric space}
318:
319: \noindent
320: The NCC spaces are exactly the affine symmetric spaces that can be
321: realized as a symmetric subspace in the distinguished boundary
322: of the complex crown $\Xi$ of the Riemannian symmetric space
323: $G/K$. We will therefore recall some basic facts about
324: $\Xi$. We refer to [GK02a] and [GK02b] as a standard source.
325:
326:
327: Let the notation be as in Subsection 1.1. Let $\a\subeq \p$ be a maximal
328: abelian subalgebra. For $\alpha \in \a^*$ let
329: $\g^\alpha=\{ X\in \g\: (\forall H\in \a)\ [H,X]=\alpha(H)X\}$ and
330: let $\Sigma:=\{\alpha\in \a^*\:\alpha\not=0,\g^\alpha\not=\{0\}\}$ be
331: the corresponding set of restricted roots.
332:
333:
334: \par Following [AG90] we define a bounded convex subset of $\a$ by
335: $$\Omega=\{ X\in \a\: (\forall \alpha\in \Sigma)\ |\alpha(X)|<{\pi\over 2}\}\ .$$
336: Denote by $K_\C$ the analytic subgroup of $G_\C$ with Lie algebra $\k_\C$.
337: Then we define a $G-K_\C$ double coset domain in $G_\C$ by
338:
339: $$\tilde \Xi=G\exp(i\Omega)K_\C$$
340: and recall that $\tilde\Xi$ is open in $G_\C$ [KS01a]. In particular the domain
341: $$\Xi=\tilde\Xi/K_\C$$
342: is an open $G$-invariant subset of $G_\C/K_\C$ containing $G/K$ as a totally
343: real submanifold. We refer to $\Xi$ as the {\it complex crown} of the
344: Riemannian symmetric space $G/K$ (cf.\ [AG90]). Observe that the definition
345: of $\Xi$ and $\tilde \Xi$ is independent of the choice of $\a\subeq \p$.
346: For a subset $\omega\subeq \a$ we define a tube domain in $A_\C=\exp(\a_\C)$
347: by
348: $$T( \omega) =A\exp(i\omega)$$
349: and notice that $T (2\Omega)$ is biholomorphic to $\a+i 2\Omega$ via the
350: exponential map.
351:
352: \msk Fix a positive system $\Sigma^+$ of $\Sigma$ and define a subalgebra
353: $\n$ of $\g$ by
354: $$\n=\bigoplus_{\alpha\in \Sigma^+} \g^\alpha\ .$$
355: Write $N_\C$ for the analytic subgroup of $G_\C$ with Lie algebra $\n_\C$.
356: Then it follows from [GK02b] that
357: $$\tilde\Xi\subeq N_\C T(\Omega) K_\C\leqno(1.2.1)$$
358: or even more precisely
359: $$\tilde\Xi=\left[\bigcap_{g\in G} g N_\C T(\Omega) K_\C \right]_0\leqno (1.2.2)$$
360: where the subscript ${}_0$ denotes the connected component of $[\cdot]$ containing $G$.
361:
362: \subheadline{1.3. The distinguished boundary of $\Xi$}
363:
364: \noindent
365: Write $\oline \Xi$ and $\partial \Xi$ for the closure respectively the boundary of $\Xi$ in
366: $G_\C/K_\C$. The {\it distinguished boundary} of $\Xi$ is a certain finite
367: union of $G$-orbits in $\partial \Xi$ which features many properties of
368: a Shilov boundary. It was introduced and investigated in [GK02a]
369: and the objective of this subsection is to recall its definition and basic
370: properties.
371:
372:
373: \par Let
374: ${\cal W}=N_K(\a)/ Z_K(\a)$ be the Weyl group of $\a$ in $G$.
375: Write $\oline\Omega$ for the closure of $\Omega$ and notice that
376: $\oline\Omega$ is a ${\cal W}$-invariant compact convex set. Denote by
377: $\partial_e\Omega$ the set of extreme points of $\oline \Omega$. Then
378: there exists $X_1,\ldots ,X_n\in\partial_e\Omega$ such
379: that
380: $$\partial_e\Omega={\cal W}(X_1)\amalg\ldots\amalg{\cal W}(X_n)\ .$$
381: Define the {\it distinguished boundary} of $\Xi$ in $G_\C/K_\C$
382: by
383: $$\partial_d\Xi\:=G\exp(i\partial_e\Omega)K_\C/K_\C\ .$$
384: We refer to [GK02a] for detailed information about $\partial_d\Xi$
385: and recall here only the facts that we need.
386:
387: \par For $1\leq j\leq n$ let
388: $z_j\:=\exp(iX_j)K_\C \in \partial_d\Xi$. If $G_\C$ is not simply connected it
389: can happen that $G(z_j)=G(z_k)$ for some $j\neq k$. But after relabelling
390: the $z_j$ we can assume that there is an $m\leq n$ such that
391: $G(z_j)\not=G(z_k)$ for $1\le j,k\le m$, $j\not= k$, and
392: $$\partial_d\Xi= G(z_1)\amalg\ldots\amalg G(z_m)\ .$$
393: Denote by $H_j$ the isotropy subgroup of $G$ in $z_j$. Then as $G$-spaces we have
394:
395: $$\partial_d\Xi= G/H_1\amalg\ldots \amalg G/ H_m \ .$$
396: As a consequence of the complete classification of $\partial_d\Xi$ in
397: [GK02a] we obtain the following fact.
398:
399:
400: \Proposition 1.3.1. For the distinguished boundary
401: $\partial_d\Xi$ of $\Xi$ the following assertions hold:
402: \item{(i)} If one of the boundary components
403: $G/H_j$ of $\partial_d\Xi$ is a symmetric space, then it is a
404: non-compactly causal symmetric space.
405: \item{(ii)} Every
406: non-compactly causal symmetric space of the form $G/H$
407: is locally isomorphic to a $G$-orbit in the distinguished boundary
408: of $\partial_d\Xi$ of $\Xi$. \qed
409:
410:
411:
412: \subheadline{1.4. The domains $\Xi_H$}
413:
414: \noindent
415: We keep the notation from Subsections 1.2 - 1.3. {}From now on we
416: fix an element $X_{\hbox{\fiverm H}}\in \partial_e\Omega$
417: and set $x_{\hbox{\fiverm H}}=\exp(iX_{\hbox{\fiverm H}})$, $z_{\hbox{\fiverm H}}=
418: x_{\hbox{\fiverm H}}K_\C$. As the notation suggests we denote by $H<G$ the stabilizer
419: of $z_{\hbox{\fiverm H}}\in \partial_d\Xi$ in $G$.
420:
421: \par For the rest of this paper we will employ the following assumptions:
422: $G_\C$ is simply connected and $G/H\simeq G(z_{\hbox{\fiverm H}})$ is an NCC symmetric space (cf.\
423: Proposition 1.3.1). Notice that this implies in particular $H=G^\tau$.
424: Recall the element $T_0$ from Definition 1.1.1 and notice that we have
425: (up to sign) $X_{\hbox{\fiverm H}}={\pi\over 2}T_0$.
426:
427:
428:
429: \par We define a domain $\Omega_{\hbox{\fiverm H}}\subeq \Omega$
430: by
431:
432: $$\Omega_{\hbox{\fiverm H}}=\Int \left(\conv\{{\cal W}(X_{\hbox{\fiverm H}})\}\right)\ .$$
433: Here $\conv\{\cdot\}$ denotes the convex hull of $\{ \cdot\}$ and $\Int(\cdot)$ denotes
434: the interior of $(\cdot)$. {}From the definition we immediately obtain that:
435:
436: \msk
437: \item{(1.4.1)} $\Omega_{\hbox{\fiverm H}}$ is open in $\a$.
438: \item{(1.4.2)} $0\in \Omega_{\hbox{\fiverm H}}$ (because $ X_{\hbox{\fiverm H}}\neq 0$
439: and ${\cal W}(X_{\hbox{\fiverm H}})$ meets every Weyl chamber).
440: \item{(1.4.3)} The set of extremal points of $\oline{\Omega_{\hbox{\fiverm H}}}$ is ${\cal W}(X_{\hbox{\fiverm H}})$.
441: \msk
442:
443: Let us illustrate the geometry for one example.
444:
445: \Example 1.4.1. Let $G=\Sl(3,\R)$. Then $\a$ is two-dimensional
446: and $\Sigma$ is a root system of type $A_2$. We have $\partial_e\Omega={\cal W}(X_1)\amalg {\cal W}(X_2)$ and the
447: corresponding isotropy subgroups are given by $H_1=\SO(1,2)$ and $H_2=\SO(2,1)$.
448: With $H=H_1$ the geometry of $\Omega$ and $\Omega_H$ is depicted as follows:
449:
450: \gsk
451: \epsfbox{hex.eps}
452: \gsk\gsk
453: \qed
454:
455: \nin Let us define a domain $\Xi_H\subeq G_\C/ K_\C$ by
456: $$\Xi_H=G\exp(i\Omega_{\hbox{\fiverm H}})K_\C/ K_\C\, .$$
457: The domain $\Xi_H$ has the following properties:
458: \msk
459: \item{(1.4.4)} $\Xi_H$ is $G$-invariant (clear from the definition).
460: \item{(1.4.5)} $\Xi_H$ is open in $G_\C/ K_\C$ (follows from (1.4.1) and [AG90]).
461: \item{(1.4.6)} $G/K\subeq \Xi_H$ is a totally real submanifold (follows from (1.4.2) and (1.4.5)).
462: \item{(1.4.7)} $\Xi_H$ is Stein (follows from the convexity of $\Omega_{\hbox{\fiverm H}}$ and [GK02b]).
463: \item{(1.4.8)} $\Xi_H\subeq \Xi$ (because $\Omega_{\hbox{\fiverm H}}\subeq \Omega$).
464: \item{(1.4.9)} $\Xi_H=\Xi$ iff $\Sigma$ is of type $C_n$ (cf.\ [KS01b]).
465: \msk
466:
467: \nin Write $\oline{\Xi_H}$ for the closure of $\Xi_H$ in $G_\C/ K_\C$ and define the
468: {\it distinguished boundary} of $\Xi_H$ by
469: $$\partial_d\Xi_H=G(z_{\hbox{\fiverm H}})\, .$$
470: Notice that $\partial_d\Xi_H\simeq G/H$ as $G$-spaces. Let us remark further that
471: $\partial_d\Xi_H\subeq \partial_d\Xi$.
472:
473: \par The distinguished boundary $\partial_d\Xi_H$ of $\Xi_H$ can be
474: considered as some sort of Shilov boundary of $\Xi_H$. More precisely,
475: mimicking the argument in [GK02a, Th.\ 2.3] we obtain that
476:
477: $$\sup_{z\in \Xi_H} |f(z)|=\sup_{z\in \partial_d\Xi_H|} |f(z)|\leqno (1.4.10)$$
478: for all bounded holomorphic functions $f$ on $\Xi_H$ which continuously extend to $\oline{\Xi_H}$.
479:
480:
481:
482:
483: \sectionheadline{2. Holomorphic $H$-spherical distributions}
484:
485: \noindent
486: In the section we assume that $G/H$ is NCC symmetric space
487: realized as $G(z_{\hbox{\fiverm H}})$ in the distinguished boundary of $\Xi$.
488: \par Recall
489: that a representation $(\pi ,V)$ of $G$ is called {\it admissible} if
490: the multiplicity of each $K$-type is finite and of {\it finite length} if
491: the associated Harish-Chandra module of $K$-finite vectors $V_K$ is of finite length.
492: Our aim in this section is to
493: associate to a non-zero $K$-fixed vector $v_{\hbox{\fiverm K}}$ in an admissible
494: representation $(\pi,V)$ of finite length a certain canonical $H$-spherical
495: distribution vector $v_{\hbox{\fiverm H}}\in (V^{-\infty})^H$. For irreducible representations
496: $\pi$ the vector $v_{\hbox{\fiverm H}}$
497: is unique in the sense that it allows analytic continuation
498: of generalized matrix coefficients on $G/H$ to holomorphic
499: functions on $\Xi_H$.
500: \par We let $X_{\hbox{\fiverm H}},x_{\hbox{\fiverm H}}$, and $z_{\hbox{\fiverm H}}$ be as in
501: the last subsection and recall that $x_{\hbox{\fiverm H}}^{-1}H_\C x_{\hbox{\fiverm H}}=K_\C$.
502:
503: \msk\nin {\bf General Remark:} In all results of this section which involve the domain
504: $\Xi_H$ one can replace $\Xi_H$ by the bigger domain $\Xi$. This holds in particular
505: for the results in Subsection 2.2.
506:
507:
508: \subheadline{2.1. The definition of the holomorphic $H$-spherical distribution vector}
509:
510:
511: \noindent
512: Before we discuss the general case let us assume for the moment
513: that $V$ is irreducible and finite dimensional. Then
514: $(\pi, V)$ extends to a holomorphic representation of $G_\C$
515: which we also denote by $(\pi, V)$.
516: If $L<G$ is a subgroup of $G$ then we write $V^L$ for the subspace of
517: $V$ fixed by $L$.
518: Then the mapping
519: $$V^K\to V^H, \ \ v_{\hbox{\fiverm K}}\mapsto v_{\hbox{\fiverm H}}\:=\pi(x_{\hbox{\fiverm H}})v_{\hbox{\fiverm K}}$$
520: sets up an isomorphism between the $K$-spherical and $H$-spherical
521: vectors of $V$.
522: The obvious problem in the general case is, that $\pi(x_{\hbox{\fiverm H}})v$ is
523: not necessarily defined as an element in $V$.
524: \par
525:
526: We are now going to develop an appropriate generalization of the mapping
527: $v_{\hbox{\fiverm K}}\mapsto v_{\hbox{\fiverm H}}$ for an admissible finite length
528: representation $(\pi ,{\cal H})$ of $G$ in a Hilbert space ${\cal H}$.
529: Denote by ${\cal H}^\infty$ respectively ${\cal H}^\omega$ the space of smooth respectively
530: analytic vectors in ${\cal H}$. Their strong anti-duals, i.e., the space
531: of continuous conjugate linear maps into $\C$, are denoted by ${\cal H}^{-\infty}$
532: respectively ${\cal H}^{-\omega}$ and referred
533: to as the $G$-modules of {\it distribution
534: vectors}, respectively {\it hyperfunction vectors} of $(\pi, {\cal H})$.
535: Notice the chain of continuous inclusions ${\cal H}^{\omega}\into {\cal H}^{\infty}\into {\cal H}\into
536: {\cal H}^{-\infty}\into {\cal H}^{-\omega}$. Here the inclusion
537: ${\cal H}\into {\cal H}^{-\infty}$ is the natural one, $v\mapsto (u\mapsto \la v,u\ra)$. We
538: will also use the notation $u\mapsto \la v,u\ra$ for $v\in {\cal H}^{-\infty}$. Define a representation
539: $\pi^0$ of $G$ on $\cal H$ by
540: $$\eqalign{
541: \la \pi^0(g) v,u\ra &=\la \pi (g^{-1})^*v,u\ra \cr
542: &= \la v,\pi (g^{-1})u\ra\cr}\qquad (g\in G; \ u, v\in {\cal H})\ .$$
543: Hence the natural representation $\pi^{-\infty}$ of $G$ on ${\cal H}^{-\infty}$ is an extension of the representation
544: $\pi^0$ on ${\cal H}$ to ${\cal H}^{-\infty}$. The representation $\pi^0$ is
545: called the {\it conjugate dual representation} of $\pi$.
546: Notice that $(\pi^0, {\cal H})$ is admissible and of finite length if and
547: only if the same holds for $(\pi ,{\cal H})$. Note also that $\pi =\pi^0$ if and only if
548: $\pi $ is unitary. Notice that if $u,v\in {\cal H}$ then $\la v ,u\ra =\overline{\la u, v\ra}$. Accordingly,
549: if $u\in {\cal H}^{\infty}$ and $v\in {\cal H}^{-\infty}$, then we write
550: $$\la u,v\ra :=\overline{\la v,u\ra}\, .$$
551:
552:
553: \par
554: Denote by ${\cal H}_K$ the $(\g,K)$-module of $K$-finite vectors in ${\cal H}$. Note that by
555: our assumption that ${\cal H}$ is admissible and of finite length it follows that ${\cal H}_K\subseteq {\cal H}^\omega$.
556: But usually we cannot find $H$-fixed vectors in ${\cal H}$ but only
557: in the larger space of distribution vectors. Another
558: complication arises as the space of $({\cal H}^{-\infty})^H$ of $H$-invariants in
559: ${\cal H}^{-\infty}$ is finite dimensional but in general not one-dimensional.
560: For ``generic'' principal series representations of $G$ one has
561: $$\dim ({\cal H}^{-\infty})^H=\left| {\cal W}/ {\cal W}_0\right|$$
562: with ${\cal W}_0=N_{K\cap H}(\a)/ Z_{K\cap H}(\a)$ the {\it little Weyl group}.
563: Our correspondence will be that we associate to a
564: $K$-fixed vector $v_{\hbox{\fiverm K}}$ a unique $H$-fixed distribution vector $v_{\hbox{\fiverm H}}$.
565: If $\pi$ is irreducible, then this
566: distribution $v_{\hbox{\fiverm H}}$ is the (up to scalar) unique element of $({\cal H}^{-\infty})^H$
567: for which generalized matrix coefficients extend to holomorphic functions
568: from $G/H$ to $\Xi$. For the proof we will need the following {\it
569: Automatic Continuity} Theorem of van den Ban, Brylinski and Delorme (c.f. [vdBD88, Th. 2.1] and
570: [BD92, Th.\ 1] for the version used here).
571: \par If $V$ is a complex vector space, then let us denote by
572: $V^{\oline *}$ its algebraic anti-dual. Set
573: $({\cal H}^{-\infty})^{\h}=\{v\in {\cal H}^{-\infty}\: (\forall X\in \h)\, d\pi^{-\infty}(X)v=0\}$.
574:
575:
576:
577: \Theorem 2.1.1. {\bf (Automatic Continuity)}
578: Let $(\pi, {\cal H})$ be an admissible representation of $G$ with finite length. Then
579: $$({\cal H}^{-\infty})^{H_0}\simeq ({\cal H}_K^{\oline *})^\h\ , \leqno (2.1.1) $$
580: meaning every $\h$-fixed anti-linear functional on ${\cal H}_K$ admits
581: a unique extension to a continuous and $H_0$-fixed anti-linear functional
582: on ${\cal H}^\infty$. In particular we have that
583: $$({\cal H}^{-\infty})^{H_0}\simeq ({\cal H}^{-\omega})^{H_0}\ .\ \leqno(2.1.2)$$\qed
584:
585: \par Recall the
586: $G-K_\C$ double coset domain $\tilde \Xi=G\exp(i\Omega)K_\C$
587: in $G_\C$ and the complex crown $\Xi=\tilde\Xi/ K_\C$. Our methods use holomorphic extensions of
588: representations. We recall therefore some results from [KS01a] \S 4 and in particular Theorem 3.1:
589:
590: \Theorem 2.1.2. Let $(\pi, {\cal H})$ be an admissible Hilbert representation of $G$
591: with finite length. Then the following assertions hold:
592:
593: \item{(i)} For every $v\in {\cal H}_K$ the orbit
594: mapping $G\to {\cal H}, \ g\mapsto \pi(g)v$ extends to a $G$-equivariant holomorphic
595: mapping $\tilde \Xi\to {\cal H}$.
596:
597: \item{(ii)} Let $v,w\in {\cal H}_K$. Then the restricted matrix coefficient
598: $A\to \C, \ a\mapsto \la \pi (a)v, w\ra $ extends to a holomorphic mapping to
599: the abelian tube domain $T (2\Omega)=A\exp(2i\Omega)$.\qed
600:
601: We will from now on assume that $\pi\res_K$ is unitary. This is no restriction
602: in view of Weyl's unitarity trick.
603:
604: \par
605: As before we identify $G/H$ with the subset $G(z_{\hbox{\fiverm H}})$ of $\overline{\Xi_H}$.
606: For $0\leq t<1$ we set $a_t=\exp(itX_{\hbox{\fiverm H}})\in \exp(i\Omega)$ and notice that
607: $\lim_{t\to 1} a_t =x_{\hbox{\fiverm H}}$.
608:
609: \Theorem 2.1.3. Let $(\pi, {\cal H})$ be a $K$-spherical admissible representation
610: of $G$ with finite length. Let $v_K\in {\cal H}^K$ be non-zero. Then the anti-linear functional
611:
612: $$v_{\hbox{\fiverm H}}^\omega\: {\cal H}^\omega \to \C , \ \ v\mapsto \lim_{t\nearrow 1} \la \pi^0(a_t)v_{\hbox{\fiverm K}}, v\ra$$
613: is well defined, non-zero and admits a non-trivial extension to an
614: $H$-fixed distribution vector of $(\pi, {\cal H})$. \qed
615:
616: \ssk \nin {\bf Note:} If $(\pi, {\cal H})$ is unitary, then $\pi^0(a_t)=\pi(a_t)$ and so
617: $v_{\hbox{\fiverm H}}^\omega(v)=\lim_{t\nearrow 1} \la \pi(a_t)v_{\hbox{\fiverm K}}, v\ra$ for all $v\in {\cal H}^\omega$.
618:
619:
620: \Proof. We first show that $v_{\hbox{\fiverm H}}^\omega$ is well defined.
621: Let $v\in {\cal H}^\omega$. As $v$ is
622: analytic, we find an $0<\epsilon <1$ such that
623: $\pi (a_\eps)v$ is defined. Notice that $\pi^0 (a_{1-\eps})v_{\hbox{\fiverm K}}$ is defined
624: by Theorem 2.1.2 (i).
625: Then we have for all $\eps<t<1$ that
626: $$\la \pi^0(a_t)v_{\hbox{\fiverm K}}, v\ra= \la \pi^0(a_{t-\eps})v_{\hbox{\fiverm K}}, \pi(a_\eps)v\ra$$
627: and hence
628: $$\lim_{t\nearrow 1}\la \pi^0(a_t)v_{\hbox{\fiverm K}}, v\ra=\la \pi^0(a_{1-\eps})v_{\hbox{\fiverm K}},
629: \pi(a_\eps)v\ra\ .$$
630: Thus $v_{\hbox{\fiverm H}}^\omega$ is defined.
631:
632: \par Next we show that $v_{\hbox{\fiverm H}}^\omega$ is fixed by $H$.
633: As $G/H$ is NCC, it follows that $H=H_0 Z_{H\cap K}(\a )$ [H\'O96, p.79]. Further it is clear from the definition
634: that $v_{\hbox{\fiverm H}}^\omega$it is fixed by $Z_{H\cap K}(\a )$. Hence it is
635: enough to prove that $v_{\hbox{\fiverm H}}^\omega$
636: is $H_0$-fixed, i.e. annihilated by $\h$. For that let $Y\in \h$ and
637: $v\in {\cal H}^\omega$. Then
638:
639: $$\eqalign{v_{\hbox{\fiverm H}}^\omega(d\pi(Y)v)&=\lim_{t\nearrow 1}\la \pi^0(a_t)v_{\hbox{\fiverm K}}, d\pi(Y)v\ra=
640: -\lim_{t\nearrow 1}\la d\pi^0(Y)\pi^0(a_t)v_{\hbox{\fiverm K}}, v\ra\cr
641: &= - \lim_{t\nearrow 1}\la\pi^0(a_t)d\pi^0(\Ad(a_t)^{-1} Y)v_{\hbox{\fiverm K}}, v\ra\cr
642: &= -\lim_{t\nearrow 1}\la\pi^0(a_{t-\eps})d\pi^0(\Ad(a_t)^{-1} Y)v_{\hbox{\fiverm K}}, \pi(a_\eps)v\ra\ .\cr}$$
643: {}From Theorem 2.1.2 (i) we now obtain that
644: $$\pi^0(a_{t-\eps})d\pi^0(\Ad(a_t)^{-1} Y)v_{\hbox{\fiverm K}}\to
645: \pi^0(a_{1-\eps})d\pi^0(\Ad(x_{\hbox{\fiverm H}})^{-1} Y)v_{\hbox{\fiverm K}}\ .$$
646: But $\Ad (x_{\hbox{\fiverm H}})^{-1}Y\in \k_\C$ and hence
647: $d\pi^0(\Ad (x_{\hbox{\fiverm H}})^{-1}Y)v_{\hbox{\fiverm K}}=0$.
648: We thus get:
649: $$\eqalign{v_{\hbox{\fiverm H}}^\omega(d\pi(Y)v)
650: &=-\lim_{t\nearrow 1}\la\pi^0(a_{t-\eps})d\pi^0(\Ad(a_t)^{-1} Y)v_{\hbox{\fiverm K}}, \pi(a_\eps)v\ra\cr
651: &= -\la\pi^0(a_{1-\eps})d\pi^0(\Ad(x_{\hbox{\fiverm H}})^{-1} Y)v_{\hbox{\fiverm K}}, \pi(a_\eps)v\ra\cr
652: &= 0\, .\cr}$$
653:
654:
655: \par Finally, let us show that $v_{\hbox{\fiverm H}}^\omega\neq 0$. Suppose the contrary, i.e.
656: $v_{\hbox{\fiverm H}}^\omega=0$. Then it follows that
657: $$(\forall u\in {\cal H}_K) \qquad \la v_{\hbox{\fiverm H}}^\omega, u\ra =0\ . \leqno (2.1.3)$$
658: Now consider the function
659: $$f\: \R +i]-2,2[\to \C , \ \ z\mapsto \la \pi^0(\exp(zX_{\hbox{\fiverm H}}))v_{\hbox{\fiverm K}},
660: v_{\hbox{\fiverm K}}\ra\ .$$
661: According to Theorem 2.1.2 (ii) the function $f$ is well defined and holomorphic.
662: It is clear that $f\not\equiv 0$ as $f(0)=\la v_{\hbox{\fiverm K}},v_{\hbox{\fiverm K}}\ra>0$.
663: But (2.1.3) implies that $f^{(n)}(i)=0$ for all $n\in \N_0$; a contradiction
664: to $f\not\equiv 0$. \qed
665:
666: In the sequel we write $v_{\hbox{\fiverm H}}$ for the $H$-fixed distribution vector
667: obtained from $v_{\hbox{\fiverm H}}^\omega$. We will call
668: $v_{\hbox{\fiverm H}}$ the {\it holomorphic $H$-spherical distribution vector} of $(\pi, {\cal H})$
669: corresponding to $v_{\hbox{\fiverm K}}$.
670:
671:
672: \Remark 2.1.4. By Theorem 2.1.3 we have
673: $$v_{\hbox{\fiverm H}}^\omega=\hbox{w}-\lim_{t\nearrow 1} \pi^0(a_t)v_{\hbox{\fiverm K}}, $$
674: i.e. $v_{\hbox{\fiverm H}}^\omega$ is the weak limit of $\pi^0(a_t)v_{\hbox{\fiverm K}}$ for $t\to 1$ in the
675: locally convex space ${\cal H}^{-\omega}$.
676: It is possible to strengthen this convergence: Let $v\in {\cal H}^\omega$ and $C\subeq G$
677: a compact subset. Then for all $\eps>0$ there exists $0<s<1$ such that for all
678: $0<s<t<1$:
679:
680: $$\sup_{g\in C} |\la v_{\hbox{\fiverm H}}^\omega, \pi(g)v\ra -\la \pi^0(a_t)v_{\hbox{\fiverm K}}, \pi(g)v\ra|<\eps\
681: . \leqno(2.1.4)$$
682: In fact this follows from a simple modification of the first part of the
683: proof of Theorem 2.1.3: we only have to observe that for $v\in {\cal H}^\omega$
684: there exists a $0<\delta<1$ such that $\pi(a_\delta)\pi(g)v$ exists for all
685: $g\in C$.
686:
687:
688: \par Supported by calculations in the rank one case we conjecture
689: that one actually has
690: $ \pi(a_t)v_{\hbox{\fiverm K}}\to v_{\hbox{\fiverm H}}$ in ${\cal H}^{-\infty}$ weakly (and hence strongly by the
691: Banach-Steinhaus Theorem which applies as ${\cal H}^\infty$ is a Fr\'echet space ). \qed
692:
693: Let us illustrate the situation by the discussion of one example.
694:
695: \Example 2.1.5. Here we will determine an explicit analytic description
696: of $v_{\hbox{\fiverm H}}$ for unitary principal series of $G=\Sl(2,\R)$.
697: Let $(\pi_\lambda, {\cal H}_\lambda)$ denote a unitary spherical
698: principal series of $G$ with parameter $\lambda\in i\a^*$.
699: Then $\pi_\lambda^0=\pi_\lambda$ as $\pi_\lambda$ is unitary. In the sequel
700: we will identify $\a_\C^*$ with $\C$ in such a way that $\rho\in \a^*$ corresponds to $1$.
701: With our choice of $\a$ to be
702: $$\a=\left\{ \pmatrix{s & 0\cr 0& -s\cr}\: s\in \R\right\}$$
703: this identification is given by
704: $$\lambda \mapsto \lambda \pmatrix{1 & 0 \cr 0 & -1\cr}\, .$$
705:
706: \par We will use the noncompact realization of ${\cal H}_\lambda =L^2(\R)$ of
707: $\pi_\lambda$. Then for $g=\pmatrix {a & b\cr c& d\cr}\in G$ the operator
708: $\pi_\lambda(g)$ is given by
709:
710: $$(\pi_\lambda(g)f)(x)=|bx+d|^{-1-\overline{\lambda}} f\left({ax+c\over bx +d}\right)
711: \qquad (f\in L^2(\R), x\in \R)\ .$$
712: A normalized $K$-spherical vector is then given by
713:
714: $$v_{\hbox{\fiverm K}}(x)={1\over \sqrt\pi} (1+x^2)^{-{1\over 2}(1+\overline{\lambda})}\ .$$
715: \par Notice that $H=\SO(1,1)$ and (up to sign) we have
716: $X_{\hbox{\fiverm H}}=\pmatrix {\pi\over 4 & 0\cr 0& -{\pi\over 4}\cr}$.
717: Thus for $0\leq t<1$ the element $a_t\in
718: \exp(i\Omega)$ is given by
719:
720: $$a_t=\pmatrix{ e^{i{\pi\over 4}t} & 0 \cr 0 & e^{-i{\pi\over 4}t} \cr}\ .$$
721: Then we have
722: $$v_{\hbox{\fiverm H}}^\omega=\hbox{w}-\lim_{t\nearrow 1} \pi_\lambda(a_t)v_{\hbox{\fiverm K}}$$
723: or
724: $$v_{\hbox{\fiverm H}}^\omega(x)=\hbox{w}- \lim_{t\nearrow 1}{e^{i{\pi\over 4}t
725: (1+\overline{\lambda})}\over \sqrt\pi} (1+e^{i\pi t}x^2)^{-{1\over 2}(1+\overline{\lambda})}\ .$$
726: A simple calculation then shows that $v_{\hbox{\fiverm H}}^\omega$ and $v_{\hbox{\fiverm H}}$
727: are given by the locally integrable function
728:
729: $$v_{\hbox{\fiverm H}} (x)=\cases {{e^{i{\pi\over 4}(1+\overline{\lambda})}\over \sqrt\pi}
730: (1- x^2)^{-{1\over 2}(1+\overline{\lambda})} & for $|x|<1$, \cr
731: 0 & for $|x|=1$, \cr
732: {e^{-i{\pi\over 4}(1+\overline{\lambda})} \over \sqrt \pi}
733: (x^2-1)^{-{1\over 2}(1+\overline{\lambda})} & for $|x|>1$\ .\cr} $$
734: A basis of $({\cal H}_\lambda^{-\infty})^H$ is
735: given by $v_{\hbox{\fiverm H,1}}, v_{\hbox{\fiverm H, 2}}$ where
736: $$v_{\hbox{\fiverm H,1}}(x)=\cases { {1\over \sqrt\pi} (1- x^2)^{-{1\over 2}(1+\overline{\lambda})} & for $|x|<1$, \cr
737: 0 & for $|x|\geq 1$. \cr}$$
738: and
739: $$v_{\hbox{\fiverm H,2}}(x)=\cases { {1\over \sqrt\pi}(x^2-1)^{-{1\over 2}(1+\overline{\lambda})} & for $|x|>1$, \cr
740: 0 & for $|x|\leq 1$. \cr}$$
741: These two basis vectors are chosen such that they have support in the
742: open $H$-orbits on $\P^1(\R)$, namely $]-1, 1[$ and $\P^1(\R)\bs [-1, 1]$.
743: Notice that $v_{\hbox{\fiverm H}}$ is a non-trivial linear combination of $v_{\hbox{\fiverm H, 1}}$ and
744: $v_{\hbox{\fiverm H,2}}$ and that
745: $v_{\hbox{\fiverm H}}$ has full support on $\R$.
746: Another interesting feature of this example is that we have here
747: $$v_{\hbox{\fiverm H}} =\hbox{w}-\lim_{t\nearrow 1}
748: \pi_\lambda(a_t)v_{\hbox{\fiverm K}} \qquad \hbox{in ${\cal H}_\lambda^{-\infty}$}$$
749: and hence also strongly by the Banach-Steinhaus Theorem (compare with the conjecture
750: stated at the end of Remark 2.1.4) . \qed
751:
752:
753:
754:
755:
756:
757: \subheadline{2.2. Holomorphic extension of matrix coefficienst on $G/H$ to $\Xi_H$}
758:
759: \noindent
760: In this subsection we clarify the role of the holomorphic distribution vector
761: $v_{\hbox{\fiverm H}}$ in view of holomorphic extensions of matrix coefficients from
762: $G/H$ to $\Xi$. If $U$ is a complex manifold, then we denote by
763: ${\cal O}(U)$ the space of holomorphic functions $f:U\to \C$.
764:
765: \Definition 2.2.1. Let $f$ be a continuous function on $G/H$. Then we say
766: that $f$ has a {\it holomorphic extension} to $\Xi_H$ if there exists an
767: $\tilde f\in {\cal O}(\Xi_H)$ such that for all compact subsets
768: $C\subeq G$ one has
769: $$\lim_{t\nearrow 1} \sup_{g\in C} |f(gH) -\tilde f(ga_tK_\C)|=0\ .\leqno(2.2.1)$$
770: \qed
771:
772: Notice that (2.2.1) implies that
773: $$f(gH)=\lim_{t\nearrow 1} \tilde f(ga_tK_\C)$$
774: for all $g\in G$.
775: Furthermore the holomorphic extensions are unique by the following lemma:
776:
777:
778: \Lemma 2.2.2. {\rm\bf (Identity Theorem for holomorphic extensions)} Let $f\in C(G/H)$
779: and assume that $f$ has
780: a holomorphic extension $\tilde f\in {\cal O}(\Xi_H)$. Then $f\equiv 0$ implies
781: $\tilde f\equiv 0$. In particular, the holomorphic extension $f\in C(G/H)$
782: is unique if it exists.
783:
784: \Proof. This is easily reduced to the one-dimensional case as
785: follows. Define an abelian tube domain
786: $T=\exp(\R X_{\hbox{\fiverm H}} + ]-1,1[ iX_{\hbox{\fiverm H}})$ and set $\partial_s
787: T =\exp(\R X_{\hbox{\fiverm H}} + iX_{\hbox{\fiverm H}})$. We realize
788: $T\subeq \Xi_H$ and $\partial_sT\subeq G/H$ through the $T$-orbit , respectively $\partial_s T$-orbit,
789: through $K_\C\in \Xi_H$, respectively $z_{\hbox{\fiverm H}}\in G/H$.
790: Let $f\in C(G/H)$ and assume that $f$ has a holomorphic extension $\tilde f$.
791: Let $\phi=f\res_{\partial_s T}$. Then $\phi$ has a holomorphic extension to
792: $T$ given by $\tilde \phi=\tilde f\res_T$. By the well known one-dimensional
793: situation we have $\tilde\phi\equiv 0$ if $\phi\equiv 0$. Thus if $f\equiv 0$, we obtain
794: $\tilde f\res_T\equiv 0$. Replacing $f$ by $f_g$ where $f_g(xH)=f(gxH)$ for $g\in G$, then
795: the above discussion implies that
796: $\tilde f\res_{GT}\equiv 0$ and hence $\tilde f\equiv 0$ as $GT$ contains the totally
797: real submanifold $G/K$ of $\Xi_H$. \qed
798:
799: We assume that the complex conjugation $\g_\C\to\g_\C, \ X\mapsto \oline X$ with respect to
800: the real form $\g$ of $\g_\C$ lifts to a conjugation $g\mapsto \oline g$ of $G_\C$.
801: Notice that this is always satisfied if $G_\C$ is the universal complexification of $G$.
802: Notice also that $\overline{x}\in \tilde{\Xi}$ for all $x\in\tilde{\Xi}$.
803: Let $u,v\in {\cal H}_K$. Then the functions
804: $$\tilde{\Xi}\ni x\mapsto\la \pi^0(x)u,v\ra , \la u,\pi(\overline{x}^{-1})v \ra\in \C $$
805: are well defined and holomorphic by Theorem 2.1.2. Both of them agree if $x\in G$ and
806: hence they agree on all of $\tilde{\Xi}$.
807: %The same argument using Theorem 2.1.3 and (2.1.4) we then obtain
808: %the following extension result:
809:
810: \Proposition 2.2.3. Let $(\pi, {\cal H})$ be a admissible $K$-spherical representation of $G$
811: with finite length. Let $v\in {\cal H}^\omega$. Then the following assertions hold:
812:
813: \item{(i)} The
814: matrix coefficient
815: $$f_{v, v_{\hbox{\fiverm H}}}\: G/H\to \C, \ \ gH\mapsto \la \pi (g^{-1})v, v_{\hbox{\fiverm H}}\ra$$
816: admits a holomorphic extension $\tilde f_{v, v_{\hbox{\fiverm H}}}$.
817: Moreover, we have
818: $$\tilde f_{v, v_{\hbox{\fiverm H}}}(xK_\C)=\la v, \pi^0(\overline x)v_{\hbox{\fiverm K}}\ra $$
819: for all $xK_\C\in \Xi_H$.
820: \item{(ii)} The
821: matrix coefficient
822: $$g_{v_{\hbox{\fiverm H}},v }\: G/H\to \C, \ \ gH\mapsto \la \pi^0(g)v_{\hbox{\fiverm H}},v\ra$$
823: admits a holomorphic extension $\tilde g_{v_{\hbox{\fiverm H}},v}$.
824: Moreover, we have
825: $$\tilde g_{v_{\hbox{\fiverm H}},v}(xK_\C)=\la \pi^0(x )v_{\hbox{\fiverm K}},v\ra \quad\hbox{\rm and}
826: \quad \tilde{g}_{v_{\hbox{\fiverm H}},v}(xK_\C)
827: =\overline{\tilde f_{v, v_{\hbox{\fiverm H}}}(\overline{x}K_\C)}$$
828: for all $xK_\C\in \Xi_H$.%\qed
829:
830: \Proof. We will only show (i) as the proof for (ii) is the same. By Theorem 2.1.2 (i) it follows that
831: $$\tilde{f}(xK_\C)=\la v,\pi^0(\bar{x})v_{\hbox{\fiverm K}}\ra$$
832: exists and is holomorphic on $\Xi_H$. Let $g\in G$ and $0<t<1$. Then
833: $$\la \pi (g^{-1})v,\pi^0(a_t)v_{\hbox{\fiverm K}}\ra =\la v,\pi^0(ga_t)v_{\hbox{\fiverm K}}\ra
834: =\tilde{f}(ga_tK_\C)\, .$$
835: Taking the limit at $t\to 1$ and using the remarks just before (2.1.4) it follows that
836: $$\lim_{t\nearrow 1} \tilde{f}(ga_tK_\C)=\la \pi (g^{-1})v,v_{\hbox{\fiverm H}}\ra$$
837: and the convergence is uniform on compact subsets in $G$.
838: \qed
839:
840: Denote by $({\cal H}^{-\infty})^H_{\rm hol}\subset ({\cal H}^{-\infty})^H$
841: the space of $H$-invariant distribution vectors $\eta$ such
842: that the function
843: $$G/H\ni x \mapsto g_{\eta ,v}(x):=\la \pi^0(x)\eta ,v\ra$$
844: has a holomorphic extension to $\Xi_H$
845: for all $v\in {\cal H}^\omega$. Notice that for $g\in G$ and $x\in \Xi_H$ we have
846: $$\tilde g_{\eta ,\pi (g)v}(x)=\tilde g_{\eta ,v}(g^{-1}x)\, .\leqno(2.2.2)$$
847: Our next task is to prove a converse of Proposition 2.2.3, namely that
848: the map
849: $${\cal H}^K\ni v_{\hbox{\fiverm K}}\mapsto v_{\hbox{\fiverm H}}\in ({\cal H}^{-\infty})^H_{\rm hol}$$
850: is an isomorphism. In particular only the $H$-invariant distribution
851: vectors constructed in Theorem 2.1.3 have holomorphic extension. Thus if
852: $\dim {\cal H}^K=1$, as in the case of the principal series
853: representations of $G$, the space $({\cal H}^{-\infty})^H_{\rm hol}$ is
854: also one-dimensional, i.e., there is (up to scalar) a unique $H$-spherical distribution vector which allows
855: holomorphic extension of the smooth matrix coefficients.
856: \par
857: As before we consider the pairing
858: $$ ({\cal H}^{-\infty})^H\times {\cal H}^\infty \to C^\infty(G/H), \ \ (\eta,v)\mapsto
859: g_{\eta,v}; \ g_{\eta,v}(gH)=\la \pi^0(g)\eta,v\ra\ .$$
860:
861:
862: \Theorem 2.2.4. Let $(\pi , {\cal H})$ be an admissible representation of $G$ with finite length. Then the map
863: $${\cal H}^K\ni v_{\hbox{\fiverm K}}\mapsto v_{\hbox{\fiverm H}}\in ({\cal H}^{-\infty})^H_{\rm hol}$$
864: is a linear isomorphism.
865:
866: \Proof. It follows from Theorem 2.1.3 that ${\cal H}^K\ni v_{\hbox{\fiverm K}}\mapsto
867: v_{\hbox{\fiverm H}}\in({\cal H}^{-\infty})^H_{\rm hol}$ is well defined and injective.
868: It is also clear that the map is linear. It remains to show that the map is onto.
869: For that let $\eta\in ({\cal H}^{-\infty})^H_{\rm hol}$. Define a conjugate linear map
870: $\tilde\eta : {\cal H}_K\to \C$ by
871: $$\tilde\eta(u)= \tilde g_{\eta ,u}(K_\C)\, .$$
872: Then it follows from (2.2.2) that $\tilde\eta$ is $K$-invariant. Thus we find a unique
873: $v_{\hbox{\fiverm K}}\in {\cal H}_K$ such that
874: $\tilde\eta(u)=\la v_{\hbox{\fiverm K}}, u\ra$ for all $u\in {\cal H}_K$.
875: In particular, it follows that
876: $$(\forall u\in {\cal H}_K)\qquad
877: \tilde g_{\eta ,u}(K_\C)= \tilde g_{v_{\hbox{\fiverm H}} ,u}(K_\C)\ .\leqno(2.2.3)$$
878: \par Fix $w\in {\cal H}_K$. We claim that $\tilde g_{\eta ,w}=\tilde g_{v_{\hbox{\fiverm H}} ,w}$.
879: In fact, it follows from (2.2.2) and (2.2.3) that all derivatives
880: of $\tilde g_{\eta ,w}$ and $\tilde g_{v_{\hbox{\fiverm H}} ,w}$ coincide at $K_\C \in\Xi_H$.
881: Thus $\tilde g_{\eta ,w}=\tilde g_{v_{\hbox{\fiverm H}} ,w}$ by Taylor's Theorem.
882:
883: \par {} It follows from our claim that $g_{\eta ,w}=g_{v_{\hbox{\fiverm H}} ,w}$ for all
884: $w\in {\cal H}_K$. In particular, we obtain that
885: $$(\forall w\in {\cal H}_K)\qquad
886: \la \eta, w\ra =g_{\eta,w}(H)=g_{v_{\hbox{\fiverm H}}, w}(H)=\la v_{\hbox{\fiverm H}}, w\ra\ , $$
887: and so $\eta=v_{\hbox{\fiverm H}}$, concluding the proof of the theorem.\qed
888:
889:
890: \Corollary 2.2.5. {\rm\bf (Multiplicity one)} Let $(\pi, {\cal H})$ be an irreducible $K$-spherical
891: Hilbert representation of $G$. Then $\dim ({\cal H}^{-\infty})_{\rm hol}^H =1$.
892:
893: \Proof. According to [H84, Ch.\ IV, Th. 4.5(iii)] we have that $\dim {\cal H}^K=1$.
894: Thus the assertion follows from Theorem 2.2.4.\qed
895:
896:
897:
898:
899: \subheadline{2.3. Distributional characters and boundary values}
900:
901: \noindent
902: Denote by $dg$ and $dh$ Haar measures on $G$ and $H$. Notice
903: that both $G$ and $H$ are unimodular and hence a left Haar measure is
904: also a right Haar measure. Denote by $dgH$ a invariant measure on $G/H$, which
905: we will normalize in a moment.
906: Recall that the mapping
907: $$C_c^\infty (G)\to C_c^\infty (G/H), \ \ f\mapsto f^H; \ f^H(xH)=\int_H f(xh)\ dh$$
908: is continuous and onto. We will normalize the measure $dgH$ in such a way that
909: $$\int_G f(g) \ dg =\int_{G/H} f^H(gH)\ dgH$$
910: holds for all $f\in C_c(G)$.
911:
912:
913: \par In this section $(\pi, {\cal H})$ will denote a {\it unitary}
914: admissible representation representation of $G$ with finite length.
915: Further we will assume that ${\cal H}^K\not=\{0\}$. Notice that
916: $\pi$ unitary implies $\pi=\pi^0$.
917: \par For $f\in C_c^\infty(G)$ let us recall
918: the mollifying property: $\pi(f){\cal H}^{-\infty}\subeq
919: {\cal H}^\infty$. Thus the mapping
920: $$\Theta_{\pi,v_{\hbox{\fiverm H}}} \: C_c^\infty(G)\to \C, \ \ f\mapsto
921: \la \pi(f)v_{\hbox{\fiverm H}}, v_{\hbox{\fiverm H}}\ra\, .$$
922: is well defined. It is known that $\Theta_\pi=\Theta_{\pi,v_{\hbox{\fiverm H}}}$ is a $H$-bi-invariant
923: positive definite distribution on $G$. The $H$-bi-invariance implies
924: that $\Theta_\pi(f)$ does only depend on $f^H$. We can therefore define a $H$-invariant
925: distribution on $G/H$, also denoted by $\Theta_\pi$, by
926: $$\Theta_\pi (f^H)=\Theta_\pi(f)\, .$$
927:
928: \par On the other hand we notice that $\pi (x)v_{\hbox{\fiverm K}}\in {\cal H}^\omega$
929: for all $x\in G\exp (i\Omega_H)K_\C$ and hence
930: $x\mapsto \la v_{\hbox{\fiverm H}},\pi (x)v_{\hbox{\fiverm K}}\ra$
931: descends to a well defined and anti-holomorphic function on $\Xi_H$.
932: Here $v_{\hbox{\fiverm K}}$ and $v_{\hbox{\fiverm H}}$ correspond to each other
933: according to Theorem 2.2.4. We can therefore define the holomorphic function
934: $\theta_\pi=\theta_{\pi,v_H}:\Xi_H\to \C$ by
935: $$\theta_\pi(xK_\C)= \overline{\la v_{\hbox{\fiverm H}}, \pi(x)v_{\hbox{\fiverm K}}\ra }=\la \pi (x)v_{\hbox{\fiverm K}} ,
936: v_{\hbox{\fiverm H}}\ra \, .$$
937: Then $\theta_\pi$ is left $H$-invariant.
938:
939:
940: Our next aim is to show that $\Theta_\pi$ is given by the limit
941: operation and convolution:
942:
943: $$\Theta_\pi(f)=\lim_{t\nearrow 1}\int_{G/H}
944: f(gH)\ \overline{\theta_\pi(g^{-1}a_t)} \ dgH$$
945: for suitable regular functions $f$.
946:
947: \par
948: As we have only established the convergence $\pi(a_t)v_{\hbox{\fiverm K}}\to v_{\hbox{\fiverm H}}$ in
949: ${\cal H}^{-\omega}$ and not in ${\cal H}^{-\infty}$, we cannot work
950: with test-functions but must use an appropriate space of
951: analytic vectors.
952: For that let us write
953: $L^1(G)^{\omega, \omega}$ for the space
954: of analytic vectors for the left-right regular representation of $G\times G$ on $L^1(G)$.
955: Notice that $L^1(G)^{\omega, \omega}$ is an algebra under convolution
956: which is invariant under the natural involution $f\mapsto f^*$ with $f^*(x)=\oline {f(x^{-1})}$.
957: Its importance lies in the fact that $\pi(f){\cal H}^{-\omega}\subeq {\cal H}^\omega$ holds for all
958: $f\in L^1(G)^{\omega,\omega}$ (cf.\ Proposition A.4.1 in the appendix).
959:
960: \par Write $L^1(G/H)^\omega$ for the space of analytic vectors
961: for the left regular representation of $G$ on $L^1(G/H)$.
962: According to Proposition A.3.2 below, the averaging map $f\mapsto f^H$ maps
963: $L^1(G)^{\omega,\omega}$ into $L^1(G/H)^\omega$.
964: Finally let us define the space:
965: $${\cal A}^1(G/H)=\{f^H\in L^1(G/H)^\omega\mid f\in L^1(G)^{\omega,\omega}\}\ . $$
966:
967: \par {}From our discussion above we conclude that the mapping
968: $$\Theta_\pi^\omega\: L^1(G)^{\omega,\omega} \to \C, \ \ f\mapsto \la \pi(f)v_{\hbox{\fiverm H}}, v_{\hbox{\fiverm H}}\ra$$
969: is well defined and $H$ bi-invariant. In particular $\Theta_\pi^\omega(f)$ depends only on $f^H$ and therefore
970: factors to ${\cal A}^1(G/H)$. We denote the corresponding map again by $\Theta_\pi^\omega$.
971:
972:
973:
974: \Theorem 2.3.1. Let $(\pi,{\cal H})$ be an unitary admissible representation
975: of $G$ of finite length with ${\cal H}^K\not=\{0\}$. Let $v_{\hbox{\fiverm K}}\in {\cal H}^K$. Then
976:
977: $$(\forall f\in {\cal A}^1(G/H))\qquad \Theta_\pi^\omega(f)=\lim_{t\nearrow 1}\int_{G/H}
978: f(gH)\ \overline{\theta_\pi(g^{-1}a_t)}\ dgH\ .$$
979:
980: \Proof. Let $F\in L^1(G)^{\omega,\omega}$ be such that $f=F^H$.
981: As $\pi(F)v_{\hbox{\fiverm H}}\in {\cal H}^\omega$ (cf.\ Proposition A.4.1), it follows from
982: Theorem 2.1.3 that
983: $$\Theta_\pi^\omega(F)=\lim_{t\nearrow 1}\la \pi(F)v_{\hbox{\fiverm H}}, \pi(a_t)v_{\hbox{\fiverm K}}\ra\ .$$
984: Thus:
985:
986: $$\eqalign{\Theta_\pi^\omega(F)
987: &=\lim_{t\nearrow 1}\la v_{\hbox{\fiverm H}}, \int_G \oline{F(g^{-1})}\pi(g)\pi(a_t)v_{\hbox{\fiverm K}}\ dg\ra\cr
988: &=\lim_{t\nearrow 1}\int_G F(g^{-1})\ \la v_{\hbox{\fiverm H}},\pi(g)\pi(a_t)v_{\hbox{\fiverm K}}\ra \ dg\cr
989: &=\lim_{t\nearrow 1}\int_G F(g)\ \la v_{\hbox{\fiverm H}},\pi(g^{-1})\pi(a_t)v_{\hbox{\fiverm K}}\ra \ dg\cr
990: &=\lim_{t\nearrow 1}\int_G F(g) \ \overline{\theta_\pi(g^{-1}a_t)}\ dg\ .\cr}\leqno (2.3.1)$$
991: Fix $0\leq t<1$. We claim that $g\mapsto \theta_\pi(g^{-1}a_t)$ is a bounded
992: function on $G$. Indeed, for $g\in G$ we have
993: $$\eqalign{|\theta_\pi(g^{-1}a_t)|&=
994: |\la \pi(g^{-1}a_t)v_{\hbox{\fiverm K}},v_{\hbox{\fiverm K}}\ra|=
995: |\la \pi(a_t)v_{\hbox{\fiverm K}},\pi(g)v_{\hbox{\fiverm K}}\ra|\cr
996: &\leq \|\pi(a_t)v_{\hbox{\fiverm K}}\|\cdot
997: \|\pi(g)v_{\hbox{\fiverm K}}\|\leq \|\pi(a_t)v_{\hbox{\fiverm K}}\|\cdot \| v_{\hbox{\fiverm K}}\|\cr}$$
998: and our claim follows from $\|\pi(a_t)v_{\hbox{\fiverm K}}\|<\infty$.
999: Combining (2.3.1) with our claim then yields
1000: $$\eqalign{\Theta_\pi^\omega(F)&=
1001: \lim_{t\nearrow 1}\int_G F(g) \ \overline{\theta_\pi(g^{-1}a_t)}\ dg\cr
1002: & =\lim_{t\nearrow 1}\int_{G/H} f(gH)\ \overline{\theta_\pi(g^{-1}a_t)}\ dgH\ ,\cr}$$
1003: completing the proof of the theorem. \qed
1004:
1005: We finish this subsection by the following simple remark.
1006:
1007: \Lemma 2.3.2. Suppose that $\pi$ is irreducible. Then $\Theta_\pi$ is an eigendistribution
1008: of the algebra $\D(G/H)$ of invariant differential operators on $G/H$.
1009:
1010: \Proof. Recall the surjective homomorphisms ${\cal U}(\g_\C)^{\frak h}
1011: \to \D(G/H)$
1012: and ${\cal U}(\g_\C)^{\frak k}\to \D(G/K)$. We have $x_{\hbox{\fiverm H}}^{-1}H_\C x_{\hbox{\fiverm H}}
1013: =K_\C$. Hence, ${\rm Ad}(x_{\hbox{\fiverm H}}^{-1})$ defines an isomorphism
1014: $${\rm Ad}(x_{\hbox{\fiverm H}}^{-1})\:
1015: {\cal U}(\g_\C)^{\frak h}\to {\cal U}(\g_\C)^{\frak k}\, .$$
1016: \par In order to prove the lemma it is sufficient to show
1017: that $v_{\hbox{\fiverm H}}$ is an eigenvector for each
1018: $d\pi^0(u)$, $u\in {\cal U}(\g_\C)^{\frak h}$.
1019: \par Notice that for each $\tilde u\in {\cal U}(\g_\C)^{\frak k}$
1020: there exists a constant $c(\tilde u)$ such that
1021: $d\pi^0(\tilde u)v_{\hbox{\fiverm K}}
1022: = c(\tilde u) v_{\hbox{\fiverm H}}$.
1023: For $u\in{\cal U}(\g_\C)^{\frak h}$ we now obtain that
1024:
1025: $$\eqalign{d\pi^0(u)v_{\hbox{\fiverm H}}&=
1026: \lim_{t\nearrow 1}d\pi^0(u)\pi^0(a_t)v_{\hbox{\fiverm K}}\cr
1027: &=\lim_{t\nearrow 1}\pi^0(a_t) (d\pi^0(\Ad (a_t^{-1})u)
1028: v_{\hbox{\fiverm K}}\cr
1029: &=\lim_{t\nearrow 1}\pi^0(a_t) (d\pi^0(\Ad (x_{\hbox {\fiverm H}}
1030: )^{-1}u)v_{\hbox{\fiverm K}}\cr
1031: &= c(\Ad(x_{\hbox {\fiverm H}}^{-1})u)v_{\hbox{\fiverm H}}\ .\cr}$$
1032: This completes the proof of the lemma.\qed
1033:
1034: \subheadline{2.4. Principal series representations}
1035:
1036: \noindent
1037: In this section we consider the case where $\pi=\pi_\lambda$ is a {\it spherical principal series representation}.
1038: In particular we will be discuss the dependence of $v_{\hbox{\fiverm H}}$
1039: on the spectral parameter $\lambda$.
1040:
1041: \par Let us first recall some well known facts about the principal series representations.
1042: For $\alpha\in \Sigma$ let $m_\alpha=\dim \g^\alpha$ and
1043: $\rho\:={1\over 2}\sum_{\alpha\in \Sigma^+} m_\alpha \alpha$. Write
1044: $M=Z_K(\a)$ and denote by $\kappa\: G\to K$ and $a\: G\to A$ the projections onto $K$, resp. $A$,
1045: associated to the Iwasawa decomposition $G=NAK$. Note that $a$ and $\kappa$ have
1046: unique holomorphic extension
1047: to $\tilde\Xi$ also denoted by $a$ and $\kappa$ (cf.\ (1.2.1) and [KS01a]). As we are assuming that
1048: $G\subseteq G_\C$ with $G_\C$ simply connected and $H=G^\tau$, we have $M=Z_H(A)$. In particular
1049: $M\subseteq H\cap K$.
1050:
1051: \par Define a minimal parabolic subgroup of $G$ by $P_{\rm min}=MAN$.
1052: For $\lambda\in \a_\C^*$ let
1053: $${\cal D}_\lambda=\{f\in C^\infty(G)\: (\forall g\in G)(\forall man\in P)
1054: \ f(mang)=a^{\rho-\lambda} f(g)\}\ .$$
1055: The group $G$ acts on ${\cal D}_\lambda$ by right translation. Denote the
1056: corresponding representation by $\pi_\lambda^\infty$, i.e., $\pi_\lambda^\infty(g)f(x)=f(xg)$.
1057: Denote by $\tilde{\cal H}_\lambda$ the completion of $\cal D_\lambda$ in the norm
1058: corresponding to the inner product
1059: $$\la f,g\ra =\int_K f(k)\overline{g(k)}\, dk\, .$$
1060: Then $\pi_\lambda^\infty$ extends to a representation $\pi_\lambda$ of $G$ in ${\cal H}_\lambda$.
1061: We refer to $(\tilde {\cal H}_\lambda ,\pi_\lambda)$ as the {\it spherical principal series representation of $G$
1062: with parameter $\lambda$}. The principal series representations $(\pi_\lambda, \tilde {\cal H}_\lambda)$
1063: are admissible and of finite length; they are irreducible for generic
1064: parameters $\lambda$ and unitary if $\lambda\in i\a^*$. It is also well known that
1065: $\tilde{\cal H}_\lambda^\infty =\cal{D}_\lambda$ excusing our above notation $\pi_\lambda^\infty$.
1066:
1067: \par The restriction map $\tilde{\cal{H}}_\lambda\ni f\mapsto f\res_{K}\in L^2(K)$ is
1068: injective by the left $NA$-covariance of $f$. Furthermore $f\res_{K}$ is left $M$-invariant. Hence
1069: the restriction map defines an isometry $\tilde{\cal H}_\lambda\hookrightarrow L^2(M\bs K)$. On
1070: the other hand if $F\in L^2(M\bs K)$ then we can define $f\in \tilde{\cal H}_\lambda$ by
1071: $f(nak)=a^{\rho -\lambda}F(Mk)$. Hence
1072: $\tilde{\cal H}_\lambda \simeq L^2(M\bs K)$. In this realization we
1073: have
1074: $$[\pi_\lambda(g)f](Mk)=a(kg)^{\rho -\lambda}f(M\kappa (kg))\, .$$
1075: Hence the Hilbert space is the same for all $\lambda$ but the formula for the representation
1076: depends on $\lambda$.
1077: Notice that for $k\in K$ this simplifies to
1078: $(\pi_\lambda(k)f)(Mx)= f(Mxk)$. We write ${\cal H}_\lambda =L^2(M\bs K)$ to indicate
1079: the role of $L^2(M\bs K)$ as the representation space for $\pi_\lambda$ and call it
1080: the {\it compact realization of} $\pi_\lambda$.
1081: As a consequence of this discussion, we see that $v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}=\1_{M\bs K}$ is a
1082: normalized $K$-fixed vector in
1083: ${\cal H}_\lambda$ and in fact ${\cal H}_\lambda^K=\C v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}$.
1084: We write $v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}$
1085: and $v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}^\omega$
1086: instead of $v_{\hbox{\fiverm H}}$ and $v_{\hbox{\fiverm H}}^\omega$ to
1087: indicate the dependence of $\lambda$.
1088:
1089: \par We have $\pi_\lambda^0=\pi_{-\oline \lambda}$. Thus in the compact realization we have
1090:
1091: $$[\pi_\lambda^0(a_t)v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}](Mk)=a(ka_t)^{\rho+\oline \lambda}\leqno(2.4.1)$$
1092: for all $0\leq t<1$ and so
1093: $$\la v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}^\omega, v\ra =\lim_{t\nearrow 1} \int_{M\bs K} a(ka_t)^{\rho+\oline \lambda}
1094: \, \oline{ v(Mk)} \ dMk \qquad (v\in {\cal H}_\lambda^\omega)\ .\leqno(2.4.2)$$
1095: \par In the sequel it will be importnat
1096: that ${\cal H}_\lambda^\omega=C^\omega(M\bs K)$ is independent of $\lambda$.
1097: The following theorem specifies the dependence of $v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}$ on $\lambda$:
1098:
1099: \Theorem 2.4.1. The mapping
1100: $$\a_\C^*\to \coprod_{\lambda\in \a_\C^*} ({\cal H}_\lambda^{-\infty})^H,
1101: \ \ \lambda\mapsto v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}$$
1102: is weakly anti-holomorphic in the sense that for all $v\in C^\omega(M\bs K)$ the mapping
1103: $$\a_\C^*\to\C, \ \ \lambda\mapsto \la v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}, v\ra $$
1104: is anti-holomorphic.
1105:
1106:
1107: \Proof. Let $v\in C^\omega(M\bs K)$. It is convenient to consider
1108: $v$ as an $M$-invariant function on $K$. As $v$ is analytic, there exists
1109: an open $K\times K$-invariant neighborhood ${\cal U}$ of $K$ in $K_\C$ such that
1110: $v$ extends to a holomorphic $M$-invariant function $\tilde v$ on ${\cal U}$.
1111:
1112: \par Let $\eps>0$. Then it follows from the compactness of $Ka_\eps$ and (1.2) that
1113: we can choose $\eps>0$ small enough such that $\kappa(Ka_\eps)\subeq {\cal U}$.
1114:
1115: \par Now consider $v$ as an element of ${\cal H}_\lambda^\omega$. We claim
1116: that $\pi_\lambda(a_\eps)v$ exists. In fact, using our introductory remarks, we have
1117:
1118: $$[\pi_\lambda(a_\eps)v](Mk)=a(ka_\eps)^{\rho-\lambda} \tilde v(\kappa(ka_\eps))\ .$$
1119:
1120: \par With (2.4.1) we now compute
1121:
1122: $$\eqalign{\la v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}, v\ra&=\la \pi_\lambda(a_{1-\eps})^0v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}, \pi_\lambda(a_\eps)v\ra \cr
1123: &=\int_{M\bs K} a(ka_{1-\eps})^{\rho+\oline \lambda}\cdot
1124: \oline {a(ka_\eps)^{\rho-\lambda}}\cdot \oline {\tilde v(\kappa(ka_\eps))}\ dMk\ .\cr}\leqno(2.4.3)$$
1125: By our remarks at the beginning of the proof, we have
1126: $$\sup_{k\in K} |\tilde v(\kappa(ka_\eps))|<\infty\ .\leqno(2.4.4) $$
1127: Notice that (1.2) implies that both $a(Ka_{1-\eps})$ and $a(Ka_\eps)$ are compact
1128: subsets of $T(\Omega)$. Thus, if $C\subeq \a_\C^*$ is a compact subset, then
1129:
1130: $$\sup_{\lambda\in C}\sup_{k\in K} |a(ka_{1-\eps})^{\rho+\oline \lambda}|<\infty,
1131: \qquad\hbox{and}\qquad \sup_{\lambda\in C}\sup_{k\in K} |a(ka_\eps)^{\rho -\lambda}|<\infty
1132: \ .\leqno(2.4.5)$$
1133: Therefore, if we use the estimates (2.4.4) and (2.4.5), it follows from (2.4.3) that
1134: $\lambda\mapsto \la v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}, v\ra $
1135: is anti-holomorphic.\qed
1136:
1137:
1138:
1139:
1140:
1141:
1142:
1143: \subheadline{2.5. Integral representation and asymptotic behaviour of $\theta_\pi$}
1144:
1145: \noindent
1146: Previously we have defined a $H$-invariant holomorphic function
1147: $\theta_\pi$ for unitary representations $\pi$. For non-unitary $\pi$ we define
1148: $\theta_\pi$ by
1149:
1150: $$\theta_\pi(xK_\C)=\la \pi(x)v_{\hbox{\fiverm K}}, v_{\hbox{\fiverm H}}\ra \qquad (xK_\C\in \Xi_H)\ .$$
1151: Clearly, $\theta_\pi$ is a holomorphic function on $\Xi_H$. Moreover, if $\pi$ is unitary, then
1152: $\theta_\pi$ is $H$-invariant.
1153:
1154:
1155: \par In this subsection we will give an integral representation of
1156: the functions $\theta_\pi$ for principal series represntations $\pi$.
1157: This will also allow us to read off the asymptotic behaviour of $\theta_\pi$.
1158:
1159: \ssk Recall the definition of the spherical function $\phi_\lambda$ of parameter
1160: $\lambda\in\a_\C^*$ by
1161:
1162: $$\phi_\lambda(g)=\la \pi_\lambda(g)v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}},
1163: v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}\ra=\int_K a(kg)^{\rho-\lambda}\ dk\qquad (g\in G)\ .$$
1164:
1165: It follows from Theorem 2.1.2 that $\phi_\lambda$ admits a holomorphic extension
1166: to $\tilde \Xi_H$ (or $\Xi_H$ if we wish to consider $\phi_\lambda$ as a function on $G/K$).
1167: Also $\phi_\lambda\res_A$ extends holomorphically to the tube $T(2\Omega)=A\exp(2i\Omega)$.
1168: Notice that
1169: $z_{\hbox{\fiverm H}}\in T(2\Omega)$. All
1170: mentioned holomorphic extensions of $\phi_\lambda$ are also denoted by $\phi_\lambda$.
1171:
1172:
1173:
1174: \par In the sequel we abbreviate and write $\theta_\lambda$ instead
1175: of $\theta_{\pi_\lambda}$. The next result is immediate
1176: from the definitions, Theorem 2.1.2 and the formula (2.4.1).
1177:
1178: \Theorem 2.5.1. Let $(\pi_\lambda, {\cal H}_\lambda)$ be a principal series representation
1179: with parameter $\lambda\in \a_\C^*$. Then for all $xK_\C\in \Xi_H$ we
1180: have
1181: $$\eqalign{
1182: \theta_\lambda(xK_\C)&=\lim_{t\nearrow 1} \int_K a(kx)^{\rho-\lambda} \oline{a(ka_t)^{\rho+\oline \lambda}} \ dk\cr
1183: &=\lim_{t \nearrow 1}\varphi_\lambda (a_tx)\, .}$$
1184: Furthermore,
1185: $$\theta_\lambda(aK_\C)=\phi_\lambda(x_{\hfH}a)$$
1186: for all $a\in T(\Omega)=A\exp (i\Omega)$. Here
1187: $\phi_\lambda$ denotes the holomorphically extended spherical function to $T(2\Omega)$.
1188:
1189: \Proof. Let $x\in\Xi_H$. Then we have
1190: $$\eqalign{
1191: \theta_\lambda (x)&=\la \pi_\lambda (x)v_{\hfK},v_{\hfH}\ra\cr
1192: &=\lim_{t\nearrow 1}\la \pi_\lambda (x)v_{\hfK},\pi^0_{\lambda} (a_t)v_{\hfK}\ra\cr
1193: &=\lim_{t\nearrow 1}\int_K a (kx)^{\rho -\lambda}\, \overline{a(ka_t)^{\rho +\overline{\lambda}}}\, dk\, .\cr }$$
1194: But we can also write the third line as
1195: $$\eqalign{
1196: \theta_\lambda (x)&=\lim_{t\nearrow 1}\la \pi_\lambda (x)v_{\hfK},\pi^0_{\lambda}(a_t)v_{\hfK}\ra \cr
1197: &=\lim_{t\nearrow 1}\la \pi_\lambda(a_tx)v_{\hfK},v_{\hfK}\ra\cr
1198: &=\lim_{t\nearrow 1}\varphi_\lambda (a_t x)
1199: \, .\cr}$$
1200: The last statement follows now from Theorem 2.1.2, part (ii).
1201: \qed
1202: To discuss the asymptotic expansions of $\theta_\lambda$ along a
1203: positive Weyl chamber we first have to recall
1204: some facts on the Harish-Chandra expansion of the spherical functions.
1205: For that let $\a_+=\{ X\in \a\: (\forall \alpha\in \Sigma^+) \ \alpha(X)>0\}$ and set
1206: $A^+=\exp(\a_+)$. Further we define $\Lambda=\N_0[\Sigma^+]$.
1207: If $\mu\in \Lambda$, then we define a meromorphic function
1208: $\Gamma_\mu(\lambda)$ in the parameter $\lambda\in \a_\C^*$ by
1209: $\Gamma_0(\lambda)=1$ and then recursively by
1210:
1211: $$\Gamma_\mu(\lambda)={2\over \la \mu, \mu-\lambda\ra} \sum_{\alpha\in \Sigma^+}
1212: m_\alpha\sum_{k\in \N} \Gamma_{\mu-2k\alpha} \la \mu+\rho-2k\alpha-\lambda,\alpha\ra\ .$$
1213: We call $\lambda\in \a_\C^*$ generic if $\Gamma_\mu(\cdot)$ is holomorphic
1214: at $\lambda$ for all $\mu\in \Lambda$.
1215: For generic $\lambda\in \a_\C^*$ we define the Harish-Chandra $\Phi$-function
1216: on $A^+$ by
1217: $$\Phi_\lambda(a)=a^{\lambda-\rho} \sum_{\mu\in \Lambda} \Gamma_\mu(\lambda) a^{-\mu} \qquad (a\in A^+)\ .$$
1218: This series is locally absolutely convergent. In particular, we see that
1219: $\Phi_\lambda$ extends to a holomorphic function on
1220: $A^+\exp(2i\Omega)\subeq A_\C$ which we also denote by $\Phi_\lambda$.
1221: Finally, with $\c(\lambda)$ the familiar Harish-Chandra $\c$-function on $G/K$, we have
1222: for all generic parameters $\lambda\in \a_\C^*$ that
1223: $$\phi_\lambda(a)=\sum_{w\in {\cal W}} \c(w\lambda)\Phi_ {w\lambda}(a)\qquad (a\in A^+)\ .$$
1224: Combining these facts with Theorem 2.5.1 we now obtain that:
1225:
1226: \Theorem 2.5.2. Let $\lambda\in \a_\C^*$ be a generic parameter. Then the following
1227: assertions hold.
1228:
1229: \item{(i)} For $a\in A^+\exp(i\Omega)$ we have
1230: $$\theta_\lambda(aK_\C)=\phi_\lambda(z_{\hbox{\fiverm H}}a)=\sum_{w\in W} \c(w\lambda) (z_{\hbox{\fiverm H}}a)^{w\lambda
1231: -\rho}\Phi_{w\lambda}(z_{\hbox{\fiverm H}}a)\ .$$
1232: \item{(ii)} Suppose that $\la \Re\lambda,\alpha\ra>0$ for all $\alpha\in \Sigma^+$.
1233: Fix $Y\in\a_+$. Then
1234: $$\lim_{t\to \infty} e^{t(\rho -\lambda)(Y)} \theta_\lambda(\exp(tY)K_\C) =
1235: \c(\lambda)\cdot z_{\hbox{\fiverm H}}^{\lambda -\rho} \ .$$\qed
1236:
1237:
1238: \subheadline{2.6. $H$-orbit coefficients of the holomorphic distribution vector}
1239:
1240: \noindent
1241: As we have remarked already earlier the space $({\cal H}_\lambda^{-\infty })^H$
1242: has dimension $\left |{\cal W}/ {\cal W}_0\right|$ for generic $\lambda$.
1243: One can parametrize $({\cal H}_\lambda^{-\infty })^H$ through the open
1244: $H$-orbits in the flag manifold $P_{\rm min}\bs G$. These orbits haven been parametrized
1245: by Rossmann and Matsuki (cf. [M79]); they are given by
1246: $$P_{\rm min} wH \qquad (w\in {\cal W}/ {\cal W}_0)\ .$$
1247:
1248: \par For $\lambda\in \a_\C^*$ and $w\in {\cal W}/ {\cal W}_0$ define
1249: a right $H$-invariant function on $G$ by
1250:
1251: $$\eta_{\lambda, w}(x)=\cases{ a^{\rho+\oline \lambda} & for $x=manwh\in MAN wH$ \cr
1252: 0 & otherwise\ .\cr}$$
1253:
1254: For $\lambda\in \a^*$ we will use the notation $\lambda <<0$ if
1255: $\la \lambda,\alpha\ra<<0$ for all $\alpha\in \Sigma^+$.
1256: Then it is known that for $\lambda<<0$ the functions $\eta_{\lambda, w}$ are
1257: continuous and define $H$-fixed distribution vectors of $\pi_\lambda$
1258: [\'O87].
1259: Moreover, the distributions $\eta_{\lambda, w}$ admit continuation
1260: in $\lambda$ to a weakly anti-meromorphic function on $\a_\C^*$. For generic $\lambda\in \a_\C^*$
1261: we have
1262: $$({\cal H}_\lambda^{-\infty})^H=\bigoplus_{w\in {\cal W}/{\cal W}_0} \C \eta_{\lambda,w}$$
1263: and the mapping
1264: $$j_\lambda\: \C^{\left |{\cal W}/{\cal W}_0\right|}\to ({\cal H}_\lambda^{-\infty})^H,
1265: \ \ (c_w)_w\mapsto \sum_{w\in {\cal W}/{\cal W}_0} c_w \eta_{\lambda,w}$$
1266: is a bijection for generic $\lambda$, weakly anti-meromorphic in $\lambda$ [vdB88].
1267: For $\lambda<<0$ the inverse of $j_\lambda$
1268: is given by the evaluation mapping
1269: $${\rm ev}\: ({\cal H}_\lambda^{-\infty})^H\to \C^{\left |{\cal W}/{\cal W}_0\right|},
1270: \ \ \eta\mapsto (\eta(w))_w\ .$$
1271:
1272: \par On the other hand we know that the weakly holomorphic distribution
1273: vector $v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}$
1274: depends weakly anti-holomorphically on $\lambda\in \a_\C^*$ (cf.\ Theorem 2.4.1).
1275: The next theorem gives us the coefficients of $v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}$
1276: in terms of the basis $(\eta_{\lambda,w})$ of $({\cal H}_\lambda^{-\infty})^H$.
1277: We note that for $\lambda<<0$ the distribution $v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}$ is given through
1278: the bounded measurable function
1279:
1280: $$v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}(k)=\lim_{t\nearrow 1} a(ka_t)^{\rho+\oline \lambda}\ .\leqno(2.6.1)$$
1281:
1282:
1283: \Theorem 2.6.1. For generic parameters $\lambda\in \a_\C^*$ we have
1284:
1285: $$v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}=\sum_{w\in {\cal W}/ {\cal W}_0}
1286: z_{\hbox{\fiverm H}}^{w^{-1}(\rho +\oline\lambda)}\cdot \eta_{\lambda,w}\ .$$
1287:
1288: \Proof. For $\lambda\in \a_\C^*$ generic write
1289: $$v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}=\sum_{w\in {\cal W}/ {\cal W}_0} c_{\lambda, w} \cdot \eta_{\lambda,w}$$
1290: for the basis expansion. As the coefficients $ c_{\lambda, w}$ depend weakly
1291: anti-meromorphically on $\lambda$, it is sufficient to show that
1292: $c_{\lambda, w}=z_{\hbox{\fiverm H}}^{w^{-1}(\rho
1293: +\oline \lambda)}$
1294: for $\lambda<<0$. Then (2.6.1) implies that
1295:
1296: $$c_{\lambda,w}=v_{\hfH}(w)=\lim_{t\nearrow 1} v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}(wa_t)=a(wz_{\hbox{\fiverm H}})^{\rho+\oline \lambda}=
1297: z_{\hbox{\fiverm H}}^{w^{-1}(\rho+\oline \lambda)}\ , $$
1298: as was to be shown. \qed
1299:
1300: \Remark 2.6.2. Let us go back to Example 2.1.5 for $G=\Sl(2,\R)$. Here
1301: ${\cal W}_0=\{\1\}$ and so ${\cal W}/{\cal W}_0=\{ \1, w\}$ where $w$ is the non-trivial element in
1302: the Weyl group which acts by multiplication by $-1$.
1303: The distributions $v_{\hbox{\fiverm H,1}}$ and $v_{\hbox{\fiverm H,2}}$ from Example 2.1.5 are given in
1304: the above notation by
1305:
1306:
1307: $$v_{\hbox{\fiverm H,1}}=\eta_{\lambda,\1}\qquad\hbox{and}\qquad v_{\hbox{\fiverm H,2}}=\eta_{\lambda,w}\ .$$
1308: In Example 2.1.5 we did show that
1309: $$v_{\hbox{\fiverm H}}=v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}=e^{i{\pi\over 4}
1310: (1+\overline{\lambda})}v_{\hbox{\fiverm H,1}}+ e^{-i{\pi\over 4}(1+\overline{\lambda})} v_{\hbox{\fiverm H,2}}\. $$
1311: As $z_{\hbox{\fiverm H}}^{\rho+\overline{\lambda}}=e^{i{\pi\over 4}(1+\overline{\lambda})}$, we hence see that
1312: the above formula is a special case of Theorem 2.6.1.\qed
1313:
1314:
1315: \subheadline{2.7. Transformation under the intertwining matrix}
1316:
1317: \noindent
1318: For $w\in {\cal W}$ and $\lambda\in \a_\C^*$ generic we have an intertwining operator
1319: $$A(\lambda, w\lambda)\: (\pi_\lambda, {\cal H}_\lambda^\infty)\to (\pi_{w\lambda},
1320: {\cal H}_{w\lambda}^\infty)\ .$$
1321: Notice that the Hilbert space ${\cal H}_\lambda$ is independent of $\lambda$ because
1322: we use the compact realzation. We can therefore speak about meromorphic maps
1323: from $\a_\C^*$ into the space of bounded operators from
1324: ${\cal H}_\lambda$ into ${\cal H}_\mu$. In this sense
1325: it is well known that the map $\a_\C^*\ni \lambda\mapsto A(\lambda, w\lambda)$ is meromorphic.
1326: Dualizing, we obtain an anti-meromorphic family of intertwining operators
1327: $$A(\lambda, w\lambda)^*\: {\cal H}_{w\lambda}^{-\infty}\to {\cal H}_\lambda^{-\infty}\ . $$
1328: Restricting $A(\lambda ,w\lambda)^*$ to the space of $H$-invariant distribution
1329: vectors we obtain a linear bijection, say
1330:
1331: $$A_H(\lambda, w\lambda)^*\: ({\cal H}_{w\lambda}^{-\infty})^H\to ({\cal H}_\lambda^{-\infty})^H\ .$$
1332: Often one refers to $A_H^*(\lambda, w\lambda)$ as the {\it intertwining matrix}.
1333: In terms of the basic distribution vectors $(\eta_{w\lambda, w'})_{w'}$ respectively
1334: $(\eta_{\lambda, w'})_{w'}$
1335: the operator $A_H(\lambda, w\lambda)^*$ has an unknown, seemingly complicated expression.
1336: In this section we will show that the the intertwining matrix
1337: maps the holomorphic distribution vector
1338: $v_{{\hbox{\fiverm H}, \scriptscriptstyle{w\lambda}}}$
1339: to a multiple of $v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}$.
1340: In order to describe this multiple more precisely we need more notations.
1341:
1342: \par For $w\in {\cal W}$ define a subgroup of $\oline N=\theta(N)$ by
1343: $$\oline N_w=\oline N\cap wNw^{-1}\ .$$
1344: For $\Re \lambda<<0$ we define functions
1345: $$\c_w(\lambda)=\int_{\oline N_w} a(\oline n)^{\rho-\lambda} \ d\oline n\ .$$
1346: If $w=w_0$ is the longest element in ${\cal W}$, then we write
1347: $\c(\lambda)$ instead of $\c_{w_0}(\lambda)$ and remark that
1348: $\c(\lambda)$ is the familiar Harish-Chandra $c$-function on $G/K$.
1349: The functions $\c_w(\lambda)$ admit meromorphic continuation to $\a_\C^*$
1350: and can be explicitely computed (Gindikin-Karpelevic formula).
1351:
1352: \par With this notation the intertwinig operators $A(\lambda,w\lambda)$ for $\lambda<<0$
1353: are defined by
1354: $$[A(\lambda, w\lambda)f](x) =\int_{\oline N_w} f(\oline n w x) \ d\oline n
1355: \qquad (f\in {\cal D}_\lambda, x\in G)\ .$$
1356:
1357:
1358: \Theorem 2.7.1. Assume that $\lambda\in \a_\C^*$ is generic. Then
1359: $$A_H(\lambda, w\lambda)^*v_{{\hbox{\fiverm H}, \scriptscriptstyle{w\lambda}}}
1360: = \c_w(\oline\lambda)\cdot v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}\ .$$
1361:
1362:
1363: \Proof. It is well known - and follows immediately from
1364: the definition - that $A(\lambda, w\lambda)v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}=\c_w(\lambda)
1365: v_{{\hfK,w\lambda}}$. But then, as $A(\lambda, w\lambda )^*v_{\hfK,w\lambda}$ is $K$-invariant, we also get
1366: $$\eqalign{
1367: \la A(\lambda ,w\lambda )^*v_{\hfK,w\lambda},v_{\hfK,\lambda}\ra &= \la v_{\hfK,w\lambda},A(\lambda,w\lambda)v_{\hfK,\lambda}\ra\cr
1368: &=\la v_{\hfK,w\lambda },\c_w(\lambda )v_{\hfK,w\lambda}\ra\cr
1369: &=\overline{\c_w (\lambda )}\, .\cr}
1370: $$
1371: Noticing that $\overline{\c_w(\lambda)}=\c_w(\overline{\lambda})$ it follows that
1372: $$A(\lambda ,w\lambda )^*v_{\hfK,w\lambda}=\c_w(\overline{\lambda})v_{\hfK,\lambda}\, .$$
1373: Finally, using that $A(\lambda ,w\lambda )^*$ is an intertwining operator, we get for
1374: a $K$-finite $u$ :
1375: $$\eqalign{
1376: \la A_H^*(\lambda ,w\lambda)v_{\hfH,w\lambda},u\ra &=
1377: \la v_{\hfH,w\lambda},A(\lambda ,w\lambda )u\ra\cr
1378: &=\lim_{t\nearrow 1}\la \pi_{w\lambda }^0(a_t)v_{\hfK,w\lambda},A(\lambda,w\lambda )u\ra\cr
1379: &=\lim_{t\nearrow 1}\la v_{\hfK,w\lambda},\pi_{w\lambda }(a_t)^{-1}A(\lambda,w\lambda )u\ra\cr
1380: &=\lim_{t\nearrow 1}\la v_{\hfK,w\lambda},A(\lambda,w\lambda )\pi_{w\lambda }(a_t)^{-1}u\ra\cr
1381: &=\lim_{t\nearrow 1}\la A(\lambda,w\lambda )^*v_{\hfK,w\lambda},\pi_{w\lambda }(a_t)^{-1}u\ra\cr
1382: &= \c_w(\overline{\lambda})\lim_{t\nearrow 1}\la \pi_{\lambda }(a_t)v_{\hfK,\lambda },u\ra\cr
1383: &=\c_w (\overline{\lambda}) \la v_{\hfH,\lambda },u\ra \, .\cr}
1384: $$
1385: Hence $A_H(\lambda,w\lambda)^*v_{{\hbox{\fiverm H}, \scriptscriptstyle{w\lambda}}}= \c_w(\oline \lambda)\cdot
1386: v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}$, as was to be shown. \qed
1387:
1388:
1389: \subheadline{2.8. Relation to the horospherical picture}
1390:
1391: \noindent
1392: In this subsection we explain the construction of the
1393: holomorphic distribution vector $v_{\hbox{\fiverm H}}$ from
1394: the horospherical point of view.
1395:
1396: \par Let $\kappa\in \a_\C^*$ be a complex parameter.
1397: We call the holomorphic function
1398: $$ \tilde \psi _\kappa(nak)= a^\kappa=\exp(< \kappa , \log a>),
1399: \qquad n\in N_{\Bbb C}, a\in T(\Omega), k\in K_\Bbb C$$ on $\tilde
1400: \Xi $ (see\ (1.2.1)) the {\it holomorphic horospherical function
1401: with parameter $\kappa$}.
1402:
1403: \par This function can be pushed down to $\Xi$ as a holomorphic locally $N_\C$-invariant function
1404: $\psi_\kappa$. The function $\psi_\kappa$ is a holomorphic extension of the usual
1405: horospherical ($N$-invariant, $A$-homogeneous) function on $G/K$,
1406: corresponding to spherical principal representations related to $\kappa$.
1407: For certain values of the parameter $\kappa$ the function $\psi_\kappa$ has boundary distribution values
1408: $\psi_{\kappa, \hbox{\fiverm H}}$ on $G/H$. To
1409: understand the structure of those distributions, let us remark that a Zariski open part
1410: of G/H is the disjoint union of domains $Y_j= NAy_j$ where $y_1,
1411: \cdots, y_k$ correspond to the vertices of $\Omega_H$
1412: (${\cal W}$-equivalent). On each $Y_j$ we have an $N$-invariant, $A$-homogeneous
1413: distribution with parameter $\kappa$ (such
1414: distributions on $Y_j$ are unique up to a multiplicative
1415: constant).
1416:
1417: The function $\tilde \psi_\kappa$ can be also pushed down on a domain
1418: $D_H$ in $N_\C M_\C \bs G_\C$ as a holomorphic function
1419: $\psi_\kappa$. This function with parameters holomorphically
1420: extends $K$-invariant vectors in the principal spherical
1421: representations on $MN\bs G $. The boundary values of this holomorphic
1422: function give an $H$-invariant distribution, the domains $Y_j$
1423: correspond to the $H$-orbits on $MN\bs G$ and we have the corresponding
1424: decomposition of $\psi_\kappa$.
1425:
1426:
1427:
1428:
1429:
1430: \sectionheadline{3. An application: Hardy spaces for NCC symmetric spaces}
1431:
1432: \noindent
1433: In this section we apply our theory
1434: developed in Section 2 to associate
1435: to every NCC symmetric space $G/H$ a Hardy space ${\cal H}^2(\Xi_H)$.
1436: The Hardy space is a $G$-invariant Hilbert space
1437: of holomorphic functions on $\Xi_H$ featuring a boundary value mapping which gives
1438: an isometric embedding of ${\cal H}^2(\Xi_H)$ into the most-continuous
1439: spectrum $L^2(G/H)_{\rm mc}$ of $L^2(G/H)$. Hence we give a
1440: realization of a part of $L^2(G/H)_{\rm mc}$ in a space
1441: of holomorphic function on $\Xi_H$, generalizing and extending our
1442: previous results from [GK\'O01] to all NCC spaces.
1443:
1444: \par This section is organized as follows: After a brief digression on
1445: Hilbert spaces of holomorphic functions on $\Xi_H$, we give
1446: an adhoc definition of the Hardy space through the spectral measure. Then,
1447: after recalling the theory of the most-continuous spectrum, we will show that
1448: there is a boundary value mapping embedding ${\cal H}^2(\Xi_H)$ isometrically
1449: into $L^2(G/H)_{\rm mc}$.
1450:
1451:
1452:
1453: \subheadline{3.1. $G$-invariant Hilbert spaces of holomorphic functions on $\Xi_H$}
1454:
1455: \noindent
1456: In this section we briefly recall the abstract theory of $G$-invariant
1457: Hilbert spaces of holomorphic functions, specialized to the complex manifold $\Xi_H$
1458: (see also [FT99] and [K99] for the general theory).
1459:
1460: \par In the sequel we will consider ${\cal O}(\Xi_H)$ as a Fr\'echet space with the
1461: topology of compact convergence. We let $G$ act on ${\cal O}(\Xi_H)$ by the
1462: left regular representation $L$:
1463: $$(L(g)f)(z)=f(g^{-1}z)\qquad (g\in G, f\in {\cal O}(\Xi_H), z\in \Xi_H)\ .\leqno(3.1.1)$$
1464: By a {\it $G$-invariant Hilbert space
1465: of holomorphic functions on $\Xi_H$} we understand a Hilbert space ${\cal H}\subeq {\cal O}(\Xi_H)$
1466: such that:
1467:
1468:
1469: \ssk
1470: \item{(IH1)} The inclusion ${\cal H}\into {\cal O}(\Xi_H)$ is continuous.
1471: \item{(IH2)} The Hilbert space ${\cal H}$ is invariant under $L$ and the the
1472: corresponding representation of $G$ is unitary.
1473: \ssk
1474: It follows from (IH1) that for every $z\in \Xi_H$ the point
1475: evaluation ${\cal H}\to \C, \ f\mapsto f(z)$,
1476: is continuous. Thus, there exists a ${\cal K}_z\in {\cal H}$ such that
1477: $\la f, {\cal K}_z\ra =f(z)$ holds for every $f\in {\cal H}$. In
1478: this way we obtain a function
1479: $${\cal K}\: \Xi_H \times \Xi_H\to \C, \ \ (z,w)\mapsto
1480: {\cal K}(z,w)=\la {\cal K}_w, {\cal K}_z\ra\ .$$
1481: The function ${\cal K}$ is holomorphic in the first variable and anti-holomorphic in the second variable.
1482: It follows from (IH2) that ${\cal K}$ is $G$-invariant, i.e., ${\cal K}(gz,gw)=
1483: {\cal K}(z,w)$ holds for all $g\in G$ and all $z,w\in \Xi_H$.
1484: We call ${\cal K}$ the {\it Cauchy-Szeg\"o kernel} of ${\cal H}$ and note that
1485: ${\cal H}$ is determined by ${\cal K}$.
1486:
1487: \msk To describe the spectral resolution of ${\cal K}$ denote by $\hat G_s$ the $K$-spherical unitary dual of $G$.
1488: We view $\hat G_s$ as a subset of $\a_\C^*/ {\cal W}$ using the
1489: parametrization of the spherical principal series. Notice that the topology on
1490: $\hat G_s$ coincides with the topology induced from
1491: $\a_\C^*/{\cal W}$.
1492: Slightly abusing our notation from Subsection 2.4, we denote by
1493: $(\pi_\lambda, {\cal H}_\lambda)$ a representative of $\lambda\in \hat G_s$.
1494: For $\lambda\in \hat G_s$ define the $G$-invariant kernel ${\cal K}_\lambda$ by
1495: $${\cal K}_\lambda(xK_\C, yK_\C)=\la \pi_\lambda(\oline y)v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}},
1496: \pi_\lambda(\oline x)v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}\ra
1497: \qquad (xK_\C, yK_\C\in \Xi_H)\ .$$
1498: Write $\Xi_H^{\rm opp}$ for $\Xi_H$ but endowed with the
1499: opposite complex structure. We recall that the map
1500: $$\hat G_s \to {\cal O}(\Xi_H\times \Xi_H^{\rm opp}),
1501: \ \ \lambda\mapsto {\cal K}_\lambda$$
1502: is continuous [KS01b, Sect. 5]
1503: (it follows from the fact that the spherical
1504: functions $\phi_\lambda$ and their holomorphic continuations
1505: are continuous in $\lambda$).
1506: Then by [KS01b, Th.\ 5.1] there exists a unique Borel measure $\mu$ on
1507: $\hat G_s$ such that
1508:
1509: $${\cal K}(z,w)=\int_{\hat G_s} {\cal K}_\lambda(z,w)\ d\mu(\lambda) \qquad (z,w\in \Xi_H)\leqno(3.1.2)$$
1510: with the right hand side converging absolutely on compact subsets of $\Xi_H\times \Xi_H$.
1511: Equivalently phrased, the mapping
1512:
1513: $$\Phi\: \int_{\hat G_s}^\oplus {\cal H}_\lambda\ d\mu(\lambda)\to {\cal H}
1514: , \ \ s=(s_\lambda)_\lambda\mapsto \left(xK_\C\mapsto \int_{\hat G_s} \la \pi_\lambda(x^{-1})s_\lambda,
1515: v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}\ra \ d\mu(\lambda)\right)\leqno (3.1.3)$$
1516: is a $G$-equivariant unitary isomorphism. In the sequel we refer to the measure $\mu$ as
1517: the {\it Plancherel measure} of ${\cal H}$.
1518:
1519: \ssk In [KS01b]
1520: a criterion was given on a Borel measure $\mu$ on $\hat G_s$ to be a Plancherel measure
1521: for an invariant Hilbert space ${\cal H}={\cal H}(\mu)$ on $\Xi$. This criterion can be easily adapted to
1522: invariant Hilbert spaces on $\Xi_H$. Let us provide the necessary modifications.
1523: \par Define a norm $\|\cdot\|_{\hbox{\fiverm H}}$ on $\a_\C^*$ by
1524:
1525: $$\|\lambda\|_{\hbox{\fiverm H}}\:=\sup_{w\in {\cal W}/{\cal W}_0} |\lambda(wX_{\hbox{\fiverm H}})|\qquad (\lambda\in \a_\C^*) \ .$$
1526: Then [KS01b, Prop. 5.4] and its proof readily gives the following generalization:
1527:
1528: \Proposition 3.1.1. Let $\mu$ be a Borel measure on $\hat G_s$ with the property
1529: $$(\forall 0\leq c<2)\qquad \int_{\hat G_s} e^{c\|\Im \lambda\|_{\hbox{\fiverm H}}}\ d\mu(\lambda)<\infty\ .\leqno(3.1.4)$$
1530: Then $\mu$ is the Plancherel measure of an
1531: invariant Hilbert space ${\cal H}(\mu)$ on $\Xi_H$. \qed
1532:
1533: \subheadline{3.2. The definition of the Hardy space}
1534:
1535: \noindent
1536: We are now ready to give the definition of the Hardy space on $\Xi_H$.
1537: Let us denote by $i\a_+^*$ an open Weyl chamber in $i\a^*$. In the sequel we will
1538: consider $i\a_+^*$ mainly as a subset of $\hat G_s$. Let
1539: $\cal O$ be a neighborhood of $i\a^*$ such that $\sum_{w\in {\cal W}/ {\cal W}_0} z_{\hbox{\fiverm H}}^{-2w^{-1}\lambda}$
1540: %Change GO: Added a minus sign
1541: has a holomorphic square root
1542: ${\bf z}_{\hfH}(\lambda)$ on $\cal O$. Define a holomorphic function $\cH$ on
1543: $\cal O$ by
1544: $$\cH (\lambda)=\c(\lambda)\cdot {\bf z}_{\hfH}(\lambda )$$
1545: and define a Borel measure $\mu$ on $i\a_+^*$ by
1546: $$d\mu (\lambda )=\frac{d\lambda}{|\cH (\lambda )|^2} \leqno(3.2.1)$$
1547: where $d\lambda$ denotes the Lebesgue measure. Then we have:
1548:
1549:
1550: \Lemma 3.2.1. The measure $\mu$ satisfies the condition (3.1.4); in particular
1551: $\mu$ is the Plancherel measure of an invariant Hilbert space ${\cal H}(\mu)$ on $\Xi_H$.
1552:
1553:
1554: \Proof. Recall the growth behaviour of the $\c$-function on the imaginary axis: There
1555: exists constants $C, N>0$ such that
1556: $$(\forall \lambda\in i\a^*)\qquad {1\over |\c(\lambda)|^2} \leq C (1+|\lambda|^N)\ . $$
1557: Moreover for $\lambda \in i\a^*$ one has
1558: $z_{\hbox{\fiverm H}}^{\lambda}=e^{\lambda (iX_{\hbox{\fiverm H}})}>0$.
1559: Hence
1560: $$\sum_{w\in {\cal W}/ {\cal W}_0} |z_{\hbox{\fiverm H}}^{-w^{-1}\lambda}|^2\geq
1561: e^{2\|{\rm Im}\lambda\|_{\hbox{\fiverm H}}}\. $$
1562: %Chane GO: I made several changes here in the proof.
1563:
1564: Combining these two facts now yields that $\mu$ satisfies (3.1.4).\qed
1565:
1566: Using Proposition 3.1.1 and Lemma 3.2.1 we now can give an adhoc-definition of the Hardy
1567: space on $\Xi_H$.
1568:
1569: \Definition 3.2.2. {\bf(Hardy space)} Let $G/H$ be a NCC symmetric space and $\Xi_H$ its associated domain
1570: in $G_\C/ K_\C$. Then we define the {\it Hardy space} ${\cal H}^2(\Xi_H)$ on $\Xi_H$ by
1571: $${\cal H}^2(\Xi_H)={\cal H}(\mu)$$
1572: with $\mu$ as in (3.2.1). \qed
1573:
1574:
1575: Recall the Cauchy-Szeg\"o kernel ${\cal K}(z,w)$ of the invariant
1576: Hilbert space ${\cal H}^2(\Xi_H)$ from Subsection 3.1.
1577:
1578: \Lemma 3.2.3. Let ${\cal K}$ be the Cauchy-Szeg\"o kernel of ${\cal H}^2(\Xi_H)$.
1579: Then the limits
1580: $$\Psi(z)=\lim_{t\nearrow 1} {\cal K}(z,a_tK_\C)\qquad (z\in \Xi_H)$$
1581: exist locally uniformly. In particular, $\Psi\:\Xi_H\to \C$
1582: is an $H$-invariant holomorphic function.
1583:
1584: \Proof. Fix $z\in \Xi_H$ and let $U\subeq \Xi_H$ be a compact
1585: neighborhood of $z$. Choose $\eps>0$ small enough such that
1586: $a_\eps U\subeq \Xi_H$. Then, by $G$-invariance, we have
1587: $${\cal K}(z, a_tK_\C)={\cal K}(a_\eps z, a_{-\eps}a_tK_\C)=
1588: {\cal K}(a_\eps z,a_{t-\eps}K_\C)$$
1589: for all $\eps <t<1$. The claim follows now, because
1590: $]\eps, 1+\eps [\ni t\mapsto {\cal K}(a_\eps z,a_{t-\eps}K_\C)\in \C$
1591: is continuous, and hence
1592: $\lim_{t\nearrow 1}{\cal K}(z, a_tK_\C)={\cal K}(a_{\eps} z, a_{1-\eps}K_\C)$
1593: exists and the convergence is uniform on compact subsets.\qed
1594: %Also recall the inequality
1595: %$|{\cal K}(z,w)|\leq \sqrt{{\cal K}(z,z)}\cdot\sqrt{{\cal K}(w,w)}$, valid
1596: %for all positive definite kernels.
1597: %Then we have
1598: %
1599: %$$\eqalign{\left|\Psi(z)\right|&=|\lim_{t\nearrow 1} {\cal K}(z, a_tK_\C)|
1600: %=\left|{\cal K}(a_\eps z, a_{1-\eps}K_\C)\right|\cr
1601: %&\leq \sqrt{{\cal K}(a_\eps z, a_\eps z)} \cdot \sqrt{{\cal K}(a_{1-\eps}K_\C, a_{1-\eps}K_\C )}\ .\cr} $$
1602: %Now the assertion follows from (3.1.2).\qed
1603:
1604: We refer to
1605: $\Psi$ as the {\it Cauchy-Szeg\"o function} of ${\cal H}^2(\Xi_H)$.
1606: As ${\cal K}$ is $G$-invariant, it follows that ${\cal K}$
1607: can be reconstructed from $\Psi$.
1608: Moreover, as $H_\C T(\Omega_{\hbox{\fiverm H}})K_\C/ K_\C$
1609: meets $\Xi_H$ in an open set, we conclude that $\Psi$ is
1610: uniquely determined by its restriction to $T(\Omega_{\hbox{\fiverm H}})K_\C/ K_\C \subeq \Xi_H$.
1611:
1612: Using Theorem 2.5.2 (i) we finally obtain the spectral
1613: resolution of $\Psi$.
1614:
1615:
1616: \Theorem 3.2.4. For $a\in T(\Omega_{\hbox{\fiverm H}})$ we have
1617:
1618: $$\Psi(aK_\C)=\int_{i\a_+^*} \phi_\lambda(z_{\hbox{\fiverm H}}a)
1619: \ {d\lambda\over |\cH(\lambda)|^2}, $$
1620: where the integrals on the right hand side converge uniformly
1621: and absolutely on compact subsets of $T(\Omega_{\hbox{\fiverm H}})$.\qed
1622:
1623:
1624: \ssk We now discuss the boundary value map $b\: {\cal H}^2(\Xi_H)\to
1625: L^2(G/H)_{\rm mc}$. As usual, this boundary value map can be
1626: nicely defined pointwise only on an appropriate dense subspace of ${\cal H}^2(\Xi_H)$.
1627: Write ${\cal H}^2(\Xi_H)^\omega$ for the analytic
1628: vectors of the left regular representation
1629: $(L, {\cal H}^2(\Xi_H))$. Fix $f\in
1630: {\cal H}^2(\Xi_H)^\omega$. Then for every
1631: compact subset $C\subeq G$ there exists
1632: an $0<\eps<1$ such that $L(a_{-\eps}g^{-1})f$ exists
1633: for all $g\in G$. In particular, if $0<\eps\leq t<1$, then
1634: $$f(ga_tK_\C)=f(ga_\eps a_{t-\eps} K_\C)=
1635: [L(a_{-\eps}g^{-1})f)](a_{t-\eps}K_\C)\ .$$
1636: and so
1637: $$\lim_{t\nearrow 1} f(ga_tK_\C)=
1638: [L(a_{-\eps}g^{-1})f)](a_{1-\eps}K_\C)\ .$$
1639: It follows that we have a well defined
1640: $G$-equivariant boundary value map:
1641:
1642: $$b^\omega\: {\cal H}^2(\Xi_H)^\omega\to C(G/H),
1643: \ \ b^\omega(f)(gH)=
1644: \lim_{t\nearrow 1} f(ga_tK_\C)\ \ .\leqno(3.2.2)$$
1645:
1646: \msk Recall from (3.1.3) the isomorphism
1647: $\Phi\:
1648: \int_{i\a_+^*}^\oplus{\cal H}_\lambda\ d\mu(\lambda)
1649: \to {\cal H}^2(\Xi_H)={\cal H}^2(\mu)$:
1650: $$s=(s_\lambda)_\lambda\mapsto \left(xK_\C\mapsto \int_{\hat G_s} \la \pi_\lambda(x^{-1})s_\lambda,
1651: v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}\ra \ d\mu(\lambda)\right)\, .$$
1652: It is
1653: useful to have the corresponding formula for $b^\omega$
1654: on the space of
1655: sections with values in
1656: $\left(\int_{i\a_+^*}^\oplus{\cal H}_\lambda\ d\mu(\lambda)\right)^\omega$, i.e., on
1657: the space of
1658: analytic sections.
1659: In this regard, it is better to replace
1660: ${\cal H}^2(\Xi_H)^\omega$ by some smaller
1661: but dense subspace ${\cal H}^2(\Xi_H)_0$.
1662: In order to define ${\cal H}^2(\Xi_H)_0$
1663: we have to introduce some terminology.
1664: For a section $s=(s_\lambda)_\lambda\in
1665: \int_{i\a_+^*}^\oplus{\cal H}_\lambda\
1666: d\mu(\lambda)$ we define its support
1667: by $\supp(s)=\oline{\{\lambda\in i\a_+^*\: s_\lambda
1668: \neq 0\}}$. Futhermore we shall use
1669: the identifications ${\cal H}_\lambda^\omega=
1670: C^\omega(M\bs K)$ for $\lambda\in i\a_+^*$.
1671: Recall that if $f\in{\cal H}^2(\Xi_H)^\omega$
1672: and $s=(s_\lambda)_\lambda=\Phi^{-1}(f)$, then
1673: almost each stalk $s_\lambda$ is an analytic
1674: vector, i.e. $s_\lambda\in C^\omega(M\bs K)$.
1675: The subspace ${\cal H}^2(\Xi_H)_0$ is
1676: then defined by
1677: $${\cal H}^2(\Xi_H)_0=
1678: \left\{ f\in {\cal H}^2(\Xi_H)^\omega\quad : \quad \eqalign{&
1679: f \quad \hbox {is $K$-finite, }\cr
1680: & s=(s_\lambda)_\lambda=\Phi^{-1}(f)
1681: \quad \hbox{has compact support, }\cr
1682: & s\: i\a_+^*\to C^\omega(M\bs K)
1683: \quad \hbox{is weakly smooth.}}\right\}$$
1684: It is an easy verification that
1685: ${\cal H}^2(\Xi_H)_0$ is a dense subspace
1686: of ${\cal H}^2(\Xi_H)$. Write $b_0\: {\cal H}^2(\Xi_H)_0\to
1687: C(G/H)$ for the restriction of $b^\omega$ to ${\cal H}^2
1688: (\Xi_H)_0$.
1689:
1690: \par In the sequel we will often identify a function
1691: $f\in {\cal H}^2(\Xi_H)_0$ with its corresponding
1692: section $s=(s_\lambda)=\Phi^{-1}(f)$.
1693: We then claim that
1694:
1695: $$b_0\: {\cal H}^2(\Xi_H)_0\to C(G/H), \ \ s=(s_\lambda)
1696: \mapsto\left(gH\mapsto\int_{i\a_+^*}
1697: \la \pi_\lambda(g^{-1})s_\lambda, v_{{\hbox{\fiverm H},
1698: \scriptscriptstyle{\lambda}}}\ra \ d\mu(\lambda)\right)\leqno(3.2.3)$$
1699: Notice that it is a priori not even clear that the
1700: right hand side of (3.2.3) is
1701: well defined. To establish (3.2.3) fix
1702: $f\in {\cal H}^2(\Xi_H)_0$ and $g\in G$.
1703: As $f$ is an analytic
1704: vector for the left regular representation
1705: $(L, {\cal H}^2(\Xi_H))$ it follows that
1706: there exists an $0<\eps<1$ such that $L(a_\eps g^{-1})f$
1707: exists. Using standard procedures one deduces
1708: that
1709: $\pi_\lambda(a_\eps g^{-1})s_\lambda$ exists
1710: for almost all $\lambda$. In particular
1711: $s_\lambda\in {\cal H}_\lambda$
1712: is analytic for almost all $\lambda$.
1713: Furthermore,
1714: $L(a_\eps g^{-1})f$ corresponds to the section
1715: $(\pi_\lambda(a_\eps g^{-1})s_\lambda)_\lambda$ and so
1716: $$\|L(a_\eps g^{-1})f\|^2=\int_{i\a_+^*}
1717: \|\pi_\lambda(a_\eps g^{-1})s_\lambda\|^2 \ d\mu(\lambda)
1718: <\infty\ .\leqno(3.2.4)$$
1719:
1720: With the
1721: convention $\pi_\lambda(a_1)
1722: v_{{\hbox{\fiverm K }, \scriptscriptstyle{\lambda}}}
1723: =v_{{\hbox{\fiverm H }, \scriptscriptstyle{\lambda}}}$
1724: we then have for all $\eps\leq t\leq 1$ and almost all $\lambda$
1725: the estimate
1726: $$\eqalign{|\la \pi_\lambda(g^{-1})s_\lambda,
1727: \pi_\lambda(a_t)v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}\ra|
1728: &= |\la \pi_\lambda(a_\eps g^{-1})s_\lambda,
1729: \pi_\lambda(a_{t-\eps}) v_{{\hbox{\fiverm K},
1730: \scriptscriptstyle{\lambda}}}\ra|\cr
1731: &\leq \|\pi_\lambda(a_\eps g^{-1})s_\lambda\|
1732: \cdot \|\pi_\lambda(a_{t-\eps}) v_{{\hbox{\fiverm K},
1733: \scriptscriptstyle{\lambda}}}\|\cr
1734: &\leq M\cdot \|\pi_\lambda(a_\eps g^{-1})s_\lambda\|\cr}\leqno
1735: (3.2.5)$$
1736: with $M=\sup_{\lambda\in \supp(s)\atop
1737: \eps\leq t\leq 1}
1738: \|\pi_\lambda(a_{t-\eps}) v_{{\hbox{\fiverm K},
1739: \scriptscriptstyle{\lambda}}}\|<\infty$ as $\supp(s)$ is
1740: compact.
1741: \par Recall our notion of holomorphic extension from
1742: Definition 2.2.1. As almost
1743: each stalk $s_\lambda$ is
1744: an analytic vector in ${\cal H}_\lambda$, it follows
1745: from estimates (3.2.4-5) and the compactness
1746: of $\supp (s)$ that
1747: $$\eqalign{\int_{i\a_+^*}
1748: \la \pi_\lambda(g^{-1})s_\lambda,
1749: v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}\ra \ d\mu(\lambda)
1750: &=
1751: \int_{\supp (s)}
1752: \lim_{t\nearrow 1} \la \pi_\lambda(g^{-1})s_\lambda,
1753: \pi(a_t)v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}\ra \ d\mu(\lambda)\cr
1754: &=
1755: \lim_{t\nearrow 1}
1756: \int_{\supp(s)}\la \pi_\lambda(g^{-1})s_\lambda,
1757: \pi_\lambda(a_t)v_{{\hbox{\fiverm K}, \scriptscriptstyle{\lambda}}}\ra \ d\mu(\lambda)\cr
1758: &= b^\omega(f)(gH)\ .\cr}$$
1759: As $f$ and $g$ were arbitray, this completes the proof
1760: of (3.2.3).
1761:
1762:
1763:
1764: \subheadline{3.3. The Plancherel Theorem for $L^2(G/H)_{\rm mc}$}
1765:
1766: \noindent
1767: Before we can show that $b_0$ has image in $L^2(G/H)_{\rm mc}$ and
1768: extends to an isometric embedding, we need to recall some facts about the
1769: most continuous spectrum $L^2(G/H)_{\rm mc}$ of $L^2(G/H)$ (cf.\ [vdBS97a] and [D98]).
1770: All the results collected below are proved in [vdBS97a] or might be considered as special
1771: cases of [D98]. The crucial way where our assumption that $G/H$
1772: is NCC, $H=G^\tau$, and $G\subseteq G_\C$ with $G_\C$ simply connected, enters is the fact
1773: that $Z_H(\a)=Z_K(\a)$ and $H=Z_H(\a )H_0$.
1774:
1775: \ssk Recall from Subsection 2.7 the mapping
1776: $j(\lambda)\: \C^{\left| {\cal W}/ {\cal W}_0\right|}\to ({\cal H}_\lambda^{-\infty})^H$
1777: and the intertwining matrix $A_H(\lambda, w\lambda)^*\: ({\cal H}_{w\lambda}^{-\infty})^H\to
1778: ({\cal H}_\lambda^{-\infty})^H$ both defined for generic $\lambda\in \a_\C^*$, and all $w\in {\cal W}$.
1779: For generic $\lambda$ we define
1780: $$j^0(\lambda)\: \C^{\left| {\cal W}/ {\cal W}_0\right|}\to ({\cal H}_\lambda^{-\infty})^H$$
1781: by
1782: $$j^0(\lambda)\:=[A_H(w_0\lambda, \lambda)^*]^{-1}\circ j(w_0\lambda)\leqno(3.3.1)$$
1783: with $w_0\in {\cal W}$ the longest element. Then $j^0$ has no poles on $i\a^*$
1784: (cf.\ [vdBS97b, Th.\ 1]).
1785: Denote by $(\e_w)_{w\in {\cal W}/ {\cal W}_0}$ the canonical basis of the Hilbert space
1786: $\C^{\left| {\cal W}/ {\cal W}_0\right|}$. For $w\in {\cal W}/{\cal W}_0$ define
1787: $$\eta_{\lambda,w}^0\:=j^0(\lambda) \e_w\in ({\cal H}_\lambda^{-\infty})^H\ .$$
1788: Define a Hilbert space structure on $\Hom (\C^{\left| {\cal W}/ {\cal W}_0\right|}, {\cal H}_\lambda)$
1789: using the identification
1790: $$\Hom (\C^{\left| {\cal W}/ {\cal W}_0\right|}, {\cal H}_\lambda)\simeq {\cal H}_\lambda\otimes
1791: [\C^{\left| {\cal W}/ {\cal W}_0\right|}]^*\ .$$
1792: \par Write ${\cal S}(G/H)$ for the Schwartz space on $G/H$
1793: and $p_{\rm mc}\: L^2(G/H)\to L^2(G/H)_{\rm mc}$ for
1794: the orthogonal projection on the most continuous spectrum.
1795: Set ${\cal S}_{\rm mc}(G/H)=p_{\rm mc}({\cal S}(G/H))$.
1796: Then for functions $f\in {\cal S}_{\rm mc}(G/H)$
1797: the Fourier transform is defined by
1798: $${\cal F}(f)=\left(\pi_\lambda(f)j^0(\lambda)\right )_\lambda \ .\leqno(3.3.2) $$
1799: By [D98, Th.\ 3] or [vdBS97a,Cor. 18.2 and Prop. 18.3],
1800: ${\cal F}$ extends to a $G$-equivariant unitary
1801: isomorphism
1802: $${\cal F}\: L^2(G/H)_{\rm mc} \to \int_{i\a_+^*}^\oplus
1803: \Hom (\C^{\left| {\cal W}/ {\cal W}_0\right|}, {\cal H}_\lambda) \ d\lambda\ . \leqno(3.3.3)$$
1804: In particular, we have (using suitable normalization of measures) that
1805: $$\|f\|^2= \int_{i\a_+^*}
1806: \|{\cal F}(f)(\lambda)\|^2 \ d\lambda \leqno(3.3.4) $$
1807: for all $f\in {\cal S}_{\rm mc}(G/H)$.
1808: \par Next we wish to describe ${\cal F}^{-1}$.
1809: Let $(\e_w^*)_{w\in {\cal W}/ {\cal W}_0}$ be the
1810: dual basis of $(\e_w)_{w\in {\cal W}/ {\cal W}_0}$. Then a section $s$ of
1811: $\int_{i\a_+^*}^\oplus
1812: \Hom (\C^{\left| {\cal W}/ {\cal W}_0\right|}, {\cal H}_\lambda) \ d\lambda$ can be written as
1813: $s=(\sum_{w\in {\cal W}/ {\cal W}_0} s_{\lambda, w}\otimes \e_w^*)_\lambda$
1814: with $s_{\lambda, w}\in {\cal H}_\lambda$ for all $\lambda\in i\a_+^*$ and
1815: $w\in {\cal W}/ {\cal W}_0$. Recall that if
1816: $s=(\sum_{w\in {\cal W}/ {\cal W}_0} s_{\lambda, w}\otimes \e_w^*)_\lambda$ is a smooth vector, then
1817: $s_{\lambda, w}$ is a smooth vector in ${\cal H}_\lambda$
1818: for almost all $\lambda$.
1819: In the sequel we will use the identification
1820: ${\cal H}_\lambda^\infty=C^\infty(M\bs K)$. Define a
1821: subspace of $\left(\int_{i\a_+^*}^\oplus
1822: \Hom (\C^{\left| {\cal W}/ {\cal W}_0\right|}, {\cal H}_\lambda) \ d\lambda\right)^\infty$
1823: by
1824: $$
1825: %\eqalign{
1826: {\cal H}_0=\left\{ s%=(\sum_{w\in {\cal W}/ {\cal W}_0} s_{\lambda, w}\otimes \e_w^*)_\lambda &
1827: \in \left(\int_{i\a_+^*}^\oplus
1828: \Hom (\C^{\left| {\cal W}/ {\cal W}_0\right|},
1829: {\cal H}_\lambda) \ d\lambda\ \right)^\infty\quad :\quad
1830: \eqalign{& s \ \hbox{is $K$-finite,}\
1831: \supp (s) \ \hbox{is compact,}\cr
1832: & s\: i\a_+^*\to \Hom(\C^{\left| {\cal W}/ {\cal W}_0\right|},
1833: C^\infty(M\bs K)) \cr
1834: & \hbox{is weakly smooth}. \cr}\right\}$$
1835: It is not hard to see that ${\cal H}_0$ is
1836: a dense subspace in $\int_{i\a_+^*}^\oplus
1837: \Hom (\C^{\left| {\cal W}/ {\cal W}_0\right|},
1838: {\cal H}_\lambda) \ d\lambda$.
1839: Then for an element $s\in {\cal H}_0$ the inverse
1840: Fourier-transform is given by [D98, Th.\ 3]
1841:
1842: $${\cal F}^{-1}(s)(gH)=\int_{i\a_+^*}\sum_{w\in {\cal W}/ {\cal W}_0} \la \pi_\lambda(g^{-1}) s_{\lambda, w},
1843: \eta_{\lambda, w}^0\ra \ d\lambda\ .\leqno(3.3.5)$$
1844: Moreover [D98, Th.\ 3] implies that
1845: $${\cal F}^{-1}({\cal H}_0)\subeq
1846: {\cal S}_{\rm mc}(G/H)\ ;
1847: \leqno(3.3.6)$$
1848: in particular ${\cal F}({\cal F}^{-1}(s))$ is
1849: given by the formula (3.3.2) for $s\in {\cal H}_0$.
1850:
1851: \Remark 3.3.1. We have normalized the invariant
1852: measure on $G/H$ and the measure $d\lambda$ on ${\frak a}^*$ so that (3.3.3) and
1853: (3.3.4) holds without any additional constants. This is possible, because we
1854: are only working with the principial series of representations
1855: and the most continuous part of the spectrum. In general, one
1856: has to take into account the order of several Weyl groups.
1857: We refer to Theorem 31 and Remark 32 in [vdB00] for general
1858: discussion on the normalization of measures.\qed
1859:
1860: \subheadline{3.4. Isometry of the boundary value mapping}
1861:
1862: \noindent
1863: In this subsection we complete our
1864: discussion of the boundary value mapping begun in Subsection 3.2.
1865:
1866:
1867:
1868: \Theorem 3.4.1. {\rm \bf(Isometry of the boundary value mapping)}
1869: The boundary value mapping, initially defined
1870: by
1871: $$b_0\: {\cal H}^2(\Xi_H)_0\to C(G/H), \ \ b_0(f)(gH)=\lim_{t\nearrow 1}
1872: f(ga_tK_\C)$$
1873: (cf.\ {\rm (3.2.2-3)}) extends to a $G$-equivariant isometric embedding
1874: $$b\: {\cal H}^2(\Xi_H)\to L^2(G/H)_{\rm mc}\ .$$
1875:
1876: \Proof. For each $\lambda\in i\a_+^*$ define a vector
1877: $\b(\lambda)\in (\C^{\left |{\cal W}
1878: / {\cal W}_0\right|})^*$ by
1879: $$\b(\lambda)= \c(w_0\lambda)
1880: \sum_{w\in {\cal W}/ {\cal W}_0} z_{\hbox{\fiverm H}}^{-w^{-1}(w_0\lambda+\rho)}\e_w^*\ .$$
1881: Notice, that for $\lambda\in i\a^*$ we have
1882: $$|z_{\hbox{\fiverm H}}^{-2w^{-1}\lambda}|=
1883: z_{\hbox{\fiverm H}}^{-w^{-1}(\lambda+\rho)}\overline{z_{\hbox{\fiverm H}}^{-w^{-1}(\lambda+\rho)}}
1884: =z_{\hbox{\fiverm H}}^{-w^{-1}(\lambda+\rho)}z_{\hbox{\fiverm H}}^{w^{-1}(\bar{\lambda}+\rho)}\, .$$
1885: Therefore,
1886: employing the Maass-Selberg relation for $\c(\lambda)$ we obtain
1887: $$\|\b(\lambda)\|^2=|\c(\lambda)|^2 \cdot \sum_{w\in {\cal W}/ {\cal W}_0}
1888: |z_{\hbox{\fiverm H}}^{-w^{-1}\lambda}|^2 =|\c_{G/H}(\lambda)|^2\ .$$
1889: In particular we see that we have an
1890: $G$-equivariant isometric embedding of
1891: direct integrals
1892: $$\iota\: \int_{i\a_+^*}^\oplus {\cal H}_\lambda\ d\mu(\lambda)
1893: \to \int_{i\a_+^*}^\oplus \Hom(\C^{\left |{\cal W}
1894: / {\cal W}_0\right|}, {\cal H}_\lambda)\ d\lambda,
1895: \ \ s=(s_\lambda)_\lambda\mapsto (s_\lambda\otimes \b(\lambda))_\lambda\ .$$
1896: {}From the definition of the spaces ${\cal H}^2(\Xi_H)_0$
1897: and ${\cal H}_0$ it is then clear that
1898: $$\iota \left(\Phi^{-1}({\cal H}^2(\Xi_H)_0)\right)\subeq
1899: {\cal H}_0\ .\leqno(3.4.1)$$
1900:
1901: \par As
1902: $$\iota (s)_\lambda = \c (w_0\lambda)\sum_{w\in {\cal W}/{\cal W}_0}
1903: z_{\hbox{\fiverm H}}^{-w^{-1}(w_0\lambda+\rho)}s_\lambda\otimes \e_w^*\ , $$
1904: we get by (3.2.3), (3.3.5), Theorem
1905: 2.6.1 and Theorem 2.7.1 that
1906: $$\eqalign{[{\cal F}^{-1}(\iota (s))](gH) &=\int_{i\a_+^*}
1907: \sum_{w\in {\cal W}/{\cal W}_0}
1908: \c (w_0\lambda)
1909: z_{\hbox{\fiverm H}}^{-w^{-1}(w_0\lambda+\rho)}
1910: \la \pi_\lambda (g^{-1})s_\lambda, \eta^0_{\lambda, w}\ra \ d\lambda\cr
1911: &=\int_{i\a_+^*}
1912: \sum_{w\in {\cal W}/{\cal W}_0}
1913: \la \pi_\lambda (g^{-1})s_\lambda, \c (w_0\bar{\lambda})
1914: z_{\hbox{\fiverm H}}^{w^{-1}(w_0\bar{\lambda}+\rho)}\eta^0_{\lambda, w}\ra \ d\lambda\cr
1915: &=\int_{i\a_+^*}
1916: \la \pi_\lambda (g^{-1})s_\lambda, v_{{\hbox{\fiverm H}, \scriptscriptstyle{\lambda}}}\ra \ d\lambda
1917: \cr
1918: &=b_0(s)(gH) }
1919: $$
1920:
1921:
1922: \par From this and (3.3.6) it follows that
1923: $b_0(f)\in S_{\rm mc}(G/H)$; in particular we
1924: have
1925: $b_0(f)\in L^2(G/H)_{\rm mc}$. Finally,
1926: $$\|b_0(f)\|_{L^2(G/H)_{\rm mc}}=
1927: \|{\cal F}^{-1} (\iota (s))\|=\|\iota(s)\|=\|s\|=
1928: \|\Phi(s)\|=\|f\|$$
1929: as ${\cal F}$, $\iota$ and $\Phi$ are isometric.
1930: This completes the proof of the theorem.\qed
1931:
1932:
1933:
1934: \Remark 3.4.2. The domain $\Xi_H$ is maximal in the sense that
1935: generic functions in ${\cal H}^2(\Xi_H)$ do not extend holomorphically
1936: over $\Xi_H$.\qed
1937:
1938:
1939:
1940:
1941: \subheadline{3.5. Concluding remarks and the example of $G=\Sl(2,\R)$}
1942:
1943: \noindent
1944: In [GK\'O01] we defined a Hardy space ${\cal H}^2(\Xi)$ on $\Xi$ for the
1945: cases where $\Xi=\Xi_H$. Let us briefly summarize its construction in order
1946: to put it into perspective with the results in this section.
1947:
1948: \par Geometrically the situation $\Xi=\Xi_H$ is equivalent to the fact
1949: that $\Xi$ is homogeneous
1950: for a bigger Hermitian group $S\supeq G$ (cf.\ [KS01b]). More precisely, if $U<S$ denotes an appropriate
1951: maximal compact subgroup with $K\subeq U$ then $\Xi$ is $G$-biholomorphic
1952: to the Hermitian symmetric space $S/U$. For example if $G$ is Hermitian, then
1953: $S=G\times G$ and $\Xi\simeq G/K\times \oline{G/K}$.
1954:
1955: \par The assumption $\Xi=\Xi_H$ thus allows us to identify $\Xi$ with a bounded
1956: symmetric domain ${\cal D}\simeq S/U$. Within this identification one shows that
1957: $\partial_d\Xi\simeq G/H$ becomes a Zariski-open subset in the Shilov boundary
1958: $\partial_s{\cal D}$ of ${\cal D}$.
1959:
1960:
1961: \par The identification of $\Xi$ with ${\cal D}$ was used in [GK\'O01] in a crucial way:
1962: One can transfer the action of an appropriate compression-semigroup $\Gamma\supeq G$ on ${\cal D}$
1963: to $\Xi$ and use this to give a definition of a Hardy space as follows:
1964:
1965: $${\cal H}^2(\Xi)=\{ f\in {\cal O}(\Xi)\: \|f\|^2=\sup_{\gamma\in \Int\Gamma}
1966: \int_{G/H} |f(\gamma gz_{\hbox{\fiverm H}})|^2\ dgH<\infty\}\ .\leqno(3.5.1)$$
1967: In [GK\'O01] we have shown -- with entirely different methods -- that the Hardy space defined as
1968: in (3.5.1) has the following properties:
1969: \msk
1970: \item{(3.5.2)} ${\cal H}^2(\Xi)$ is a Hilbert space of holomorphic functions
1971: featuring an isometric boundary value mapping $b\:{\cal H}^2(\Xi)\into L^2(G/H)_{\rm mc}$.
1972: Moreover, $\im b$ is a multiplicity one subspace of {\it full spectrum}.
1973: \item{(3.5.3)} The Hardy space ${\cal H}^2(\Xi)$ is $G$-isometric to
1974: $L^2(G/K)$ through a transform of Segal-Barg\-mann type.
1975: \item{(3.5.4)} ${\cal H}^2(\Xi)$ is $G$-isometric to the classical Hardy space
1976: ${\cal H}^2({\cal D})$ through an explicitely given mapping.
1977: \msk
1978: In particular for $\Xi=\Xi_H$ it follows from Theorem 3.4.1 and
1979: (3.5.2) that the definition of (3.5.1) coincides
1980: with our spectral definition of the Hardy space in Definition 3.2.2. For the
1981: cases where $\Xi\neq \Xi_H$ there is no apparent semigroup action on $\Xi_H$ and a definition
1982: of ${\cal H}^2(\Xi_H)$ in the flavour of (3.5.1) seems presently not possible.
1983:
1984: \par Notice that (3.5.3) implies that the Plancherel measure of ${\cal H}^2(\Xi)$ has
1985: support equal to $i\a_+^*$. However, in [GK\'O01] we could not determine this
1986: measure explicitely. With the new approach given in this section this difficulty
1987: is already taken care of with the definition of the Hardy space.
1988: \par The explicit isomorphism of ${\cal H}^2(\Xi)$ with the classical
1989: Hardy space ${\cal H}^2({\cal D})$ allows us to find also a nice closed
1990: expression for the Cauchy-Szeg\"o function $\Psi$ (cf.\ [GK\'O01, Th.\ 5.7 and
1991: Ex.\ 5.10]). Combining this closed expression with the spectral resolution
1992: of $\Psi$ in Theorem 3.2.4 one obtains interesting identities for
1993: (generalized) hypergeometric functions. For example for $G=\Sl(2,\R)$ one obtains
1994: the following formula:
1995: $${1-\tanh^2 t\over 1+\tanh^2 t} ={\pi\over 2}\int_0^\infty F\left({1\over 4}+i{\lambda\over 4},
1996: {1\over 4}-i{\lambda\over 4}, 1; -\sinh^2\left(2t+i{\pi\over 2}\right)\right)
1997: \cdot\left|{\Gamma\left({i\lambda+1\over2}\right)\over \Gamma\left({i\lambda\over2}\right)} \right|^2
1998: \ {d\lambda\over \cosh {\pi\over 2}\lambda}$$
1999: for all $t\in \R+i]-{\pi\over 4}, {\pi\over 4}[$. Here $F$ denotes the Gau\3 hypergeometric function.
2000:
2001:
2002: \sectionheadline{A. Appendix: Analytic vectors for representations}
2003:
2004: In this appendix we will summarize some facts on analytic vectors
2005: for representations. None of the results collected below is new,
2006: however some of them might be hard to find explicitely in the
2007: literature. In order to keep the exposition short, we will omit
2008: proofs and often do not make the most general assumptions.
2009: A more detailed account containing complete proofs can be found
2010: in the forthcoming survey [K\'O03].
2011:
2012: \subheadline{A.1. Definition and topology of analytic vectors}
2013:
2014: Throughout this appendix $G$ will denote a connected unimodular Lie group
2015: with $G\subeq G_\C$.
2016:
2017: \par Let $E$ be a complex Banach space and
2018: $\Gl(E)$ the group of continuous invertible operators on $E$.
2019: By a (Banach) representation of $(\pi, E)$ of $G$ we will understand
2020: a group homomorphism $\pi\: G\to \Gl(E)$ such that
2021: for all $v\in E$ the orbit mapping
2022: $$\gamma_v\: G\to E, \ \ g\mapsto \pi(g)v$$
2023: is continuous.
2024:
2025: \par A vector $v\in E$ is called {\it analytic} if $\gamma_v$ is an
2026: analytic $E$-valued map or, equivalently, if there exists
2027: an open neighborhood $U$ of $\1$ in $G_\C$ and a $G$-equivariant
2028: holomorphic mapping
2029: $$\gamma_{v,U}\: GU\to E$$
2030: such that $\gamma_{v,U}(\1)=v$. In particular, $\gamma_{v, U}\res_G=\gamma_v$.
2031:
2032: \par The vector space of all analytic vectors for $(\pi, E)$ is denoted
2033: by $E^\omega$. We recall a fundamental result of Nelson which
2034: states that $E^\omega$ is dense in $E$.
2035:
2036: \ssk Next we are going to recall the definition of the topology
2037: on $E^\omega$.
2038: \par For a complex manifold
2039: $M$ let us denote by ${\cal O}(M,E)$ the space of all $E$-valued
2040: holomorphic mappings on $E$. Topologically we consider ${\cal O}(M,E)$
2041: as a Fr\'echet space with the topology of compact convergence.
2042:
2043: \par For any open neighborhood $U$
2044: of $\1$ in $G_\C$ we write $E_U$ for the subspace
2045: of $E^\omega$ for which $\gamma_{v,U}$ exists. Then we obtain a linear
2046: embedding
2047:
2048: $$\eta_U\: E_U\to {\cal O}(GU, E), \ \ v\mapsto \gamma_{v,U}\ .$$
2049: The image of $\eta_U$ is closed and hence ${\cal O}(GU,E)$ induces
2050: a Fr\'echet topology on $E_U$. Notice that for $U_1\subeq U_2$
2051: we obtain a continuous embedding $E_{U_2}\to E_{U_1}$ via restriction.
2052: Thus
2053: $$E^\omega=\lim_{U\to \{1\}} E_U=\bigcup_U E_U $$
2054: and we can equip $E^\omega$ with the inductive limit topology, i.e.
2055: the finest topology on $E^\omega$ for which all inclusion
2056: mappings $E_U\to E^\omega$ become continuous. Notice that this
2057: turns $E^\omega$ into a locally convex topological vector space.
2058:
2059: \par By $E^{-\omega}$ we will denote the antidual of $E^\omega$, i.e.
2060: the space of all antilinear continuous functionals on $E^\omega$.
2061: The space $E^{-\omega}$ is referred to as the space of {\it hyperfunction
2062: vectors} of the representation $(\pi, E)$. We equip
2063: $E^{-\omega}$ with the topology of bounded convergence.
2064:
2065:
2066: \subheadline{A.2. Analytic vectors for $L^1(G/H)$}
2067:
2068: \par Let $H<G$ be a closed subgroup such that $G/H$
2069: carries a $G$-invariant measure. We write $L^1(G/H)$ for the corresponding
2070: Banach space of integrable functions and $(L, L^1(G/H))$ for the left regular representation
2071: of $G$ on $L^1(G/H)$, i.e.,
2072: $$(L(g)f)(xH)=f(g^{-1}xH)\qquad (g,x\in G, f\in L^1(G/H))\ .$$
2073:
2074: Further it is convenient to assume that $G/H\subeq G_\C/ H_\C$.
2075: Then we have the following characterization of the analytic
2076: vectors:
2077:
2078: \Proposition A.2.1. Let $U$ be an open neighborhood of $\1$ in $G_\C$.
2079: Then $f\in L^1(G/H)_U$ if and only if
2080: there exists a holomorphic function $\tilde f$ on the open set
2081: $$U^{-1} GH_\C/ H_\C \subeq G_\C/ H_\C$$
2082: with the following properties:
2083: \item{(1)} $\tilde f\res_{G/H}=f$.
2084: \item{(2)} For all $x\in U$ the map
2085: $$\tilde f_x\: G/H\to\C, \ \ gH\mapsto \tilde f(x^{-1}gH)$$
2086: belongs to $L^1(G/H)$.
2087: \item{(3)} For all compact subsets $U^c\subeq U$ we have
2088: $$\sup_{x\in U^c} \|\tilde f_x\|<\infty\ .$$ \qed
2089:
2090: There are two types of homogeneous spaces $G/H$ which will
2091: be of particular interest for us. The first is when $H={\1}$. Then $L^1(G)^\omega$
2092: denotes the analytic for the left regular representation of $G$ on $L^1(G)$.
2093: The second case is for $G=H\times H$ and $H<G$ the diagonal subgroup. In this
2094: case $G/H\simeq H$ and $L$ becomes left-right regular representation of $H\times H$
2095: on $H$. Here we shall write $L^1(H)^{\omega, \omega}$ for the analytic vectors.
2096:
2097:
2098:
2099: \subheadline{A.3. Averaging properties}
2100:
2101: Recall that the average map
2102: $$C_c(G)\to C_c(G/H), \ \ f\mapsto f^H; f^H(xH)=\int_H f(xh)\ dh$$
2103: is contiunuous and onto. Further, this map extends to a surjective contraction of Banach spaces
2104: $L^1(G)\to L^1(G/H)$. We will show that the averaging operator
2105: maps analytic vectors into analytic vectors.
2106:
2107: \par A standard application of the Bergman estimate gives:
2108:
2109: \Lemma A.3.1. Let $U\subeq G_\C$ be an open neighborhood
2110: of $\1$. Then for any pair of compact subsets
2111: $U_1, U_2\subeq U$ with $U_1\subeq \Int U_2$ there
2112: exists a constant $C>0$ such that for all $f\in L^1(G)_U$ we have that
2113: $$(\forall x\in U_1^{-1}G) \qquad \int_H |\tilde f(xh)|\ dh\leq
2114: C \sup_{x\in U_2}\|\tilde f_x\|, $$
2115: where $\tilde f$ denotes the extension of $f$ to a holomorphic
2116: function on $U^{-1}G$ (cf.\ Proposition A.2.1).\qed
2117:
2118: Combining Lemma A.3.1 with Proposition A.2.1 we obtain:
2119:
2120: \Proposition A.3.2. Let $U\subeq G_\C$ be an open neighborhood
2121: of $\1$. Then for every
2122: $f\in L^1(G)_U$ and $g\in G$ the integral
2123: $f^H(g)=\int_H f(gh)\ dh$ converges absolutely and $f^H\in L^1(G/H)_U$.
2124: In particular, there is a well defined mapping
2125: $$L^1(G)^\omega\to L^1(G/H)^\omega, \ \ f\mapsto f^H\ .$$\qed
2126:
2127:
2128:
2129: \subheadline{A.4. Mollifying properties}
2130:
2131: In this section $E={\cal H}$ will be a Hilbert space and $(\pi, {\cal H})$ a unitary
2132: representation of $G$.
2133: For $f\in L^1(G)$ one defines a continuous operator $\pi(f)\: {\cal H}\to {\cal H}$ by
2134:
2135: $$\pi(f)v=\int_G f(g) \pi(g)v\ dg\qquad (v\in {\cal H})\ .$$
2136: Notice that this defines a $*$-representation of the Banach algebra
2137: $L^1(G)$, i.e. we have $\pi(f*g)=\pi(f)\pi(g)$ and $\pi(f)^*=\pi(f^*)$ with
2138: $f^*(x)=\oline {f(x^{-1}) }$.
2139:
2140: \par Recall that $L^1(G)^{\omega, \omega}$ denotes the analytic
2141: vectors for the left-right regular representation of $G\times G$ on
2142: $L^1(G)$. It is easy to see
2143: that $L^1(G)^{\omega,\omega}$ is $*$-closed subalgebra of $L^1(G)$.
2144:
2145: \par Let $f\in L^1(G)^{\omega,\omega}$. It follows readily
2146: from Proposition A.2.1 and the definition of analytic vectors
2147: that $\pi^\omega(f)$ maps ${\cal H}$ continuously into ${\cal H}^\omega$.
2148: In particular the restriction $\pi^\omega(f)\:=\pi(f)\res_{{\cal H}^\omega}$
2149: gives rise to a continuos operator $\pi^\omega(f)\: {\cal H}^\omega\to {\cal H}^\omega$.
2150: Hence we have an algebra representation:
2151:
2152: $$\pi^\omega\: L^1(G)^{\omega,\omega}\to \End({\cal H}^\omega), \ \ f\mapsto \pi^\omega(f)\ .$$
2153: The corresponding dual representation is given by
2154:
2155: $$\pi^{-\omega}\: L^1(G)^{\omega,\omega}\to \End({\cal H}^{-\omega}); \ \
2156: \pi^{-\omega}(f)\lambda=\lambda\circ \pi^\omega(f^*)\ .$$
2157: Another application of Proposition A.2.1 then gives us
2158: the mollifying property:
2159:
2160: \Proposition A.4.1. Let $(\pi, {\cal H})$ be a unitary
2161: representation of a unimodular Lie group $G$. Then we have for all
2162: $f\in L^1(G)^{\omega,\omega}$ that
2163: $$\pi^{-\omega}(f){\cal H}^{-\omega}\subeq {\cal H}^\omega\ .$$\qed
2164:
2165: \nin {\bf Note:} For $f\in L^1(G)^{\omega, \omega}$, it is often convenient to write $\pi(f)$
2166: instead of $\pi^{-\omega}(f)$. We will use this convention throughout Section 2 in the main text.
2167:
2168:
2169:
2170: \def\entries{
2171:
2172:
2173: \[AG90 Akhiezer, D.\ N., and S.\ G.\ Gindikin, {\it On Stein
2174: extensions of
2175: real symmetric spaces},
2176: Math.\ Ann.\ {\bf 286}, 1--12, 1990
2177:
2178: \[vdB88 van den Ban, E., {\it The principal series for a reductive symmetric space I,
2179: $H$-fixed distribution vextors}, Ann. sci. \'Ec. Norm. Sup. {\bf 4}, {\bf 21}
2180: (1988), 359--412
2181:
2182: \[vdB00 ---, {\it The Plancherel theorem for a reductive symmetric space}, Lectures
2183: for the European School of Group Theory. August 14--26,
2184: 2000, SDU-Odense University.
2185: http://www.math.uu.nl/people/ban/publ.html
2186:
2187: \[vdBD88 van den Ban, E., and P. Delorme, {\it Quelques propri\'et\'es des repr\'esentations
2188: sph\'eriques pour les espaces sym\'etriques r\'eductifs}, J. Funct. Anal. {\bf 80} (1988),
2189: 284--307
2190:
2191: \[vdBS97a van den Ban, E., and H.\ Schlichtkrull, {\it The most
2192: continuous part of the Plan\-che\-rel decomposition for a reductive
2193: symmetric space}, Ann. of Math. {\bf (2) 145} (1997), no. {\bf 2},
2194: 267--364
2195:
2196: \[vdBS97b ---, {\it Fourier transform on a semisimple symmetric space}, Invent. Math. {\bf 130} (1997), no. {\bf 3},
2197: 517--574
2198:
2199: \[BD92 Brylinski, J.-L., and P. Delorme, {\it Vecteurs distributions $H$-invariants pour les s\'eries
2200: principales g\'en\'eralis\'ees d'espaces sym\'etriques r\'eductifs et
2201: prolongement m\'eromorphe d'int\'egralesd'Eisenstein}, Invent. Math. {\bf 109} (1992), no. {\bf 3}, 619--664
2202:
2203: \[D98 Delorme, P., {\it Formule de Plancherel pour les espaces
2204: sym\'etriques r\'eductifs}, Ann. of Math. {\bf (2) 147}
2205: (1998), no. {\bf 2}, 417--452
2206:
2207: \[FT99 Faraut, J., and E. G. F. Thomas, {\it Invariant Hilbert spaces of
2208: holomorphic functions}, J. Lie Theory {\bf 9} (1999), no. {\bf 2}, 383--402
2209:
2210:
2211: \[GK02a Gindikin, S., and B.\ Kr\"otz, {\it Complex crowns of Riemannian
2212: symmetric spaces and non-compactly causal symmetric spaces},
2213: Trans. Amer. Math. Soc. {\bf 354} (2002), no. {\bf 8},
2214: 3299--3327
2215:
2216: \[GK02b ---, {\it Invariant Stein domains in Stein symmetric spaces
2217: and a non-linear complex convexity theorem}, IMRN {\bf 18} (2002), 959--971
2218:
2219: \[GK\'O083 Gindikin, S., B.\ Kr\"otz and G.\ \'Olafsson, {\it Hardy spaces for
2220: non-compactly causal symmetric spaces and the most continuous spectrum}, Math. Ann. {\bf 327} (2003), 25--66
2221:
2222: \[H84 Helgason, S., ``Groups and Geometric Analysis'', Academic Press, 1984
2223:
2224: \[H\'O96 Hilgert, J.\ and
2225: G.\ \'Olafsson, ``Causal Symmetric Spaces, Geometry and
2226: Harmonic Analysis,'' Acad. Press, 1996
2227:
2228: \[K99 Kr\"otz, B., {\it The Plancherel theorem for biinvariant Hilbert
2229: spaces}, Publ. Res. Inst. Math. Sci. {\bf 35} (1999), no. {\bf 1},
2230: 91--122
2231:
2232: \[K\'O03 Kr\"otz, B., and G. \'Olafsson, {\it Analytic vectors for representations -- a
2233: survey}, in preparation
2234:
2235: \[KS01a Kr\"otz, B., and R.J. Stanton, {\it Holomorphic extension of
2236: representations: (I)
2237: automorphic functions}, Annals of Mathematics, to appear
2238:
2239: \[KS01b Kr\"otz, B., and R.J. Stanton, {\it Holomorphic extensions of
2240: representations: (II) geometry and harmonic analysis}, preprint
2241:
2242: \[M79 Matsuki, T., {\it The orbits of affine symmetric spaces under the action of minimal parabolic
2243: subgroups}, J. Math. Soc. Japan {\bf 31} (1979), 331--357
2244:
2245: \[\'O87 \'Olafsson, G., {\it Fourier and Poisson transformation associated to
2246: semisimple symmetric space}, Invent Math. {\bf 90} (1987), 605--629
2247:
2248:
2249:
2250: }
2251:
2252:
2253: {\sectionheadline{\bf References}
2254: \frenchspacing
2255: \entries\par}
2256: \dlastpage
2257: \bye
2258:
2259: