Function spaces of harmonic and analytic functions in infinitely
many variables
Citation for published version (APA):
Martens, F. J. L. (1987). Function spaces of harmonic and analytic functions in infinitely many variables. (EUT report. WSK, Dept. of Mathematics and Computing Science; Vol. 87-WSK-02). Technische Universiteit Eindhoven.
Document status and date: Published: 01/01/1987
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
providing details and we will investigate your claim.
University of Technology
Netherlands
Department of Mathematics and Computing Science
Function spaces of harmonic
and analytic functions
in infintely many variables
byEJ. L. Martens
AMS Subject Classifications: 81 B05, 46F05, 46F10
EUT Report 87-WSK-D2 ISSN 0167-9708
Coden: TEUEDE Maart 1987
Contents
O. Introduction
1. Preliminaries
2. Spaces of functions related to spaces of harmonic functions
3. A space of functions on a real separable infinite dimensional Hilbert space
4. A space of functions on the unit ball in IR£2
5. Special isomorphisms between
M,
the symmetric Fock space, theIN
Bargmann space and (IR ,d~)
6. Characterization of
M
and related spacesAcknowledgement References 1 3 15 33 56 61 79 96 97
Introduction
In this university report we study spaces of functions on subsets of Hilbert spaces. There are three main subjects in this paper.
The first subject is the theory of reproducing kernels. Most results on
re-~roducing kernels which we apply, have already been stated in Aronszajns paper [Ar]. We apply the theory of reproducing kernels to Hilbert spaces of harmonic functions.
The second subject is the harmonic or spherical harmonic function. As a guide and starting point we used the book Spherical Harmonics of Muller [MuJ.
The third subject is the representation of some special groups in the intro-duced spaces. The papers of KOno and Orihara [Ko] and [Or] served as a source of inspiration.
The purposes of this report are the introduction of spaces of functions of infinitely many variables, which are Hilbert spaces, analyticity spaces or trajectory spaces. These spaces are characterized both functional analytical-ly and anaanalytical-lyticalanalytical-ly.
Now we present a short summary of each section separately.
Section 1 deals with reproducing kernels, representations, analyticity on locally convex vector spaces and Gegenbauer polynomials.
In section 2 we introduce a Hilbert space ~ of harmonic functions on lRq and we study its reproducing kernel. We also consider the usual representa-tion of the orthogonal group on lRq in ~.
In section 3 we define a Hilbert space
M
of functions of an infinite number of variables, which contains all spaces A~, q E IN. The fact that the set of harmonic functions is dense in M is used to construct specific orthonormal bases and to prove some general properties of the elements ofM.
In section 4 we define a Hilbert space
N
of functions on B the unit ball in (U'2' We give results on this spaceN
similar to the results of'the spaceM
as defined in section 3.In section 5 we consider isomorphisms between M and some well known spaces, such as the Pock space, the Bargmann space and a space of square integrable functions with respect to a Gaussian measure. For the latt.er type of space we refer to [Ko] and [OrJ.
In the Bargmann space of infinite order the harmonic 90lynomiAls are dense (Theorem 3.8). This is not true in a Bargmann space of finite order.
In section 6 characterizations of the space
M
and related spaces are present-ed. Now analyticity plays a more important role than harmoniticity. We have two kinds of characterizations: Estimates with respect to reproducing kernels or a combination of a growth condition and a (real) analyticity condition.1. Preliminaries
In the first part of this section we present some general results on repro-ducing kernels and unitary representations of topological groups.
In the second part of this section we mention some recurrence relations of Gegenbauer polynomials.
In the third part we introduce the notion of analyticity on a locally convex topological vector space.
Finally we discuss an integral identity for c1-functions on a neighbourhood of the unit sphere Sq-l in IRq.
Reproducing kernels
In this section F denotes a class of functions from a set D into ¢ which establishes a complex Hilbert space with inner product (.,.) F and norm
n.
If",
Definition 1.1. The function K: D x D ~ t is called a reproducing kernel of F iff
(a) V D: K{"Y) E F
y£
f (y) •
We list some properties of reproducing kernels. These results can also be found in [ArJ or [YosJ.
(1.1) For
W
cF
let Span(W) denote the linear span ofW,
i.e.Span(W)
n
{ I
j=l The subspace Span({K(',y)
a.w.
J J
Y ED}) is dense in
F.
(1.2) Let the space F have a reproducing kernel K. Then for all x,y E D II K(' ,y)
I~ =
KCy,y)K(x,y) K
(1.4) The space F has a reproducing kernel iff for each y E D the evaluation functional f ~ fey), f E
F,
is continuous.(l.S) Let
F
have a reproducing kernel K. Then we have(1.6) Let Fl be a closed subspace of F, let F have a reproducing kernel K on D x D and let P
1 be the orthogonal projection from
F
ontoFl.
Then Fl has a reproducing kernel Kl with Kl (.,y) = Pi (K(.,y» and
Y E D, f E
F •
(1.7) Let Fl and F2 be two mutually orthogonal closed subspaces of the function space F with reproducing kernels Kl and K
2, respectively. Then F admits a reproducing kernel K with K = Kl + K
2.
(1.8) Let
F
have a reproducing kernel K and let . (~) n n be an orthonormal basis inF.
The following equalities are valid:I
I(j)n (x)12
= K(x,x)n
I
~n(Y)~n(x) = K(x,y)n
x,y ED.
(1.9) Let (~) be an orthonormal basis in
F.
n n
The space
F
admits a reproducing kernel iff for all x,y € D the seriesL
~ (y)~ (x) is pointwise convergent. n nn
(1.10) Let DO denote a set and let K denote a function from DO x DO into ~ with the property that for each n ~ 0, a
j ( €, Yj ( D, 1 $ j n
L
ajatK(Y~'Yj) ~
0 . j,tThen there exists a vector space FO of functions on DO' which is a Hilbert space and which admits K as a reproducing kernel. This space FO is uniquely determined.
(1.11) The elements of a Hilbert space with reproducing kernel can be characterized with the aid of the reproducing kernel:
(1. 12)
Theorem 1.2. Let
F
denote a Hilbert space of functions from a set 0 into ¢ with reproducing kernel K •. Let f denote a function from D into C. Then f € F iff there exists c > 0 such that for each t E lN and all ct. c Il,Y
j 0 with J tI
L
ctl(Yj )I
j=l Proof. *: Let t c lN, ct- - -
j straightforward estimate tI
L
ct/(Yj )I
j=1 1 S js
t R.-
~ s c(I
~ctjK(Yk'Yj) } k, j=l E (;, Yj € 0 where 1 :< j t ==I (
I
u.K(.,y.) ,f)FI
s j=l J ) t S IIfIF
II
I
K{.,yj)IIFs
t. Then we have the-: Let
W
denote the Span({K(',y) l y e D}). The linear functionalm:
W
~ C is defined by t t m(L
ct j K ( • , y j) ) =L
ct. j=l j=l J R. c lN, ct. E 1:, Yj f. D where 1 s j s t. If J tL
j=1 a.K(',y.)= a
I J ) then R.I
j=l a.f(y.) ==a
J )because of (1.12), so the functional m is properly defined.
By assumption (1.12) and the properties of the re~roducing kernel of F it follows that for all w E
W
1m
(w)I
~ ell w II •The space
W
is dense inF,
so by the Riesz' representation theorem there exists 9 EF
such that for all w EW
mew) == (w,g)F •
In particular for each y E D we have fey) = m(K(·,y» (K(· ,y) ,g)
This implies f == g, whence f E
F.
rJ
Representations
Let
G
be a group with identity element e.The Banach algebra of bounded operators on a Hilbert space X will be denoted by
L(X)
and the identity on X by I or IX'Definition 1.3. Let
G be a topological group, let X be a complex Hilbert
space and let IT denote a mapping fromG into
L(X).
The mapping IT is called a representation of the group
G in X iff
(a) (b) (c)V
hG:
TI(gh)=
TI(g)IT(h) g, E II(e) == IV X= 9 + IT(g)x, 9 EGis continuous at e •
XE
A representation IT is called unitary iff
(d)
V
G: [IT(g)]*
=
IT(g -1 ) . gELet
F
denote a Hilbert function space with reproducing kernel K: D x D ~~ <Cand let G denote a group of transformations on D. Then under certain condi-tions on K there exists a canonical unitary representation of
G
inF.
(1. 13) (1. 14)
(1. 15)
( 1. 16)
Theorem 1.4~ Let G denote a topological group of transformations on O. Let F have a reproducing kernel K with the properties:
(a) V
R € G V x,y€ 0: K(Rx,Ry)
=
K(x,y)(b) Vy€O: The F-valued mapping R + K(-,Ry), REG is continuous at I D. Then the mapping IT: G+
L(F)
defined byX E 0, f E F, REG is a unitary representation of G in F.
Proof. Put W = Span({K(',y) lYE oJ).
For each REG the operator 'IT (R): W + W is defined by
t
IT(R) (
I
(),K(·,y.)) =J ] Cl.K(-,Ry.), J ] R, ?: 0, Cl. E
e,
y.€ 0 •
J J
It is easy to see that
-1 [1T(R)f](x) = feR xl, 1T(RS)f = 1T(R)1T(S)f, fEW X E 0, f €
W,
REG R,S E G, fEW -1 ( 1T (R)f , g)F
=
(f, IT (R ) g) F' f,g EW,
R E:G .
We only prove the last statement. Let f,g E
W
withk
f =
I
Cl.K(-,a.) andj=1 J J
Then for each R E: G
(1T(R)f,g)F "" g -1 (f,lT(R )g)F' m
=
I
B,Q,K(' ,b,Q,).
1=1The statements (1.14)-(1.16) imply for all REG, fEW: Ihr (R) f If
=
II f If .The set
W
is dense inF.
Let Ti(R) denote the unique unitary extension of1T (R) to
F.
Let REG, let f E F and let XED. Then we have [1f (R)
fJ
(x) (n(R)f,K(o,x»F -1=
(f,1T(R )K(-,x»F -1 (f,K(o,R x»F f(R-1X) • In the sequel we denote IT(R) by IT(R).For all R,S E
G
it follows from (1.14) to (1.16)II (RS) II(R)IT(S)
Let f E
F.
Finally we want to prove that R + IT(R)f is continuous at 1 0,
E
Let £ > O. We take 9 E W such that II f - 9 II <
'4 "
Letm
9
I
j=l
CX,K( ,y,)
J J with cxj E ~, CX, ~ J 0, v, - J ED.
Let Uj be an open neighbourhood of 10 such that for all R E U
j
m
Put U = n U
j < Then U is an open neighbourhood of 1D"
j=l For R E U we have m
~
t
+I
I
CX]'I
II K ( • , RYJ') - K ( • , y J,)n
F+:i
< £ " 0 j=lLet V be a subspace of the Hilbert space X and let A E L(X) be such that A (V) c V.
This subspace V is called A-invariant.
The restriction of
A
toV
will be denoted byAlv'
It is clear thatAlv
EL(V).
*
If also A (V) c V then V is s.aid to reduce A. If U is a unitary operator then each V-invariant subspace is also reducing.
Definition 1.5. Let
n
denote a unitary representation of a topological groupG in the Hilbert space
X.
The representation
n
is called irreducible iff each closed subspaceV
ofX,
which is TI(g)-invariant for all g EG,
equals {a} orX.
Lemma 1.6. Schur's lemma.
Let IT be a irreducible unitary representation of a topological group
G
in the Hilbert spaceX.
LetA
EL(X)
be such that for all g EG
the operators A andneg)
commute.Then there exists
A
E ~ such thatA = AI .
• See [so], I, Satz 2.3.
-Theorem 1.7. Gurevi~.
Let IT be a unitary representation of a compact topological group
G in
the separable Hilbert spaceX.
Then there exist finite dimensional invariant subsoaces ' - n
X ,
n E IN, ofX such that
(a) (b)X
1X
n m X = (9 X n n n:F
m(c) g + IT (g)
Ix '
g EG,
is an irreducible unitary representation ofG
n
in X for all n E IN . n
(1.17)
( 1.18)
(1.19)
Gegenbauer polynomials
For q 2 2 and n ;:::
°
we define the function pq: :IF -+ lR byn
Lemma 1.8. Let q ~ 2.
Then the following statements are valid. (a) pq is a polynomial of degree n .
n
(b)
°
I
(c) 1 , n E IN •
Proof. See
[MuJ,
p. 16-18.n
:f
mIn [MOS], § 5.3 the Gegenbauer polynomials cA of degree n, n
~
0, and ordern
o
A, A ;:::
-~, are introduced as special cases of Jacobi polynomials. These poly-nomials satisfy 1f
cA (t) cA (t) (1 _ t2)A-~dt =° ,
n m n ;. m -1""
I
CA(t)rn = 1,
n=O n (1 - 2tr + r2)A A ;. 0 I t E [-l,lJ,Irl
< 1""
I
CO(t)rn -log (1 - 2tr + r ) 2 n=O n Lemma 1. 9. F ix q ;:: 3. Then pq = r(n + l)r(q - 2) c~q-l n r{n + q - 2) n n ~ 0 p2 =:; ~ cO n 2 n ' n 2: 1 1(1. 20)
(1.21)
Proof. 14' rom Lemma 1. 8 (b) and re lation ( 1. 17) we obtain a q E lR such that
n
~rom Lemma 1.S(c) and relation (1.18) it follows (aq}-l
=
c~q-l(l)=
r(n + q - 2)n n r(n+ 1) ( q - 2 ) '
The proof for p2 runs similarly.
n
We present two recurrence relations for the pq,s. By pq we mean (Pq) '.
n n n
Lemma 1.10. Let q ~ 2 and n ~ O. Define P~1 (t) = O. Then we have
(2n + q - 2)t pq(t)
n (n + q -
2)P~+1
(t) + nP~-l
(t)and
(2n + q -
2)P~(t)
=(n(~; ~)2) P~+l
(t) - (n + : _ 3)P~-l
(t)Proof. From [MOS], § 5.3, we have
2(n +
A)tC~(t)
= (n +1)C~+1
(t) + (n + 2A -1)C~_1
(t) andWith the aid of Lemma 1.8 we get the desired results.
Lemma 1.11. Let (t) be a Cauchy sequence in IR with limit t. q q
Then for all n ~ 0
Proof. Since pq - 1 we have lim P6(t ) = 1.
0 q-l-<X> q Assume for 0 ::; j ::; n lim P~(t ) t j q-)<X) ] q Because of Lemma 1.10
o
t E IRo
pq 1 (t ) n+ q (2n + q - 2) t pCJ (t ) (n + q - 2) q n q (n + n q - 2) pq l{t ) n- q whence lim pq l{t ) n+ q q~ n+l = t
1<' or q,m € IN and n ::>: m define the function Aq+1: [-1,1] -l>
m.
byn,m (1.22) Aq+1 (t) n,m Lemma 1.12. Let q,m,n 1,n2 t IN with nl,n2 ~ m. Then (a) + 1 (b) Aq+1 (t)
rem
+ ( _ _ _ _ _ ) 1:1(1 _ t2)m/2 • m,mr
(m +!)!;
Proof.(a) See [MU], Lemmas 14 and 15 and pages 3 and 4.
(b) Because of Lemma 1.10 the coefficient of tm in pq+1 is positive.
m
By (1.22) there exists a > 0 such that
Aq+1(t)
m,m
Since -1 we have Analyticity q + 1rem
+ - 2 - ) 1:1 ( )r
(m +!)
!;/r
(m +The notion of analyticity will be generalized.
fEfini tion 1. 13. Let W denote an open subset of
«:q.
The function f: W -+ C is called analytic on W iff for all w € W there
exists an open neighbourhood N of w such that for all Z E N
w w
f (z)
where the series is absolutely convergent.
k
(z - w ) q q q
fEfinition 1. 14. Let V denote a complex (real) locally convex topological vector space and let W denote an open subset of V' •
The function f: W -+ it is called analytic on W iff f is continuous on W and for all w ( Wand v (' '/ there exist an open neighbourhood . M w,v of
o
in it such thatA
-+ f(w +AV)
is analytic on M (is extendible to anw,v analytic function on M ) .
w,v
If f is analytic on an open subset W in
c
q because of definition 1.13, then certainly f is analytic because of definition 1.14.Theorem 1.15. Hartogs theorem.
If u is a complex valued function defined on an open set ~ c Cq and u is analytic in each variable z. when the other variables are given
ar-J
bitrary values, then u is analytic in Q.
Proof. See [HoJ, theorem 2.2.8.
An integral identity
[J
Let sq-l and sq denote the unit sphere and the
o~en
unit ball in mq, respec-tively.q-l
The usual surface measure on S will b~denoted by dO q_1 .
1~~~~j~j~ Fo~ a C'-function on a neighbourhood of sq the normal d~rivative in a point I; E sq"1 wi 11 be denoted by
d f (0 v
Theorem 1.16. Let f and 9 be continuously differentiable functions on a neighbourhood of the closed unit ball Bq• Let j c IN with 1 ~ j ~ q. Then we have the following identity
a
[ax.
fJ(I;)g J f ( ~)ax.
9J (
~) dO' 1 (~) + q-J +J
l;.[a J v f(l;)g + f ( t;) a v 9 ( t;)J
do q-1 (l;) + Proof. Put sq-1 + (q - 1)J
t;.f(t.:)g{t;)dO J q-1(1;;) • q-l S[~.
fJ( 9 J () f (I;) [-" - 9J (
l;) do 1 ( l;) • ox. q-]Straightforward calculation yields
M =
I
[dax.
(f. -g) ] ( 1;) d a q-l ( t;) J q-l SI
k=l q-l S Bf
-3 -()
[ 3 -a -(f.g) ] (x) xkdx -~ x. q J + qI
[ax. (f.g) J(x)dxa
-
=
J (Gauss theorem)J
q-l S Bf
~-aar ()
-(
f . g)-
(x) x. Jdx + x. ~ k q Jf
a
-(q-l) [ax.(f.g)](X)dX Js·[a
f(l;)g{t;) + f(l;)a g(t;)JdO I(!;) +J v v
q-+ (q - 1)
J
l;jf(t;)9(I;)doq_I (1;) • sq-1
2. Spaces of functions related to spaces of harmonic functions
In this section we introduce spaces of fUnctions which are restrictions of harmonic functions to sq-l or are harmonic functions on IRq. Most results are extensions of results in [Mu] and [EG].
We use the following notations where we take q £ lN, q > 1, fixed. q-1 S (a,r) dx dO' q-1 2rrq / 2 0' q-l
r
(s.) 2 f.:,r
a
2 q 2 j=l ox. Ja
q,v=
r
x. l. j=1 Harm(W) (...
) II • IIa
ax. l., the sphere with centre a and radius r in IRq , the unit sphere in IRq
, the open ball with centre a and radius r in IRq
the open unit ball in IRq
the usual Lebesgue measure in IRq
, the usual (q-1)-dimensional surface measure
q-l on S
Note that dx
q-l , the total surface measure of S
, the q-dimensional Laplacian
, the q-dimensional normal derivative , the space of complex valued functions f
which are harmonic on a neighbourhood of each point of W c IRq
Note that f € Harm (sq) implies
f E Harm(Sq(O,l + Ell for some E > O.
, the compact group of orthogonal transforma-tions from IR q onto IR q
, subgroup of O(IRq) consisting of all rota-tions R(det(R}
=
1), the inner product on IRq , the norm on IRq •
(2. 1)
q-1 q-l
Consider the Hilbert space L
2(S )
=
L2(S ,dOq_t) of square integrableq-l
functions on S with the natural inner product
f ( ~) g ( ~) do 1 ( t;;) ,
q-We denote the corresponding norm by
II. II
1 •q sq- q-1
For each R E O(IR ) we define the operator L
R: L2(S ) Then we have: - LRS LRLS - LIf
=
f , q R, S € 0 (IR ). q-l f € L 2(S ).The mapping R + LRf is continuous - The operator (L
R)* is equal to
L
T the surface measure do 1
q-R
q-l for each f E L
2(S ).
because of the rotational invariance of
q q-1
So R + LR is a unitary representation of O(IR ) in L
2(S ).
Since O(IRq) is a compact group, theorem 1.7 yields the existence of mutual-ly orthogonal finite dimensional subspaces Vq n
~
0, of L (sq-1) such thatn' 2
- L
2(Sq-1) = $
V
q•
n=O n
The subspaces V~, n ~ 0 remain invariant under the operators L
R, R € O(IR q
), and the representation R + LRI is irreducible.
V
q nIn [Vi], Chapter IX, § 3 the following has been shown:
The subspaces
V
q can be selected in such a way that for each f EV
q theren n
exists a unique harmonic homogeneous polynomial p of degree n in q variables (i.e. p(AX)
=
Anp(X) and [6~](X) = 0, A € IR, x E IRq) such thatf(~) = p(~) ,
This leads to the following definition:
Definition 2.1. Fix q € IN and n ~ O.
Let ~ denote the space of harmonic homogeneous polynomials of degree n
n
in q variables.
Put dq dim(~) dim(Vq).
Lemma 2.2. Let q ~ 2.
The numbers dq can be explicitly calculated. We have
n
dq = (2n + q - 2)f(n + q - 2)/(r(n + l)r(q - 1»
n
dq
o
1.
Proof. See [Mu], pp. 3 and 4.
n ~ 1
Since
V
q is finite dimensional, it follows from remark (1.9) thatV
q has an n
reproducing kernel. The following lemma yields an explicit formula for this reproducing kernel.
Lemma 2.3. Let {eq .
I
1 ~ j ~ dq } be any orthonormal basis of ~.n,] n n Then we have dq n
-I
j=lHere the polynomial pq is defined by (1.17).
n
Proof. See [Mu], theorem 2.
Corollary 2.4. Let f E
V
q•n
Then
If(e;) I
Proof. See property (1.5).
q-l
~,n E S .
Let Dq denote the orthogonal prOjection from L (sq-t) onto
li
q•n 2 n
Corollary 2.5. Let f E L
2(Sq-l).
For each
n
E sq-l we have the equalityo
o
Proof. See property (1.6).
o
In the following lemma we present the generating function of the polynomials pq.n
Lemma 2.6. Let t,t € IR with It I < 1 and It I ~ 1. Then we have
L
n=O
For fixed t the series is uniformly convergent in t ,
ITI
~ 1. Proof. From Lemmas 1.9 and 2.2 it followsq q (2n + q - 2) c~q-l
d·P =
-n n (q - 2) n
Relation (1.17) gives the wanted result. IJ
Corollary 2.7. Let t f IR with It
I
< 1. Then we haveL
n=O
1 +t
(1 - t) q-1
Proof. Take t 1 in Lemma 2.6.
o
Now we are in a position to link to each f € L (sq-l) a harmonic function
~f
2
on a ball Bq(O,r) where r ~ 1 depends on f. q-l
Definition 2.8. Let f E L
2(S ). Then
Let
r = sup{p ~ 1
(2.2)
We remark that the series in (2.2) is absolutely and uniformly convergent on each ball Bq(O,p) with p < r. This is a consequence of the definition of r and the corollaries 2.4 and 2.7.
q-1
Theorem 2.9. Let f E L
2(S ) and let r be as in definition 2.8. (a) The function Hqf is harmonic on Bq(O,r) .
(b) If there exist 5 > 1 and a harmonic function g on Bq(O,s) such that
g(~)
f(~)
for~
E sq-l, then(i) r ~ 5
(ii) [Hqf] (x)
=
9 (x) , X E B (0,5). q(c) For y € Bq the following integral representation holds:
[Hqf](y) =
~
J
q-l q-1 S 1 - II Y Ir f (~)
do (~
) •"~
_ yf
q-lProof. (a) and (b): See [Gr], Lemma 2.1. (c): Take y E Bq, Y ~ O.
Then we have
""
dqJ
::
I
~"Ylr
f(~)P~(II~
II
.~)dOq_l (~)
n=O q-1 S q-l (Corollary 2.5) 1
J
1 - Ill' Ir f (~) do 1 (~)=
0 q-l (1 +lIy
It - 2(y.E;:»Q/2 Q-q-l S (Lemma 2.6) -=---.::...~ f(~)da 1 (~) .II
~-
yIF
q-=Moreover
Let n ~ 1.
[Iflf]
(x) [Hqf] (0) So we see that forLet f €
V
q• n = II x IPf (II: Ir=
0 . all 11 € S q-l f(~)dcr 1 (~) . q-ThenWe remark that the mapping
Ifl:
L2(sq-l) + Harm(B q
) is injective.
- -1
Next we consider the spaces Harm (Bq), Ifl(L
2(Sq » and Harm(B
q
). The first and the second are contained in the latter. So we get
o
This triple of spaces fits in a functional analytic setting which we describe now.
In [EG] the analyticity space
S
B and the trajectory spaceT
Bareintro-Y, Y,
duced for Y a Hilbert space and B a self-adjoint operator. Here
I
tB{f € Y 3
t>0: f E D(e )}
and
They constitute the triple
As we shall see
for a suitable self-adjoint operator
A.
Definition 2.10. The self-adjoint operator A in L (sq-l) is defined by
2
Af
I
f E D(A)n=O
The operator A has the following properties:
2 ~
(2.3) - The operator A is equal to -~(q - 1)1 + {~(q - 1) I + ~LB} where ~B de-q-l
notes the Laplace-Beltrami operator of the sphere S • (2.4) - For each f E D(A) with ~f E Harm (Bq)
Af (i;) [d ~fJ(i;)
q,v
because
a
h = nh for all h E Hq.q,v n
(2.5) - Fix t > O.
For all f € D(e(log t)A)
Remark:
D(e(log t)A) =
From (2.5) it follows
The Space
T
1 can be represented by Harm(F3q). To this en.d we extend q-L 2(S ),A the operatorH
q: L 2(Sq-l)~
Harm(B q) tension will be denoted bygq.
to an operator on
T
Theex-L
2(sq-l),A"
For each F €
T
we defineq-1
< L2 (5 ) ,A
x [F (-log II x ID] ( 9 )
[~FJ<O) = [F(1)](0) , 1 > 0 • We observe that F (-log II x ID E S q-1 •
The function gqF is harmonic on Bq•
Further gq is a linear bijection from
T
1 onto Harm (Bq) • Also thisq-is a consequence of (2.5). L2 (S ),A
If moreover F E L
2(sq-l) (i.e. F(t) e-tAf for some f E L2(Sq-l» then
'if1F
Ff!f.
Next we introduce some operators in L (sq-l) which will be needed later on.
2
Definition 2.11. We define the operators
P
k and
~,
1~
k~
q, in L2(Sq-l) byLemma 2.12. Let f E ~ and let k E 1N with 1 ~ k $ q. m
The following relations hold true:
~~f
=
(2m+
q - 2)Dq (0 f)J\. m-l -X
Proof. Fix ~ E sq-l.
Applying corollary 2.5 and lemma 1.10 we get
[~fJ (t;) o q-l
f
q-l S q ~,.P (n.t;)f(n)do len) J\. m q-l:" [(m +q - 2) • q "k - (m + 1) P m+l (n. t;) (m +q m · q -3) Pm-1 (n.1;;) J(2m +q fen) -2) dOq _1 (n) ~(m +q -2) pq ( •• ~) _ m dY k (m + 1) m+l (m +q p q ( •• t;) ] (11) f ( n) do 1 (n) . 3) m-l 2m +q -2To this end we remark that
[ a
(JIlf) ] (nlq,v
=
m (EfIf) (n) = mf (n)and also that
[-f-(Hqf) ](n)
=
[Pkf] (n).
Yk So we get cr q-lf
nkf(n)a
[m+q-2 pq (e.l;) _ _ m_ pq (-.l;)](n) dcr len) + q,v m + 1 m+l m+q-3 m-l (2m+q-2) q-q-l S dq +_m_J
[(m+ q -2) Pmq+1(n.l;) - m pq (n.l;)]' cr 1 (m+l) (m+q-3) m-l q- q-l S Now we have dq mJ
[(m+q -2) oq Ll=G
m :; 1 Pm+1 (n.l;) q q-l n kf (n) m eq (m+q-3) Pm-l (n.l;)](n.~) (2m+q-2) dcrq _1 (n) • SApplying Lemma 1.10 twice yields
dq
=~
J
q-l q-l S whence :iI: ... em + q -n,
(2m +q -2) (m +1) dq mf
pq (T'!. t;;) [ ... ('''f(n) + (2m.+q) 11kf (nl Jdcr q_1 (11) + cr m+1 k . q-l sq-l(2.6) dq m m (2m+q-2)(m+q-3) (1q_l
J
P~-l (n.~)
[-(Pkf) (n) +2nk f (n) ]dO"q_l (n). sq-l S' P f E:V
q and ~ + q - 2 dq l.nce k m-l (2Q, + q - 2) (9, + 1 ) R,=
1 d q R,~
0 we get (2Q. +q) H1 ' + ___ 1 _ _ _ Pkf) --(-2m--.-.:;;=-q----:-D~_l (~f)
Application of the projection D~_1 on both sides of the latter equality yields
If we substitute this in (2.6) we arrive at the other wanted relation.
0
In the final part of this section we discuss another Hilbert space which con-tains all ~ as mutually orthogonal subspaces. Let
n
with natural inner product
( 21T)
f
exp (-II x Ir /2) f (x) 9 dx, IRq
and with corresponding norm " • "
M2 (IRq)
2 13 Let ~ denote the algebraic direct sum of the Hq,s, i.e •
.;;;....;c.~~..;;..;;;;...;...;..;;.-"'-:....;;;..;~- R, n
each f E ~ is of the form f
=
I
n=Of with R, ~ 0, f E: Hq, 1 ~ n ~ £.
n n n
Observe that Hq is the space of all harmonic polynomials in q variables. Let
J..fI
denote the closure of Hq in M2 (IRq) •Our first aim is a characterization of the elements of ~. Lemma 2.14. Let 9 E
V
q and h EV
q•n m
Then we have the following equality
2 (n+m) /2.
r
(n +~ +q)2.1Tq/2
9 (~) h (~) dO" 1 (f,;)
q-Pro.cE., We. have, 1 .
J
exp(-lIxlfI2) (Hqg) (x)UfIh)
(x)dx (211') q/2 IRq OC> = I'J
( r 2 ) n+m+q-l q/2 exp -:2
r (2'11') .o
I
g(~)h(~)dcrq_l (~)dr
sq-l 2 (n+m)/2 q/2 2.'11' <X>I
exp -s s()
(n+m+q-2) /2f
o
sq-1 2(n+m)/2.r
(n +~ +q) = ---~--~~-2.n-g/2J
g(~)h(~)dcr 1(~) q-q-l S Corollary 2.15. Let n ~ m. <X> Then ~ ~ Hq in ~. Moreover ~fI = ~ Hq•n m n=O n
Proof.
V
q 1V
q in L2(SQ-l).
- - n m
The second statement is trivial.
Corollary 2.16. Let n ~ O. Let {eq .
I
1 ~ j ~ dq} be an orthonormal basisn,J n in
V
q. Then n is an orthonormal basis inH
q• n Co.rollary 2. 17. Let n ~ 1.The subspace
H
q has a reproducing kernel Kq withn n
o
x .".. 0 .". yx ""
0 or·y = 0o
o
(2.7)
Proof. Let {eq .
I
1 ~ j ~ dq} be an orthonormal basis of ~. Fix x,y E IRqn,] n n
with x
f:
0f:
y.Because of corollary 2.16 and property 1.8
In case y
=
0 or x q/2 dq 2.nIIx If
I!If _n_
pq(~ 2nr(n+!l) y O'q_l nUx!! 2 (Lemma 2.3)o
we have Kq(x,y) no.
Of course the kernel Kci of
Hci
satisfiesy
• lIy I~ ;:
Because of property (1.5) we know that for each f E
Hq
n
X E
Lemma 2 1 . Let n ~ 0 and let x E IRq. Then
1 we have
(2.8) Because q ~ 2 we a. + 9: -2a. + q So we get Kq(x,x) n Further we have q K1 (x,x) and q KO (x,x) =
rei)
(2n +q -2)r(n
+q -2) 2nr(n +!)
r(n + t)r(kI -
1) (2n + q - 2) (n - 2 + q - 1) (2n - 2 + q) (2 (n - 2) + q)have for each (l ~ 0
1 :5
s....::..1:.
. q :5(~)n-1
IIxIf
n q n! :5 II xIt
(because dq 1 q) 1 :5 ( q.9....:....l
) -1Corollary 2.19. Let f € ~ and
x
€ IRq. Thenn n-1
I
f (x)I
:<;; ( q -q 1) -2- II x Ifll f IIrnr
M 2(IR q )Proof. From inequality (2.7).
(q - 1) II x Ifn q r(n + 1) •
Let
n n
Theorem 2.20. Let f € ~.
Then f is a harmonic function on IRq which satisfies
I
f (x)I
Proof. Pix x E IRq. For all g €
ftt·
we haven-1
o
o
I
I
(Corollary 2. 19) ()O
s (
I
n=O / qexp(~
q - 1 qSo the linear functional t : ~ ~ € defined by x t (g) = x is continuous. 00
I
[Q~g]
(x) I n=OFor all 9 E Hq we have t (g) = g(x). x
m
For m € IN put f
=
I
[Qqf].m n=O n
Then f + f, m + "" in norm and f + ~ t (f), m + "", uniformly on compacta in
m m x x
IRq. Since the f 's are harmonic, the function f is harmonic. It is clear
m
that f satisfies the inequality (2.8).
0
Next we prove the converse of theorem 2.20. We need an auxiliary result.
Lemma 2.21. Let r > 1. Let f E Harm(Bq(O,r» and 9 E Harm(Bq). For each P,
1
s
p < rProof. We have
=
f(~)g(i;)da 1 (~) =
q-f(pE;;> [e-(log p)Ag](t:;)da 1(~) =
q-[e-(log
p)Af(p.)J(t:;)g(~)da
1(t:;)Proof. From theorem 2.20 it follows that
if!
c M2 (mq) n Harm(mq). Letq q
.a
q-lf E M2 (m ) n Harm(m ) and assume that f .l M-. For all 11 E S a n d n ~ 0
We have 1 =
---=-...,.-(21T)q/2
J
exp(-lIxIf/2)f(X)K~(X,n)dX
=m
qo
From Lemma 2.21 it follows that
So for all 11 € sq-l and n ~ 0
a
[ "" 2 r n+q-l exp(-2)
r n r a q-lf
q-l S So fl q-1 .l L2(Sq-l). Hence for each n E sq-l fen) = O. Since f is harmonic,
S
we conclude f
=
O.$.E;tna~ks.
- From Theorem 2.20 it follows that for all x E IRq the linear !unctionals f ~ f(x)t f E ~ are continuous. So the space ~ has a reprOducirtgkernel
00 e n=O
H
q it follows from n~.~n<.~+·" that n00
L
n=O
We have for f E
rfI
n(lemma 2.14) •
So i t follows that Bq is a compact injective operator from
MJ.
into L2(Sq-l). Next we define a unitary representation of O(IRq) in
MJ..
For all f E ~~ and R E O(IRq) the function f(RT.) (~f1.
The following definition makes sense:
Definition 2.23. The representation rrq: O(IRq) + L(~~) is defined by
It is easy to see that
n
q is a unitary representation of O(IRq) in ~. Each finite dimensional subspacerfI
is invariant under the operators rrq(R) ,n R E O(IRq).
The representation R + rrq(R)
IrfI'
R E O(IRq),iS irreducible, because rrq(R) :=::n
=
HqL
Bq and because the representation R +L
I '
R E O(IRq) is irreducible.R 00 R~
r1oreover
MJ.
=
$rfI.
See theorem 1.6 (Gurevic). nn=O n
Finally we give another description of the inner product (-,.)
restrict-ed to Hq. M2 (IRq)
Definition 2.24. Let p be a polynomial in q variables.
a
Let p(3) denote the differential operator p(--- , .•• ,
aX
1
Lemma 2.25. Let f,g E
rfI.
Then (f,g) :=:: [f(3)9J (0) M 2(IR q ) d
ax) .
qprodf. If f € ~ and 9 € ~ with n# m, then (f,g) q = 0 and
n
m
[f(a)g](O)
=
O. M2 (m )So we restrict ourselves to f,g €
Because of Lemma 2.14 we have
/fl.
n = ---=-0' q-1 f(t;)g(E;)dO' 1 (t;)q-The homogeneous polynomial f is a linear combination of monomials of the form n II xk . with 1 S k.~ q. j=1 J q-l J Lete - 1 on S • Put 2nr (n + 9.) 1
f
n a=
II t;k. g dO' 1 (t;)r
(9.) a q-l q-2 q-1 J S Then we have 2nr
(n + ~) 1 n a=
[ IT ~ e,g]q f(9.) tJ j=1 2 q-l jBecause ~ is self-adjoint and because of Lemma 2.12 n
.
2nr(n + 9.) n-1 a = _ _ _ --=;:..2_ a [IT~
e,D~+l (~
g) + Dq 1(~
g) ] r(9.) q-l j=1 j n n- n q 2 n-lRemembering that II ~.e ~ V~+1 and applying Lemma 2.12 again we ~et j=1 J n-l
f-
[IT~.e,
q-l 1 J 2n-1r(n - 1 + 9.) n-l _ _ _ _ _ _ 2 _ _ 1_ [ IT °q_l j=1Since
P
k 9 € H~_l we can repeat this argument n-l times and finally we get n1 n (l
=
-=---r
e , a q-l ITP
k g] j=l j q 1 = -a q-lf
q-l S n ( IT~
g)(~)da
1(~)
aX.. q-j=l k J n [IT-f-
g]
(0) j=l Xk jSince f is a linear combination of monomials we get (f,g) q
=
[f(Cl)g](O) .M2 (IR )
3. A space of functions on a real separable infinite dimensional Hilbert space In the first part of this section we construct a complex Hilbert space of functions of infinitely many real variables with the aid of the Hilbert spaces of harmonic functions, introduced in the previous section.
In the second part of this section a unitary representation of the full or-thogonal group on IRt2 in this Hilbert space will be introduced.
We use the following conventions and notations:
(3.1) IRt2 denotes the real Hilbert space of square summable sequences with inner product and norm denoted by ( ••• ) and II • II respectively.
For a,b,c,d E IRt2 we define
(a + ib. c + id) = (a.c) + (b.d) + i [ (b.c) - (a.d)] .
(3.2) (ej)j denotes the standard orthonormal basis of IRt2; i.e. ej
(3.3) IRtc denotes the subspace of IRt2 consisting of all finite sequences x i.e, only a finite number of entries of x are non zero.
(3.4) The space IRq is embedded in IRt2 identifying (x1' ... ,x
q) and j=1
~
x,e. J J E IRt2' Thus we get for exampleIRq = IRt and so on . c
(3.5) O(IRt
2) denotes the group of all orthogonal operators (i.e. bijective isome-tries) from IRt2 onto IRt
2. (3.6)
o (
IR t ) == {V E 0 ( IR t2)
I
V ( IR t )=
IR t } •c c c
(3.7) Eq denotes the orthogonal prOjection from IRt2 onto IRq. (3.8) The group O(IRq) is embedded in O(IRt
2) identifying R E O(IR q
) and the ortho-gonal operator RE + (I - E ) E O(IRt
2).
00 q q
Remark: u O(IRq) is a proper subset of O(IRt ). c q=l
Example: Define U E O{IRt 2) by
00
Then U
t
u q=ln E IN •
O(IRq) and U E O(IRt ).
(3.9)
(3.10)
(3. 11)
(3.12)
We start with the definition of an inner product space in which all spaces ~ of the previous section can be embedded.
Definition 3.1. The linear space F(IRt2) consists of all functions f: IRt2 ~ ~ with the following properties:
(a) f is continuous on IR!2' (b)
(c)
3 : f = foE
p>O p
I
e -II x If121
f (x) 12dx < CD for p IS IN such IRPIn F(IR!2) an inner product is defined by
(f,g) F(IRt )
2
that f
The corresponding norm will be denoted by II. I'F (IR t ).
2
(Remark: CD
f
e-u2/2dU =(2n)~)
._00
Definition 3.2. Let q € IN.
The mapping embq: ~ E F(IR!2) is defined by embqf = foE
q
foE • P
If no confusion arises, we identify f € ~ and embqf € F(IRt
2) and we denote embqf by f.
So ~ will be considered as a proper subspace of F(IRt
2). With these conven-tions we have the following relaconven-tions:
(f,g) F(IR! ) 2 (f,g) q M2 (IR ) and
tfI
c Hq+P c F(IRR. 2) n n for q,p E IN for q,p € IN II f I'FCIR! ) = 2 f,g € ~Lemma 3.3. Let m,n,p,q E IN with m
F
n. Then HP 1. Hq in F {IR 9..2} •
n m
Proof. Let r max{p,q}.
By (3.10) HP c Hr and ~ c Hr ,
n n m m
Since Hr 1. Hr in ~f, by (3.9) we conclude HP .1 ~ in F(IR9..
2).
n m n m
In the sequel the following subspaces of F(IR9..
2) will be of importance. Let
H
denote the subspacen
Let
H
denote the subspace~ ~.
q=l 00 00 u q=lo
We recall that ~ Span ( u
n=O ~) t
n i.e. the space of all harmonic polynomials in q variables.
Observe that
H
is the direct sum of theH
'5, soH
is the space of all har-nmonic polynomials of an arbitrary number of variables. For the spaces of de-finition 3.4 we construct a Hilbert completion with respect to II. If(IRt )'
2
To this end some auxiliary results are needed,
Lemma 3 ,Let f E
H
and let x E IRt2' ThenI
f (x)I
~ exp (-2-)IIx
Ir
II
f If (IR9. )2
Proof, Since f E H I there exists apE IN such that f € ~ c ~f! for all q ? p,
From theorem 2.20 it follows
I
f (x)I
SoI
f (x)I
If
(E x)I
p IIxli
2 exp (-2-)11 fll F(IRt ) • 2o
(a) Then (fk(x»k is a convergent sequence in ~ for all x € IR~2.
(b) Moreover the function f: IR~2 ~ ~ defined by
f (x)
=
lim fk (x) ,k-"
satisfies the following conditions:
( -I) ... If( ) I x ~ exp (II - 2 -x If)l' ~m Ilf IL k 't-(IR ~ ) ,
and k+<x> 2
(ii) f is continuous on IR~2.
Proof. (a): Fix x E IR~2.
By lemma 3.5
So (fk(x»k is a convergent sequence. (b) (i): Observe that
If(x) I = lim Ifk(x) I ~
k-..
IIx If ,
:s; exp (-2-) hm
II
fk IhIR 11.) •k+<x> 2
(ii): Fix x € IR11.2 and e: > O.
Let k > 0 such that for all R, ~ k
2 II f k - f IL R, 'f (IR R, ) :s; exp -( (II x
II
+ 2 1) ) ~ b •2
Let 1 > 0 > 0 be such that Ifk(x) - fk(y) I <:
f
for all y ( IRt2 with IIx-yll<:o.Then for all t <:: k and all y wi th II x - y
II
<: 0:::;
~
ex(II
x If _01
xII
+ 1) 2) + .£ +~
(hl_
<II x II + 1) 2, <:~
So for all y €
mt
2 with II x - y II < 0
If(x) - fey)
I
< e: •o
Defini tion 3.7. Let M denote the Hilbert completion with respect to II • II F (:IR R,2)
of
H
which consists of all functions on mt2 which are pointwise limits of Cauchy sequences in
H.
The corresponding inner product and norm in
M
will again be denoted by ( • ) fA and II • 1M,
Let
M
denote the closure ofH
inM.
n n
Each element of M is the pointwise limit of a sequence of harmonic functions each of an arbitrary but finite number of variables.
Since
H
LH ,
n ~ m, andH =
n m u
nEIN
H
n it follows that M n L M , m n ~ m and that
M =: q) M. n=O n
The space ~~ is a closed subspace of
M
because of (3.9). SinceM
consistsn
of homogeneous functions, we have M n ~
=
~.n n
(3.13) Let pq,
P
and pq denote the orthogonal projections fromM
onto ~,M
andn n n n
W
respectively.Lemma 3.8. Let n € IN.
(a) pq +
P ,
q + 00, strongly .n n
(b) pq + I, q + 00, strongly .
Proof. (a): Take f E
M.
Let e: > O. We choose g €H
such that IIf - gil <t
Take q E IN with g E Hq. Then for p ~ q
IlPPf - P f II ~ IIPP(f -g)II
M + IIPPg - P gllM + lip {g - f)IIM
n n
M
n n n n<
Lemma 3.9. Let x € IR1
2. (a) For all f € M
I
IIx IfIf(x) ~ exp (-2-l
II
filM (b) For all f € Mn
Proof. (al: This is a direct consequence of theorem 3.6(b) (i).
(b): Fix f €
M .
n
Then (pqf) is a Cauchy sequence with limit f. From corollary 2.19 it follows
n q I f (x)
I
=
lim I [pqf] (x) I == limI
[Pqf] (E xlI
$ n n q q--- q -liE xllns...:.J..
n -1 q II pllfll <_ ~ lim sup ( q )M
q-+«>Irl!
nBecause of the previous lemma
M
f n ~ 0, and M have reproducing kernels. no
The computation of these reproducing kernels requires the following results.
Lemma 3.10. Let f €
M
and let y € IRt2. Then
Proof. Follows from property (1.6).
Lemma 3.11. Let y €
nu
2• 2n
Then
(K~(YIY»q
is a convergent sequence in IR with limitII~I~
Proof. Let n ~ 1. From the proof of lemma 2.18 it follows
liE
ylln (2n +q -2) (n -2 +q -1) ... (q -1) ---.,;q=..,.-_ (2n -2 +q) (2(n -2) +q) ••• (q) n! _ II ,,112nSo lim Kq(y,y) n - ~. n! The case n = 0 is trivial. q
-o
Defini Hon 3. 12. The functions Kn' n :2: 0 and K from
:m 1/,2
x:m 1/,2 into mare
defined by
n
K ( n x,y -) _ (x.y) n!
K(x,y) = exp(x.y)
Theorem 3.13. Let y Em1/,2 and let n :2: O. Then we have (a) (b) (c) (d) n Kn ( 0 , y) E M and
II
K (., y)II
M =!Lx!L .
n n.frlT
For each f E M nII ¥II
2 K ( • , y) E M and II K ( 0 , y) II M = exp (-2-) • For each f E M fey) = (f,K(o,y»M •Proof (a): For q,p E IN we have
=
Kq+P(y y)n '
From Lemma 3.11 it follows that (Kq(o y» is a Cauchy sequence. Because of n ' q
Lemma 3.9 lim Kq(o,y) exists pointwise and is an element of M •
q~ n n
Let x E m2
2, Since
So with Lemma 3.11 and Lemma 1.11
lim
K~(X'Y) = ~Ixllnllylln (II~II
q-l'CO
y n
1fYII)
n
11
K ( n ., y)11
M=
lim IIKqnO, Y {)11 -
M - - -
lIyllq-l'CO
Irl!
(b): Fix f EM. Then
n fey)
=
lim q~ lim q~ [pIf] (y) = n (f , K~ ( 0 , y) )M
(f,Kn(o,y»M (c); From (a) it follows that00
exp(O.y) EM.
2 So K(-,y) E
M
and IIK(·,y)IIM
=exp(II~1I
). E Y q liE yll) q 1 n = - , (x,y) n.(d) Since point evaluation in M is continuous and Af
=
K is the reproducing kernel of M.$
n=O
M
it follows thatn
o
Due to the fact that the elements of
M
can be approximated by harmonic func-tions in finitely many variables, certain properties of harmonic funcfunc-tions carryover to the functions of M. The following two theorems are results of this type. The first theorem is a generalization of the classical Mean-value theorem for harmonic functions.Theorem 3.14. Let f E M, let x E IR12 and let r > O. Then
f (x) lim 1
f
f (.;) do 1 (.;>=
.
q~ C1 r q- q-q-1 q-l S (E x,r) q(3.14) Proof. Put
a.
q 1f
q-l S (E x,r) q f{~)do 1(~) •q-The sequence (Pqf)q is a Cauchy sequence in M with limit f. Since [Pqf] (x)
=
[pqf](E xl and since pqf is harmonic we haveq
Ia.
-[pqf](x)I
= 1I
(f (~) - [Pqf](t;) )do 1 (~)I
q q-l q-o q-1 r q-l S (E x,r) qJ
2s 1 exp(tilL)lif - pqfll,.fOq_l u;;)
q-l 2 o q-1 r q-1 S (E x,r) q So lim
a.
- [pqf] (x)o
or q-+«> q lim Ci lim [Pqf] (x) f (x).
q-+<x> q q-+«>The second theorem is a weak version of the min-max principle of harmonic functions.
Theorem 3.15. Let D be a bounded open subset of IR~2 with boundary
r
such that D is weakly closed in IRi2. As for harmonic functions, we have V D
I
f (x)I
s
XE sup
ncr
If
(n)
I .
Proof. Let r > 0 be such that D c {x ( IR~2
I
II xii < r}. Then we have2
V - If(x) I ~ eXP (2)lIfIl
M
XEDSince
r
=
D\D, d=
sup If(n) I < 00.nEr
Now suppose the assertion (3.14) were not true.
Then there exists y E D such that if(y)
I
> d. Let p > 0 be such that:S;
:'
If(y) I = d + P. Since D is open there exists z E E D for sufficiently large q
q such that
(a) If(z) - fey) I <
t
p(b)
From (b) i t follows
1 < - p
3 for all XED .
We obtain
IcFlf](z) I = If(y) - fey) + fez) - fez) + CPqf](z)
I
~?: If(y) I - If(y) - fez) I - If(z) - [Pqf](z) I >
1 1
> d + p -
3
p -3
p =Further we have
sup ICpqf](n)
1'5
sup <If(n)1+
ICpqf](n) - fen)1<
nEr nEr
1 < d + } p . So we are in the following position:
There exists a function, viz pqf, on IRq which is harmonic and satisfies ICPqf](z) I > d +
t
p and sup ICpqf](n) I < d +t
p •nEr Since D is open in IRt
2, observe that EqD is open in IRq. We have z E E D.
q
We are ready if we can prove that the boundary of E D is contained in E
r.
q q
Now Eq is a compact operator and D is weakly compact in IRt
2, whence Eq(D) is compact in IRq
Therefore E (0) is closed in IRq and E (D) c E (0). Since
q q q
~\E (D) C {E (O}}\E (D) C E (D\D)
q q q q q
the boundary
r
of E (D) is a subset of E(r).
q q q
E
(n
(3.15) This is contradictory to
I
[Pqf](z)I
> d +~
p > sup nEE (f) qI
[Pqf] (n)I
2: supI
[Pqf] (n)I .
nEfqo
Next we present two orthonormal bases in
M.
One of them arises from an expli-cit construction, the other is given expliexpli-citly. This construction is by in-duction and is based on a generating principle due to Muller. This principle is formulated in the following lemma.Lemma 3.16. Let f denote the restriction to sq-l of a harmonic homogeneous polynomial p of degree m in q variables.
Let n 2: m. Let F: sq + ~ be defined by F(/1 -
t2~
+ te 1)=
Aq+l(t)f(~)
q+ n,m
q-l
~ E S , t E [-l,lJ •
(For the definition of Aq+1 see (1.20).) n,m
Then F is equal to a restriction to sq of a unique harmonic homogeneous polynomial of degree n in q+l variables.
Proof. See [Mu], definition 4 and lemma 15.
o
Before we proceed we mention that the numbers dq
n dim
(H
q
) satisfy the
rela-n tion n
L
m=O as a consequence of corollary 2.7. Definition 3.17. Let eO,l=
1.Let {e 1,e 2} be an orthonormal basis for H2, n 2: 1.
n, n, n
Let n,i E IN with i > 2.
There exist unique q E IN such that
Because of the relation (3.15) there exist unique m,j E IN u {oJ with
o
$ m < nand 1 $ j $ dq such that m(3. 16)
n
i
=
L
k=m+l
The function e 0: IRt2 + ~ is defined by
n,~ E q+ lx ~
a
e 0 (x) n,~ te 1 +11 -
t2~,
t E [-1,1],~
E sq-1 q+a
So from {eO,l} U {en,i
I
n E IN, {eO 1} U {e 0 n E IN, 1 ~ i ~, n,~
E q+ tX
1 ~ i ~ dq} we obtain the set n
dq+1 } and so on.
n
O.
{ }
H
2.Theorem 3.18. For each n E IN let e 1,e 2 be an orthonormal basis for n
, n, n,
Let e 0' n,i E IN, i ~ 3 be constructed as in definition 3.17.
n,~
Then we have
(a) For all n,q E IN with normal basis of ~. n q ~ 2 the set {e 0 n,~ I 1 ~ i ~ d q } is an ortho-n
(b) For all n E IN the set {e 0 i E IN} is an orthonormal basis of
n,~
M .
n(c) The set {eo, 1 } U {e
n, i n,i E IN} is an orthonormal basis of
M.
Proof. (a): The proof is by induction. First we note that
mal basis of
H2.
for all n E IN the set {e 0
I
1~
i~
d2
} is an
orthonor-n,~ n
n
Next we assume that for all n E IN the set {e 0
I
1n,~ is an
orthonor-mal basis of ~.
n
Starting from this assumption we show that {e 0
I
1~
i~
dq
+1} is an
ortho-n,~ n
normal basis of Hq+1 for all n E IN.
n
Fix n E IN. Let il E IN with
d~
< il~ d~+l.
As we have seen there exist unique m1,jl such that e 0 is given by formula
n'~l
(3.16) in which we put i i
1, m = mt and j = jl.
Because of lemma 3.16 e 0 E Hq+l.
Because of lemma 1.12(b) for all x E IRq+1\{O}
IITt(n +
~2)
L 1""2 n q+ x
( + 1 ) II xII A (-II -II .eq+1)
r(n + _q ____ ) n,m x
2
So also the e . with 1 ~ i ~ dq are given by formula (3.16) in which we put
nt~ n
m == nand j == i.
k . . l ' i dq+1
Ta e ~1,12 E IN with ~ 1
1, 2 ~ n .
For j = 1,2 take the unique mn,jn E IN U {a} such that e . is given by
.. .. . . n,l£,
formula (3.16) in which we put i == i£" m == m£, and J == J
t . We note that 1 ~ jt ~ dq , t 1,2. mR. Put a, (e . , e . )
M"
n'~l n'~2 Lemma 2.14 says + 1f
e i (ll)e . (n)dO" (n) n , 1 n'~2 qTake variables t E [-l,lJ and
~
E sq-l such that nand dO" (n) '" (1 - t2) 2 dO" 1
(~)dt
q q-2n--,----,.--:"- r
(n + 2n (q+1) /2 -1 1f
q-lf
SBecause of the induction assumption we get
Finally, Lemma 1.12 and Lemma 2.14 yield
a
=
{ I
dq+1} ,~+1Thus we have proved e . 1 ~ i s i s an orthonormal basis of
n-n,~ n n for arbitrary n. (b): Fix n € IN. Because u q~2 sis of M • n
tf.
is dense in M , the set {e . i E IN} is an orthonormalba-n n n,~
(c): Since M (i
n=O
M this statement follows trivially from (b).
n
The second basis consists of homogeneous polynomials which are not harmonic in general.
Definition 3.19. Let MI denote the set of multi indices
00 Let Is I
=
I
j=O Sj' s!=
s 1 ! s2 ! •.. anda
s,t IT <5 1 s.,t. j= J J Put MI == {s E MIn
I
lsi=
n} for n ~ O. For each s € MI we define q>: s IRR,2 ~ It by lP (x)=
s II j=l s. J (x.e . ) JPut ~ = {q>
I
s € MI} and ~=
{q>I
s € MI }.s n s n
Theorem 3.20. Let n E IN. Then we have (a) Vs EMI : q>s c Mn·
n
(b) The set ~ is an orthonormal basis of M •
n n
(c) The set 4> is an orthonormal basis of
M.
q
Proof. (a): We first construct auxiliary harmonic functions gs,k with s € MI and k,q E lN large enough.
n Fix k E IN.
For 1 s j s k, q ~ k we define the operators D~,k E L(IRR,2) by
f
m 0 q o::+1
1 s j - 1 D. k e . mk m ~ J, J+ j S m ~ q - 1o
(3.17) (3.18) (3.19) (3.20) (3.21) The q Dj,keh operators 0, k q J, q 0, k maps J, Dq , k ' e
=
), J 0, have Span k 'i' q 2 LII
D4 kxll = j=l J, E Dq+P=
Dq q j,k j,k h ,. j + Ink, 0 :5m:5q- 1.
the following properties{e, Ink
I
o
:5 m :5 q - 1} isometrically onto IRq.
J+II
xii 2 for all x E IRkq .k
For all SEMI and k,q E IN with
n
g~,k:
IRJl.2 -+ (£ byI
SJ' - n and q ~ k we define the function j=l k gq k(x) = s, q q IT K (0, kx,e,) 1 S j ) ' JBecause of (3.17) the j-th factor is a harmonic homogeneous polynomial of degree s, in the variables x x x So gq is a harmonic
ho-J j ' j+k'···' j+(q-l)k· s,k
mogeneous polynomial of degree n in the variables x
1' ••• ,xqk.
k k
Claim: For S,t E MI , k E IN with
I
s,I
t,=
n, p ~ 1 and q ~ k wen j"'l ) j=1 J have Proof: Put (l Then we have (s!t!)
~
( l=
--~--~--~ (21T) (q+p) k/2 Because of (3.20) q q+p K t j ) ' (0, kx,e,) J=
k q II K (e"e,)6 t 1 tj J ) Sj' jII
xii 2f
IR (q+p) /k 2 k q+p q+p q q II K (0, kx,e,)Kt (0, kx,e,)dx. , 1 s, J, J , ) , J J = ) J eApplication of the transformation of variables x -+ (Y1'.'.'Yk) with
y,
=
Dq+Px yields because of (3.19)k
- ( I
j=l 2 lIy.1I )/2
J k e 2 lIy.1I _ _ J _ q+p qIT K {y.,e.)K (y"e.)dY1, .. dYk
1 s, J J t. J J j= J J k = (s!t!)
~
IT j=l [ 1 ~+pI
e 2 q+p q K (y.,e.)K t (y.,e.)dy.J s, J J . J J J s! IRq+p k q IT K (e.,e.)o t . 1 s. J J s " . J= J ] JThis proves the claim.
k
Let <P € ¢> • Let k E IN with I s , = n.
s n J
j=l
We will prove that gq ~ <P , q ~ ~, in s,k s
From claim (3.21) i t follows that for q
J J
M •
n ~ k, P ~a
== s! k IT j=l q+p K (e"e.) Sj J J - s! k IT j=l q K (e.,e,) Sj ) JBecause of Lemma 3.11 the sequence k
IT
j=l (gq k is a Cauchy sequence in
M •
Lets, q n
strongly to Dj,k € L(IR!2) defined by
{ e, D, ke. mk = e J J, J+ m e m+1 m
=
0 1 ~ m < j m ~ jKq (e"e,» is convergent. Hence s. ) J q
J q
1 ~ j ~ k. The operator Dj,k tends
=