Citation for published version (APA):
Eijndhoven, van, S. J. L., & Graaf, de, J. (1984). On distribution spaces based on Jacobi polynomials. (EUT-Report; Vol. 84-WSK-01). Technische Hogeschool Eindhoven.
Document status and date: Published: 01/01/1984 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.
ONDERAFDELING DER WISKUNDE EN INFORMATICA
DEPARlllliNT OF MATHEMATICS AND COMPUTING SCIENCE
On distribution spaces based on Jacobi polynomials
by
S.J.L. van Eijndhoven and J. de Graaf
EUT Report 84-WSK-OI ISSN 0167-9708 Coden: TEUEDE
Eindhoven March 1984
by
Summary
Preliminaries
Analyticity spaces and trajectory spaces Entireness spaces and ultra-trajectory spaces
Some notations
Survey of results
I. Classification of analyticity spaces and entireness spaces
2. The classification of analyticity spaces and entireness spaces generated by the Jacobi polynomials
3. The characterization of the analyticity spaces and entireness spaces based on the Jacobi polynomials, with applications to classical analysis 2 5 8 10 17 33 44 3.1. The characterization of 3.2. The characterization of
s (
)
andT(L2
([-n,n]),6V) L 2 [-norr] ,6v the spaces 46 2 _1 ( 2 . d2 d)V and L2 ( [-1 ,1] ,(I-x ) 2dx) ,I - (I-x ) - 2 +x dx \ dx ( 2-!
(
2 d2 d)V)
T,L
2 ([-I,I],(I-x) dx, -(I-x) dx2 +K dx 'and related classical analytic results 61
4. Hyperfunctions and trajectory spaces 77
5. Ultra-hyperfunctions and ultra-trajectory spaces 98
6. Representation of some large continuous groups as groups of
continuous linear operators on spaces of (ultra-) hyperfunctions 120
Appendix 125
The Jacobi operator
A a,S
2 d2 d
- (I-x) - «S-a) - (a+S+2)x) dx ' dx2
a>-I,S>-I,
1S self-adjoint and positive 1n the Hilbert space
X
a,S
We study the analyticity domain DW(Av Q) and the entireness domain a,1-'
00 v
D (exp(A Q» of the v-th power, v ~
!,
of the operator A Qin X Qa,1-' a,1-' a,1-'
It is shown that for fixed v the analyticity and entireness domains do not depend on a and S. Moreover,for each fixed v the mentioned domains are characterized as spaces of analytic functions of suitable well-described
(growth-) classes.
Next the spaces DW(Av Q) and Doo(exp(Av Q» are considered as test spaces for
a,1-' a,1-'
distribution theories. They are very special examples of general functional analytic constructions as g1ven by De Graaf and Van Eijndhoven.
The distribution spaces (dual spaces of
described in detail and many natural examples of continuous linear function-als and (extendible) continuous linear mappings are given. These examples are based on simple geometric and analytic considerations. Further, expansions of distributions in Jacobi polynomials and Jacobi functions of the second kind are studied and sharp estimates on the expansion coefficients are produced. For v
=
!
the relation with (ultra-) hyperfunction theory is discussed.Finally, some large groups of analytic functions are represented as (extendi-ble) linear operators ,on the mentioned spaces.
In the past decennia generalized functions have been introduced in mAny different ways. We mention Schwartz's distribution theory in which general-ized functions are regarded as continuous linear functionals on some locally convex space of good functions, Jones' theory on generalised functions, in which generalized functions are weak limits of regular sequences of good
functions, and the theory on hyperfunctions in which generalized functions are 'boundary values' of analytic functions defined in a region of the complex plane. Probably the easiest way to introduce generalized functions is by means of formal series expansions with respect to some orthonormal basis in a Hilbert space. Here we shall clarify this method a bit more.
00
Let (vn)n=O be an orthonormal basis in a Hilbert space X. The Riesz-Fischer
00
I
a v where the sequences n=O n na class which contains
l2.
A candidate for the dual of E00
considering sequences (b )00 0 for which
I
I~ bI
< 00n n= n=O n n
E.
Let 0 denote the vector space of the related series00
(an)n=O belong to can be obtained by
00
for all (an)n=O ~n
00
theorem says that each f in X is represented in
l2
by the l2-sequenceN
«f,vn»noo=O' and Ilf -
I
(f,v)v 11+0 as N + 00. Now consider a vector spacen=O n n
E
the elements of which are formal seriesI
b v . Then it is clear that 0 c X c E. Thus we get a functional analytic n=O n nanalogue of the so called pans ion theory [7] developed by Korevaar. We note that Korevaar looks at the Hilbert space L
2(IR) with the Hermite functions as an orthonormal basis.
Analyticity spaces and trajectory spaces
A simplified version of the theory [6] on generalized functions developed by De Graaf can be described as follows. The space
E
consists of all formal00
series
I
a v where the sequences (an)noo=O satisfy n=O n n(0. 1)
-;\ t n
Vt>O: sup (Ia Ie ) < 00.
nEIN n
of the positive self-adjoint operator
H
I
n=O spectruma
< A O~ Al ~ A2 ~ •.. and { A I n E IN u {a}} is the nHere (An):=O is a fixed sequence of positive numbers with the property that 00 -;\ t
e n < 00 for all t > O. The set
loosely defined by (0.2)
Hv
n Avn n n=0,1,2, . . . .
00 00
Let (an)n=O be a sequence with property (0.1). Then to (an)n=O we let cor-respond the mapping F: (0,00) + X defined by
F(t) 00
I
n=O -A t n e a v n n-TH
It is clear that the mapping F satisfies F(t+T) = e F(t), t,T >
o.
-TH
Conversely, if a mapping G: (0,00) + X satisfies G(t+T) = e G(t) for all
can be identified with the set of all So the set of all sequences
a v .
n n
t,T > 0, then there exists a sequence (a )00 0 which obeys condition (0.1) 00 -A t n n=
and for all t > 0, G(t) =
I
e n n=O 00(an)n=O determined by condition (0.1)
mappings F from (0,00) into X with the mentioned property.
Since each positive self-adjoint operator
A
generates a one parameter sem1-group of so-called smoothing operators, the previous considerations lead to the following more general definition.(0.4)
DefinitionLet A be a positive self-adjoint operator in X. Then the space T~,A'
the trajectory space, consists of all mappings F: (0,00) + X with the
called property that
Heuristically, the space
T
x
A can be regarded as the space of 'initialcon-,
du
ditions' u(o) to the evolution equation dt - Au, which give rise to a solution t ~ u(t), t > 0, through X.
The topology for the space
Tx
,
A is generated by the seminormsF t+ IIF (t)IIX' t > 0 , F E TX,A •
With these sem~norms
Tx,A
becomes a Frechet space. Moreover,Tx,A
is ~10nteliff the operators e-tA, t > 0, are compact;
TX,A
is nuclear iff the operators -tAe t > 0, are Hilbert-Schmidt. (Observe that
T
XH
is nuclear.)00 _~ t
Let f E X, f
L
(f,v)v. satisfy (f,v ) = O(e n) for some t > O. Thenn=O n oon n for every F E
T
XH'
F =L
anvn, we have " n=O I N ( N 2 -2AT)2(
N 2I \;
(f,v)1
~I
la l e n \L
1
(f,v)1
n=O n n n=O n n=O n
If we take 0 < T < t the right-hand side converges if N + 00. So the series
00
a
(f,v ) is absolutely convergent. We remark that the order estimate onn n
00
«f,v» 0 implies that f is contained in the domain of each
n n=
unbounded self-adjoint operator e
TH ,
0 < T < t. This leads to the generalI
n=O
the sequence
(0.5) Definition
u
t>O
-tA
{e
w.
I t > 0, W E xl .It is clear that Sx A is a dense subspace of x. Since Sx A consists of
,
,
precisely all analytic vectors of A, we call the spaces of type Sx A
analyt-,
icity spaces.
The natural topology for the space Sx A is the inductive limit topology
,
induced by the spaces e-tA(x) with Hilbert space topology generated by the
. tA tA -tA
1nner product (f,g)t := (e f,e g), f,g E e (X). This inductive limit is not strict. The construction of seminorms which generate the inductive limit topology has led to a thorough description of several topological features of SX,A' For example, the sequence (fn)nElN is a null sequence in SX,A iff
tA
there exists t > 0 such that (e f ) :IN is a null sequence in x. Further, n nE
we note that Sx A is complete, bornological and barreled.
,
On Sx A,
xT
X,
A we introduce the pairing <","> by(0.6) <f,G>
The definition (0.6) makes sense for T > 0 sufficiently small and it does not depend on the choice of T. Through this pairing, all strongly continuous
functionals on SX,A are represented by the members of
Tx,A'
Conversely, all continuous functionals onTX,A
are represented by the members of SX,A" The spaces Sx A and,
T
X,
A are in duality.Entireness spaces and ultra-trajectory spaces
Another distribution theory which is a considerable generalization of tHe theory of tempered distributions, has been developed by Van Eijndhoven, see
[2]. In order to introduce this theory along the lines of formal series expansions we start with a fixed sequence of positive numbers (~n):=O which are ordered, 0 < ~O ~ ~I ~ ••• and ~n + 00, n + 00. Further, there has to be
~ e-~ntO
to > 0 such that L < 00. Now for the space
E
we take the vectorn=O 00
space of all formal series
I
a v where the sequences (a )00 0 satisfyn=O n n n n= (0.7.a) 3t>0: sup la Ie-~nt < 00 nElNu{O} n or equivalently (O.7.b) 00 -2~ s 3 V •
I
la1
2 e n t>O s~t' n=O n 00To a sequence (an)n=O with the property (O.7.b) we associate a mapping ~
from [t,oo) into X as follows:
Note that Hs) 00
I
n=O s ~ t .On the other hand, let ~: [t,oo) + X satisfy
Then there exists a sequence (b)oo which satisfies (O.7.b) and n n=O'
'P(s) 00
I
n=O -fl nS e b v n n s ?': tWe thus arrive at the following general definition.
(0.8) Definition
Let
A
be a positive self-adjoint operator in X. Let Xt' t > 0, denote the space of all mappings ~ from [t,oo) into X satisfying
Then the space a(X,A) is defined to be the inductive limit
a(x,A)
(We note that X c X for all t, 0 < t < T. )
t T
The spaee Xt IS a Hilbert space with inner product (iP,'P)(t) We note that X
t is a copy of X.
The space a(X,A) is called the ultra-trajectory space. Inspired by [6] explicit semihorms have been constructed which generate a locally convex topology equivalent to the inductive limit topology. It has been proved that a(x,A) is complete, bornological and barreled. Further, a(X,A) is Montel iff
-tA .
tA
e IS compact for some t > 0; a(X,A) is nuclear iff e- is
Hilbert-Schmidt for some t > O.
Suppose that the Fourier coefficients (f,v ), n E IN u {O}, of an element
n
-fl t
f E X satisfy Vt>O: (f,v
n) = O(e n) or, equivalently, Vt>O:
00 2 f l t 00
I
I
(f,vn
)!
e n < 00. Then for any sequence (an)n=O satisfying (0.7) then=O 00
series
I
a
(f,v ) converges absolutely. So the following definition seemsn=O n n
(0.9) Definition
Let A be a positive self-adjoint operator in x. Then the space T(X,A) is defined by
T(X,A) A
00
D(ee ) ) •
The space T(X,A) is called the entireness space, because it contains all entire vectors of A. We note that wis an entire vector of A iff
V 3 V IIAn II :<=::: n'. anb.
a>O b>O nEIN: w. With the seminorms
tA
wi+ lie wll
X ' W E TeX,A), t > 0 ,
Tex,A) becomes a Frechet space.
The pairing between the spaces Tex,A) and aex,A) is defined as follows. Let w E TeX,A) and let ~ E a(X,A). Then
eO.IO)
where s > 0 has to be taken so large that ~ EX. We note that (0.10) does
s
not depend on the choice of s. With this pairing a(X,A) is a representation of the strong dual of TeX,A) and, conversely, T(X,A) is a representation of the strong dual of aex,A).
From a topological point of view the spaces Tex,A) and
T
X,
A'
and aex,A) and Sx A have similar properties. So the theory [2] can be considered as a kind,
of reverse of the theory [6]. For instance, a sequence e~) IN in aex,A) isn nE a null sequence iff there exists s > 0 such that (~) IN c X and
n nE s
II~ (s)1I -+ 0 as n -+ 00. Finally we mention the following inclusion scheme
n
Tex,A) c Sx A c X c
T
x
A c aex,A) .SOME NOTATIONS
In this paper we consider the Hilbert spaces
and the positive self-adjoint operator A S 1n X
a, a,S
2 d2 d
A
= -
(I-x) --- - «S-a) - (a+S+2)x) dxa,S dx2
where we take a and S larger than -I. The ouerator A. a,S has a discrete
spectrum: {n(n+a+S+l)
I
n E IN u {O}}. Its normalized eigenvectors are the normalized Jacobi polynomialsR~a,S),
1
ra + 8 + 2n + I f(n+l)f(n+a+S+I)]2 p(a,S) L 2a+S+1 f(n+a+l)f(n+S+I) n
where
(d. [8], p. 209).
In this report we shall work out the following program: Classification of the space
take v 2
!
and a,S> -I. Weand (y,o) and for all
v
Sx
(A )V and T(X S,(A S)) where weB' S a, a,
a, a,
get the following result: For all pairs (a, S)
and, also, V T(Xa,B' (Aa,s)) \! T(X 0' (A 0) ) y, y,
Y Cl,S and C a,S d2 1 d d
de 2 + (8-0.) sin e
ere -
(0.+13+1) cot e de It is clear thatn(n+a+S+l)R(a,S)(cos e) •
n
Moreover, for all v
and V T(Y Q , (C (3) )
=
0.,1-' Cl, V T(Yy,<5'(Cy,<5) )Characterization of the spaces S (C _l)V and Y_1 _1' I
2 ' 2 2, 2
for each v ~
!,
1n classical analytic terms. Here we employ the relations. and
(_1 _1)
W
R 2, 2 (cos e) = - cos ne
n I T ' n E 1N
By means of this characterization we can also describe the spaces
Relations between the spaces
T
1X Q'(A (3)2
0.,1-' a, some classes of hyperfunctions, respectively
1
and 0"(X Q , (A (3) 2) wi th
0.,1-' a, ultra-hyperfunctions.
SURVEY OF RESULTS
This University Report is a contribution to our research project of making a
link between our general functional analytic theory on analyticity spaces,
trajectory spaces, entireness spaces and ultra-trajectory spaces on one hand
and "classical" analysis and distribution theory on the other hand. Similar
results have been achieved in the papers [I], [3], [4] and [5] where test
spaces have been introduced in which the Hermite functions and the Laguerre
functions serve as bases in the way pointed out in the preliminaries. In the
present paper the main emphasis is on the test spaces in which the Jacobi
polynomials establish bases. So the elements of these test function spaces
are determined by sequences of expansion coefficients of certain growth
orders.
We give a detailed classification of such test spaces and characterize them
as classes of analytic functions which satisfy specific growth conditions.
This characterization enables to describe a variety of continuous linear
functionals and continuous linear operators on these spaces in classical
analytic terms. As a further result we find conditions on the asymptotics of
the coefficients in the Jacobi series expansion of an analytic (entire)
function which belongs to such a growth class. Finally, we show that the
duals of some of our test spaces can be represented by spaces of (ultra-)
hyperfunctions. Each (ultra-) hyperfunction can be expanded in so-called
Jacobi functions of the second kind. We describe the convergence of these
expansions in terms of complex analysis.
Now we go into more detail.
In Chapter I we study the general classification problem which can be stated
as follows. Let there be given a separable Hilbert space X and a positive
,(X,A)
=
,(Y,B), both set theoretically and topologically. For instance, ifW
is a continuous bijection on Sx A' respectively ,(x,A), then we may take,
Y=x
W
andB
=W-1AW.
Herex
W
denotes the completion of Sx A (,(x,A»,
with respect to the norm II'IIW= IIW'IIX' Special attention is given to the case
X
=
lZ and A=
Awith Aa diagonal operator in lZ' Also we mention the case that B is a perturbation of A. This has been investigated in an earlier paper[4] and quoted in Chapter I for completene·ss. Further, we !lay attention to the interrelation between the spaces
Tx,A
andTy,B
which both are representa-tions of the dual S~,
A' if Sx A,
=
Sy,
B'In Chapter Z we apply the abstract results on classification of analyticity spaces and entireness spaces to the following concrete case.
Let Xa,s be the Hilbert space Lz([-I,I],(I-x)a(l+x)Sdx) and let Aa,S be the positive self-adjoint operator
A
a,S
Z d7 d d
- (I-x) --- - (S-a) -- + (a+S+Z)x dx
dxZ dx
as introduced in the Preliminaries. We show that for all pairs (a,S), (y,o) with a,S,y,o > -I and for all v 2
!
and
The coordinate transformation x cos e transforms the Hilbert space X ~n
a,S
a
S
Ya,S
=
LZ([O,n],(1 - cos e) (I + cos e) sin e de)and the operator A becomes a,S
C
a,S
d2 1 d d
de 2 + (S-a) sin e
de -
(a+S+I) cot e de We have for all pairs (a,S), (y,o), a,S,y,o > -I, and all v ~and
V
T(y 0' (C 0) )
y, y,
In order to apply the general theorems of Chapter 1, we derive rather subtle estimates for the matrix entries of the operators
d:
and xd~
with respect to the Jacobi polynomial bases. (See the Appendix to this paper.)Chapter 3 deals with the characterization problem. We look for a description
in terms of classical analysis, of Sy (C )V and T(Y (3'(C a)v).
a' (3 a, a,fJ
a,fJ a,
tion of the spaces
v the spaces
Sx
(A )v, T(X a,(A a))'a' (3 a,fJ a,fJ
a,fJ a,
To this end, we start with a
characteriza-,(_£)V
de2
and
These spaces can be presented as classes of 2TI-periodic analytic functions of well-defined growth behaviour dependent on v. Since
S v
Ya,fJa,(Ca,fJa) SY (C I)V
I I ' I -2'-2 -2,-2
the space S ~s the subspace of
S
(
d2)V which consistsYa, fJa'(Ca, fJa)V
L ([
2 -TI , TIJ) , - --2de
of all even functions in
S
(
d2 )V' Similarly, T(Ya ,S' (CC':,(3)v) L2 ([ -TI , TI J), - - 2characterization of the spaces
Sx
(A )V and T(Xa,~,(Aa,B)V) 18 then(X,r~ (X,(~
obtained by means of the conformal mapping w
=
cos z.Important consequences of this characterization are theorems on approxima-tion of analytic (entire) funcapproxima-tions by means of (normalized) Jacobi poly-nomials. We mention the following.
Let f be an entire function satisfying 2v
If(w)
I
s A eXP(B(log Iwl)2V-I) , Iwl 2 I ,where A,B > 0 and v >
!.
Let a,S> -I. Thenf 00
I
n=O a(a'S)R(a,S) n nwhere the coefficients a(a,S) satisfy
n
sup
la~a,S)
I exp(nv(t-s» < 00 nE1N2v-1
for all 0 < s < t with t
=
(2v)-2V(2~-I)
The seriesexp (-BOOg 00
I
n=O a(a,S)R(a.,S)(w) n nfor some v >
!
and some t > O. Then the func-b(a'S)R(a,S)(w) is entire analytic andn n
converges uniformly on ~. (a,S)
Let a.,S > -I, and let (bn )nElNu{O} be a sequence for which sup (lb(a,S)/ exp(nVt» < 00
nElN n
tion f defined by f(w)
=
I
n=O2v If(w) I S A
I (
2v)-2 V)2V-I Here Af > 0, and for B
f any number larger than (2v-l) t can be taken.
Let f be an analytic function in an open neighbourhood of the interval a(a,S)R(a,S) where the coefficients
n n
00
< 00 for some t > 0 depending on a
on a sufficiently small
neighbour-I
n=O exp(nt)) [-1,1]. Let a,S> -1. Then f
(a S ) .
I
(a S)I
a ' sat1sfy sup (a '
n nElN n
and S. The series converges uniformly
hood of [-1,1]. (Cf. [9], where sharper results have been obtained.)
Let a,S> -I and let (b(a,S))oo c ~lNu{O}. Then
n n=O
represents an entire function iff for all t > 0, < 00.
since
e5
z
for all v >
!,
we get:3 A ,.0 B,w 00
I
I
n=O 2v s~,w
eXP(B(IOg(max(l, [z 1)))2V-I)With the aid of these classical analytic descriptions and with elementary geometrical considerations we introduce natural classes of continuous linear
f~nctionals and continuous (extendible) linear mappings on all
S-
andIn Chapters 4 and 5 we limit ourselves to the case v ~. The duals
s'
Y Q'(C Q)~ a,1-' a,1-' ands'
1 XQ,(A
0)" a,I-' a,1-'are linked to three classes of hyperfunctions. The representation of the considered linear functionals as hyperfunctions is by means of contour integrals.
In Chapter 4 we introduce a natural Frechet topology on each of these hyper-function spaces. As a consequence we get the following classical result (see
[9], p. 250): A function 8 which is analytic on the region ~, [-I,J] with
8(00) = 0, can be expanded in a series of associated Jacobi functions, Q(a,S)
n defined by 'Q(a,s)(w) I n
=
21fi If
-Jhave also been obtained
I
(a S) -ntI
sup a ' e < 00nEIN n
outside each open converges uniformly
series satisfy: Vt>O (a,S). h" a ~n t ~s n a(a'S)'Q(a,S) n n 00
I
n=Oof [-J,I]. Results in this directions and the series
neighbourhood The coefficients
in [J0] .
In Chapter 5 the duals T'(L2([-1f,n]), - de 2} },( d2\4\ T'(Y ! a,S,(Ca,S)2) and
are treated in the same way as the corresponding S'-spaces in the previous chapter. These duals can be represented by classes of so-called ultra-hyperfunctions. Each space of ultra-hyperfunctions can be regarded as an inductive limit of Banach spaces. It follows that the spaces o(L2([-1f,1f]), (-
d:)~\),
o(Y o,(CQ)~)
and o(X o,(A 0)4) are homeomorphicde a,I-' a,I-' a,1-' a,I-'
to certain ultra-hyperfunction spaces. Finally, classical results are ob-tained concerning the expansion in associated Jacobi function series of functions which are analytic at infinity.
In Chapter 6 we deal with two large continuous groups of analytic functions.
These groups can be represented as groups of continuous linear mappings on
1. CLASSIFICATIONS OF ANALYTICITY SPACES AND ENTIRENESS SPACES
In this chapter, we discuss two methods to classify analyticity spaces
and
entireness spaces. First, we consider the case of a general positive self-adjoint operator in a Hilbert space. Next, we turn to a more concrete situation.Let X denote a Hilbert space and let A be a positive self-adjoint operator 1n X. In our paper [5] we have proved the following classification theorems,
in which a central role is played by perturbations.
(1.1) Theorem
Let P be a linear operator in X with O(P) ~
Sx
AV
where we take v > 0 fixed.,
Let the following conditions be satisfied:
(i) There exists a Hilbert space Y such that the operator exp(-tAv) maps X into Y for all t > O.
(ii) The operator A + P defined on
Sx
,
AV
can be extended to a positive self-adjoint operator in Y (denoted by A + P also).(iii) There exists a monotone non-increasing function ~ on the open interval (0,I) such that
Remark. In fact we proved that under the conditions of Theorem (I. I) the following result is valid: V 0 3~ 0 3 0 such that the operator
t> t> T>
V ~ V V .
exp(TA )exp(t(A+P) )exp(-tA ) is bounded on X.
(1.2) Theorem
Let P be a linear operator in X such that O(P) ~ exp(-tAV) (X) for some t > 0 and some v > 0 fixed. Let the following conditions be satisfied:
(i) There exists a Hilbert space Y and there exists t > 0 such that the
operator exp(-tAv) maps X into Y.
(ii) The operator A + P defined on T(X,Av) is entendible to a positive and
self-adjoint operator in Y.
(iii) There exist positive constants d and r
Oand there exists q, 0 < q < v such that
v -I v q
Ilexp(rA )PA exp(-rA)1I < dr .
A
v vThen T(X, ) C T(Y,(A+P) ).
Remark. In fact we proved that under the conditions of Theorem (1.2) the following result is valid: V~ 0 V 0 3 0 such that the operator
t> T> t>
v ~ v v
exp(TA )exp(t(A+P) )exp(-tA ) is bounded on X.
Next we consider another way of classifying analyticity spaces and entireness spaces. We prove that each continuous bijection
W
on Sx A,
(T(x,A)) gives rise to a Hilbert space xWand a positive self-adjoint onerator AWin xWsuch that S A
=
SX, XW,AW
W W
(T(x,A)
=
T(X ,A )).We denote the inner product of X by (.,.). Let
W
be a continuous bijection -Ion the analyticity space
Sx
A. Since also its inverse mappingW
iscon-,
tinuous on
Sx
,
A the sesquilinear formturns Sx
,
A
into a pre-Hilbert space. By xW
we denote the Hilbert completion of Sx A with respect to ( ,,
)W.
Then we have(1.3) Theorem -I
The operator
W AW
with domain Sx A is positive and essentially self-adjoint,
in
XW.
IfAW
denotes its unique positive self-adjoint extension~n
X
W
,
thenSX,A
=S
W W·
X ,AProof
Let f
W
-Ig E Sx A' Since g,
E Sx A there exist a,b,
> 0 such thatHence
n E IN.
for all n E IN. Thus it follows that the analyticity domain of the operator
W-IAW
in xW
contains Sx,
A'
Since Sx A,
~s
dense inx
W
,
the analyticity domain-) W -I
of
W AW
is dense in X . So following [II], Theorem 8.31,W AW
isessen-• 11 If d" .
x
W • 1 hW-IA(t)'
. .
Sx
Wt~a y se -a Jo~nt ~n • It ~s c ear t at ~s pos~t~ve on XAC •
,
Hence the unique self-adjoint extension
AW
is positive. Since S contains xW
A(JJprecisely all analytic vectors of
A
W
,
we obtain SX,A CSXW,AW'
on'SX,A we have for all t > 0Now observe that the linear mapping (JJ on Sx
,
A
can be extended to a boundedW
linear operator from X onto X. We denote this extension also by
W.
It follows that for all X E x(JJThe assertion
S
~S
is then obtained from the observation thatx,A
xW
AW
W '
1Wx
E X for all X E X and thatW-
is continuous onSX,A.
n
Since
Sx A
,
elements GS W W'
X
,A
E T
x
,
Athe elements f E
Sx A
can be paired both with the,
TA
<f,G> = (e f,G(T», T > 0 sufficiently small, and with the elements H (
<f,H>W
A
W
(eP f,H(p»W' p > 0 sufficiently small.
Let £ be a continuous linear functional on
Sx
A.
Then there exists,
Tf£} E
Tx,A
such that £(f) = <f,T(£», f ESX,A'
and alsoTW
(£) ETxW,AW
such that £(f) =<f,TW(£»W'
f ESx A.
SO the antilinear mappings T: £ ~ T(£),
and
TW:
£ ~TW
(£) are isomorphisms from S~,A ontoTX,A
and from S~,A ontoT W W'
respectively.X ,A
We shall investigate the relationship between T(£) and
TW
(£). To this end, we define the mappingj
on Tx
,
A by-I
j(F): t ~
W
F(t) , t > 0, F E Tx
,
A •W
It is clear that (j(F»(t) E
Sx,A
C X for all t > 0, and also that for allt,T > 0
(j(F»(t+T)
W
-I F(t+T)Hence j(F) E
T W W.
Further it follows that j 1S a continuous linear mapp1ngX ,A
and that j is invertible with
-I
Now let f E S~
,
A'
Then for all f ESx
,
A'
-I '
<f,j(T«W
)
f))>WWe thus obtain
In the same way
It leads to the following result
-I '
T
W
=joT
0(W
)
T-I
j 0 T
W 0 W'
which expresses the relation between the representations
T
andTW'
We now repeat the above considerations for the entireness space T(X,A).-)
Let
V
be a continuous bijection on the Frechet space T(X,A). ThenV
is also continuous on T(x,A). With the inner productT(x,A) becomes a pre-Hilbert space. Let
xV
denote the Hilbert spacecomple--I
tion of T(X,A) with respect to
(·,·)V'
It is clear that the operatorV AV
with domain T(X,A) c
xV
is symmetric and positive. Similarly to Theorem (1.3) we have(1.4) Theorem
-I V
The operator
V AV
~s essentially self-adjoint and positive. Let A denote its unique self-adjoint extension inXV.
ThenProof
T(X,A) = T(X
V V
,A ) .By standard arguments it can be proved that T(X,A) is contained in the
-I V
analyticity domain of the operator
V AV.
Since T(X,A) is dense in X , it follows thatV-1AV
is essentially self-adjoint. The operatorV
defined on T(X,A), can be extended to a bounded operator from xV into X. We denote thisextension also by
V.
Conversely, for each g E X the linear functionalo f (Vf ) . . V h . h xV h h
~: ~ ,g ~s cont~nuous on X . So t ere ex~sts E sue t at
g
(Vf,g)
=
(f,h)V •Hence g
Vh.
Therefore we write h-I V g E X# V g E X . -I V g, and We thus obtain V-I(e-tA(X) ) and finally
T(X,A) V-I (n e-tA (X) )
t>O
n
V-I (e-tA(x)) t>On
t>O (V- Ie-tAV) (XV) =
n
t>Ov
V T(X ,A ) •o
Next we apply Theorems (1.3) and (1.4) to the following concrete case. Let
H
be a positive self-adjoint operator in the separable Hilbert space X, and letH
have a discrete spectrum. Then in X there exists a complete< 00
for all t > O. The spaces ,(X,H)
ex> -Akt
t > 0 such that
I
e k=OWe define the unitary operator
U
from X into i2 by such that He = A e • The spacesn n n
orthonormal basis (en):=o and there are positive numbers Ak , k E IN u {OJ,
00 -Akt
Sx
HandT
XH
are nuclear iffI
e" k=O
and o(X,H) are nuclear iff there exists
Uw = «~:I,
e ))
00 0 'n n= W EX.
Then
and
Here A denotes the diagonal operator with matrix
k,l E IN u {OJ •
So, instead of the spaces SX,H and ,(x,H), we can study the spaces Si
2,A and ,(i2,A) as well. For instance, in Si A there exists a natural identification
2'
between continuous linear mappings on this space and infinite matrices. In the sequel we make no distinction between a linear operator in
i
2 and its corresponding matrix. Let A be a diagonal 00 -A t and whereI
e k k=OV
kk > 0 and with matrix with A kk = Ak, k E IN u {OJ, where 0 < AO ~ Al ~ •••for positive constants c and a. Then we have
(1.5) Lemma
Let v > O. Then
Sf
AV 2' ProofFor each t > 0, we define the subspaces R](t) and R
2(t) as follows:
{(f
J.) E f2
I
jElNu{O}sup(ILl
J exp(A~ t»J < oo}{(f
J.)
E f
2
I
jElNu{O}sup(ILl
J exp«V .. )JJ via t» < co}.The following relations are not hard to prove:
U R] (t) t>O
Let g E R1(t). Then there exists K] > 0 such that
Ig·1 v t) Ig·1 v -v K] ~ exp(A. exp«cA.) t c ) ~ J J J J Ig.1 I v / a - v ~ exp (z(V •• ) t c ) J JJ
for all J > jo with jo E IN so large that
(D .• )via :s: JJ v 2(cA.) J I -v I -v Hence g E R 2(ztc ), and R1(t) c R2(ztc ). Conversely, let h E R
I I
h . exp«
V. . )vIa t):<::;J JJ
Moreover, there exists JO E 1N such that for all J > jo
V•. ~! CA~
JJ J
So for J > jo we obtain
I I
h . exp« .. )
V vIa t ) ~I I
h . exp(A.v(2Ic)V t ) .J JJ J J
v v
Thus we find that h E R]«!c) t) and R
2(t) c R] «!c) t). Now our assertion
is proved by taking intersections or unions.
o
A. = j. Observe that for all v,t > 0,
J
in
l2'
We assume that the entriesK
mnIn the remaining part of this chapter we consider the diagonal matrix Awith
00 •V t
I
e-J < 00, It implies that thej=O
and the entireness space
,(l2,A
v) are both nuclear. analyticity spaceSl
A
V2'
Let K
=
(K ) be a linear operator mnsatisfy the following conditions:
(I .6) K = 0 for m > n
mn
3
c
>0 Vm,n: IKmm - Knnl > C3D>0 3y>O V 0< V : IKmnl :<::; DnY ,
m, -m<n n
It is clear that the numbers K , n E 1N u {a}, are eigenvalues of Kwith
nn
eigenvectors u(n), say,
u~n)
= 0 for j > n. If we take u(n)=
I, then theJ n
(n)
u , m
=
O,l, ••.,n, have to satisfym
(K - K )u(n) + K u(n) + ... + K u(n) 0
mm nn m m,m+1 m+] mn n
for m O,I, •.. ,n-l. Next, we define the linear operator $ ($ ) by mn
S mn
if m > n
if 0 :::; m :::; n •
Then algebraically we have the relation KS =
SM,
whereM
denotes the diagon-al matrixMk
= ~k' k E IN u {a}. We note thatM
is injective on finite sequences.In the remaining part of this chapter we take ~ > I fixed. We want to prove that S is a continuous bijection on Sf A~ and on T(f2,A~). To this end,
2' first note that
S mn For t > 0 we put n-m
K
-K
L\'
Km,m+k Sm+,nk • mm nn k=l tThen for 0 the following recurrent relation is valid mn
t n-m t
o K K
I
exp (-t«m+k)f.I-m~))K +k 0mn mm - nn k=l m,m m+k,n
We take a fixed Nt E IN so large that for all n > Nt
%
ny+1 exp(-(p-l)t nP-1) < 1 where we set p(1.7) Lemma
min(~,2). It leads to the following result.
~-l
Proof
t
Let n > Nt' Since ann I, the inequality is valid for m = n. Hent~, with 'backward induction' it follows that
exp(tm~-I(n-m» lot
I
~ mn • exp(tm~-I (n-m».exp(-t(m+k)~-I (n-m-k»} $ n-m $f
nYI
exp(-nt«m+kl~-l
- mll-I)) $ k=l n-m~
%
nYI
exp (-ntk (p-1)(m+k) p-2) $ k=I n-m $%
nYI
exp(- (p-I)t n P-Ik) k=I $f
ny+I exp(-(p-I)t nP-I) < I • I f we put Lt := maxla~nl,
O$n$Nt O$m$Nt Then Lemma (1.7) gives(1.8) Lemma
n
exp ( - (nll-mll)t)
I
SI
~ Ct • mn
The same techniques as used above apply in the proof of the following lemma: (1.9) Lemma Let V= (V ) satisfy mn V mn and, also, if m > n if m = n Vt>O 3
B >0: sup
I
exp (- t (nil-mil))VmnI
< Btt n,mE'lN u{O}
Then V ~s ~nvert~• ·ble andV-1 satis' f '~es
(V-I)
mn
o
if m > n ,and
Proof
Vt>O VE>O: sup
(I
(V-l)mnl
exp(-nll(t+E) + milt)) < 00 •n,m
Let (W ) denote the inverse matrix of (V ) (which exists algebraically!).
mn mn Then we have W mm W mn W mn Now we put
o
i f m > n nI
Vmt Wn9, if 0 ~ m < n • 9,=m+1 t w mn and t ermnWe want to estimate wt • To this end note first that mn t w mn O:;:;m<n. Hence by assumption
IW~nl
where we may as well suppose that B t > 1. We assert that
Iw
tI :;:;
2n-m+1(Bt)n-m, 0 :;:; m :;:; n. To show this we use back-mn ward induction: 1 < 2 n 00
I
B~~+m+1 2-~+m:;:;I
~=m+1 k=l (1.10) CorollaryLet
V
be as in Lemma (l.9). Then-k 2 1 •
n
v
V • t>O r::>0' and < 00Proof
Let t > 0 and let E > O. Let N,M E IN. Then we estimate as follows:
(
M~N
2\4
\m
;=0
IVmnexp(-(t+E)n~
+tm~)
I }
~
,
So, by Lemma (1.9), Also, we have M N 1(
~ I 2\2 1..I
(V- )mn exp(- (t+E)n]l + tm~)I )
:;;
m,n=O<; sup (I(V-I)
exp(-(t+E)n~
+(t+48)m~)
• mnn,mElN
Hence, IIIetA~ V-I e-(t+E)A~111 < co
If we apply the previous results to the diagonalizing operator S, we get
(1.11) Theorem
-I
The linear operator
S
and the inverseS
satisfyand
o
y V •
and any t >
a.
From the characterization of continuous linear mappings on analyticity
spaces and the entireness spaces (cf. [6c] and [2]) it follows that
(1.12) Corollary
I. The operator
S
is a continuous bijection on S with continuous f.2,All-I
inverse
S .
II. The operator
S
is a continuous bijection on T(l2,All ) with continuous inverseS .
-1Now we assume in addition that the matrix entries
K
n E IN u {a} of the nn'matrix (K ) introduced in (1.6), satisfy mn K nn a = cn (1 + 0(1)) -1
In the first part of this section the Hilbert space (f. 2)S
for some positive constant c. If we define the diagonal operator Mby
Mk = Kkk' k E" IN u fa}, then Sf. All = Sf. MlJ/a by Lemma (1.5). Moreover,
2' 2'
the operator
S
is continuous on Sf. Mll/a (T(f. 2,M
ll/a)) and has a continuous 2'
inverse. We have the relation
K
=
SMS-
1•has been
intro--1
duced. It follows that K is a positive self-adjoint operator in (f. 2)S which satisfies Kll/a
=
SMll/aS- 1• It leads to the following classificationresult.
(1.13) Theorem
II.
Proof
2. THE CLASSIFICATION OF ANALYTICITY SPACES AND ENTIRENESS SPACES GENERATED BY THE JACOBI POLYNOMIALS
The two classification methods for analyticity and entireness spaces dis-cussed in the previous chapter, lead to the classification of the spaces
S and
X Q,(A Q)V
ex,I-' ex,I-'
the Hilbert space
v
T(X Q,(A Q) ) with ex,S> ex,I-' ex,I-'
L
z([-1,1],(1-x)ex(1+x}S dx}
-1 and v ~ ~. Here X Q denotes
ex,I-'
and
A
Q the positiveself-ex,1-' adjoint operator
2 dZ d d
- (l-x ) dxZ + (ex+S+Z)x dx - (S-ex) dx •
It is well known that the Jacobi polynomials p(ex,S). n E IN u {O}, are the
n
eigenfunctions of A Q with eigenvalues n(n+ex+S+1}. (For the definition of
ex,I-' (ex S)
P , we refer to [8], p. Z08.}
n
Instead of p(ex,S) we rather consider the normalized Jacobi polynomials
n
1
{zn + ex + S + 1 f(n+1)f(n+a.+S+1}}2 p(a.,S)Zex+S+1 f(n+ex+1)f(n+S+l) n
which constitute an orthonormal basis in X S. ex,
The normalized Chebyshev polynomials T
R-!'-!
establish a special classn n
of Jacobi polynomials. They satisfy
(Z. I) T (cos 8)
n
w
cos n8 , n ~ I , 8 E [O,7T] ,Yr,
2 -!
Lz([-l.l].(l-x) dx) and, also, that
A_I
_1 =2, 2
d Z
x dx • The eigenvalues of
A
I I are the numbers n ,-2,-2 We note that X I I - 2 '-2 Z dZ (I-x) - + dxZ n E IN u {a}.
If we consider the transformation x forms into
cos e, then the operator A Q
trans-a, ...
C a,S
1 d d
(S-a) sin e de - (a+S+I) cot e
d8
So, with the aid of (2.1) it follows that the matrix of A S with respect to
a,
the orthonormal basis (T )00
a
satisfiesn n=
o
if n > m (2.2) (Aa,S T ,T )x = n(n+a+S+l) n n I -~ -2, I(Aa,S T ,T )x ~ DnH
a
~ 'rn < n n m I I -2,-2for some positive constant D. So we can apply the theory of the previous chapter. It yields an operator S with the property
a,S
A
=
S A (S )-1a,S a,S a,S a,S
where
A
S denotes the diagonal operatorA
T = n(n+a+S+l)T •a, a,S n n
Moreover, for all ~ > 1 the operator Sa,S is a continuous bijection on the as well as on the entireness space
-1
inverse
S
u.a, ...
We note that we have proved in chapter 1 the following sharper result:
(2.3) < 00
< 00
We find that
(2.4) p(a,S)
n Tn + ( 5 )a,S n-],n Tn-] + ••• + (5a,1JQ)1,n T] + ($a,iJ0)0,n TO
1S an eigenfunction of Aa,S. In order to arrive at the spaces
Sx
(A )v/2/2
a,S' a,Sand T(X Q,(A Q)V ) we have to normalize the polynomials p(a,S) with
a,1J a,1J n
respect to the norm of X Q. To put it differently, we have to compute the
a,1J constants d(a,S) which satisfy
n
p(a,S)
n
d(a,S)R(a,S)
n n
Following [8], p. 210. the following relations are valid for all n E IN:
T (x) n and n-] 1(2 n n-] 2
V;
x + ••• x + ••• So we obtain T (x) n Therefore 2n-] 2 n-] x + ••• • p(a,S) n with T + ( $ ) T + ••• n a,S n-],n n-] 2n-] 2 22n-] IWith the aid of Lemma (a.II) and the result (a.13.ii) of the Appendix it follows that there are constants q Q and p Q > 0 such that
a,fJ a,fJ -I q Q(n+l) a,fJ and
Id
(0.,S)I
~
p S (n+ I) • n 0.,So the diagonal operator V
S
with entries dCa,S) and its inverse V-I0., n a,S
satisfy the following
(2.5) V
E>O·
IIIV
a , S exp (-E:(A- 2 ' - 2I ,)]J./2)III < 00VE>O·
II'V-
IS exp (- E:(A I I ) ]J. /2)III
< 000., -2,-2
for each ]J. > I.
So if we put
W
S
V
Q' thenW
Q is a continuous bijection both ona,S a,S a,fJ a,fJ
-I
S
A
and on T(X I"A
I I ) with continuous inverse W Q.X 1 I I I -2,-2 -2'-2 a,fJ
- 2 ' - 2 ' -2·-2
Further we obtain from (2.3) and (2.5):
(2.6) Lemma
with 11 > I.
Since W T = R(a,S) n E IN u {OJ, and since
A
a,S n n ' a,S
obtGin from Theorem (1.3) and Theorem 1.4)
W A W-I we
(Z.7) Theorem
Let ~ > 1. Then for all a,S> -I
and
s
X , (A )~/Z a,S a,S- s
/
- Z _1 ( Z dZ d)~
2L
Z( [-1 , 1] , (l-x ) 2 dx), - (l-x )-Z
+x dx dx .If we apply the transformation x = cos 8, we get
(Z.8) Corollary
Let ~ > I. Then for all a,S> -I
and
s
y (C )~/Z a,S' a,S ~/Z ,(y o,(C 0) ) a,I-' a,I-'s
( dZ)f1/Z
L Z([O,rr],d8), - - Z d8 '(
(
dZ)~/Z)
, LZ([O,rr] ,d8), - - Z . de .Here Ya,S denotes the Hilbert space Lz([O,TI],(l - cos 8)a(1 + sin 8)S de).
The classification method, based on the use of continuous bijections is not refined enough to obtain the classification of the spaces
S
I andX 0' (A 0)2
I a,I-' a,I-'
,(X o,(A 0)2). For this case we use the other classification method, based a,I-' a,I-'
on perturbations. With Theorems (1.1) and (l.Z) we will obtain the results of Theorem Z.7 for ~
=
1. We note that the spacesT
x
(A
)4
and!
a,S' a,Bo(xa,1-'o,(Aa,B) ) will be related to the spaces of (ultra-) hyperfunctions in the Chapters 6 and 7.
In the Appendix we have snown that the matrix elements of the operators
d
(2.9) n3 / 2(n_k) ! (k+I)2 if k ~ n ifO~k<n
where G Q is some positive constant;
a,I-'
a
if k > n (2.10)I
«XV)R(a,S) R.(a,S))I
~ n n '-1c a.,S if k=
n (n)3/2 H Q 1 a,1-' (k+I)2 ifa
~ k < nfor some positive constant H Q.
a,I-'
For each a,S> -I we put
A
Q=
A Q + 1, and furthera,I-' a,I-'
pY,O {(o-y) - (S-a)}V - {(o+y) - (S+a)}(xV) • a,S
Then we have the relation
A
A
+pY,o
y,o a,S a,S
With the aid of the estimates (2.9) and (2.10) we prove the following auxiliary results.
(2. II) Lemma
Let a,S> -I and let
!
~ v ~ I. Then there exists a positive constant e Qa,1-' such that for all r >
a
~ v ~ -I
II exp (r (A Q) )V(A Q)
a,I-' a,I-'
~ v
exp(-r.{A Q}}II Q
a.,I-' a,I-'
-3/2
~ e r .
a,S Proof
o
i f k :f:. t+n, t E 1N u {O}v v
• exp -r(«nH)(nH+IX+S+))+ I) - (R, (t+IX+8+ 1)+ I) )
First we shall prove that W(IX,S)(r) is a bounded linear operator on
X
Q forn IX,~
all r > O. Therefore, note first that
«n+R,)(n+R,+M8+1 )+1)v - (t(R,+a+8+1 )+1)v
v-I = v[«n+t) (n+i+IX+8+1)+I) - (i(t+IX+8+1)+I)]~(n,R,)
for some number ~(n,t) E [R,(t+a+8+1)+I,(n+i)(n+i+a+8+1)+I]. We get the following estimation
v-I ~ vn(n+i+a+S+I) «n+i) (n+i+a+S+I)+I) ~ vn In addition, for all n E IN we have by the estimate (2.9)
{
(nH) 3/2 n
sup G
iEINU{O} IX,S (R,+I)!
I
(n+ t) (n+ R,+a+S+ I)
(IX S) 1 -vrn
So the norm of the operator
W '
(r) is smaller than G Q n2 e forn n,~
each n E ill. Hence the series
I
IIW(n,S) (r)1I converges. Thus we obtain then=1 n following estimation
00
<; G I n ! exp (-vrn)
a,S n=I
for some positive constant e Q.
a,I-'
-3/2 <; e r
a,S
o
Compairing the bounds for the matrix elements of the operator xV and of the operatorV,
the reader will see that the same method as used in the proof of Lemma (2.11) applies in the proof of the following result.(2.12) Lemma
Let a,S> -I and let! <; v <; I. Then there exists a positive constant f Q
a,I-' such that for all r > 0
-I
:<; f r + I .
a,S
Remark. From the preceding lemmas it follows that
V
and xV are well-defined continuous linear mappings on the analyticity spacesS
~A v and onX Q' ( Q)
~ V a,I-' a,I-'
the entireness spaces ,(X Q,(A Q) ) with! <; v <; I and a,S> -I.
a,1-' a,I-'
Remark. The Lemmas (2.11) and (2.12) also hold for v > I. However, for simplicity in the estimation we have given the proof in case of ! :<; v :<; only.
We want to apply Theorem (1.1) and Theorem (1.2) to the perturbations
Py,o.
a,STherefore we need the following lemma.
(2.13) Lemma
Let
a,S,y,o
> -1 and let v > O. Then for every t > 0 the linear operator~ v
exp(-t(Aa,Q) ) maps X .Q continuously into X ~.
Proof
By [8], p. 217, there exist q > 0 and c > 0 such that
max
-I$;x$;1
So with the aid of (a.13.i) we derive
Now let t E IN be so large that t > q+3j2. Then for each f E X Q
a,I-'
1
J
~ - t 2 y 0
I«Aa,S) f)(x) 1 (I-x) (I+x) dx$;
-I 1 $;
J
(l-x)[y]+I(I+x)[o]+I!«A )-t f)(x)\2(I_x)-I+Y-[Y](I+x)-I+o-[o] dx $; a,S -I 1$; 2[y]+[o]+2 max (1«A
a S)-t f)(x)1 2)
f
(l_x)-I+y-[y](I+x)-I+o-[6] dx $; -1$;x$;1 '-I
$; h max
I
«A Q)t f)(x)12 y,o -1$;x$;1 a,I-'for some constant h > 0 which does not depend on f E X Q.
y,o a,I-'
Consider the following estimation
max -1$;x$;1 I «A Q)-t f) (x) 1 2 a,I-' max -1$;x$;1
I
I
(f R(a,S)) ( 1 )t R(a,s)(X)12 $; , n a,S n(n+a+S+I) + 1 n n=O$;
(~
I (f Ra,S) If I )t max IRn(a,S) (x) 1)2 $; n__LO ' n a,S \n(n+a+S+I) + 1 -)$;x$;1:0; Ilf
II~,
S 00 ( 2!LI
I I ) ( I +n) 2q+ I n (n+a+ s+I) + n=O := k IIfl12 Qy,o a.,I-'
Thus we have found that there exists a constant oy,o > 0 independent on the a.,S
choice of f such that
:0; cry,
°
Ilfll Q 'a,S a.,I-'
~ -£
SO (A Q) maps X Q continuously into X
a,I-' a,I-' y,o
~ 'J
Finally we note that exp(-t(A Q) ) can be written as
a,I-'
~ 'J exp(-t(A Q»
a.,I-'
Since the operator between {.} maps X Q continuously into itself the proof
a.,I-' is complete.
(2.14) Theorem
Let a.,S,y,o > -I and let ~ :0; 'J :0; I. Then
n
and, also
Proof
Following Lemma (I.S) it is sufficient to prove that
Sx
(A )'J a,S' a,S=
Sx
(A )'J and that T(X S,(A S)'J) = T(X o,(A o)'J)· By the LemmasY,o' y,o a, a, y, y,
(2.1), (2.12) and (2.13) the following assertions are valid:
The perturbation Py,o =«o-y) - (S-a»D - «o+y) - (S+a.)xD satisfies a,S
D(PY'o) :>
Sx
~ v andA
A
+ pY,o is a positiveself-a,S a S,(Aa
,
,
(3) y,o a, 13 a,Sadjoint operator in X 0'
Y,
~ v
X Scontinuously in X
For all t > 0, exp(-t(A (3)) maps Y,o'
a, a,
Then by Theorem (1.1),
Sx
(A
)va,S' a,S
~ v ~ v
T(Xa,S,(Aa,s) ) C T(Xy,o,(Ay,o) ). If we interchange the roles of (a,S) tained.
Application of the transformation x
(2.15) Corollary
C
Sx
(A
)V, and by Theorem (1.2), y,o' y,oand (y,o), the wanted result is
ob-o
cos 6 yields
Let a,S,y,o > -I and let 4 ~ v ~ I. Then
and
v T(Y o,(C ~))
y, y,u
For a = 13 = -4, we have seen that
R~4,-4(cos
6) =~
and R: 4,-4(cos 6) ==
W
cos n6, n E: IN. So it becomes rather easy to characterize the spacesS . and
Y, 1,(C I .)V
-2,-2 -2,-2
Theorem (2.7) and Theorem (2.14) yield the other values of a,S> -1.
same analytic description for the
Further, by means of the conformal mapping w
=
cos z, we get thecharacteri-zation of the spaces
S
andXa,S,(Aa,S)V be carried out in the next chapter.
T(X (3,(A S)v). This program will
3. THE CHARACTERIZATION OF THE ANALYTICITY SPACES AND ENTIRENESS
SPACES BASED ON THE JACOBI POLYNOMIALS, WITH APPLICATIONS TO CLASSICAL ANALYSIS
In this chapter we intend to derive characterizations in classical analytic
terms of the elements in the analyticity spaces Sy (C )v and in the a,S' a,S
entireness spaces T(Y S,(C S)v) where we consider a,S> -I and v ~
!.
a, a,
Following the results obtained in the previous chapter we only have to
consider one pair Y
Q,C
.0. It is rather natural therefore to study thea,I-' a,I-'
case a = 13
=
-2,I ~.e.,• the so-called Chebyshev case.For convenience we put Y= Y_!,_! = LZ([O,TI]) and
C
=C_!,_!
The normalized eigenfunctions c of the operatorC
given byn c (8) n
'fli
cos n8 ,\ I
.!.
V
'TT ' 8 E [O,TI] , n E 1N,establish an orthonormal basis ~n Y. We want to determine the analytic behaviour of the series
00 Z t+
L
n=O a c (z) n n 00for all coefficients (an)n=O satisfying the following order estimate
O(exp(-tnZv) , n E 1N u {a}, v ~
!
Since cos nz
z it is necessary and sufficient to characterize
the analytic behaviour of the type of series
00 n=-oo b n ~nz e where b 11
O
Zvconsists of all even functions in S
L
Z( [-'IT, 'IT]) ,~v
entireness space T(Y,CV) consists of all even functions in T(L
Z([-'lT,n]),6 v). 1 ina
We observe that the functions e : 8 t+ - - e ,
e
E [-'IT, 'IT], nEll, establishn
If.IT
an orthonormal basis for the Hilbert space L2([-'lT,'IT]). They are the
eigen-2
functions of the positive self-adjoint operator
~
= -~
in LZ([-'lT,n])Z
de
with eigenvalues n , n E~. It is clear that the analyticity space Sy
,
c
v and, similarly, that theThe program of this chapter is the following. In a separate section we characterize
These spaces
the elements of
SLz([-'lT,'IT]),~V
and T(L 2 ([-n,n]),6v), v~ ~.
correspond to classes of Z'IT-periodic, analytic functions which mostly satisfy certain growth conditions. In a rather natural way we define norms in these classes of analytic functions. This leads to an alternative description of sequential convergence in the corresponding analyticity spaces and the corresponding entireness spaces.
As we have seen all results carryover to the spaces
Sy,c
v and T(Y,CV), v ~ ~. We thus obtain vector spaces of even, 2n-periodic, analytic functions with their natural norms.We next employ the conformal mapping w = cos z to the function classes v
associated to Sy
,
CV
and T(Y,C ). The thus obtained classes of analytic functions lead to a characterization of the spaces Sx (A )V andv a.,S' a.,S
T(X Q,(A Q» for v ~ ~ and a.,S > -1. Moreover we derive a classical
a.,1J a.,1J
description of sequential convergence in these spaces. As a result we are able to connect the growth behaviour of analytic (entire) functions with the growth of their expansion coefficients with respect to bases of Jacobi
polynomials (R(a.,S».
3.1. The characterization of SL [ ] v and T(L
2([-n,n]),6v)
2 ( -n, n ),6
First we discuss the case v
(3. 1• I) Lemma
~. To this end, consider the following lemma:
Let (a)oo E
~
Z. Then for the expression n n=-oo 00 f(z) -L
n=-oo 1nz a e n Proof 00 I ~) f(x+iy)I
an n=-oo {x+iy Iyl ::; T < t}. For ythe following assertions are valid.
I. f is an analytic function on an open strip 11m zl < t, t > 0, and it
tends to boundary functions x ~ f(x±it) in L
2([-n,n])-sense iff (an et In
I
):=-00 E 12'II. f is an analytic function on an open strip 11m zl < to' to > 0, iff for
t1nl)00 f)
all t with 0 < t < to the sequence (an e n=-oo E ~2 (or equivalently,
l ).00
e-ny einx converges un1formly on a. 1 1 'str1ps
t t (or y ~ -t) the sequence (an e-ny):=_oo tends
• f) h (a e-nt)oo (or (a ent)oo ).
1n ~2-sense to t e sequence n n=-oo n n=-oo
I ~) Both (an ent)oon=-oo and (an e-nt)oon=-oo have to be l2-sequences. Hence (an e1n1t):=_00 is an 12-sequence.
l~) Similar to I ~).
For all y E (-t,t) we have
II) ~) (a n open. E 12 and hence n
f
/f(x+iy)12 dx < 00. Therefore -n(a e 1n1Y) E l since the interval (-t,t) is
n 00
We introduce the following notation: St denotes the symmetric open strip of width Zt around the real axis, i.e.
St
= {
Z E II:I
I
1m ZI
< t} .(3.].2) Definition
Let t > O. Then the class A(St,27f-per) consists of 27f-periodic functions for which there exists £ > 0 such that f is analytic on St+£.
The class A(~,27f-per) consists of all 27f-periodic entire analytic functions.
(3.1.3) Theorem
Let t > O. Then f E L2([-7f,7f]} can be extended to f E A(St,27f-per) iff
!
f E exp(-t~2)(SL2([_7f,7f]),~~). Proof
Immediately from Lemma 3.1.1.
We note that
o
!
U exp (- T~2) (L 2 ( [ -7f , 7f]) ) T>t
We obtain the following characterization:
(3.14) Theorem
1. iff there exists t > 0 such that f E A(St,27f-per).
Next we shall show that the 'functional analytic' topologies of the spaces
I
SL2([-7f,7f]),~~ and T(L2([-7f,7f]),~2) are equivalent to topologies which are related to the analytic properties of the functions in these spaces.
denote the Hilbert space t6!
t6~
(e f,e g)L where ( , )L • 2(D
2 the norm 1n L2,
t([-rr,rr]) product (f,g)t 1 = , 2 L2([-rr,rr]). We denote denotes the inner product ofTo this end, let
L~ ~ ~
([-rr, rr]), t > 0,I '
-tl:l2
e (L2([-rr,rr])) with inner
by II'lit 1
, 2
In the space A(St,2rr-per) we define the following norm
sup If(z)
I ,
11m z
I
~tNote that with the usual identifications
U A(St,2rr-per) t>O and
n
A(St,2rr-per) t>O as sets. ( I ) Now let f E L 2 2t([-rr,rr]) for some fixed t > O. From the characterization 1n
,
Lemma (1.3.1) it follows that f E A(S ,2rr-per) for all T, 0 < T < t. Let T, T
with 0 < T < t be fixed, and consider the following estimation
p (f) T sup 11m ZI~T 00
I
L
n=-oo (f,e )e (z) I ~ n n e21nlt (f,e ) nThus we obtain with a
=
f!rr
-rr (I - e-2(t-T) _1) 2 the inequality t,Tp (f) ~ at IIf lit I
Conversely, let f c A(St,2n-per)
(1)
have f E L22 ([-n,n]). Write f =
, t
for some t > O. Then by Lemma (1.3.1) we
00 a e . Then a is given by n n n n=-oo a n n = _1
f
ffiT
-n f(x)e-inx dx , n E: 7l •We have the following relations
a n n
=
_1 Jf/fiT
-n f(x)e-inx dx f (xt~t. )e-~n x_~t. ( +' )dx tnt e -n 1TJ
f(xtit)e-inx dx • So for all n E INThus we obtain for all T,
a
< T < t,(**) IIfll I T,2
= 2n at P (f) •
,T t
The inequalities (*) and (**) yield the following result.
(3. 1.5) Theorem
1.
1
-t112
Let t > O. Le; (fn)nEIN be a sequence in e (SL
2([-n,n]),I1!)' The sequence (etI12f n )nEIN is a null sequence in SL
2([-n,1T]),I1! iff there exists £ >
a
such that pt+£(fn) + 0, i.e., fn(z) +a
uniformly onI
II. The sequence (g) 'IN is a null sequence in T(L2([-n,n]),~2) iff for all nnE.
t > 0, pt(gn) ~ 0, i.e., gn(z) ~
°
uniformly on each strip 11m zl ~ t.Remark. The sequence (f) IN is a null sequence in
S
! iffn nE L2([-n,n]),~2
(fn ) c A(St,2n-per) for some t >
°
and pt(fn) ~ 0, i.e., fn(z) ~ 0 uniformly on 11mzl
~ t.In the remaining part of this section we investigate the case v >
!.
The next two lemmas are of main importance.(3. 1 •6) Lemma
00 Z
Let (a )n n=-oo E ~ . Then for the expression
f (z) 00 n=-oo inz a e n z = x+iy E (C ,
the following assertions are valid:
I. Let v ~ I. Then f is a 2n-periodic analytic function of growth behaviour
-n iff n
f
If(x+iy) 1 2 dx 2v~
A2 exp (2Bly!2V-I) , A ~ 0 , B >°
for t ( 1) 2V-I (2v)-2 v\2v ; II. Let v >!.
If (a n 2v 00 exp(tInl»
n=-oothen f is a 2n-periodic function of growth behaviour (*). Moreover, if f satisfies the growth condition (*), then the sequence (an):=-oo satisfies
Proof
I .) Consider the following estimation
2v A2
~
exp (- 2BIy ,2v-1 ) 'ITr
If(x+iy) 1 2 dx J -rr 2v = _1 exp (- 2BIy12v-] \r
?'1T ) n=-OO 2v ] ooI ({ -2v-1 -2V(2V - 1)2V-I 2V} 2 1 1 - 2ny - 2(2v) B i nI •=
2TI'
n=-oo exp - B YIf we now take
( )
2V-]
2v - 1 2v-1
Yk = - 2vB. sgn(k)lkl , k Ell,
then in the k-th term the factor between curly brackets equals I. Hence
(
) 2V-I for all k E ll. This proves that for t = (2v) -2v\ 2v ; 1
2v 00
(a exp(tlnl» belongs to f .
n n=-oo 00
the sequence
I ~) The analyticity of f is trivial. We estimate as follows:
-'IT 'IT
f
1f(x+iy) 12 dx = _Ir
2'IT
I
~ -21f lIan
The proof is complete if we can show that the series between {.} is uniformly bounded for all y E JR. Then just take
I
(
2V)-2V\2V-I
B (2v-l) J
t .'
In order to show this uniform boundedness, consider the function 2v x ~ exp(-2tlxl - 2xy) , X E JR, at the 2v 2t (2v-1 )(lxl)2V-I exp 2vt
y. This function attains its maX1mum
I
I
y 12v-1- 2vtl sign(y). Left to this point the function increases, point x =
for fixed
right to this point it decreases. Then we get the estimate
00 2v
I
e-2tlnl -2ny n=-oo 2v~
exp(2t(2V_I)(lxl)2V~I)
+ 2 \ \2vt 00f
2v -2tlxl +2xy e dx . - 0 0Finally, standard asymptotic techniques yield
00 n=-oo 2v -2tlnl -2ny e I - v 2v
~
C(I + lyl)2v-1 exp (2t(2V_I)fJ.xl)2v-l\
\2vt /
So for v ~ I the series {.} is uniformly bounded in y.
II. In order to prove the first part of the statement we estimate as follows: