• Aucun résultat trouvé

THE ASYMPTOTICS OF SPHERICAL FUNCTIONS AND THE CENTRAL LIMIT THEOREM

N/A
N/A
Protected

Academic year: 2022

Partager "THE ASYMPTOTICS OF SPHERICAL FUNCTIONS AND THE CENTRAL LIMIT THEOREM"

Copied!
12
0
0

Texte intégral

(1)

A NNALES DE L ’ INSTITUT F OURIER

G ENKAI Z HANG

The asymptotics of spherical functions and the central limit theorem on symmetric cones

Annales de l’institut Fourier, tome 45, no2 (1995), p. 565-575

<http://www.numdam.org/item?id=AIF_1995__45_2_565_0>

© Annales de l’institut Fourier, 1995, tous droits réservés.

L’accès aux archives de la revue « Annales de l’institut Fourier » (http://annalif.ujf-grenoble.fr/) implique l’accord avec les conditions gé- nérales d’utilisation (http://www.numdam.org/legal.php). Toute utilisa- tion commerciale ou impression systématique est constitutive d’une in- fraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.

Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques

http://www.numdam.org/

(2)

45, 2 (1995), 565-575

THE ASYMPTOTICS OF SPHERICAL FUNCTIONS AND THE CENTRAL LIMIT THEOREM

ON SYMMETRIC CONES

by Genkai ZHANG

0. Introduction.

In multivariate statistical analysis the spaces of positive definite matrices (real or complex) are of considerable interests. One studies the spherical polynomials, central limit theorem on them. See [M]. A natural generalization of those spaces is the cone of positive elements in a formally real Jordan algebra. This class of cones includes cones of positive definite real, complex and quaternion matrices and the forward light cone. In this paper we will prove a kind of central limit theorem on any such symmetric cones.

In his paper [T2] Terras obtained a central limit theorem for rotation- invariant random variables on the space of n x n real positive definite (symmetric) matrices in the case n = 3. This has been recently generalized to any n by Richards [R]. In this paper we will generalize this central limit theorem to the symmetric cone of positive elements in a Jordan algebra.

The main tools used in [R] are the Binet-Cauchy theorem and certain coefficients in some recurrence formulas of zonal (spherical) polynomials in [Ku]. Recently we have found some recurrence formulas on any simple and formally real Jordan algebra [Z]. To prove our central limit theorem we will

Key words'. Spherical functions — Central limit theorem — Jordan algebra — Symmetric spaces - Heat equation - Spherical polynomial.

Math. classification: 43A65 - 62E15.

(3)

use those recurrence formulas and a recent result of Faraut and Koranyi [FK] on expansion of reproducing kernels.

Acknowledgment. — The author would like to thank Jonathan Arazy for comments on the paper and Bent 0rsted for constant encour- agement.. He also thanks Donald. St. P. Richards for drawing his attention to the work of P. Graczyk ([Gl] and [G2]). He acknowledges the financial support from Odense University, Denmark, where this work was done.

1. Preliminaries.

In this section we recall some facts about the symmetric cone in a Jordan algebra. Most results in this section can be found, e.g. in [UU] and [KS] (on formally real Jordan algebras) and [Tl] (in the special case of real symmetric matrices).

Let X be a simple and formally real Jordan algebra [BK] with product x o y and unit element e. Basic examples are the space X = 7<y(SC) of all self-adjoint r x r-matrices over K = R,C or El (quaternions) with the anti-commutator product x o y = (xy + y x ) / 2 , and the "spin factor"

X = R1"^ of all vectors x = (a;o,^i,... ,^'n) = (^05 ^Q? with product x o y = (xovo + x ' y ' , XQV' + y^x'). The open set

P = {x2^ C X invertible}

is a convex cone. For the matrix algebras, P is the cone of positive definite matrices, whereas for the spin factor P is the forward light cone. The linear automorphism group

GL(P)={geGL(X):g(P)=P}

is transitive on P. Let G be the identity component of GL(P). G has Lie algebra

g = {M e gW : exp(tM) € GL(P), for all t e R}

with the commutator bracket [M, N] = MN — NM. The Lie algebra g has a Cartan decomposition g = k4- p, where kis the Lie algebra of the

"orthogonal" group

K == 0(P) = {g € G : ge == e}

and p = {MX : x C X} consists of all multiplication operators M^y=xoy

(4)

on X. Moreover P = G/K is now a Riemannian symmetric space.

Let e i , . . . , Or be a maximal orthogonal system of idempotents in X.

Then we have

e = ei + ... +67..

Let

(1.1) X= ^ ^

l^<j<r

be the Peirce decomposition induced by { e i , . . . , 67.}. Here X^ = X,, = [x C X : e, ox = ^^

are the joint eigenspaces of the multiplication operators {Me.Y^i on Jsr- The integer r is independent of the choice of the maximal orthogonal system of idempotents and is called the rank of X.

Let

a = RMei + ... 4- MMe,

be the subspace of p generated by M^,..., M^. Then a is maximal abelian in p. So r is the rank of the Riemannian symmetric space P. We have thus a root space decomposition of (g, a) [HI]. We let {7j}^i be the dual basis in a* of{MeJ^i, i.e.

7j(^eJ=^fc.

Then the corresponding root system consists of 7J ~ 7fc, j ^ k with common multiplicity a. The Weyl group is the symmetric group 67.. For example in the case of matrix algebra 7<r(K) the rank is r and the multiplicity is a = 1,2 or 4 according to K = R, C, or HI respectively.

We fix an ordering of a* by letting

7l < 72 < • • • < 7r.

/"y . __ '"VL. ^"

Therefore, the positive roots are 3 , j > k. Let p = ^ p^ be, as

2 ~ .7=1

usual, the half sum of positive roots

q ^ . _ ^ g _ . ^1

^ 2 ^ 2 2 ^ V 2 yj>fc j=i 7^'

We will now describe the spherical functions and the heat equation on the symmetric space P. We need the notion of Jordan algebra determinants.

(5)

Let P(x) := 2(Ma;)2—Ma;2. The determinant of it is up to a constant a unique (integer) power of an irreducible polynomial A (a;) of degree r on X, normalized so that A(e) = 1. A (a?) is called the determinant of the Jordan algebra X. It can also be defined as the unique irreducible polynomial given by the "Cramer's rule":

_i cofactorofrc , „. . . . . x = ——————, [x € Ainvertible)

A(a;) normalized so that A(e) = 1.

Let Xij be given in the Peirce decomposition (1.1) we put X, = ^ X^e

M<J

and let Pj be the orthogonal projection of X on Xj; we define A,(aO=Ax,(P,aO

where ^x- is the corresponding determinant in Xj.

For any t = ( t i , . . . , tr) € C7" we define

(1.2) A^^n^w

J=l

r

Let A = S ^j7j ^ ^5 ^^e complexification of a*, we let s = j=i

(^i, S 2 , . . . , 5y.) be defined by

(1.3) Sj = \j 4- pj - Xj-^-i - pj+i,

where we put Sr = Ay. + p r ' The spherical function h\ is then h),(x)= { ^{kx)dk= ( \\^(kx)dk.

J K ^j^i

Then h\(x) are J^-invariant eigenfunctions of all the G-invariant dif- ferential operators on P. The Helgason-Fourier transform of a J^-invariant function / on P is defined by

fW = / f{x)h^x)d-x^

Jr [x)ti\(x)d x, iv

where d* is the G-invariant measure on P.

For later use we also introduce the spherical polynomials, which is the analytic continuation of h\ (in A). Let m = ( m i , . . . , rrir) be a r-tuple of integers with m\ > m^ > ... > mr > 0. The functions

<M^)= / n^"^ 1 ^)^

^3=1

(6)

where m^-n = 0, are the spherical polynomials on P. In particular if we put

i^E^.

k=l

the functions

<M^)= { Aij(Jb;)dfc= I ^j(kx)dk JK ~ JK

are called the fundamental spherical polynomials on P.

Finally we recall the heat equation on P. Let L be the Laplace- Beltrami operator on P. We normalize the J^-invariant norm on p so that norm M^ has norm 1. Then M e i , . . . , M^ are orthonormal basis of a. This also gives a normalization of the Riemannian metric on the cone P = G / K .

Then (f)\ are eigenfunctions of L:

^'(i^-^)^.

j'=i

The heat equation is defined

f Qtu{x, t) = Lu(x, t), (x, t) C P x R+

\u{x^)=f{x).

2. The main theorem and the proof.

Before stating our main theorem we introduce some notation. Let

r

H = ^ hjMej € a. We let J(expH) be the function defined on A by j=i

j(exp^) = —n^ 1 -e^Q-n^ 01 ^^.

2 V 2 / 2>J 1=1

Our main theorem is the following central limit theorem.

THEOREM 1. — Let Ym, m >_ 1 be a sequence of independent, identical distributed, K-invariant random variables in P, with common density function f(Y). Assume that f(Y) satisfies

( hif(exp H)J{exp H)dh^ • • • dhr = 0

JQL

(7)

/ hihjf(exp H)J(exp H)dhi' • • dhr = 0, i -^ j Jo.

and the convolution Sm = YI o Y^ o • - - o Ym is normalized as in (3.39) of [Tl] and that Sm has density function f^. Then for any measurable subset S c P

/^(y)d*(y)^exp^^x t G_^^F^,(Y)d\Y)

JS 9D(2 + ar) JS 2(2+ar) 2+ar

as m —> oo.

In the above, Gf is the fundamental solution of the heat equation on P with Helgason-Fourier transform

s-.m^(^-^)).

\ i=l /

Further Fi has Helgason-Fourier transform

I r \

Ft(\)=exp t( ^ AzA,) .

\ l<fc<j^r /

Following [Tl] and [R] we need to find the expansion of the spherical functions h\(expH) on a around H ^= 0. Recall that spherical functions h\(expH) are Weyl group invariant in A and H G a. The expansion is

r

(recalling that H = ^ hjMej) j=i

r r

h^(expH) =1 + (an ^A, + 012) (^^z)

i=l %=1

r r r

+ (a2i^A? + 022 ^ A,A^ + a23^A, + 024} (i^)

z==l l^i<j<r 1=1 i=l

4- -P(Ai,..., Ay.) ( Y^ ^z/^j) + higher terms;

Ki<j<r

namely we need to find the six coefficients an, 0:12, Q;2i, 0^221 <^23) o'24- We fix X 6 Q^ and 5 as in (1.3). Consider the formula (1.2) in the polar coordinates

r

(2.1) A^(fca.e)=I'[A^(te.e) j'=i

and put a = expH. Following [R] we take logarithm differentiation to obtain

1 f) r 1 Q

(2.2) ————-——As(fca-e) =Y^5,————-——^j(ka'e).

v / A^(fca • e) Qh\ — ' L^ Aj (ka - e) Qh\ J l

(8)

Integrating over K and putting H = 0 we get -h\{a'e)

(2.3) ^

=E 5

^/M

^ll( ae )

<9/ii j^=o

j'=i Jf=o

and

We adopt the notation [R] to denote

^(fc)=^-A,(fca.e)

/;W=^A,(fca.e)

f(/) = / /(^

JK

H=0

H=0

for a function / on K.

We need a result of Faraut and Koranyi [FK].

LEMMA 1. — For a = expH we have

(2.4) f[(l - te^) = ^(-1^ (^Ma. e)t^

j=i j=o VJ/

Proof. — This is a special case of Theorem 3.8 in [FK]. See also Remark 2 after the theorem. We give here a simple independent proof. We know that the polynomial A(e — z) is a polynomial of degree r on X and is K invariant. Thus we have an expansion

r

A(e-^)=^c,^(^', j=o

for some constants cj. Now take z = a ' e the above equation reads

n(i-^)=^c^(^'.

J=l j=0

Evaluating the equation at H = 0 and noting that <^(e) = 1 we find the coefficients Cj. D Lemma 1 has also been proved previously by J. Arazy and Z. Yan [A].

Now we differentiate the expansion in Lemma 1 and set H = 0,

-t(i-.r-,S(-i).(;:;)^(..

^ij(a-e) t3'.

H=0

(9)

however the LHS of the above is

-((i-o-^a- 1 )^;:;)^'

j=o v / thus we find

f(/0=——<Ma-e) = j / r .

<^1 — H=0

Using this we see

Oill = -, Oi2 =0.

V

Note that if we differentiate (2.4) twice we get

W)=|^<Mo-e)

H=0

= j / r .

Now the coefficients of 2^ h2 is 1=1

a

2 (2.5)

We have

Q ( 5 i , . . . , s ^ ) = -^h^a-e) H=O

Q(.i,...,^)=<p((^y^)')+^(^(/;)-^(^^^

j'=i j=i

We need to find

W^ W)

2

)-

We start to calculate £{f[f'j)- It follows from §1 that Ai^+H = AiAj.

Differentiating at H = 0 and integrating we find

(2.6) -i-^^ • e) = ^(^0 + <?(/;) + 2^ (/{')<?(/;).

^"'i ^=o

To find out the LHS we need the a result proved in [Z] (see also [St]).

LEMMA 2. — We have the recurrence formula of the product 0ii<^ij of the spherical polynomials

^i = ^T^)^^ + ^r^T)^

for the product (^11 (^u.

(10)

The lemma can also be written as

^l±lrL A. A.

r

- ^ .

rdT^)^^ = ^^ - rO-TtJ)01^' Thus

^^ t -'„„ =(+jJ)(2j+ ^: ) |, (r - J)u+)

Now we substitute this to (2.6) to get

In particular

^'OT

W)'^-

Thus the coefficient of AiA2 m (2.5) is 2a22=2(f((/^)-£((,i 2a22=2(f((/^)-£((/02))

2a 2r + ar2 and the coefficient of A^ is

2a2i=£((/{)2) _ 2 + Q

~~ 2r + or2' The coefficient of Ai in (2.5) is

2023 = £(f[) - £ ((/Q2) - a^f(/^) + ^(r - l)f(/{)

J=2 z

=0.

On the other hand, we have Q(si,... ,Sr) = 0 for s = 0 or for

^ = -|(^ - 7-2—)' that is r

0 = Q'21 ^ A^ + 0:22 ^ A^A^ 4- 0:24 j=l Ki<j<r

which gives us

a(r2 - 1)

Q'24 = -

48(2 + ar)'

(11)

Finally we get the required expansion

^ ( e x p ^ ^ + ^ ^ A ^ ^ ^ ) 1=1 %=i

(

2+0 ^""^ , 2 CL ^--\

+ 2(27^^or2)^ ^ +2r^^ar2 2^ A^

v / %=1 l<z<7<r

_

a48(24-ar)A2^

(

r2

z

l

L\(\-h

2

}

+ P ( A i , . . . , Ay.) ( Y^ ft^j) + higher terms.

<i<j<r

The rest of the proof is similar to that in [Tl].

BIBLIOGRAPHY

[A] J. ARAZY, Personal communication.

[BK] H. BRAUN and M. KOECHER, Jordan-Algebren, Springer, Berlin, Heidelberg, New York, 1966.

[FK] J. FARAUT and A. KORANYI, Function spaces and reproducing kernels on bounded symmetric domains, J. Func. Anal., 89 (1990), 64-89.

[Gl] P. GRACZYK, A central limit theorem on the space of positive definite symmetric matrices, Ann. Inst. Fourier, 42-1,2 (1992), 857-874.

[G2] P. GRACZYK, Dispersions and a central limit theorem, preprint.

[HI] S. HELGASON, Differential Geometry, Lie Groups and Symmetric Spaces, Academic Press, London, 1978.

[H2] S. HELGASON, Groups and geometric analysis, Academic Press, London, 1984.

[KS] B. KOSTANT and S. SAHI, The Capelli Identity, Tube Domains, and the Generalized Laplace Transform, Adv. Math., 87 (1991), 71-92.

[Ku] H. B. KUSHNER, The linearization of the product of two zonal polynomials, SIAM J. Math. Anal., 19 (1988), 686-717.

[L] 0. LOOS, Bounded Symmetric Domains and Jordan Pairs, University of California, Irvine, 1977.

[M] R. J. MURIHEAD, Aspects of multivariate statistical theory, John Wiley & Sons, New-York, 1982.

[R] D. St. P. RICHARDS, The central limit theorem on spaces of positive definite matrices, J. Multivariate Anal., 23 (1987), 13-36.

[St] R. P. STANLEY, Some Combinatorial Properties of Jack Symmetric Functions, Adv.

Math., 77 (1989), 76-115.

[Tl] A. TERRAS, Harmonic analysis on symmetric spaces and applications, II, Springer- Verlag, New-York, 1985.

(12)

[T2] A. TERRAS, Asymptotics of special functions and the central limit theorem on the space 'Pn of positive n x n matrices, J. Multivariate Anal., 23 (1987) 13—36.

[UU] A. UNTERBERGER and H. UPMEIER, Berezin transform and invariant differential operators, preprint.

[Ul] H. UPMEIER, Jordan algebras in analysis, operator theory, and quantum mechanics, Regional conference series in mathematics, n° 67, Amer. Math. Soc., 1987.

[U2] H. UPMEIER, Toeplitz operators on bounded symmetric domains, Tran. Amer.

Math. Soc., 280 (1983), 221-237.

[Vr] L. VRETARE, Elementary spherical functions on symmetric spaces, Math. Scand., 39 (1976), 343-358.

[Z] G. ZHANG, Some recurrence formula for spherical polynomials on tube domains, Trans. Amer. Math. Soc., to appear.

Manuscrit recu Ie 18 octobre 1994.

Genkai ZHANG, School of Mathematics University of New South Wales Kensington, NSW 2033 (Australie).

genkai@solution.maths.unsw.edu.au

Références

Documents relatifs

Abstract: We established the rate of convergence in the central limit theorem for stopped sums of a class of martingale difference sequences.. Sur la vitesse de convergence dans le

Key words and phrases: Central limit theorem, spatial processes, m- dependent random fields, weak mixing..

The quantities g a (i, j), which appear in Theorem 1, involve both the state of the system at time n and the first return time to a fixed state a. We can further analyze them and get

The proof relies on the correct aymptotics of the variance of the eigenvalue counting function of GUE matrices due to Gustavsson, and its extension to large families of Wigner

Various techniques have been developed toward this goal: moment method for polynomial functions ϕ (see [2]), Stieltj`es transform for analytical functions ϕ (see [4]) and

We established the rate of convergence in the central limit theorem for stopped sums of a class of martingale difference sequences..  2004

In this paper we prove a central limit theorem of Lindeberg-Feller type on the space Pn- It generalizes a theorem obtained by Faraut ([!]) for n = 2. To state and prove this theorem

Since every separable Banach space is isometric to a linear subspace of a space C(S), our central limit theorem gives as a corollary a general central limit theorem for separable