ON A ZONAL POLYNOMIAL INTEGRAL

Transcription

1 ON A ZONAL POLYNOMIAL INTEGRAL A. K. GUPTA AND D. G. KABE Received September 00 and in revised form 9 July 003 A certain multiple integral occurring in the studies of Beherens-Fisher multivariate problem has been evaluated by Mathai et al. 995 in terms of invariant polynomials. However, this paper explicitly evaluates the context integral in terms of zonal polynomials, thus establishing a relationship between zonal polynomial integrals and invariant polynomial integrals.. Introduction Consider two p p independent random matrices S and S.The matrix S, with noncentrality parameter matrix Ω and covariance matrix Σ, is Wishart with n degrees of freedom and S is Wishart with covariance matrix Σ and n degrees of freedom. Then the random matrix F = S / S S / has the density gf=k F /n p S /n +n p { exp trσ S } trσ S/ FS / 0 F n ; 4 ΩΣ / S / FS / ds = k θ δ,λ φ C φλ δ,λ:φ k!l! n / λ C φi C δ,λ φ Σ S, Σ ΩΣ / S ds,. Copyright c 003 Hindawi Publishing Corporation Journal of Applied Mathematics 003: Mathematics Subject Classification: 6H0, 6H URL:

2 570 On a zonal polynomial integral where Λ is the p p diagonal matrix of the roots of F and K as a generic letter denotes the normalizing constants of density functions in this paper. Mathai et al. [4, equations 5.5., 5.5., pages 74 75] do not evaluate the integral.. However, they derive the density of the roots of F in terms of a series of invariant polynomials of two matrix arguments. This paper explicitly obtains the density of F in terms of zonal polynomials, a result which has not been available in the literature so far, not even in terms of invariant polynomials. An expression for the density of the roots of F is also derived in terms of invariant polynomials of two arguments, which is slightly different from the one presented by Mathai et al. [4, equation 5.5., page 74]. The next section states some useful results and the main results of the paper are derived in Section 3. Although it is not always possible to derive invariant polynomial results in terms of zonal polynomials, sometimes it is as done in this paper. We assume that all matrices occurring in this paper are full-rank matrices, except the noncentrality parameter Ω which may have any rank.. Some useful results If P is any p p matrix, then it is known that see Gupta and Nagar [] K exp{trpv} I VV /n p dv = 0 F n; 4 P P.. Now we can write P =P P / H for some p p orthogonal H, see G. A. Korn and T. M. Korn [3, page 4]. It follows that K exp{trqpv } I VV /n p dv = K exp { trqp P / HV } I VV /n p dv = K exp { trp P / Q V } I VV /n p dv = 0 F n; 4 P PQ Q = 0 F n; 4 Q QP P = K exp{trpqv} I VV /n p dv..

3 A. K. Gupta and D. G. Kabe 57 Again it follows from. that K exp { trqp P / HV } I VV /n p dv = K exp { tr H P P / Q } I VV /n p dv = K exp { tr H P P / Q Q / R } I VV /n p dv = 0 F n; 4 R Q QRH P PH = 0 F n;q QP P,.3 where Rp p and Hp p are any two arbitrary orthogonal matrices which may depend on Q Q and P P, respectively. Now it follows from.3 that 0F n; 4 P PQ QT T = 0 F n; 4 H P PHR Q QRU T TU,.4 where H, R, and U are any p p orthogonal matrices, and note that 0F n; 4 ABC = 0 F n; 4,.5 where is any permutation of ABC and also holds for any permutation of four symmetric matrices. Obviously,.4 in terms of zonal polynomials yields From.6, note that C θ Q QP PT T=C θ R Q QRH P PHU T TU..6 exp{trp PQ QR R} = 0 F exp{tr } = t=0 t! C θp PQ QR R,.7 where is any permutation of P PQ QR R. Further, it is easy to verify that 0F n; 4 P PQ QT TG G = 0 F n; 4 P PQ QG GT T.8 because.8 is true for any permutation of the four distinct matrices in.8. Now we proceed with the main result.

4 57 On a zonal polynomial integral 3. Main results In view of.4,.5, and.7 and using., the density of F is given by gf=k F /n p { exp trσ S } trσ / FΣ / S S /n +n p 0 F n ; 4 Σ / ΩΣ / F / SF ds / = K F /n p Σ +Σ / FΣ / /n +n F n + n, n ; F/ Σ / ΩΣ / F / Σ +Σ / FΣ / = K F /n p Σ /n +n F n + n, n ; +Σ / FΣ / Σ / ΩΣ / / F Σ / ΩΣ / / Σ +Σ / FΣ /. 3. Now if Σ / Σ Σ/ = θi for any scalar θ, then note that gf=k F /n p θi + F /n +n F n + n, n ; ΩFθI + F. 3. Thus the density of Λ=diagλ,...,λ p of the roots of F is gλ = K Λ /n p θi +Λ /n +n i<j F n + n, n ; ΩΛθI +Λ. 3.3 λi λ j The expression is not very suitable for obtaining the density of Λ for the expression 3.. ForsomeQ p p orthogonal, let QFQ be still denoted by F,letLbe the p p diagonal matrix of the roots of Σ / ΩΣ /, and let be the diagonal matrix of the roots of Σ / Σ Σ/. Then the

5 A. K. Gupta and D. G. Kabe 573 density of this F is gf=k F /n p +F /n +n F n + n, n ; LF + F. 3.4 Mathai et al. [4, equation A.3.5, page 338] showed that +F a F a,u;lf + F = δ,λ;φ a φ C δ,λ φ F, LF. 3.5 kk!l!u λ We have observed earlier that C θ Q QP P=C θ R Q QRH P PH and thus the invariant polynomial argument for matrices can also be replaced by their roots. Thus the density of Λ for 3.4 is gλ = K Λ /n p i<j λi λ j δ,λ;φ / n + n φ Cδ,λ φ Λ,LΛ k!l!. /n λ 3.6 However, 3.6 is not suitable for obtaining the density of trf. Acknowledgment The authors are thankful to the referee for a critical reading of the paper. References [] Y. Chikuse, Distributions of some matrix variates and latent roots in multivariate Behrens-Fisher discriminant analysis, Ann. Statist. 9 98, no., [] A. K. Gupta and D. K. Nagar, Matrix Variate Distributions, Chapman & Hall/CRC Monographs and Surveys in Pure and Applied Mathematics, vol. 04, Chapman & Hall/CRC, Boca Raton, Fla, USA, 000. [3] G. A. Korn and T. M. Korn, Mathematical Handbook for Scientists and Engineers, McGraw-Hill, New York, 96. [4] A. M. Mathai, S. B. Provost, and T. Hayakawa, Bilinear Forms and Zonal Polynomials, Lecture Notes in Statistics, vol. 0, Springer-Verlag, New York, 995. A. K. Gupta: Department of Mathematics and Statistics, Bowling Green State University, Bowling Green, OH , USA address: gupta@bgnet.bgsu.edu D. G. Kabe: Department of Mathematics and Computer Science, Saint Mary s University, Halefax, Nova Scotia, Canada B3H 3C3