See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/303993136

A Note on the Existence of the Multivariate Gamma Distribution

Article · June 2016

CITATIONS READS

2 169

1 author:

Thomas Royen
Fachhochschule Bingen
25 PUBLICATIONS   293 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Thomas Royen on 24 June 2016.

The user has requested enhancement of the downloaded file.

 -1-

A Note on the Existence of the Multivariate Gamma Distribution

Thomas Royen
Fachhochschule Bingen, University of Applied Sciences
e-mail: thomas.royen@t-online.de

Abstract. The p - variate gamma distribution in the sense of Krishnamoorthy and Parthasarathy exists for all
positive integer degrees of freedom  and at least for all real values   p  2, p  2. For special structures
of the “associated“ covariance matrix it also exists for all positive  . In this paper a relation between central and
non-central multivariate gamma distributions is shown, which implies the existence of the p - variate gamma
distribution at least for all non-integer  greater than the integer part of ( p  1) / 2 without any additional
assumptions for the associated covariance matrix.

1. Introduction

The p-variate chi-square distribution (or more precisely: “Wishart-chi-square distribution“) with  degrees of
freedom and the “associated“ covariance matrix  (the  p2 ( , ) - distribution) is defined as the joint distribu-
tion of the diagonal elements of a Wp ( , ) - Wishart matrix. Its probability density (pdf) has the Laplace trans-
form (Lt)

| I p  2T | /2 , (1.1)

with the ( p p) - identity matrix I p ,   , T  diag (t1 ,..., t p ),t j  0, and the associated covariance matrix
, which is assumed to be non-singular throughout this paper.

The p - variate gamma distribution in the sense of Krishnamoorthy and Parthasarathy [4] with the associated
covariance matrix  and the “degree of freedom”   2 (the  p ( , ) - distribution) can be defined by the
Lt

| I p  T | (1.2)

of its pdf
g ( x1,..., x p ; , ). (1.3)

For values   2  this distribution differs from the  p2 ( , ) - distribution only by a scale factor 2, but in
this paper we are only interested in positive non-integer values 2 for which | I p  T | is the Lt of a pdf and
not of a function also assuming any negative values. These values  are called here “admissible values”. The
admissibility of all values 2  p  1 follows from the existence of the Wp (2 , ) - distribution, and the ad-
missibility of 2  ( p  2, p  1) follows from formula (1.4) below. Smaller values of  are admissible at least
with some additional assumptions for . Sufficient and necessary conditions for  entailing infinite divisibility
Key words and phrases: multivariate gamma distribution, non-central multivariate gamma distribution, Wishart
distribution
2010 Mathematics Subject Classifications: 60E05, 62E10
 -2-

of the Lt | I p  T |1 (i.e. all   0 are admissible) are found in [1] and [3]. According to [1], | I p  T |1 is
infinitely divisible if and only if there exists any signature matrix S  diag (s1 ,..., s p ),s j  1, for which S 1S
is an M-matrix. For the infinite divisibility of a more general class of multivariate gamma distributions see [2].

At least all values 2  m  1,m  ,m  p, are admissible for “ m - factorial” covariance matrices  
W 2  AAT (  W W  I p  BBT ,B  WA with rows b j ) with a suitable matrix W  diag (w1 ,..., wp ),
wj  0, and a real ( p  m) - matrix A of the lowest possible rank m. This follows from the representation

g ( x1 ,..., x p ; , )  E  p
j 1
w2j g (w2j x j , 12 b j Sb jT  (1.4)

of the  p ( , ) - pdf (see [7] and [8]), where


g ( x, y)  e y  n0 g  n ( x) n!
yn

is the non-central gamma density with the non-centrality parameter y and the central gamma densities g  n ,
and the expectation refers to the Wm (2 , I m ) - Wishart matrix S . With W 2   I p , where  is the lowest
eigenvalue of , it follows m  p  1. A special case – entailing infinite divisibility – is a one-factorial  
W 2  aaT with a real column a.

In the following section it will be shown that all values 2  [( p  1) / 2] (the integer part of ( p  1) / 2 ) are
admissible without any further assumptions for . This result is obtained as a corollary of a relation between
central and non-central multivariate gamma densities given in theorem 1. This relation was already derived in a
similar form for integer values   2 in [6], but the non-integer values require a different proof by means of
the Lt.

2. A Sufficient Condition for the Existence of the  p ( , ) - Distribution

Let be given any partition
 11 12 
  (2.1)
  21  22 

of the non-singular ( p  p) - covariance matrix  with ( pi  pi ) - matrices ii and set

0  11  1222121 , (2.2)

which is a non-singular covariance matrix too.

For 2  max( p1 1, p2 1) the Wishart Wp2 (2 , 22 ) - pdf and the non-central Wp1 (2 , 0 , ) - pdf exist
where the symmetrical positive semi-definite ( p1  p1 ) - matrix  is any “non-centrality matrix” of rank k  p1 .
The diagonal of a random matrix Z has a non-central  p1 ( , 0 , ) - distribution if 2Z has a Wp1 (2 , 0 , ) -
distribution, which exists (apart from integer values 2 with rank()  min( p1, 2 ) ) for all   ( p1  1) / 2
with rank()  p1 (see [5]).
 -3-

The corresponding  p1 ( , 0 , ) - pdf

g ( x1 ,..., x p1 ;  , 0 , ) (2.3)

has the Lt

| I1  0T1 | etr(T1 ( I1  0T1 )1 ) (2.4)

with T1  diag (t1 ,..., t p1 ). (In the literature the “non-centrality matrix” is frequently defined in a different non-
symmetrical way.) Let p2 denote the set of all ( p2  p2 ) - correlation matrices C and let

2Y  2 X 1/2CX 1/2 (2.5)

be a Wp2 (2 , 22 ) - matrix with a random C  p2 and X  diag ( X p1 1 ,..., X p ), where the elements X j
have the gamma-densities  jj1 g ( jj1 x j )   jj xj 1 exp( jj1 x j ) / ( ) with the  jj from the diagonal of
 22 . The density of Y is given by

( p2 ( ))1 | 22 | | Y | ( p2 1)/2 etr(221Y ) (2.6)

with the multivariate gamma function  p2 ( )   p2 ( p2 1)/4  j 21    2 .

p
 j 1

Now, with the notations from (2,1), (2.2), (2.3), (2.5), (2.6), we show for the  p ( , ) - pdf from (1.3):

Theorem 1. If p  p1  p2 and 2  max( p1 1, p2 1), then

g ( x1 ,..., x p ;  , )  ( p2 ( )) 1 |  22 | 

 g ( x1,..., x p1 ; , 0 , 1222

1 1/2 1
X CX 1/222 21 ) | X | 1| C | ( p2 1)/2 etr(22
1 1/2
X CX 1/2 )dC.
p2

Remark. The simple special case with p2  1,C  1 (and  22  1 ) was already given by theorem 2 in [11].

As a corollary we obtain

Theorem 2. The function g ( x1 ,..., x p ; , ) with the Lt | I p  T | from (1.2) is a  p ( , ) - pdf at least
for 2   ([( p  1) / 2], ).

Proof of theorem 2. Choose for p1 or p2 in theorem 1 the value [( p  1) / 2].

Proof of theorem 1. The equation in theorem 1 will be verified by the Lt of both sides. The left side has the Lt
| I p  T | and we get with the Schur complement for determinants

I1  0T1  12221 21T1 12T2

| I p  T | 
21T1 I 2  22T2
 -4-

1
| I 2  22T2 | | I1  0T1  1222 21T1  12T2 ( I 2  22T2 )1 21T1 |

1
| I 2  22T2 |  | I1  0T1  1222 (I 2  22T2 (I 2  22T2 )1 )21T1 |
1
| I 2  22T2 |  | I1  0T1  1222 ( I 2  22T2 )1 21T1 |

1
| I1  0T1 |  | I 2  22T2 |  | I1  1222 ( I 2  22T2 )1 21T1 ( I1  0T1 )1 | . (2.7)

With (2.4), (2.5) and (2.6) we find for the right side the Lt

| I1  0T1 | ( p2 ( )) 1 | 22 |  

 etr(T1 ( I1  0T1 )1 1222

1 1
Y 22 21 ) | Y | ( p2 1)/2 etr(22
1
Y  T2Y )dY 
Y 0

| I1  0T1 | ( p2 ( )) 1 |  22 | 

 | Y | ( p2 1)/2 etr  (22121T1 ( I1  0T1 )1 12221  221  T2 )Y dY 

Y 0

| I1  0T1 | | 21T1 ( I1  0T1 )1 12221  I 2  22T2 | ,

(see e.g. formula (2.2.6) in [12]), and

| I1  0T1 |  | I 2  22T2 |  | I 2  21T1 ( I1  0T1 )1 1222

1
( I2  22T2 )1 |

1
| I1  0T1 |  | I 2  22T2 |  | I1  1222 ( I 2  22T2 )1 21T1 ( I1  0T1 )1 |,

which coincides with formula (2.7), where the last identity follows from the general equation

I1 A12
| I 2  B21 A12 || I1  A12 B21 |.
 B21 I2

Some further remarks. If   W 2  AAT is m - factorial, where A – and consequently B  WA – may con-
tain any mixture of real or pure imaginary columns, then the function with the representation from (1.4) has
again the Lt | I p  T | and it is a  p ( , ) - pdf at least for all values 2  m  1  [( p  1) / 2]. For

special structures of  smaller values of 2 are possible, e.g. with an m - factorial   W 2  AAT with a
real matrix A and m  1  [( p  1) / 2] . Furthermore, let be, e.g., p2  p1 ,0  W02  A0 A0T with a real
( p1  m0 ) - matrix A0 of rank m0 and m12  rank(12 )  p2 . Then, at least 2  max(m0  m12 1, p2 1) is
admissible and max(m0  m12 1, p2 1)  [( p 1) / 2] is possible for low values of m0 and m12 .

On the other hand it is at present an open question if there exist some ( p  p) - covariance matrices  for
which 2 is inadmissible for some values 2  ([( p  3) / 2],[( p 1) / 2]), p  5.

A consequence of theorem 2 is the extension of the inequality

 -5-

G p ( x1,..., x p ; , )  G p1 ( x1,..., x p1 ;  , 11 )G p p1 ( x p1 1,..., x p ;  , 22 ),

x1 ,..., x p  0,rank(12 )  0, (2.8)

for the p - variate cumulative  p ( , ) - distribution function G p . This inequality was proved for 2  and
for all values 2  p  2 in [9] (see also [10]), and it implies the famous Gaussian correlation inequality for
2  1. Now, this inequality can be extended to all non-integer values 2  [( p  1) / 2] without any further
assumptions for .

References

[1] Bapat, R. B., Infinite divisibility of multivariate gamma distributions and M-matrices, Sankhyā Series A 51
(1989), 73-78.
[2] Bernardoff, P., Which multivariate gamma distributions are infinitely divisible?, Bernoulli 12 (2006),169-189.
[3] Griffiths, R. C., Characterization of infinitely divisible multivariate gamma distributions, J. Multivarate Anal.
15 (1984), 13-20.
[4] Krishnamoorthy, A. S. and Parthasarathy, M., A multivariate gamma type distribution, Ann. Math. Stat. 22
(1951), 549-557.
[5] Letac, G. and Massam, H., Existence and non-existence of the non-central Wishart distribution, (2011),
arXiv:1108.2849.
[6] Miller, K. S. and Sackrowitz, H., Relationships between biased and unbiased Rayleigh distributions, SIAM
J. Appl. Math. 15 (1967), 1490-1495.
[7] Royen, T., On some central and non-central multivariate chi-square distributions, Statistica Sinica 5 (1995),
373-397.
[8] Royen, T., Integral representations and approximations for multivariate gamma distributions, Ann. Inst.
Statist. Math. 59 (2007), 499-513.
[9] Royen, T., A simple proof of the Gaussian correlation conjecture extended to some multivariate gamma
distributions, Far East J. Theor. Stat. 48 (2014), 139-145. (preprint: arXiv:1408.1028)
[10] Royen, T., Some probability inequalities for multivariate gamma and normal distributions, Far East J.
Theor. Stat. 51 (2015), 17-36.
[11] Royen, T., Non-central multivariate chi-square and gamma distributions, (2016), arXiv:1604.06906.
[12] Siotani, M., Hayakawa, T. and Fujikoshi, Y., Modern Multivariate Statistical Analysis: A Graduate Course
and Handbook, American Science Press, Inc. (1985), Columbus, Ohio, U.S.A.
View publication stats