Matrix variate beta distribution explained

In statistics, the matrix variate beta distribution is a generalization of the beta distribution. If

U

is a

p x p

positive definite matrix with a matrix variate beta distribution, and

a,b>(p-1)/2

are real parameters, we write

U\simBp\left(a,b\right)

(sometimes
I\left(a,b\right)
B
p
). The probability density function for

U

is:
-1
\left\{\beta
p\left(a,b\right)\right\}

\det\left(U\right)a-(p+1)/2

b-(p+1)/2
\det\left(I
p-U\right)

.

Here

\betap\left(a,b\right)

is the multivariate beta function:
\beta
p\left(a,b\right)=\Gammap\left(a\right)\Gammap\left(b\right)
\Gammap\left(a+b\right)

where

\Gammap\left(a\right)

is the multivariate gamma function given by

\Gammap\left(a\right)=\pip(p-1)/4

p\Gamma\left(a-(i-1)/2\right).
\prod
i=1

Theorems

Distribution of matrix inverse

If

U\simBp(a,b)

then the density of

X=U-1

is given by
1
\betap\left(a,b\right)

\det(X)-(a+b)

b-(p+1)/2
\det\left(X-I
p\right)

provided that

X>Ip

and

a,b>(p-1)/2

.

Orthogonal transform

If

U\simBp(a,b)

and

H

is a constant

p x p

orthogonal matrix, then

HUHT\simB(a,b).

Also, if

H

is a random orthogonal

p x p

matrix which is independent of

U

, then

HUHT\simBp(a,b)

, distributed independently of

H

.

If

A

is any constant

q x p

,

q\leqp

matrix of rank

q

, then

AUAT

has a generalized matrix variate beta distribution, specifically

AUAT\sim

T,0\right)
GB
q\left(a,b;AA
.

Partitioned matrix results

If

U\simBp\left(a,b\right)

and we partition

U

as

U=\begin{bmatrix} U11&U12\\ U21&U22\end{bmatrix}

where

U11

is

p1 x p1

and

U22

is

p2 x p2

, then defining the Schur complement

U22 ⋅

as

U22-U21{U11

}^U_ gives the following results:

U11

is independent of

U22 ⋅

U11\sim

B
p1

\left(a,b\right)

U22 ⋅ \sim

B
p2

\left(a-p1/2,b\right)

U21\midU11,U22 ⋅

has an inverted matrix variate t distribution, specifically

U21\midU11,U22 ⋅ \sim

IT
p2,p1
\left(2b-p+1,0,I
p2

-U22 ⋅ ,U11

(I
p1

-U11)\right).

Wishart results

Mitra proves the following theorem which illustrates a useful property of the matrix variate beta distribution. Suppose

S1,S2

are independent Wishart

p x p

matrices

S1\simWp(n1,\Sigma),S2\simWp(n2,\Sigma)

. Assume that

\Sigma

is positive definite and that

n1+n2\geqp

. If

U=S-1/2

-1/2
S
1\left(S

\right)T,

where

S=S1+S2

, then

U

has a matrix variate beta distribution

Bp(n1/2,n2/2)

. In particular,

U

is independent of

\Sigma

.

See also

References