"Monte Carlo Estimates of the Log Determinant of ... - Spatial Statistics
"Monte Carlo Estimates of the Log Determinant of ... - Spatial Statistics
"Monte Carlo Estimates of the Log Determinant of ... - Spatial Statistics
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
52 R.P. Barry, R.K. Pace / Linear Algebra and its Applications 289 (1999) 41±54<br />
x 0 Ax ˆ z 2 1 k A;i ‡ ‡ z 2 n k A;i;<br />
where …z 1 ; . . . ; z n † are N…0; I†. Then<br />
x 0 Ax<br />
x 0 x ˆ z2 1 k A;1 ‡ ‡ z 2 n k A;n<br />
z 2 1 ‡ ‡ ;<br />
z2 n<br />
where …z 1 ; . . . ; z n † is N…0; I†. This fraction can be rewritten<br />
z 2 1<br />
P n k A;1 ‡ z2 2<br />
P n<br />
iˆ1 z2 i<br />
iˆ1 z2 i<br />
k A;2 ‡ ‡<br />
z2 n<br />
P n<br />
iˆ1 z2 i<br />
k A;n :<br />
As each z 2 i<br />
is chi-square with one degree <strong>of</strong> freedom, and z 2 i<br />
is independent <strong>of</strong> z 2 j<br />
for i 6ˆ j, <strong>the</strong> coecients W 1 ; . . . ; W n have <strong>the</strong> Dirichlet…1=2† distribution<br />
(Johnson and Kotz, 1972, p. 231).<br />
The mean, variance and covariance <strong>of</strong> n-variate Dirichlet…1=2† random<br />
variables are:<br />
EW i ˆ 1<br />
n ; Var…W …1=2†…n=2 1=2† 2…n 1†<br />
i† ˆ ˆ<br />
…n=2† 2 …n=2 ‡ 1† n 2 …n ‡ 2† ;<br />
2<br />
Cov…W i ; W j † ˆ<br />
n 2 …n ‡ 2†<br />
(Johnson and Kotz, 1972, p. 233). For additional information on <strong>the</strong> Dirichlet<br />
distribution, see Narayanan (1990). (<br />
Result 2. For any real n n matrix A and x N n …0; I†<br />
E x0 Ax<br />
x 0 x ˆ tr…A†=n:<br />
Pro<strong>of</strong>. For A symmetric we can use Result 1 to get<br />
E x0 Ax<br />
x 0 x ˆ EW 1k A;1 ‡ ‡ EW n k A;n ˆ …k A;1 ‡ ‡ k A;n †=n ˆ tr…A†=n<br />
If A is not symmetric, <strong>the</strong>n consider B ˆ …1=2†…A t ‡ A†. Clearly B is symmetric<br />
and has <strong>the</strong> same trace as A. However, for any vector c; c 0 Ac ˆ c 0 Bc. Thus<br />
x 0 Ax=x 0 x must have <strong>the</strong> same distribution as x 0 Bx=x 0 x and <strong>the</strong>refore is an<br />
unbiased estimator <strong>of</strong> tr…A†=n. (<br />
Result 3. For real symmetric A,<br />
Var x0 Ax<br />
x 0 x ˆ 2VAR…k† ;<br />
n ‡ 2<br />
where VAR…k† is <strong>the</strong> population variance <strong>of</strong> <strong>the</strong> eigenvalues <strong>of</strong> A: