multidimensional Gaussian integral
Let and .
Theorem 1
Let be a symmetric positive definite
(http://planetmath.org/PositiveDefinite) matrix and, where.Then
(1) |
where .
Proof. is real and symmetric (since . For convenience, let . We can decompose into , where is an orthonormal () matrix of the eigenvectors of and is a diagonal matrix
of the eigenvalues
of . Then
(2) |
Because is orthonormal, we have . Now define a new vector variable , and substitute:
(3) | ||||
(4) |
where is the determinant of the Jacobian matrix . In this case, and thus .
Since is diagonal, the integral may be separated into the product of independent
Gaussian distributions, each of which we can integrate separately using the well-known formula
(6) |
Carrying out this program, we get
(7) | ||||
(8) | ||||
(9) | ||||
(10) |
Now, we have , so this becomes
(12) |
Substituting back in for , we get
(13) |
as promised.