A Gaussian integral identity

This integral pops up all the time in Bayesian data analysis:

\int{N(\mathbf{a}|\mathbf{Xb},\mathbf{\Sigma}_2)N(\mathbf{b}|\mathbf{c},\mathbf{\Sigma}_1) d\mathbf{b}}

I got tired of rederiving the solution, so I am posting it here for easy reference:

= \left((2\pi)^{d}|\mathbf{\Sigma}_1||\mathbf{\Sigma}_2||\mathbf{H}|\right)^{-1/2}\exp(-\frac{1}{2} E)

Where d is the dimensionality of \mathbf{a}, |\cdot| is the matrix determinant and:

E = \mathbf{a'\Sigma_2^{-1}a} + \mathbf{c'\Sigma_1^{-1}c} - \mathbf{x'Hx}

\mathbf{x} = \mathbf{H}^{-1}(\mathbf{X'\Sigma_2^{-1}a + \Sigma_1^{-1}c})

\mathbf{H} = \mathbf{X'\Sigma_2^{-1}X+\Sigma_1^{-1}}

Leave a comment

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s