A well-known theorem in probability states: Let x be an Nx1 Gaussian vector with mean vector m and covariance matrix Cx .If we linearly transform x as y=Px using a square NxN transform matrix P, then y is also Gaussian with mean vector Pm and covariance matrix Cy=PCxPT.

This theorem has an elegant proof using the multivariate Gaussian probability density function (pdf).

It is also known that the above theorem is also valid for non-square (NxM) transform matrices, P. 

Does anyone know any reference book containing the proof for the more general case of non-square linear transformations?

Similar questions and discussions