 Rank normal form

Every matrix of rank $r$ is equivalent to a block matrix with $I_r$ in its upper-left corner and zeros elsewhere. This follows by choosing bases adapted to the kernel and image of the represented linear map.