On rank-2 Nonnegative Matrix Factorizations and their variants

We consider the problem of finding the best nonnegative rank-2 approximation of an arbitrary nonnegative matrix. We first revisit the theory, including an explicit parametrization of all possible nonnegative factorizations of a nonnegative matrix of rank 2. Based on this result, we construct a cheaply computable (albeit suboptimal) nonnegative rank-2 approximation for an arbitrary nonnegative matrix input. This can then be used as a starting point for the Alternating Nonnegative Least Squares method to find a nearest approximate nonnegative rank-2 factorization of the input; heuristically, our newly proposed initial value results in both improved computational complexity and enhanced output quality. We provide extensive numerical experiments to support these claims. Motivated by graph-theoretical applications, we also study some variants of the problem, including matrices with symmetry constraints.