Nonnegative ranks, decompositions, and factorizations of nonnegative matrices (Q686375)

Let \(A\) be an \(m \times n\) nonnegative matrix with elements in an ordered field. The smallest nonnegative integer \(q\) for which there exist \(q\) nonnegative column vectors such that each column of \(A\) can be expressed as a nonnegative linear combination of those vectors is called the nonnegative column rank, \(c-\text{rank}_ +A\). The nonnegative row rank, \(r-\text{rank}_ +A\), is defined in a similar manner. Two other nonnegative ranks are defined for \(A\) and all four are proved to be equal. The authors show that in calculating these rank it is sufficient to calculate the appropriate nonnegative rank of a bivariate probability matrix or a stochastic matrix associated with \(A\). They prove that if the (ordinary) rank, rank \(A\), does not exceed 2 then \(\text{rank} A= \text{rank}_ +A\). Finally, they show that the nonnegative rank can be computed exactly over the reals by a finite algorithm. They pose the problem of determining whether or not the nonnegative ranks of a rational matrix, over the reals and rationals, respectively, are equal.