On the possible ranks among matrices with a given pattern

Abstract

Given are tight upper and lower bounds for the minimum rank among all matrices with a prescribed zero-nonzero pattern. The upper bound is based upon solving for a matrix with a given null space and, with optimal choices, produces the correct minimum rank. It leads to simple, but often accurate, bounds based upon overt statistics of the pattern. The lower bound is also conceptually simple. Often, the lower and an upper bound coincide, but examples are given in which they do not.