Maximal extension for linear spaces of matrics with large rank

For every 0 < k < min{m,n} and any linear subspace E of real m × n matrices whose non-zero elements have rank greater than k, we show that there is a maximal extension Emax satisfying the same rank condition, and that the dimension of Emax is not less than (m – k)(n – k). We apply this result to the study of quasiconvex functions defined on the complement Exs22A5 of E in the form F(X) = f(PExs22A5(X)), where PExs22A5 is the orthgonal projection to Exs22A5.