I have large sparse square matrix n by n, its rank is slightly below n, let's say m. I want to make it non-singular by removing rows and columns by a certain rule. The rule is that if you remove ith row, you must remove ith column as well, so that the matrix is always square. This is effectively removing a node in an adjacency graph.
My first question is: does there always exist such a combination of n-m rows and columns I can remove such that the remaining m by m submatrix is structurally non singular.
My second questions is: is there an effective algorithm to obtain a p by p non-singular submatrix without removing excessive amount of rows and columns
To provide more context, the matrix I'm dealing with is about 1000 by 1000 with sparsity close to 0.05