Consider a large (in the region of 500 by 500 up to 2000 by 2000) real, square symmetric matrix.
What would be a good way of algorithmically determining if it has any pair of eigenvals with opposite signs? I want to avoid finding the whole spectrum if at all possible.