Thanks Steven. I figured it was something like that. I was hoping that since the non-symmetry was just numerical error I would still get the eigenvalues sorted in the same order as the symmetric case. It does seem unsatisfying that such a small numerical error can cause such different behavior in the function. I think for my code I'll need to make sure I symmetrize all the inputs (since I really do want a consistent order).
Cheers, Ethan
