Open dkweiss31 opened 4 years ago
As far as I know, there are no all-purpose methods that can accomplish this. As noted by the Netlib team, the user should take extra precautions to use these methods. I would be interested though if there are new possibilities in the literature.
I am interested in the generalized eigenvalue problem
Hx=\lambda Mx
where H is symmetric positive definite (and complex in general) but M is symmetric indefinite, i.e., it has negative eigenvalues. As far as I can tell, no linear combination\alpha H +\Beta M
is positive definite.I can solve for the eigenvalues using
scipy.linalg.ordqz(H, M, sort=sorter)
with an appropriatesorter
function, but this has the disadvantage of returning all the eigenvalues. For large applications, this will not suffice, as I am only interested in the smallest eigenvalues.I understand that there exists a symmetric indefinite Lanczos method, which is more in line with what I am interested in. I am wondering if this algorithm has been implemented somewhere I can interface with it, or if this is something new that could be added to scipy.