Consider implementing implicit solve_sylvester using least squares
If the Hamiltonian is dense but the interesting subspace is small compared to its size, using MUMPS is inefficient for solve_sylvester
, whereas computing full diagonalization is unnecessary. Instead we may use least squares solver to apply a Green's function at an eigenenergy, see e.g. scipy.linalg.lstsq
. That function doesn't seem to store the decomposition, and we'd need that, but otherwise it's
With a dense problem this only brings benefits in very few cases though.
Edited by Anton Akhmerov