search

MatthewRHermes / mrh

MRH's research code
Other
19 stars 30 forks source link

CI Vector Indeces Permuted in LASSI CI Vectors #70

Closed dking072 closed 9 months ago

dking072 commented 9 months ago

With many fragments/rootspaces, for the civecs_lassi returned with the code below:

lsi = lassi.LASSI(las) energies_lassi, civecs_lassi = lsi.kernel()

the ordering of the rows of this matrix do not correspond to the ordering of the las states in las.e_states

MatthewRHermes commented 9 months ago

Since users or callers may potentially input model states in any arbitrary order, lassi first implicitly permutes the basis functions to block-diagonalize the Hamiltonian. Said permutation involves calling python set on lists of quantum numbers which will tend to result in different Hilbert spaces getting addressed in a more-or-less arbitrary order. This wouldn't be a problem if I correctly inverted the permutation before returning, but here I goofed and just used the permutation index vector itself instead of its inverse. Anyway hopefully 8e76434 fixes it.

MatthewRHermes commented 9 months ago

OP is, to my knowledge, the only person who has ever actually attempted to use LASSI with a model space spanning more than one whole-molecule Hilbert space, which is quite a strange feature and I don't remember why I thought it was important. So this bug shouldn't have affected anyone else.

dking072 commented 9 months ago

Pulling dev fixes this, thanks!