Deprecate determinant addresses as input to fermionic solver functions

[caleb-johnson]

solve_fermion and optimize_orbitals should take a bitstring_matrix as input to define the subspace onto which to project/diagonalize the Hamiltonian, same as solve_qubit. Users should not have to convert to determinant addresses before using those functions, as this is an unnecessary step and can be error prone, especially if not using the helper functions.

[caleb-johnson]

@garrison would you mind glancing at the changes in the fermion module and tell me whether you agree that this is a clean way to deprecate (it was your idea ;) ).

I believe there is a breaking change here for users who previously specified all their positional args as kwargs, as we now require the first arg to be positional. I suppose I could update the release note to reflect this

[kevinsung]

I'm not sure I agree with this change. The PySCF SCI solver takes strings as input, so it seems more user-friendly for solve_fermion, which has a similar functionality, to take the same kind of input. If I'm not mistaken, the motivation for this change is that postselect_and_subsample returns a bitstring matrix instead of strings. Can we instead modify that function to return strings?

[caleb-johnson]

I'm not sure I agree with this change. The PySCF SCI solver takes strings as input, so it seems more user-friendly for solve_fermion, which has a similar functionality, to take the same kind of input. If I'm not mistaken, the motivation for this change is that postselect_and_subsample returns a bitstring matrix instead of strings. Can we instead modify that function to return strings?

The motivation is that solve_qubit takes the bitstring_matrix, and we want this technique to extend beyond chemistry. So by making this change, the solvers can have a more similar interface. I find creation of addresses, and the possibility for errors at that step, to be unnecessary in the workflow

[caleb-johnson]

If I'm not mistaken, the motivation for this change is that postselect_and_subsample returns a bitstring matrix instead of strings. Can we instead modify that function to return strings?

Post-selection and subsampling is a part of the more general technique, so we don't want to have any chemistry lingo in there. I think that interface should stay as-is. I'd really just like to remove an unnecessary step in the workflow that doesn't really add any understanding or information (addresses and bitstring_matrix contain the same info).

[kevinsung]

I see, that makes sense. I'm fine with this change then.