[WIP] clearer trexio eri interface

[kgasperich]

This is just one example of a way to address the issues brought up in Issue #103

The changes here are just the first implementation I thought of that adds some clarity/safety, so I'm not strongly attached to the way I've done it here, and someone more familiar with pyscf might know of a cleaner way to handle it, especially regarding the ao2mo part. I'm not familiar with all of the possible options for input/output symmetries, so it might make sense to pass some more args through to that function to allow for various possibilities.

I also haven't changed/added any tests here, but I could do that if there is any interest in merging the PR.

[NastaMauger]

Hello @kgasperich

I think to work effectively, we might want to add a chemist/physicist notation feature, either directly in TREXIO or within the PySCF-Tools. This would help standardize how we handle the ERIs and ensure clarity when dealing with different notation/softwares.

In my PR. after the write_eri function, I added the following:

idx=idx.reshape((num_integrals,4))
for i in range(num_integrals):
    idx[i,1],idx[i,2]=idx[i,2],idx[i,1] 

idx=idx.flatten()

with trexio.File(filename, 'w', back_end=_mode(backend)) as tf: 
    trexio.write_mo_2e_int_eri(tf, 0, num_integrals, idx, eri.ravel())

I believe this change should be incorporated. Let me know your thoughts.

Best

[kgasperich]

Hi @NastaMauger

That change should work to swap between phys/chem notation.
I think with numpy slicing you can accomplish the same thing with idx[:,[1,2]] = idx[:,[2,1]] to avoid the loop, but it's really just a style preference, and it's fine either way.

Also, I think idx.flatten() will return a copy (whereas idx.ravel() would return a view). It's probably not worth worrying about, but maybe there are cases where avoiding the copy would be better.