Setup qcel.models.Molecule with non-contiguous fragments about qcelemental HOT 5 OPEN

There's a PR up @awvwgk , just in case you didn't get pinged.

from qcelemental.

Ah yes, in the beginning it was requested that fragments be allowed to be noncontiguous. So the good news is that I wrote the hard part (molparse validator) to accommodate. Then we ran into there being zero downstream programs that could actually work with noncontig. I can probably expose throw_reorder to the pydantic model, but results should be checked carefully at first. So you've got a downstream that can use noncontig? I can see how dispersion on fragments would be a good case, esp. as geometry is fixed.

from qcelemental.

Thanks for the context, yes I tried testing non-contiguous fragments for a workflow which can work with those robustly, but directly hit a hard-stop. Now I work around by handling the fragment information outside of the qcel models because qcel is still more convenient than rolling my own.

from qcelemental.

Ok, I looked a bit more, and it's the molparse internal representation (aka psi's) that works in fragment separators (requires contiguous) rather than fragment lists. Thus, molparse always contiguizes, and throw_reorder merely toggles between let-it-reorder and stop-in-tracks.

So, an allow_noncontiguous_fragments_but_reorder=True is not too hard (just lifts the throw), but it's not keeping with good qcschema practice (changing up the input for convenience), and it's probably not what your workflow wants, is it?. A proper handling of noncontig would need more extensive renovation.

Is this xtb that can work with non-contig or a workflow w/o a program in particular? For now, your workaround is probably best.

from qcelemental.

I could live with qcel reordering the atoms, just leaving them in their order would be preferred, of course.

Generally, the reordering seems like something psi4 should apply as needed, similar as we handle ghost atoms in DFT-D4, qcel / qcng could provide a convenience function for permuting atoms (together with all associated properties).

from qcelemental.