Suggestions on how to reconstruct a crystal from pairwise distances and unit cell parameters

[sgbaird]

I'm doing generative modeling via xtal2png which encodes the information below (see image). I'm running into the problem where reconstructed crystals are mostly exhibiting P(1) symmetry sparks-baird/xtal2png#79, which isn't desired. While I'm working on this from the generative modeling side of things (improve representation, find better generative modeling parameters & models), I'm also wondering if there's some post-processing that might be feasible with PyXtal.

Right now, I'm ignoring cell volume, space group, and pairwise distances when doing the reconstruction. I've considered incorporating these (especially space group information sparks-baird/xtal2png#79 and pairwise distances sparks-baird/xtal2png#83) into the reconstruction process.

Note that I'm also doing surrogate DFT relaxation as a post-processing step via m3gnet.

[qzhu2017]

I am not really sure about you request. If you are truly interested in this function, we can set a video meeting to discuss it.

[sgbaird]

@qzhu2017 that would be great actually.

sterling.baird@utah.edu

[sgbaird]

@qzhu2017 thanks for meeting up to chat about this! Looking forward to doing some more brainstorming.

[sgbaird]

@qzhu2017 curious if you had any additional thoughts on this? See also the comment in sparks-baird/xtal2png#83 (comment).

Any thoughts on how space group number could be used to guide reconstruction?