search

MaterSim / PyXtal

A code to generate atomic structure with symmetry
MIT License
253 stars 68 forks source link

Suggestions on how to reconstruct a crystal from pairwise distances and unit cell parameters #199

Open sgbaird opened 2 years ago

sgbaird commented 2 years ago

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 https://github.com/sparks-baird/xtal2png/issues/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 https://github.com/sparks-baird/xtal2png/issues/79 and pairwise distances https://github.com/sparks-baird/xtal2png/issues/83) into the reconstruction process.

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

qzhu2017 commented 2 years ago

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 commented 2 years ago

@qzhu2017 that would be great actually.

sterling.baird@utah.edu

sgbaird commented 2 years ago

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

sgbaird commented 2 years ago

@qzhu2017 curious if you had any additional thoughts on this? See also the comment in https://github.com/sparks-baird/xtal2png/issues/83#issuecomment-1169591055.

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