You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There are two ways to generate 3D structures from Wyckoff representations: DiffCSP++ and CHGNet. They later can be relaxed with CHGNet and/or DFT.
123
123
#### DiffCSP++
124
124
Wyckoffs can be relaxed with modified [DiffCSP++ code](https://github.com/kazeevn/DiffCSPNew/tree/master)
125
+
#### CrySPR + MACE
126
+
[CrySPR](https://chemrxiv.org/engage/chemrxiv/article-details/66b308a501103d79c5fd9b91) scheme using [`pyxtal`](https://pyxtal.readthedocs.io/en/latest/index.html) and a [MACE](https://github.com/ACEsuit/mace) ML force field is integrated directly into the package. Install the optional extra first:
URL-based models are downloaded once and cached in `~/.cache/wyckoff_transformer/mace_models/`. A local path is accepted too: `--model /path/to/model.model`.
142
+
143
+
Output layout is identical to the CHGNet variant below. Key options:
144
+
-`--n-trials N` — number of random PyXtal trials per structure (default 6)
145
+
-`--fmax F` — force convergence criterion in eV/Å (default 0.01)
146
+
-`--model-name NAME` — label for the results CSV (default: model file stem)
[CrySPR](https://chemrxiv.org/engage/chemrxiv/article-details/66b308a501103d79c5fd9b91) scheme that combines [`pyxtal`](https://pyxtal.readthedocs.io/en/latest/index.html) with CHGNet (or any other machine-learning interatomic potentials)
0 commit comments