Generalize forcefields for generic ASE calculator support #940
Conversation
|
@esoteric-ephemera Great! |
|
Hi @JaGeo, restored the |
|
Hi @esoteric-ephemera, this is really nice to see in atomate2. Your implementation looks great to me. Would you be also happy to add static makers? In terms of your todos:
I'm not sure if this makes sense? We'd still need to make subclasses for specific implementations anyway, so I don't think a base implementation would add anything?
This would be great if you can figure it out. |
Sounds good, I'll just port the MD maker over to Static makers have been added + tests |
|
Hi @esoteric-ephemera. Thanks again for this. Is it ready to merge? |
* starting point for input set docs * remove sentence * correct InputGenerator name in figure * adjust figure size * reflect changes of vasp inputsets now in pymatgen * fix linting * change torchdata version according to #940 * change torchdata version according to #940 * Update pyproject.toml resolve conflict with changed matgl version
|
Hi @utf should be ready to merge |
src/atomate2/ase/jobs.py
Outdated
| AseStructureTaskDoc or AseMoleculeTaskDoc | ||
| """ | ||
| return AseTaskDoc.to_mol_or_struct_metadata_doc( | ||
| getattr(self.calculator, "name", self.calculator.__class__), |
There was a problem hiding this comment.
i think this line will fail pydantic validation if a calculator should ever not have a name. __class__ is not a string, we want the __name__ attribute
There was a problem hiding this comment.
good catch!
|
Thanks @esoteric-ephemera, this is very exciting to have! |
|
In that context: should we document this and a few other of our new workflows that are not VASP-based more clearly? I think a lot of people have interest in this general interface but might not discover it easily |
|
@JaGeo agreed, this should probably go in the docs for clarity. Also happy to add to the working paper |
|
If you're happy to add to the paper, I think it should be highlighted. |
|
And thanks @esoteric-ephemera (bit confused today 🙈 or maybe as usual) |
* starting point for input set docs * remove sentence * correct InputGenerator name in figure * adjust figure size * reflect changes of vasp inputsets now in pymatgen * fix linting * change torchdata version according to materialsproject#940 * change torchdata version according to materialsproject#940 * Update pyproject.toml resolve conflict with changed matgl version
…roject#940) - Add ASE base calculator classes from forcefield library - Add tests for ASE calculators, refactor forcefields - Add static makers - Add tblite dependence to ASE in toml - Migrate forcefield MD base class to ASE - Add barebones tests for ASE MD jobs - Generalize to allow molecules in ASE calculations - Make tblite dependence optional - Move forcefield strict version pins to separate toml block - Refactor naming, enumify and lazy load MD dynamics modules - Make Ase{Structure,Molecule}TaskDoc construction a single function - Change ASE and forcefield result from dict to BaseModel - Ensure ASE and forcefield result docs are subscriptable - Add "ionic_steps" to DATA_OBJECTS - Add 'elapsed_time' and 'tags' to ASE schemas - Add timing to ASE runs - Define basic ASE job schema; remove task_doc_kwargs from ASE in favor of explicit kwargs - Revise ionic_step storage (only for MD) - Add option to pass forcefield name as str without MLFF prefix or as enum - Add deprecation warnings to all predefined MLFF makers (static, relax, MD) - Correct typing of atoms in AseRelaxer - Add tblite to optional ase-ext and strict requirements - Remove dependence on deprecated forcefield makers - Make default calc kwargs and ensure these are loaded when not specified by user - Remove Frechet cell filter safety checks - Make MD ensembles an enum - Move ASE tests to separate test run - Add ASE to phonon supported codes, enforce string literal in BasePhononMaker - Fix potential Pydantic validation error if Ase(MD|Relax)Maker.calculator has no name --------- Co-authored-by: orionarcher <[email protected]> Co-authored-by: Janosh Riebesell <[email protected]>
This is a refactor to better support generic ASE calculators in atomate2, as well as molecules with ASE calculators.
Major changes:
schemas/utilshave been ported from theforcefieldslibrary to a separateaselibraryForceFieldTaskDocumentfrom the baseAseStructureTaskDocclassStructure, andMolecule-specific ASE task doc classesForcefieldResultandForcefieldObjectclasses alias ASE parent classes + deprecation warningforcefieldstoase.MolecularDynamicsobjectsMost of the other moved classes are internals for atomate2, so I don't expect many important consequences there