MACE

References: Batatia2022

This section specifies the input parameters for MACE potential type, a higher-order equivariant message-passing neural network. The MACE model must be exported to a TorchScript file that takes a single dictionary argument (see the create_cp2k_model.py helper). Requires linking with libtorch library from https://pytorch.org/cppdocs/installing.html. [Edit on GitHub]

Keywords

Keyword descriptions

ATOMS

Type: string[ ]
Usage: ATOMS {KIND 1} {KIND 2} .. {KIND N}

Description: Defines the atomic kinds involved in the MACE potential. Provide a list of each element, making sure that the mapping from the ATOMS list to MACE atom types is correct. This mapping should also be consistent for the atomic coordinates as specified in the sections COORDS or TOPOLOGY.

Mentions:MACE

[Edit on GitHub]

POT_FILE_NAME

Type: string
Aliases: MODEL_FILE_NAME
Usage: POT_FILE_NAME {FILENAME}

Description: Specifies the filename that contains the exported MACE model. MACE models use standardized units (Angstrom for length, eV for energy, eV/Angstrom for forces), so no unit keywords are required.

Mentions:MACE

[Edit on GitHub]