xnn.ffnn.models.ffield#
ReaxFF parameter libraries: the SEAMM .frc format and ReaxFF-nn JSON.
A ReaxFF model is fully specified by its parameter library. Two on-disk formats are supported:
the MolSSI/SEAMM
.frcforce-field format (xnn.ffnn.common.frc), in which the standard ReaxFF parameter blocks appear as named sections (#reaxff_general_parameters,#reaxff_atomic_parameters_1-8…#reaxff_hydrogen-bond_parameters) with every parameter identified by name rather than by column position. Published fields translated to this format ship with xnn (ReaxFF("CHO_cho_2008"); seelist_forcefields());the JSON parameter-library format used by ReaxFF-nn parameter sets (Guo et al., Comput. Mater. Sci. 172, 109393, 2020; Xue et al., PCCP 23, 19457, 2021), which stores the same parameters as a flat
"<name>_<type>"dictionary plus the neural-network weight matrices and the function/layer selectors. This is also the formatFFieldLibrary.save()writes, since network weights have no place in a.frcfile.
Parameters are kept in the flat naming convention of ReaxFF libraries
("Desi_C-C", "val_C", "theta0_H-C-H", …) and in the file’s
units (kcal/mol where applicable – the model converts to eV when it
assembles its tensors). read_ffield() returns a FFieldLibrary;
the completion helpers below apply the same conventions ReaxFF codes apply
after reading a library (off-diagonal combination rules, hydrogen-bond
defaults, torsion wildcard resolution).
Functions
|
Add placeholder parameters for unlisted |
|
Fill missing off-diagonal pair parameters in place. |
|
Build a complete per-bond-type cutoff table for the given species. |
|
Drop torsion types that are reversed spellings of an earlier entry. |
|
Heuristic fallback cutoff for a species pair, in Angstrom. |
|
Build a classical |
|
Read a ReaxFF parameter library. |
|
Look up a torsion parameter, resolving reversal and wildcards. |
|
Build a generic starting-point (seed) parameter library. |
|
Write a classical |
Classes
|
A parsed ReaxFF parameter library. |