xnn.ffnn.models.ffield.template_library#

xnn.ffnn.models.ffield.template_library(species, *, nn=True, messages=1, message_function=3, energy_function=1, bo_function=0, mf_layer=(8, 1), be_layer=(8, 1), seed=0)[source]#

Build a generic starting-point (seed) parameter library.

The library carries plausible, untrained values for every classical parameter – magnitudes follow the published general-parameter and hydrocarbon tables of van Duin et al. (2001), radii follow covalent radii, and the EEM electronegativities / hardnesses are the standard Pearson values – plus, when nn is on, randomly initialized ReaxFF-nn message / bond-energy networks. It is a seed for training (the point of the ffnn family), a template whose values can be replaced by a published parameterization, and the fixture for self-contained tests. It is not a validated force field.

Parameters:

species (sequence of str) – Chemical symbols to parameterize (organic elements are tabulated; others fall back to carbon-like values).
nn (bool, optional) – Include ReaxFF-nn network weights, by default True.
messages (int, optional) – Message-passing steps T, by default 1.
message_function (int, optional) – ReaxFF-nn function-form selectors, by default 3 / 1 / 0 (the combination used by published ReaxFF-nn parameter sets).
energy_function (int, optional) – ReaxFF-nn function-form selectors, by default 3 / 1 / 0 (the combination used by published ReaxFF-nn parameter sets).
bo_function (int, optional) – ReaxFF-nn function-form selectors, by default 3 / 1 / 0 (the combination used by published ReaxFF-nn parameter sets).
mf_layer (tuple, optional) – (width, n_hidden) of the message and bond-energy networks, by default (8, 1).
be_layer (tuple, optional) – (width, n_hidden) of the message and bond-energy networks, by default (8, 1).
seed (int, optional) – Seed for the network-weight initialization, by default 0.

Returns:

The template library.

Return type:

FFieldLibrary