Model
Edge Basis
Basic
r_max
Type: floatDefault: n/aThe cutoff radius within which an atom is considered a neighbor.
irreps_edge_sh
Type: Irreps or intDefault: n/aThe irreps to use for the spherical harmonic projection of the edges. If an integer, specifies all spherical harmonics up to and including that integer as \(\ell_{\text{max}}\). If provided as explicit irreps, all multiplicities should be 1.
num_basis
Type: intDefault:8
The number of radial basis functions to use.
chemical_embedding_irreps_out
Type: IrrepsDefault: n/aThe size of the linear embedding of the chemistry of an atom.
Advanced
BesselBasis_trainable
Type: boolDefault:True
Whether the Bessel radial basis should be trainable.
basis
Type: typeDefault:<class 'nequip.nn.radial_basis.BesselBasis'>
The radial basis to use.
Convolution
Basic
num_layers
Type: intDefault:3
The number of convolution layers.
Advanced
invariant_layers
Type: intDefault:1
The number of hidden layers in the radial neural network.
invariant_neurons
Type: intDefault:8
The width of the hidden layers of the radial neural network.
resnet
Type: boolDefault:False
nonlinearity_type
Type: strDefault:gate
nonlinearity_scalars
Type: dictDefault:{'e': 'ssp', 'o': 'tanh'}
nonlinearity_gates
Type: dictDefault:{'e': 'ssp', 'o': 'abs'}
use_sc
Type: boolDefault:True