xnn.transformer.featurizers.radial#

Exponential-normal radial basis (TorchMD-Net / TensorNet / BAMBOO).

The exponential-normal smearing expands an interatomic distance onto a set of Gaussians placed in exponential distance space rather than linearly, so the basis is dense at short range (where the physics is stiff) and sparse near the cutoff. It is the radial basis shared by the graph-transformer potentials (TorchMD-Net ExpNormalSmearing, TensorNet, and BAMBOO’s GET). The whole basis is smoothly damped to zero at the cutoff by the shared Behler CosineCutoff.

Classes

ExpNormalSmearing([n_rbf, cutoff, ...])

Exponential-normal radial basis expansion (TorchMD-Net-style).