xnn.dnn.models.hdnnp#

High-Dimensional Neural Network Potential (Behler-Parrinello, 2007).

A thin composition over a featurizer: it pairs a per-atom invariant descriptor (here RadialSymmetryFunctions) with per-element atomic networks. The shared body (DescriptorPotential) lives in dnn/models/base.py; swap the featurizer for an AEV and you have ANI (see ani.py) – the model code is identical.

Classes

HDNNP(species[, cutoff, etas, rs, hidden])

High-Dimensional Neural Network Potential (Behler-Parrinello, 2007).