representation.SchNet

class representation.SchNet(*args: Any, **kwargs: Any)

SchNet architecture for learning representations of atomistic systems

References:

[1]

Schütt, Arbabzadah, Chmiela, Müller, Tkatchenko: Quantum-chemical insights from deep tensor neural networks. Nature Communications, 8, 13890. 2017.

[2]

Schütt, Kindermans, Sauceda, Chmiela, Tkatchenko, Müller: SchNet: A continuous-filter convolutional neural network for modeling quantum interactions. In Advances in Neural Information Processing Systems, pp. 992-1002. 2017.

[3]

Schütt, Sauceda, Kindermans, Tkatchenko, Müller: SchNet - a deep learning architecture for molceules and materials. The Journal of Chemical Physics 148 (24), 241722. 2018.

Parameters:

• n_atom_basis – number of features to describe atomic environments. This determines the size of each embedding vector; i.e. embeddings_dim.

• n_interactions – number of interaction blocks.

• radial_basis – layer for expanding interatomic distances in a basis set

• cutoff_fn – cutoff function

• n_filters – number of filters used in continuous-filter convolution

• shared_interactions – if True, share the weights across interaction blocks and filter-generating networks.

• max_z – maximal nuclear charge

• activation – activation function

• activate_charge_spin_embedding – if True, charge and spin embeddings are added to nuclear embeddings taken from SpookyNet Implementation

• embedding – type of nuclear embedding to use (simple is simple embedding and complex is the one with electron configuration)