core.models.gemnet_oc.layers.radial_basis

Copyright (c) Meta, Inc. and its affiliates. This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree.

Classes

PolynomialEnvelope	Polynomial envelope function that ensures a smooth cutoff.
ExponentialEnvelope	Exponential envelope function that ensures a smooth cutoff,
GaussianBasis	Base class for all neural network modules.
SphericalBesselBasis	First-order spherical Bessel basis
BernsteinBasis	Bernstein polynomial basis,
RadialBasis

Module Contents

class core.models.gemnet_oc.layers.radial_basis.PolynomialEnvelope(exponent: int)

Bases: torch.nn.Module

Polynomial envelope function that ensures a smooth cutoff.

Parameters:

exponent (int) – Exponent of the envelope function.

p

a: float

b: float

c: float

forward(d_scaled: torch.Tensor) → torch.Tensor

class core.models.gemnet_oc.layers.radial_basis.ExponentialEnvelope

Bases: torch.nn.Module

Exponential envelope function that ensures a smooth cutoff, as proposed in Unke, Chmiela, Gastegger, Schütt, Sauceda, Müller 2021. SpookyNet: Learning Force Fields with Electronic Degrees of Freedom and Nonlocal Effects

forward(d_scaled: torch.Tensor) → torch.Tensor

class core.models.gemnet_oc.layers.radial_basis.GaussianBasis(start: float = 0.0, stop: float = 5.0, num_gaussians: int = 50, trainable: bool = False)

Bases: torch.nn.Module

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing to nest them in a tree structure. You can assign the submodules as regular attributes:

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will have their parameters converted too when you call to(), etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

Variables:

training (bool) – Boolean represents whether this module is in training or evaluation mode.

coeff

forward(dist: torch.Tensor) → torch.Tensor

class core.models.gemnet_oc.layers.radial_basis.SphericalBesselBasis(num_radial: int, cutoff: float)

Bases: torch.nn.Module

First-order spherical Bessel basis

Parameters:

• num_radial (int) – Number of basis functions. Controls the maximum frequency.

• cutoff (float) – Cutoff distance in Angstrom.

norm_const

frequencies

forward(d_scaled: torch.Tensor) → torch.Tensor

class core.models.gemnet_oc.layers.radial_basis.BernsteinBasis(num_radial: int, pregamma_initial: float = 0.45264)

Bases: torch.nn.Module

Bernstein polynomial basis, as proposed in Unke, Chmiela, Gastegger, Schütt, Sauceda, Müller 2021. SpookyNet: Learning Force Fields with Electronic Degrees of Freedom and Nonlocal Effects

Parameters:

• num_radial (int) – Number of basis functions. Controls the maximum frequency.

• pregamma_initial (float) – Initial value of exponential coefficient gamma. Default: gamma = 0.5 * a_0**-1 = 0.94486, inverse softplus -> pregamma = log e**gamma - 1 = 0.45264

pregamma

softplus

forward(d_scaled: torch.Tensor) → torch.Tensor

class core.models.gemnet_oc.layers.radial_basis.RadialBasis(num_radial: int, cutoff: float, rbf: dict[str, str] | None = None, envelope: dict[str, str | int] | None = None, scale_basis: bool = False)

Bases: torch.nn.Module

Parameters:

• num_radial (int) – Number of basis functions. Controls the maximum frequency.

• cutoff (float) – Cutoff distance in Angstrom.

• rbf (dict = {"name": "gaussian"}) – Basis function and its hyperparameters.

• envelope (dict = {"name": "polynomial", "exponent": 5}) – Envelope function and its hyperparameters.

• scale_basis (bool) – Whether to scale the basis values for better numerical stability.

inv_cutoff

scale_basis

forward(d: torch.Tensor) → torch.Tensor