torchani.ase#

Calculator subclass, used for interfacing with the ASE library

The ASE library can be used to perform molecular dynamics. For more information consult the user guide.

Functions

to_ase

from_ase

Classes

Calculator

TorchANI calculator for ASE

class torchani.ase.Calculator(model, overwrite=False, stress_kind='scaling')[source]#

TorchANI calculator for ASE

ANI models can be converted to their ASE Calculator form by calling the ANI.ase method.

import torchani
model = torchani.models.ANI1x()
calc = model.ase()  # Convert model into its "ASE Calculator" form
Parameters:
  • model (torchani.arch.ANI) – neural network potential model that convert coordinates into energies.

  • overwrite (bool) – Deprecated since version 2.9.0. After wrapping atoms into central box, replace the original positions in the ase.Atoms object with the wrapped positions.

  • stress_kind ("fdotr"|"scaling"|"numerical") – Strategy used to calculate stress. The “fdotr” approach only requires the cell volume, so it can be used with domain-decomposition approaches when running in parallel on multiple GPUs. The “scaling” approach uses the analytic gradient with respect to the strain \(\epsilon\), evaluated at \(\epsilon = 0\), after applying the deformation \(h' = h(I + \epsilon)\) to the cell and \(\mathbf{r}_i^T = \mathbf{r}_i^T(I + \epsilon)\) to the atomic coordinates.

implemented_properties: list[str] = ['energy', 'free_energy', 'forces', 'stress']#

Properties calculator can handle (energy, forces, …)

calculate(atoms=None, properties=['energy'], system_changes=['positions', 'numbers', 'cell', 'pbc', 'initial_charges', 'initial_magmoms'])[source]#

Do the calculation.

properties: list of str

List of what needs to be calculated. Can be any combination of ‘energy’, ‘forces’, ‘stress’, ‘dipole’, ‘charges’, ‘magmom’ and ‘magmoms’.

system_changes: list of str

List of what has changed since last calculation. Can be any combination of these six: ‘positions’, ‘numbers’, ‘cell’, ‘pbc’, ‘initial_charges’ and ‘initial_magmoms’.

Subclasses need to implement this, but can ignore properties and system_changes if they want. Calculated properties should be inserted into results dictionary like shown in this dummy example:

self.results = {'energy': 0.0,
                'forces': np.zeros((len(atoms), 3)),
                'stress': np.zeros(6),
                'dipole': np.zeros(3),
                'charges': np.zeros(len(atoms)),
                'magmom': 0.0,
                'magmoms': np.zeros(len(atoms))}

The subclass implementation should first call this implementation to set the atoms attribute and create any missing directories.