torch_sim.optimizers

Optimizers for geometry relaxations.

This module provides optimization algorithms for atomic structures in a batched format, enabling efficient relaxation of multiple atomic structures simultaneously. It includes several gradient-based methods with support for both atomic position and unit cell optimization.

The module offers:

• Standard gradient descent for atomic positions

• Gradient descent with unit cell optimization

• FIRE (Fast Inertial Relaxation Engine) optimization with unit cell parameters

• FIRE optimization with Frechet cell parameterization for improved cell relaxation

ASE-style FIRE: https://gitlab.com/ase/ase/-/blob/master/ase/optimize/fire.py?ref_type=heads Velocity Verlet-style FIRE: https://doi.org/10.1103/PhysRevLett.97.170201

Functions

fire	Initialize a batched FIRE optimization.
frechet_cell_fire	Initialize a batched FIRE optimization with Frechet cell parameterization.
gradient_descent	Initialize a batched gradient descent optimization.
unit_cell_fire	Initialize a batched FIRE optimization with unit cell degrees of freedom.
unit_cell_gradient_descent	Initialize a batched gradient descent optimization with unit cell parameters.

Classes

FireState	State information for batched FIRE optimization.
FrechetCellFIREState	State class for batched FIRE optimization with Frechet cell derivatives.
GDState	State class for batched gradient descent optimization.
UnitCellFireState	State information for batched FIRE optimization with unit cell degrees of freedom.
UnitCellGDState	State class for batched gradient descent optimization with unit cell.
ase_fire_key	str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
vv_fire_key	str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str