gradient_descent_step¶
- torch_sim.optimizers.gradient_descent.gradient_descent_step(state, model, *, pos_lr=0.01, cell_lr=0.1)[source]¶
Perform one gradient descent optimization step.
Updates atomic positions and optionally cell parameters based on the filter.
- Parameters:
model (ModelInterface) – Model that computes energies, forces, and optionally stress
state (OptimState | CellOptimState) – Current optimization state
pos_lr (float | Tensor) – Learning rate(s) for atomic positions
cell_lr (float | Tensor) – Learning rate(s) for cell optimization (ignored if no cell filter)
- Returns:
Updated OptimState after one optimization step
- Return type: