nvalchemi.dynamics.FIRE2#

class nvalchemi.dynamics.FIRE2(model, dt, delaystep=_FIRE2_DEFAULTS['delaystep'], dtgrow=_FIRE2_DEFAULTS['dtgrow'], dtshrink=_FIRE2_DEFAULTS['dtshrink'], alphashrink=_FIRE2_DEFAULTS['alphashrink'], alpha0=_FIRE2_DEFAULTS['alpha0'], tmax=_FIRE2_DEFAULTS['tmax'], tmin=_FIRE2_DEFAULTS['tmin'], maxstep=_FIRE2_DEFAULTS['maxstep'], n_steps=None, hooks=None, convergence_hook=None, **kwargs)[source]#

Fixed-cell FIRE2 geometry optimizer.

Parameters:
  • model (BaseModelMixin) – The neural network potential model.

  • dt (float or torch.Tensor) – Initial adaptive timestep [M] or scalar. With by_group=True, M is the number of graph groups rather than the number of graphs.

  • delaystep (int) – Minimum steps before adaptation. Default 60.

  • dtgrow (float) – Timestep growth factor. Default 1.05.

  • dtshrink (float) – Timestep shrink factor. Default 0.75.

  • alphashrink (float) – Alpha decrease factor. Default 0.985.

  • alpha0 (float) – Initial mixing parameter. Default 0.09.

  • tmax (float) – Maximum timestep. Default 0.08.

  • tmin (float) – Minimum timestep. Default 0.005.

  • maxstep (float) – Maximum displacement per step. Default 0.1.

  • n_steps (int, optional) – Total steps for run().

  • hooks (list[Hook], optional) – Initial hooks.

  • convergence_hook (ConvergenceHook or dict, optional) – Convergence criterion.

  • by_group (bool, optional) – If True, update each group of graphs as a single unit, sharing the adaptive timestep, mixing parameter, and adaptation counters across the graphs in that group. Requires a valid layout set with batch.set_group_layout(). A configured convergence hook must use the same by_group setting. Default False. Forwarded through **kwargs to BaseDynamics.

  • **kwargs – Forwarded to BaseDynamics.

__needs_keys__#

{"forces"}.

Type:

set[str]

__provides_keys__#

{"positions", "velocities"}.

Type:

set[str]

post_update(batch)[source]#

No-op; forces from new positions are used on the next step.

Parameters:

batch (Batch)

Return type:

None

pre_update(batch)[source]#

Full FIRE2 step using current forces.

Parameters:

batch (Batch) – Current batch; positions and velocities updated in-place.

Return type:

None

samples_equilibrium: ClassVar[bool] = False#

A relaxation descends to a minimum rather than sampling an ensemble.