GustafssonController

class cubie.integrators.GustafssonController(precision: type[float16] | type[float32] | type[float64] | dtype[float16] | dtype[float32] | dtype[float64], dt: float = None, n_states: int = 1, **kwargs)

Bases: BaseAdaptiveStepController

Adaptive controller using Gustafsson acceleration.

build_controller(precision: type[float16] | type[float32] | type[float64] | dtype[float16] | dtype[float32] | dtype[float64], clamp: Callable, min_step_shrink: float, max_step_growth: float, dt_min: float, dt_max: float, algorithm_order: int, safety: float, error_norm: Callable) → ControllerCache

Create the device function for the Gustafsson controller.

Parameters:

• precision – Precision callable used to coerce scalars on device.

• clamp – Callable that clamps proposed step sizes.

• min_step_shrink – Most the step may shrink per adjustment.

• max_step_growth – Most the step may grow per adjustment.

• dt_min – Minimum permissible step size.

• dt_max – Maximum permissible step size.

• algorithm_order – Order of the integration algorithm.

• safety – Safety factor used when scaling the step size.

• error_norm – Device function returning the mean squared scaled error.

Returns:

CUDA device function implementing the Gustafsson controller.

Return type:

Callable

property newton_target_iters: int

Return the Newton-work reference used for gain damping.