Linear optimizers

Linear optimizers are offered to fit parts of the MTP parameters deterministically.

class motep.optimizers.lls.LLSOptimizerBase(loss: LossFunctionBase, *, minimized: list[str] | None = None, **kwargs: dict[str, Any])

Bases: ParallelOptimizerBase

Abstract base class for linear optimizers.

• LLSOptimizer

• Level2MTPOptimizer

minimized

Properties whose errors are minimized. The elements must be some of energy, forces, stress, and mgrad (magnetic moment gradients, for magnetic potentials). Default: ["energy", "forces", "stress", "mgrad"].

Type:

list[str]

class motep.optimizers.lls.LLSOptimizer(loss: LossFunctionBase, *, minimized: list[str] | None = None, **kwargs: dict[str, Any])

Bases: LLSOptimizerBase

Optimizer based on linear least squares (LLS).

The elements of the optimized attribute must be:

• species_coeffs

• moment_coeffs

radial_coeffs cannot be optimized with this optimizer.

class motep.optimizers.level2mtp.Level2MTPOptimizer(loss: LossFunctionBase, *, minimized: list[str] | None = None, **kwargs: dict[str, Any])

Bases: LLSOptimizerBase

Optimizer for Level 2 MTP.

moment_coeffs and radial_coeffs are automatically set to be a level 2 MTP. Further, species_coeffs can also be optimized.