jaxmat.solvers.custom_optimistix_solvers

jaxmat.solvers.custom_optimistix_solvers#

GaussNewtonTrustRegion(rtol, atol, norm=<function max_norm>, linear_solver=AutoLinearSolver(well_posed=True))[source]#

Gauss-Newton descent globalised with a classical trust region.

Return type:

AbstractLeastSquaresSolver

Parameters:

  • rtol (float)

  • atol (float)

  • norm (Callable[[PyTree], Shaped[Array, '']])

  • linear_solver (AbstractLinearSolver)

NewtonTrustRegion(rtol, atol, norm=<function max_norm>, linear_solver=AutoLinearSolver(well_posed=True))[source]#

Full Newton descent globalised with a classical trust region.

Derived from optimistix.LevenbergMarquardt but using a full Newton descent instead of the damped one.

Return type:

AbstractLeastSquaresSolver

Parameters:

  • rtol (float)

  • atol (float)

  • norm (Callable[[PyTree], Shaped[Array, '']])

  • linear_solver (AbstractLinearSolver)

BFGSLinearTrustRegion(rtol, atol, norm=<function max_norm>, use_inverse=True)[source]#

BFGS quasi-Newton descent globalised with a linear trust region.

Return type:

AbstractMinimiser

Parameters:

  • rtol (float)

  • atol (float)

  • norm (Callable[[PyTree], Shaped[Array, '']])

  • use_inverse (bool)

Contents