Equality Constrained Composite-Step RSNK¶
-
class
kona.algorithms.
CompositeStepRSNK
(primal_factory, state_factory, eq_factory, ineq_factory, optns={})[source]¶ Bases:
kona.algorithms.base_algorithm.OptimizationAlgorithm
A reduced-space composite-step optimization algorithm for PDE-governed equality constrained problems, globalized using a trust-region approach.
This implementation is based on the composite-step algorithm proposed by `Heinkenschloss and Ridzal<http://epubs.siam.org/doi/abs/10.1137/130921738>`_. However, we have omitted the inexactness corrections for simplicity and implemented a 2nd order adjoint approach for producing the necessary matrix-vector products.
Variables: - factor_matrices (bool) – Boolean flag for matrix-based PDE solvers.
- min_radius, max_radius (radius,) – Trust radius parameters.
- mu_max, mu_pow (mu,) – Augmented Lagrangian constraint factor parameters.
- normal_KKT (
AugmentedKKTMatrix
) – Matrix object for the normal step system. - tangent_KKT (
LagrangianHessian
) – Matrix object for the tangent step system. - globalization (string) – Flag to determine solution globalization type.