Unconstrained Reduced-Space Quasi-Newton

class kona.algorithms.ReducedSpaceQuasiNewton(primal_factory, state_factory, eq_factory, ineq_factory, optns=None)[source]

Bases: kona.algorithms.base_algorithm.OptimizationAlgorithm

Unconstrained optimization using quasi-Newton in the reduced space, globalized using either back-tracking or Strong Wolfe line search on the objective as the merit function.

This algorithm can leverage both limited-memory BFGS and limited-memory Symmetric Rank 1 approximations of the Hessian.

Variables:
  • factor_matrices (bool) – Boolean flag for matrix-based PDE solvers.
  • min_radius, max_radius (radius,) – Trust radius parameters.
  • mu_init, mu_max, mu_pow, eta (mu,) – Augmented Lagrangian constraint factor parameters.
  • grad_scale, feas_scale (scale,) – Optimality metric normalization factors.
  • approx_hessian (QuasiNewtonApprox-like) – The quasi-Newton approximation object for the Hessian.
  • globalization (string) – Flag to determine solution globalization type.
solve()[source]