QuasiNewtonApprox (base class)

class kona.linalg.matrices.hessian.basic.QuasiNewtonApprox(vector_factory, optns={})

Bases: kona.linalg.matrices.hessian.basic.BaseHessian

Base class for quasi-Newton approximations of the Hessian

Variables:

• max_stored (int) – Maximum number of corrections stored.
• norm_init (float) – Initial norm of design component of gradient.
• init_hessian (KonaVector) – Initial (diagonal) Hessian approximation (stored as a vector).
• s_list (list of KonaVector) – Difference between subsequent solutions: \(s_k = x_{k+1} - x_k\)
• y_list (list of KonaVector) – Difference between subsequent gradients: \(y_k = g_{k+1} - g_k\)

add_correction(s_new, y_new)

Adds a new correction to the Hessian approximation.

Parameters:

• s_new (KonaVector) – Difference between subsequent solutions.
• y_new (KonaVector) – Difference between subsequent gradients.