StructuredLBFGSDirectionParams< Conf > Struct Template Reference

#include <alpaqa/inner/directions/panoc/structured-lbfgs.hpp>

Detailed Description

template<Config Conf>
struct alpaqa::StructuredLBFGSDirectionParams< Conf >

Parameters for the StructuredLBFGSDirection class.

Definition at line 15 of file structured-lbfgs.hpp.

Collaboration diagram for StructuredLBFGSDirectionParams< Conf >:

Public Types
enum	FailurePolicy { FallbackToProjectedGradient , UseScaledLBFGSInput }

Public Attributes
real_t	hessian_vec_factor = 0
	Set this option to a nonzero value to include the Hessian-vector product \( \nabla^2_{x_\mathcal{J}x_\mathcal{K}}\psi(x) q_\mathcal{K} \) from equation 12b in [4], scaled by this parameter.
bool	hessian_vec_finite_differences = true
	If hessian_vec_factor is nonzero, set this option to true to approximate that term using finite differences instead of using AD.
bool	full_augmented_hessian = true
	If hessian_vec_factor is nonzero and hessian_vec_finite_differences is true, set this option to true to compute the exact Hessian of the augmented Lagrangian, false to approximate it using the Hessian of the Lagrangian.
enum alpaqa::StructuredLBFGSDirectionParams::FailurePolicy	failure_policy = FallbackToProjectedGradient
	What to do when L-BFGS failed (e.g.

Member Enumeration Documentation

FailurePolicy

template<Config Conf>

enum FailurePolicy

Enumerator
FallbackToProjectedGradient	If L-BFGS fails, propagate the failure and tell PANOC that no accelerated step is available, causing it to accept the projected gradient step instead.
UseScaledLBFGSInput	If L-BFGS fails, return \( q_\mathcal{J} = -\gamma\nabla_{x_\mathcal{J}}\psi(x^k) -\gamma\nabla^2_{x_\mathcal{J}x_\mathcal{K}}\psi(x) q_\mathcal{K} \) as the accelerated step (effectively approximating \( \nabla_{x_\mathcal{J}x_\mathcal{J}} \approx \gamma I \)).

Definition at line 32 of file structured-lbfgs.hpp.

Member Data Documentation

hessian_vec_factor

template<Config Conf>

real_t hessian_vec_factor = 0

Set this option to a nonzero value to include the Hessian-vector product \( \nabla^2_{x_\mathcal{J}x_\mathcal{K}}\psi(x) q_\mathcal{K} \) from equation 12b in [4], scaled by this parameter.

Set it to zero to leave out that term (this usually only slightly increases the number of iterations, and eliminates one Hessian-vector product per iteration, improving the overall runtime).

Definition at line 23 of file structured-lbfgs.hpp.

hessian_vec_finite_differences

template<Config Conf>

bool hessian_vec_finite_differences = true

If hessian_vec_factor is nonzero, set this option to true to approximate that term using finite differences instead of using AD.

Definition at line 26 of file structured-lbfgs.hpp.

full_augmented_hessian

template<Config Conf>

bool full_augmented_hessian = true

If hessian_vec_factor is nonzero and hessian_vec_finite_differences is true, set this option to true to compute the exact Hessian of the augmented Lagrangian, false to approximate it using the Hessian of the Lagrangian.

Definition at line 31 of file structured-lbfgs.hpp.

failure_policy

template<Config Conf>

enum alpaqa::StructuredLBFGSDirectionParams::FailurePolicy failure_policy = FallbackToProjectedGradient

What to do when L-BFGS failed (e.g.

if there were no pairs (s, y) with positive curvature).

---

The documentation for this struct was generated from the following file:

• src/alpaqa/include/alpaqa/inner/directions/panoc/structured-lbfgs.hpp