L-BFGS optimizer for unconstrained minimization. More...

#include <lbfgs.h>

Public Types
using	Objective = std::function<double(const Vec&, Vec&)>
	Objective function signature: f(x, gradient) -> energy.
using	Callback = std::function<bool(int, const Vec&, double, const Vec&)>
	Optional callback after each iteration: callback(iter, x, f, g) -> continue?

Public Member Functions
	LBFGS (const LBFGSSettings &settings={})
LBFGSResult	minimize (Objective f, const Vec &x0, int max_iter=200)
	Minimize the objective function starting from x0.
LBFGSResult	minimize (Objective f, const Vec &x0, Callback callback, int max_iter=200)
	Minimize with iteration callback.
const LBFGSSettings &	settings () const
	Access settings.
LBFGSSettings &	settings ()

Detailed Description

L-BFGS optimizer for unconstrained minimization.

Implements the Limited-memory Broyden-Fletcher-Goldfarb-Shanno algorithm with strong Wolfe line search.

Usage:

LBFGS optimizer(settings);
auto result = optimizer.minimize(
    [](const Vec& x, Vec& g) {
        // Compute f(x) and gradient g
        return f;
    },
    x0, max_iter);

Member Typedef Documentation

◆ Callback

using occ::mults::LBFGS::Callback = std::function<bool(int, const Vec&, double, const Vec&)>

Optional callback after each iteration: callback(iter, x, f, g) -> continue?

◆ Objective

using occ::mults::LBFGS::Objective = std::function<double(const Vec&, Vec&)>

Objective function signature: f(x, gradient) -> energy.

Constructor & Destructor Documentation

◆ LBFGS()

occ::mults::LBFGS::LBFGS ( const LBFGSSettings & settings = {} )

explicit

Member Function Documentation

◆ minimize() [1/2]

LBFGSResult occ::mults::LBFGS::minimize	(	Objective	f,
		const Vec &	x0,
		Callback	callback,
		int	max_iter = 200 )

Minimize with iteration callback.

Parameters

f	Objective function
x0	Initial parameters
callback	Called after each iteration; return false to stop
max_iter	Maximum number of iterations

Returns: Optimization result

◆ minimize() [2/2]

LBFGSResult occ::mults::LBFGS::minimize	(	Objective	f,
		const Vec &	x0,
		int	max_iter = 200 )

Minimize the objective function starting from x0.

Parameters

f	Objective function computing f(x) and storing gradient in g
x0	Initial parameters
max_iter	Maximum number of iterations

Returns: Optimization result

◆ settings() [1/2]

LBFGSSettings & occ::mults::LBFGS::settings ( )

inline

◆ settings() [2/2]

const LBFGSSettings & occ::mults::LBFGS::settings ( ) const