#include <OptimisationProblem.h>

Inheritance diagram for dolfin::OptimisationProblem:

Collaboration diagram for dolfin::OptimisationProblem:

Public Member Functions
	OptimisationProblem ()
	Constructor.
virtual	~OptimisationProblem ()
	Destructor.
virtual double	f (const GenericVector &x)=0
	Compute the objective function :math:f(x).
virtual void	form (GenericMatrix &A, GenericMatrix &P, GenericVector &b, const GenericVector &x)
virtual void	F (GenericVector &b, const GenericVector &x)=0
	Compute the gradient :math:F(x) = f'(x).
virtual void	J (GenericMatrix &A, const GenericVector &x)=0
	Compute the Hessian :math:J(x) = f''(x).
virtual void	J_pc (GenericMatrix &P, const GenericVector &x)
Public Member Functions inherited from dolfin::NonlinearProblem
	NonlinearProblem ()
	Constructor.
virtual	~NonlinearProblem ()
	Destructor.

Detailed Description

This is a base class for nonlinear optimisation problems which return the real-valued objective function :math:f(x), its gradient :math:F(x) = f'(x) and its Hessian :math:J(x) = f''(x)

Member Function Documentation

◆ F()

virtual void dolfin::OptimisationProblem::F	(	GenericVector &	b,
		const GenericVector &	x )

pure virtual

Compute the gradient :math:F(x) = f'(x).

Implements dolfin::NonlinearProblem.

◆ form()

virtual void dolfin::OptimisationProblem::form	(	GenericMatrix &	A,
		GenericMatrix &	P,
		GenericVector &	b,
		const GenericVector &	x )

inlinevirtual

Function called by the solver before requesting F, J or J_pc. This can be used to compute F, J and J_pc together. Preconditioner matrix P can be left empty so that A is used instead

Reimplemented from dolfin::NonlinearProblem.

◆ J()

virtual void dolfin::OptimisationProblem::J	(	GenericMatrix &	A,
		const GenericVector &	x )

pure virtual

Compute the Hessian :math:J(x) = f''(x).

Implements dolfin::NonlinearProblem.

◆ J_pc()

virtual void dolfin::OptimisationProblem::J_pc	(	GenericMatrix &	P,
		const GenericVector &	x )

inlinevirtual

Compute J_pc used to precondition J. Not implementing this or leaving P empty results in system matrix A being used to construct preconditioner.

Note that if nonempty P is not assembled on first call then a solver implementation may throw away P and not call this routine ever again.