belfem::opt::Optimizer Class Reference

driver that optimizes an Objective using a chosen Algorithm. More...

#include <cl_Optimizer.hpp>

Public Member Functions
	Optimizer (Objective &aObjective, const Algorithm aAlgorithm)
	~Optimizer ()=default
void	set_bounds (const Vector< real > &aLowerBounds, const Vector< real > &aUpperBounds)
	set the lower and upper bounds of the design variables.
Status	optimize (Vector< real > &aX, real &aValue)
	run the optimization, starting from and writing back into aX.
const string &	errmsg () const
	detailed diagnostic from the solver for the most recent optimize() call ( e.g.
real &	ftol_rel ()
	relative tolerance on the objective value ( ftol_rel )
real &	xtol_rel ()
	relative tolerance on the design vector ( xtol_rel )
uint &	max_eval ()
	maximum number of objective evaluations; 0 means unlimited
bool &	maximize ()
	flag selecting maximization ( true ) over minimization ( false )
const Algorithm &	algorithm () const
	the algorithm used by this optimizer

Detailed Description

driver that optimizes an Objective using a chosen Algorithm.

This is the counterpart of ode::Integrator: it owns the numerical bookkeeping (bounds, tolerances, evaluation budget) while remaining agnostic of what is being optimized. The concrete solver library (nlopt) is a private implementation detail of cl_Optimizer.cpp.

The Objective is borrowed by reference (non-owning) and must outlive every call to optimize(), just as ode::Integrator borrows its ODE.

Constructor & Destructor Documentation

Optimizer()

belfem::opt::Optimizer::Optimizer	(	Objective &	aObjective,
		const Algorithm	aAlgorithm )

~Optimizer()

belfem::opt::Optimizer::~Optimizer ( )

default

Member Function Documentation

algorithm()

const Algorithm & belfem::opt::Optimizer::algorithm ( ) const

inline

the algorithm used by this optimizer

errmsg()

const string & belfem::opt::Optimizer::errmsg ( ) const

inline

detailed diagnostic from the solver for the most recent optimize() call ( e.g.

which bound fails lb <= ub ); empty when no detail was reported. Complements the generic error_message( status ).

ftol_rel()

real & belfem::opt::Optimizer::ftol_rel ( )

inline

relative tolerance on the objective value ( ftol_rel )

max_eval()

uint & belfem::opt::Optimizer::max_eval ( )

inline

maximum number of objective evaluations; 0 means unlimited

maximize()

bool & belfem::opt::Optimizer::maximize ( )

inline

flag selecting maximization ( true ) over minimization ( false )

optimize()

Status belfem::opt::Optimizer::optimize	(	Vector< real > &	aX,
		real &	aValue )

run the optimization, starting from and writing back into aX.

Parameters

aX	on entry the initial guess, on exit the optimum
aValue	on exit the objective value at aX

Returns

the termination status

aX and aValue are only meaningful when is_usable( status ) is true; on failure they must not be trusted. When the solver reports a detailed diagnostic, it is available via errmsg().

set_bounds()

void belfem::opt::Optimizer::set_bounds	(	const Vector< real > &	aLowerBounds,
		const Vector< real > &	aUpperBounds )

set the lower and upper bounds of the design variables.

Each vector must have length objective.dimension().

Setting aLowerBounds( i ) == aUpperBounds( i ) fixes design variable i: nlopt eliminates that dimension internally, so parameters can be pinned without changing the Objective.

xtol_rel()

real & belfem::opt::Optimizer::xtol_rel ( )

inline

relative tolerance on the design vector ( xtol_rel )

---

The documentation for this class was generated from the following files:

• src/numerics/opt/cl_Optimizer.hpp
• src/numerics/opt/cl_Optimizer.cpp