Molecular optimizer derived from the QuasiNewton optimizer. More...

#include <molecular_optimizer.h>

Inheritance diagram for madness::MolecularOptimizer:

Collaboration diagram for madness::MolecularOptimizer:

[legend]

Public Member Functions
	MolecularOptimizer (World &world, const commandlineparser &parser, const std::shared_ptr< MolecularOptimizationTargetInterface > &tar)
	same ctor as the QuasiNewton optimizer

bool	converged () const

bool	converged (const Tensor< double > &displacement) const

double	gradient_norm () const

bool	optimize (Tensor< double > &x)
	optimize the underlying molecule

void	set_hessian (const Tensor< double > &hess)
	set an (initial) hessian

double	value () const

Public Member Functions inherited from madness::OptimizerInterface
virtual	~OptimizerInterface ()

Static Public Member Functions
static Tensor< double >	compute_frequencies (const Molecule &molecule, const Tensor< double > &hessian, Tensor< double > &normalmodes, const std::vector< std::string > &remove_dof={}, const bool print_hessian=false)
	returns the vibrational frequencies

static Tensor< double >	compute_reduced_mass (const Molecule &molecule, const Tensor< double > &normalmodes)

static Tensor< double >	projector_external_dof (const Molecule &mol, const std::vector< std::string > &remove_dof)
	compute the projector to remove transl. and rot. degrees of freedom

static void	remove_external_dof (Tensor< double > &hessian, const Molecule &mol, const std::vector< std::string > &remove_dof)
	remove translational degrees of freedom from the hessian

Public Attributes
MolecularOptimizationParameters	parameters

Private Member Functions
Tensor< double >	new_search_direction2 (const Tensor< double > &g, const Tensor< double > &hessian) const
	effectively invert the hessian and multiply with the gradient

bool	optimize_conjugate_gradients (Tensor< double > &x)

bool	optimize_quasi_newton (Tensor< double > &x)

Private Attributes
double	f =1.e10

double	gnorm =1.e10

Tensor< double >	h

std::shared_ptr< MolecularOptimizationTargetInterface >	target
	How to update the hessian: BFGS or SR1.

Detailed Description

Molecular optimizer derived from the QuasiNewton optimizer.

Essentially the QuasiNewton optimizer, but with the additional feature of projecting out rotational and translational degrees of freedom

Constructor & Destructor Documentation

◆ MolecularOptimizer()

madness::MolecularOptimizer::MolecularOptimizer	(	World &	world,
		const commandlineparser &	parser,
		const std::shared_ptr< MolecularOptimizationTargetInterface > &	tar
	)

inline

same ctor as the QuasiNewton optimizer

Member Function Documentation

◆ compute_frequencies()

static Tensor< double > madness::MolecularOptimizer::compute_frequencies	(	const Molecule &	molecule,
		const Tensor< double > &	hessian,
		Tensor< double > &	normalmodes,
		const std::vector< std::string > &	remove_dof = `{}`,
		const bool	print_hessian = `false`
	)

inlinestatic

returns the vibrational frequencies

Parameters

[in]	hessian	the hessian matrix (not mass-weighted)
[out]	normalmodes	the normal modes
[in]	project_tr	whether to project out translation and rotation
[in]	print_hessian	whether to print the hessian matrix

Returns: the frequencies in atomic units

Referenced by optimize_quasi_newton().

◆ compute_reduced_mass()

static Tensor< double > madness::MolecularOptimizer::compute_reduced_mass	(	const Molecule &	molecule,
		const Tensor< double > &	normalmodes
	)

inlinestatic

References madness::constants::atomic_mass_in_au, madness::copy(), madness::inner(), L, molecule, mu, and projector_external_dof().

◆ converged() [1/2]

bool madness::MolecularOptimizer::converged ( ) const

inlinevirtual

Implements madness::OptimizerInterface.

References gradient_norm().

Referenced by optimize_conjugate_gradients(), and optimize_quasi_newton().

◆ converged() [2/2]

bool madness::MolecularOptimizer::converged ( const Tensor< double > & displacement ) const

inline

References madness::displacement(), gradient_norm(), parameters, and madness::MolecularOptimizationParameters::tol().

◆ gradient_norm()

double madness::MolecularOptimizer::gradient_norm ( ) const

inlinevirtual

Implements madness::OptimizerInterface.

References gnorm.

Referenced by converged(), and converged().

◆ new_search_direction2()

Tensor< double > madness::MolecularOptimizer::new_search_direction2	(	const Tensor< double > &	g,
		const Tensor< double > &	hessian
	)		const

inlineprivate

effectively invert the hessian and multiply with the gradient

References std::abs(), e(), madness::g, madness::MolecularOptimizationParameters::gradient_precision(), madness::inner(), parameters, madness::MolecularOptimizationParameters::printtest(), madness::syev(), and v.

Referenced by optimize_quasi_newton().

◆ optimize()

bool madness::MolecularOptimizer::optimize ( Tensor< double > & x )

inlinevirtual

optimize the underlying molecule

Parameters

[in] x the coordinates to compute energy and gradient

Implements madness::OptimizerInterface.

References converge(), and optimize_quasi_newton().

Referenced by main().

◆ optimize_conjugate_gradients()

bool madness::MolecularOptimizer::optimize_conjugate_gradients ( Tensor< double > & x )

inlineprivate

References beta, madness::MolecularOptimizationParameters::cg_method(), converged(), madness::displacement(), energy, madness::Tensor< T >::fill(), gnorm, madness::inner(), madness::MolecularOptimizationParameters::maxiter(), madness::Tensor< T >::normf(), parameters, projector_external_dof(), madness::MolecularOptimizationParameters::remove_dof(), madness::BaseTensor::size(), and target.

◆ optimize_quasi_newton()

bool madness::MolecularOptimizer::optimize_quasi_newton ( Tensor< double > & x )

inlineprivate

References madness::Tensor< T >::absmax(), madness::constants::au2invcm, compute_frequencies(), converged(), madness::BaseTensor::dim(), e(), f, gnorm, h, madness::QuasiNewton::hessian_update_bfgs(), madness::QuasiNewton::hessian_update_sr1(), madness::inner(), madness::QuasiNewton::line_search(), madness::MolecularOptimizationParameters::maxiter(), new_search_direction2(), madness::Tensor< T >::normf(), parameters, madness::print(), madness::MolecularOptimizationParameters::printtest(), projector_external_dof(), madness::MolecularOptimizationParameters::remove_dof(), remove_external_dof(), madness::Tensor< T >::scale(), madness::BaseTensor::size(), madness::syev(), target, madness::Tensor< T >::trace(), madness::MolecularOptimizationParameters::update(), v, and madness::MolecularOptimizationParameters::value_precision().

Referenced by optimize().

◆ projector_external_dof()

static Tensor< double > madness::MolecularOptimizer::projector_external_dof	(	const Molecule &	mol,
		const std::vector< std::string > &	remove_dof
	)

inlinestatic

compute the projector to remove transl. and rot. degrees of freedom

taken from http://www.gaussian.com/g_whitepap/vib.htm I don't really understand the concept behind the projectors, but it seems to work, and it is not written down explicitly anywhere. NOTE THE ERROR IN THE FORMULAS ON THE WEBPAGE !

Parameters

[in] do_remove_dof which dof to remove: x,y,z,Rx,Ry,Rz (transl/rot)

References madness::_(), madness::constants::atomic_mass_in_au, madness::Molecule::center_of_mass(), e(), madness::Molecule::get_atom(), madness::Atom::get_mass_in_au(), I, madness::inner(), madness::Molecule::moment_of_inertia(), madness::Molecule::natom(), norm(), madness::Tensor< T >::normf(), p(), madness::print(), madness::Tensor< T >::scale(), madness::syev(), madness::commandlineparser::tolower(), madness::Molecule::translate(), madness::transpose(), v, madness::Atom::x, madness::Atom::y, and madness::Atom::z.

Referenced by madness::Nemo::compute_IR_intensities(), compute_reduced_mass(), optimize_conjugate_gradients(), optimize_quasi_newton(), and remove_external_dof().

◆ remove_external_dof()

static void madness::MolecularOptimizer::remove_external_dof	(	Tensor< double > &	hessian,
		const Molecule &	mol,
		const std::vector< std::string > &	remove_dof
	)