File GeometryOptimizer.h

Copyright

This code is licensed under the 3-clause BSD license.

Copyright ETH Zurich, Laboratory for Physical Chemistry, Reiher Group.

See LICENSE.txt for details.

namespace Scine

This header file contains functions that allow for common notation for common things that can be done at a different degree of derivatives.

This header contains alias definitions defining which classes to use for the different degrees of derivatives.

namespace Utils
class GeometryOptimizerBase
#include <GeometryOptimizer.h>

The base class for all Geometry optimizers.

The purpose of the geometry optimizers is to wrap technical details needed for the actual optimization into a class that is easily used and has a small set of data needed in its constructor, is mainly configured via its settings, and then exposes the optimization through a simple function accepting a geometry (AtomCollection)

The main purpose of this base class is to hide the template parameter(s) of the derived class(es).

Subclassed by Scine::Utils::GeometryOptimizer< OptimizerType >

Public Functions

GeometryOptimizerBase()

Default constructor.

virtual ~GeometryOptimizerBase()

Virtual default destructor.

virtual int optimize(AtomCollection &atoms) = 0

The main functionality of the geometry optimizer.

This function wraps the optimize functions of the underlying optimizer.

Return

int The final number of optimization cycles carried out.

Parameters

virtual void setSettings(const Settings &settings) = 0

Function to apply the given settings to underlying classes.

Parameters
  • settings: The new settings.

virtual Settings getSettings() const = 0

Get the public settings as a Utils::Settings object.

Return

Settings A settings object with the current settings.

virtual void addObserver(std::function<void(const int&, const double&, const Eigen::VectorXd&)> function) = 0

Add an observer function that will be triggered in each iteration.

Parameters
  • function: A function to be executed in every loop of the optimization. The function will have access to the current cycle count, the current value and to a const reference of the current parameters.

virtual void clearObservers() = 0

Clear all existing observer functions.

For optimization problems with very fast evaluations of the underlying function the removal of all observers can increase performance as the observers are given as std::functions and can not be added via templates.

Public Members

bool transformCoordinates = true

Switch to transform the coordinates from Cartesian into an internal space.

The optimization will be carried out in the internal coordinate space possibly accellerating convergence.

Public Static Attributes

constexpr const char *geooptTransfromCoordinatesKey = "geoopt_transfrom_coordinates"
template<class OptimizerType, class ConvergenceCheckType>
class GeometryOptimizerSettings : public Scine::Utils::Settings
#include <GeometryOptimizer.h>

Settings for a GeometryOptimizer.

Uses template arguments in order to automatically include the settings of underlying objects into the given settings.

Template Parameters

Public Functions

GeometryOptimizerSettings(const GeometryOptimizerBase &base, const OptimizerType &optimizer, const ConvergenceCheckType &check)

Construct a new GeometryOptimizerSettings object.

Sets the default values of the settings to the current values set in the objects given to the constructor.

Parameters
  • base: The geometry optimizer.

  • optimizer: The optimizer.

  • check: The convergence check criteria.

template<class OptimizerType>
class GeometryOptimizer : public Scine::Utils::GeometryOptimizerBase
#include <GeometryOptimizer.h>

Basically just a templated version of the base class GeometryOptimizerBase, where the template defines the actual optimizer used in the geometry optimization.

Template Parameters
  • OptimizerType: Expects any of the Optimizer classes. Note that some special optimizers may not yet be supported or may need additional specialization.

Public Functions

GeometryOptimizer(Core::Calculator &calculator)

Construct a new GeometryOptimizer object.

Parameters
  • calculator: The calculator to be used for the single point/gradient calculations.

virtual int optimize(AtomCollection &atoms)

See GeometryOptimizerBase::optimize().

Return

int The final number of optimization cycles carried out.

Parameters

virtual void setSettings(const Settings &settings)

Function to apply the given settings to underlying classes.

Parameters
  • settings: The new settings.

virtual Settings getSettings() const

Get the public settings as a Utils::Settings object.

Return

Settings A settings object with the current settings.

virtual void addObserver(std::function<void(const int&, const double&, const Eigen::VectorXd&)> function)

Add an observer function that will be triggered in each iteration.

Parameters
  • function: A function to be executed in every loop of the optimization. The function will have access to the current cycle count, the current value and to a const reference of the current parameters.

virtual void clearObservers()

Clear all existing observer functions.

For optimization problems with very fast evaluations of the underlying function the removal of all observers can increase performance as the observers are given as std::functions and can not be added via templates.

template<>
int optimize(AtomCollection &atoms)

The main functionality of the geometry optimizer.

This function wraps the optimize functions of the underlying optimizer.

Return

int The final number of optimization cycles carried out.

Parameters

template<>
int optimize(AtomCollection &atoms)

The main functionality of the geometry optimizer.

This function wraps the optimize functions of the underlying optimizer.

Return

int The final number of optimization cycles carried out.

Parameters

Public Members

OptimizerType optimizer

The underlying optimizer, public in order to change it’s settings.

GradientBasedCheck check

The underlying convergence check, public in order to change it’s settings.

Private Members

Core::Calculator &_calculator