Open Qmin: energyMinimizerGradientDescent Class Reference

Implement energy minimization via the FIRE algorithm. More...

#include <energyMinimizerGradientDescent.h>

Inheritance diagram for energyMinimizerGradientDescent:

Collaboration diagram for energyMinimizerGradientDescent:

[legend]

Public Member Functions
	energyMinimizerGradientDescent ()
	The basic constructor. More...

	energyMinimizerGradientDescent (shared_ptr< simpleModel > system)
	The basic constructor that feeds in a target system to minimize. More...

virtual void	initializeParameters ()
	Sets a bunch of default parameters that do not depend on the number of degrees of freedom. More...

void	initializeFromModel ()
	Set a bunch of default initialization parameters (if the State is available to determine the size of vectors) More...

void	setGradientDescentParameters (scalar deltaT, scalar forceCutoff)
	Set a lot of parameters! More...

void	setForceCutoff (scalar fc)
	Set the force cutoff. More...

void	gradientDescentStep ()
	an interface to call either the CPU or GPU FIRE algorithm More...

void	gradientDescentCPU ()
	Perform a velocity Verlet step on the CPU. More...

void	gradientDescentGPU ()
	Perform a velocity Verlet step on the GPU. More...

void	minimize ()
	Minimize to either the force tolerance or the maximum number of iterations. More...

virtual void	performUpdate ()
	The "intergate equatios of motion just calls minimize. More...

virtual scalar	getMaxForce ()
	Return the maximum force. More...

virtual scalar	getClassSize ()

virtual void	integrateEquationOfMotion ()

virtual void	integrateEOMGPU ()

virtual void	integrateEOMCPU ()

virtual void	Update (int timestep)
	The fundamental function that a controlling Simulation can call. More...

void	setSimulation (shared_ptr< basicSimulation > _sim)
	set the simulation More...

virtual void	setModel (shared_ptr< simpleModel > _model)
	virtual function to allow the model to be a derived class More...

void	setPeriod (int _p)
	set the period More...

void	setPhase (int _p)
	set the phase More...

virtual void	spatialSorting ()
	allow for spatial sorting to be called if necessary... More...

virtual void	setReproducible (bool rep)
	Allow for a reproducibility call to be made. More...

virtual void	setGPU (bool _useGPU=true)
	Enforce GPU operation. More...

int	getNdof ()
	Get the number of degrees of freedom of the equation of motion. More...

void	setNdof (int _n)
	Set the number of degrees of freedom of the equation of motion. More...

virtual void	setDeltaT (scalar dt)
	allow all updaters to potentially implement an internal time scale More...

virtual void	setNThreads (int n)
	allow for setting multiple threads More...

void	setMaximumIterations (int maxIt)
	Set the maximum number of iterations before terminating (or set to -1 to ignore) More...

int	getCurrentIterations ()

int	getMaxIterations ()

void	setCurrentIterations (int newIterations)

int	getNTotal ()
	communicate the number of non-object sites across ranks More...

Public Attributes
GPUArray< dVec >	displacement
	an array of displacements More...

shared_ptr< basicSimulation >	sim
	A pointer to the governing simulation. More...

shared_ptr< simpleModel >	model
	A pointer to a simpleModel that the updater acts on. More...

vector< scalar >	updaterData

int	iterations
	The number of iterations performed. More...

Protected Attributes
scalar	forceMax
	sqrt(force.force) / N_{dof} More...

scalar	forceCutoff
	The cutoff value of the maximum force. More...

shared_ptr< kernelTuner >	dotProductTuner
	kernel tuner for performance More...

GPUArray< scalar >	sumReductionIntermediate
	Utility array for simple reductions. More...

GPUArray< scalar >	sumReductionIntermediate2

int	nThreads =1
	number of threads to use More...

int	Period
	The period of the updater... the updater will work every Period timesteps. More...

int	Phase
	The phase of the updater... the updater will work every Period timesteps offset by a phase. More...

bool	useGPU
	whether the updater does its work on the GPU or not More...

bool	neverGPU
	whether the updater never does work on the GPU More...

int	Ndof
	some measure of the number of degrees of freedom the equations of motion might need to know about locally More...

int	nTotal
	the total number of non-object sites across all ranks More...

bool	reproducible
	whether the RNGs give reproducible results More...

scalar	deltaT
	The internal time step size. More...

int	maxIterations
	The maximum number of iterations allowed. More...

Detailed Description

Implement energy minimization via the FIRE algorithm.

This class uses a simple gradient descent algorithm The class is in the same framework as the simpleEquationOfMotion class, so it is called by calling performTimestep on a Simulation object that has been given the minimizer and the configuration to minimize. Each timestep, though, is a complete minimization (i.e. will run for the maximum number of iterations, or until a target tolerance has been acheived, or whatever stopping condition the user sets.

Constructor & Destructor Documentation

◆ energyMinimizerGradientDescent() [1/2]

energyMinimizerGradientDescent::energyMinimizerGradientDescent ( )

inline

The basic constructor.

References initializeParameters().

◆ energyMinimizerGradientDescent() [2/2]

energyMinimizerGradientDescent::energyMinimizerGradientDescent ( shared_ptr< simpleModel > system )

The basic constructor that feeds in a target system to minimize.

Initialize the minimizer with a reference to a target system, set a bunch of default parameters. Of note, the current default is CPU operation

References initializeFromModel(), initializeParameters(), and updater::setModel().

Member Function Documentation

◆ initializeParameters()

void energyMinimizerGradientDescent::initializeParameters ( )

virtual

Sets a bunch of default parameters that do not depend on the number of degrees of freedom.

Initialize the minimizer with some default parameters. that do not depend on Ndof

References dotProductTuner, updater::iterations, updater::setDeltaT(), setForceCutoff(), updater::setGPU(), updater::setMaximumIterations(), and updater::updaterData.

Referenced by energyMinimizerGradientDescent().

◆ initializeFromModel()

void energyMinimizerGradientDescent::initializeFromModel ( )

virtual

Set a bunch of default initialization parameters (if the State is available to determine the size of vectors)

Initialize the minimizer with some default parameters.

Precondition: requires a Simple2DModel (to set N correctly) to be already known

Reimplemented from equationOfMotion.

References equationOfMotion::displacement, updater::model, updater::Ndof, updater::neverGPU, GPUArray< T >::noGPU, GPUArray< T >::resize(), sumReductionIntermediate, and sumReductionIntermediate2.

Referenced by energyMinimizerGradientDescent(), and minimize().

◆ setGradientDescentParameters()

void energyMinimizerGradientDescent::setGradientDescentParameters	(	scalar	deltaT,
		scalar	forceCutoff
	)

Set a lot of parameters!

References updater::deltaT, forceCutoff, updater::setDeltaT(), and setForceCutoff().

◆ setForceCutoff()

void energyMinimizerGradientDescent::setForceCutoff ( scalar fc )

inline

Set the force cutoff.

References forceCutoff.

Referenced by initializeParameters(), and setGradientDescentParameters().

◆ gradientDescentStep()

void energyMinimizerGradientDescent::gradientDescentStep ( )

an interface to call either the CPU or GPU FIRE algorithm

Call the correct FIRE step routine

References gradientDescentCPU(), gradientDescentGPU(), and updater::useGPU.

Referenced by minimize().

◆ gradientDescentCPU()

void energyMinimizerGradientDescent::gradientDescentCPU ( )

Perform a velocity Verlet step on the CPU.

Perform a GD minimization step on the CPU

References updater::deltaT, dot(), forceMax, updater::model, updater::Ndof, updater::nTotal, scalar, updater::sim, and updater::updaterData.

Referenced by gradientDescentStep().

◆ gradientDescentGPU()

void energyMinimizerGradientDescent::gradientDescentGPU ( )

Perform a velocity Verlet step on the GPU.

Perform a GD minimization step on the GPU

References updater::deltaT, forceMax, gpu_gpuarray_dVec_dot_products(), updater::model, updater::Ndof, updater::nTotal, scalar, updater::sim, sumReductionIntermediate, sumReductionIntermediate2, and updater::updaterData.

Referenced by gradientDescentStep().

◆ minimize()

void energyMinimizerGradientDescent::minimize ( )

Minimize to either the force tolerance or the maximum number of iterations.

Perform a GD minimization sequence...attempts to get attain: (1/N)\sum_i|f_i|^2 < forceCutoff

References forceCutoff, forceMax, gradientDescentStep(), initializeFromModel(), updater::iterations, updater::model, updater::Ndof, and updater::sim.

Referenced by performUpdate().

◆ performUpdate()

virtual void energyMinimizerGradientDescent::performUpdate ( )

inlinevirtual

The "intergate equatios of motion just calls minimize.

Reimplemented from equationOfMotion.

References minimize().

◆ getMaxForce()

virtual scalar energyMinimizerGradientDescent::getMaxForce ( )

inlinevirtual

Return the maximum force.

Reimplemented from updater.

References forceMax.

◆ getClassSize()

virtual scalar energyMinimizerGradientDescent::getClassSize ( )

inlinevirtual

Reimplemented from equationOfMotion.

References equationOfMotion::getClassSize(), GPUArray< T >::getNumElements(), scalar, sumReductionIntermediate, and sumReductionIntermediate2.

◆ integrateEquationOfMotion()

virtual void equationOfMotion::integrateEquationOfMotion ( )

inlinevirtualinherited

References equationOfMotion::initializeFromModel(), equationOfMotion::integrateEOMCPU(), equationOfMotion::integrateEOMGPU(), updater::model, updater::Ndof, and updater::useGPU.

Referenced by energyMinimizerFIRE::minimize(), and equationOfMotion::performUpdate().

◆ integrateEOMGPU()

virtual void equationOfMotion::integrateEOMGPU ( )

inlinevirtualinherited

Reimplemented in velocityVerlet.

Referenced by equationOfMotion::integrateEquationOfMotion().

◆ integrateEOMCPU()

virtual void equationOfMotion::integrateEOMCPU ( )

inlinevirtualinherited

Reimplemented in velocityVerlet.

Referenced by equationOfMotion::integrateEquationOfMotion().

◆ Update()

virtual void updater::Update ( int timestep )

inlinevirtualinherited

The fundamental function that a controlling Simulation can call.

References updater::performUpdate(), updater::Period, and updater::Phase.

◆ setSimulation()

void updater::setSimulation ( shared_ptr< basicSimulation > _sim )

inlineinherited

set the simulation

References updater::sim.

◆ setModel()

virtual void updater::setModel ( shared_ptr< simpleModel > _model )

inlinevirtualinherited

virtual function to allow the model to be a derived class

References updater::initializeFromModel(), and updater::model.

Referenced by energyMinimizerFIRE::energyMinimizerFIRE(), energyMinimizerGradientDescent(), energyMinimizerLoLBFGS::energyMinimizerLoLBFGS(), and energyMinimizerNesterovAG::energyMinimizerNesterovAG().

◆ setPeriod()

void updater::setPeriod ( int _p )

inlineinherited

set the period

References updater::Period.

◆ setPhase()

void updater::setPhase ( int _p )

inlineinherited

set the phase

References updater::Phase.

◆ spatialSorting()

virtual void updater::spatialSorting ( )

inlinevirtualinherited

allow for spatial sorting to be called if necessary...

◆ setReproducible()

virtual void updater::setReproducible ( bool rep )

inlinevirtualinherited

Allow for a reproducibility call to be made.

References updater::reproducible.

◆ setGPU()

virtual void updater::setGPU ( bool _useGPU = true )

inlinevirtualinherited

Enforce GPU operation.

References updater::useGPU.

Referenced by energyMinimizerNesterovAG::initializeParameters(), energyMinimizerLoLBFGS::initializeParameters(), initializeParameters(), and energyMinimizerFIRE::initializeParameters().

◆ getNdof()

int updater::getNdof ( )

inlineinherited

Get the number of degrees of freedom of the equation of motion.

References updater::Ndof.

◆ setNdof()

void updater::setNdof ( int _n )

inlineinherited

Set the number of degrees of freedom of the equation of motion.

References updater::Ndof.

◆ setDeltaT()

virtual void updater::setDeltaT ( scalar dt )

inlinevirtualinherited

allow all updaters to potentially implement an internal time scale

Reimplemented in energyMinimizerFIRE.

References updater::deltaT.

Referenced by initializeParameters(), and setGradientDescentParameters().

◆ setNThreads()

virtual void updater::setNThreads ( int n )

inlinevirtualinherited

allow for setting multiple threads

References n, and updater::nThreads.

◆ setMaximumIterations()

void updater::setMaximumIterations ( int maxIt )

inlineinherited

Set the maximum number of iterations before terminating (or set to -1 to ignore)

References updater::maxIterations.

Referenced by energyMinimizerNesterovAG::initializeParameters(), energyMinimizerLoLBFGS::initializeParameters(), initializeParameters(), and energyMinimizerFIRE::initializeParameters().

◆ getCurrentIterations()

int updater::getCurrentIterations ( )

inlineinherited

References updater::iterations.

◆ getMaxIterations()

int updater::getMaxIterations ( )

inlineinherited

References updater::maxIterations.

◆ setCurrentIterations()

void updater::setCurrentIterations ( int newIterations )

inlineinherited

References updater::iterations.

◆ getNTotal()

int updater::getNTotal ( )

inherited

communicate the number of non-object sites across ranks

References access_location::host, updater::model, updater::Ndof, updater::nTotal, access_mode::read, updater::sim, and updater::updaterData.

Member Data Documentation

◆ forceMax

scalar energyMinimizerGradientDescent::forceMax

protected

sqrt(force.force) / N_{dof}

Referenced by getMaxForce(), gradientDescentCPU(), gradientDescentGPU(), and minimize().

◆ forceCutoff

scalar energyMinimizerGradientDescent::forceCutoff

protected

The cutoff value of the maximum force.

Referenced by minimize(), setForceCutoff(), and setGradientDescentParameters().

◆ dotProductTuner

shared_ptr<kernelTuner> energyMinimizerGradientDescent::dotProductTuner

protected

kernel tuner for performance

Referenced by initializeParameters().

◆ sumReductionIntermediate

GPUArray<scalar> energyMinimizerGradientDescent::sumReductionIntermediate

protected

Utility array for simple reductions.

Referenced by getClassSize(), gradientDescentGPU(), and initializeFromModel().

◆ sumReductionIntermediate2

GPUArray<scalar> energyMinimizerGradientDescent::sumReductionIntermediate2

protected

Referenced by getClassSize(), gradientDescentGPU(), and initializeFromModel().

◆ displacement

GPUArray<dVec> equationOfMotion::displacement

inherited

an array of displacements

Referenced by energyMinimizerAdam::adamStepCPU(), energyMinimizerAdam::adamStepGPU(), equationOfMotion::getClassSize(), energyMinimizerAdam::initializeFromModel(), equationOfMotion::initializeFromModel(), initializeFromModel(), energyMinimizerFIRE::initializeFromModel(), velocityVerlet::integrateEOMCPU(), and velocityVerlet::integrateEOMGPU().

◆ sim

shared_ptr<basicSimulation> updater::sim

inherited

◆ model

shared_ptr<simpleModel> updater::model

inherited

A pointer to a simpleModel that the updater acts on.

Referenced by energyMinimizerAdam::adamStepCPU(), energyMinimizerAdam::adamStepGPU(), energyMinimizerFIRE::fireStepCPU(), energyMinimizerFIRE::fireStepGPU(), updater::getNTotal(), gradientDescentCPU(), gradientDescentGPU(), energyMinimizerAdam::initializeFromModel(), energyMinimizerNesterovAG::initializeFromModel(), energyMinimizerLoLBFGS::initializeFromModel(), equationOfMotion::initializeFromModel(), initializeFromModel(), energyMinimizerFIRE::initializeFromModel(), updater::initializeFromModel(), velocityVerlet::integrateEOMCPU(), velocityVerlet::integrateEOMGPU(), equationOfMotion::integrateEquationOfMotion(), energyMinimizerLoLBFGS::lineSearchCPU(), energyMinimizerLoLBFGS::LoLBFGSStepCPU(), energyMinimizerLoLBFGS::LoLBFGSStepGPU(), energyMinimizerAdam::minimize(), energyMinimizerNesterovAG::minimize(), energyMinimizerLoLBFGS::minimize(), minimize(), energyMinimizerFIRE::minimize(), energyMinimizerNesterovAG::nesterovStepCPU(), energyMinimizerNesterovAG::nesterovStepGPU(), and updater::setModel().

◆ updaterData

vector<scalar> updater::updaterData

inherited

Referenced by energyMinimizerFIRE::fireStepCPU(), energyMinimizerFIRE::fireStepGPU(), updater::getNTotal(), gradientDescentCPU(), gradientDescentGPU(), initializeParameters(), and energyMinimizerFIRE::initializeParameters().

◆ iterations

int updater::iterations

inherited

The number of iterations performed.

Referenced by energyMinimizerFIRE::fireStepCPU(), energyMinimizerFIRE::fireStepGPU(), updater::getCurrentIterations(), energyMinimizerAdam::initializeFromModel(), energyMinimizerNesterovAG::initializeParameters(), energyMinimizerLoLBFGS::initializeParameters(), initializeParameters(), energyMinimizerFIRE::initializeParameters(), energyMinimizerLoLBFGS::LoLBFGSStepCPU(), energyMinimizerLoLBFGS::LoLBFGSStepGPU(), energyMinimizerAdam::minimize(), energyMinimizerNesterovAG::minimize(), energyMinimizerLoLBFGS::minimize(), minimize(), energyMinimizerFIRE::minimize(), and updater::setCurrentIterations().

◆ nThreads

int updater::nThreads =1

protectedinherited

number of threads to use

Referenced by updater::setNThreads().

◆ Period

int updater::Period

protectedinherited

The period of the updater... the updater will work every Period timesteps.

Referenced by updater::setPeriod(), updater::Update(), and updater::updater().

◆ Phase

int updater::Phase

protectedinherited

The phase of the updater... the updater will work every Period timesteps offset by a phase.

Referenced by updater::setPhase(), updater::Update(), and updater::updater().

◆ useGPU

bool updater::useGPU

protectedinherited

whether the updater does its work on the GPU or not

Referenced by energyMinimizerFIRE::fireStep(), gradientDescentStep(), equationOfMotion::integrateEquationOfMotion(), energyMinimizerAdam::minimize(), energyMinimizerNesterovAG::minimize(), energyMinimizerLoLBFGS::minimize(), updater::setGPU(), and updater::updater().

◆ neverGPU

bool updater::neverGPU

protectedinherited

whether the updater never does work on the GPU

Referenced by equationOfMotion::initializeFromModel(), initializeFromModel(), and energyMinimizerFIRE::initializeFromModel().

◆ Ndof

int updater::Ndof

protectedinherited

some measure of the number of degrees of freedom the equations of motion might need to know about locally

Referenced by energyMinimizerAdam::adamStepCPU(), energyMinimizerAdam::adamStepGPU(), energyMinimizerFIRE::fireStepCPU(), energyMinimizerFIRE::fireStepGPU(), updater::getNdof(), updater::getNTotal(), gradientDescentCPU(), gradientDescentGPU(), energyMinimizerAdam::initializeFromModel(), energyMinimizerNesterovAG::initializeFromModel(), energyMinimizerLoLBFGS::initializeFromModel(), equationOfMotion::initializeFromModel(), initializeFromModel(), energyMinimizerFIRE::initializeFromModel(), updater::initializeFromModel(), energyMinimizerFIRE::initializeParameters(), velocityVerlet::integrateEOMCPU(), velocityVerlet::integrateEOMGPU(), equationOfMotion::integrateEquationOfMotion(), energyMinimizerLoLBFGS::lineSearchCPU(), energyMinimizerLoLBFGS::LoLBFGSStepCPU(), energyMinimizerLoLBFGS::LoLBFGSStepGPU(), energyMinimizerAdam::minimize(), energyMinimizerNesterovAG::minimize(), energyMinimizerLoLBFGS::minimize(), minimize(), energyMinimizerFIRE::minimize(), energyMinimizerNesterovAG::nesterovStepCPU(), energyMinimizerNesterovAG::nesterovStepGPU(), and updater::setNdof().

◆ nTotal

int updater::nTotal

protectedinherited

the total number of non-object sites across all ranks

Referenced by energyMinimizerFIRE::fireStepCPU(), energyMinimizerFIRE::fireStepGPU(), updater::getNTotal(), gradientDescentCPU(), gradientDescentGPU(), and energyMinimizerFIRE::initializeParameters().

◆ reproducible

bool updater::reproducible

protectedinherited

whether the RNGs give reproducible results

Referenced by updater::setReproducible(), and updater::updater().

◆ deltaT

scalar updater::deltaT

protectedinherited

The internal time step size.

Referenced by energyMinimizerAdam::adamStepCPU(), energyMinimizerAdam::adamStepGPU(), energyMinimizerFIRE::fireStepCPU(), energyMinimizerFIRE::fireStepGPU(), gradientDescentCPU(), gradientDescentGPU(), velocityVerlet::integrateEOMCPU(), velocityVerlet::integrateEOMGPU(), energyMinimizerLoLBFGS::minimize(), energyMinimizerFIRE::minimize(), energyMinimizerNesterovAG::nesterovStepCPU(), energyMinimizerNesterovAG::nesterovStepGPU(), energyMinimizerAdam::setAdamParameters(), energyMinimizerFIRE::setDeltaT(), updater::setDeltaT(), energyMinimizerFIRE::setFIREParameters(), setGradientDescentParameters(), energyMinimizerLoLBFGS::setLoLBFGSParameters(), and energyMinimizerNesterovAG::setNesterovAGParameters().

◆ maxIterations

int updater::maxIterations

protectedinherited

The maximum number of iterations allowed.

Referenced by updater::getMaxIterations(), energyMinimizerAdam::minimize(), energyMinimizerNesterovAG::minimize(), energyMinimizerLoLBFGS::minimize(), and updater::setMaximumIterations().

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

Public Member Functions

Public Attributes

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ energyMinimizerGradientDescent() [1/2]

◆ energyMinimizerGradientDescent() [2/2]

Member Function Documentation

◆ initializeParameters()

◆ initializeFromModel()

◆ setGradientDescentParameters()

◆ setForceCutoff()

◆ gradientDescentStep()

◆ gradientDescentCPU()

◆ gradientDescentGPU()

◆ minimize()

◆ performUpdate()

◆ getMaxForce()

◆ getClassSize()

◆ integrateEquationOfMotion()

◆ integrateEOMGPU()

◆ integrateEOMCPU()

◆ Update()

◆ setSimulation()

◆ setModel()

◆ setPeriod()

◆ setPhase()

◆ spatialSorting()

◆ setReproducible()

◆ setGPU()

◆ getNdof()

◆ setNdof()

◆ setDeltaT()

◆ setNThreads()

◆ setMaximumIterations()

◆ getCurrentIterations()

◆ getMaxIterations()

◆ setCurrentIterations()

◆ getNTotal()

Member Data Documentation

◆ forceMax

◆ forceCutoff

◆ dotProductTuner

◆ sumReductionIntermediate

◆ sumReductionIntermediate2

◆ displacement

◆ sim

◆ model

◆ updaterData

◆ iterations

◆ nThreads

◆ Period

◆ Phase

◆ useGPU

◆ neverGPU

◆ Ndof

◆ nTotal

◆ reproducible

◆ deltaT

◆ maxIterations