Implementation of TimeStepper which advances an ItoSolver. More...

#include <CD_BrownianWalkerStepper.H>

Inheritance diagram for Physics::BrownianWalker::BrownianWalkerStepper:

Collaboration diagram for Physics::BrownianWalker::BrownianWalkerStepper:

[legend]

Public Member Functions
	BrownianWalkerStepper ()
	Default constructor – parses basic options.

	BrownianWalkerStepper (RefCountedPtr< ItoSolver > &a_solver)
	Full constructor. Calls the basic constructor and sets solver.

virtual	~BrownianWalkerStepper ()
	Destructor (does nothing)

void	initialData () override
	Fill solver with initial data.

void	allocate () override
	Allocate storage for solvers and TimeStepper.

void	postInitialize () override
	Perform post-initialization routines. This does nothing.

bool	loadBalanceThisRealm (const std::string a_realm) const override
	Check if the input realm should be load balanced. More...

Vector< long int >	getCheckpointLoads (const std::string a_realm, const int a_level) const override
	Get computational loads to be checkpointed. This is used by Driver both for plotting computational loads AND for checkpointing the loads for restart purposes. More...

void	loadBalanceBoxes (Vector< Vector< int >> &a_procs, Vector< Vector< Box >> &a_boxes, const std::string a_realm, const Vector< DisjointBoxLayout > &a_grids, const int a_lmin, const int a_finestLevel) override
	Load balance grid boxes for a specific realm. More...

void	postCheckpointSetup () override
	Perform post-initialization routines. More...

int	getNumberOfPlotVariables () const override
	Get the number of plot variables for this time stepper. More...

Vector< std::string >	getPlotVariableNames () const override
	Get plot variable names.

void	writePlotData (LevelData< EBCellFAB > &a_output, int &a_icomp, const std::string a_outputRealm, const int a_level) const override
	Write plot data to output holder. More...

Real	computeDt () override
	Compute a time step to be used by Driver. More...

void	synchronizeSolverTimes (const int a_step, const Real a_time, const Real a_dt) override
	Synchronize solver times. More...

void	printStepReport () override
	Print a step report – this prints the local and global number of particles in the simulation. More...

void	registerOperators () override
	Register operators – we only need the ones the solver needs to have the solver do this.

void	registerRealms () override
	Register realms to be used in the simulation. More...

void	parseRuntimeOptions () override
	Parse runtime options.

bool	needToRegrid () override
	Always returns false.

void	preRegrid (const int a_lbase, const int a_oldFinestLevel) override
	Perform pre-regrid operations. More...

void	setupSolvers () override
	Set up the Ito solver. This instantiates the solver and parses solver options.

Real	advance (const Real a_dt) override
	Advancement method. More...

void	regrid (const int a_lmin, const int a_oldFinestLevel, const int a_newFinestLevel) override
	Regrid method. This just regrids the solver. More...

void	postRegrid () override
	Post-regrid operations.

Public Member Functions inherited from TimeStepper
	TimeStepper ()
	Default constructor (does nothing)

virtual	~TimeStepper ()
	Default destructor (does nothing)

void	setAmr (const RefCountedPtr< AmrMesh > &a_amr)
	Set AmrMesh. More...

void	setComputationalGeometry (const RefCountedPtr< ComputationalGeometry > &a_computationalGeometry)
	Set the computational geometry. More...

virtual void	prePlot ()
	An option for calling special functions prior to plotting data. Called by Driver in the IMMEDIATELY before writing the plot file.

virtual void	postPlot ()
	An option for calling special functions prior to plotting data. Called by Driver in the IMMEDIATELY after writing the plot file.

Protected Types
enum	LoadBalancingMethod { Mesh , Particle }
	For classifying load balancing methods.

Protected Member Functions
void	setAdvectionDiffusion ()
	Set advection and diffusion fields.

void	setVelocity ()
	Set the Brownian walker advection field.

void	setDiffusion ()
	Set the ItoSolver "diffusion field".

void	makeSuperParticles ()
	Make super particles.

void	loadBalanceBoxesMesh (Vector< Vector< int >> &a_procs, Vector< Vector< Box >> &a_boxes, const std::string a_realm, const Vector< DisjointBoxLayout > &a_grids, const int a_lmin, const int a_finestLevel)
	Method that shows how to load balance the application using the number of particles per cell stored on the mesh. More...

void	loadBalanceBoxesParticles (Vector< Vector< int >> &a_procs, Vector< Vector< Box >> &a_boxes, const std::string a_realm, const Vector< DisjointBoxLayout > &a_grids, const int a_lmin, const int a_finestLevel)
	Method that shows how to load balance the application using the number of particles per cell stored on the mesh. More...

Protected Attributes
LoadBalancingMethod	m_whichLoadBalance
	Load balancing method.

std::string	m_realm
	Realm where the particles live.

phase::which_phase	m_phase
	Phase where the particles live.

bool	m_loadBalance
	If true, application will be load balanced.

RefCountedPtr< ItoSolver >	m_solver
	Reference to underlying solver.

RefCountedPtr< ItoSpecies >	m_species
	Reference to underlying species.

EBAMRCellData	m_regridPPC
	Number of particles per cell. Used only during regrids.

Real	m_diffCo
	Fluid diffusion coefficient.

Real	m_mobility
	Particle mobility coefficients.

Real	m_omega
	Angular velocity for advection field.

int	m_ppc
	Number of particles per cell.

Real	m_cfl
	Effective CFL number for particle hops.

Protected Attributes inherited from TimeStepper
int	m_verbosity
	Class verbosity.

int	m_timeStep
	Time step.

Real	m_time
	TIme.

Real	m_dt
	Previous time step size.

RefCountedPtr< AmrMesh >	m_amr
	AmrMesh.

RefCountedPtr< ComputationalGeometry >	m_computationalGeometry
	Computational geometry.

Detailed Description

Implementation of TimeStepper which advances an ItoSolver.

This class is fairly simple – we want to advance particles using an Ito kernel dX = V*dt + sqrt(2*D*dt)*N0 where N0 is a Gaussian random number. The particle velocity in the ItoSolver is V = mu*v(Xp) where mu is a particle mobility and v is a "velocity field". The ItoSolver has various way of representing the velocity, but here we consider a constant mobility. The diffusion coefficient is also constant (the value is fetched from the input script).

Note that while the ItoSolver CAN interpolate the particle diffusion coefficient from a mesh quantity – this physics module sets it directly.

This physics module shows how to load balance an application introspectively based on the number of particles assigned to each rank. We support two methods, either by estimating the number of particles on the new mesh, or counting them directly.

Member Function Documentation

◆ advance()

Real BrownianWalkerStepper::advance ( const Real a_dt )

overridevirtual

Advancement method.

Returns: a_dt

Implements TimeStepper.

◆ computeDt()

Real BrownianWalkerStepper::computeDt ( )

overridevirtual

Compute a time step to be used by Driver.

This returns m_cfl*ItoSolver::computeDt (i.e. m_cfl is a CFL-like number)

Implements TimeStepper.

◆ getCheckpointLoads()

Vector< long int > BrownianWalkerStepper::getCheckpointLoads	(	const std::string	a_realm,
		const int	a_level
	)		const

overridevirtual

Get computational loads to be checkpointed. This is used by Driver both for plotting computational loads AND for checkpointing the loads for restart purposes.

This uses the number of computational particles as a load metric.

Parameters

[in]	a_realm	Realm
[in]	a_level	Grid level

Reimplemented from TimeStepper.

◆ getNumberOfPlotVariables()

int BrownianWalkerStepper::getNumberOfPlotVariables ( ) const

overridevirtual

Get the number of plot variables for this time stepper.

Only the solver plots to file so this just returns the number of plot variables specified for the solver

Implements TimeStepper.

◆ loadBalanceBoxes()

void BrownianWalkerStepper::loadBalanceBoxes	(	Vector< Vector< int >> &	a_procs,
		Vector< Vector< Box >> &	a_boxes,
		const std::string	a_realm,
		const Vector< DisjointBoxLayout > &	a_grids,
		const int	a_lmin,
		const int	a_finestLevel
	)

overridevirtual

Load balance grid boxes for a specific realm.

This will count the number of particles in each box and then load balance them.

Parameters

[out]	a_procs	MPI ranks owning the various grid boxes.
[out]	a_boxes	Grid boxes on every level (obtain them with a_grids[lvl].boxArray())
[in]	a_realm	Realm identifier
[in]	a_grids	Original grids
[in]	a_lmin	Coarsest grid level that changed
[in]	a_finestLevel	New finest grid level

Note: This should only be called if a_realm==m_realm and m_loadBalance==true

Reimplemented from TimeStepper.

◆ loadBalanceBoxesMesh()

void BrownianWalkerStepper::loadBalanceBoxesMesh	(	Vector< Vector< int >> &	a_procs,
		Vector< Vector< Box >> &	a_boxes,
		const std::string	a_realm,
		const Vector< DisjointBoxLayout > &	a_grids,
		const int	a_lmin,
		const int	a_finestLevel
	)

protected

Method that shows how to load balance the application using the number of particles per cell stored on the mesh.

This routine will load balance based on an estimated number of particles per cell. This number is computed ON THE MESH, i.e. it does not directly count the particles. Rather, we happen to know that ItoSolver deposited his particle numbers to the mesh before in the preRegrid, and we use that information to estimate the number of particles per patch in the new grids.

This is different from the other method which assign particles to the new grids and then counts them. Because this routine does not remap particles (which can be expensive), this routine is typically faster than loadBalanceBoxesParticles. loadBalanceBoxesParticles

Parameters

[out]	a_procs	MPI ranks owning the various grid boxes.
[out]	a_boxes	Grid boxes on every level (obtain them with a_grids[lvl].boxArray())
[in]	a_realm	Realm identifier
[in]	a_grids	Original grids
[in]	a_lmin	Coarsest grid level that changed
[in]	a_finestLevel	New finest grid level

◆ loadBalanceBoxesParticles()

void BrownianWalkerStepper::loadBalanceBoxesParticles	(	Vector< Vector< int >> &	a_procs,
		Vector< Vector< Box >> &	a_boxes,
		const std::string	a_realm,
		const Vector< DisjointBoxLayout > &	a_grids,
		const int	a_lmin,
		const int	a_finestLevel
	)

protected

Method that shows how to load balance the application using the number of particles per cell stored on the mesh.

This routine will load balance by directly counting the particles on the new mesh. The result of that is that we regrid a particle data holder twice.

Because this routine ends up remapping particles twice (once here, and once in BrownianWalkerStepper::regrid), this routine is usually slower than loadBalanceBoxesMesh when used at large particle counts. In addition, loadBalanceBoxesMesh does it's best at estimating the final number of computatational particles after particle merging/splitting, while this routine only counts the original computational particles before the regrid.

Parameters

[out]	a_procs	MPI ranks owning the various grid boxes.
[out]	a_boxes	Grid boxes on every level (obtain them with a_grids[lvl].boxArray())
[in]	a_realm	Realm identifier
[in]	a_grids	Original grids
[in]	a_lmin	Coarsest grid level that changed
[in]	a_finestLevel	New finest grid level

◆ loadBalanceThisRealm()

bool BrownianWalkerStepper::loadBalanceThisRealm ( const std::string a_realm ) const

overridevirtual

Check if the input realm should be load balanced.

Parameters

[in] a_realm Realm to be load balanced.

This returns true if a_realm==m_realm and m_loadBalance==true.

Reimplemented from TimeStepper.

◆ postCheckpointSetup()

void BrownianWalkerStepper::postCheckpointSetup ( )

overridevirtual

Perform post-initialization routines.

This remaps the particles and makes new superparticles. It also sets the velocity fields.

Implements TimeStepper.

◆ preRegrid()

void BrownianWalkerStepper::preRegrid	(	const int	a_lbase,
		const int	a_oldFinestLevel
	)

overridevirtual

Perform pre-regrid operations.

Parameters

[in]	a_lmin	The coarsest level that changes
[in]	a_oldFinestLevel	The finest level before the regrid.

Implements TimeStepper.

◆ printStepReport()

void BrownianWalkerStepper::printStepReport ( )

overridevirtual

Print a step report – this prints the local and global number of particles in the simulation.

This is called by Driver after time step.

Implements TimeStepper.

◆ registerRealms()

void BrownianWalkerStepper::registerRealms ( )

overridevirtual

Register realms to be used in the simulation.

The primal realm is always register – but you can also register a separate particle realm.

Implements TimeStepper.

◆ regrid()

void BrownianWalkerStepper::regrid	(	const int	a_lmin,
		const int	a_oldFinestLevel,
		const int	a_newFinestLevel
	)

overridevirtual

Regrid method. This just regrids the solver.

Parameters

[in]	a_lmin	The coarsest level that changed.
[in]	a_oldFinestLevel	The finest level before the regrid.
[in]	a_newFinestLevel	The finest level after the regrid.

Implements TimeStepper.

◆ synchronizeSolverTimes()

void BrownianWalkerStepper::synchronizeSolverTimes	(	const int	a_step,
		const Real	a_time,
		const Real	a_dt
	)

overridevirtual

Synchronize solver times.

Parameters

[in]	a_step	Time step
[in]	a_time	Time (in seconds)
[in]	a_dt	Time step that was used.

Implements TimeStepper.

◆ writePlotData()

void BrownianWalkerStepper::writePlotData	(	LevelData< EBCellFAB > &	a_output,
		int &	a_icomp,
		const std::string	a_outputRealm,
		const int	a_level
	)		const

overridevirtual

Write plot data to output holder.

Parameters

[in,out]	a_output	Output data holder.
[in,out]	a_icomp	Starting component in a_output to begin at.
[in]	a_outputRealm	Realm where a_output belongs
[in]	a_level	Grid level

Implements TimeStepper.

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

Physics/BrownianWalker/CD_BrownianWalkerStepper.H
Physics/BrownianWalker/CD_BrownianWalkerStepper.cpp

Public Member Functions

Protected Types

Protected Member Functions

Protected Attributes

Detailed Description

Member Function Documentation

◆ advance()

◆ computeDt()

◆ getCheckpointLoads()

◆ getNumberOfPlotVariables()

◆ loadBalanceBoxes()

◆ loadBalanceBoxesMesh()

◆ loadBalanceBoxesParticles()

◆ loadBalanceThisRealm()

◆ postCheckpointSetup()

◆ preRegrid()

◆ printStepReport()

◆ registerRealms()

◆ regrid()

◆ synchronizeSolverTimes()

◆ writePlotData()