#include <JacobiPC.H>

Inheritance diagram for JacobiPC< T, Ops >:

Public Types
using	RT = typename T::value_type

Public Types inherited from Preconditioner< T, Ops >
using	RT = typename T::value_type

Public Member Functions
	JacobiPC ()=default

	~JacobiPC () override=default

	JacobiPC (const JacobiPC &)=delete

JacobiPC &	operator= (const JacobiPC &)=delete

	JacobiPC (JacobiPC &&) noexcept=delete

JacobiPC &	operator= (JacobiPC &&) noexcept=delete

void	Define (const T &, Ops *) override
	Define the preconditioner.

void	Update (const T &a_U) override
	Update the preconditioner.

void	Apply (T &, const T &) override
	Solve (I + M) x = b via damped Jacobi iteration.

void	printParameters () const override
	Print parameters.

bool	IsDefined () const override
	Check if the nonlinear solver has been defined.

RT	computeResidualNorm (const T &a_x, const T &a_b) const
	Compute \|\|b - (I+M)*x\|\| for residual monitoring. Uses the ghost-padded m_x_ghost (which must already be populated and boundary-filled) for stencil neighbor access. For diagonal-only mass matrices (ncomp=1), uses the fast diagonal formula from x directly.

Public Member Functions inherited from Preconditioner< T, Ops >
	Preconditioner ()=default
	Default constructor.

virtual	~Preconditioner ()=default
	Default destructor.

	Preconditioner (const Preconditioner &)=default

Preconditioner &	operator= (const Preconditioner &)=default

	Preconditioner (Preconditioner &&) noexcept=default

Preconditioner &	operator= (Preconditioner &&) noexcept=default

virtual void	getPCMatrix (amrex::Gpu::DeviceVector< int > &, amrex::Gpu::DeviceVector< int > &, amrex::Gpu::DeviceVector< int > &, amrex::Gpu::DeviceVector< RT > &, int &, int &)
	Get the sparse matrix form of the preconditioner.

virtual void	setName (const std::string &)
	Set the name for screen output and parsing inputs.

void	CurTime (const RT a_time)
	Set the current time.

void	CurTimeStep (const RT a_dt)
	Set the current time step size.

Protected Types
using	MFArr = amrex::Array<amrex::MultiFab,3>

Protected Member Functions
void	readParameters ()

void	copyToGhostAndFill (const amrex::Vector< amrex::Array< amrex::MultiFab *, 3 > > &a_x_mfarrvec)

Protected Attributes
bool	m_is_defined = false

bool	m_verbose = false

int	m_max_iter = 200

RT	m_omega = RT(1.0)

bool	m_adaptive_omega = true

RT	m_atol = 1.0e-16

RT	m_rtol = 1.0e-4

Ops *	m_ops = nullptr

int	m_num_amr_levels = 0

const amrex::Vector< amrex::Array< amrex::MultiFab , 3 > >	m_bcoefs = nullptr

bool	m_has_offdiag = false

bool	m_work_defined = false

amrex::Vector< amrex::Array< amrex::MultiFab, 3 > >	m_work

amrex::Vector< amrex::Array< amrex::MultiFab, 3 > >	m_x_ghost

int	m_stencil_width = 0

Protected Attributes inherited from Preconditioner< T, Ops >
RT	m_time = 0.0

RT	m_dt = 0.0

Detailed Description

template<class T, class Ops>
class JacobiPC< T, Ops >

Point-Jacobi Preconditioner.

Approximately solves (I + M) x = b using damped Jacobi iterations, where M is the scaled mass matrix (plasma response dJ/dE).

When the mass matrix is purely diagonal (mass_matrices_pc_width=0), the solve is exact in one step: x = b / (1 + M_diag). When the mass matrix has off-diagonal stencil entries, damped Jacobi iterations are performed.

Adaptive omega (enabled by default): the relaxation parameter omega starts at 1.0 and is automatically reduced (by a factor of 0.9) when the Jacobi residual increases, indicating that the current omega is too large for the spectral properties of the mass matrix. The adapted omega is cached for subsequent Apply() calls. This handles marginally diagonally dominant or non-symmetric mass matrices. When the user explicitly sets pc_jacobi.omega, adaptive behavior is disabled (can be re-enabled with pc_jacobi.adaptive_omega = true).

Matrix Elements Layout: The mass matrix MultiFab stores stencil coefficients as components. For a field direction dim, the per-spatial-direction component counts are given by GetMassMatricesPCnComp(dim, d). The total number of components is the product over spatial dimensions. The diagonal component is at index (nComp-1)/2. The stencil offset for component c is decoded from the component index using the per-direction counts.

Note: WarpXSolverVec MultiFabs have zero ghost cells. This class allocates its own ghost-padded temporaries (m_x_ghost) for stencil neighbor access. All FillBoundary calls operate on these temporaries.

Member Typedef Documentation

◆ MFArr

template<class T, class Ops>

using JacobiPC< T, Ops >::MFArr = amrex::Array<amrex::MultiFab,3>

protected

◆ RT

template<class T, class Ops>

using JacobiPC< T, Ops >::RT = typename T::value_type

Constructor & Destructor Documentation

◆ JacobiPC() [1/3]

template<class T, class Ops>

JacobiPC< T, Ops >::JacobiPC ( )

default

◆ ~JacobiPC()

template<class T, class Ops>

JacobiPC< T, Ops >::~JacobiPC ( )

overridedefault

◆ JacobiPC() [2/3]

template<class T, class Ops>

JacobiPC< T, Ops >::JacobiPC ( const JacobiPC< T, Ops > & )

delete

◆ JacobiPC() [3/3]

template<class T, class Ops>

JacobiPC< T, Ops >::JacobiPC ( JacobiPC< T, Ops > && )

deletenoexcept

Member Function Documentation

◆ Apply()

template<class T, class Ops>

void JacobiPC< T, Ops >::Apply	(	T &	a_x,
		const T &	a_b )

overridevirtual

Solve (I + M) x = b via damped Jacobi iteration.

Implements Preconditioner< T, Ops >.

◆ computeResidualNorm()

template<class T, class Ops>

auto JacobiPC< T, Ops >::computeResidualNorm	(	const T &	a_x,
		const T &	a_b ) const

Compute ||b - (I+M)*x|| for residual monitoring. Uses the ghost-padded m_x_ghost (which must already be populated and boundary-filled) for stencil neighbor access. For diagonal-only mass matrices (ncomp=1), uses the fast diagonal formula from x directly.

◆ copyToGhostAndFill()

template<class T, class Ops>

void JacobiPC< T, Ops >::copyToGhostAndFill ( const amrex::Vector< amrex::Array< amrex::MultiFab *, 3 > > & a_x_mfarrvec )

protected

◆ Define()

template<class T, class Ops>

void JacobiPC< T, Ops >::Define	(	const T &	,
		Ops *	)

overridevirtual

Define the preconditioner.

Implements Preconditioner< T, Ops >.

◆ IsDefined()

template<class T, class Ops>

bool JacobiPC< T, Ops >::IsDefined ( ) const

inlinenodiscardoverridevirtual

Check if the nonlinear solver has been defined.

Implements Preconditioner< T, Ops >.

◆ operator=() [1/2]

template<class T, class Ops>

JacobiPC & JacobiPC< T, Ops >::operator= ( const JacobiPC< T, Ops > & )

delete

◆ operator=() [2/2]

template<class T, class Ops>

JacobiPC & JacobiPC< T, Ops >::operator= ( JacobiPC< T, Ops > && )

deletenoexcept

◆ printParameters()

template<class T, class Ops>

void JacobiPC< T, Ops >::printParameters ( ) const

overridevirtual

Print parameters.

Reimplemented from Preconditioner< T, Ops >.

◆ readParameters()

template<class T, class Ops>

void JacobiPC< T, Ops >::readParameters ( )

protected

◆ Update()

template<class T, class Ops>

void JacobiPC< T, Ops >::Update ( const T & a_U )

overridevirtual

Update the preconditioner.

Implements Preconditioner< T, Ops >.

Member Data Documentation

◆ m_adaptive_omega

template<class T, class Ops>

bool JacobiPC< T, Ops >::m_adaptive_omega = true

protected

◆ m_atol

template<class T, class Ops>

RT JacobiPC< T, Ops >::m_atol = 1.0e-16

protected

◆ m_bcoefs

template<class T, class Ops>

const amrex::Vector<amrex::Array<amrex::MultiFab*,3> >* JacobiPC< T, Ops >::m_bcoefs = nullptr

protected

◆ m_has_offdiag

template<class T, class Ops>

bool JacobiPC< T, Ops >::m_has_offdiag = false

protected

◆ m_is_defined

template<class T, class Ops>

bool JacobiPC< T, Ops >::m_is_defined = false

protected

◆ m_max_iter

template<class T, class Ops>

int JacobiPC< T, Ops >::m_max_iter = 200

protected

◆ m_num_amr_levels

template<class T, class Ops>

int JacobiPC< T, Ops >::m_num_amr_levels = 0

protected

◆ m_omega

template<class T, class Ops>

RT JacobiPC< T, Ops >::m_omega = RT(1.0)

protected

◆ m_ops

template<class T, class Ops>

Ops* JacobiPC< T, Ops >::m_ops = nullptr

protected

◆ m_rtol

template<class T, class Ops>

RT JacobiPC< T, Ops >::m_rtol = 1.0e-4

protected

◆ m_stencil_width

template<class T, class Ops>

int JacobiPC< T, Ops >::m_stencil_width = 0

protected

◆ m_verbose

template<class T, class Ops>

bool JacobiPC< T, Ops >::m_verbose = false

protected

◆ m_work

template<class T, class Ops>

amrex::Vector<amrex::Array<amrex::MultiFab,3> > JacobiPC< T, Ops >::m_work

protected

◆ m_work_defined

template<class T, class Ops>

bool JacobiPC< T, Ops >::m_work_defined = false

protected

◆ m_x_ghost

template<class T, class Ops>

amrex::Vector<amrex::Array<amrex::MultiFab,3> > JacobiPC< T, Ops >::m_x_ghost

protected

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

/home/docs/checkouts/readthedocs.org/user_builds/warpx/checkouts/6814/Source/NonlinearSolvers/JacobiPC.H

Public Types

Public Member Functions

Protected Types

Protected Member Functions

Protected Attributes

Detailed Description

Member Typedef Documentation

◆ MFArr

◆ RT

Constructor & Destructor Documentation

◆ JacobiPC() [1/3]

◆ ~JacobiPC()

◆ JacobiPC() [2/3]

◆ JacobiPC() [3/3]

Member Function Documentation

◆ Apply()

◆ computeResidualNorm()

◆ copyToGhostAndFill()

◆ Define()

◆ IsDefined()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ printParameters()

◆ readParameters()

◆ Update()

Member Data Documentation

◆ m_adaptive_omega

◆ m_atol

◆ m_bcoefs

◆ m_has_offdiag

◆ m_is_defined

◆ m_max_iter

◆ m_num_amr_levels

◆ m_omega

◆ m_ops

◆ m_rtol

◆ m_stencil_width

◆ m_verbose

◆ m_work

◆ m_work_defined

◆ m_x_ghost