Description

An iterative solver based on modified version of ADMM Alternating Direction Method of Multipliers.

See ChSystemDescriptor for more information about the problem formulation and the data structures passed to the solver.

#include <ChSolverADMM.h>

Inheritance diagram for chrono::ChSolverADMM:

[legend]

Collaboration diagram for chrono::ChSolverADMM:

[legend]

Public Types
enum	AdmmStepType { NONE = 0, BALANCED_UNSCALED, BALANCED_FAST, BALANCED_RANGE }
	Types of step adaption policies for the ADMM solver.

enum	AdmmAcceleration { BASIC = 0, NESTEROV }
	Types of ADMM enhancements.

Public Types inherited from chrono::ChSolver
enum	Type { Type::PSOR, Type::PSSOR, Type::PJACOBI, Type::PMINRES, Type::BARZILAIBORWEIN, Type::APGD, Type::ADMM, Type::SPARSE_LU, Type::SPARSE_QR, Type::PARDISO_MKL, Type::MUMPS, Type::GMRES, Type::MINRES, Type::BICGSTAB, CUSTOM }
	Available types of solvers. More...

Public Member Functions
	ChSolverADMM ()
	Default constructor: uses the SparseQR direct solver from Eigine, slow and UNOPTIMAL. More...

	ChSolverADMM (std::shared_ptr< ChDirectSolverLS > my_LS_engine)
	Constructor where you can pass a better direct solver than the default SparseQR solver. More...

virtual Type	GetType () const override
	Return type of the solver.

virtual double	Solve (ChSystemDescriptor &sysd) override
	Performs the solution of the problem. More...

void	SetDiagonalPreconditioner (bool mp)
	Enable the diagonal preconditioner.

bool	GetDiagonalPreconditioner ()

void	SetRho (double mr)
	Set the initial ADMM step. Could change later if adaptive step is used.

double	GetRho ()

void	SetRhoBilaterals (double mr)
	Set the ADMM step for bilateral constraints only.

double	GetRhoBilaterals ()

void	SetSigma (double mr)
	Set the sigma regularization.

double	GetSigma ()

void	SetStepAdjustEach (int mr)
	Set how frequently the rho step is adjusted.

int	GetStepAdjustEach ()

void	SetStepAdjustThreshold (double mr)
	Set the step adjust threshold T, ex. More...

double	GetStepAdjustThreshold ()

void	SetStepAdjustMaxfactor (double mr)
	Set the step adjust max factor F, if new step scaling is higher than F it is clamped to F, if lower than 1/F it is clamped to F, to avoid exploding results.

double	GetStepAdjustMaxfactor ()

void	SetStepAdjustPolicy (AdmmStepType mr)
	Set the step adjust max factor F, if new step scaling is higher than F it is clamped to F, if lower than 1/F it is clamped to F, to avoid exploding results.

AdmmStepType	GetStepAdjustPolicy ()

void	SetAcceleration (AdmmAcceleration mr)
	Set the type of ADMM iteration, enabling acceleration variants.

AdmmAcceleration	GetAcceleration ()

void	SetTolerancePrimal (double mr)
	Set the absolute tolerance for the primal residual (force impulses error), if the iteration falls below this and the dual tolerance, it stops iterating.

double	GetTolerancePrimal ()

void	SetToleranceDual (double mr)
	Set the absolute tolerance for the dual residual (constraint speed error), if the iteration falls below this and the primal tolerance, it stops iterating.

double	GetToleranceDual ()

double	GetErrorPrimal () const
	Return the primal residual (force impulses error) reached during the last solve.

double	GetErrorDual () const
	Return the dual residual (constraint speed error) reached during the last solve.

virtual double	GetError () const override
	Return the number of iterations performed during the last solve. More...

virtual void	ArchiveOut (ChArchiveOut &archive_out) override
	Method to allow serialization of transient data to archives.

virtual void	ArchiveIn (ChArchiveIn &archive_in) override
	Method to allow de serialization of transient data from archives.

Public Member Functions inherited from chrono::ChIterativeSolverVI
void	SetOmega (double mval)
	Set the overrelaxation factor (default: 1.0). More...

void	SetSharpnessLambda (double mval)
	Set the sharpness factor (default: 1.0). More...

virtual void	SetMaxIterations (int max_iterations) override
	Set the maximum number of iterations.

void	SetRecordViolation (bool mval)
	Enable/disable recording of the constraint violation history. More...

double	GetOmega () const
	Return the current value of the overrelaxation factor.

double	GetSharpnessLambda () const
	Return the current value of the sharpness factor.

virtual int	GetIterations () const override
	Return the number of iterations performed during the last solve.

const std::vector< double > &	GetViolationHistory () const
	Access the vector of constraint violation history. More...

const std::vector< double > &	GetDeltalambdaHistory () const
	Access the vector with history of maximum change in Lagrange multipliers Note that collection of constraint violations must be enabled through SetRecordViolation.

Public Member Functions inherited from chrono::ChIterativeSolver
void	SetTolerance (double tolerance)
	Set the tolerance threshold used by the stopping criteria.

void	EnableDiagonalPreconditioner (bool val)
	Enable/disable use of a simple diagonal preconditioner (default: true). More...

void	EnableWarmStart (bool val)
	Enable/disable warm starting by providing an initial guess (default: false). More...

int	GetMaxIterations () const
	Get the current maximum number of iterations.

double	GetTolerance () const
	Get the current tolerance value.

Public Member Functions inherited from chrono::ChSolver
virtual ChDirectSolverLS *	AsDirect ()
	Downcast to ChDirectSolver.

virtual bool	Setup (ChSystemDescriptor &sysd)
	This function does the setup operations for the solver. More...

void	SetVerbose (bool mv)
	Set verbose output from solver.

void	EnableWrite (bool val, const std::string &frame, const std::string &out_dir=".")
	Enable/disable debug output of matrix, RHS, and solution vector.

Additional Inherited Members
Protected Member Functions inherited from chrono::ChIterativeSolverVI
virtual bool	IsIterative () const override
	Return true if iterative solver.

virtual bool	IsDirect () const override
	Return true if direct solver.

virtual ChIterativeSolver *	AsIterative () override
	Downcast to ChIterativeSolver.

void	AtIterationEnd (double mmaxviolation, double mdeltalambda, unsigned int iternum)
	This method MUST be called by all iterative methods INSIDE their iteration loops (at the end). More...

virtual bool	SolveRequiresMatrix () const override
	Indicate whether ot not the Solve() phase requires an up-to-date problem matrix. More...

Protected Member Functions inherited from chrono::ChIterativeSolver
	ChIterativeSolver (int max_iterations, double tolerance, bool use_precond, bool warm_start)

void	WriteMatrices (ChSystemDescriptor &sysd, bool one_indexed=true)

double	CheckSolution (ChSystemDescriptor &sysd, const ChVectorDynamic<> &x)

Protected Attributes inherited from chrono::ChIterativeSolverVI
int	m_iterations
	number of iterations performed by the solver during last call to Solve()

double	m_omega
	over-relaxation factor

double	m_shlambda
	sharpness factor

bool	record_violation_history

std::vector< double >	violation_history

std::vector< double >	dlambda_history

Protected Attributes inherited from chrono::ChIterativeSolver
bool	m_use_precond
	use diagonal preconditioning?

bool	m_warm_start
	use initial guesss?

int	m_max_iterations
	maximum number of iterations

double	m_tolerance
	tolerance threshold in stopping criteria

Protected Attributes inherited from chrono::ChSolver
bool	verbose

bool	write_matrix

std::string	output_dir

std::string	frame_id

Constructor & Destructor Documentation

◆ ChSolverADMM() [1/2]

chrono::ChSolverADMM::ChSolverADMM ( )

Default constructor: uses the SparseQR direct solver from Eigine, slow and UNOPTIMAL.

Prefer the other constructor where you can pass the MKL direct solver.

◆ ChSolverADMM() [2/2]

chrono::ChSolverADMM::ChSolverADMM ( std::shared_ptr< ChDirectSolverLS > my_LS_engine )

Constructor where you can pass a better direct solver than the default SparseQR solver.

The custom direct solver will be used in the factorization and solves in the inner loop.

Member Function Documentation

◆ GetError()

virtual double chrono::ChSolverADMM::GetError ( ) const

inlineoverridevirtual

Return the number of iterations performed during the last solve.

Return the tolerance error reached during the last solve (here refer to the ADMM dual residual)

Implements chrono::ChIterativeSolver.

◆ SetStepAdjustThreshold()

void chrono::ChSolverADMM::SetStepAdjustThreshold ( double mr )

inline

Set the step adjust threshold T, ex.

1.5; if new step scaling is in the interval [1/T, T], no adjustment is done anyway, to save CPU effort.

◆ Solve()

double chrono::ChSolverADMM::Solve ( ChSystemDescriptor & sysd )

overridevirtual

Performs the solution of the problem.

Returns: the maximum constraint violation after termination, as dual (speed) residual

Parameters

sysd	system description with constraints and variables

Implements chrono::ChSolver.

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

/builds/uwsbel/chrono/src/chrono/solver/ChSolverADMM.h
/builds/uwsbel/chrono/src/chrono/solver/ChSolverADMM.cpp