mrpt::math::CSparseMatrix::CholeskyDecomp Class Reference

Detailed Description

Auxiliary class to hold the results of a Cholesky factorization of a sparse matrix.

This implementation does not allow updating/downdating.

Usage example:

CSparseMatrix  SM(100,100);
SM.insert_entry(i,j, val); ...
SM.compressFromTriplet();
 
// Do Cholesky decomposition:
           CSparseMatrix::CholeskyDecomp  CD(SM);
CD.get_inverse();
...

Note

Only the upper triangular part of the input matrix is accessed.

This class was initially adapted from "robotvision", by Hauke Strasdat, Steven Lovegrove and Andrew J. Davison. See http://www.openslam.org/robotvision.html

This class designed to be "uncopiable".

See also

The main class: CSparseMatrix

Definition at line 440 of file CSparseMatrix.h.

#include <mrpt/math/CSparseMatrix.h>

Public Member Functions
	CholeskyDecomp (const CSparseMatrix &A)
	Constructor from a square definite-positive sparse matrix A, which can be use to solve Ax=b The actual Cholesky decomposition takes places in this constructor. More...
	CholeskyDecomp (const CholeskyDecomp &A)=delete
CholeskyDecomp &	operator= (const CholeskyDecomp &)=delete
virtual	~CholeskyDecomp ()
	Destructor. More...
CMatrixDouble	get_L () const
	Return the L matrix (L*L' = M), as a dense matrix. More...
void	get_L (CMatrixDouble &out_L) const
	Return the L matrix (L*L' = M), as a dense matrix. More...
template<class VECTOR >
VECTOR	backsub (const VECTOR &b) const
	Return the vector from a back-substitution step that solves: Ux=b. More...
void	backsub (const Eigen::VectorXd &b, Eigen::VectorXd &result_x) const
	Return the vector from a back-substitution step that solves: Ux=b. More...
void	backsub (const double *b, double *result, const size_t N) const
	overload for double pointers which assume the user has reserved the output memory for result More...
void	update (const CSparseMatrix &new_SM)
	Update the Cholesky factorization from an updated vesion of the original input, square definite-positive sparse matrix. More...

Private Attributes
css *	m_symbolic_structure
csn *	m_numeric_structure
const CSparseMatrix *	m_originalSM
	A const reference to the original matrix used to build this decomposition. More...

Constructor & Destructor Documentation

CholeskyDecomp() [1/2]

CSparseMatrix::CholeskyDecomp::CholeskyDecomp

(

const CSparseMatrix &

SM

)

Constructor from a square definite-positive sparse matrix A, which can be use to solve Ax=b The actual Cholesky decomposition takes places in this constructor.

Constructor from a square semidefinite-positive sparse matrix.

Note

Only the upper triangular part of the matrix is accessed.

Exceptions

std::runtime_error	On non-square input matrix.
mrpt::math::CExceptionNotDefPos	On non-definite-positive matrix as input.

The actual Cholesky decomposition takes places in this constructor.

Exceptions

std::runtime_error	On non-square input matrix.
mrpt::math::CExceptionNotDefPos	On non-semidefinite-positive matrix as input.

Definition at line 356 of file CSparseMatrix.cpp.

References ASSERT_, mrpt::math::CSparseMatrix::cols(), mrpt::math::CSparseMatrix::isColumnCompressed(), m_numeric_structure, m_originalSM, m_symbolic_structure, mrpt::math::CSparseMatrix::rows(), and mrpt::math::CSparseMatrix::sparse_matrix.

CholeskyDecomp() [2/2]

mrpt::math::CSparseMatrix::CholeskyDecomp::CholeskyDecomp ( const CholeskyDecomp &  A )

delete

~CholeskyDecomp()

CSparseMatrix::CholeskyDecomp::~CholeskyDecomp ( )

virtual

Destructor.

Definition at line 378 of file CSparseMatrix.cpp.

Member Function Documentation

backsub() [1/3]

void CSparseMatrix::CholeskyDecomp::backsub	(	const double *	b,
		double *	result,
		const size_t	N
	)		const

overload for double pointers which assume the user has reserved the output memory for result

Return the vector from a back-substitution step that solves: Ux=b

Definition at line 400 of file CSparseMatrix.cpp.

References ASSERT_.

backsub() [2/3]

void CSparseMatrix::CholeskyDecomp::backsub	(	const Eigen::VectorXd &	b,
		Eigen::VectorXd &	result_x
	)		const

Return the vector from a back-substitution step that solves: Ux=b.

Return the vector from a back-substitution step that solves: Ux=b

Vectors can be Eigen::VectorXd or mrpt::math::CVectorDouble

Definition at line 391 of file CSparseMatrix.cpp.

References ASSERT_.

backsub() [3/3]

template<class VECTOR >

VECTOR mrpt::math::CSparseMatrix::CholeskyDecomp::backsub ( const VECTOR &  b ) const

inline

Return the vector from a back-substitution step that solves: Ux=b.

Definition at line 480 of file CSparseMatrix.h.

get_L() [1/2]

CMatrixDouble mrpt::math::CSparseMatrix::CholeskyDecomp::get_L ( ) const

inline

Return the L matrix (L*L' = M), as a dense matrix.

Definition at line 468 of file CSparseMatrix.h.

Referenced by TEST().

get_L() [2/2]

void CSparseMatrix::CholeskyDecomp::get_L

(

CMatrixDouble &

out_L

)

const

Return the L matrix (L*L' = M), as a dense matrix.

Definition at line 385 of file CSparseMatrix.cpp.

References mrpt::math::CSparseMatrix::cs2dense().

operator=()

CholeskyDecomp& mrpt::math::CSparseMatrix::CholeskyDecomp::operator= ( const CholeskyDecomp &  )

delete

update()

void CSparseMatrix::CholeskyDecomp::update

(

const CSparseMatrix &

new_SM

)

Update the Cholesky factorization from an updated vesion of the original input, square definite-positive sparse matrix.

NOTE: This new matrix MUST HAVE exactly the same sparse structure than the original one.

Definition at line 424 of file CSparseMatrix.cpp.

References ASSERTMSG_, and mrpt::math::CSparseMatrix::sparse_matrix.

Member Data Documentation

m_numeric_structure

csn* mrpt::math::CSparseMatrix::CholeskyDecomp::m_numeric_structure

private

Definition at line 444 of file CSparseMatrix.h.

Referenced by CholeskyDecomp().

m_originalSM

const CSparseMatrix* mrpt::math::CSparseMatrix::CholeskyDecomp::m_originalSM

private

A const reference to the original matrix used to build this decomposition.

Definition at line 447 of file CSparseMatrix.h.

Referenced by CholeskyDecomp().

m_symbolic_structure

css* mrpt::math::CSparseMatrix::CholeskyDecomp::m_symbolic_structure

private

Definition at line 443 of file CSparseMatrix.h.

Referenced by CholeskyDecomp().