Lapack_Cholesky< Size, Precision > Class Template Reference

Decomposes a positive-semidefinite symmetric matrix A (such as a covariance) into L*L^T, where L is lower-triangular. More...

#include <Lapack_Cholesky.h>

List of all members.

Public Member Functions
template<class P2 , class B2 >
	Lapack_Cholesky (const Matrix< Size, Size, P2, B2 > &m)
	Lapack_Cholesky (int size)
template<class P2 , class B2 >
void	compute (const Matrix< Size, Size, P2, B2 > &m)
void	do_compute ()
int	rank () const
template<int Size2, typename P2 , typename B2 >
Vector< Size, Precision >	backsub (const Vector< Size2, P2, B2 > &v) const
template<int Size2, int Cols2, typename P2 , typename B2 >
Matrix< Size, Cols2, Precision, ColMajor >	backsub (const Matrix< Size2, Cols2, P2, B2 > &m) const
template<int Size2, typename P2 , typename B2 >
Precision	mahalanobis (const Vector< Size2, P2, B2 > &v) const
Matrix< Size, Size, Precision >	get_L () const
Precision	determinant () const
Matrix	get_inverse () const

---

Detailed Description

template<int Size, typename Precision = DefaultPrecision>
class TooN::Lapack_Cholesky< Size, Precision >

Decomposes a positive-semidefinite symmetric matrix A (such as a covariance) into L*L^T, where L is lower-triangular.

Also can compute A = S*S^T, with S lower triangular. The LDL^T form is faster to compute than the class Cholesky decomposition. The decomposition can be used to compute A^-1*x, A^-1*M, M*A^-1*M^T, and A^-1 itself, though the latter rarely needs to be explicitly represented. Also efficiently computes det(A) and rank(A). It can be used as follows:

// Declare some matrices.
Matrix<3> A = ...; // we'll pretend it is pos-def
Matrix<2,3> M;
Matrix<2> B;
Vector<3> y = make_Vector(2,3,4);
// create the Cholesky decomposition of A
Cholesky<3> chol(A);
// compute x = A^-1 * y
x = cholA.backsub(y);
//compute A^-1
Matrix<3> Ainv = cholA.get_inverse();

Cholesky decomposition of a symmetric matrix. Only the lower half of the matrix is considered This uses the non-sqrt version of the decomposition giving symmetric M = L*D*L.T() where the diagonal of L contains ones

Parameters:

Size	the size of the matrix
Precision	the precision of the entries in the matrix and its decomposition