devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

move sparse solvers from unsupported/ to main Eigen/ and remove the "not stable yet" warning

Browse Source
This commit is contained in:
Gael Guennebaud
2011-11-12 14:11:27 +01:00
parent dcb66d6b40
commit 53fa851724
62 changed files with 206 additions and 1336 deletions
Download Patch File Download Diff File Expand all files Collapse all files

139
unsupported/Eigen/src/IterativeSolvers/BasicPreconditioners.h
View File

@@ -1,139 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_BASIC_PRECONDITIONERS_H
#define EIGEN_BASIC_PRECONDITIONERS_H
/** \brief A preconditioner based on the digonal entries
*
* This class allows to approximately solve for A.x = b problems assuming A is a diagonal matrix.
* In other words, this preconditioner neglects all off diagonal entries and, in Eigen's language, solves for:
* \code
* A.diagonal().asDiagonal() . x = b
* \endcode
*
* \tparam _Scalar the type of the scalar.
*
* This preconditioner is suitable for both selfadjoint and general problems.
* The diagonal entries are pre-inverted and stored into a dense vector.
*
* \note A variant that has yet to be implemented would attempt to preserve the norm of each column.
*
*/
template <typename _Scalar>
class DiagonalPreconditioner
{
typedef _Scalar Scalar;
typedef Matrix<Scalar,Dynamic,1> Vector;
typedef typename Vector::Index Index;
public:
typedef Matrix<Scalar,Dynamic,Dynamic> MatrixType;
DiagonalPreconditioner() : m_isInitialized(false) {}
template<typename MatrixType>
DiagonalPreconditioner(const MatrixType& mat) : m_invdiag(mat.cols())
{
compute(mat);
}
Index rows() const { return m_invdiag.size(); }
Index cols() const { return m_invdiag.size(); }
template<typename MatrixType>
DiagonalPreconditioner& compute(const MatrixType& mat)
{
m_invdiag.resize(mat.cols());
for(int j=0; j<mat.outerSize(); ++j)
{
typename MatrixType::InnerIterator it(mat,j);
while(it && it.index()!=j) ++it;
if(it && it.index()==j)
m_invdiag(j) = Scalar(1)/it.value();
else
m_invdiag(j) = 0;
}
m_isInitialized = true;
return *this;
}
template<typename Rhs, typename Dest>
void _solve(const Rhs& b, Dest& x) const
{
x = m_invdiag.array() * b.array() ;
}
template<typename Rhs> inline const internal::solve_retval<DiagonalPreconditioner, Rhs>
solve(const MatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "DiagonalPreconditioner is not initialized.");
eigen_assert(m_invdiag.size()==b.rows()
&& "DiagonalPreconditioner::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval<DiagonalPreconditioner, Rhs>(*this, b.derived());
}
protected:
Vector m_invdiag;
bool m_isInitialized;
};
namespace internal {
template<typename _MatrixType, typename Rhs>
struct solve_retval<DiagonalPreconditioner<_MatrixType>, Rhs>
: solve_retval_base<DiagonalPreconditioner<_MatrixType>, Rhs>
{
typedef DiagonalPreconditioner<_MatrixType> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
}
/** \brief A naive preconditioner which approximates any matrix as the identity matrix
*
* \sa class DiagonalPreconditioner
*/
class IdentityPreconditioner
{
public:
IdentityPreconditioner() {}
template<typename MatrixType>
IdentityPreconditioner(const MatrixType& ) {}
template<typename MatrixType>
IdentityPreconditioner& compute(const MatrixType& ) { return *this; }
template<typename Rhs>
inline const Rhs& solve(const Rhs& b) const { return b; }
};
#endif // EIGEN_BASIC_PRECONDITIONERS_H

261
unsupported/Eigen/src/IterativeSolvers/BiCGSTAB.h
View File

@@ -1,261 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_BICGSTAB_H
#define EIGEN_BICGSTAB_H
namespace internal {
/** \internal Low-level bi conjugate gradient stabilized algorithm
* \param mat The matrix A
* \param rhs The right hand side vector b
* \param x On input and initial solution, on output the computed solution.
* \param precond A preconditioner being able to efficiently solve for an
* approximation of Ax=b (regardless of b)
* \param iters On input the max number of iteration, on output the number of performed iterations.
* \param tol_error On input the tolerance error, on output an estimation of the relative error.
*/
template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
void bicgstab(const MatrixType& mat, const Rhs& rhs, Dest& x,
const Preconditioner& precond, int& iters,
typename Dest::RealScalar& tol_error)
{
using std::sqrt;
using std::abs;
typedef typename Dest::RealScalar RealScalar;
typedef typename Dest::Scalar Scalar;
typedef Matrix<Scalar,Dynamic,1> VectorType;
RealScalar tol = tol_error;
int maxIters = iters;
int n = mat.cols();
VectorType r = rhs - mat * x;
VectorType r0 = r;
RealScalar r0_sqnorm = r0.squaredNorm();
Scalar rho = 1;
Scalar alpha = 1;
Scalar w = 1;
VectorType v = VectorType::Zero(n), p = VectorType::Zero(n);
VectorType y(n), z(n);
VectorType kt(n), ks(n);
VectorType s(n), t(n);
RealScalar tol2 = tol*tol;
int i = 0;
do
{
Scalar rho_old = rho;
rho = r0.dot(r);
Scalar beta = (rho/rho_old) * (alpha / w);
p = r + beta * (p - w * v);
y = precond.solve(p);
v.noalias() = mat * y;
alpha = rho / r0.dot(v);
s = r - alpha * v;
z = precond.solve(s);
t.noalias() = mat * z;
kt = precond.solve(t);
ks = precond.solve(s);
w = kt.dot(ks) / kt.squaredNorm();
x += alpha * y + w * z;
r = s - w * t;
++i;
} while ( r.squaredNorm()/r0_sqnorm > tol2 && i<maxIters );
tol_error = sqrt(r.squaredNorm()/r0_sqnorm);
//tol_error = sqrt(abs(absNew / absInit));
iters = i;
}
}
template< typename _MatrixType,
typename _Preconditioner = DiagonalPreconditioner<typename _MatrixType::Scalar> >
class BiCGSTAB;
namespace internal {
template< typename _MatrixType, typename _Preconditioner>
struct traits<BiCGSTAB<_MatrixType,_Preconditioner> >
{
typedef _MatrixType MatrixType;
typedef _Preconditioner Preconditioner;
};
}
/** \brief A bi conjugate gradient stabilized solver for sparse square problems
*
* This class allows to solve for A.x = b sparse linear problems using a bi conjugate gradient
* stabilized algorithm. The vectors x and b can be either dense or sparse.
*
* \tparam _MatrixType the type of the sparse matrix A, can be a dense or a sparse matrix.
* \tparam _Preconditioner the type of the preconditioner. Default is DiagonalPreconditioner
*
* The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
* and setTolerance() methods. The default are 1000 max iterations and NumTraits<Scalar>::epsilon()
* for the tolerance.
*
* This class can be used as the direct solver classes. Here is a typical usage example:
* \code
* int n = 10000;
* VectorXd x(n), b(n);
* SparseMatrix<double> A(n,n);
* // fill A and b
* BiCGSTAB<SparseMatrix<double> > solver;
* solver(A);
* x = solver.solve(b);
* std::cout << "#iterations: " << solver.iterations() << std::endl;
* std::cout << "estimated error: " << solver.error() << std::endl;
* // update b, and solve again
* x = solver.solve(b);
* \endcode
*
* By default the iterations start with x=0 as an initial guess of the solution.
* One can control the start using the solveWithGuess() method. Here is a step by
* step execution example starting with a random guess and printing the evolution
* of the estimated error:
* * \code
* x = VectorXd::Random(n);
* solver.setMaxIterations(1);
* int i = 0;
* do {
* x = solver.solveWithGuess(b,x);
* std::cout << i << " : " << solver.error() << std::endl;
* ++i;
* } while (solver.info()!=Success && i<100);
* \endcode
* Note that such a step by step excution is slightly slower.
*
* \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
*/
template< typename _MatrixType, typename _Preconditioner>
class BiCGSTAB : public IterativeSolverBase<BiCGSTAB<_MatrixType,_Preconditioner> >
{
typedef IterativeSolverBase<BiCGSTAB> Base;
using Base::mp_matrix;
using Base::m_error;
using Base::m_iterations;
using Base::m_info;
using Base::m_isInitialized;
public:
typedef _MatrixType MatrixType;
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::Index Index;
typedef typename MatrixType::RealScalar RealScalar;
typedef _Preconditioner Preconditioner;
public:
/** Default constructor. */
BiCGSTAB() : Base() {}
/** Initialize the solver with matrix \a A for further \c Ax=b solving.
*
* This constructor is a shortcut for the default constructor followed
* by a call to compute().
*
* \warning this class stores a reference to the matrix A as well as some
* precomputed values that depend on it. Therefore, if \a A is changed
* this class becomes invalid. Call compute() to update it with the new
* matrix A, or modify a copy of A.
*/
BiCGSTAB(const MatrixType& A) : Base(A) {}
~BiCGSTAB() {}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
* \a x0 as an initial solution.
*
* \sa compute()
*/
template<typename Rhs,typename Guess>
inline const internal::solve_retval_with_guess<BiCGSTAB, Rhs, Guess>
solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
{
eigen_assert(m_isInitialized && "BiCGSTAB is not initialized.");
eigen_assert(Base::rows()==b.rows()
&& "BiCGSTAB::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval_with_guess
<BiCGSTAB, Rhs, Guess>(*this, b.derived(), x0);
}
/** \internal */
template<typename Rhs,typename Dest>
void _solveWithGuess(const Rhs& b, Dest& x) const
{
for(int j=0; j<b.cols(); ++j)
{
m_iterations = Base::m_maxIterations;
m_error = Base::m_tolerance;
typename Dest::ColXpr xj(x,j);
internal::bicgstab(*mp_matrix, b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
}
m_isInitialized = true;
m_info = m_error <= Base::m_tolerance ? Success : NoConvergence;
}
/** \internal */
template<typename Rhs,typename Dest>
void _solve(const Rhs& b, Dest& x) const
{
x.setOnes();
_solveWithGuess(b,x);
}
protected:
};
namespace internal {
template<typename _MatrixType, typename _Preconditioner, typename Rhs>
struct solve_retval<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
: solve_retval_base<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
{
typedef BiCGSTAB<_MatrixType, _Preconditioner> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
}
#endif // EIGEN_BICGSTAB_H

255
unsupported/Eigen/src/IterativeSolvers/ConjugateGradient.h
View File

@@ -1,255 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_CONJUGATE_GRADIENT_H
#define EIGEN_CONJUGATE_GRADIENT_H
namespace internal {
/** \internal Low-level conjugate gradient algorithm
* \param mat The matrix A
* \param rhs The right hand side vector b
* \param x On input and initial solution, on output the computed solution.
* \param precond A preconditioner being able to efficiently solve for an
* approximation of Ax=b (regardless of b)
* \param iters On input the max number of iteration, on output the number of performed iterations.
* \param tol_error On input the tolerance error, on output an estimation of the relative error.
*/
template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
EIGEN_DONT_INLINE
void conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
const Preconditioner& precond, int& iters,
typename Dest::RealScalar& tol_error)
{
using std::sqrt;
using std::abs;
typedef typename Dest::RealScalar RealScalar;
typedef typename Dest::Scalar Scalar;
typedef Matrix<Scalar,Dynamic,1> VectorType;
RealScalar tol = tol_error;
int maxIters = iters;
int n = mat.cols();
VectorType residual = rhs - mat * x; //initial residual
VectorType p(n);
p = precond.solve(residual); //initial search direction
VectorType z(n), tmp(n);
RealScalar absNew = internal::real(residual.dot(p)); // the square of the absolute value of r scaled by invM
RealScalar absInit = absNew; // the initial absolute value
int i = 0;
while ((i < maxIters) && (absNew > tol*tol*absInit))
{
tmp.noalias() = mat * p; // the bottleneck of the algorithm
Scalar alpha = absNew / p.dot(tmp); // the amount we travel on dir
x += alpha * p; // update solution
residual -= alpha * tmp; // update residue
z = precond.solve(residual); // approximately solve for "A z = residual"
RealScalar absOld = absNew;
absNew = internal::real(residual.dot(z)); // update the absolute value of r
RealScalar beta = absNew / absOld; // calculate the Gram-Schmidit value used to create the new search direction
p = z + beta * p; // update search direction
i++;
}
tol_error = sqrt(abs(absNew / absInit));
iters = i;
}
}
template< typename _MatrixType, int _UpLo=Lower,
typename _Preconditioner = DiagonalPreconditioner<typename _MatrixType::Scalar> >
class ConjugateGradient;
namespace internal {
template< typename _MatrixType, int _UpLo, typename _Preconditioner>
struct traits<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> >
{
typedef _MatrixType MatrixType;
typedef _Preconditioner Preconditioner;
};
}
/** \brief A conjugate gradient solver for sparse self-adjoint problems
*
* This class allows to solve for A.x = b sparse linear problems using a conjugate gradient algorithm.
* The sparse matrix A must be selfadjoint. The vectors x and b can be either dense or sparse.
*
* \tparam _MatrixType the type of the sparse matrix A, can be a dense or a sparse matrix.
* \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
* or Upper. Default is Lower.
* \tparam _Preconditioner the type of the preconditioner. Default is DiagonalPreconditioner
*
* The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
* and setTolerance() methods. The default are 1000 max iterations and NumTraits<Scalar>::epsilon()
* for the tolerance.
*
* This class can be used as the direct solver classes. Here is a typical usage example:
* \code
* int n = 10000;
* VectorXd x(n), b(n);
* SparseMatrix<double> A(n,n);
* // fill A and b
* ConjugateGradient<SparseMatrix<double> > cg;
* cg(A);
* x = cg.solve(b);
* std::cout << "#iterations: " << cg.iterations() << std::endl;
* std::cout << "estimated error: " << cg.error() << std::endl;
* // update b, and solve again
* x = cg.solve(b);
* \endcode
*
* By default the iterations start with x=0 as an initial guess of the solution.
* One can control the start using the solveWithGuess() method. Here is a step by
* step execution example starting with a random guess and printing the evolution
* of the estimated error:
* * \code
* x = VectorXd::Random(n);
* cg.setMaxIterations(1);
* int i = 0;
* do {
* x = cg.solveWithGuess(b,x);
* std::cout << i << " : " << cg.error() << std::endl;
* ++i;
* } while (cg.info()!=Success && i<100);
* \endcode
* Note that such a step by step excution is slightly slower.
*
* \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
*/
template< typename _MatrixType, int _UpLo, typename _Preconditioner>
class ConjugateGradient : public IterativeSolverBase<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> >
{
typedef IterativeSolverBase<ConjugateGradient> Base;
using Base::mp_matrix;
using Base::m_error;
using Base::m_iterations;
using Base::m_info;
using Base::m_isInitialized;
public:
typedef _MatrixType MatrixType;
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::Index Index;
typedef typename MatrixType::RealScalar RealScalar;
typedef _Preconditioner Preconditioner;
enum {
UpLo = _UpLo
};
public:
/** Default constructor. */
ConjugateGradient() : Base() {}
/** Initialize the solver with matrix \a A for further \c Ax=b solving.
*
* This constructor is a shortcut for the default constructor followed
* by a call to compute().
*
* \warning this class stores a reference to the matrix A as well as some
* precomputed values that depend on it. Therefore, if \a A is changed
* this class becomes invalid. Call compute() to update it with the new
* matrix A, or modify a copy of A.
*/
ConjugateGradient(const MatrixType& A) : Base(A) {}
~ConjugateGradient() {}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
* \a x0 as an initial solution.
*
* \sa compute()
*/
template<typename Rhs,typename Guess>
inline const internal::solve_retval_with_guess<ConjugateGradient, Rhs, Guess>
solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
{
eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
eigen_assert(Base::rows()==b.rows()
&& "ConjugateGradient::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval_with_guess
<ConjugateGradient, Rhs, Guess>(*this, b.derived(), x0);
}
/** \internal */
template<typename Rhs,typename Dest>
void _solveWithGuess(const Rhs& b, Dest& x) const
{
m_iterations = Base::m_maxIterations;
m_error = Base::m_tolerance;
for(int j=0; j<b.cols(); ++j)
{
m_iterations = Base::m_maxIterations;
m_error = Base::m_tolerance;
typename Dest::ColXpr xj(x,j);
internal::conjugate_gradient(mp_matrix->template selfadjointView<UpLo>(), b.col(j), xj,
Base::m_preconditioner, m_iterations, m_error);
}
m_isInitialized = true;
m_info = m_error <= Base::m_tolerance ? Success : NoConvergence;
}
/** \internal */
template<typename Rhs,typename Dest>
void _solve(const Rhs& b, Dest& x) const
{
x.setOnes();
_solveWithGuess(b,x);
}
protected:
};
namespace internal {
template<typename _MatrixType, int _UpLo, typename _Preconditioner, typename Rhs>
struct solve_retval<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner>, Rhs>
: solve_retval_base<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner>, Rhs>
{
typedef ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> Dec;
EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec()._solve(rhs(),dst);
}
};
}
#endif // EIGEN_CONJUGATE_GRADIENT_H

225
unsupported/Eigen/src/IterativeSolvers/IterativeSolverBase.h
View File

@@ -1,225 +0,0 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_ITERATIVE_SOLVER_BASE_H
#define EIGEN_ITERATIVE_SOLVER_BASE_H
/** \brief Base class for linear iterative solvers
*
* \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
*/
template< typename Derived>
class IterativeSolverBase
{
public:
typedef typename internal::traits<Derived>::MatrixType MatrixType;
typedef typename internal::traits<Derived>::Preconditioner Preconditioner;
typedef typename MatrixType::Scalar Scalar;
typedef typename MatrixType::Index Index;
typedef typename MatrixType::RealScalar RealScalar;
public:
Derived& derived() { return *static_cast<Derived*>(this); }
const Derived& derived() const { return *static_cast<const Derived*>(this); }
/** Default constructor. */
IterativeSolverBase()
: mp_matrix(0)
{
init();
}
/** Initialize the solver with matrix \a A for further \c Ax=b solving.
*
* This constructor is a shortcut for the default constructor followed
* by a call to compute().
*
* \warning this class stores a reference to the matrix A as well as some
* precomputed values that depend on it. Therefore, if \a A is changed
* this class becomes invalid. Call compute() to update it with the new
* matrix A, or modify a copy of A.
*/
IterativeSolverBase(const MatrixType& A)
{
init();
compute(A);
}
~IterativeSolverBase() {}
/** Initializes the iterative solver with the matrix \a A for further solving \c Ax=b problems.
*
* Currently, this function mostly initialized/compute the preconditioner. In the future
* we might, for instance, implement column reodering for faster matrix vector products.
*
* \warning this class stores a reference to the matrix A as well as some
* precomputed values that depend on it. Therefore, if \a A is changed
* this class becomes invalid. Call compute() to update it with the new
* matrix A, or modify a copy of A.
*/
Derived& compute(const MatrixType& A)
{
mp_matrix = &A;
m_preconditioner.compute(A);
m_isInitialized = true;
m_info = Success;
return derived();
}
/** \internal */
Index rows() const { return mp_matrix->rows(); }
/** \internal */
Index cols() const { return mp_matrix->cols(); }
/** \returns the tolerance threshold used by the stopping criteria */
RealScalar tolerance() const { return m_tolerance; }
/** Sets the tolerance threshold used by the stopping criteria */
Derived& setTolerance(RealScalar tolerance)
{
m_tolerance = tolerance;
return derived();
}
/** \returns a read-write reference to the preconditioner for custom configuration. */
Preconditioner& preconditioner() { return m_preconditioner; }
/** \returns a read-only reference to the preconditioner. */
const Preconditioner& preconditioner() const { return m_preconditioner; }
/** \returns the max number of iterations */
int maxIterations() const { return m_maxIterations; }
/** Sets the max number of iterations */
Derived& setMaxIterations(int maxIters)
{
m_maxIterations = maxIters;
return derived();
}
/** \returns the number of iterations performed during the last solve */
int iterations() const
{
eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
return m_iterations;
}
/** \returns the tolerance error reached during the last solve */
RealScalar error() const
{
eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
return m_error;
}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
*
* \sa compute()
*/
template<typename Rhs> inline const internal::solve_retval<Derived, Rhs>
solve(const MatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
eigen_assert(rows()==b.rows()
&& "IterativeSolverBase::solve(): invalid number of rows of the right hand side matrix b");
return internal::solve_retval<Derived, Rhs>(derived(), b.derived());
}
/** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
*
* \sa compute()
*/
template<typename Rhs>
inline const internal::sparse_solve_retval<IterativeSolverBase, Rhs>
solve(const SparseMatrixBase<Rhs>& b) const
{
eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
eigen_assert(rows()==b.rows()
&& "IterativeSolverBase::solve(): invalid number of rows of the right hand side matrix b");
return internal::sparse_solve_retval<IterativeSolverBase, Rhs>(*this, b.derived());
}
/** \returns Success if the iterations converged, and NoConvergence otherwise. */
ComputationInfo info() const
{
eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
return m_info;
}
/** \internal */
template<typename Rhs, typename DestScalar, int DestOptions, typename DestIndex>
void _solve_sparse(const Rhs& b, SparseMatrix<DestScalar,DestOptions,DestIndex> &dest) const
{
eigen_assert(rows()==b.rows());
int rhsCols = b.cols();
int size = b.rows();
Eigen::Matrix<DestScalar,Dynamic,1> tb(size);
Eigen::Matrix<DestScalar,Dynamic,1> tx(size);
for(int k=0; k<rhsCols; ++k)
{
tb = b.col(k);
tx = derived().solve(tb);
dest.col(k) = tx.sparseView(0);
}
}
protected:
void init()
{
m_isInitialized = false;
m_maxIterations = 1000;
m_tolerance = NumTraits<Scalar>::epsilon();
}
const MatrixType* mp_matrix;
Preconditioner m_preconditioner;
int m_maxIterations;
RealScalar m_tolerance;
mutable RealScalar m_error;
mutable int m_iterations;
mutable ComputationInfo m_info;
mutable bool m_isInitialized;
};
namespace internal {
template<typename Derived, typename Rhs>
struct sparse_solve_retval<IterativeSolverBase<Derived>, Rhs>
: sparse_solve_retval_base<IterativeSolverBase<Derived>, Rhs>
{
typedef IterativeSolverBase<Derived> Dec;
EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
template<typename Dest> void evalTo(Dest& dst) const
{
dec().derived()._solve_sparse(rhs(),dst);
}
};
}
#endif // EIGEN_ITERATIVE_SOLVER_BASE_H