• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
5 //
6 // This Source Code Form is subject to the terms of the Mozilla
7 // Public License v. 2.0. If a copy of the MPL was not distributed
8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
9 
10 #ifndef EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
11 #define EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
12 
13 namespace Eigen {
14 
15 namespace internal {
16 
17 /** \internal Low-level conjugate gradient algorithm for least-square problems
18   * \param mat The matrix A
19   * \param rhs The right hand side vector b
20   * \param x On input and initial solution, on output the computed solution.
21   * \param precond A preconditioner being able to efficiently solve for an
22   *                approximation of A'Ax=b (regardless of b)
23   * \param iters On input the max number of iteration, on output the number of performed iterations.
24   * \param tol_error On input the tolerance error, on output an estimation of the relative error.
25   */
26 template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
27 EIGEN_DONT_INLINE
least_square_conjugate_gradient(const MatrixType & mat,const Rhs & rhs,Dest & x,const Preconditioner & precond,Index & iters,typename Dest::RealScalar & tol_error)28 void least_square_conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
29                                      const Preconditioner& precond, Index& iters,
30                                      typename Dest::RealScalar& tol_error)
31 {
32   using std::sqrt;
33   using std::abs;
34   typedef typename Dest::RealScalar RealScalar;
35   typedef typename Dest::Scalar Scalar;
36   typedef Matrix<Scalar,Dynamic,1> VectorType;
37 
38   RealScalar tol = tol_error;
39   Index maxIters = iters;
40 
41   Index m = mat.rows(), n = mat.cols();
42 
43   VectorType residual        = rhs - mat * x;
44   VectorType normal_residual = mat.adjoint() * residual;
45 
46   RealScalar rhsNorm2 = (mat.adjoint()*rhs).squaredNorm();
47   if(rhsNorm2 == 0)
48   {
49     x.setZero();
50     iters = 0;
51     tol_error = 0;
52     return;
53   }
54   RealScalar threshold = tol*tol*rhsNorm2;
55   RealScalar residualNorm2 = normal_residual.squaredNorm();
56   if (residualNorm2 < threshold)
57   {
58     iters = 0;
59     tol_error = sqrt(residualNorm2 / rhsNorm2);
60     return;
61   }
62 
63   VectorType p(n);
64   p = precond.solve(normal_residual);                         // initial search direction
65 
66   VectorType z(n), tmp(m);
67   RealScalar absNew = numext::real(normal_residual.dot(p));  // the square of the absolute value of r scaled by invM
68   Index i = 0;
69   while(i < maxIters)
70   {
71     tmp.noalias() = mat * p;
72 
73     Scalar alpha = absNew / tmp.squaredNorm();      // the amount we travel on dir
74     x += alpha * p;                                 // update solution
75     residual -= alpha * tmp;                        // update residual
76     normal_residual = mat.adjoint() * residual;     // update residual of the normal equation
77 
78     residualNorm2 = normal_residual.squaredNorm();
79     if(residualNorm2 < threshold)
80       break;
81 
82     z = precond.solve(normal_residual);             // approximately solve for "A'A z = normal_residual"
83 
84     RealScalar absOld = absNew;
85     absNew = numext::real(normal_residual.dot(z));  // update the absolute value of r
86     RealScalar beta = absNew / absOld;              // calculate the Gram-Schmidt value used to create the new search direction
87     p = z + beta * p;                               // update search direction
88     i++;
89   }
90   tol_error = sqrt(residualNorm2 / rhsNorm2);
91   iters = i;
92 }
93 
94 }
95 
96 template< typename _MatrixType,
97           typename _Preconditioner = LeastSquareDiagonalPreconditioner<typename _MatrixType::Scalar> >
98 class LeastSquaresConjugateGradient;
99 
100 namespace internal {
101 
102 template< typename _MatrixType, typename _Preconditioner>
103 struct traits<LeastSquaresConjugateGradient<_MatrixType,_Preconditioner> >
104 {
105   typedef _MatrixType MatrixType;
106   typedef _Preconditioner Preconditioner;
107 };
108 
109 }
110 
111 /** \ingroup IterativeLinearSolvers_Module
112   * \brief A conjugate gradient solver for sparse (or dense) least-square problems
113   *
114   * This class allows to solve for A x = b linear problems using an iterative conjugate gradient algorithm.
115   * The matrix A can be non symmetric and rectangular, but the matrix A' A should be positive-definite to guaranty stability.
116   * Otherwise, the SparseLU or SparseQR classes might be preferable.
117   * The matrix A and the vectors x and b can be either dense or sparse.
118   *
119   * \tparam _MatrixType the type of the matrix A, can be a dense or a sparse matrix.
120   * \tparam _Preconditioner the type of the preconditioner. Default is LeastSquareDiagonalPreconditioner
121   *
122   * \implsparsesolverconcept
123   *
124   * The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
125   * and setTolerance() methods. The defaults are the size of the problem for the maximal number of iterations
126   * and NumTraits<Scalar>::epsilon() for the tolerance.
127   *
128   * This class can be used as the direct solver classes. Here is a typical usage example:
129     \code
130     int m=1000000, n = 10000;
131     VectorXd x(n), b(m);
132     SparseMatrix<double> A(m,n);
133     // fill A and b
134     LeastSquaresConjugateGradient<SparseMatrix<double> > lscg;
135     lscg.compute(A);
136     x = lscg.solve(b);
137     std::cout << "#iterations:     " << lscg.iterations() << std::endl;
138     std::cout << "estimated error: " << lscg.error()      << std::endl;
139     // update b, and solve again
140     x = lscg.solve(b);
141     \endcode
142   *
143   * By default the iterations start with x=0 as an initial guess of the solution.
144   * One can control the start using the solveWithGuess() method.
145   *
146   * \sa class ConjugateGradient, SparseLU, SparseQR
147   */
148 template< typename _MatrixType, typename _Preconditioner>
149 class LeastSquaresConjugateGradient : public IterativeSolverBase<LeastSquaresConjugateGradient<_MatrixType,_Preconditioner> >
150 {
151   typedef IterativeSolverBase<LeastSquaresConjugateGradient> Base;
152   using Base::matrix;
153   using Base::m_error;
154   using Base::m_iterations;
155   using Base::m_info;
156   using Base::m_isInitialized;
157 public:
158   typedef _MatrixType MatrixType;
159   typedef typename MatrixType::Scalar Scalar;
160   typedef typename MatrixType::RealScalar RealScalar;
161   typedef _Preconditioner Preconditioner;
162 
163 public:
164 
165   /** Default constructor. */
166   LeastSquaresConjugateGradient() : Base() {}
167 
168   /** Initialize the solver with matrix \a A for further \c Ax=b solving.
169     *
170     * This constructor is a shortcut for the default constructor followed
171     * by a call to compute().
172     *
173     * \warning this class stores a reference to the matrix A as well as some
174     * precomputed values that depend on it. Therefore, if \a A is changed
175     * this class becomes invalid. Call compute() to update it with the new
176     * matrix A, or modify a copy of A.
177     */
178   template<typename MatrixDerived>
179   explicit LeastSquaresConjugateGradient(const EigenBase<MatrixDerived>& A) : Base(A.derived()) {}
180 
181   ~LeastSquaresConjugateGradient() {}
182 
183   /** \internal */
184   template<typename Rhs,typename Dest>
185   void _solve_with_guess_impl(const Rhs& b, Dest& x) const
186   {
187     m_iterations = Base::maxIterations();
188     m_error = Base::m_tolerance;
189 
190     for(Index j=0; j<b.cols(); ++j)
191     {
192       m_iterations = Base::maxIterations();
193       m_error = Base::m_tolerance;
194 
195       typename Dest::ColXpr xj(x,j);
196       internal::least_square_conjugate_gradient(matrix(), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
197     }
198 
199     m_isInitialized = true;
200     m_info = m_error <= Base::m_tolerance ? Success : NoConvergence;
201   }
202 
203   /** \internal */
204   using Base::_solve_impl;
205   template<typename Rhs,typename Dest>
206   void _solve_impl(const MatrixBase<Rhs>& b, Dest& x) const
207   {
208     x.setZero();
209     _solve_with_guess_impl(b.derived(),x);
210   }
211 
212 };
213 
214 } // end namespace Eigen
215 
216 #endif // EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
217