00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011 #ifndef EIGEN_GMRES_H
00012 #define EIGEN_GMRES_H
00013
00014 namespace Eigen {
00015
00016 namespace internal {
00017
00055 template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
00056 bool gmres(const MatrixType & mat, const Rhs & rhs, Dest & x, const Preconditioner & precond,
00057 int &iters, const int &restart, typename Dest::RealScalar & tol_error) {
00058
00059 using std::sqrt;
00060 using std::abs;
00061
00062 typedef typename Dest::RealScalar RealScalar;
00063 typedef typename Dest::Scalar Scalar;
00064 typedef Matrix < Scalar, Dynamic, 1 > VectorType;
00065 typedef Matrix < Scalar, Dynamic, Dynamic > FMatrixType;
00066
00067 RealScalar tol = tol_error;
00068 const int maxIters = iters;
00069 iters = 0;
00070
00071 const int m = mat.rows();
00072
00073 VectorType p0 = rhs - mat*x;
00074 VectorType r0 = precond.solve(p0);
00075
00076
00077 VectorType w = VectorType::Zero(restart + 1);
00078
00079 FMatrixType H = FMatrixType::Zero(m, restart + 1);
00080 VectorType tau = VectorType::Zero(restart + 1);
00081 std::vector < JacobiRotation < Scalar > > G(restart);
00082
00083
00084 VectorType e;
00085 RealScalar beta;
00086 r0.makeHouseholder(e, tau.coeffRef(0), beta);
00087 w(0)=(Scalar) beta;
00088 H.bottomLeftCorner(m - 1, 1) = e;
00089
00090 for (int k = 1; k <= restart; ++k) {
00091
00092 ++iters;
00093
00094 VectorType v = VectorType::Unit(m, k - 1), workspace(m);
00095
00096
00097 for (int i = k - 1; i >= 0; --i) {
00098 v.tail(m - i).applyHouseholderOnTheLeft(H.col(i).tail(m - i - 1), tau.coeffRef(i), workspace.data());
00099 }
00100
00101
00102 VectorType t=mat*v;
00103 v=precond.solve(t);
00104
00105
00106 for (int i = 0; i < k; ++i) {
00107 v.tail(m - i).applyHouseholderOnTheLeft(H.col(i).tail(m - i - 1), tau.coeffRef(i), workspace.data());
00108 }
00109
00110 if (v.tail(m - k).norm() != 0.0) {
00111
00112 if (k <= restart) {
00113
00114
00115 VectorType e(m - k - 1);
00116 RealScalar beta;
00117 v.tail(m - k).makeHouseholder(e, tau.coeffRef(k), beta);
00118 H.col(k).tail(m - k - 1) = e;
00119
00120
00121 v.tail(m - k).applyHouseholderOnTheLeft(H.col(k).tail(m - k - 1), tau.coeffRef(k), workspace.data());
00122
00123 }
00124 }
00125
00126 if (k > 1) {
00127 for (int i = 0; i < k - 1; ++i) {
00128
00129 v.applyOnTheLeft(i, i + 1, G[i].adjoint());
00130 }
00131 }
00132
00133 if (k<m && v(k) != (Scalar) 0) {
00134
00135 G[k - 1].makeGivens(v(k - 1), v(k));
00136
00137
00138 v.applyOnTheLeft(k - 1, k, G[k - 1].adjoint());
00139 w.applyOnTheLeft(k - 1, k, G[k - 1].adjoint());
00140
00141 }
00142
00143
00144 H.col(k - 1).head(k) = v.head(k);
00145
00146 bool stop=(k==m || abs(w(k)) < tol || iters == maxIters);
00147
00148 if (stop || k == restart) {
00149
00150
00151 VectorType y = w.head(k);
00152 H.topLeftCorner(k, k).template triangularView < Eigen::Upper > ().solveInPlace(y);
00153
00154
00155 VectorType x_new = y(k - 1) * VectorType::Unit(m, k - 1);
00156
00157
00158 x_new.tail(m - k + 1).applyHouseholderOnTheLeft(H.col(k - 1).tail(m - k), tau.coeffRef(k - 1), workspace.data());
00159
00160 for (int i = k - 2; i >= 0; --i) {
00161 x_new += y(i) * VectorType::Unit(m, i);
00162
00163 x_new.tail(m - i).applyHouseholderOnTheLeft(H.col(i).tail(m - i - 1), tau.coeffRef(i), workspace.data());
00164 }
00165
00166 x += x_new;
00167
00168 if (stop) {
00169 return true;
00170 } else {
00171 k=0;
00172
00173
00174 VectorType p0=mat*x;
00175 VectorType p1=precond.solve(p0);
00176 r0 = rhs - p1;
00177
00178 w = VectorType::Zero(restart + 1);
00179 H = FMatrixType::Zero(m, restart + 1);
00180 tau = VectorType::Zero(restart + 1);
00181
00182
00183 RealScalar beta;
00184 r0.makeHouseholder(e, tau.coeffRef(0), beta);
00185 w(0)=(Scalar) beta;
00186 H.bottomLeftCorner(m - 1, 1) = e;
00187
00188 }
00189
00190 }
00191
00192
00193
00194 }
00195
00196 return false;
00197
00198 }
00199
00200 }
00201
00202 template< typename _MatrixType,
00203 typename _Preconditioner = DiagonalPreconditioner<typename _MatrixType::Scalar> >
00204 class GMRES;
00205
00206 namespace internal {
00207
00208 template< typename _MatrixType, typename _Preconditioner>
00209 struct traits<GMRES<_MatrixType,_Preconditioner> >
00210 {
00211 typedef _MatrixType MatrixType;
00212 typedef _Preconditioner Preconditioner;
00213 };
00214
00215 }
00216
00262 template< typename _MatrixType, typename _Preconditioner>
00263 class GMRES : public IterativeSolverBase<GMRES<_MatrixType,_Preconditioner> >
00264 {
00265 typedef IterativeSolverBase<GMRES> Base;
00266 using Base::mp_matrix;
00267 using Base::m_error;
00268 using Base::m_iterations;
00269 using Base::m_info;
00270 using Base::m_isInitialized;
00271
00272 private:
00273 int m_restart;
00274
00275 public:
00276 typedef _MatrixType MatrixType;
00277 typedef typename MatrixType::Scalar Scalar;
00278 typedef typename MatrixType::Index Index;
00279 typedef typename MatrixType::RealScalar RealScalar;
00280 typedef _Preconditioner Preconditioner;
00281
00282 public:
00283
00285 GMRES() : Base(), m_restart(30) {}
00286
00297 GMRES(const MatrixType& A) : Base(A), m_restart(30) {}
00298
00299 ~GMRES() {}
00300
00303 int get_restart() { return m_restart; }
00304
00308 void set_restart(const int restart) { m_restart=restart; }
00309
00315 template<typename Rhs,typename Guess>
00316 inline const internal::solve_retval_with_guess<GMRES, Rhs, Guess>
00317 solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
00318 {
00319 eigen_assert(m_isInitialized && "GMRES is not initialized.");
00320 eigen_assert(Base::rows()==b.rows()
00321 && "GMRES::solve(): invalid number of rows of the right hand side matrix b");
00322 return internal::solve_retval_with_guess
00323 <GMRES, Rhs, Guess>(*this, b.derived(), x0);
00324 }
00325
00327 template<typename Rhs,typename Dest>
00328 void _solveWithGuess(const Rhs& b, Dest& x) const
00329 {
00330 bool failed = false;
00331 for(int j=0; j<b.cols(); ++j)
00332 {
00333 m_iterations = Base::maxIterations();
00334 m_error = Base::m_tolerance;
00335
00336 typename Dest::ColXpr xj(x,j);
00337 if(!internal::gmres(*mp_matrix, b.col(j), xj, Base::m_preconditioner, m_iterations, m_restart, m_error))
00338 failed = true;
00339 }
00340 m_info = failed ? NumericalIssue
00341 : m_error <= Base::m_tolerance ? Success
00342 : NoConvergence;
00343 m_isInitialized = true;
00344 }
00345
00347 template<typename Rhs,typename Dest>
00348 void _solve(const Rhs& b, Dest& x) const
00349 {
00350 x = b;
00351 if(x.squaredNorm() == 0) return;
00352 _solveWithGuess(b,x);
00353 }
00354
00355 protected:
00356
00357 };
00358
00359
00360 namespace internal {
00361
00362 template<typename _MatrixType, typename _Preconditioner, typename Rhs>
00363 struct solve_retval<GMRES<_MatrixType, _Preconditioner>, Rhs>
00364 : solve_retval_base<GMRES<_MatrixType, _Preconditioner>, Rhs>
00365 {
00366 typedef GMRES<_MatrixType, _Preconditioner> Dec;
00367 EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
00368
00369 template<typename Dest> void evalTo(Dest& dst) const
00370 {
00371 dec()._solve(rhs(),dst);
00372 }
00373 };
00374
00375 }
00376
00377 }
00378
00379 #endif // EIGEN_GMRES_H