Home | History | Annotate | Download | only in MatrixFunctions
      1 // This file is part of Eigen, a lightweight C++ template library
      2 // for linear algebra.
      3 //
      4 // Copyright (C) 2009-2011 Jitse Niesen <jitse (at) maths.leeds.ac.uk>
      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_MATRIX_FUNCTION
     11 #define EIGEN_MATRIX_FUNCTION
     12 
     13 #include "StemFunction.h"
     14 #include "MatrixFunctionAtomic.h"
     15 
     16 
     17 namespace Eigen {
     18 
     19 /** \ingroup MatrixFunctions_Module
     20   * \brief Class for computing matrix functions.
     21   * \tparam  MatrixType  type of the argument of the matrix function,
     22   *                      expected to be an instantiation of the Matrix class template.
     23   * \tparam  AtomicType  type for computing matrix function of atomic blocks.
     24   * \tparam  IsComplex   used internally to select correct specialization.
     25   *
     26   * This class implements the Schur-Parlett algorithm for computing matrix functions. The spectrum of the
     27   * matrix is divided in clustered of eigenvalues that lies close together. This class delegates the
     28   * computation of the matrix function on every block corresponding to these clusters to an object of type
     29   * \p AtomicType and uses these results to compute the matrix function of the whole matrix. The class
     30   * \p AtomicType should have a \p compute() member function for computing the matrix function of a block.
     31   *
     32   * \sa class MatrixFunctionAtomic, class MatrixLogarithmAtomic
     33   */
     34 template <typename MatrixType,
     35 	  typename AtomicType,
     36           int IsComplex = NumTraits<typename internal::traits<MatrixType>::Scalar>::IsComplex>
     37 class MatrixFunction
     38 {
     39   public:
     40 
     41     /** \brief Constructor.
     42       *
     43       * \param[in]  A       argument of matrix function, should be a square matrix.
     44       * \param[in]  atomic  class for computing matrix function of atomic blocks.
     45       *
     46       * The class stores references to \p A and \p atomic, so they should not be
     47       * changed (or destroyed) before compute() is called.
     48       */
     49     MatrixFunction(const MatrixType& A, AtomicType& atomic);
     50 
     51     /** \brief Compute the matrix function.
     52       *
     53       * \param[out] result  the function \p f applied to \p A, as
     54       * specified in the constructor.
     55       *
     56       * See MatrixBase::matrixFunction() for details on how this computation
     57       * is implemented.
     58       */
     59     template <typename ResultType>
     60     void compute(ResultType &result);
     61 };
     62 
     63 
     64 /** \internal \ingroup MatrixFunctions_Module
     65   * \brief Partial specialization of MatrixFunction for real matrices
     66   */
     67 template <typename MatrixType, typename AtomicType>
     68 class MatrixFunction<MatrixType, AtomicType, 0>
     69 {
     70   private:
     71 
     72     typedef internal::traits<MatrixType> Traits;
     73     typedef typename Traits::Scalar Scalar;
     74     static const int Rows = Traits::RowsAtCompileTime;
     75     static const int Cols = Traits::ColsAtCompileTime;
     76     static const int Options = MatrixType::Options;
     77     static const int MaxRows = Traits::MaxRowsAtCompileTime;
     78     static const int MaxCols = Traits::MaxColsAtCompileTime;
     79 
     80     typedef std::complex<Scalar> ComplexScalar;
     81     typedef Matrix<ComplexScalar, Rows, Cols, Options, MaxRows, MaxCols> ComplexMatrix;
     82 
     83   public:
     84 
     85     /** \brief Constructor.
     86       *
     87       * \param[in]  A       argument of matrix function, should be a square matrix.
     88       * \param[in]  atomic  class for computing matrix function of atomic blocks.
     89       */
     90     MatrixFunction(const MatrixType& A, AtomicType& atomic) : m_A(A), m_atomic(atomic) { }
     91 
     92     /** \brief Compute the matrix function.
     93       *
     94       * \param[out] result  the function \p f applied to \p A, as
     95       * specified in the constructor.
     96       *
     97       * This function converts the real matrix \c A to a complex matrix,
     98       * uses MatrixFunction<MatrixType,1> and then converts the result back to
     99       * a real matrix.
    100       */
    101     template <typename ResultType>
    102     void compute(ResultType& result)
    103     {
    104       ComplexMatrix CA = m_A.template cast<ComplexScalar>();
    105       ComplexMatrix Cresult;
    106       MatrixFunction<ComplexMatrix, AtomicType> mf(CA, m_atomic);
    107       mf.compute(Cresult);
    108       result = Cresult.real();
    109     }
    110 
    111   private:
    112     typename internal::nested<MatrixType>::type m_A; /**< \brief Reference to argument of matrix function. */
    113     AtomicType& m_atomic; /**< \brief Class for computing matrix function of atomic blocks. */
    114 
    115     MatrixFunction& operator=(const MatrixFunction&);
    116 };
    117 
    118 
    119 /** \internal \ingroup MatrixFunctions_Module
    120   * \brief Partial specialization of MatrixFunction for complex matrices
    121   */
    122 template <typename MatrixType, typename AtomicType>
    123 class MatrixFunction<MatrixType, AtomicType, 1>
    124 {
    125   private:
    126 
    127     typedef internal::traits<MatrixType> Traits;
    128     typedef typename MatrixType::Scalar Scalar;
    129     typedef typename MatrixType::Index Index;
    130     static const int RowsAtCompileTime = Traits::RowsAtCompileTime;
    131     static const int ColsAtCompileTime = Traits::ColsAtCompileTime;
    132     static const int Options = MatrixType::Options;
    133     typedef typename NumTraits<Scalar>::Real RealScalar;
    134     typedef Matrix<Scalar, Traits::RowsAtCompileTime, 1> VectorType;
    135     typedef Matrix<Index, Traits::RowsAtCompileTime, 1> IntVectorType;
    136     typedef Matrix<Index, Dynamic, 1> DynamicIntVectorType;
    137     typedef std::list<Scalar> Cluster;
    138     typedef std::list<Cluster> ListOfClusters;
    139     typedef Matrix<Scalar, Dynamic, Dynamic, Options, RowsAtCompileTime, ColsAtCompileTime> DynMatrixType;
    140 
    141   public:
    142 
    143     MatrixFunction(const MatrixType& A, AtomicType& atomic);
    144     template <typename ResultType> void compute(ResultType& result);
    145 
    146   private:
    147 
    148     void computeSchurDecomposition();
    149     void partitionEigenvalues();
    150     typename ListOfClusters::iterator findCluster(Scalar key);
    151     void computeClusterSize();
    152     void computeBlockStart();
    153     void constructPermutation();
    154     void permuteSchur();
    155     void swapEntriesInSchur(Index index);
    156     void computeBlockAtomic();
    157     Block<MatrixType> block(MatrixType& A, Index i, Index j);
    158     void computeOffDiagonal();
    159     DynMatrixType solveTriangularSylvester(const DynMatrixType& A, const DynMatrixType& B, const DynMatrixType& C);
    160 
    161     typename internal::nested<MatrixType>::type m_A; /**< \brief Reference to argument of matrix function. */
    162     AtomicType& m_atomic; /**< \brief Class for computing matrix function of atomic blocks. */
    163     MatrixType m_T; /**< \brief Triangular part of Schur decomposition */
    164     MatrixType m_U; /**< \brief Unitary part of Schur decomposition */
    165     MatrixType m_fT; /**< \brief %Matrix function applied to #m_T */
    166     ListOfClusters m_clusters; /**< \brief Partition of eigenvalues into clusters of ei'vals "close" to each other */
    167     DynamicIntVectorType m_eivalToCluster; /**< \brief m_eivalToCluster[i] = j means i-th ei'val is in j-th cluster */
    168     DynamicIntVectorType m_clusterSize; /**< \brief Number of eigenvalues in each clusters  */
    169     DynamicIntVectorType m_blockStart; /**< \brief Row index at which block corresponding to i-th cluster starts */
    170     IntVectorType m_permutation; /**< \brief Permutation which groups ei'vals in the same cluster together */
    171 
    172     /** \brief Maximum distance allowed between eigenvalues to be considered "close".
    173       *
    174       * This is morally a \c static \c const \c Scalar, but only
    175       * integers can be static constant class members in C++. The
    176       * separation constant is set to 0.1, a value taken from the
    177       * paper by Davies and Higham. */
    178     static const RealScalar separation() { return static_cast<RealScalar>(0.1); }
    179 
    180     MatrixFunction& operator=(const MatrixFunction&);
    181 };
    182 
    183 /** \brief Constructor.
    184  *
    185  * \param[in]  A       argument of matrix function, should be a square matrix.
    186  * \param[in]  atomic  class for computing matrix function of atomic blocks.
    187  */
    188 template <typename MatrixType, typename AtomicType>
    189 MatrixFunction<MatrixType,AtomicType,1>::MatrixFunction(const MatrixType& A, AtomicType& atomic)
    190   : m_A(A), m_atomic(atomic)
    191 {
    192   /* empty body */
    193 }
    194 
    195 /** \brief Compute the matrix function.
    196   *
    197   * \param[out] result  the function \p f applied to \p A, as
    198   * specified in the constructor.
    199   */
    200 template <typename MatrixType, typename AtomicType>
    201 template <typename ResultType>
    202 void MatrixFunction<MatrixType,AtomicType,1>::compute(ResultType& result)
    203 {
    204   computeSchurDecomposition();
    205   partitionEigenvalues();
    206   computeClusterSize();
    207   computeBlockStart();
    208   constructPermutation();
    209   permuteSchur();
    210   computeBlockAtomic();
    211   computeOffDiagonal();
    212   result = m_U * (m_fT.template triangularView<Upper>() * m_U.adjoint());
    213 }
    214 
    215 /** \brief Store the Schur decomposition of #m_A in #m_T and #m_U */
    216 template <typename MatrixType, typename AtomicType>
    217 void MatrixFunction<MatrixType,AtomicType,1>::computeSchurDecomposition()
    218 {
    219   const ComplexSchur<MatrixType> schurOfA(m_A);
    220   m_T = schurOfA.matrixT();
    221   m_U = schurOfA.matrixU();
    222 }
    223 
    224 /** \brief Partition eigenvalues in clusters of ei'vals close to each other
    225   *
    226   * This function computes #m_clusters. This is a partition of the
    227   * eigenvalues of #m_T in clusters, such that
    228   * # Any eigenvalue in a certain cluster is at most separation() away
    229   *   from another eigenvalue in the same cluster.
    230   * # The distance between two eigenvalues in different clusters is
    231   *   more than separation().
    232   * The implementation follows Algorithm 4.1 in the paper of Davies
    233   * and Higham.
    234   */
    235 template <typename MatrixType, typename AtomicType>
    236 void MatrixFunction<MatrixType,AtomicType,1>::partitionEigenvalues()
    237 {
    238   using std::abs;
    239   const Index rows = m_T.rows();
    240   VectorType diag = m_T.diagonal(); // contains eigenvalues of A
    241 
    242   for (Index i=0; i<rows; ++i) {
    243     // Find set containing diag(i), adding a new set if necessary
    244     typename ListOfClusters::iterator qi = findCluster(diag(i));
    245     if (qi == m_clusters.end()) {
    246       Cluster l;
    247       l.push_back(diag(i));
    248       m_clusters.push_back(l);
    249       qi = m_clusters.end();
    250       --qi;
    251     }
    252 
    253     // Look for other element to add to the set
    254     for (Index j=i+1; j<rows; ++j) {
    255       if (abs(diag(j) - diag(i)) <= separation() && std::find(qi->begin(), qi->end(), diag(j)) == qi->end()) {
    256         typename ListOfClusters::iterator qj = findCluster(diag(j));
    257         if (qj == m_clusters.end()) {
    258           qi->push_back(diag(j));
    259         } else {
    260           qi->insert(qi->end(), qj->begin(), qj->end());
    261           m_clusters.erase(qj);
    262         }
    263       }
    264     }
    265   }
    266 }
    267 
    268 /** \brief Find cluster in #m_clusters containing some value
    269   * \param[in] key Value to find
    270   * \returns Iterator to cluster containing \c key, or
    271   * \c m_clusters.end() if no cluster in m_clusters contains \c key.
    272   */
    273 template <typename MatrixType, typename AtomicType>
    274 typename MatrixFunction<MatrixType,AtomicType,1>::ListOfClusters::iterator MatrixFunction<MatrixType,AtomicType,1>::findCluster(Scalar key)
    275 {
    276   typename Cluster::iterator j;
    277   for (typename ListOfClusters::iterator i = m_clusters.begin(); i != m_clusters.end(); ++i) {
    278     j = std::find(i->begin(), i->end(), key);
    279     if (j != i->end())
    280       return i;
    281   }
    282   return m_clusters.end();
    283 }
    284 
    285 /** \brief Compute #m_clusterSize and #m_eivalToCluster using #m_clusters */
    286 template <typename MatrixType, typename AtomicType>
    287 void MatrixFunction<MatrixType,AtomicType,1>::computeClusterSize()
    288 {
    289   const Index rows = m_T.rows();
    290   VectorType diag = m_T.diagonal();
    291   const Index numClusters = static_cast<Index>(m_clusters.size());
    292 
    293   m_clusterSize.setZero(numClusters);
    294   m_eivalToCluster.resize(rows);
    295   Index clusterIndex = 0;
    296   for (typename ListOfClusters::const_iterator cluster = m_clusters.begin(); cluster != m_clusters.end(); ++cluster) {
    297     for (Index i = 0; i < diag.rows(); ++i) {
    298       if (std::find(cluster->begin(), cluster->end(), diag(i)) != cluster->end()) {
    299         ++m_clusterSize[clusterIndex];
    300         m_eivalToCluster[i] = clusterIndex;
    301       }
    302     }
    303     ++clusterIndex;
    304   }
    305 }
    306 
    307 /** \brief Compute #m_blockStart using #m_clusterSize */
    308 template <typename MatrixType, typename AtomicType>
    309 void MatrixFunction<MatrixType,AtomicType,1>::computeBlockStart()
    310 {
    311   m_blockStart.resize(m_clusterSize.rows());
    312   m_blockStart(0) = 0;
    313   for (Index i = 1; i < m_clusterSize.rows(); i++) {
    314     m_blockStart(i) = m_blockStart(i-1) + m_clusterSize(i-1);
    315   }
    316 }
    317 
    318 /** \brief Compute #m_permutation using #m_eivalToCluster and #m_blockStart */
    319 template <typename MatrixType, typename AtomicType>
    320 void MatrixFunction<MatrixType,AtomicType,1>::constructPermutation()
    321 {
    322   DynamicIntVectorType indexNextEntry = m_blockStart;
    323   m_permutation.resize(m_T.rows());
    324   for (Index i = 0; i < m_T.rows(); i++) {
    325     Index cluster = m_eivalToCluster[i];
    326     m_permutation[i] = indexNextEntry[cluster];
    327     ++indexNextEntry[cluster];
    328   }
    329 }
    330 
    331 /** \brief Permute Schur decomposition in #m_U and #m_T according to #m_permutation */
    332 template <typename MatrixType, typename AtomicType>
    333 void MatrixFunction<MatrixType,AtomicType,1>::permuteSchur()
    334 {
    335   IntVectorType p = m_permutation;
    336   for (Index i = 0; i < p.rows() - 1; i++) {
    337     Index j;
    338     for (j = i; j < p.rows(); j++) {
    339       if (p(j) == i) break;
    340     }
    341     eigen_assert(p(j) == i);
    342     for (Index k = j-1; k >= i; k--) {
    343       swapEntriesInSchur(k);
    344       std::swap(p.coeffRef(k), p.coeffRef(k+1));
    345     }
    346   }
    347 }
    348 
    349 /** \brief Swap rows \a index and \a index+1 in Schur decomposition in #m_U and #m_T */
    350 template <typename MatrixType, typename AtomicType>
    351 void MatrixFunction<MatrixType,AtomicType,1>::swapEntriesInSchur(Index index)
    352 {
    353   JacobiRotation<Scalar> rotation;
    354   rotation.makeGivens(m_T(index, index+1), m_T(index+1, index+1) - m_T(index, index));
    355   m_T.applyOnTheLeft(index, index+1, rotation.adjoint());
    356   m_T.applyOnTheRight(index, index+1, rotation);
    357   m_U.applyOnTheRight(index, index+1, rotation);
    358 }
    359 
    360 /** \brief Compute block diagonal part of #m_fT.
    361   *
    362   * This routine computes the matrix function applied to the block diagonal part of #m_T, with the blocking
    363   * given by #m_blockStart. The matrix function of each diagonal block is computed by #m_atomic. The
    364   * off-diagonal parts of #m_fT are set to zero.
    365   */
    366 template <typename MatrixType, typename AtomicType>
    367 void MatrixFunction<MatrixType,AtomicType,1>::computeBlockAtomic()
    368 {
    369   m_fT.resize(m_T.rows(), m_T.cols());
    370   m_fT.setZero();
    371   for (Index i = 0; i < m_clusterSize.rows(); ++i) {
    372     block(m_fT, i, i) = m_atomic.compute(block(m_T, i, i));
    373   }
    374 }
    375 
    376 /** \brief Return block of matrix according to blocking given by #m_blockStart */
    377 template <typename MatrixType, typename AtomicType>
    378 Block<MatrixType> MatrixFunction<MatrixType,AtomicType,1>::block(MatrixType& A, Index i, Index j)
    379 {
    380   return A.block(m_blockStart(i), m_blockStart(j), m_clusterSize(i), m_clusterSize(j));
    381 }
    382 
    383 /** \brief Compute part of #m_fT above block diagonal.
    384   *
    385   * This routine assumes that the block diagonal part of #m_fT (which
    386   * equals the matrix function applied to #m_T) has already been computed and computes
    387   * the part above the block diagonal. The part below the diagonal is
    388   * zero, because #m_T is upper triangular.
    389   */
    390 template <typename MatrixType, typename AtomicType>
    391 void MatrixFunction<MatrixType,AtomicType,1>::computeOffDiagonal()
    392 {
    393   for (Index diagIndex = 1; diagIndex < m_clusterSize.rows(); diagIndex++) {
    394     for (Index blockIndex = 0; blockIndex < m_clusterSize.rows() - diagIndex; blockIndex++) {
    395       // compute (blockIndex, blockIndex+diagIndex) block
    396       DynMatrixType A = block(m_T, blockIndex, blockIndex);
    397       DynMatrixType B = -block(m_T, blockIndex+diagIndex, blockIndex+diagIndex);
    398       DynMatrixType C = block(m_fT, blockIndex, blockIndex) * block(m_T, blockIndex, blockIndex+diagIndex);
    399       C -= block(m_T, blockIndex, blockIndex+diagIndex) * block(m_fT, blockIndex+diagIndex, blockIndex+diagIndex);
    400       for (Index k = blockIndex + 1; k < blockIndex + diagIndex; k++) {
    401 	C += block(m_fT, blockIndex, k) * block(m_T, k, blockIndex+diagIndex);
    402 	C -= block(m_T, blockIndex, k) * block(m_fT, k, blockIndex+diagIndex);
    403       }
    404       block(m_fT, blockIndex, blockIndex+diagIndex) = solveTriangularSylvester(A, B, C);
    405     }
    406   }
    407 }
    408 
    409 /** \brief Solve a triangular Sylvester equation AX + XB = C
    410   *
    411   * \param[in]  A  the matrix A; should be square and upper triangular
    412   * \param[in]  B  the matrix B; should be square and upper triangular
    413   * \param[in]  C  the matrix C; should have correct size.
    414   *
    415   * \returns the solution X.
    416   *
    417   * If A is m-by-m and B is n-by-n, then both C and X are m-by-n.
    418   * The (i,j)-th component of the Sylvester equation is
    419   * \f[
    420   *     \sum_{k=i}^m A_{ik} X_{kj} + \sum_{k=1}^j X_{ik} B_{kj} = C_{ij}.
    421   * \f]
    422   * This can be re-arranged to yield:
    423   * \f[
    424   *     X_{ij} = \frac{1}{A_{ii} + B_{jj}} \Bigl( C_{ij}
    425   *     - \sum_{k=i+1}^m A_{ik} X_{kj} - \sum_{k=1}^{j-1} X_{ik} B_{kj} \Bigr).
    426   * \f]
    427   * It is assumed that A and B are such that the numerator is never
    428   * zero (otherwise the Sylvester equation does not have a unique
    429   * solution). In that case, these equations can be evaluated in the
    430   * order \f$ i=m,\ldots,1 \f$ and \f$ j=1,\ldots,n \f$.
    431   */
    432 template <typename MatrixType, typename AtomicType>
    433 typename MatrixFunction<MatrixType,AtomicType,1>::DynMatrixType MatrixFunction<MatrixType,AtomicType,1>::solveTriangularSylvester(
    434   const DynMatrixType& A,
    435   const DynMatrixType& B,
    436   const DynMatrixType& C)
    437 {
    438   eigen_assert(A.rows() == A.cols());
    439   eigen_assert(A.isUpperTriangular());
    440   eigen_assert(B.rows() == B.cols());
    441   eigen_assert(B.isUpperTriangular());
    442   eigen_assert(C.rows() == A.rows());
    443   eigen_assert(C.cols() == B.rows());
    444 
    445   Index m = A.rows();
    446   Index n = B.rows();
    447   DynMatrixType X(m, n);
    448 
    449   for (Index i = m - 1; i >= 0; --i) {
    450     for (Index j = 0; j < n; ++j) {
    451 
    452       // Compute AX = \sum_{k=i+1}^m A_{ik} X_{kj}
    453       Scalar AX;
    454       if (i == m - 1) {
    455 	AX = 0;
    456       } else {
    457 	Matrix<Scalar,1,1> AXmatrix = A.row(i).tail(m-1-i) * X.col(j).tail(m-1-i);
    458 	AX = AXmatrix(0,0);
    459       }
    460 
    461       // Compute XB = \sum_{k=1}^{j-1} X_{ik} B_{kj}
    462       Scalar XB;
    463       if (j == 0) {
    464 	XB = 0;
    465       } else {
    466 	Matrix<Scalar,1,1> XBmatrix = X.row(i).head(j) * B.col(j).head(j);
    467 	XB = XBmatrix(0,0);
    468       }
    469 
    470       X(i,j) = (C(i,j) - AX - XB) / (A(i,i) + B(j,j));
    471     }
    472   }
    473   return X;
    474 }
    475 
    476 /** \ingroup MatrixFunctions_Module
    477   *
    478   * \brief Proxy for the matrix function of some matrix (expression).
    479   *
    480   * \tparam Derived  Type of the argument to the matrix function.
    481   *
    482   * This class holds the argument to the matrix function until it is
    483   * assigned or evaluated for some other reason (so the argument
    484   * should not be changed in the meantime). It is the return type of
    485   * matrixBase::matrixFunction() and related functions and most of the
    486   * time this is the only way it is used.
    487   */
    488 template<typename Derived> class MatrixFunctionReturnValue
    489 : public ReturnByValue<MatrixFunctionReturnValue<Derived> >
    490 {
    491   public:
    492 
    493     typedef typename Derived::Scalar Scalar;
    494     typedef typename Derived::Index Index;
    495     typedef typename internal::stem_function<Scalar>::type StemFunction;
    496 
    497    /** \brief Constructor.
    498       *
    499       * \param[in] A  %Matrix (expression) forming the argument of the
    500       * matrix function.
    501       * \param[in] f  Stem function for matrix function under consideration.
    502       */
    503     MatrixFunctionReturnValue(const Derived& A, StemFunction f) : m_A(A), m_f(f) { }
    504 
    505     /** \brief Compute the matrix function.
    506       *
    507       * \param[out] result \p f applied to \p A, where \p f and \p A
    508       * are as in the constructor.
    509       */
    510     template <typename ResultType>
    511     inline void evalTo(ResultType& result) const
    512     {
    513       typedef typename Derived::PlainObject PlainObject;
    514       typedef internal::traits<PlainObject> Traits;
    515       static const int RowsAtCompileTime = Traits::RowsAtCompileTime;
    516       static const int ColsAtCompileTime = Traits::ColsAtCompileTime;
    517       static const int Options = PlainObject::Options;
    518       typedef std::complex<typename NumTraits<Scalar>::Real> ComplexScalar;
    519       typedef Matrix<ComplexScalar, Dynamic, Dynamic, Options, RowsAtCompileTime, ColsAtCompileTime> DynMatrixType;
    520       typedef MatrixFunctionAtomic<DynMatrixType> AtomicType;
    521       AtomicType atomic(m_f);
    522 
    523       const PlainObject Aevaluated = m_A.eval();
    524       MatrixFunction<PlainObject, AtomicType> mf(Aevaluated, atomic);
    525       mf.compute(result);
    526     }
    527 
    528     Index rows() const { return m_A.rows(); }
    529     Index cols() const { return m_A.cols(); }
    530 
    531   private:
    532     typename internal::nested<Derived>::type m_A;
    533     StemFunction *m_f;
    534 
    535     MatrixFunctionReturnValue& operator=(const MatrixFunctionReturnValue&);
    536 };
    537 
    538 namespace internal {
    539 template<typename Derived>
    540 struct traits<MatrixFunctionReturnValue<Derived> >
    541 {
    542   typedef typename Derived::PlainObject ReturnType;
    543 };
    544 }
    545 
    546 
    547 /********** MatrixBase methods **********/
    548 
    549 
    550 template <typename Derived>
    551 const MatrixFunctionReturnValue<Derived> MatrixBase<Derived>::matrixFunction(typename internal::stem_function<typename internal::traits<Derived>::Scalar>::type f) const
    552 {
    553   eigen_assert(rows() == cols());
    554   return MatrixFunctionReturnValue<Derived>(derived(), f);
    555 }
    556 
    557 template <typename Derived>
    558 const MatrixFunctionReturnValue<Derived> MatrixBase<Derived>::sin() const
    559 {
    560   eigen_assert(rows() == cols());
    561   typedef typename internal::stem_function<Scalar>::ComplexScalar ComplexScalar;
    562   return MatrixFunctionReturnValue<Derived>(derived(), StdStemFunctions<ComplexScalar>::sin);
    563 }
    564 
    565 template <typename Derived>
    566 const MatrixFunctionReturnValue<Derived> MatrixBase<Derived>::cos() const
    567 {
    568   eigen_assert(rows() == cols());
    569   typedef typename internal::stem_function<Scalar>::ComplexScalar ComplexScalar;
    570   return MatrixFunctionReturnValue<Derived>(derived(), StdStemFunctions<ComplexScalar>::cos);
    571 }
    572 
    573 template <typename Derived>
    574 const MatrixFunctionReturnValue<Derived> MatrixBase<Derived>::sinh() const
    575 {
    576   eigen_assert(rows() == cols());
    577   typedef typename internal::stem_function<Scalar>::ComplexScalar ComplexScalar;
    578   return MatrixFunctionReturnValue<Derived>(derived(), StdStemFunctions<ComplexScalar>::sinh);
    579 }
    580 
    581 template <typename Derived>
    582 const MatrixFunctionReturnValue<Derived> MatrixBase<Derived>::cosh() const
    583 {
    584   eigen_assert(rows() == cols());
    585   typedef typename internal::stem_function<Scalar>::ComplexScalar ComplexScalar;
    586   return MatrixFunctionReturnValue<Derived>(derived(), StdStemFunctions<ComplexScalar>::cosh);
    587 }
    588 
    589 } // end namespace Eigen
    590 
    591 #endif // EIGEN_MATRIX_FUNCTION
    592