• 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) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
5 // Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
6 //
7 // This Source Code Form is subject to the terms of the Mozilla
8 // Public License v. 2.0. If a copy of the MPL was not distributed
9 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
10 
11 #ifndef EIGEN_HOUSEHOLDER_SEQUENCE_H
12 #define EIGEN_HOUSEHOLDER_SEQUENCE_H
13 
14 namespace Eigen {
15 
16 /** \ingroup Householder_Module
17   * \householder_module
18   * \class HouseholderSequence
19   * \brief Sequence of Householder reflections acting on subspaces with decreasing size
20   * \tparam VectorsType type of matrix containing the Householder vectors
21   * \tparam CoeffsType  type of vector containing the Householder coefficients
22   * \tparam Side        either OnTheLeft (the default) or OnTheRight
23   *
24   * This class represents a product sequence of Householder reflections where the first Householder reflection
25   * acts on the whole space, the second Householder reflection leaves the one-dimensional subspace spanned by
26   * the first unit vector invariant, the third Householder reflection leaves the two-dimensional subspace
27   * spanned by the first two unit vectors invariant, and so on up to the last reflection which leaves all but
28   * one dimensions invariant and acts only on the last dimension. Such sequences of Householder reflections
29   * are used in several algorithms to zero out certain parts of a matrix. Indeed, the methods
30   * HessenbergDecomposition::matrixQ(), Tridiagonalization::matrixQ(), HouseholderQR::householderQ(),
31   * and ColPivHouseholderQR::householderQ() all return a %HouseholderSequence.
32   *
33   * More precisely, the class %HouseholderSequence represents an \f$ n \times n \f$ matrix \f$ H \f$ of the
34   * form \f$ H = \prod_{i=0}^{n-1} H_i \f$ where the i-th Householder reflection is \f$ H_i = I - h_i v_i
35   * v_i^* \f$. The i-th Householder coefficient \f$ h_i \f$ is a scalar and the i-th Householder vector \f$
36   * v_i \f$ is a vector of the form
37   * \f[
38   * v_i = [\underbrace{0, \ldots, 0}_{i-1\mbox{ zeros}}, 1, \underbrace{*, \ldots,*}_{n-i\mbox{ arbitrary entries}} ].
39   * \f]
40   * The last \f$ n-i \f$ entries of \f$ v_i \f$ are called the essential part of the Householder vector.
41   *
42   * Typical usages are listed below, where H is a HouseholderSequence:
43   * \code
44   * A.applyOnTheRight(H);             // A = A * H
45   * A.applyOnTheLeft(H);              // A = H * A
46   * A.applyOnTheRight(H.adjoint());   // A = A * H^*
47   * A.applyOnTheLeft(H.adjoint());    // A = H^* * A
48   * MatrixXd Q = H;                   // conversion to a dense matrix
49   * \endcode
50   * In addition to the adjoint, you can also apply the inverse (=adjoint), the transpose, and the conjugate operators.
51   *
52   * See the documentation for HouseholderSequence(const VectorsType&, const CoeffsType&) for an example.
53   *
54   * \sa MatrixBase::applyOnTheLeft(), MatrixBase::applyOnTheRight()
55   */
56 
57 namespace internal {
58 
59 template<typename VectorsType, typename CoeffsType, int Side>
60 struct traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
61 {
62   typedef typename VectorsType::Scalar Scalar;
63   typedef typename VectorsType::StorageIndex StorageIndex;
64   typedef typename VectorsType::StorageKind StorageKind;
65   enum {
66     RowsAtCompileTime = Side==OnTheLeft ? traits<VectorsType>::RowsAtCompileTime
67                                         : traits<VectorsType>::ColsAtCompileTime,
68     ColsAtCompileTime = RowsAtCompileTime,
69     MaxRowsAtCompileTime = Side==OnTheLeft ? traits<VectorsType>::MaxRowsAtCompileTime
70                                            : traits<VectorsType>::MaxColsAtCompileTime,
71     MaxColsAtCompileTime = MaxRowsAtCompileTime,
72     Flags = 0
73   };
74 };
75 
76 struct HouseholderSequenceShape {};
77 
78 template<typename VectorsType, typename CoeffsType, int Side>
79 struct evaluator_traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
80   : public evaluator_traits_base<HouseholderSequence<VectorsType,CoeffsType,Side> >
81 {
82   typedef HouseholderSequenceShape Shape;
83 };
84 
85 template<typename VectorsType, typename CoeffsType, int Side>
86 struct hseq_side_dependent_impl
87 {
88   typedef Block<const VectorsType, Dynamic, 1> EssentialVectorType;
89   typedef HouseholderSequence<VectorsType, CoeffsType, OnTheLeft> HouseholderSequenceType;
90   static inline const EssentialVectorType essentialVector(const HouseholderSequenceType& h, Index k)
91   {
92     Index start = k+1+h.m_shift;
93     return Block<const VectorsType,Dynamic,1>(h.m_vectors, start, k, h.rows()-start, 1);
94   }
95 };
96 
97 template<typename VectorsType, typename CoeffsType>
98 struct hseq_side_dependent_impl<VectorsType, CoeffsType, OnTheRight>
99 {
100   typedef Transpose<Block<const VectorsType, 1, Dynamic> > EssentialVectorType;
101   typedef HouseholderSequence<VectorsType, CoeffsType, OnTheRight> HouseholderSequenceType;
102   static inline const EssentialVectorType essentialVector(const HouseholderSequenceType& h, Index k)
103   {
104     Index start = k+1+h.m_shift;
105     return Block<const VectorsType,1,Dynamic>(h.m_vectors, k, start, 1, h.rows()-start).transpose();
106   }
107 };
108 
109 template<typename OtherScalarType, typename MatrixType> struct matrix_type_times_scalar_type
110 {
111   typedef typename ScalarBinaryOpTraits<OtherScalarType, typename MatrixType::Scalar>::ReturnType
112     ResultScalar;
113   typedef Matrix<ResultScalar, MatrixType::RowsAtCompileTime, MatrixType::ColsAtCompileTime,
114                  0, MatrixType::MaxRowsAtCompileTime, MatrixType::MaxColsAtCompileTime> Type;
115 };
116 
117 } // end namespace internal
118 
119 template<typename VectorsType, typename CoeffsType, int Side> class HouseholderSequence
120   : public EigenBase<HouseholderSequence<VectorsType,CoeffsType,Side> >
121 {
122     typedef typename internal::hseq_side_dependent_impl<VectorsType,CoeffsType,Side>::EssentialVectorType EssentialVectorType;
123 
124   public:
125     enum {
126       RowsAtCompileTime = internal::traits<HouseholderSequence>::RowsAtCompileTime,
127       ColsAtCompileTime = internal::traits<HouseholderSequence>::ColsAtCompileTime,
128       MaxRowsAtCompileTime = internal::traits<HouseholderSequence>::MaxRowsAtCompileTime,
129       MaxColsAtCompileTime = internal::traits<HouseholderSequence>::MaxColsAtCompileTime
130     };
131     typedef typename internal::traits<HouseholderSequence>::Scalar Scalar;
132 
133     typedef HouseholderSequence<
134       typename internal::conditional<NumTraits<Scalar>::IsComplex,
135         typename internal::remove_all<typename VectorsType::ConjugateReturnType>::type,
136         VectorsType>::type,
137       typename internal::conditional<NumTraits<Scalar>::IsComplex,
138         typename internal::remove_all<typename CoeffsType::ConjugateReturnType>::type,
139         CoeffsType>::type,
140       Side
141     > ConjugateReturnType;
142 
143     /** \brief Constructor.
144       * \param[in]  v      %Matrix containing the essential parts of the Householder vectors
145       * \param[in]  h      Vector containing the Householder coefficients
146       *
147       * Constructs the Householder sequence with coefficients given by \p h and vectors given by \p v. The
148       * i-th Householder coefficient \f$ h_i \f$ is given by \p h(i) and the essential part of the i-th
149       * Householder vector \f$ v_i \f$ is given by \p v(k,i) with \p k > \p i (the subdiagonal part of the
150       * i-th column). If \p v has fewer columns than rows, then the Householder sequence contains as many
151       * Householder reflections as there are columns.
152       *
153       * \note The %HouseholderSequence object stores \p v and \p h by reference.
154       *
155       * Example: \include HouseholderSequence_HouseholderSequence.cpp
156       * Output: \verbinclude HouseholderSequence_HouseholderSequence.out
157       *
158       * \sa setLength(), setShift()
159       */
160     HouseholderSequence(const VectorsType& v, const CoeffsType& h)
161       : m_vectors(v), m_coeffs(h), m_trans(false), m_length(v.diagonalSize()),
162         m_shift(0)
163     {
164     }
165 
166     /** \brief Copy constructor. */
167     HouseholderSequence(const HouseholderSequence& other)
168       : m_vectors(other.m_vectors),
169         m_coeffs(other.m_coeffs),
170         m_trans(other.m_trans),
171         m_length(other.m_length),
172         m_shift(other.m_shift)
173     {
174     }
175 
176     /** \brief Number of rows of transformation viewed as a matrix.
177       * \returns Number of rows
178       * \details This equals the dimension of the space that the transformation acts on.
179       */
180     Index rows() const { return Side==OnTheLeft ? m_vectors.rows() : m_vectors.cols(); }
181 
182     /** \brief Number of columns of transformation viewed as a matrix.
183       * \returns Number of columns
184       * \details This equals the dimension of the space that the transformation acts on.
185       */
186     Index cols() const { return rows(); }
187 
188     /** \brief Essential part of a Householder vector.
189       * \param[in]  k  Index of Householder reflection
190       * \returns    Vector containing non-trivial entries of k-th Householder vector
191       *
192       * This function returns the essential part of the Householder vector \f$ v_i \f$. This is a vector of
193       * length \f$ n-i \f$ containing the last \f$ n-i \f$ entries of the vector
194       * \f[
195       * v_i = [\underbrace{0, \ldots, 0}_{i-1\mbox{ zeros}}, 1, \underbrace{*, \ldots,*}_{n-i\mbox{ arbitrary entries}} ].
196       * \f]
197       * The index \f$ i \f$ equals \p k + shift(), corresponding to the k-th column of the matrix \p v
198       * passed to the constructor.
199       *
200       * \sa setShift(), shift()
201       */
202     const EssentialVectorType essentialVector(Index k) const
203     {
204       eigen_assert(k >= 0 && k < m_length);
205       return internal::hseq_side_dependent_impl<VectorsType,CoeffsType,Side>::essentialVector(*this, k);
206     }
207 
208     /** \brief %Transpose of the Householder sequence. */
209     HouseholderSequence transpose() const
210     {
211       return HouseholderSequence(*this).setTrans(!m_trans);
212     }
213 
214     /** \brief Complex conjugate of the Householder sequence. */
215     ConjugateReturnType conjugate() const
216     {
217       return ConjugateReturnType(m_vectors.conjugate(), m_coeffs.conjugate())
218              .setTrans(m_trans)
219              .setLength(m_length)
220              .setShift(m_shift);
221     }
222 
223     /** \brief Adjoint (conjugate transpose) of the Householder sequence. */
224     ConjugateReturnType adjoint() const
225     {
226       return conjugate().setTrans(!m_trans);
227     }
228 
229     /** \brief Inverse of the Householder sequence (equals the adjoint). */
230     ConjugateReturnType inverse() const { return adjoint(); }
231 
232     /** \internal */
233     template<typename DestType> inline void evalTo(DestType& dst) const
234     {
235       Matrix<Scalar, DestType::RowsAtCompileTime, 1,
236              AutoAlign|ColMajor, DestType::MaxRowsAtCompileTime, 1> workspace(rows());
237       evalTo(dst, workspace);
238     }
239 
240     /** \internal */
241     template<typename Dest, typename Workspace>
242     void evalTo(Dest& dst, Workspace& workspace) const
243     {
244       workspace.resize(rows());
245       Index vecs = m_length;
246       if(internal::is_same_dense(dst,m_vectors))
247       {
248         // in-place
249         dst.diagonal().setOnes();
250         dst.template triangularView<StrictlyUpper>().setZero();
251         for(Index k = vecs-1; k >= 0; --k)
252         {
253           Index cornerSize = rows() - k - m_shift;
254           if(m_trans)
255             dst.bottomRightCorner(cornerSize, cornerSize)
256                .applyHouseholderOnTheRight(essentialVector(k), m_coeffs.coeff(k), workspace.data());
257           else
258             dst.bottomRightCorner(cornerSize, cornerSize)
259                .applyHouseholderOnTheLeft(essentialVector(k), m_coeffs.coeff(k), workspace.data());
260 
261           // clear the off diagonal vector
262           dst.col(k).tail(rows()-k-1).setZero();
263         }
264         // clear the remaining columns if needed
265         for(Index k = 0; k<cols()-vecs ; ++k)
266           dst.col(k).tail(rows()-k-1).setZero();
267       }
268       else
269       {
270         dst.setIdentity(rows(), rows());
271         for(Index k = vecs-1; k >= 0; --k)
272         {
273           Index cornerSize = rows() - k - m_shift;
274           if(m_trans)
275             dst.bottomRightCorner(cornerSize, cornerSize)
276                .applyHouseholderOnTheRight(essentialVector(k), m_coeffs.coeff(k), &workspace.coeffRef(0));
277           else
278             dst.bottomRightCorner(cornerSize, cornerSize)
279                .applyHouseholderOnTheLeft(essentialVector(k), m_coeffs.coeff(k), &workspace.coeffRef(0));
280         }
281       }
282     }
283 
284     /** \internal */
285     template<typename Dest> inline void applyThisOnTheRight(Dest& dst) const
286     {
287       Matrix<Scalar,1,Dest::RowsAtCompileTime,RowMajor,1,Dest::MaxRowsAtCompileTime> workspace(dst.rows());
288       applyThisOnTheRight(dst, workspace);
289     }
290 
291     /** \internal */
292     template<typename Dest, typename Workspace>
293     inline void applyThisOnTheRight(Dest& dst, Workspace& workspace) const
294     {
295       workspace.resize(dst.rows());
296       for(Index k = 0; k < m_length; ++k)
297       {
298         Index actual_k = m_trans ? m_length-k-1 : k;
299         dst.rightCols(rows()-m_shift-actual_k)
300            .applyHouseholderOnTheRight(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
301       }
302     }
303 
304     /** \internal */
305     template<typename Dest> inline void applyThisOnTheLeft(Dest& dst) const
306     {
307       Matrix<Scalar,1,Dest::ColsAtCompileTime,RowMajor,1,Dest::MaxColsAtCompileTime> workspace;
308       applyThisOnTheLeft(dst, workspace);
309     }
310 
311     /** \internal */
312     template<typename Dest, typename Workspace>
313     inline void applyThisOnTheLeft(Dest& dst, Workspace& workspace) const
314     {
315       const Index BlockSize = 48;
316       // if the entries are large enough, then apply the reflectors by block
317       if(m_length>=BlockSize && dst.cols()>1)
318       {
319         for(Index i = 0; i < m_length; i+=BlockSize)
320         {
321           Index end = m_trans ? (std::min)(m_length,i+BlockSize) : m_length-i;
322           Index k = m_trans ? i : (std::max)(Index(0),end-BlockSize);
323           Index bs = end-k;
324           Index start = k + m_shift;
325 
326           typedef Block<typename internal::remove_all<VectorsType>::type,Dynamic,Dynamic> SubVectorsType;
327           SubVectorsType sub_vecs1(m_vectors.const_cast_derived(), Side==OnTheRight ? k : start,
328                                                                    Side==OnTheRight ? start : k,
329                                                                    Side==OnTheRight ? bs : m_vectors.rows()-start,
330                                                                    Side==OnTheRight ? m_vectors.cols()-start : bs);
331           typename internal::conditional<Side==OnTheRight, Transpose<SubVectorsType>, SubVectorsType&>::type sub_vecs(sub_vecs1);
332           Block<Dest,Dynamic,Dynamic> sub_dst(dst,dst.rows()-rows()+m_shift+k,0, rows()-m_shift-k,dst.cols());
333           apply_block_householder_on_the_left(sub_dst, sub_vecs, m_coeffs.segment(k, bs), !m_trans);
334         }
335       }
336       else
337       {
338         workspace.resize(dst.cols());
339         for(Index k = 0; k < m_length; ++k)
340         {
341           Index actual_k = m_trans ? k : m_length-k-1;
342           dst.bottomRows(rows()-m_shift-actual_k)
343             .applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
344         }
345       }
346     }
347 
348     /** \brief Computes the product of a Householder sequence with a matrix.
349       * \param[in]  other  %Matrix being multiplied.
350       * \returns    Expression object representing the product.
351       *
352       * This function computes \f$ HM \f$ where \f$ H \f$ is the Householder sequence represented by \p *this
353       * and \f$ M \f$ is the matrix \p other.
354       */
355     template<typename OtherDerived>
356     typename internal::matrix_type_times_scalar_type<Scalar, OtherDerived>::Type operator*(const MatrixBase<OtherDerived>& other) const
357     {
358       typename internal::matrix_type_times_scalar_type<Scalar, OtherDerived>::Type
359         res(other.template cast<typename internal::matrix_type_times_scalar_type<Scalar,OtherDerived>::ResultScalar>());
360       applyThisOnTheLeft(res);
361       return res;
362     }
363 
364     template<typename _VectorsType, typename _CoeffsType, int _Side> friend struct internal::hseq_side_dependent_impl;
365 
366     /** \brief Sets the length of the Householder sequence.
367       * \param [in]  length  New value for the length.
368       *
369       * By default, the length \f$ n \f$ of the Householder sequence \f$ H = H_0 H_1 \ldots H_{n-1} \f$ is set
370       * to the number of columns of the matrix \p v passed to the constructor, or the number of rows if that
371       * is smaller. After this function is called, the length equals \p length.
372       *
373       * \sa length()
374       */
375     HouseholderSequence& setLength(Index length)
376     {
377       m_length = length;
378       return *this;
379     }
380 
381     /** \brief Sets the shift of the Householder sequence.
382       * \param [in]  shift  New value for the shift.
383       *
384       * By default, a %HouseholderSequence object represents \f$ H = H_0 H_1 \ldots H_{n-1} \f$ and the i-th
385       * column of the matrix \p v passed to the constructor corresponds to the i-th Householder
386       * reflection. After this function is called, the object represents \f$ H = H_{\mathrm{shift}}
387       * H_{\mathrm{shift}+1} \ldots H_{n-1} \f$ and the i-th column of \p v corresponds to the (shift+i)-th
388       * Householder reflection.
389       *
390       * \sa shift()
391       */
392     HouseholderSequence& setShift(Index shift)
393     {
394       m_shift = shift;
395       return *this;
396     }
397 
398     Index length() const { return m_length; }  /**< \brief Returns the length of the Householder sequence. */
399     Index shift() const { return m_shift; }    /**< \brief Returns the shift of the Householder sequence. */
400 
401     /* Necessary for .adjoint() and .conjugate() */
402     template <typename VectorsType2, typename CoeffsType2, int Side2> friend class HouseholderSequence;
403 
404   protected:
405 
406     /** \brief Sets the transpose flag.
407       * \param [in]  trans  New value of the transpose flag.
408       *
409       * By default, the transpose flag is not set. If the transpose flag is set, then this object represents
410       * \f$ H^T = H_{n-1}^T \ldots H_1^T H_0^T \f$ instead of \f$ H = H_0 H_1 \ldots H_{n-1} \f$.
411       *
412       * \sa trans()
413       */
414     HouseholderSequence& setTrans(bool trans)
415     {
416       m_trans = trans;
417       return *this;
418     }
419 
420     bool trans() const { return m_trans; }     /**< \brief Returns the transpose flag. */
421 
422     typename VectorsType::Nested m_vectors;
423     typename CoeffsType::Nested m_coeffs;
424     bool m_trans;
425     Index m_length;
426     Index m_shift;
427 };
428 
429 /** \brief Computes the product of a matrix with a Householder sequence.
430   * \param[in]  other  %Matrix being multiplied.
431   * \param[in]  h      %HouseholderSequence being multiplied.
432   * \returns    Expression object representing the product.
433   *
434   * This function computes \f$ MH \f$ where \f$ M \f$ is the matrix \p other and \f$ H \f$ is the
435   * Householder sequence represented by \p h.
436   */
437 template<typename OtherDerived, typename VectorsType, typename CoeffsType, int Side>
438 typename internal::matrix_type_times_scalar_type<typename VectorsType::Scalar,OtherDerived>::Type operator*(const MatrixBase<OtherDerived>& other, const HouseholderSequence<VectorsType,CoeffsType,Side>& h)
439 {
440   typename internal::matrix_type_times_scalar_type<typename VectorsType::Scalar,OtherDerived>::Type
441     res(other.template cast<typename internal::matrix_type_times_scalar_type<typename VectorsType::Scalar,OtherDerived>::ResultScalar>());
442   h.applyThisOnTheRight(res);
443   return res;
444 }
445 
446 /** \ingroup Householder_Module \householder_module
447   * \brief Convenience function for constructing a Householder sequence.
448   * \returns A HouseholderSequence constructed from the specified arguments.
449   */
450 template<typename VectorsType, typename CoeffsType>
451 HouseholderSequence<VectorsType,CoeffsType> householderSequence(const VectorsType& v, const CoeffsType& h)
452 {
453   return HouseholderSequence<VectorsType,CoeffsType,OnTheLeft>(v, h);
454 }
455 
456 /** \ingroup Householder_Module \householder_module
457   * \brief Convenience function for constructing a Householder sequence.
458   * \returns A HouseholderSequence constructed from the specified arguments.
459   * \details This function differs from householderSequence() in that the template argument \p OnTheSide of
460   * the constructed HouseholderSequence is set to OnTheRight, instead of the default OnTheLeft.
461   */
462 template<typename VectorsType, typename CoeffsType>
463 HouseholderSequence<VectorsType,CoeffsType,OnTheRight> rightHouseholderSequence(const VectorsType& v, const CoeffsType& h)
464 {
465   return HouseholderSequence<VectorsType,CoeffsType,OnTheRight>(v, h);
466 }
467 
468 } // end namespace Eigen
469 
470 #endif // EIGEN_HOUSEHOLDER_SEQUENCE_H
471