1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2009 Claire Maurice
5 // Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
6 // Copyright (C) 2010,2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
7 //
8 // This Source Code Form is subject to the terms of the Mozilla
9 // Public License v. 2.0. If a copy of the MPL was not distributed
10 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
11 
12 #ifndef EIGEN_COMPLEX_SCHUR_H
13 #define EIGEN_COMPLEX_SCHUR_H
14 
15 #include "./HessenbergDecomposition.h"
16 
17 namespace Eigen {
18 
19 namespace internal {
20 template<typename MatrixType, bool IsComplex> struct complex_schur_reduce_to_hessenberg;
21 }
22 
23 /** \eigenvalues_module \ingroup Eigenvalues_Module
24   *
25   *
26   * \class ComplexSchur
27   *
28   * \brief Performs a complex Schur decomposition of a real or complex square matrix
29   *
30   * \tparam _MatrixType the type of the matrix of which we are
31   * computing the Schur decomposition; this is expected to be an
32   * instantiation of the Matrix class template.
33   *
34   * Given a real or complex square matrix A, this class computes the
35   * Schur decomposition: \f$ A = U T U^*\f$ where U is a unitary
36   * complex matrix, and T is a complex upper triangular matrix.  The
37   * diagonal of the matrix T corresponds to the eigenvalues of the
38   * matrix A.
39   *
40   * Call the function compute() to compute the Schur decomposition of
41   * a given matrix. Alternatively, you can use the
42   * ComplexSchur(const MatrixType&, bool) constructor which computes
43   * the Schur decomposition at construction time. Once the
44   * decomposition is computed, you can use the matrixU() and matrixT()
45   * functions to retrieve the matrices U and V in the decomposition.
46   *
47   * \note This code is inspired from Jampack
48   *
49   * \sa class RealSchur, class EigenSolver, class ComplexEigenSolver
50   */
51 template<typename _MatrixType> class ComplexSchur
52 {
53   public:
54     typedef _MatrixType MatrixType;
55     enum {
56       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
57       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
58       Options = MatrixType::Options,
59       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
60       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
61     };
62 
63     /** \brief Scalar type for matrices of type \p _MatrixType. */
64     typedef typename MatrixType::Scalar Scalar;
65     typedef typename NumTraits<Scalar>::Real RealScalar;
66     typedef typename MatrixType::Index Index;
67 
68     /** \brief Complex scalar type for \p _MatrixType.
69       *
70       * This is \c std::complex<Scalar> if #Scalar is real (e.g.,
71       * \c float or \c double) and just \c Scalar if #Scalar is
72       * complex.
73       */
74     typedef std::complex<RealScalar> ComplexScalar;
75 
76     /** \brief Type for the matrices in the Schur decomposition.
77       *
78       * This is a square matrix with entries of type #ComplexScalar.
79       * The size is the same as the size of \p _MatrixType.
80       */
81     typedef Matrix<ComplexScalar, RowsAtCompileTime, ColsAtCompileTime, Options, MaxRowsAtCompileTime, MaxColsAtCompileTime> ComplexMatrixType;
82 
83     /** \brief Default constructor.
84       *
85       * \param [in] size  Positive integer, size of the matrix whose Schur decomposition will be computed.
86       *
87       * The default constructor is useful in cases in which the user
88       * intends to perform decompositions via compute().  The \p size
89       * parameter is only used as a hint. It is not an error to give a
90       * wrong \p size, but it may impair performance.
91       *
92       * \sa compute() for an example.
93       */
94     ComplexSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
m_matT(size,size)95       : m_matT(size,size),
96         m_matU(size,size),
97         m_hess(size),
98         m_isInitialized(false),
99         m_matUisUptodate(false),
100         m_maxIters(-1)
101     {}
102 
103     /** \brief Constructor; computes Schur decomposition of given matrix.
104       *
105       * \param[in]  matrix    Square matrix whose Schur decomposition is to be computed.
106       * \param[in]  computeU  If true, both T and U are computed; if false, only T is computed.
107       *
108       * This constructor calls compute() to compute the Schur decomposition.
109       *
110       * \sa matrixT() and matrixU() for examples.
111       */
112     ComplexSchur(const MatrixType& matrix, bool computeU = true)
113       : m_matT(matrix.rows(),matrix.cols()),
114         m_matU(matrix.rows(),matrix.cols()),
115         m_hess(matrix.rows()),
116         m_isInitialized(false),
117         m_matUisUptodate(false),
118         m_maxIters(-1)
119     {
120       compute(matrix, computeU);
121     }
122 
123     /** \brief Returns the unitary matrix in the Schur decomposition.
124       *
125       * \returns A const reference to the matrix U.
126       *
127       * It is assumed that either the constructor
128       * ComplexSchur(const MatrixType& matrix, bool computeU) or the
129       * member function compute(const MatrixType& matrix, bool computeU)
130       * has been called before to compute the Schur decomposition of a
131       * matrix, and that \p computeU was set to true (the default
132       * value).
133       *
134       * Example: \include ComplexSchur_matrixU.cpp
135       * Output: \verbinclude ComplexSchur_matrixU.out
136       */
matrixU()137     const ComplexMatrixType& matrixU() const
138     {
139       eigen_assert(m_isInitialized && "ComplexSchur is not initialized.");
140       eigen_assert(m_matUisUptodate && "The matrix U has not been computed during the ComplexSchur decomposition.");
141       return m_matU;
142     }
143 
144     /** \brief Returns the triangular matrix in the Schur decomposition.
145       *
146       * \returns A const reference to the matrix T.
147       *
148       * It is assumed that either the constructor
149       * ComplexSchur(const MatrixType& matrix, bool computeU) or the
150       * member function compute(const MatrixType& matrix, bool computeU)
151       * has been called before to compute the Schur decomposition of a
152       * matrix.
153       *
154       * Note that this function returns a plain square matrix. If you want to reference
155       * only the upper triangular part, use:
156       * \code schur.matrixT().triangularView<Upper>() \endcode
157       *
158       * Example: \include ComplexSchur_matrixT.cpp
159       * Output: \verbinclude ComplexSchur_matrixT.out
160       */
matrixT()161     const ComplexMatrixType& matrixT() const
162     {
163       eigen_assert(m_isInitialized && "ComplexSchur is not initialized.");
164       return m_matT;
165     }
166 
167     /** \brief Computes Schur decomposition of given matrix.
168       *
169       * \param[in]  matrix  Square matrix whose Schur decomposition is to be computed.
170       * \param[in]  computeU  If true, both T and U are computed; if false, only T is computed.
171 
172       * \returns    Reference to \c *this
173       *
174       * The Schur decomposition is computed by first reducing the
175       * matrix to Hessenberg form using the class
176       * HessenbergDecomposition. The Hessenberg matrix is then reduced
177       * to triangular form by performing QR iterations with a single
178       * shift. The cost of computing the Schur decomposition depends
179       * on the number of iterations; as a rough guide, it may be taken
180       * on the number of iterations; as a rough guide, it may be taken
181       * to be \f$25n^3\f$ complex flops, or \f$10n^3\f$ complex flops
182       * if \a computeU is false.
183       *
184       * Example: \include ComplexSchur_compute.cpp
185       * Output: \verbinclude ComplexSchur_compute.out
186       *
187       * \sa compute(const MatrixType&, bool, Index)
188       */
189     ComplexSchur& compute(const MatrixType& matrix, bool computeU = true);
190 
191     /** \brief Compute Schur decomposition from a given Hessenberg matrix
192      *  \param[in] matrixH Matrix in Hessenberg form H
193      *  \param[in] matrixQ orthogonal matrix Q that transform a matrix A to H : A = Q H Q^T
194      *  \param computeU Computes the matriX U of the Schur vectors
195      * \return Reference to \c *this
196      *
197      *  This routine assumes that the matrix is already reduced in Hessenberg form matrixH
198      *  using either the class HessenbergDecomposition or another mean.
199      *  It computes the upper quasi-triangular matrix T of the Schur decomposition of H
200      *  When computeU is true, this routine computes the matrix U such that
201      *  A = U T U^T =  (QZ) T (QZ)^T = Q H Q^T where A is the initial matrix
202      *
203      * NOTE Q is referenced if computeU is true; so, if the initial orthogonal matrix
204      * is not available, the user should give an identity matrix (Q.setIdentity())
205      *
206      * \sa compute(const MatrixType&, bool)
207      */
208     template<typename HessMatrixType, typename OrthMatrixType>
209     ComplexSchur& computeFromHessenberg(const HessMatrixType& matrixH, const OrthMatrixType& matrixQ,  bool computeU=true);
210 
211     /** \brief Reports whether previous computation was successful.
212       *
213       * \returns \c Success if computation was succesful, \c NoConvergence otherwise.
214       */
info()215     ComputationInfo info() const
216     {
217       eigen_assert(m_isInitialized && "ComplexSchur is not initialized.");
218       return m_info;
219     }
220 
221     /** \brief Sets the maximum number of iterations allowed.
222       *
223       * If not specified by the user, the maximum number of iterations is m_maxIterationsPerRow times the size
224       * of the matrix.
225       */
setMaxIterations(Index maxIters)226     ComplexSchur& setMaxIterations(Index maxIters)
227     {
228       m_maxIters = maxIters;
229       return *this;
230     }
231 
232     /** \brief Returns the maximum number of iterations. */
getMaxIterations()233     Index getMaxIterations()
234     {
235       return m_maxIters;
236     }
237 
238     /** \brief Maximum number of iterations per row.
239       *
240       * If not otherwise specified, the maximum number of iterations is this number times the size of the
241       * matrix. It is currently set to 30.
242       */
243     static const int m_maxIterationsPerRow = 30;
244 
245   protected:
246     ComplexMatrixType m_matT, m_matU;
247     HessenbergDecomposition<MatrixType> m_hess;
248     ComputationInfo m_info;
249     bool m_isInitialized;
250     bool m_matUisUptodate;
251     Index m_maxIters;
252 
253   private:
254     bool subdiagonalEntryIsNeglegible(Index i);
255     ComplexScalar computeShift(Index iu, Index iter);
256     void reduceToTriangularForm(bool computeU);
257     friend struct internal::complex_schur_reduce_to_hessenberg<MatrixType, NumTraits<Scalar>::IsComplex>;
258 };
259 
260 /** If m_matT(i+1,i) is neglegible in floating point arithmetic
261   * compared to m_matT(i,i) and m_matT(j,j), then set it to zero and
262   * return true, else return false. */
263 template<typename MatrixType>
264 inline bool ComplexSchur<MatrixType>::subdiagonalEntryIsNeglegible(Index i)
265 {
266   RealScalar d = numext::norm1(m_matT.coeff(i,i)) + numext::norm1(m_matT.coeff(i+1,i+1));
267   RealScalar sd = numext::norm1(m_matT.coeff(i+1,i));
268   if (internal::isMuchSmallerThan(sd, d, NumTraits<RealScalar>::epsilon()))
269   {
270     m_matT.coeffRef(i+1,i) = ComplexScalar(0);
271     return true;
272   }
273   return false;
274 }
275 
276 
277 /** Compute the shift in the current QR iteration. */
278 template<typename MatrixType>
279 typename ComplexSchur<MatrixType>::ComplexScalar ComplexSchur<MatrixType>::computeShift(Index iu, Index iter)
280 {
281   using std::abs;
282   if (iter == 10 || iter == 20)
283   {
284     // exceptional shift, taken from http://www.netlib.org/eispack/comqr.f
285     return abs(numext::real(m_matT.coeff(iu,iu-1))) + abs(numext::real(m_matT.coeff(iu-1,iu-2)));
286   }
287 
288   // compute the shift as one of the eigenvalues of t, the 2x2
289   // diagonal block on the bottom of the active submatrix
290   Matrix<ComplexScalar,2,2> t = m_matT.template block<2,2>(iu-1,iu-1);
291   RealScalar normt = t.cwiseAbs().sum();
292   t /= normt;     // the normalization by sf is to avoid under/overflow
293 
294   ComplexScalar b = t.coeff(0,1) * t.coeff(1,0);
295   ComplexScalar c = t.coeff(0,0) - t.coeff(1,1);
296   ComplexScalar disc = sqrt(c*c + RealScalar(4)*b);
297   ComplexScalar det = t.coeff(0,0) * t.coeff(1,1) - b;
298   ComplexScalar trace = t.coeff(0,0) + t.coeff(1,1);
299   ComplexScalar eival1 = (trace + disc) / RealScalar(2);
300   ComplexScalar eival2 = (trace - disc) / RealScalar(2);
301 
302   if(numext::norm1(eival1) > numext::norm1(eival2))
303     eival2 = det / eival1;
304   else
305     eival1 = det / eival2;
306 
307   // choose the eigenvalue closest to the bottom entry of the diagonal
308   if(numext::norm1(eival1-t.coeff(1,1)) < numext::norm1(eival2-t.coeff(1,1)))
309     return normt * eival1;
310   else
311     return normt * eival2;
312 }
313 
314 
315 template<typename MatrixType>
316 ComplexSchur<MatrixType>& ComplexSchur<MatrixType>::compute(const MatrixType& matrix, bool computeU)
317 {
318   m_matUisUptodate = false;
319   eigen_assert(matrix.cols() == matrix.rows());
320 
321   if(matrix.cols() == 1)
322   {
323     m_matT = matrix.template cast<ComplexScalar>();
324     if(computeU)  m_matU = ComplexMatrixType::Identity(1,1);
325     m_info = Success;
326     m_isInitialized = true;
327     m_matUisUptodate = computeU;
328     return *this;
329   }
330 
331   internal::complex_schur_reduce_to_hessenberg<MatrixType, NumTraits<Scalar>::IsComplex>::run(*this, matrix, computeU);
332   computeFromHessenberg(m_matT, m_matU, computeU);
333   return *this;
334 }
335 
336 template<typename MatrixType>
337 template<typename HessMatrixType, typename OrthMatrixType>
338 ComplexSchur<MatrixType>& ComplexSchur<MatrixType>::computeFromHessenberg(const HessMatrixType& matrixH, const OrthMatrixType& matrixQ, bool computeU)
339 {
340   m_matT = matrixH;
341   if(computeU)
342     m_matU = matrixQ;
343   reduceToTriangularForm(computeU);
344   return *this;
345 }
346 namespace internal {
347 
348 /* Reduce given matrix to Hessenberg form */
349 template<typename MatrixType, bool IsComplex>
350 struct complex_schur_reduce_to_hessenberg
351 {
352   // this is the implementation for the case IsComplex = true
353   static void run(ComplexSchur<MatrixType>& _this, const MatrixType& matrix, bool computeU)
354   {
355     _this.m_hess.compute(matrix);
356     _this.m_matT = _this.m_hess.matrixH();
357     if(computeU)  _this.m_matU = _this.m_hess.matrixQ();
358   }
359 };
360 
361 template<typename MatrixType>
362 struct complex_schur_reduce_to_hessenberg<MatrixType, false>
363 {
364   static void run(ComplexSchur<MatrixType>& _this, const MatrixType& matrix, bool computeU)
365   {
366     typedef typename ComplexSchur<MatrixType>::ComplexScalar ComplexScalar;
367 
368     // Note: m_hess is over RealScalar; m_matT and m_matU is over ComplexScalar
369     _this.m_hess.compute(matrix);
370     _this.m_matT = _this.m_hess.matrixH().template cast<ComplexScalar>();
371     if(computeU)
372     {
373       // This may cause an allocation which seems to be avoidable
374       MatrixType Q = _this.m_hess.matrixQ();
375       _this.m_matU = Q.template cast<ComplexScalar>();
376     }
377   }
378 };
379 
380 } // end namespace internal
381 
382 // Reduce the Hessenberg matrix m_matT to triangular form by QR iteration.
383 template<typename MatrixType>
384 void ComplexSchur<MatrixType>::reduceToTriangularForm(bool computeU)
385 {
386   Index maxIters = m_maxIters;
387   if (maxIters == -1)
388     maxIters = m_maxIterationsPerRow * m_matT.rows();
389 
390   // The matrix m_matT is divided in three parts.
391   // Rows 0,...,il-1 are decoupled from the rest because m_matT(il,il-1) is zero.
392   // Rows il,...,iu is the part we are working on (the active submatrix).
393   // Rows iu+1,...,end are already brought in triangular form.
394   Index iu = m_matT.cols() - 1;
395   Index il;
396   Index iter = 0; // number of iterations we are working on the (iu,iu) element
397   Index totalIter = 0; // number of iterations for whole matrix
398 
399   while(true)
400   {
401     // find iu, the bottom row of the active submatrix
402     while(iu > 0)
403     {
404       if(!subdiagonalEntryIsNeglegible(iu-1)) break;
405       iter = 0;
406       --iu;
407     }
408 
409     // if iu is zero then we are done; the whole matrix is triangularized
410     if(iu==0) break;
411 
412     // if we spent too many iterations, we give up
413     iter++;
414     totalIter++;
415     if(totalIter > maxIters) break;
416 
417     // find il, the top row of the active submatrix
418     il = iu-1;
419     while(il > 0 && !subdiagonalEntryIsNeglegible(il-1))
420     {
421       --il;
422     }
423 
424     /* perform the QR step using Givens rotations. The first rotation
425        creates a bulge; the (il+2,il) element becomes nonzero. This
426        bulge is chased down to the bottom of the active submatrix. */
427 
428     ComplexScalar shift = computeShift(iu, iter);
429     JacobiRotation<ComplexScalar> rot;
430     rot.makeGivens(m_matT.coeff(il,il) - shift, m_matT.coeff(il+1,il));
431     m_matT.rightCols(m_matT.cols()-il).applyOnTheLeft(il, il+1, rot.adjoint());
432     m_matT.topRows((std::min)(il+2,iu)+1).applyOnTheRight(il, il+1, rot);
433     if(computeU) m_matU.applyOnTheRight(il, il+1, rot);
434 
435     for(Index i=il+1 ; i<iu ; i++)
436     {
437       rot.makeGivens(m_matT.coeffRef(i,i-1), m_matT.coeffRef(i+1,i-1), &m_matT.coeffRef(i,i-1));
438       m_matT.coeffRef(i+1,i-1) = ComplexScalar(0);
439       m_matT.rightCols(m_matT.cols()-i).applyOnTheLeft(i, i+1, rot.adjoint());
440       m_matT.topRows((std::min)(i+2,iu)+1).applyOnTheRight(i, i+1, rot);
441       if(computeU) m_matU.applyOnTheRight(i, i+1, rot);
442     }
443   }
444 
445   if(totalIter <= maxIters)
446     m_info = Success;
447   else
448     m_info = NoConvergence;
449 
450   m_isInitialized = true;
451   m_matUisUptodate = computeU;
452 }
453 
454 } // end namespace Eigen
455 
456 #endif // EIGEN_COMPLEX_SCHUR_H
457