xref: /aosp_15_r20/external/eigen/Eigen/src/Eigenvalues/RealQZ.h (revision bf2c37156dfe67e5dfebd6d394bad8b2ab5804d4)
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2012 Alexey Korepanov <[email protected]>
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_REAL_QZ_H
11 #define EIGEN_REAL_QZ_H
12 
13 namespace Eigen {
14 
15   /** \eigenvalues_module \ingroup Eigenvalues_Module
16    *
17    *
18    * \class RealQZ
19    *
20    * \brief Performs a real QZ decomposition of a pair of square matrices
21    *
22    * \tparam _MatrixType the type of the matrix of which we are computing the
23    * real QZ decomposition; this is expected to be an instantiation of the
24    * Matrix class template.
25    *
26    * Given a real square matrices A and B, this class computes the real QZ
27    * decomposition: \f$ A = Q S Z \f$, \f$ B = Q T Z \f$ where Q and Z are
28    * real orthogonal matrixes, T is upper-triangular matrix, and S is upper
29    * quasi-triangular matrix. An orthogonal matrix is a matrix whose
30    * inverse is equal to its transpose, \f$ U^{-1} = U^T \f$. A quasi-triangular
31    * matrix is a block-triangular matrix whose diagonal consists of 1-by-1
32    * blocks and 2-by-2 blocks where further reduction is impossible due to
33    * complex eigenvalues.
34    *
35    * The eigenvalues of the pencil \f$ A - z B \f$ can be obtained from
36    * 1x1 and 2x2 blocks on the diagonals of S and T.
37    *
38    * Call the function compute() to compute the real QZ decomposition of a
39    * given pair of matrices. Alternatively, you can use the
40    * RealQZ(const MatrixType& B, const MatrixType& B, bool computeQZ)
41    * constructor which computes the real QZ decomposition at construction
42    * time. Once the decomposition is computed, you can use the matrixS(),
43    * matrixT(), matrixQ() and matrixZ() functions to retrieve the matrices
44    * S, T, Q and Z in the decomposition. If computeQZ==false, some time
45    * is saved by not computing matrices Q and Z.
46    *
47    * Example: \include RealQZ_compute.cpp
48    * Output: \include RealQZ_compute.out
49    *
50    * \note The implementation is based on the algorithm in "Matrix Computations"
51    * by Gene H. Golub and Charles F. Van Loan, and a paper "An algorithm for
52    * generalized eigenvalue problems" by C.B.Moler and G.W.Stewart.
53    *
54    * \sa class RealSchur, class ComplexSchur, class EigenSolver, class ComplexEigenSolver
55    */
56 
57   template<typename _MatrixType> class RealQZ
58   {
59     public:
60       typedef _MatrixType MatrixType;
61       enum {
62         RowsAtCompileTime = MatrixType::RowsAtCompileTime,
63         ColsAtCompileTime = MatrixType::ColsAtCompileTime,
64         Options = MatrixType::Options,
65         MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
66         MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
67       };
68       typedef typename MatrixType::Scalar Scalar;
69       typedef std::complex<typename NumTraits<Scalar>::Real> ComplexScalar;
70       typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
71 
72       typedef Matrix<ComplexScalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> EigenvalueType;
73       typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> ColumnVectorType;
74 
75       /** \brief Default constructor.
76        *
77        * \param [in] size  Positive integer, size of the matrix whose QZ decomposition will be computed.
78        *
79        * The default constructor is useful in cases in which the user intends to
80        * perform decompositions via compute().  The \p size parameter is only
81        * used as a hint. It is not an error to give a wrong \p size, but it may
82        * impair performance.
83        *
84        * \sa compute() for an example.
85        */
86       explicit RealQZ(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime) :
m_S(size,size)87         m_S(size, size),
88         m_T(size, size),
89         m_Q(size, size),
90         m_Z(size, size),
91         m_workspace(size*2),
92         m_maxIters(400),
93         m_isInitialized(false),
94         m_computeQZ(true)
95       {}
96 
97       /** \brief Constructor; computes real QZ decomposition of given matrices
98        *
99        * \param[in]  A          Matrix A.
100        * \param[in]  B          Matrix B.
101        * \param[in]  computeQZ  If false, A and Z are not computed.
102        *
103        * This constructor calls compute() to compute the QZ decomposition.
104        */
105       RealQZ(const MatrixType& A, const MatrixType& B, bool computeQZ = true) :
106         m_S(A.rows(),A.cols()),
107         m_T(A.rows(),A.cols()),
108         m_Q(A.rows(),A.cols()),
109         m_Z(A.rows(),A.cols()),
110         m_workspace(A.rows()*2),
111         m_maxIters(400),
112         m_isInitialized(false),
113         m_computeQZ(true)
114       {
115         compute(A, B, computeQZ);
116       }
117 
118       /** \brief Returns matrix Q in the QZ decomposition.
119        *
120        * \returns A const reference to the matrix Q.
121        */
matrixQ()122       const MatrixType& matrixQ() const {
123         eigen_assert(m_isInitialized && "RealQZ is not initialized.");
124         eigen_assert(m_computeQZ && "The matrices Q and Z have not been computed during the QZ decomposition.");
125         return m_Q;
126       }
127 
128       /** \brief Returns matrix Z in the QZ decomposition.
129        *
130        * \returns A const reference to the matrix Z.
131        */
matrixZ()132       const MatrixType& matrixZ() const {
133         eigen_assert(m_isInitialized && "RealQZ is not initialized.");
134         eigen_assert(m_computeQZ && "The matrices Q and Z have not been computed during the QZ decomposition.");
135         return m_Z;
136       }
137 
138       /** \brief Returns matrix S in the QZ decomposition.
139        *
140        * \returns A const reference to the matrix S.
141        */
matrixS()142       const MatrixType& matrixS() const {
143         eigen_assert(m_isInitialized && "RealQZ is not initialized.");
144         return m_S;
145       }
146 
147       /** \brief Returns matrix S in the QZ decomposition.
148        *
149        * \returns A const reference to the matrix S.
150        */
matrixT()151       const MatrixType& matrixT() const {
152         eigen_assert(m_isInitialized && "RealQZ is not initialized.");
153         return m_T;
154       }
155 
156       /** \brief Computes QZ decomposition of given matrix.
157        *
158        * \param[in]  A          Matrix A.
159        * \param[in]  B          Matrix B.
160        * \param[in]  computeQZ  If false, A and Z are not computed.
161        * \returns    Reference to \c *this
162        */
163       RealQZ& compute(const MatrixType& A, const MatrixType& B, bool computeQZ = true);
164 
165       /** \brief Reports whether previous computation was successful.
166        *
167        * \returns \c Success if computation was successful, \c NoConvergence otherwise.
168        */
info()169       ComputationInfo info() const
170       {
171         eigen_assert(m_isInitialized && "RealQZ is not initialized.");
172         return m_info;
173       }
174 
175       /** \brief Returns number of performed QR-like iterations.
176       */
iterations()177       Index iterations() const
178       {
179         eigen_assert(m_isInitialized && "RealQZ is not initialized.");
180         return m_global_iter;
181       }
182 
183       /** Sets the maximal number of iterations allowed to converge to one eigenvalue
184        * or decouple the problem.
185       */
setMaxIterations(Index maxIters)186       RealQZ& setMaxIterations(Index maxIters)
187       {
188         m_maxIters = maxIters;
189         return *this;
190       }
191 
192     private:
193 
194       MatrixType m_S, m_T, m_Q, m_Z;
195       Matrix<Scalar,Dynamic,1> m_workspace;
196       ComputationInfo m_info;
197       Index m_maxIters;
198       bool m_isInitialized;
199       bool m_computeQZ;
200       Scalar m_normOfT, m_normOfS;
201       Index m_global_iter;
202 
203       typedef Matrix<Scalar,3,1> Vector3s;
204       typedef Matrix<Scalar,2,1> Vector2s;
205       typedef Matrix<Scalar,2,2> Matrix2s;
206       typedef JacobiRotation<Scalar> JRs;
207 
208       void hessenbergTriangular();
209       void computeNorms();
210       Index findSmallSubdiagEntry(Index iu);
211       Index findSmallDiagEntry(Index f, Index l);
212       void splitOffTwoRows(Index i);
213       void pushDownZero(Index z, Index f, Index l);
214       void step(Index f, Index l, Index iter);
215 
216   }; // RealQZ
217 
218   /** \internal Reduces S and T to upper Hessenberg - triangular form */
219   template<typename MatrixType>
hessenbergTriangular()220     void RealQZ<MatrixType>::hessenbergTriangular()
221     {
222 
223       const Index dim = m_S.cols();
224 
225       // perform QR decomposition of T, overwrite T with R, save Q
226       HouseholderQR<MatrixType> qrT(m_T);
227       m_T = qrT.matrixQR();
228       m_T.template triangularView<StrictlyLower>().setZero();
229       m_Q = qrT.householderQ();
230       // overwrite S with Q* S
231       m_S.applyOnTheLeft(m_Q.adjoint());
232       // init Z as Identity
233       if (m_computeQZ)
234         m_Z = MatrixType::Identity(dim,dim);
235       // reduce S to upper Hessenberg with Givens rotations
236       for (Index j=0; j<=dim-3; j++) {
237         for (Index i=dim-1; i>=j+2; i--) {
238           JRs G;
239           // kill S(i,j)
240           if(m_S.coeff(i,j) != 0)
241           {
242             G.makeGivens(m_S.coeff(i-1,j), m_S.coeff(i,j), &m_S.coeffRef(i-1, j));
243             m_S.coeffRef(i,j) = Scalar(0.0);
244             m_S.rightCols(dim-j-1).applyOnTheLeft(i-1,i,G.adjoint());
245             m_T.rightCols(dim-i+1).applyOnTheLeft(i-1,i,G.adjoint());
246             // update Q
247             if (m_computeQZ)
248               m_Q.applyOnTheRight(i-1,i,G);
249           }
250           // kill T(i,i-1)
251           if(m_T.coeff(i,i-1)!=Scalar(0))
252           {
253             G.makeGivens(m_T.coeff(i,i), m_T.coeff(i,i-1), &m_T.coeffRef(i,i));
254             m_T.coeffRef(i,i-1) = Scalar(0.0);
255             m_S.applyOnTheRight(i,i-1,G);
256             m_T.topRows(i).applyOnTheRight(i,i-1,G);
257             // update Z
258             if (m_computeQZ)
259               m_Z.applyOnTheLeft(i,i-1,G.adjoint());
260           }
261         }
262       }
263     }
264 
265   /** \internal Computes vector L1 norms of S and T when in Hessenberg-Triangular form already */
266   template<typename MatrixType>
computeNorms()267     inline void RealQZ<MatrixType>::computeNorms()
268     {
269       const Index size = m_S.cols();
270       m_normOfS = Scalar(0.0);
271       m_normOfT = Scalar(0.0);
272       for (Index j = 0; j < size; ++j)
273       {
274         m_normOfS += m_S.col(j).segment(0, (std::min)(size,j+2)).cwiseAbs().sum();
275         m_normOfT += m_T.row(j).segment(j, size - j).cwiseAbs().sum();
276       }
277     }
278 
279 
280   /** \internal Look for single small sub-diagonal element S(res, res-1) and return res (or 0) */
281   template<typename MatrixType>
findSmallSubdiagEntry(Index iu)282     inline Index RealQZ<MatrixType>::findSmallSubdiagEntry(Index iu)
283     {
284       using std::abs;
285       Index res = iu;
286       while (res > 0)
287       {
288         Scalar s = abs(m_S.coeff(res-1,res-1)) + abs(m_S.coeff(res,res));
289         if (s == Scalar(0.0))
290           s = m_normOfS;
291         if (abs(m_S.coeff(res,res-1)) < NumTraits<Scalar>::epsilon() * s)
292           break;
293         res--;
294       }
295       return res;
296     }
297 
298   /** \internal Look for single small diagonal element T(res, res) for res between f and l, and return res (or f-1)  */
299   template<typename MatrixType>
findSmallDiagEntry(Index f,Index l)300     inline Index RealQZ<MatrixType>::findSmallDiagEntry(Index f, Index l)
301     {
302       using std::abs;
303       Index res = l;
304       while (res >= f) {
305         if (abs(m_T.coeff(res,res)) <= NumTraits<Scalar>::epsilon() * m_normOfT)
306           break;
307         res--;
308       }
309       return res;
310     }
311 
312   /** \internal decouple 2x2 diagonal block in rows i, i+1 if eigenvalues are real */
313   template<typename MatrixType>
splitOffTwoRows(Index i)314     inline void RealQZ<MatrixType>::splitOffTwoRows(Index i)
315     {
316       using std::abs;
317       using std::sqrt;
318       const Index dim=m_S.cols();
319       if (abs(m_S.coeff(i+1,i))==Scalar(0))
320         return;
321       Index j = findSmallDiagEntry(i,i+1);
322       if (j==i-1)
323       {
324         // block of (S T^{-1})
325         Matrix2s STi = m_T.template block<2,2>(i,i).template triangularView<Upper>().
326           template solve<OnTheRight>(m_S.template block<2,2>(i,i));
327         Scalar p = Scalar(0.5)*(STi(0,0)-STi(1,1));
328         Scalar q = p*p + STi(1,0)*STi(0,1);
329         if (q>=0) {
330           Scalar z = sqrt(q);
331           // one QR-like iteration for ABi - lambda I
332           // is enough - when we know exact eigenvalue in advance,
333           // convergence is immediate
334           JRs G;
335           if (p>=0)
336             G.makeGivens(p + z, STi(1,0));
337           else
338             G.makeGivens(p - z, STi(1,0));
339           m_S.rightCols(dim-i).applyOnTheLeft(i,i+1,G.adjoint());
340           m_T.rightCols(dim-i).applyOnTheLeft(i,i+1,G.adjoint());
341           // update Q
342           if (m_computeQZ)
343             m_Q.applyOnTheRight(i,i+1,G);
344 
345           G.makeGivens(m_T.coeff(i+1,i+1), m_T.coeff(i+1,i));
346           m_S.topRows(i+2).applyOnTheRight(i+1,i,G);
347           m_T.topRows(i+2).applyOnTheRight(i+1,i,G);
348           // update Z
349           if (m_computeQZ)
350             m_Z.applyOnTheLeft(i+1,i,G.adjoint());
351 
352           m_S.coeffRef(i+1,i) = Scalar(0.0);
353           m_T.coeffRef(i+1,i) = Scalar(0.0);
354         }
355       }
356       else
357       {
358         pushDownZero(j,i,i+1);
359       }
360     }
361 
362   /** \internal use zero in T(z,z) to zero S(l,l-1), working in block f..l */
363   template<typename MatrixType>
pushDownZero(Index z,Index f,Index l)364     inline void RealQZ<MatrixType>::pushDownZero(Index z, Index f, Index l)
365     {
366       JRs G;
367       const Index dim = m_S.cols();
368       for (Index zz=z; zz<l; zz++)
369       {
370         // push 0 down
371         Index firstColS = zz>f ? (zz-1) : zz;
372         G.makeGivens(m_T.coeff(zz, zz+1), m_T.coeff(zz+1, zz+1));
373         m_S.rightCols(dim-firstColS).applyOnTheLeft(zz,zz+1,G.adjoint());
374         m_T.rightCols(dim-zz).applyOnTheLeft(zz,zz+1,G.adjoint());
375         m_T.coeffRef(zz+1,zz+1) = Scalar(0.0);
376         // update Q
377         if (m_computeQZ)
378           m_Q.applyOnTheRight(zz,zz+1,G);
379         // kill S(zz+1, zz-1)
380         if (zz>f)
381         {
382           G.makeGivens(m_S.coeff(zz+1, zz), m_S.coeff(zz+1,zz-1));
383           m_S.topRows(zz+2).applyOnTheRight(zz, zz-1,G);
384           m_T.topRows(zz+1).applyOnTheRight(zz, zz-1,G);
385           m_S.coeffRef(zz+1,zz-1) = Scalar(0.0);
386           // update Z
387           if (m_computeQZ)
388             m_Z.applyOnTheLeft(zz,zz-1,G.adjoint());
389         }
390       }
391       // finally kill S(l,l-1)
392       G.makeGivens(m_S.coeff(l,l), m_S.coeff(l,l-1));
393       m_S.applyOnTheRight(l,l-1,G);
394       m_T.applyOnTheRight(l,l-1,G);
395       m_S.coeffRef(l,l-1)=Scalar(0.0);
396       // update Z
397       if (m_computeQZ)
398         m_Z.applyOnTheLeft(l,l-1,G.adjoint());
399     }
400 
401   /** \internal QR-like iterative step for block f..l */
402   template<typename MatrixType>
step(Index f,Index l,Index iter)403     inline void RealQZ<MatrixType>::step(Index f, Index l, Index iter)
404     {
405       using std::abs;
406       const Index dim = m_S.cols();
407 
408       // x, y, z
409       Scalar x, y, z;
410       if (iter==10)
411       {
412         // Wilkinson ad hoc shift
413         const Scalar
414           a11=m_S.coeff(f+0,f+0), a12=m_S.coeff(f+0,f+1),
415           a21=m_S.coeff(f+1,f+0), a22=m_S.coeff(f+1,f+1), a32=m_S.coeff(f+2,f+1),
416           b12=m_T.coeff(f+0,f+1),
417           b11i=Scalar(1.0)/m_T.coeff(f+0,f+0),
418           b22i=Scalar(1.0)/m_T.coeff(f+1,f+1),
419           a87=m_S.coeff(l-1,l-2),
420           a98=m_S.coeff(l-0,l-1),
421           b77i=Scalar(1.0)/m_T.coeff(l-2,l-2),
422           b88i=Scalar(1.0)/m_T.coeff(l-1,l-1);
423         Scalar ss = abs(a87*b77i) + abs(a98*b88i),
424                lpl = Scalar(1.5)*ss,
425                ll = ss*ss;
426         x = ll + a11*a11*b11i*b11i - lpl*a11*b11i + a12*a21*b11i*b22i
427           - a11*a21*b12*b11i*b11i*b22i;
428         y = a11*a21*b11i*b11i - lpl*a21*b11i + a21*a22*b11i*b22i
429           - a21*a21*b12*b11i*b11i*b22i;
430         z = a21*a32*b11i*b22i;
431       }
432       else if (iter==16)
433       {
434         // another exceptional shift
435         x = m_S.coeff(f,f)/m_T.coeff(f,f)-m_S.coeff(l,l)/m_T.coeff(l,l) + m_S.coeff(l,l-1)*m_T.coeff(l-1,l) /
436           (m_T.coeff(l-1,l-1)*m_T.coeff(l,l));
437         y = m_S.coeff(f+1,f)/m_T.coeff(f,f);
438         z = 0;
439       }
440       else if (iter>23 && !(iter%8))
441       {
442         // extremely exceptional shift
443         x = internal::random<Scalar>(-1.0,1.0);
444         y = internal::random<Scalar>(-1.0,1.0);
445         z = internal::random<Scalar>(-1.0,1.0);
446       }
447       else
448       {
449         // Compute the shifts: (x,y,z,0...) = (AB^-1 - l1 I) (AB^-1 - l2 I) e1
450         // where l1 and l2 are the eigenvalues of the 2x2 matrix C = U V^-1 where
451         // U and V are 2x2 bottom right sub matrices of A and B. Thus:
452         //  = AB^-1AB^-1 + l1 l2 I - (l1+l2)(AB^-1)
453         //  = AB^-1AB^-1 + det(M) - tr(M)(AB^-1)
454         // Since we are only interested in having x, y, z with a correct ratio, we have:
455         const Scalar
456           a11 = m_S.coeff(f,f),     a12 = m_S.coeff(f,f+1),
457           a21 = m_S.coeff(f+1,f),   a22 = m_S.coeff(f+1,f+1),
458                                     a32 = m_S.coeff(f+2,f+1),
459 
460           a88 = m_S.coeff(l-1,l-1), a89 = m_S.coeff(l-1,l),
461           a98 = m_S.coeff(l,l-1),   a99 = m_S.coeff(l,l),
462 
463           b11 = m_T.coeff(f,f),     b12 = m_T.coeff(f,f+1),
464                                     b22 = m_T.coeff(f+1,f+1),
465 
466           b88 = m_T.coeff(l-1,l-1), b89 = m_T.coeff(l-1,l),
467                                     b99 = m_T.coeff(l,l);
468 
469         x = ( (a88/b88 - a11/b11)*(a99/b99 - a11/b11) - (a89/b99)*(a98/b88) + (a98/b88)*(b89/b99)*(a11/b11) ) * (b11/a21)
470           + a12/b22 - (a11/b11)*(b12/b22);
471         y = (a22/b22-a11/b11) - (a21/b11)*(b12/b22) - (a88/b88-a11/b11) - (a99/b99-a11/b11) + (a98/b88)*(b89/b99);
472         z = a32/b22;
473       }
474 
475       JRs G;
476 
477       for (Index k=f; k<=l-2; k++)
478       {
479         // variables for Householder reflections
480         Vector2s essential2;
481         Scalar tau, beta;
482 
483         Vector3s hr(x,y,z);
484 
485         // Q_k to annihilate S(k+1,k-1) and S(k+2,k-1)
486         hr.makeHouseholderInPlace(tau, beta);
487         essential2 = hr.template bottomRows<2>();
488         Index fc=(std::max)(k-1,Index(0));  // first col to update
489         m_S.template middleRows<3>(k).rightCols(dim-fc).applyHouseholderOnTheLeft(essential2, tau, m_workspace.data());
490         m_T.template middleRows<3>(k).rightCols(dim-fc).applyHouseholderOnTheLeft(essential2, tau, m_workspace.data());
491         if (m_computeQZ)
492           m_Q.template middleCols<3>(k).applyHouseholderOnTheRight(essential2, tau, m_workspace.data());
493         if (k>f)
494           m_S.coeffRef(k+2,k-1) = m_S.coeffRef(k+1,k-1) = Scalar(0.0);
495 
496         // Z_{k1} to annihilate T(k+2,k+1) and T(k+2,k)
497         hr << m_T.coeff(k+2,k+2),m_T.coeff(k+2,k),m_T.coeff(k+2,k+1);
498         hr.makeHouseholderInPlace(tau, beta);
499         essential2 = hr.template bottomRows<2>();
500         {
501           Index lr = (std::min)(k+4,dim); // last row to update
502           Map<Matrix<Scalar,Dynamic,1> > tmp(m_workspace.data(),lr);
503           // S
504           tmp = m_S.template middleCols<2>(k).topRows(lr) * essential2;
505           tmp += m_S.col(k+2).head(lr);
506           m_S.col(k+2).head(lr) -= tau*tmp;
507           m_S.template middleCols<2>(k).topRows(lr) -= (tau*tmp) * essential2.adjoint();
508           // T
509           tmp = m_T.template middleCols<2>(k).topRows(lr) * essential2;
510           tmp += m_T.col(k+2).head(lr);
511           m_T.col(k+2).head(lr) -= tau*tmp;
512           m_T.template middleCols<2>(k).topRows(lr) -= (tau*tmp) * essential2.adjoint();
513         }
514         if (m_computeQZ)
515         {
516           // Z
517           Map<Matrix<Scalar,1,Dynamic> > tmp(m_workspace.data(),dim);
518           tmp = essential2.adjoint()*(m_Z.template middleRows<2>(k));
519           tmp += m_Z.row(k+2);
520           m_Z.row(k+2) -= tau*tmp;
521           m_Z.template middleRows<2>(k) -= essential2 * (tau*tmp);
522         }
523         m_T.coeffRef(k+2,k) = m_T.coeffRef(k+2,k+1) = Scalar(0.0);
524 
525         // Z_{k2} to annihilate T(k+1,k)
526         G.makeGivens(m_T.coeff(k+1,k+1), m_T.coeff(k+1,k));
527         m_S.applyOnTheRight(k+1,k,G);
528         m_T.applyOnTheRight(k+1,k,G);
529         // update Z
530         if (m_computeQZ)
531           m_Z.applyOnTheLeft(k+1,k,G.adjoint());
532         m_T.coeffRef(k+1,k) = Scalar(0.0);
533 
534         // update x,y,z
535         x = m_S.coeff(k+1,k);
536         y = m_S.coeff(k+2,k);
537         if (k < l-2)
538           z = m_S.coeff(k+3,k);
539       } // loop over k
540 
541       // Q_{n-1} to annihilate y = S(l,l-2)
542       G.makeGivens(x,y);
543       m_S.applyOnTheLeft(l-1,l,G.adjoint());
544       m_T.applyOnTheLeft(l-1,l,G.adjoint());
545       if (m_computeQZ)
546         m_Q.applyOnTheRight(l-1,l,G);
547       m_S.coeffRef(l,l-2) = Scalar(0.0);
548 
549       // Z_{n-1} to annihilate T(l,l-1)
550       G.makeGivens(m_T.coeff(l,l),m_T.coeff(l,l-1));
551       m_S.applyOnTheRight(l,l-1,G);
552       m_T.applyOnTheRight(l,l-1,G);
553       if (m_computeQZ)
554         m_Z.applyOnTheLeft(l,l-1,G.adjoint());
555       m_T.coeffRef(l,l-1) = Scalar(0.0);
556     }
557 
558   template<typename MatrixType>
compute(const MatrixType & A_in,const MatrixType & B_in,bool computeQZ)559     RealQZ<MatrixType>& RealQZ<MatrixType>::compute(const MatrixType& A_in, const MatrixType& B_in, bool computeQZ)
560     {
561 
562       const Index dim = A_in.cols();
563 
564       eigen_assert (A_in.rows()==dim && A_in.cols()==dim
565           && B_in.rows()==dim && B_in.cols()==dim
566           && "Need square matrices of the same dimension");
567 
568       m_isInitialized = true;
569       m_computeQZ = computeQZ;
570       m_S = A_in; m_T = B_in;
571       m_workspace.resize(dim*2);
572       m_global_iter = 0;
573 
574       // entrance point: hessenberg triangular decomposition
575       hessenbergTriangular();
576       // compute L1 vector norms of T, S into m_normOfS, m_normOfT
577       computeNorms();
578 
579       Index l = dim-1,
580             f,
581             local_iter = 0;
582 
583       while (l>0 && local_iter<m_maxIters)
584       {
585         f = findSmallSubdiagEntry(l);
586         // now rows and columns f..l (including) decouple from the rest of the problem
587         if (f>0) m_S.coeffRef(f,f-1) = Scalar(0.0);
588         if (f == l) // One root found
589         {
590           l--;
591           local_iter = 0;
592         }
593         else if (f == l-1) // Two roots found
594         {
595           splitOffTwoRows(f);
596           l -= 2;
597           local_iter = 0;
598         }
599         else // No convergence yet
600         {
601           // if there's zero on diagonal of T, we can isolate an eigenvalue with Givens rotations
602           Index z = findSmallDiagEntry(f,l);
603           if (z>=f)
604           {
605             // zero found
606             pushDownZero(z,f,l);
607           }
608           else
609           {
610             // We are sure now that S.block(f,f, l-f+1,l-f+1) is underuced upper-Hessenberg
611             // and T.block(f,f, l-f+1,l-f+1) is invertible uper-triangular, which allows to
612             // apply a QR-like iteration to rows and columns f..l.
613             step(f,l, local_iter);
614             local_iter++;
615             m_global_iter++;
616           }
617         }
618       }
619       // check if we converged before reaching iterations limit
620       m_info = (local_iter<m_maxIters) ? Success : NoConvergence;
621 
622       // For each non triangular 2x2 diagonal block of S,
623       //    reduce the respective 2x2 diagonal block of T to positive diagonal form using 2x2 SVD.
624       // This step is not mandatory for QZ, but it does help further extraction of eigenvalues/eigenvectors,
625       // and is in par with Lapack/Matlab QZ.
626       if(m_info==Success)
627       {
628         for(Index i=0; i<dim-1; ++i)
629         {
630           if(m_S.coeff(i+1, i) != Scalar(0))
631           {
632             JacobiRotation<Scalar> j_left, j_right;
633             internal::real_2x2_jacobi_svd(m_T, i, i+1, &j_left, &j_right);
634 
635             // Apply resulting Jacobi rotations
636             m_S.applyOnTheLeft(i,i+1,j_left);
637             m_S.applyOnTheRight(i,i+1,j_right);
638             m_T.applyOnTheLeft(i,i+1,j_left);
639             m_T.applyOnTheRight(i,i+1,j_right);
640             m_T(i+1,i) = m_T(i,i+1) = Scalar(0);
641 
642             if(m_computeQZ) {
643               m_Q.applyOnTheRight(i,i+1,j_left.transpose());
644               m_Z.applyOnTheLeft(i,i+1,j_right.transpose());
645             }
646 
647             i++;
648           }
649         }
650       }
651 
652       return *this;
653     } // end compute
654 
655 } // end namespace Eigen
656 
657 #endif //EIGEN_REAL_QZ
658