• 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) 2012, 2013 Chen-Pang He <jdh8@ms63.hinet.net>
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_POWER
11 #define EIGEN_MATRIX_POWER
12 
13 namespace Eigen {
14 
15 template<typename MatrixType> class MatrixPower;
16 
17 /**
18  * \ingroup MatrixFunctions_Module
19  *
20  * \brief Proxy for the matrix power of some matrix.
21  *
22  * \tparam MatrixType  type of the base, a matrix.
23  *
24  * This class holds the arguments to the matrix power until it is
25  * assigned or evaluated for some other reason (so the argument
26  * should not be changed in the meantime). It is the return type of
27  * MatrixPower::operator() and related functions and most of the
28  * time this is the only way it is used.
29  */
30 /* TODO This class is only used by MatrixPower, so it should be nested
31  * into MatrixPower, like MatrixPower::ReturnValue. However, my
32  * compiler complained about unused template parameter in the
33  * following declaration in namespace internal.
34  *
35  * template<typename MatrixType>
36  * struct traits<MatrixPower<MatrixType>::ReturnValue>;
37  */
38 template<typename MatrixType>
39 class MatrixPowerParenthesesReturnValue : public ReturnByValue< MatrixPowerParenthesesReturnValue<MatrixType> >
40 {
41   public:
42     typedef typename MatrixType::RealScalar RealScalar;
43     typedef typename MatrixType::Index Index;
44 
45     /**
46      * \brief Constructor.
47      *
48      * \param[in] pow  %MatrixPower storing the base.
49      * \param[in] p    scalar, the exponent of the matrix power.
50      */
MatrixPowerParenthesesReturnValue(MatrixPower<MatrixType> & pow,RealScalar p)51     MatrixPowerParenthesesReturnValue(MatrixPower<MatrixType>& pow, RealScalar p) : m_pow(pow), m_p(p)
52     { }
53 
54     /**
55      * \brief Compute the matrix power.
56      *
57      * \param[out] result
58      */
59     template<typename ResultType>
evalTo(ResultType & res)60     inline void evalTo(ResultType& res) const
61     { m_pow.compute(res, m_p); }
62 
rows()63     Index rows() const { return m_pow.rows(); }
cols()64     Index cols() const { return m_pow.cols(); }
65 
66   private:
67     MatrixPower<MatrixType>& m_pow;
68     const RealScalar m_p;
69 };
70 
71 /**
72  * \ingroup MatrixFunctions_Module
73  *
74  * \brief Class for computing matrix powers.
75  *
76  * \tparam MatrixType  type of the base, expected to be an instantiation
77  * of the Matrix class template.
78  *
79  * This class is capable of computing triangular real/complex matrices
80  * raised to a power in the interval \f$ (-1, 1) \f$.
81  *
82  * \note Currently this class is only used by MatrixPower. One may
83  * insist that this be nested into MatrixPower. This class is here to
84  * faciliate future development of triangular matrix functions.
85  */
86 template<typename MatrixType>
87 class MatrixPowerAtomic : internal::noncopyable
88 {
89   private:
90     enum {
91       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
92       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime
93     };
94     typedef typename MatrixType::Scalar Scalar;
95     typedef typename MatrixType::RealScalar RealScalar;
96     typedef std::complex<RealScalar> ComplexScalar;
97     typedef typename MatrixType::Index Index;
98     typedef Block<MatrixType,Dynamic,Dynamic> ResultType;
99 
100     const MatrixType& m_A;
101     RealScalar m_p;
102 
103     void computePade(int degree, const MatrixType& IminusT, ResultType& res) const;
104     void compute2x2(ResultType& res, RealScalar p) const;
105     void computeBig(ResultType& res) const;
106     static int getPadeDegree(float normIminusT);
107     static int getPadeDegree(double normIminusT);
108     static int getPadeDegree(long double normIminusT);
109     static ComplexScalar computeSuperDiag(const ComplexScalar&, const ComplexScalar&, RealScalar p);
110     static RealScalar computeSuperDiag(RealScalar, RealScalar, RealScalar p);
111 
112   public:
113     /**
114      * \brief Constructor.
115      *
116      * \param[in] T  the base of the matrix power.
117      * \param[in] p  the exponent of the matrix power, should be in
118      * \f$ (-1, 1) \f$.
119      *
120      * The class stores a reference to T, so it should not be changed
121      * (or destroyed) before evaluation. Only the upper triangular
122      * part of T is read.
123      */
124     MatrixPowerAtomic(const MatrixType& T, RealScalar p);
125 
126     /**
127      * \brief Compute the matrix power.
128      *
129      * \param[out] res  \f$ A^p \f$ where A and p are specified in the
130      * constructor.
131      */
132     void compute(ResultType& res) const;
133 };
134 
135 template<typename MatrixType>
MatrixPowerAtomic(const MatrixType & T,RealScalar p)136 MatrixPowerAtomic<MatrixType>::MatrixPowerAtomic(const MatrixType& T, RealScalar p) :
137   m_A(T), m_p(p)
138 {
139   eigen_assert(T.rows() == T.cols());
140   eigen_assert(p > -1 && p < 1);
141 }
142 
143 template<typename MatrixType>
compute(ResultType & res)144 void MatrixPowerAtomic<MatrixType>::compute(ResultType& res) const
145 {
146   using std::pow;
147   switch (m_A.rows()) {
148     case 0:
149       break;
150     case 1:
151       res(0,0) = pow(m_A(0,0), m_p);
152       break;
153     case 2:
154       compute2x2(res, m_p);
155       break;
156     default:
157       computeBig(res);
158   }
159 }
160 
161 template<typename MatrixType>
computePade(int degree,const MatrixType & IminusT,ResultType & res)162 void MatrixPowerAtomic<MatrixType>::computePade(int degree, const MatrixType& IminusT, ResultType& res) const
163 {
164   int i = 2*degree;
165   res = (m_p-degree) / (2*i-2) * IminusT;
166 
167   for (--i; i; --i) {
168     res = (MatrixType::Identity(IminusT.rows(), IminusT.cols()) + res).template triangularView<Upper>()
169 	.solve((i==1 ? -m_p : i&1 ? (-m_p-i/2)/(2*i) : (m_p-i/2)/(2*i-2)) * IminusT).eval();
170   }
171   res += MatrixType::Identity(IminusT.rows(), IminusT.cols());
172 }
173 
174 // This function assumes that res has the correct size (see bug 614)
175 template<typename MatrixType>
compute2x2(ResultType & res,RealScalar p)176 void MatrixPowerAtomic<MatrixType>::compute2x2(ResultType& res, RealScalar p) const
177 {
178   using std::abs;
179   using std::pow;
180   res.coeffRef(0,0) = pow(m_A.coeff(0,0), p);
181 
182   for (Index i=1; i < m_A.cols(); ++i) {
183     res.coeffRef(i,i) = pow(m_A.coeff(i,i), p);
184     if (m_A.coeff(i-1,i-1) == m_A.coeff(i,i))
185       res.coeffRef(i-1,i) = p * pow(m_A.coeff(i,i), p-1);
186     else if (2*abs(m_A.coeff(i-1,i-1)) < abs(m_A.coeff(i,i)) || 2*abs(m_A.coeff(i,i)) < abs(m_A.coeff(i-1,i-1)))
187       res.coeffRef(i-1,i) = (res.coeff(i,i)-res.coeff(i-1,i-1)) / (m_A.coeff(i,i)-m_A.coeff(i-1,i-1));
188     else
189       res.coeffRef(i-1,i) = computeSuperDiag(m_A.coeff(i,i), m_A.coeff(i-1,i-1), p);
190     res.coeffRef(i-1,i) *= m_A.coeff(i-1,i);
191   }
192 }
193 
194 template<typename MatrixType>
computeBig(ResultType & res)195 void MatrixPowerAtomic<MatrixType>::computeBig(ResultType& res) const
196 {
197   using std::ldexp;
198   const int digits = std::numeric_limits<RealScalar>::digits;
199   const RealScalar maxNormForPade = digits <=  24? 4.3386528e-1L                            // single precision
200                                   : digits <=  53? 2.789358995219730e-1L                    // double precision
201                                   : digits <=  64? 2.4471944416607995472e-1L                // extended precision
202                                   : digits <= 106? 1.1016843812851143391275867258512e-1L    // double-double
203                                   :                9.134603732914548552537150753385375e-2L; // quadruple precision
204   MatrixType IminusT, sqrtT, T = m_A.template triangularView<Upper>();
205   RealScalar normIminusT;
206   int degree, degree2, numberOfSquareRoots = 0;
207   bool hasExtraSquareRoot = false;
208 
209   for (Index i=0; i < m_A.cols(); ++i)
210     eigen_assert(m_A(i,i) != RealScalar(0));
211 
212   while (true) {
213     IminusT = MatrixType::Identity(m_A.rows(), m_A.cols()) - T;
214     normIminusT = IminusT.cwiseAbs().colwise().sum().maxCoeff();
215     if (normIminusT < maxNormForPade) {
216       degree = getPadeDegree(normIminusT);
217       degree2 = getPadeDegree(normIminusT/2);
218       if (degree - degree2 <= 1 || hasExtraSquareRoot)
219 	break;
220       hasExtraSquareRoot = true;
221     }
222     matrix_sqrt_triangular(T, sqrtT);
223     T = sqrtT.template triangularView<Upper>();
224     ++numberOfSquareRoots;
225   }
226   computePade(degree, IminusT, res);
227 
228   for (; numberOfSquareRoots; --numberOfSquareRoots) {
229     compute2x2(res, ldexp(m_p, -numberOfSquareRoots));
230     res = res.template triangularView<Upper>() * res;
231   }
232   compute2x2(res, m_p);
233 }
234 
235 template<typename MatrixType>
getPadeDegree(float normIminusT)236 inline int MatrixPowerAtomic<MatrixType>::getPadeDegree(float normIminusT)
237 {
238   const float maxNormForPade[] = { 2.8064004e-1f /* degree = 3 */ , 4.3386528e-1f };
239   int degree = 3;
240   for (; degree <= 4; ++degree)
241     if (normIminusT <= maxNormForPade[degree - 3])
242       break;
243   return degree;
244 }
245 
246 template<typename MatrixType>
getPadeDegree(double normIminusT)247 inline int MatrixPowerAtomic<MatrixType>::getPadeDegree(double normIminusT)
248 {
249   const double maxNormForPade[] = { 1.884160592658218e-2 /* degree = 3 */ , 6.038881904059573e-2, 1.239917516308172e-1,
250       1.999045567181744e-1, 2.789358995219730e-1 };
251   int degree = 3;
252   for (; degree <= 7; ++degree)
253     if (normIminusT <= maxNormForPade[degree - 3])
254       break;
255   return degree;
256 }
257 
258 template<typename MatrixType>
getPadeDegree(long double normIminusT)259 inline int MatrixPowerAtomic<MatrixType>::getPadeDegree(long double normIminusT)
260 {
261 #if   LDBL_MANT_DIG == 53
262   const int maxPadeDegree = 7;
263   const double maxNormForPade[] = { 1.884160592658218e-2L /* degree = 3 */ , 6.038881904059573e-2L, 1.239917516308172e-1L,
264       1.999045567181744e-1L, 2.789358995219730e-1L };
265 #elif LDBL_MANT_DIG <= 64
266   const int maxPadeDegree = 8;
267   const long double maxNormForPade[] = { 6.3854693117491799460e-3L /* degree = 3 */ , 2.6394893435456973676e-2L,
268       6.4216043030404063729e-2L, 1.1701165502926694307e-1L, 1.7904284231268670284e-1L, 2.4471944416607995472e-1L };
269 #elif LDBL_MANT_DIG <= 106
270   const int maxPadeDegree = 10;
271   const double maxNormForPade[] = { 1.0007161601787493236741409687186e-4L /* degree = 3 */ ,
272       1.0007161601787493236741409687186e-3L, 4.7069769360887572939882574746264e-3L, 1.3220386624169159689406653101695e-2L,
273       2.8063482381631737920612944054906e-2L, 4.9625993951953473052385361085058e-2L, 7.7367040706027886224557538328171e-2L,
274       1.1016843812851143391275867258512e-1L };
275 #else
276   const int maxPadeDegree = 10;
277   const double maxNormForPade[] = { 5.524506147036624377378713555116378e-5L /* degree = 3 */ ,
278       6.640600568157479679823602193345995e-4L, 3.227716520106894279249709728084626e-3L,
279       9.619593944683432960546978734646284e-3L, 2.134595382433742403911124458161147e-2L,
280       3.908166513900489428442993794761185e-2L, 6.266780814639442865832535460550138e-2L,
281       9.134603732914548552537150753385375e-2L };
282 #endif
283   int degree = 3;
284   for (; degree <= maxPadeDegree; ++degree)
285     if (normIminusT <= maxNormForPade[degree - 3])
286       break;
287   return degree;
288 }
289 
290 template<typename MatrixType>
291 inline typename MatrixPowerAtomic<MatrixType>::ComplexScalar
computeSuperDiag(const ComplexScalar & curr,const ComplexScalar & prev,RealScalar p)292 MatrixPowerAtomic<MatrixType>::computeSuperDiag(const ComplexScalar& curr, const ComplexScalar& prev, RealScalar p)
293 {
294   using std::ceil;
295   using std::exp;
296   using std::log;
297   using std::sinh;
298 
299   ComplexScalar logCurr = log(curr);
300   ComplexScalar logPrev = log(prev);
301   int unwindingNumber = ceil((numext::imag(logCurr - logPrev) - RealScalar(EIGEN_PI)) / RealScalar(2*EIGEN_PI));
302   ComplexScalar w = numext::log1p((curr-prev)/prev)/RealScalar(2) + ComplexScalar(0, EIGEN_PI*unwindingNumber);
303   return RealScalar(2) * exp(RealScalar(0.5) * p * (logCurr + logPrev)) * sinh(p * w) / (curr - prev);
304 }
305 
306 template<typename MatrixType>
307 inline typename MatrixPowerAtomic<MatrixType>::RealScalar
computeSuperDiag(RealScalar curr,RealScalar prev,RealScalar p)308 MatrixPowerAtomic<MatrixType>::computeSuperDiag(RealScalar curr, RealScalar prev, RealScalar p)
309 {
310   using std::exp;
311   using std::log;
312   using std::sinh;
313 
314   RealScalar w = numext::log1p((curr-prev)/prev)/RealScalar(2);
315   return 2 * exp(p * (log(curr) + log(prev)) / 2) * sinh(p * w) / (curr - prev);
316 }
317 
318 /**
319  * \ingroup MatrixFunctions_Module
320  *
321  * \brief Class for computing matrix powers.
322  *
323  * \tparam MatrixType  type of the base, expected to be an instantiation
324  * of the Matrix class template.
325  *
326  * This class is capable of computing real/complex matrices raised to
327  * an arbitrary real power. Meanwhile, it saves the result of Schur
328  * decomposition if an non-integral power has even been calculated.
329  * Therefore, if you want to compute multiple (>= 2) matrix powers
330  * for the same matrix, using the class directly is more efficient than
331  * calling MatrixBase::pow().
332  *
333  * Example:
334  * \include MatrixPower_optimal.cpp
335  * Output: \verbinclude MatrixPower_optimal.out
336  */
337 template<typename MatrixType>
338 class MatrixPower : internal::noncopyable
339 {
340   private:
341     typedef typename MatrixType::Scalar Scalar;
342     typedef typename MatrixType::RealScalar RealScalar;
343     typedef typename MatrixType::Index Index;
344 
345   public:
346     /**
347      * \brief Constructor.
348      *
349      * \param[in] A  the base of the matrix power.
350      *
351      * The class stores a reference to A, so it should not be changed
352      * (or destroyed) before evaluation.
353      */
MatrixPower(const MatrixType & A)354     explicit MatrixPower(const MatrixType& A) :
355       m_A(A),
356       m_conditionNumber(0),
357       m_rank(A.cols()),
358       m_nulls(0)
359     { eigen_assert(A.rows() == A.cols()); }
360 
361     /**
362      * \brief Returns the matrix power.
363      *
364      * \param[in] p  exponent, a real scalar.
365      * \return The expression \f$ A^p \f$, where A is specified in the
366      * constructor.
367      */
operator()368     const MatrixPowerParenthesesReturnValue<MatrixType> operator()(RealScalar p)
369     { return MatrixPowerParenthesesReturnValue<MatrixType>(*this, p); }
370 
371     /**
372      * \brief Compute the matrix power.
373      *
374      * \param[in]  p    exponent, a real scalar.
375      * \param[out] res  \f$ A^p \f$ where A is specified in the
376      * constructor.
377      */
378     template<typename ResultType>
379     void compute(ResultType& res, RealScalar p);
380 
rows()381     Index rows() const { return m_A.rows(); }
cols()382     Index cols() const { return m_A.cols(); }
383 
384   private:
385     typedef std::complex<RealScalar> ComplexScalar;
386     typedef Matrix<ComplexScalar, Dynamic, Dynamic, 0,
387               MatrixType::RowsAtCompileTime, MatrixType::ColsAtCompileTime> ComplexMatrix;
388 
389     /** \brief Reference to the base of matrix power. */
390     typename MatrixType::Nested m_A;
391 
392     /** \brief Temporary storage. */
393     MatrixType m_tmp;
394 
395     /** \brief Store the result of Schur decomposition. */
396     ComplexMatrix m_T, m_U;
397 
398     /** \brief Store fractional power of m_T. */
399     ComplexMatrix m_fT;
400 
401     /**
402      * \brief Condition number of m_A.
403      *
404      * It is initialized as 0 to avoid performing unnecessary Schur
405      * decomposition, which is the bottleneck.
406      */
407     RealScalar m_conditionNumber;
408 
409     /** \brief Rank of m_A. */
410     Index m_rank;
411 
412     /** \brief Rank deficiency of m_A. */
413     Index m_nulls;
414 
415     /**
416      * \brief Split p into integral part and fractional part.
417      *
418      * \param[in]  p        The exponent.
419      * \param[out] p        The fractional part ranging in \f$ (-1, 1) \f$.
420      * \param[out] intpart  The integral part.
421      *
422      * Only if the fractional part is nonzero, it calls initialize().
423      */
424     void split(RealScalar& p, RealScalar& intpart);
425 
426     /** \brief Perform Schur decomposition for fractional power. */
427     void initialize();
428 
429     template<typename ResultType>
430     void computeIntPower(ResultType& res, RealScalar p);
431 
432     template<typename ResultType>
433     void computeFracPower(ResultType& res, RealScalar p);
434 
435     template<int Rows, int Cols, int Options, int MaxRows, int MaxCols>
436     static void revertSchur(
437         Matrix<ComplexScalar, Rows, Cols, Options, MaxRows, MaxCols>& res,
438         const ComplexMatrix& T,
439         const ComplexMatrix& U);
440 
441     template<int Rows, int Cols, int Options, int MaxRows, int MaxCols>
442     static void revertSchur(
443         Matrix<RealScalar, Rows, Cols, Options, MaxRows, MaxCols>& res,
444         const ComplexMatrix& T,
445         const ComplexMatrix& U);
446 };
447 
448 template<typename MatrixType>
449 template<typename ResultType>
compute(ResultType & res,RealScalar p)450 void MatrixPower<MatrixType>::compute(ResultType& res, RealScalar p)
451 {
452   using std::pow;
453   switch (cols()) {
454     case 0:
455       break;
456     case 1:
457       res(0,0) = pow(m_A.coeff(0,0), p);
458       break;
459     default:
460       RealScalar intpart;
461       split(p, intpart);
462 
463       res = MatrixType::Identity(rows(), cols());
464       computeIntPower(res, intpart);
465       if (p) computeFracPower(res, p);
466   }
467 }
468 
469 template<typename MatrixType>
split(RealScalar & p,RealScalar & intpart)470 void MatrixPower<MatrixType>::split(RealScalar& p, RealScalar& intpart)
471 {
472   using std::floor;
473   using std::pow;
474 
475   intpart = floor(p);
476   p -= intpart;
477 
478   // Perform Schur decomposition if it is not yet performed and the power is
479   // not an integer.
480   if (!m_conditionNumber && p)
481     initialize();
482 
483   // Choose the more stable of intpart = floor(p) and intpart = ceil(p).
484   if (p > RealScalar(0.5) && p > (1-p) * pow(m_conditionNumber, p)) {
485     --p;
486     ++intpart;
487   }
488 }
489 
490 template<typename MatrixType>
initialize()491 void MatrixPower<MatrixType>::initialize()
492 {
493   const ComplexSchur<MatrixType> schurOfA(m_A);
494   JacobiRotation<ComplexScalar> rot;
495   ComplexScalar eigenvalue;
496 
497   m_fT.resizeLike(m_A);
498   m_T = schurOfA.matrixT();
499   m_U = schurOfA.matrixU();
500   m_conditionNumber = m_T.diagonal().array().abs().maxCoeff() / m_T.diagonal().array().abs().minCoeff();
501 
502   // Move zero eigenvalues to the bottom right corner.
503   for (Index i = cols()-1; i>=0; --i) {
504     if (m_rank <= 2)
505       return;
506     if (m_T.coeff(i,i) == RealScalar(0)) {
507       for (Index j=i+1; j < m_rank; ++j) {
508         eigenvalue = m_T.coeff(j,j);
509         rot.makeGivens(m_T.coeff(j-1,j), eigenvalue);
510         m_T.applyOnTheRight(j-1, j, rot);
511         m_T.applyOnTheLeft(j-1, j, rot.adjoint());
512         m_T.coeffRef(j-1,j-1) = eigenvalue;
513         m_T.coeffRef(j,j) = RealScalar(0);
514         m_U.applyOnTheRight(j-1, j, rot);
515       }
516       --m_rank;
517     }
518   }
519 
520   m_nulls = rows() - m_rank;
521   if (m_nulls) {
522     eigen_assert(m_T.bottomRightCorner(m_nulls, m_nulls).isZero()
523         && "Base of matrix power should be invertible or with a semisimple zero eigenvalue.");
524     m_fT.bottomRows(m_nulls).fill(RealScalar(0));
525   }
526 }
527 
528 template<typename MatrixType>
529 template<typename ResultType>
computeIntPower(ResultType & res,RealScalar p)530 void MatrixPower<MatrixType>::computeIntPower(ResultType& res, RealScalar p)
531 {
532   using std::abs;
533   using std::fmod;
534   RealScalar pp = abs(p);
535 
536   if (p<0)
537     m_tmp = m_A.inverse();
538   else
539     m_tmp = m_A;
540 
541   while (true) {
542     if (fmod(pp, 2) >= 1)
543       res = m_tmp * res;
544     pp /= 2;
545     if (pp < 1)
546       break;
547     m_tmp *= m_tmp;
548   }
549 }
550 
551 template<typename MatrixType>
552 template<typename ResultType>
computeFracPower(ResultType & res,RealScalar p)553 void MatrixPower<MatrixType>::computeFracPower(ResultType& res, RealScalar p)
554 {
555   Block<ComplexMatrix,Dynamic,Dynamic> blockTp(m_fT, 0, 0, m_rank, m_rank);
556   eigen_assert(m_conditionNumber);
557   eigen_assert(m_rank + m_nulls == rows());
558 
559   MatrixPowerAtomic<ComplexMatrix>(m_T.topLeftCorner(m_rank, m_rank), p).compute(blockTp);
560   if (m_nulls) {
561     m_fT.topRightCorner(m_rank, m_nulls) = m_T.topLeftCorner(m_rank, m_rank).template triangularView<Upper>()
562         .solve(blockTp * m_T.topRightCorner(m_rank, m_nulls));
563   }
564   revertSchur(m_tmp, m_fT, m_U);
565   res = m_tmp * res;
566 }
567 
568 template<typename MatrixType>
569 template<int Rows, int Cols, int Options, int MaxRows, int MaxCols>
revertSchur(Matrix<ComplexScalar,Rows,Cols,Options,MaxRows,MaxCols> & res,const ComplexMatrix & T,const ComplexMatrix & U)570 inline void MatrixPower<MatrixType>::revertSchur(
571     Matrix<ComplexScalar, Rows, Cols, Options, MaxRows, MaxCols>& res,
572     const ComplexMatrix& T,
573     const ComplexMatrix& U)
574 { res.noalias() = U * (T.template triangularView<Upper>() * U.adjoint()); }
575 
576 template<typename MatrixType>
577 template<int Rows, int Cols, int Options, int MaxRows, int MaxCols>
revertSchur(Matrix<RealScalar,Rows,Cols,Options,MaxRows,MaxCols> & res,const ComplexMatrix & T,const ComplexMatrix & U)578 inline void MatrixPower<MatrixType>::revertSchur(
579     Matrix<RealScalar, Rows, Cols, Options, MaxRows, MaxCols>& res,
580     const ComplexMatrix& T,
581     const ComplexMatrix& U)
582 { res.noalias() = (U * (T.template triangularView<Upper>() * U.adjoint())).real(); }
583 
584 /**
585  * \ingroup MatrixFunctions_Module
586  *
587  * \brief Proxy for the matrix power of some matrix (expression).
588  *
589  * \tparam Derived  type of the base, a matrix (expression).
590  *
591  * This class holds the arguments to the matrix power until it is
592  * assigned or evaluated for some other reason (so the argument
593  * should not be changed in the meantime). It is the return type of
594  * MatrixBase::pow() and related functions and most of the
595  * time this is the only way it is used.
596  */
597 template<typename Derived>
598 class MatrixPowerReturnValue : public ReturnByValue< MatrixPowerReturnValue<Derived> >
599 {
600   public:
601     typedef typename Derived::PlainObject PlainObject;
602     typedef typename Derived::RealScalar RealScalar;
603     typedef typename Derived::Index Index;
604 
605     /**
606      * \brief Constructor.
607      *
608      * \param[in] A  %Matrix (expression), the base of the matrix power.
609      * \param[in] p  real scalar, the exponent of the matrix power.
610      */
MatrixPowerReturnValue(const Derived & A,RealScalar p)611     MatrixPowerReturnValue(const Derived& A, RealScalar p) : m_A(A), m_p(p)
612     { }
613 
614     /**
615      * \brief Compute the matrix power.
616      *
617      * \param[out] result  \f$ A^p \f$ where \p A and \p p are as in the
618      * constructor.
619      */
620     template<typename ResultType>
evalTo(ResultType & res)621     inline void evalTo(ResultType& res) const
622     { MatrixPower<PlainObject>(m_A.eval()).compute(res, m_p); }
623 
rows()624     Index rows() const { return m_A.rows(); }
cols()625     Index cols() const { return m_A.cols(); }
626 
627   private:
628     const Derived& m_A;
629     const RealScalar m_p;
630 };
631 
632 /**
633  * \ingroup MatrixFunctions_Module
634  *
635  * \brief Proxy for the matrix power of some matrix (expression).
636  *
637  * \tparam Derived  type of the base, a matrix (expression).
638  *
639  * This class holds the arguments to the matrix power until it is
640  * assigned or evaluated for some other reason (so the argument
641  * should not be changed in the meantime). It is the return type of
642  * MatrixBase::pow() and related functions and most of the
643  * time this is the only way it is used.
644  */
645 template<typename Derived>
646 class MatrixComplexPowerReturnValue : public ReturnByValue< MatrixComplexPowerReturnValue<Derived> >
647 {
648   public:
649     typedef typename Derived::PlainObject PlainObject;
650     typedef typename std::complex<typename Derived::RealScalar> ComplexScalar;
651     typedef typename Derived::Index Index;
652 
653     /**
654      * \brief Constructor.
655      *
656      * \param[in] A  %Matrix (expression), the base of the matrix power.
657      * \param[in] p  complex scalar, the exponent of the matrix power.
658      */
MatrixComplexPowerReturnValue(const Derived & A,const ComplexScalar & p)659     MatrixComplexPowerReturnValue(const Derived& A, const ComplexScalar& p) : m_A(A), m_p(p)
660     { }
661 
662     /**
663      * \brief Compute the matrix power.
664      *
665      * Because \p p is complex, \f$ A^p \f$ is simply evaluated as \f$
666      * \exp(p \log(A)) \f$.
667      *
668      * \param[out] result  \f$ A^p \f$ where \p A and \p p are as in the
669      * constructor.
670      */
671     template<typename ResultType>
evalTo(ResultType & res)672     inline void evalTo(ResultType& res) const
673     { res = (m_p * m_A.log()).exp(); }
674 
rows()675     Index rows() const { return m_A.rows(); }
cols()676     Index cols() const { return m_A.cols(); }
677 
678   private:
679     const Derived& m_A;
680     const ComplexScalar m_p;
681 };
682 
683 namespace internal {
684 
685 template<typename MatrixPowerType>
686 struct traits< MatrixPowerParenthesesReturnValue<MatrixPowerType> >
687 { typedef typename MatrixPowerType::PlainObject ReturnType; };
688 
689 template<typename Derived>
690 struct traits< MatrixPowerReturnValue<Derived> >
691 { typedef typename Derived::PlainObject ReturnType; };
692 
693 template<typename Derived>
694 struct traits< MatrixComplexPowerReturnValue<Derived> >
695 { typedef typename Derived::PlainObject ReturnType; };
696 
697 }
698 
699 template<typename Derived>
700 const MatrixPowerReturnValue<Derived> MatrixBase<Derived>::pow(const RealScalar& p) const
701 { return MatrixPowerReturnValue<Derived>(derived(), p); }
702 
703 template<typename Derived>
704 const MatrixComplexPowerReturnValue<Derived> MatrixBase<Derived>::pow(const std::complex<RealScalar>& p) const
705 { return MatrixComplexPowerReturnValue<Derived>(derived(), p); }
706 
707 } // namespace Eigen
708 
709 #endif // EIGEN_MATRIX_POWER
710