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