1// This file is part of Eigen, a lightweight C++ template library 2// for linear algebra. 3// 4// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> 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_LLT_H 11#define EIGEN_LLT_H 12 13namespace Eigen { 14 15namespace internal{ 16template<typename MatrixType, int UpLo> struct LLT_Traits; 17} 18 19/** \ingroup Cholesky_Module 20 * 21 * \class LLT 22 * 23 * \brief Standard Cholesky decomposition (LL^T) of a matrix and associated features 24 * 25 * \param MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition 26 * \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper. 27 * The other triangular part won't be read. 28 * 29 * This class performs a LL^T Cholesky decomposition of a symmetric, positive definite 30 * matrix A such that A = LL^* = U^*U, where L is lower triangular. 31 * 32 * While the Cholesky decomposition is particularly useful to solve selfadjoint problems like D^*D x = b, 33 * for that purpose, we recommend the Cholesky decomposition without square root which is more stable 34 * and even faster. Nevertheless, this standard Cholesky decomposition remains useful in many other 35 * situations like generalised eigen problems with hermitian matrices. 36 * 37 * Remember that Cholesky decompositions are not rank-revealing. This LLT decomposition is only stable on positive definite matrices, 38 * use LDLT instead for the semidefinite case. Also, do not use a Cholesky decomposition to determine whether a system of equations 39 * has a solution. 40 * 41 * Example: \include LLT_example.cpp 42 * Output: \verbinclude LLT_example.out 43 * 44 * \sa MatrixBase::llt(), class LDLT 45 */ 46 /* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH) 47 * Note that during the decomposition, only the upper triangular part of A is considered. Therefore, 48 * the strict lower part does not have to store correct values. 49 */ 50template<typename _MatrixType, int _UpLo> class LLT 51{ 52 public: 53 typedef _MatrixType MatrixType; 54 enum { 55 RowsAtCompileTime = MatrixType::RowsAtCompileTime, 56 ColsAtCompileTime = MatrixType::ColsAtCompileTime, 57 Options = MatrixType::Options, 58 MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime 59 }; 60 typedef typename MatrixType::Scalar Scalar; 61 typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar; 62 typedef typename MatrixType::Index Index; 63 64 enum { 65 PacketSize = internal::packet_traits<Scalar>::size, 66 AlignmentMask = int(PacketSize)-1, 67 UpLo = _UpLo 68 }; 69 70 typedef internal::LLT_Traits<MatrixType,UpLo> Traits; 71 72 /** 73 * \brief Default Constructor. 74 * 75 * The default constructor is useful in cases in which the user intends to 76 * perform decompositions via LLT::compute(const MatrixType&). 77 */ 78 LLT() : m_matrix(), m_isInitialized(false) {} 79 80 /** \brief Default Constructor with memory preallocation 81 * 82 * Like the default constructor but with preallocation of the internal data 83 * according to the specified problem \a size. 84 * \sa LLT() 85 */ 86 LLT(Index size) : m_matrix(size, size), 87 m_isInitialized(false) {} 88 89 LLT(const MatrixType& matrix) 90 : m_matrix(matrix.rows(), matrix.cols()), 91 m_isInitialized(false) 92 { 93 compute(matrix); 94 } 95 96 /** \returns a view of the upper triangular matrix U */ 97 inline typename Traits::MatrixU matrixU() const 98 { 99 eigen_assert(m_isInitialized && "LLT is not initialized."); 100 return Traits::getU(m_matrix); 101 } 102 103 /** \returns a view of the lower triangular matrix L */ 104 inline typename Traits::MatrixL matrixL() const 105 { 106 eigen_assert(m_isInitialized && "LLT is not initialized."); 107 return Traits::getL(m_matrix); 108 } 109 110 /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A. 111 * 112 * Since this LLT class assumes anyway that the matrix A is invertible, the solution 113 * theoretically exists and is unique regardless of b. 114 * 115 * Example: \include LLT_solve.cpp 116 * Output: \verbinclude LLT_solve.out 117 * 118 * \sa solveInPlace(), MatrixBase::llt() 119 */ 120 template<typename Rhs> 121 inline const internal::solve_retval<LLT, Rhs> 122 solve(const MatrixBase<Rhs>& b) const 123 { 124 eigen_assert(m_isInitialized && "LLT is not initialized."); 125 eigen_assert(m_matrix.rows()==b.rows() 126 && "LLT::solve(): invalid number of rows of the right hand side matrix b"); 127 return internal::solve_retval<LLT, Rhs>(*this, b.derived()); 128 } 129 130 #ifdef EIGEN2_SUPPORT 131 template<typename OtherDerived, typename ResultType> 132 bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const 133 { 134 *result = this->solve(b); 135 return true; 136 } 137 138 bool isPositiveDefinite() const { return true; } 139 #endif 140 141 template<typename Derived> 142 void solveInPlace(MatrixBase<Derived> &bAndX) const; 143 144 LLT& compute(const MatrixType& matrix); 145 146 /** \returns the LLT decomposition matrix 147 * 148 * TODO: document the storage layout 149 */ 150 inline const MatrixType& matrixLLT() const 151 { 152 eigen_assert(m_isInitialized && "LLT is not initialized."); 153 return m_matrix; 154 } 155 156 MatrixType reconstructedMatrix() const; 157 158 159 /** \brief Reports whether previous computation was successful. 160 * 161 * \returns \c Success if computation was succesful, 162 * \c NumericalIssue if the matrix.appears to be negative. 163 */ 164 ComputationInfo info() const 165 { 166 eigen_assert(m_isInitialized && "LLT is not initialized."); 167 return m_info; 168 } 169 170 inline Index rows() const { return m_matrix.rows(); } 171 inline Index cols() const { return m_matrix.cols(); } 172 173 template<typename VectorType> 174 LLT rankUpdate(const VectorType& vec, const RealScalar& sigma = 1); 175 176 protected: 177 178 static void check_template_parameters() 179 { 180 EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar); 181 } 182 183 /** \internal 184 * Used to compute and store L 185 * The strict upper part is not used and even not initialized. 186 */ 187 MatrixType m_matrix; 188 bool m_isInitialized; 189 ComputationInfo m_info; 190}; 191 192namespace internal { 193 194template<typename Scalar, int UpLo> struct llt_inplace; 195 196template<typename MatrixType, typename VectorType> 197static typename MatrixType::Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) 198{ 199 using std::sqrt; 200 typedef typename MatrixType::Scalar Scalar; 201 typedef typename MatrixType::RealScalar RealScalar; 202 typedef typename MatrixType::Index Index; 203 typedef typename MatrixType::ColXpr ColXpr; 204 typedef typename internal::remove_all<ColXpr>::type ColXprCleaned; 205 typedef typename ColXprCleaned::SegmentReturnType ColXprSegment; 206 typedef Matrix<Scalar,Dynamic,1> TempVectorType; 207 typedef typename TempVectorType::SegmentReturnType TempVecSegment; 208 209 Index n = mat.cols(); 210 eigen_assert(mat.rows()==n && vec.size()==n); 211 212 TempVectorType temp; 213 214 if(sigma>0) 215 { 216 // This version is based on Givens rotations. 217 // It is faster than the other one below, but only works for updates, 218 // i.e., for sigma > 0 219 temp = sqrt(sigma) * vec; 220 221 for(Index i=0; i<n; ++i) 222 { 223 JacobiRotation<Scalar> g; 224 g.makeGivens(mat(i,i), -temp(i), &mat(i,i)); 225 226 Index rs = n-i-1; 227 if(rs>0) 228 { 229 ColXprSegment x(mat.col(i).tail(rs)); 230 TempVecSegment y(temp.tail(rs)); 231 apply_rotation_in_the_plane(x, y, g); 232 } 233 } 234 } 235 else 236 { 237 temp = vec; 238 RealScalar beta = 1; 239 for(Index j=0; j<n; ++j) 240 { 241 RealScalar Ljj = numext::real(mat.coeff(j,j)); 242 RealScalar dj = numext::abs2(Ljj); 243 Scalar wj = temp.coeff(j); 244 RealScalar swj2 = sigma*numext::abs2(wj); 245 RealScalar gamma = dj*beta + swj2; 246 247 RealScalar x = dj + swj2/beta; 248 if (x<=RealScalar(0)) 249 return j; 250 RealScalar nLjj = sqrt(x); 251 mat.coeffRef(j,j) = nLjj; 252 beta += swj2/dj; 253 254 // Update the terms of L 255 Index rs = n-j-1; 256 if(rs) 257 { 258 temp.tail(rs) -= (wj/Ljj) * mat.col(j).tail(rs); 259 if(gamma != 0) 260 mat.col(j).tail(rs) = (nLjj/Ljj) * mat.col(j).tail(rs) + (nLjj * sigma*numext::conj(wj)/gamma)*temp.tail(rs); 261 } 262 } 263 } 264 return -1; 265} 266 267template<typename Scalar> struct llt_inplace<Scalar, Lower> 268{ 269 typedef typename NumTraits<Scalar>::Real RealScalar; 270 template<typename MatrixType> 271 static typename MatrixType::Index unblocked(MatrixType& mat) 272 { 273 using std::sqrt; 274 typedef typename MatrixType::Index Index; 275 276 eigen_assert(mat.rows()==mat.cols()); 277 const Index size = mat.rows(); 278 for(Index k = 0; k < size; ++k) 279 { 280 Index rs = size-k-1; // remaining size 281 282 Block<MatrixType,Dynamic,1> A21(mat,k+1,k,rs,1); 283 Block<MatrixType,1,Dynamic> A10(mat,k,0,1,k); 284 Block<MatrixType,Dynamic,Dynamic> A20(mat,k+1,0,rs,k); 285 286 RealScalar x = numext::real(mat.coeff(k,k)); 287 if (k>0) x -= A10.squaredNorm(); 288 if (x<=RealScalar(0)) 289 return k; 290 mat.coeffRef(k,k) = x = sqrt(x); 291 if (k>0 && rs>0) A21.noalias() -= A20 * A10.adjoint(); 292 if (rs>0) A21 *= RealScalar(1)/x; 293 } 294 return -1; 295 } 296 297 template<typename MatrixType> 298 static typename MatrixType::Index blocked(MatrixType& m) 299 { 300 typedef typename MatrixType::Index Index; 301 eigen_assert(m.rows()==m.cols()); 302 Index size = m.rows(); 303 if(size<32) 304 return unblocked(m); 305 306 Index blockSize = size/8; 307 blockSize = (blockSize/16)*16; 308 blockSize = (std::min)((std::max)(blockSize,Index(8)), Index(128)); 309 310 for (Index k=0; k<size; k+=blockSize) 311 { 312 // partition the matrix: 313 // A00 | - | - 314 // lu = A10 | A11 | - 315 // A20 | A21 | A22 316 Index bs = (std::min)(blockSize, size-k); 317 Index rs = size - k - bs; 318 Block<MatrixType,Dynamic,Dynamic> A11(m,k, k, bs,bs); 319 Block<MatrixType,Dynamic,Dynamic> A21(m,k+bs,k, rs,bs); 320 Block<MatrixType,Dynamic,Dynamic> A22(m,k+bs,k+bs,rs,rs); 321 322 Index ret; 323 if((ret=unblocked(A11))>=0) return k+ret; 324 if(rs>0) A11.adjoint().template triangularView<Upper>().template solveInPlace<OnTheRight>(A21); 325 if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,-1); // bottleneck 326 } 327 return -1; 328 } 329 330 template<typename MatrixType, typename VectorType> 331 static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma) 332 { 333 return Eigen::internal::llt_rank_update_lower(mat, vec, sigma); 334 } 335}; 336 337template<typename Scalar> struct llt_inplace<Scalar, Upper> 338{ 339 typedef typename NumTraits<Scalar>::Real RealScalar; 340 341 template<typename MatrixType> 342 static EIGEN_STRONG_INLINE typename MatrixType::Index unblocked(MatrixType& mat) 343 { 344 Transpose<MatrixType> matt(mat); 345 return llt_inplace<Scalar, Lower>::unblocked(matt); 346 } 347 template<typename MatrixType> 348 static EIGEN_STRONG_INLINE typename MatrixType::Index blocked(MatrixType& mat) 349 { 350 Transpose<MatrixType> matt(mat); 351 return llt_inplace<Scalar, Lower>::blocked(matt); 352 } 353 template<typename MatrixType, typename VectorType> 354 static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma) 355 { 356 Transpose<MatrixType> matt(mat); 357 return llt_inplace<Scalar, Lower>::rankUpdate(matt, vec.conjugate(), sigma); 358 } 359}; 360 361template<typename MatrixType> struct LLT_Traits<MatrixType,Lower> 362{ 363 typedef const TriangularView<const MatrixType, Lower> MatrixL; 364 typedef const TriangularView<const typename MatrixType::AdjointReturnType, Upper> MatrixU; 365 static inline MatrixL getL(const MatrixType& m) { return m; } 366 static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); } 367 static bool inplace_decomposition(MatrixType& m) 368 { return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; } 369}; 370 371template<typename MatrixType> struct LLT_Traits<MatrixType,Upper> 372{ 373 typedef const TriangularView<const typename MatrixType::AdjointReturnType, Lower> MatrixL; 374 typedef const TriangularView<const MatrixType, Upper> MatrixU; 375 static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); } 376 static inline MatrixU getU(const MatrixType& m) { return m; } 377 static bool inplace_decomposition(MatrixType& m) 378 { return llt_inplace<typename MatrixType::Scalar, Upper>::blocked(m)==-1; } 379}; 380 381} // end namespace internal 382 383/** Computes / recomputes the Cholesky decomposition A = LL^* = U^*U of \a matrix 384 * 385 * \returns a reference to *this 386 * 387 * Example: \include TutorialLinAlgComputeTwice.cpp 388 * Output: \verbinclude TutorialLinAlgComputeTwice.out 389 */ 390template<typename MatrixType, int _UpLo> 391LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a) 392{ 393 check_template_parameters(); 394 395 eigen_assert(a.rows()==a.cols()); 396 const Index size = a.rows(); 397 m_matrix.resize(size, size); 398 m_matrix = a; 399 400 m_isInitialized = true; 401 bool ok = Traits::inplace_decomposition(m_matrix); 402 m_info = ok ? Success : NumericalIssue; 403 404 return *this; 405} 406 407/** Performs a rank one update (or dowdate) of the current decomposition. 408 * If A = LL^* before the rank one update, 409 * then after it we have LL^* = A + sigma * v v^* where \a v must be a vector 410 * of same dimension. 411 */ 412template<typename _MatrixType, int _UpLo> 413template<typename VectorType> 414LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma) 415{ 416 EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorType); 417 eigen_assert(v.size()==m_matrix.cols()); 418 eigen_assert(m_isInitialized); 419 if(internal::llt_inplace<typename MatrixType::Scalar, UpLo>::rankUpdate(m_matrix,v,sigma)>=0) 420 m_info = NumericalIssue; 421 else 422 m_info = Success; 423 424 return *this; 425} 426 427namespace internal { 428template<typename _MatrixType, int UpLo, typename Rhs> 429struct solve_retval<LLT<_MatrixType, UpLo>, Rhs> 430 : solve_retval_base<LLT<_MatrixType, UpLo>, Rhs> 431{ 432 typedef LLT<_MatrixType,UpLo> LLTType; 433 EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs) 434 435 template<typename Dest> void evalTo(Dest& dst) const 436 { 437 dst = rhs(); 438 dec().solveInPlace(dst); 439 } 440}; 441} 442 443/** \internal use x = llt_object.solve(x); 444 * 445 * This is the \em in-place version of solve(). 446 * 447 * \param bAndX represents both the right-hand side matrix b and result x. 448 * 449 * \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD. 450 * 451 * This version avoids a copy when the right hand side matrix b is not 452 * needed anymore. 453 * 454 * \sa LLT::solve(), MatrixBase::llt() 455 */ 456template<typename MatrixType, int _UpLo> 457template<typename Derived> 458void LLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const 459{ 460 eigen_assert(m_isInitialized && "LLT is not initialized."); 461 eigen_assert(m_matrix.rows()==bAndX.rows()); 462 matrixL().solveInPlace(bAndX); 463 matrixU().solveInPlace(bAndX); 464} 465 466/** \returns the matrix represented by the decomposition, 467 * i.e., it returns the product: L L^*. 468 * This function is provided for debug purpose. */ 469template<typename MatrixType, int _UpLo> 470MatrixType LLT<MatrixType,_UpLo>::reconstructedMatrix() const 471{ 472 eigen_assert(m_isInitialized && "LLT is not initialized."); 473 return matrixL() * matrixL().adjoint().toDenseMatrix(); 474} 475 476/** \cholesky_module 477 * \returns the LLT decomposition of \c *this 478 */ 479template<typename Derived> 480inline const LLT<typename MatrixBase<Derived>::PlainObject> 481MatrixBase<Derived>::llt() const 482{ 483 return LLT<PlainObject>(derived()); 484} 485 486/** \cholesky_module 487 * \returns the LLT decomposition of \c *this 488 */ 489template<typename MatrixType, unsigned int UpLo> 490inline const LLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo> 491SelfAdjointView<MatrixType, UpLo>::llt() const 492{ 493 return LLT<PlainObject,UpLo>(m_matrix); 494} 495 496} // end namespace Eigen 497 498#endif // EIGEN_LLT_H 499