Belos Package Browser (Single Doxygen Collection)  Development
BelosRCGSolMgr.hpp
Go to the documentation of this file.
1 //@HEADER
2 // ************************************************************************
3 //
4 // Belos: Block Linear Solvers Package
5 // Copyright 2004 Sandia Corporation
6 //
7 // Under the terms of Contract DE-AC04-94AL85000 with Sandia Corporation,
8 // the U.S. Government retains certain rights in this software.
9 //
10 // Redistribution and use in source and binary forms, with or without
11 // modification, are permitted provided that the following conditions are
12 // met:
13 //
14 // 1. Redistributions of source code must retain the above copyright
15 // notice, this list of conditions and the following disclaimer.
16 //
17 // 2. Redistributions in binary form must reproduce the above copyright
18 // notice, this list of conditions and the following disclaimer in the
19 // documentation and/or other materials provided with the distribution.
20 //
21 // 3. Neither the name of the Corporation nor the names of the
22 // contributors may be used to endorse or promote products derived from
23 // this software without specific prior written permission.
24 //
25 // THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "AS IS" AND ANY
26 // EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28 // PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL SANDIA CORPORATION OR THE
29 // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
30 // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
31 // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
32 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
33 // LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
34 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
35 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 //
37 // Questions? Contact Michael A. Heroux (maherou@sandia.gov)
38 //
39 // ************************************************************************
40 //@HEADER
41 
42 #ifndef BELOS_RCG_SOLMGR_HPP
43 #define BELOS_RCG_SOLMGR_HPP
44 
49 #include "BelosConfigDefs.hpp"
50 #include "BelosTypes.hpp"
51 
52 #include "BelosLinearProblem.hpp"
53 #include "BelosSolverManager.hpp"
54 
55 #include "BelosRCGIter.hpp"
61 #include "BelosStatusTestCombo.hpp"
63 #include "BelosOutputManager.hpp"
64 #include "Teuchos_BLAS.hpp"
65 #include "Teuchos_LAPACK.hpp"
66 #include "Teuchos_as.hpp"
67 #ifdef BELOS_TEUCHOS_TIME_MONITOR
68 #include "Teuchos_TimeMonitor.hpp"
69 #endif
70 
110 namespace Belos {
111 
113 
114 
122  RCGSolMgrLinearProblemFailure(const std::string& what_arg) : BelosError(what_arg)
123  {}};
124 
131  class RCGSolMgrLAPACKFailure : public BelosError {public:
132  RCGSolMgrLAPACKFailure(const std::string& what_arg) : BelosError(what_arg)
133  {}};
134 
141  class RCGSolMgrRecyclingFailure : public BelosError {public:
142  RCGSolMgrRecyclingFailure(const std::string& what_arg) : BelosError(what_arg)
143  {}};
144 
146 
147 
148  // Partial specialization for unsupported ScalarType types.
149  // This contains a stub implementation.
150  template<class ScalarType, class MV, class OP,
151  const bool supportsScalarType =
154  class RCGSolMgr :
155  public Details::SolverManagerRequiresRealLapack<ScalarType, MV, OP,
156  Belos::Details::LapackSupportsScalar<ScalarType>::value &&
157  ! Teuchos::ScalarTraits<ScalarType>::isComplex>
158  {
159  static const bool scalarTypeIsSupported =
162  typedef Details::SolverManagerRequiresRealLapack<ScalarType, MV, OP,
164 
165  public:
167  base_type ()
168  {}
171  base_type ()
172  {}
173  virtual ~RCGSolMgr () {}
174  };
175 
176  // Partial specialization for real ScalarType.
177  // This contains the actual working implementation of RCG.
178  // See discussion in the class documentation above.
179  template<class ScalarType, class MV, class OP>
180  class RCGSolMgr<ScalarType, MV, OP, true> :
181  public Details::SolverManagerRequiresRealLapack<ScalarType, MV, OP, true> {
182  private:
188 
189  public:
190 
192 
193 
199  RCGSolMgr();
200 
224 
226  virtual ~RCGSolMgr() {};
228 
230 
231 
233  return *problem_;
234  }
235 
237  Teuchos::RCP<const Teuchos::ParameterList> getValidParameters() const;
238 
241 
248  return Teuchos::tuple(timerSolve_);
249  }
250 
256  return achievedTol_;
257  }
258 
260  int getNumIters() const {
261  return numIters_;
262  }
263 
265  bool isLOADetected() const { return false; }
266 
268 
270 
271 
273  void setProblem( const Teuchos::RCP<LinearProblem<ScalarType,MV,OP> > &problem ) { problem_ = problem; }
274 
276  void setParameters( const Teuchos::RCP<Teuchos::ParameterList> &params );
277 
279 
281 
282 
288  void reset( const ResetType type ) {
289  if ((type & Belos::Problem) && !Teuchos::is_null(problem_)) problem_->setProblem();
290  else if (type & Belos::RecycleSubspace) existU_ = false;
291  }
293 
295 
296 
314  ReturnType solve();
315 
317 
320 
322  std::string description() const;
323 
325 
326  private:
327 
328  // Called by all constructors; Contains init instructions common to all constructors
329  void init();
330 
331  // Computes harmonic eigenpairs of projected matrix created during one cycle.
332  // Y contains the harmonic Ritz vectors corresponding to the recycleBlocks eigenvalues of smallest magnitude.
333  void getHarmonicVecs(const Teuchos::SerialDenseMatrix<int,ScalarType> &F,
336 
337  // Sort list of n floating-point numbers and return permutation vector
338  void sort(std::vector<ScalarType>& dlist, int n, std::vector<int>& iperm);
339 
340  // Initialize solver state storage
341  void initializeStateStorage();
342 
343  // Linear problem.
345 
346  // Output manager.
349 
350  // Status test.
355 
356  // Current parameter list.
358 
359  // Default solver values.
361  static const int maxIters_default_;
362  static const int blockSize_default_;
363  static const int numBlocks_default_;
364  static const int recycleBlocks_default_;
365  static const bool showMaxResNormOnly_default_;
366  static const int verbosity_default_;
367  static const int outputStyle_default_;
368  static const int outputFreq_default_;
369  static const std::string label_default_;
371 
372  //
373  // Current solver values.
374  //
375 
378 
384 
387 
390 
391  int numBlocks_, recycleBlocks_;
393  int verbosity_, outputStyle_, outputFreq_;
394 
396  // Solver State Storage
398  // Search vectors
400  //
401  // A times current search direction
403  //
404  // Residual vector
406  //
407  // Preconditioned residual
409  //
410  // Flag indicating that the recycle space should be used
411  bool existU_;
412  //
413  // Flag indicating that the updated recycle space has been created
414  bool existU1_;
415  //
416  // Recycled subspace and its image
418  //
419  // Recycled subspace for next system and its image
421  //
422  // Coefficients arising in RCG iteration
426  //
427  // Solutions to local least-squares problems
429  //
430  // The matrix U^T A U
432  //
433  // LU factorization of U^T A U
435  //
436  // Data from LU factorization of UTAU
438  //
439  // The matrix (AU)^T AU
441  //
442  // The scalar r'*z
444  //
445  // Matrices needed for calculation of harmonic Ritz eigenproblem
447  //
448  // Matrices needed for updating recycle space
455  ScalarType dold;
457 
458  // Timers.
459  std::string label_;
461 
462  // Internal state variables.
464  };
465 
466 
467 // Default solver values.
468 template<class ScalarType, class MV, class OP>
471 
472 template<class ScalarType, class MV, class OP>
474 
475 template<class ScalarType, class MV, class OP>
477 
478 template<class ScalarType, class MV, class OP>
480 
481 template<class ScalarType, class MV, class OP>
483 
484 template<class ScalarType, class MV, class OP>
486 
487 template<class ScalarType, class MV, class OP>
489 
490 template<class ScalarType, class MV, class OP>
492 
493 template<class ScalarType, class MV, class OP>
495 
496 template<class ScalarType, class MV, class OP>
497 const std::string RCGSolMgr<ScalarType,MV,OP,true>::label_default_ = "Belos";
498 
499 template<class ScalarType, class MV, class OP>
501 
502 // Empty Constructor
503 template<class ScalarType, class MV, class OP>
505  achievedTol_(0.0),
506  numIters_(0)
507 {
508  init();
509 }
510 
511 // Basic Constructor
512 template<class ScalarType, class MV, class OP>
516  problem_(problem),
517  achievedTol_(0.0),
518  numIters_(0)
519 {
520  init();
521  TEUCHOS_TEST_FOR_EXCEPTION(problem_ == Teuchos::null, std::invalid_argument, "Problem not given to solver manager.");
522 
523  // If the parameter list pointer is null, then set the current parameters to the default parameter list.
524  if ( !is_null(pl) ) {
525  setParameters( pl );
526  }
527 }
528 
529 // Common instructions executed in all constructors
530 template<class ScalarType, class MV, class OP>
532 {
533  outputStream_ = outputStream_default_;
534  convtol_ = convtol_default_;
535  maxIters_ = maxIters_default_;
536  numBlocks_ = numBlocks_default_;
537  recycleBlocks_ = recycleBlocks_default_;
538  verbosity_ = verbosity_default_;
539  outputStyle_= outputStyle_default_;
540  outputFreq_= outputFreq_default_;
541  showMaxResNormOnly_ = showMaxResNormOnly_default_;
542  label_ = label_default_;
543  params_Set_ = false;
544  P_ = Teuchos::null;
545  Ap_ = Teuchos::null;
546  r_ = Teuchos::null;
547  z_ = Teuchos::null;
548  existU_ = false;
549  existU1_ = false;
550  U_ = Teuchos::null;
551  AU_ = Teuchos::null;
552  U1_ = Teuchos::null;
553  Alpha_ = Teuchos::null;
554  Beta_ = Teuchos::null;
555  D_ = Teuchos::null;
556  Delta_ = Teuchos::null;
557  UTAU_ = Teuchos::null;
558  LUUTAU_ = Teuchos::null;
559  ipiv_ = Teuchos::null;
560  AUTAU_ = Teuchos::null;
561  rTz_old_ = Teuchos::null;
562  F_ = Teuchos::null;
563  G_ = Teuchos::null;
564  Y_ = Teuchos::null;
565  L2_ = Teuchos::null;
566  DeltaL2_ = Teuchos::null;
567  AU1TUDeltaL2_ = Teuchos::null;
568  AU1TAU1_ = Teuchos::null;
569  AU1TU1_ = Teuchos::null;
570  AU1TAP_ = Teuchos::null;
571  FY_ = Teuchos::null;
572  GY_ = Teuchos::null;
573  APTAP_ = Teuchos::null;
574  U1Y1_ = Teuchos::null;
575  PY2_ = Teuchos::null;
576  AUTAP_ = Teuchos::null;
577  AU1TU_ = Teuchos::null;
578  dold = 0.;
579 }
580 
581 template<class ScalarType, class MV, class OP>
583 {
584  // Create the internal parameter list if ones doesn't already exist.
585  if (params_ == Teuchos::null) {
586  params_ = Teuchos::rcp( new Teuchos::ParameterList(*getValidParameters()) );
587  }
588  else {
589  params->validateParameters(*getValidParameters());
590  }
591 
592  // Check for maximum number of iterations
593  if (params->isParameter("Maximum Iterations")) {
594  maxIters_ = params->get("Maximum Iterations",maxIters_default_);
595 
596  // Update parameter in our list and in status test.
597  params_->set("Maximum Iterations", maxIters_);
598  if (maxIterTest_!=Teuchos::null)
599  maxIterTest_->setMaxIters( maxIters_ );
600  }
601 
602  // Check for the maximum number of blocks.
603  if (params->isParameter("Num Blocks")) {
604  numBlocks_ = params->get("Num Blocks",numBlocks_default_);
605  TEUCHOS_TEST_FOR_EXCEPTION(numBlocks_ <= 0, std::invalid_argument,
606  "Belos::RCGSolMgr: \"Num Blocks\" must be strictly positive.");
607 
608  // Update parameter in our list.
609  params_->set("Num Blocks", numBlocks_);
610  }
611 
612  // Check for the maximum number of blocks.
613  if (params->isParameter("Num Recycled Blocks")) {
614  recycleBlocks_ = params->get("Num Recycled Blocks",recycleBlocks_default_);
615  TEUCHOS_TEST_FOR_EXCEPTION(recycleBlocks_ <= 0, std::invalid_argument,
616  "Belos::RCGSolMgr: \"Num Recycled Blocks\" must be strictly positive.");
617 
618  TEUCHOS_TEST_FOR_EXCEPTION(recycleBlocks_ >= numBlocks_, std::invalid_argument,
619  "Belos::RCGSolMgr: \"Num Recycled Blocks\" must be less than \"Num Blocks\".");
620 
621  // Update parameter in our list.
622  params_->set("Num Recycled Blocks", recycleBlocks_);
623  }
624 
625  // Check to see if the timer label changed.
626  if (params->isParameter("Timer Label")) {
627  std::string tempLabel = params->get("Timer Label", label_default_);
628 
629  // Update parameter in our list and solver timer
630  if (tempLabel != label_) {
631  label_ = tempLabel;
632  params_->set("Timer Label", label_);
633  std::string solveLabel = label_ + ": RCGSolMgr total solve time";
634 #ifdef BELOS_TEUCHOS_TIME_MONITOR
635  timerSolve_ = Teuchos::TimeMonitor::getNewCounter(solveLabel);
636 #endif
637  }
638  }
639 
640  // Check for a change in verbosity level
641  if (params->isParameter("Verbosity")) {
642  if (Teuchos::isParameterType<int>(*params,"Verbosity")) {
643  verbosity_ = params->get("Verbosity", verbosity_default_);
644  } else {
645  verbosity_ = (int)Teuchos::getParameter<Belos::MsgType>(*params,"Verbosity");
646  }
647 
648  // Update parameter in our list.
649  params_->set("Verbosity", verbosity_);
650  if (printer_ != Teuchos::null)
651  printer_->setVerbosity(verbosity_);
652  }
653 
654  // Check for a change in output style
655  if (params->isParameter("Output Style")) {
656  if (Teuchos::isParameterType<int>(*params,"Output Style")) {
657  outputStyle_ = params->get("Output Style", outputStyle_default_);
658  } else {
659  outputStyle_ = (int)Teuchos::getParameter<Belos::OutputType>(*params,"Output Style");
660  }
661 
662  // Reconstruct the convergence test if the explicit residual test is not being used.
663  params_->set("Output Style", outputStyle_);
664  outputTest_ = Teuchos::null;
665  }
666 
667  // output stream
668  if (params->isParameter("Output Stream")) {
669  outputStream_ = Teuchos::getParameter<Teuchos::RCP<std::ostream> >(*params,"Output Stream");
670 
671  // Update parameter in our list.
672  params_->set("Output Stream", outputStream_);
673  if (printer_ != Teuchos::null)
674  printer_->setOStream( outputStream_ );
675  }
676 
677  // frequency level
678  if (verbosity_ & Belos::StatusTestDetails) {
679  if (params->isParameter("Output Frequency")) {
680  outputFreq_ = params->get("Output Frequency", outputFreq_default_);
681  }
682 
683  // Update parameter in out list and output status test.
684  params_->set("Output Frequency", outputFreq_);
685  if (outputTest_ != Teuchos::null)
686  outputTest_->setOutputFrequency( outputFreq_ );
687  }
688 
689  // Create output manager if we need to.
690  if (printer_ == Teuchos::null) {
691  printer_ = Teuchos::rcp( new OutputManager<ScalarType>(verbosity_, outputStream_) );
692  }
693 
694  // Convergence
695  typedef Belos::StatusTestCombo<ScalarType,MV,OP> StatusTestCombo_t;
696  typedef Belos::StatusTestGenResNorm<ScalarType,MV,OP> StatusTestResNorm_t;
697 
698  // Check for convergence tolerance
699  if (params->isParameter("Convergence Tolerance")) {
700  convtol_ = params->get("Convergence Tolerance",convtol_default_);
701 
702  // Update parameter in our list and residual tests.
703  params_->set("Convergence Tolerance", convtol_);
704  if (convTest_ != Teuchos::null)
705  convTest_->setTolerance( convtol_ );
706  }
707 
708  if (params->isParameter("Show Maximum Residual Norm Only")) {
709  showMaxResNormOnly_ = Teuchos::getParameter<bool>(*params,"Show Maximum Residual Norm Only");
710 
711  // Update parameter in our list and residual tests
712  params_->set("Show Maximum Residual Norm Only", showMaxResNormOnly_);
713  if (convTest_ != Teuchos::null)
714  convTest_->setShowMaxResNormOnly( showMaxResNormOnly_ );
715  }
716 
717  // Create status tests if we need to.
718 
719  // Basic test checks maximum iterations and native residual.
720  if (maxIterTest_ == Teuchos::null)
721  maxIterTest_ = Teuchos::rcp( new StatusTestMaxIters<ScalarType,MV,OP>( maxIters_ ) );
722 
723  // Implicit residual test, using the native residual to determine if convergence was achieved.
724  if (convTest_ == Teuchos::null)
725  convTest_ = Teuchos::rcp( new StatusTestResNorm_t( convtol_, 1 ) );
726 
727  if (sTest_ == Teuchos::null)
728  sTest_ = Teuchos::rcp( new StatusTestCombo_t( StatusTestCombo_t::OR, maxIterTest_, convTest_ ) );
729 
730  if (outputTest_ == Teuchos::null) {
731 
732  // Create the status test output class.
733  // This class manages and formats the output from the status test.
734  StatusTestOutputFactory<ScalarType,MV,OP> stoFactory( outputStyle_ );
735  outputTest_ = stoFactory.create( printer_, sTest_, outputFreq_, Passed+Failed+Undefined );
736 
737  // Set the solver string for the output test
738  std::string solverDesc = " Recycling CG ";
739  outputTest_->setSolverDesc( solverDesc );
740  }
741 
742  // Create the timer if we need to.
743  if (timerSolve_ == Teuchos::null) {
744  std::string solveLabel = label_ + ": RCGSolMgr total solve time";
745 #ifdef BELOS_TEUCHOS_TIME_MONITOR
746  timerSolve_ = Teuchos::TimeMonitor::getNewCounter(solveLabel);
747 #endif
748  }
749 
750  // Inform the solver manager that the current parameters were set.
751  params_Set_ = true;
752 }
753 
754 
755 template<class ScalarType, class MV, class OP>
758 {
760 
761  // Set all the valid parameters and their default values.
762  if(is_null(validPL)) {
763  Teuchos::RCP<Teuchos::ParameterList> pl = Teuchos::parameterList();
764  pl->set("Convergence Tolerance", convtol_default_,
765  "The relative residual tolerance that needs to be achieved by the\n"
766  "iterative solver in order for the linear system to be declared converged.");
767  pl->set("Maximum Iterations", maxIters_default_,
768  "The maximum number of block iterations allowed for each\n"
769  "set of RHS solved.");
770  pl->set("Block Size", blockSize_default_,
771  "Block Size Parameter -- currently must be 1 for RCG");
772  pl->set("Num Blocks", numBlocks_default_,
773  "The length of a cycle (and this max number of search vectors kept)\n");
774  pl->set("Num Recycled Blocks", recycleBlocks_default_,
775  "The number of vectors in the recycle subspace.");
776  pl->set("Verbosity", verbosity_default_,
777  "What type(s) of solver information should be outputted\n"
778  "to the output stream.");
779  pl->set("Output Style", outputStyle_default_,
780  "What style is used for the solver information outputted\n"
781  "to the output stream.");
782  pl->set("Output Frequency", outputFreq_default_,
783  "How often convergence information should be outputted\n"
784  "to the output stream.");
785  pl->set("Output Stream", outputStream_default_,
786  "A reference-counted pointer to the output stream where all\n"
787  "solver output is sent.");
788  pl->set("Show Maximum Residual Norm Only", showMaxResNormOnly_default_,
789  "When convergence information is printed, only show the maximum\n"
790  "relative residual norm when the block size is greater than one.");
791  pl->set("Timer Label", label_default_,
792  "The string to use as a prefix for the timer labels.");
793  // pl->set("Restart Timers", restartTimers_);
794  validPL = pl;
795  }
796  return validPL;
797 }
798 
799 // initializeStateStorage
800 template<class ScalarType, class MV, class OP>
802 
803  // Check if there is any multivector to clone from.
804  Teuchos::RCP<const MV> rhsMV = problem_->getRHS();
805  if (rhsMV == Teuchos::null) {
806  // Nothing to do
807  return;
808  }
809  else {
810 
811  // Initialize the state storage
812  TEUCHOS_TEST_FOR_EXCEPTION(static_cast<ptrdiff_t>(numBlocks_) > MVT::GetGlobalLength(*rhsMV),std::invalid_argument,
813  "Belos::RCGSolMgr::initializeStateStorage(): Cannot generate a Krylov basis with dimension larger the operator!");
814 
815  // If the subspace has not been initialized before, generate it using the RHS from lp_.
816  if (P_ == Teuchos::null) {
817  P_ = MVT::Clone( *rhsMV, numBlocks_+2 );
818  }
819  else {
820  // Generate P_ by cloning itself ONLY if more space is needed.
821  if (MVT::GetNumberVecs(*P_) < numBlocks_+2) {
822  Teuchos::RCP<const MV> tmp = P_;
823  P_ = MVT::Clone( *tmp, numBlocks_+2 );
824  }
825  }
826 
827  // Generate Ap_ only if it doesn't exist
828  if (Ap_ == Teuchos::null)
829  Ap_ = MVT::Clone( *rhsMV, 1 );
830 
831  // Generate r_ only if it doesn't exist
832  if (r_ == Teuchos::null)
833  r_ = MVT::Clone( *rhsMV, 1 );
834 
835  // Generate z_ only if it doesn't exist
836  if (z_ == Teuchos::null)
837  z_ = MVT::Clone( *rhsMV, 1 );
838 
839  // If the recycle space has not been initialized before, generate it using the RHS from lp_.
840  if (U_ == Teuchos::null) {
841  U_ = MVT::Clone( *rhsMV, recycleBlocks_ );
842  }
843  else {
844  // Generate U_ by cloning itself ONLY if more space is needed.
845  if (MVT::GetNumberVecs(*U_) < recycleBlocks_) {
846  Teuchos::RCP<const MV> tmp = U_;
847  U_ = MVT::Clone( *tmp, recycleBlocks_ );
848  }
849  }
850 
851  // If the recycle space has not be initialized before, generate it using the RHS from lp_.
852  if (AU_ == Teuchos::null) {
853  AU_ = MVT::Clone( *rhsMV, recycleBlocks_ );
854  }
855  else {
856  // Generate AU_ by cloning itself ONLY if more space is needed.
857  if (MVT::GetNumberVecs(*AU_) < recycleBlocks_) {
858  Teuchos::RCP<const MV> tmp = AU_;
859  AU_ = MVT::Clone( *tmp, recycleBlocks_ );
860  }
861  }
862 
863  // If the recycle space has not been initialized before, generate it using the RHS from lp_.
864  if (U1_ == Teuchos::null) {
865  U1_ = MVT::Clone( *rhsMV, recycleBlocks_ );
866  }
867  else {
868  // Generate U1_ by cloning itself ONLY if more space is needed.
869  if (MVT::GetNumberVecs(*U1_) < recycleBlocks_) {
870  Teuchos::RCP<const MV> tmp = U1_;
871  U1_ = MVT::Clone( *tmp, recycleBlocks_ );
872  }
873  }
874 
875  // Generate Alpha_ only if it doesn't exist, otherwise resize it.
876  if (Alpha_ == Teuchos::null)
877  Alpha_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_, 1 ) );
878  else {
879  if ( (Alpha_->numRows() != numBlocks_) || (Alpha_->numCols() != 1) )
880  Alpha_->reshape( numBlocks_, 1 );
881  }
882 
883  // Generate Beta_ only if it doesn't exist, otherwise resize it.
884  if (Beta_ == Teuchos::null)
885  Beta_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_ + 1, 1 ) );
886  else {
887  if ( (Beta_->numRows() != (numBlocks_+1)) || (Beta_->numCols() != 1) )
888  Beta_->reshape( numBlocks_ + 1, 1 );
889  }
890 
891  // Generate D_ only if it doesn't exist, otherwise resize it.
892  if (D_ == Teuchos::null)
893  D_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_ , 1 ) );
894  else {
895  if ( (D_->numRows() != numBlocks_) || (D_->numCols() != 1) )
896  D_->reshape( numBlocks_, 1 );
897  }
898 
899  // Generate Delta_ only if it doesn't exist, otherwise resize it.
900  if (Delta_ == Teuchos::null)
901  Delta_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, numBlocks_ + 1 ) );
902  else {
903  if ( (Delta_->numRows() != recycleBlocks_) || (Delta_->numCols() != (numBlocks_ + 1)) )
904  Delta_->reshape( recycleBlocks_, numBlocks_ + 1 );
905  }
906 
907  // Generate UTAU_ only if it doesn't exist, otherwise resize it.
908  if (UTAU_ == Teuchos::null)
909  UTAU_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, recycleBlocks_ ) );
910  else {
911  if ( (UTAU_->numRows() != recycleBlocks_) || (UTAU_->numCols() != recycleBlocks_) )
912  UTAU_->reshape( recycleBlocks_, recycleBlocks_ );
913  }
914 
915  // Generate LUUTAU_ only if it doesn't exist, otherwise resize it.
916  if (LUUTAU_ == Teuchos::null)
917  LUUTAU_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, recycleBlocks_ ) );
918  else {
919  if ( (LUUTAU_->numRows() != recycleBlocks_) || (LUUTAU_->numCols() != recycleBlocks_) )
920  LUUTAU_->reshape( recycleBlocks_, recycleBlocks_ );
921  }
922 
923  // Generate ipiv_ only if it doesn't exist, otherwise resize it.
924  if (ipiv_ == Teuchos::null)
925  ipiv_ = Teuchos::rcp( new std::vector<int>(recycleBlocks_) );
926  else {
927  if ( (int)ipiv_->size() != recycleBlocks_ ) // if ipiv not correct size, always resize it
928  ipiv_->resize(recycleBlocks_);
929  }
930 
931  // Generate AUTAU_ only if it doesn't exist, otherwise resize it.
932  if (AUTAU_ == Teuchos::null)
933  AUTAU_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, recycleBlocks_ ) );
934  else {
935  if ( (AUTAU_->numRows() != recycleBlocks_) || (AUTAU_->numCols() != recycleBlocks_) )
936  AUTAU_->reshape( recycleBlocks_, recycleBlocks_ );
937  }
938 
939  // Generate rTz_old_ only if it doesn't exist
940  if (rTz_old_ == Teuchos::null)
942  else {
943  if ( (rTz_old_->numRows() != 1) || (rTz_old_->numCols() != 1) )
944  rTz_old_->reshape( 1, 1 );
945  }
946 
947  // Generate F_ only if it doesn't exist
948  if (F_ == Teuchos::null)
949  F_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_+recycleBlocks_, numBlocks_+recycleBlocks_ ) );
950  else {
951  if ( (F_->numRows() != (numBlocks_+recycleBlocks_)) || (F_->numCols() != numBlocks_+recycleBlocks_) )
952  F_->reshape( numBlocks_+recycleBlocks_, numBlocks_+recycleBlocks_ );
953  }
954 
955  // Generate G_ only if it doesn't exist
956  if (G_ == Teuchos::null)
957  G_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_+recycleBlocks_, numBlocks_+recycleBlocks_ ) );
958  else {
959  if ( (G_->numRows() != (numBlocks_+recycleBlocks_)) || (G_->numCols() != numBlocks_+recycleBlocks_) )
960  G_->reshape( numBlocks_+recycleBlocks_, numBlocks_+recycleBlocks_ );
961  }
962 
963  // Generate Y_ only if it doesn't exist
964  if (Y_ == Teuchos::null)
965  Y_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_+recycleBlocks_, recycleBlocks_ ) );
966  else {
967  if ( (Y_->numRows() != (numBlocks_+recycleBlocks_)) || (Y_->numCols() != numBlocks_+recycleBlocks_) )
968  Y_->reshape( numBlocks_+recycleBlocks_, numBlocks_+recycleBlocks_ );
969  }
970 
971  // Generate L2_ only if it doesn't exist
972  if (L2_ == Teuchos::null)
973  L2_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_+1, numBlocks_ ) );
974  else {
975  if ( (L2_->numRows() != (numBlocks_+1)) || (L2_->numCols() != numBlocks_) )
976  L2_->reshape( numBlocks_+1, numBlocks_ );
977  }
978 
979  // Generate DeltaL2_ only if it doesn't exist
980  if (DeltaL2_ == Teuchos::null)
981  DeltaL2_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, numBlocks_ ) );
982  else {
983  if ( (DeltaL2_->numRows() != (recycleBlocks_)) || (DeltaL2_->numCols() != (numBlocks_) ) )
984  DeltaL2_->reshape( recycleBlocks_, numBlocks_ );
985  }
986 
987  // Generate AU1TUDeltaL2_ only if it doesn't exist
988  if (AU1TUDeltaL2_ == Teuchos::null)
989  AU1TUDeltaL2_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, numBlocks_ ) );
990  else {
991  if ( (AU1TUDeltaL2_->numRows() != (recycleBlocks_)) || (AU1TUDeltaL2_->numCols() != (numBlocks_) ) )
992  AU1TUDeltaL2_->reshape( recycleBlocks_, numBlocks_ );
993  }
994 
995  // Generate AU1TAU1_ only if it doesn't exist
996  if (AU1TAU1_ == Teuchos::null)
997  AU1TAU1_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, recycleBlocks_ ) );
998  else {
999  if ( (AU1TAU1_->numRows() != (recycleBlocks_)) || (AU1TAU1_->numCols() != (recycleBlocks_) ) )
1000  AU1TAU1_->reshape( recycleBlocks_, recycleBlocks_ );
1001  }
1002 
1003  // Generate GY_ only if it doesn't exist
1004  if (GY_ == Teuchos::null)
1005  GY_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_ + recycleBlocks_, recycleBlocks_ ) );
1006  else {
1007  if ( (GY_->numRows() != (numBlocks_ + recycleBlocks_)) || (GY_->numCols() != (recycleBlocks_) ) )
1008  GY_->reshape( numBlocks_+recycleBlocks_, recycleBlocks_ );
1009  }
1010 
1011  // Generate AU1TU1_ only if it doesn't exist
1012  if (AU1TU1_ == Teuchos::null)
1013  AU1TU1_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, recycleBlocks_ ) );
1014  else {
1015  if ( (AU1TU1_->numRows() != (recycleBlocks_)) || (AU1TU1_->numCols() != (recycleBlocks_) ) )
1016  AU1TU1_->reshape( recycleBlocks_, recycleBlocks_ );
1017  }
1018 
1019  // Generate FY_ only if it doesn't exist
1020  if (FY_ == Teuchos::null)
1021  FY_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_ + recycleBlocks_, recycleBlocks_ ) );
1022  else {
1023  if ( (FY_->numRows() != (numBlocks_ + recycleBlocks_)) || (FY_->numCols() != (recycleBlocks_) ) )
1024  FY_->reshape( numBlocks_+recycleBlocks_, recycleBlocks_ );
1025  }
1026 
1027  // Generate AU1TAP_ only if it doesn't exist
1028  if (AU1TAP_ == Teuchos::null)
1029  AU1TAP_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, numBlocks_ ) );
1030  else {
1031  if ( (AU1TAP_->numRows() != (recycleBlocks_)) || (AU1TAP_->numCols() != (numBlocks_) ) )
1032  AU1TAP_->reshape( recycleBlocks_, numBlocks_ );
1033  }
1034 
1035  // Generate APTAP_ only if it doesn't exist
1036  if (APTAP_ == Teuchos::null)
1037  APTAP_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_, numBlocks_ ) );
1038  else {
1039  if ( (APTAP_->numRows() != (numBlocks_)) || (APTAP_->numCols() != (numBlocks_) ) )
1040  APTAP_->reshape( numBlocks_, numBlocks_ );
1041  }
1042 
1043  // If the subspace has not been initialized before, generate it using the RHS from lp_.
1044  if (U1Y1_ == Teuchos::null) {
1045  U1Y1_ = MVT::Clone( *rhsMV, recycleBlocks_ );
1046  }
1047  else {
1048  // Generate U1Y1_ by cloning itself ONLY if more space is needed.
1049  if (MVT::GetNumberVecs(*U1Y1_) < recycleBlocks_) {
1050  Teuchos::RCP<const MV> tmp = U1Y1_;
1051  U1Y1_ = MVT::Clone( *tmp, recycleBlocks_ );
1052  }
1053  }
1054 
1055  // If the subspace has not been initialized before, generate it using the RHS from lp_.
1056  if (PY2_ == Teuchos::null) {
1057  PY2_ = MVT::Clone( *rhsMV, recycleBlocks_ );
1058  }
1059  else {
1060  // Generate PY2_ by cloning itself ONLY if more space is needed.
1061  if (MVT::GetNumberVecs(*PY2_) < recycleBlocks_) {
1062  Teuchos::RCP<const MV> tmp = PY2_;
1063  PY2_ = MVT::Clone( *tmp, recycleBlocks_ );
1064  }
1065  }
1066 
1067  // Generate AUTAP_ only if it doesn't exist
1068  if (AUTAP_ == Teuchos::null)
1069  AUTAP_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, numBlocks_ ) );
1070  else {
1071  if ( (AUTAP_->numRows() != (recycleBlocks_)) || (AUTAP_->numCols() != (numBlocks_) ) )
1072  AUTAP_->reshape( recycleBlocks_, numBlocks_ );
1073  }
1074 
1075  // Generate AU1TU_ only if it doesn't exist
1076  if (AU1TU_ == Teuchos::null)
1077  AU1TU_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycleBlocks_, recycleBlocks_ ) );
1078  else {
1079  if ( (AU1TU_->numRows() != (recycleBlocks_)) || (AU1TU_->numCols() != (recycleBlocks_) ) )
1080  AU1TU_->reshape( recycleBlocks_, recycleBlocks_ );
1081  }
1082 
1083 
1084  }
1085 }
1086 
1087 template<class ScalarType, class MV, class OP>
1089 
1091  std::vector<int> index(recycleBlocks_);
1092  ScalarType one = Teuchos::ScalarTraits<ScalarType>::one();
1093  ScalarType zero = Teuchos::ScalarTraits<ScalarType>::zero();
1094 
1095  // Count of number of cycles performed on current rhs
1096  int cycle = 0;
1097 
1098  // Set the current parameters if they were not set before.
1099  // NOTE: This may occur if the user generated the solver manager with the default constructor and
1100  // then didn't set any parameters using setParameters().
1101  if (!params_Set_) {
1102  setParameters(Teuchos::parameterList(*getValidParameters()));
1103  }
1104 
1106  "Belos::RCGSolMgr::solve(): Linear problem is not a valid object.");
1108  "Belos::RCGSolMgr::solve(): Linear problem is not ready, setProblem() has not been called.");
1109  TEUCHOS_TEST_FOR_EXCEPTION((problem_->getLeftPrec() != Teuchos::null)&&(problem_->getRightPrec() != Teuchos::null),
1111  "Belos::RCGSolMgr::solve(): RCG does not support split preconditioning, only set left or right preconditioner.");
1112 
1113  // Grab the preconditioning object
1114  Teuchos::RCP<OP> precObj;
1115  if (problem_->getLeftPrec() != Teuchos::null) {
1116  precObj = Teuchos::rcp_const_cast<OP>(problem_->getLeftPrec());
1117  }
1118  else if (problem_->getRightPrec() != Teuchos::null) {
1119  precObj = Teuchos::rcp_const_cast<OP>(problem_->getRightPrec());
1120  }
1121 
1122  // Create indices for the linear systems to be solved.
1123  int numRHS2Solve = MVT::GetNumberVecs( *(problem_->getRHS()) );
1124  std::vector<int> currIdx(1);
1125  currIdx[0] = 0;
1126 
1127  // Inform the linear problem of the current linear system to solve.
1128  problem_->setLSIndex( currIdx );
1129 
1130  // Check the number of blocks and change them if necessary.
1131  ptrdiff_t dim = MVT::GetGlobalLength( *(problem_->getRHS()) );
1132  if (numBlocks_ > dim) {
1133  numBlocks_ = Teuchos::asSafe<int>(dim);
1134  params_->set("Num Blocks", numBlocks_);
1135  printer_->stream(Warnings) <<
1136  "Warning! Requested Krylov subspace dimension is larger than operator dimension!" << std::endl <<
1137  " The maximum number of blocks allowed for the Krylov subspace will be adjusted to " << numBlocks_ << std::endl;
1138  }
1139 
1140  // Initialize storage for all state variables
1141  initializeStateStorage();
1142 
1143  // Parameter list
1144  Teuchos::ParameterList plist;
1145  plist.set("Num Blocks",numBlocks_);
1146  plist.set("Recycled Blocks",recycleBlocks_);
1147 
1148  // Reset the status test.
1149  outputTest_->reset();
1150 
1151  // Assume convergence is achieved, then let any failed convergence set this to false.
1152  bool isConverged = true;
1153 
1154  // Compute AU = A*U, UTAU = U'*AU, AUTAU = (AU)'*(AU)
1155  if (existU_) {
1156  index.resize(recycleBlocks_);
1157  for (int i=0; i<recycleBlocks_; ++i) { index[i] = i; }
1158  Teuchos::RCP<const MV> Utmp = MVT::CloneView( *U_, index );
1159  index.resize(recycleBlocks_);
1160  for (int i=0; i<recycleBlocks_; ++i) { index[i] = i; }
1161  Teuchos::RCP<MV> AUtmp = MVT::CloneViewNonConst( *AU_, index );
1162  // Initialize AU
1163  problem_->applyOp( *Utmp, *AUtmp );
1164  // Initialize UTAU
1165  Teuchos::SerialDenseMatrix<int,ScalarType> UTAUtmp( Teuchos::View, *UTAU_, recycleBlocks_, recycleBlocks_ );
1166  MVT::MvTransMv( one, *Utmp, *AUtmp, UTAUtmp );
1167  // Initialize AUTAU ( AUTAU = AU'*(M\AU) )
1168  Teuchos::SerialDenseMatrix<int,ScalarType> AUTAUtmp( Teuchos::View, *AUTAU_, recycleBlocks_, recycleBlocks_ );
1169  if ( precObj != Teuchos::null ) {
1170  index.resize(recycleBlocks_);
1171  for (int i=0; i<recycleBlocks_; ++i) { index[i] = i; }
1172  index.resize(recycleBlocks_);
1173  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1174  Teuchos::RCP<MV> PCAU = MVT::CloneViewNonConst( *U1_, index ); // use U1 as temp storage
1175  OPT::Apply( *precObj, *AUtmp, *PCAU );
1176  MVT::MvTransMv( one, *AUtmp, *PCAU, AUTAUtmp );
1177  } else {
1178  MVT::MvTransMv( one, *AUtmp, *AUtmp, AUTAUtmp );
1179  }
1180  }
1181 
1183  // RCG solver
1184 
1186  rcg_iter = Teuchos::rcp( new RCGIter<ScalarType,MV,OP>(problem_,printer_,outputTest_,plist) );
1187 
1188  // Enter solve() iterations
1189  {
1190 #ifdef BELOS_TEUCHOS_TIME_MONITOR
1191  Teuchos::TimeMonitor slvtimer(*timerSolve_);
1192 #endif
1193 
1194  while ( numRHS2Solve > 0 ) {
1195 
1196  // Debugging output to tell use if recycle space exists and will be used
1197  if (printer_->isVerbosity( Debug ) ) {
1198  if (existU_) printer_->print( Debug, "Using recycle space generated from previous call to solve()." );
1199  else printer_->print( Debug, "No recycle space exists." );
1200  }
1201 
1202  // Reset the number of iterations.
1203  rcg_iter->resetNumIters();
1204 
1205  // Set the current number of recycle blocks and subspace dimension with the RCG iteration.
1206  rcg_iter->setSize( recycleBlocks_, numBlocks_ );
1207 
1208  // Reset the number of calls that the status test output knows about.
1209  outputTest_->resetNumCalls();
1210 
1211  // indicate that updated recycle space has not yet been generated for this linear system
1212  existU1_ = false;
1213 
1214  // reset cycle count
1215  cycle = 0;
1216 
1217  // Get the current residual
1218  problem_->computeCurrResVec( &*r_ );
1219 
1220  // If U exists, find best soln over this space first
1221  if (existU_) {
1222  // Solve linear system UTAU * y = (U'*r)
1223  Teuchos::SerialDenseMatrix<int,ScalarType> Utr(recycleBlocks_,1);
1224  index.resize(recycleBlocks_);
1225  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1226  Teuchos::RCP<const MV> Utmp = MVT::CloneView( *U_, index );
1227  MVT::MvTransMv( one, *Utmp, *r_, Utr );
1228  Teuchos::SerialDenseMatrix<int,ScalarType> UTAUtmp( Teuchos::View, *UTAU_, recycleBlocks_, recycleBlocks_ );
1229  Teuchos::SerialDenseMatrix<int,ScalarType> LUUTAUtmp( Teuchos::View, *LUUTAU_, recycleBlocks_, recycleBlocks_ );
1230  LUUTAUtmp.assign(UTAUtmp);
1231  int info = 0;
1232  lapack.GESV(recycleBlocks_, 1, LUUTAUtmp.values(), LUUTAUtmp.stride(), &(*ipiv_)[0], Utr.values(), Utr.stride(), &info);
1234  "Belos::RCGSolMgr::solve(): LAPACK GESV failed to compute a solution.");
1235 
1236  // Update solution (x = x + U*y)
1237  MVT::MvTimesMatAddMv( one, *Utmp, Utr, one, *problem_->getCurrLHSVec() );
1238 
1239  // Update residual ( r = r - AU*y )
1240  index.resize(recycleBlocks_);
1241  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1242  Teuchos::RCP<const MV> AUtmp = MVT::CloneView( *AU_, index );
1243  MVT::MvTimesMatAddMv( -one, *AUtmp, Utr, one, *r_ );
1244  }
1245 
1246  if ( precObj != Teuchos::null ) {
1247  OPT::Apply( *precObj, *r_, *z_ );
1248  } else {
1249  z_ = r_;
1250  }
1251 
1252  // rTz_old = r'*z
1253  MVT::MvTransMv( one, *r_, *z_, *rTz_old_ );
1254 
1255  if ( existU_ ) {
1256  // mu = UTAU\(AU'*z);
1257  Teuchos::SerialDenseMatrix<int,ScalarType> mu( Teuchos::View, *Delta_, recycleBlocks_, 1);
1258  index.resize(recycleBlocks_);
1259  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1260  Teuchos::RCP<const MV> AUtmp = MVT::CloneView( *AU_, index );
1261  MVT::MvTransMv( one, *AUtmp, *z_, mu );
1262  Teuchos::SerialDenseMatrix<int,ScalarType> LUUTAUtmp( Teuchos::View, *LUUTAU_, recycleBlocks_, recycleBlocks_ );
1263  char TRANS = 'N';
1264  int info;
1265  lapack.GETRS( TRANS, recycleBlocks_, 1, LUUTAUtmp.values(), LUUTAUtmp.stride(), &(*ipiv_)[0], mu.values(), mu.stride(), &info );
1267  "Belos::RCGSolMgr::solve(): LAPACK GETRS failed to compute a solution.");
1268  // p = z - U*mu;
1269  index.resize( 1 );
1270  index[0] = 0;
1271  Teuchos::RCP<MV> Ptmp = MVT::CloneViewNonConst( *P_, index );
1272  MVT::MvAddMv(one,*z_,zero,*z_,*Ptmp);
1273  MVT::MvTimesMatAddMv( -one, *U_, mu, one, *Ptmp );
1274  } else {
1275  // p = z;
1276  index.resize( 1 );
1277  index[0] = 0;
1278  Teuchos::RCP<MV> Ptmp = MVT::CloneViewNonConst( *P_, index );
1279  MVT::MvAddMv(one,*z_,zero,*z_,*Ptmp);
1280  }
1281 
1282  // Set the new state and initialize the solver.
1283  RCGIterState<ScalarType,MV> newstate;
1284 
1285  // Create RCP views here
1286  index.resize( numBlocks_+1 );
1287  for (int ii=0; ii<(numBlocks_+1); ++ii) { index[ii] = ii; }
1288  newstate.P = MVT::CloneViewNonConst( *P_, index );
1289  index.resize( recycleBlocks_ );
1290  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1291  newstate.U = MVT::CloneViewNonConst( *U_, index );
1292  index.resize( recycleBlocks_ );
1293  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1294  newstate.AU = MVT::CloneViewNonConst( *AU_, index );
1295  newstate.Alpha = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *Alpha_, numBlocks_, 1 ) );
1296  newstate.Beta = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *Beta_, numBlocks_, 1 ) );
1297  newstate.D = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *D_, numBlocks_, 1 ) );
1298  newstate.Delta = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *Delta_, recycleBlocks_, numBlocks_, 0, 1 ) );
1299  newstate.LUUTAU = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *LUUTAU_, recycleBlocks_, recycleBlocks_ ) );
1300  // assign the rest of the values in the struct
1301  newstate.curDim = 1; // We have initialized the first search vector
1302  newstate.Ap = Ap_;
1303  newstate.r = r_;
1304  newstate.z = z_;
1305  newstate.existU = existU_;
1306  newstate.ipiv = ipiv_;
1307  newstate.rTz_old = rTz_old_;
1308 
1309  rcg_iter->initialize(newstate);
1310 
1311  while(1) {
1312 
1313  // tell rcg_iter to iterate
1314  try {
1315  rcg_iter->iterate();
1316 
1318  //
1319  // check convergence first
1320  //
1322  if ( convTest_->getStatus() == Passed ) {
1323  // We have convergence
1324  break; // break from while(1){rcg_iter->iterate()}
1325  }
1327  //
1328  // check for maximum iterations
1329  //
1331  else if ( maxIterTest_->getStatus() == Passed ) {
1332  // we don't have convergence
1333  isConverged = false;
1334  break; // break from while(1){rcg_iter->iterate()}
1335  }
1337  //
1338  // check if cycle complete; update for next cycle
1339  //
1341  else if ( rcg_iter->getCurSubspaceDim() == rcg_iter->getMaxSubspaceDim() ) {
1342  // index into P_ of last search vector generated this cycle
1343  int lastp = -1;
1344  // index into Beta_ of last entry generated this cycle
1345  int lastBeta = -1;
1346  if (recycleBlocks_ > 0) {
1347  if (!existU_) {
1348  if (cycle == 0) { // No U, no U1
1349 
1350  Teuchos::SerialDenseMatrix<int,ScalarType> Ftmp( Teuchos::View, *F_, numBlocks_, numBlocks_ );
1351  Teuchos::SerialDenseMatrix<int,ScalarType> Gtmp( Teuchos::View, *G_, numBlocks_, numBlocks_ );
1352  Teuchos::SerialDenseMatrix<int,ScalarType> Dtmp( Teuchos::View, *D_, numBlocks_, 1 );
1353  Teuchos::SerialDenseMatrix<int,ScalarType> Alphatmp( Teuchos::View, *Alpha_, numBlocks_, 1 );
1354  Teuchos::SerialDenseMatrix<int,ScalarType> Betatmp( Teuchos::View, *Beta_, numBlocks_, 1 );
1355  Ftmp.putScalar(zero);
1356  Gtmp.putScalar(zero);
1357  for (int ii=0;ii<numBlocks_;ii++) {
1358  Gtmp(ii,ii) = (Dtmp(ii,0) / Alphatmp(ii,0))*(1 + Betatmp(ii,0));
1359  if (ii > 0) {
1360  Gtmp(ii-1,ii) = -Dtmp(ii,0)/Alphatmp(ii-1,0);
1361  Gtmp(ii,ii-1) = -Dtmp(ii,0)/Alphatmp(ii-1,0);
1362  }
1363  Ftmp(ii,ii) = Dtmp(ii,0);
1364  }
1365 
1366  // compute harmonic Ritz vectors
1367  Teuchos::SerialDenseMatrix<int,ScalarType> Ytmp( Teuchos::View, *Y_, numBlocks_, recycleBlocks_ );
1368  getHarmonicVecs(Ftmp,Gtmp,Ytmp);
1369 
1370  // U1 = [P(:,1:end-1)*Y];
1371  index.resize( numBlocks_ );
1372  for (int ii=0; ii<numBlocks_; ++ii) { index[ii] = ii; }
1373  Teuchos::RCP<const MV> Ptmp = MVT::CloneView( *P_, index );
1374  index.resize( recycleBlocks_ );
1375  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1376  Teuchos::RCP<MV> U1tmp = MVT::CloneViewNonConst( *U1_, index );
1377  MVT::MvTimesMatAddMv( one, *Ptmp, Ytmp, zero, *U1tmp );
1378 
1379  // Precompute some variables for next cycle
1380 
1381  // AU1TAU1 = Y'*G*Y;
1382  Teuchos::SerialDenseMatrix<int,ScalarType> GYtmp( Teuchos::View, *GY_, numBlocks_, recycleBlocks_ );
1383  GYtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Gtmp,Ytmp,zero);
1384  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TAU1tmp( Teuchos::View, *AU1TAU1_, recycleBlocks_, recycleBlocks_ );
1385  AU1TAU1tmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Ytmp,GYtmp,zero);
1386 
1387  // AU1TU1 = Y'*F*Y;
1388  Teuchos::SerialDenseMatrix<int,ScalarType> FYtmp( Teuchos::View, *FY_, numBlocks_, recycleBlocks_ );
1389  FYtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Ftmp,Ytmp,zero);
1390  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TU1tmp( Teuchos::View, *AU1TU1_, recycleBlocks_, recycleBlocks_ );
1391  AU1TU1tmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Ytmp,FYtmp,zero);
1392 
1393  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TAPtmp( Teuchos::View, *AU1TAP_, recycleBlocks_, 1 );
1394  // Must reinitialize AU1TAP; can become dense later
1395  AU1TAPtmp.putScalar(zero);
1396  // AU1TAP(:,1) = Y(end,:)' * (-1/Alpha(end));
1397  ScalarType alphatmp = -1.0 / Alphatmp(numBlocks_-1,0);
1398  for (int ii=0; ii<recycleBlocks_; ++ii) {
1399  AU1TAPtmp(ii,0) = Ytmp(numBlocks_-1,ii) * alphatmp;
1400  }
1401 
1402  // indicate that updated recycle space now defined
1403  existU1_ = true;
1404 
1405  // Indicate the size of the P, Beta structures generated this cycle
1406  lastp = numBlocks_;
1407  lastBeta = numBlocks_-1;
1408 
1409  } // if (cycle == 0)
1410  else { // No U, but U1 guaranteed to exist now
1411 
1412  // Finish computation of subblocks
1413  // AU1TAP = AU1TAP * D(1);
1414  Teuchos::SerialDenseMatrix<int,ScalarType> Dtmp( Teuchos::View, *D_, numBlocks_, 1 );
1415  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TAPtmp( Teuchos::View, *AU1TAP_, recycleBlocks_, numBlocks_ );
1416  AU1TAPtmp.scale(Dtmp(0,0));
1417 
1418  Teuchos::SerialDenseMatrix<int,ScalarType> Alphatmp( Teuchos::View, *Alpha_, numBlocks_, 1 );
1419  Teuchos::SerialDenseMatrix<int,ScalarType> Betatmp( Teuchos::View, *Beta_, numBlocks_+1, 1 );
1420  Teuchos::SerialDenseMatrix<int,ScalarType> APTAPtmp( Teuchos::View, *APTAP_, numBlocks_, numBlocks_ );
1421  APTAPtmp.putScalar(zero);
1422  for (int ii=0; ii<numBlocks_; ii++) {
1423  APTAPtmp(ii,ii) = (Dtmp(ii,0) / Alphatmp(ii,0))*(1 + Betatmp(ii+1,0));
1424  if (ii > 0) {
1425  APTAPtmp(ii-1,ii) = -Dtmp(ii,0)/Alphatmp(ii-1,0);
1426  APTAPtmp(ii,ii-1) = -Dtmp(ii,0)/Alphatmp(ii-1,0);
1427  }
1428  }
1429 
1430  // F = [AU1TU1 zeros(k,m); zeros(m,k) diag(D)];
1431  Teuchos::SerialDenseMatrix<int,ScalarType> Ftmp( Teuchos::View, *F_, (numBlocks_+recycleBlocks_), (numBlocks_+recycleBlocks_) );
1432  Teuchos::SerialDenseMatrix<int,ScalarType> F11( Teuchos::View, *F_, recycleBlocks_, recycleBlocks_ );
1433  Teuchos::SerialDenseMatrix<int,ScalarType> F12( Teuchos::View, *F_, recycleBlocks_, numBlocks_, 0, recycleBlocks_ );
1434  Teuchos::SerialDenseMatrix<int,ScalarType> F21( Teuchos::View, *F_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1435  Teuchos::SerialDenseMatrix<int,ScalarType> F22( Teuchos::View, *F_, numBlocks_, numBlocks_, recycleBlocks_, recycleBlocks_ );
1436  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TU1tmp( Teuchos::View, *AU1TU1_, recycleBlocks_, recycleBlocks_ );
1437  F11.assign(AU1TU1tmp);
1438  F12.putScalar(zero);
1439  F21.putScalar(zero);
1440  F22.putScalar(zero);
1441  for(int ii=0;ii<numBlocks_;ii++) {
1442  F22(ii,ii) = Dtmp(ii,0);
1443  }
1444 
1445  // G = [AU1TAU1 AU1TAP; AU1TAP' APTAP];
1446  Teuchos::SerialDenseMatrix<int,ScalarType> Gtmp( Teuchos::View, *G_, (numBlocks_+recycleBlocks_), (numBlocks_+recycleBlocks_) );
1447  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TAU1tmp( Teuchos::View, *AU1TAU1_, recycleBlocks_, recycleBlocks_ );
1448  Teuchos::SerialDenseMatrix<int,ScalarType> G11( Teuchos::View, *G_, recycleBlocks_, recycleBlocks_ );
1449  Teuchos::SerialDenseMatrix<int,ScalarType> G12( Teuchos::View, *G_, recycleBlocks_, numBlocks_, 0, recycleBlocks_ );
1450  Teuchos::SerialDenseMatrix<int,ScalarType> G21( Teuchos::View, *G_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1451  Teuchos::SerialDenseMatrix<int,ScalarType> G22( Teuchos::View, *G_, numBlocks_, numBlocks_, recycleBlocks_, recycleBlocks_ );
1452  G11.assign(AU1TAU1tmp);
1453  G12.assign(AU1TAPtmp);
1454  // G21 = G12'; (no transpose operator exists for SerialDenseMatrix; Do copy manually)
1455  for (int ii=0;ii<recycleBlocks_;++ii)
1456  for (int jj=0;jj<numBlocks_;++jj)
1457  G21(jj,ii) = G12(ii,jj);
1458  G22.assign(APTAPtmp);
1459 
1460  // compute harmonic Ritz vectors
1461  Teuchos::SerialDenseMatrix<int,ScalarType> Ytmp( Teuchos::View, *Y_, (recycleBlocks_+numBlocks_), recycleBlocks_ );
1462  getHarmonicVecs(Ftmp,Gtmp,Ytmp);
1463 
1464  // U1 = [U1 P(:,2:end-1)]*Y;
1465  index.resize( numBlocks_ );
1466  for (int ii=0; ii<numBlocks_; ++ii) { index[ii] = ii+1; }
1467  Teuchos::RCP<const MV> Ptmp = MVT::CloneView( *P_, index );
1468  index.resize( recycleBlocks_ );
1469  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1470  Teuchos::RCP<MV> PY2tmp = MVT::CloneViewNonConst( *PY2_, index );
1471  Teuchos::SerialDenseMatrix<int,ScalarType> Y2( Teuchos::View, *Y_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1472  index.resize( recycleBlocks_ );
1473  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1474  Teuchos::RCP<MV> U1tmp = MVT::CloneViewNonConst( *U1_, index );
1475  index.resize( recycleBlocks_ );
1476  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1477  Teuchos::RCP<MV> U1Y1tmp = MVT::CloneViewNonConst( *U1Y1_, index );
1478  Teuchos::SerialDenseMatrix<int,ScalarType> Y1( Teuchos::View, *Y_, recycleBlocks_, recycleBlocks_ );
1479  MVT::MvTimesMatAddMv( one, *Ptmp, Y2, zero, *PY2tmp );
1480  MVT::MvTimesMatAddMv( one, *U1tmp, Y1, zero, *U1Y1tmp );
1481  MVT::MvAddMv(one,*U1Y1tmp, one, *PY2tmp, *U1tmp);
1482 
1483  // Precompute some variables for next cycle
1484 
1485  // AU1TAU1 = Y'*G*Y;
1486  Teuchos::SerialDenseMatrix<int,ScalarType> GYtmp( Teuchos::View, *GY_, (numBlocks_+recycleBlocks_), recycleBlocks_ );
1487  GYtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Gtmp,Ytmp,zero);
1488  AU1TAU1tmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Ytmp,GYtmp,zero);
1489 
1490  // AU1TAP = zeros(k,m);
1491  // AU1TAP(:,1) = Y(end,:)' * (-1/Alpha(end));
1492  AU1TAPtmp.putScalar(zero);
1493  ScalarType alphatmp = -1.0 / Alphatmp(numBlocks_-1,0);
1494  for (int ii=0; ii<recycleBlocks_; ++ii) {
1495  AU1TAPtmp(ii,0) = Ytmp(numBlocks_+recycleBlocks_-1,ii) * alphatmp;
1496  }
1497 
1498  // AU1TU1 = Y'*F*Y;
1499  Teuchos::SerialDenseMatrix<int,ScalarType> FYtmp( Teuchos::View, *FY_, (numBlocks_+recycleBlocks_), recycleBlocks_ );
1500  FYtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Ftmp,Ytmp,zero);
1501  //Teuchos::SerialDenseMatrix<int,ScalarType> AU1TU1tmp( Teuchos::View, *AU1TU1_, recycleBlocks_, recycleBlocks_ );
1502  AU1TU1tmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Ytmp,FYtmp,zero);
1503 
1504  // Indicate the size of the P, Beta structures generated this cycle
1505  lastp = numBlocks_+1;
1506  lastBeta = numBlocks_;
1507 
1508  } // if (cycle != 1)
1509  } // if (!existU_)
1510  else { // U exists
1511  if (cycle == 0) { // No U1, but U exists
1512  Teuchos::SerialDenseMatrix<int,ScalarType> Alphatmp( Teuchos::View, *Alpha_, numBlocks_, 1 );
1513  Teuchos::SerialDenseMatrix<int,ScalarType> Betatmp( Teuchos::View, *Beta_, numBlocks_, 1 );
1514  Teuchos::SerialDenseMatrix<int,ScalarType> Dtmp( Teuchos::View, *D_, numBlocks_, 1 );
1515  Teuchos::SerialDenseMatrix<int,ScalarType> APTAPtmp( Teuchos::View, *APTAP_, numBlocks_, numBlocks_ );
1516  APTAPtmp.putScalar(zero);
1517  for (int ii=0; ii<numBlocks_; ii++) {
1518  APTAPtmp(ii,ii) = (Dtmp(ii,0) / Alphatmp(ii,0))*(1 + Betatmp(ii,0));
1519  if (ii > 0) {
1520  APTAPtmp(ii-1,ii) = -Dtmp(ii,0)/Alphatmp(ii-1,0);
1521  APTAPtmp(ii,ii-1) = -Dtmp(ii,0)/Alphatmp(ii-1,0);
1522  }
1523  }
1524 
1525  Teuchos::SerialDenseMatrix<int,ScalarType> L2tmp( Teuchos::View, *L2_, numBlocks_+1, numBlocks_ );
1526  L2tmp.putScalar(zero);
1527  for(int ii=0;ii<numBlocks_;ii++) {
1528  L2tmp(ii,ii) = 1./Alphatmp(ii,0);
1529  L2tmp(ii+1,ii) = -1./Alphatmp(ii,0);
1530  }
1531 
1532  // AUTAP = UTAU*Delta*L2;
1533  Teuchos::SerialDenseMatrix<int,ScalarType> AUTAPtmp( Teuchos::View, *AUTAP_, recycleBlocks_, numBlocks_ );
1534  Teuchos::SerialDenseMatrix<int,ScalarType> UTAUtmp( Teuchos::View, *UTAU_, recycleBlocks_, recycleBlocks_ );
1535  Teuchos::SerialDenseMatrix<int,ScalarType> Deltatmp( Teuchos::View, *Delta_, recycleBlocks_, numBlocks_+1 );
1536  Teuchos::SerialDenseMatrix<int,ScalarType> DeltaL2tmp( Teuchos::View, *DeltaL2_, recycleBlocks_, numBlocks_ );
1537  DeltaL2tmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Deltatmp,L2tmp,zero);
1538  AUTAPtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,UTAUtmp,DeltaL2tmp,zero);
1539 
1540  // F = [UTAU zeros(k,m); zeros(m,k) diag(D)];
1541  Teuchos::SerialDenseMatrix<int,ScalarType> Ftmp( Teuchos::View, *F_, (numBlocks_+recycleBlocks_), (numBlocks_+recycleBlocks_) );
1542  Teuchos::SerialDenseMatrix<int,ScalarType> F11( Teuchos::View, *F_, recycleBlocks_, recycleBlocks_ );
1543  Teuchos::SerialDenseMatrix<int,ScalarType> F12( Teuchos::View, *F_, recycleBlocks_, numBlocks_, 0, recycleBlocks_ );
1544  Teuchos::SerialDenseMatrix<int,ScalarType> F21( Teuchos::View, *F_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1545  Teuchos::SerialDenseMatrix<int,ScalarType> F22( Teuchos::View, *F_, numBlocks_, numBlocks_, recycleBlocks_, recycleBlocks_ );
1546  F11.assign(UTAUtmp);
1547  F12.putScalar(zero);
1548  F21.putScalar(zero);
1549  F22.putScalar(zero);
1550  for(int ii=0;ii<numBlocks_;ii++) {
1551  F22(ii,ii) = Dtmp(ii,0);
1552  }
1553 
1554  // G = [AUTAU AUTAP; AUTAP' APTAP];
1555  Teuchos::SerialDenseMatrix<int,ScalarType> Gtmp( Teuchos::View, *G_, (numBlocks_+recycleBlocks_), (numBlocks_+recycleBlocks_) );
1556  Teuchos::SerialDenseMatrix<int,ScalarType> G11( Teuchos::View, *G_, recycleBlocks_, recycleBlocks_ );
1557  Teuchos::SerialDenseMatrix<int,ScalarType> G12( Teuchos::View, *G_, recycleBlocks_, numBlocks_, 0, recycleBlocks_ );
1558  Teuchos::SerialDenseMatrix<int,ScalarType> G21( Teuchos::View, *G_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1559  Teuchos::SerialDenseMatrix<int,ScalarType> G22( Teuchos::View, *G_, numBlocks_, numBlocks_, recycleBlocks_, recycleBlocks_ );
1560  Teuchos::SerialDenseMatrix<int,ScalarType> AUTAUtmp( Teuchos::View, *AUTAU_, recycleBlocks_, recycleBlocks_ );
1561  G11.assign(AUTAUtmp);
1562  G12.assign(AUTAPtmp);
1563  // G21 = G12'; (no transpose operator exists for SerialDenseMatrix; Do copy manually)
1564  for (int ii=0;ii<recycleBlocks_;++ii)
1565  for (int jj=0;jj<numBlocks_;++jj)
1566  G21(jj,ii) = G12(ii,jj);
1567  G22.assign(APTAPtmp);
1568 
1569  // compute harmonic Ritz vectors
1570  Teuchos::SerialDenseMatrix<int,ScalarType> Ytmp( Teuchos::View, *Y_, (recycleBlocks_+numBlocks_), recycleBlocks_ );
1571  getHarmonicVecs(Ftmp,Gtmp,Ytmp);
1572 
1573  // U1 = [U P(:,1:end-1)]*Y;
1574  index.resize( recycleBlocks_ );
1575  for (int ii=0; ii<(recycleBlocks_); ++ii) { index[ii] = ii; }
1576  Teuchos::RCP<const MV> Utmp = MVT::CloneView( *U_, index );
1577  index.resize( numBlocks_ );
1578  for (int ii=0; ii<numBlocks_; ++ii) { index[ii] = ii; }
1579  Teuchos::RCP<const MV> Ptmp = MVT::CloneView( *P_, index );
1580  index.resize( recycleBlocks_ );
1581  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1582  Teuchos::RCP<MV> PY2tmp = MVT::CloneViewNonConst( *PY2_, index );
1583  Teuchos::SerialDenseMatrix<int,ScalarType> Y2( Teuchos::View, *Y_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1584  index.resize( recycleBlocks_ );
1585  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1586  Teuchos::RCP<MV> UY1tmp = MVT::CloneViewNonConst( *U1Y1_, index );
1587  Teuchos::SerialDenseMatrix<int,ScalarType> Y1( Teuchos::View, *Y_, recycleBlocks_, recycleBlocks_ );
1588  index.resize( recycleBlocks_ );
1589  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1590  Teuchos::RCP<MV> U1tmp = MVT::CloneViewNonConst( *U1_, index );
1591  MVT::MvTimesMatAddMv( one, *Ptmp, Y2, zero, *PY2tmp );
1592  MVT::MvTimesMatAddMv( one, *Utmp, Y1, zero, *UY1tmp );
1593  MVT::MvAddMv(one,*UY1tmp, one, *PY2tmp, *U1tmp);
1594 
1595  // Precompute some variables for next cycle
1596 
1597  // AU1TAU1 = Y'*G*Y;
1598  Teuchos::SerialDenseMatrix<int,ScalarType> GYtmp( Teuchos::View, *GY_, (numBlocks_+recycleBlocks_), recycleBlocks_ );
1599  GYtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Gtmp,Ytmp,zero);
1600  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TAU1tmp( Teuchos::View, *AU1TAU1_, recycleBlocks_, recycleBlocks_ );
1601  AU1TAU1tmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Ytmp,GYtmp,zero);
1602 
1603  // AU1TU1 = Y'*F*Y;
1604  Teuchos::SerialDenseMatrix<int,ScalarType> FYtmp( Teuchos::View, *FY_, (numBlocks_+recycleBlocks_), recycleBlocks_ );
1605  FYtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Ftmp,Ytmp,zero);
1606  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TU1tmp( Teuchos::View, *AU1TU1_, recycleBlocks_, recycleBlocks_ );
1607  AU1TU1tmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Ytmp,FYtmp,zero);
1608 
1609  // AU1TU = UTAU;
1610  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TUtmp( Teuchos::View, *AU1TU_, recycleBlocks_, recycleBlocks_ );
1611  AU1TUtmp.assign(UTAUtmp);
1612 
1613  // dold = D(end);
1614  dold = Dtmp(numBlocks_-1,0);
1615 
1616  // indicate that updated recycle space now defined
1617  existU1_ = true;
1618 
1619  // Indicate the size of the P, Beta structures generated this cycle
1620  lastp = numBlocks_;
1621  lastBeta = numBlocks_-1;
1622  }
1623  else { // Have U and U1
1624  Teuchos::SerialDenseMatrix<int,ScalarType> Alphatmp( Teuchos::View, *Alpha_, numBlocks_, 1 );
1625  Teuchos::SerialDenseMatrix<int,ScalarType> Betatmp( Teuchos::View, *Beta_, numBlocks_+1, 1 );
1626  Teuchos::SerialDenseMatrix<int,ScalarType> Dtmp( Teuchos::View, *D_, numBlocks_, 1 );
1627  Teuchos::SerialDenseMatrix<int,ScalarType> APTAPtmp( Teuchos::View, *APTAP_, numBlocks_, numBlocks_ );
1628  for (int ii=0; ii<numBlocks_; ii++) {
1629  APTAPtmp(ii,ii) = (Dtmp(ii,0) / Alphatmp(ii,0))*(1 + Betatmp(ii+1,0));
1630  if (ii > 0) {
1631  APTAPtmp(ii-1,ii) = -Dtmp(ii,0)/Alphatmp(ii-1,0);
1632  APTAPtmp(ii,ii-1) = -Dtmp(ii,0)/Alphatmp(ii-1,0);
1633  }
1634  }
1635 
1636  Teuchos::SerialDenseMatrix<int,ScalarType> L2tmp( Teuchos::View, *L2_, numBlocks_+1, numBlocks_ );
1637  for(int ii=0;ii<numBlocks_;ii++) {
1638  L2tmp(ii,ii) = 1./Alphatmp(ii,0);
1639  L2tmp(ii+1,ii) = -1./Alphatmp(ii,0);
1640  }
1641 
1642  // M(end,1) = dold*(-Beta(1)/Alpha(1));
1643  // AU1TAP = Y'*[AU1TU*Delta*L2; M];
1644  Teuchos::SerialDenseMatrix<int,ScalarType> DeltaL2( Teuchos::View, *DeltaL2_, recycleBlocks_, numBlocks_ );
1645  Teuchos::SerialDenseMatrix<int,ScalarType> Deltatmp( Teuchos::View, *Delta_, recycleBlocks_, numBlocks_+1 );
1646  DeltaL2.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Deltatmp,L2tmp,zero);
1647  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TUDeltaL2( Teuchos::View, *AU1TUDeltaL2_, recycleBlocks_, numBlocks_ );
1648  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TUtmp( Teuchos::View, *AU1TU_, recycleBlocks_, recycleBlocks_ );
1649  AU1TUDeltaL2.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,AU1TUtmp,DeltaL2,zero);
1650  Teuchos::SerialDenseMatrix<int,ScalarType> Y1( Teuchos::View, *Y_, recycleBlocks_, recycleBlocks_ );
1651  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TAPtmp( Teuchos::View, *AU1TAP_, recycleBlocks_, numBlocks_ );
1652  AU1TAPtmp.putScalar(zero);
1653  AU1TAPtmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Y1,AU1TUDeltaL2,zero);
1654  Teuchos::SerialDenseMatrix<int,ScalarType> Y2( Teuchos::View, *Y_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1655  ScalarType val = dold * (-Betatmp(0,0)/Alphatmp(0,0));
1656  for(int ii=0;ii<recycleBlocks_;ii++) {
1657  AU1TAPtmp(ii,0) += Y2(numBlocks_-1,ii)*val;
1658  }
1659 
1660  // AU1TU = Y1'*AU1TU
1661  Teuchos::SerialDenseMatrix<int,ScalarType> Y1TAU1TU( Teuchos::View, *GY_, recycleBlocks_, recycleBlocks_ );
1662  Y1TAU1TU.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Y1,AU1TUtmp,zero);
1663  AU1TUtmp.assign(Y1TAU1TU);
1664 
1665  // F = [AU1TU1 zeros(k,m); zeros(m,k) diag(D)];
1666  Teuchos::SerialDenseMatrix<int,ScalarType> Ftmp( Teuchos::View, *F_, (numBlocks_+recycleBlocks_), (numBlocks_+recycleBlocks_) );
1667  Teuchos::SerialDenseMatrix<int,ScalarType> F11( Teuchos::View, *F_, recycleBlocks_, recycleBlocks_ );
1668  Teuchos::SerialDenseMatrix<int,ScalarType> F12( Teuchos::View, *F_, recycleBlocks_, numBlocks_, 0, recycleBlocks_ );
1669  Teuchos::SerialDenseMatrix<int,ScalarType> F21( Teuchos::View, *F_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1670  Teuchos::SerialDenseMatrix<int,ScalarType> F22( Teuchos::View, *F_, numBlocks_, numBlocks_, recycleBlocks_, recycleBlocks_ );
1671  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TU1tmp( Teuchos::View, *AU1TU1_, recycleBlocks_, recycleBlocks_ );
1672  F11.assign(AU1TU1tmp);
1673  for(int ii=0;ii<numBlocks_;ii++) {
1674  F22(ii,ii) = Dtmp(ii,0);
1675  }
1676 
1677  // G = [AU1TAU1 AU1TAP; AU1TAP' APTAP];
1678  Teuchos::SerialDenseMatrix<int,ScalarType> Gtmp( Teuchos::View, *G_, (numBlocks_+recycleBlocks_), (numBlocks_+recycleBlocks_) );
1679  Teuchos::SerialDenseMatrix<int,ScalarType> G11( Teuchos::View, *G_, recycleBlocks_, recycleBlocks_ );
1680  Teuchos::SerialDenseMatrix<int,ScalarType> G12( Teuchos::View, *G_, recycleBlocks_, numBlocks_, 0, recycleBlocks_ );
1681  Teuchos::SerialDenseMatrix<int,ScalarType> G21( Teuchos::View, *G_, numBlocks_, recycleBlocks_, recycleBlocks_, 0 );
1682  Teuchos::SerialDenseMatrix<int,ScalarType> G22( Teuchos::View, *G_, numBlocks_, numBlocks_, recycleBlocks_, recycleBlocks_ );
1683  Teuchos::SerialDenseMatrix<int,ScalarType> AU1TAU1tmp( Teuchos::View, *AU1TAU1_, recycleBlocks_, recycleBlocks_ );
1684  G11.assign(AU1TAU1tmp);
1685  G12.assign(AU1TAPtmp);
1686  // G21 = G12'; (no transpose operator exists for SerialDenseMatrix; Do copy manually)
1687  for (int ii=0;ii<recycleBlocks_;++ii)
1688  for (int jj=0;jj<numBlocks_;++jj)
1689  G21(jj,ii) = G12(ii,jj);
1690  G22.assign(APTAPtmp);
1691 
1692  // compute harmonic Ritz vectors
1693  Teuchos::SerialDenseMatrix<int,ScalarType> Ytmp( Teuchos::View, *Y_, (recycleBlocks_+numBlocks_), recycleBlocks_ );
1694  getHarmonicVecs(Ftmp,Gtmp,Ytmp);
1695 
1696  // U1 = [U1 P(:,2:end-1)]*Y;
1697  index.resize( numBlocks_ );
1698  for (int ii=0; ii<numBlocks_; ++ii) { index[ii] = ii+1; }
1699  Teuchos::RCP<const MV> Ptmp = MVT::CloneView( *P_, index );
1700  index.resize( recycleBlocks_ );
1701  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1702  Teuchos::RCP<MV> PY2tmp = MVT::CloneViewNonConst( *PY2_, index );
1703  index.resize( recycleBlocks_ );
1704  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1705  Teuchos::RCP<MV> U1tmp = MVT::CloneViewNonConst( *U1_, index );
1706  index.resize( recycleBlocks_ );
1707  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1708  Teuchos::RCP<MV> U1Y1tmp = MVT::CloneViewNonConst( *U1Y1_, index );
1709  MVT::MvTimesMatAddMv( one, *Ptmp, Y2, zero, *PY2tmp );
1710  MVT::MvTimesMatAddMv( one, *U1tmp, Y1, zero, *U1Y1tmp );
1711  MVT::MvAddMv(one,*U1Y1tmp, one, *PY2tmp, *U1tmp);
1712 
1713  // Precompute some variables for next cycle
1714 
1715  // AU1TAU1 = Y'*G*Y;
1716  Teuchos::SerialDenseMatrix<int,ScalarType> GYtmp( Teuchos::View, *GY_, (numBlocks_+recycleBlocks_), recycleBlocks_ );
1717  GYtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Gtmp,Ytmp,zero);
1718  AU1TAU1tmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Ytmp,GYtmp,zero);
1719 
1720  // AU1TU1 = Y'*F*Y;
1721  Teuchos::SerialDenseMatrix<int,ScalarType> FYtmp( Teuchos::View, *FY_, (numBlocks_+recycleBlocks_), recycleBlocks_ );
1722  FYtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,Ftmp,Ytmp,zero);
1723  AU1TU1tmp.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,Ytmp,FYtmp,zero);
1724 
1725  // dold = D(end);
1726  dold = Dtmp(numBlocks_-1,0);
1727 
1728  // Indicate the size of the P, Beta structures generated this cycle
1729  lastp = numBlocks_+1;
1730  lastBeta = numBlocks_;
1731 
1732  }
1733  }
1734  } // if (recycleBlocks_ > 0)
1735 
1736  // Cleanup after end of cycle
1737 
1738  // P = P(:,end-1:end);
1739  index.resize( 2 );
1740  index[0] = lastp-1; index[1] = lastp;
1741  Teuchos::RCP<const MV> Ptmp2 = MVT::CloneView( *P_, index );
1742  index[0] = 0; index[1] = 1;
1743  MVT::SetBlock(*Ptmp2,index,*P_);
1744 
1745  // Beta = Beta(end);
1746  (*Beta_)(0,0) = (*Beta_)(lastBeta,0);
1747 
1748  // Delta = Delta(:,end);
1749  if (existU_) { // Delta only initialized if U exists
1750  Teuchos::SerialDenseMatrix<int,ScalarType> mu1( Teuchos::View, *Delta_, recycleBlocks_, 1, 0, 0 );
1751  Teuchos::SerialDenseMatrix<int,ScalarType> mu2( Teuchos::View, *Delta_, recycleBlocks_, 1, 0, numBlocks_ );
1752  mu1.assign(mu2);
1753  }
1754 
1755  // Now reinitialize state variables for next cycle
1756  newstate.P = Teuchos::null;
1757  index.resize( numBlocks_+1 );
1758  for (int ii=0; ii<(numBlocks_+1); ++ii) { index[ii] = ii+1; }
1759  newstate.P = MVT::CloneViewNonConst( *P_, index );
1760 
1761  newstate.Beta = Teuchos::null;
1762  newstate.Beta = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *Beta_, numBlocks_, 1, 1, 0 ) );
1763 
1764  newstate.Delta = Teuchos::null;
1765  newstate.Delta = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *Delta_, recycleBlocks_, numBlocks_, 0, 1 ) );
1766 
1767  newstate.curDim = 1; // We have initialized the first search vector
1768 
1769  // Pass to iteration object
1770  rcg_iter->initialize(newstate);
1771 
1772  // increment cycle count
1773  cycle = cycle + 1;
1774 
1775  }
1777  //
1778  // we returned from iterate(), but none of our status tests Passed.
1779  // something is wrong, and it is probably our fault.
1780  //
1782  else {
1783  TEUCHOS_TEST_FOR_EXCEPTION(true,std::logic_error,
1784  "Belos::RCGSolMgr::solve(): Invalid return from RCGIter::iterate().");
1785  }
1786  }
1787  catch (const std::exception &e) {
1788  printer_->stream(Errors) << "Error! Caught std::exception in RCGIter::iterate() at iteration "
1789  << rcg_iter->getNumIters() << std::endl
1790  << e.what() << std::endl;
1791  throw;
1792  }
1793  }
1794 
1795  // Inform the linear problem that we are finished with this block linear system.
1796  problem_->setCurrLS();
1797 
1798  // Update indices for the linear systems to be solved.
1799  numRHS2Solve -= 1;
1800  if ( numRHS2Solve > 0 ) {
1801  currIdx[0]++;
1802  // Set the next indices.
1803  problem_->setLSIndex( currIdx );
1804  }
1805  else {
1806  currIdx.resize( numRHS2Solve );
1807  }
1808 
1809  // Update the recycle space for the next linear system
1810  if (existU1_) { // be sure updated recycle space was created
1811  // U = U1
1812  index.resize(recycleBlocks_);
1813  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1814  MVT::SetBlock(*U1_,index,*U_);
1815  // Set flag indicating recycle space is now defined
1816  existU_ = true;
1817  if (numRHS2Solve > 0) { // also update AU, UTAU, and AUTAU
1818  // Free pointers in newstate
1819  newstate.P = Teuchos::null;
1820  newstate.Ap = Teuchos::null;
1821  newstate.r = Teuchos::null;
1822  newstate.z = Teuchos::null;
1823  newstate.U = Teuchos::null;
1824  newstate.AU = Teuchos::null;
1825  newstate.Alpha = Teuchos::null;
1826  newstate.Beta = Teuchos::null;
1827  newstate.D = Teuchos::null;
1828  newstate.Delta = Teuchos::null;
1829  newstate.LUUTAU = Teuchos::null;
1830  newstate.ipiv = Teuchos::null;
1831  newstate.rTz_old = Teuchos::null;
1832 
1833  // Reinitialize AU, UTAU, AUTAU
1834  index.resize(recycleBlocks_);
1835  for (int i=0; i<recycleBlocks_; ++i) { index[i] = i; }
1836  Teuchos::RCP<const MV> Utmp = MVT::CloneView( *U_, index );
1837  index.resize(recycleBlocks_);
1838  for (int i=0; i<recycleBlocks_; ++i) { index[i] = i; }
1839  Teuchos::RCP<MV> AUtmp = MVT::CloneViewNonConst( *AU_, index );
1840  // Initialize AU
1841  problem_->applyOp( *Utmp, *AUtmp );
1842  // Initialize UTAU
1843  Teuchos::SerialDenseMatrix<int,ScalarType> UTAUtmp( Teuchos::View, *UTAU_, recycleBlocks_, recycleBlocks_ );
1844  MVT::MvTransMv( one, *Utmp, *AUtmp, UTAUtmp );
1845  // Initialize AUTAU ( AUTAU = AU'*(M\AU) )
1846  Teuchos::SerialDenseMatrix<int,ScalarType> AUTAUtmp( Teuchos::View, *AUTAU_, recycleBlocks_, recycleBlocks_ );
1847  if ( precObj != Teuchos::null ) {
1848  index.resize(recycleBlocks_);
1849  for (int i=0; i<recycleBlocks_; ++i) { index[i] = i; }
1850  index.resize(recycleBlocks_);
1851  for (int ii=0; ii<recycleBlocks_; ++ii) { index[ii] = ii; }
1852  Teuchos::RCP<MV> LeftPCAU = MVT::CloneViewNonConst( *U1_, index ); // use U1 as temp storage
1853  OPT::Apply( *precObj, *AUtmp, *LeftPCAU );
1854  MVT::MvTransMv( one, *AUtmp, *LeftPCAU, AUTAUtmp );
1855  } else {
1856  MVT::MvTransMv( one, *AUtmp, *AUtmp, AUTAUtmp );
1857  }
1858  } // if (numRHS2Solve > 0)
1859 
1860  } // if (existU1)
1861  }// while ( numRHS2Solve > 0 )
1862 
1863  }
1864 
1865  // print final summary
1866  sTest_->print( printer_->stream(FinalSummary) );
1867 
1868  // print timing information
1869 #ifdef BELOS_TEUCHOS_TIME_MONITOR
1870  // Calling summarize() can be expensive, so don't call unless the
1871  // user wants to print out timing details. summarize() will do all
1872  // the work even if it's passed a "black hole" output stream.
1873  if (verbosity_ & TimingDetails)
1874  Teuchos::TimeMonitor::summarize( printer_->stream(TimingDetails) );
1875 #endif
1876 
1877  // get iteration information for this solve
1878  numIters_ = maxIterTest_->getNumIters();
1879 
1880  // Save the convergence test value ("achieved tolerance") for this solve.
1881  {
1882  using Teuchos::rcp_dynamic_cast;
1883  typedef StatusTestGenResNorm<ScalarType,MV,OP> conv_test_type;
1884  // testValues is nonnull and not persistent.
1885  const std::vector<MagnitudeType>* pTestValues =
1886  rcp_dynamic_cast<conv_test_type>(convTest_)->getTestValue();
1887 
1888  TEUCHOS_TEST_FOR_EXCEPTION(pTestValues == NULL, std::logic_error,
1889  "Belos::RCGSolMgr::solve(): The convergence test's getTestValue() "
1890  "method returned NULL. Please report this bug to the Belos developers.");
1891 
1892  TEUCHOS_TEST_FOR_EXCEPTION(pTestValues->size() < 1, std::logic_error,
1893  "Belos::RCGSolMgr::solve(): The convergence test's getTestValue() "
1894  "method returned a vector of length zero. Please report this bug to the "
1895  "Belos developers.");
1896 
1897  // FIXME (mfh 12 Dec 2011) Does pTestValues really contain the
1898  // achieved tolerances for all vectors in the current solve(), or
1899  // just for the vectors from the last deflation?
1900  achievedTol_ = *std::max_element (pTestValues->begin(), pTestValues->end());
1901  }
1902 
1903  if (!isConverged) {
1904  return Unconverged; // return from RCGSolMgr::solve()
1905  }
1906  return Converged; // return from RCGSolMgr::solve()
1907 }
1908 
1909 // Compute the harmonic eigenpairs of the projected, dense system.
1910 template<class ScalarType, class MV, class OP>
1914  // order of F,G
1915  int n = F.numCols();
1916 
1917  // The LAPACK interface
1919 
1920  // Magnitude of harmonic Ritz values
1921  std::vector<MagnitudeType> w(n);
1922 
1923  // Sorted order of harmonic Ritz values
1924  std::vector<int> iperm(n);
1925 
1926  // Compute k smallest harmonic Ritz pairs
1927  // SUBROUTINE DSYGV( ITYPE, JOBZ, UPLO, N, A, LDA, B, LDB, W, WORK, LWORK, INFO )
1928  int itype = 1; // solve A*x = (lambda)*B*x
1929  char jobz='V'; // compute eigenvalues and eigenvectors
1930  char uplo='U'; // since F,G symmetric, reference only their upper triangular data
1931  std::vector<ScalarType> work(1);
1932  int lwork = -1;
1933  int info = 0;
1934  // since SYGV destroys workspace, create copies of F,G
1937 
1938  // query for optimal workspace size
1939  lapack.SYGV(itype, jobz, uplo, n, G2.values(), G2.stride(), F2.values(), F2.stride(), &w[0], &work[0], lwork, &info);
1941  "Belos::RCGSolMgr::solve(): LAPACK SYGV failed to query optimal work size.");
1942  lwork = (int)work[0];
1943  work.resize(lwork);
1944  lapack.SYGV(itype, jobz, uplo, n, G2.values(), G2.stride(), F2.values(), F2.stride(), &w[0], &work[0], lwork, &info);
1946  "Belos::RCGSolMgr::solve(): LAPACK SYGV failed to compute eigensolutions.");
1947 
1948 
1949  // Construct magnitude of each harmonic Ritz value
1950  this->sort(w,n,iperm);
1951 
1952  // Select recycledBlocks_ smallest eigenvectors
1953  for( int i=0; i<recycleBlocks_; i++ ) {
1954  for( int j=0; j<n; j++ ) {
1955  Y(j,i) = G2(j,iperm[i]);
1956  }
1957  }
1958 
1959 }
1960 
1961 // This method sorts list of n floating-point numbers and return permutation vector
1962 template<class ScalarType, class MV, class OP>
1963 void RCGSolMgr<ScalarType,MV,OP,true>::sort(std::vector<ScalarType>& dlist, int n, std::vector<int>& iperm)
1964 {
1965  int l, r, j, i, flag;
1966  int RR2;
1967  double dRR, dK;
1968 
1969  // Initialize the permutation vector.
1970  for(j=0;j<n;j++)
1971  iperm[j] = j;
1972 
1973  if (n <= 1) return;
1974 
1975  l = n / 2 + 1;
1976  r = n - 1;
1977  l = l - 1;
1978  dRR = dlist[l - 1];
1979  dK = dlist[l - 1];
1980 
1981  RR2 = iperm[l - 1];
1982  while (r != 0) {
1983  j = l;
1984  flag = 1;
1985 
1986  while (flag == 1) {
1987  i = j;
1988  j = j + j;
1989 
1990  if (j > r + 1)
1991  flag = 0;
1992  else {
1993  if (j < r + 1)
1994  if (dlist[j] > dlist[j - 1]) j = j + 1;
1995 
1996  if (dlist[j - 1] > dK) {
1997  dlist[i - 1] = dlist[j - 1];
1998  iperm[i - 1] = iperm[j - 1];
1999  }
2000  else {
2001  flag = 0;
2002  }
2003  }
2004  }
2005  dlist[i - 1] = dRR;
2006  iperm[i - 1] = RR2;
2007  if (l == 1) {
2008  dRR = dlist [r];
2009  RR2 = iperm[r];
2010  dK = dlist[r];
2011  dlist[r] = dlist[0];
2012  iperm[r] = iperm[0];
2013  r = r - 1;
2014  }
2015  else {
2016  l = l - 1;
2017  dRR = dlist[l - 1];
2018  RR2 = iperm[l - 1];
2019  dK = dlist[l - 1];
2020  }
2021  }
2022  dlist[0] = dRR;
2023  iperm[0] = RR2;
2024 }
2025 
2026 // This method requires the solver manager to return a std::string that describes itself.
2027 template<class ScalarType, class MV, class OP>
2029 {
2030  std::ostringstream oss;
2031  oss << "Belos::RCGSolMgr<...,"<<Teuchos::ScalarTraits<ScalarType>::name()<<">";
2032  return oss.str();
2033 }
2034 
2035 } // end Belos namespace
2036 
2037 #endif /* BELOS_RCG_SOLMGR_HPP */
Teuchos::RCP< MV > r
The current residual.
Collection of types and exceptions used within the Belos solvers.
Teuchos::RCP< std::ostream > outputStream_
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > rTz_old
void GESV(const OrdinalType n, const OrdinalType nrhs, ScalarType *A, const OrdinalType lda, OrdinalType *IPIV, ScalarType *B, const OrdinalType ldb, OrdinalType *info) const
Belos&#39;s basic output manager for sending information of select verbosity levels to the appropriate ou...
Teuchos::RCP< MV > P
The current Krylov basis.
Teuchos::ScalarTraits< MagnitudeType > MT
Teuchos::RCP< std::vector< int > > ipiv
Data from LU factorization of U^T A U.
Class which manages the output and verbosity of the Belos solvers.
void SYGV(const OrdinalType itype, const char JOBZ, const char UPLO, const OrdinalType n, ScalarType *A, const OrdinalType lda, ScalarType *B, const OrdinalType ldb, ScalarType *W, ScalarType *WORK, const OrdinalType lwork, OrdinalType *info) const
void setProblem(const Teuchos::RCP< LinearProblem< ScalarType, MV, OP > > &problem)
Set the linear problem that needs to be solved.
Teuchos::RCP< Teuchos::Time > timerSolve_
ScalarType * values() const
bool is_null(const boost::shared_ptr< T > &p)
RCGSolMgrRecyclingFailure is thrown when any problem occurs in using/creating the recycling subspace...
static const bool scalarTypeIsSupported
Teuchos::RCP< StatusTest< ScalarType, MV, OP > > sTest_
MagnitudeType achievedTol_
Tolerance achieved by the last solve() invocation.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > D_
Teuchos::RCP< MV > AU
Teuchos::RCP< StatusTestOutput< ScalarType, MV, OP > > outputTest_
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > AUTAU_
T & get(ParameterList &l, const std::string &name)
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > GY_
ParameterList & set(std::string const &name, T const &value, std::string const &docString="", RCP< const ParameterEntryValidator > const &validator=null)
static RCP< Time > getNewCounter(const std::string &name)
bool is_null(const std::shared_ptr< T > &p)
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > L2_
int multiply(ETransp transa, ETransp transb, ScalarType alpha, const SerialDenseMatrix< OrdinalType, ScalarType > &A, const SerialDenseMatrix< OrdinalType, ScalarType > &B, ScalarType beta)
Teuchos::RCP< MV > U
The recycled subspace and its image.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > LUUTAU
The LU factorization of the matrix U^T A U.
Base class for Belos::SolverManager subclasses which normally can only compile with real ScalarType t...
#define TEUCHOS_TEST_FOR_EXCEPTION(throw_exception_test, Exception, msg)
A factory class for generating StatusTestOutput objects.
Implementation of the RCG (Recycling Conjugate Gradient) iterative linear solver. ...
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > Beta
This class implements the RCG iteration, where a single-std::vector Krylov subspace is constructed...
int numIters_
Number of iterations taken by the last solve() invocation.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > AUTAP_
Teuchos::ScalarTraits< ScalarType >::magnitudeType MagnitudeType
An implementation of StatusTestResNorm using a family of residual norms.
int scale(const ScalarType alpha)
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > UTAU_
int getNumIters() const
Get the iteration count for the most recent call to solve().
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > rTz_old_
Belos::StatusTest class for specifying a maximum number of iterations.
Teuchos::RCP< Teuchos::ParameterList > params_
static std::string name()
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > AU1TU1_
A factory class for generating StatusTestOutput objects.
Iterated Modified Gram-Schmidt (IMGS) implementation of the Belos::OrthoManager class.
RCGSolMgrLinearProblemFailure(const std::string &what_arg)
RCGSolMgrLinearProblemFailure is thrown when the linear problem is not setup (i.e.
MagnitudeType convtol_
Convergence tolerance (read from parameter list).
Traits class which defines basic operations on multivectors.
Teuchos::RCP< LinearProblem< ScalarType, MV, OP > > problem_
Belos::StatusTest for logically combining several status tests.
void validateParameters(ParameterList const &validParamList, int const depth=1000, EValidateUsed const validateUsed=VALIDATE_USED_ENABLED, EValidateDefaults const validateDefaults=VALIDATE_DEFAULTS_ENABLED) const
Structure to contain pointers to RCGIter state variables.
Belos concrete class for performing the RCG iteration.
MultiVecTraits< ScalarType, MV > MVT
Classical Gram-Schmidt (with DGKS correction) implementation of the Belos::OrthoManager class...
int maxIters_
Maximum iteration count (read from parameter list).
Teuchos::Array< Teuchos::RCP< Teuchos::Time > > getTimers() const
Return the timers for this object.
A Belos::StatusTest class for specifying a maximum number of iterations.
ResetType
How to reset the solver.
Definition: BelosTypes.hpp:205
bool existU
Flag to indicate the recycle space should be used.
TEUCHOS_DEPRECATED RCP< T > rcp(T *p, Dealloc_T dealloc, bool owns_mem)
Pure virtual base class which describes the basic interface for a solver manager. ...
Teuchos::RCP< MV > z
The current preconditioned residual.
Teuchos::RCP< MV > Ap
A times current search vector.
void GETRS(const char TRANS, const OrdinalType n, const OrdinalType nrhs, const ScalarType *A, const OrdinalType lda, const OrdinalType *IPIV, ScalarType *B, const OrdinalType ldb, OrdinalType *info) const
static void summarize(Ptr< const Comm< int > > comm, std::ostream &out=std::cout, const bool alwaysWriteLocal=false, const bool writeGlobalStats=true, const bool writeZeroTimers=true, const ECounterSetOp setOp=Intersection, const std::string &filter="", const bool ignoreZeroTimers=false)
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > Alpha_
int putScalar(const ScalarType value=Teuchos::ScalarTraits< ScalarType >::zero())
static const Teuchos::RCP< std::ostream > outputStream_default_
MagnitudeType achievedTol() const
Tolerance achieved by the last solve() invocation.
A linear system to solve, and its associated information.
Class which describes the linear problem to be solved by the iterative solver.
Type traits class that says whether Teuchos::LAPACK has a valid implementation for the given ScalarTy...
OperatorTraits< ScalarType, MV, OP > OPT
ReturnType
Whether the Belos solve converged for all linear systems.
Definition: BelosTypes.hpp:154
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > APTAP_
Teuchos::RCP< std::vector< int > > ipiv_
bool isLOADetected() const
Return whether a loss of accuracy was detected by this solver during the most current solve...
Iterated Classical Gram-Schmidt (ICGS) implementation of the Belos::OrthoManager class.
Teuchos::RCP< StatusTestGenResNorm< ScalarType, MV, OP > > convTest_
Teuchos::RCP< OutputManager< ScalarType > > printer_
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > Alpha
Coefficients arising in RCG iteration.
Teuchos::RCP< StatusTestOutput< ScalarType, MV, OP > > create(const Teuchos::RCP< OutputManager< ScalarType > > &printer, Teuchos::RCP< StatusTest< ScalarType, MV, OP > > test, int mod, int printStates)
Create the StatusTestOutput object specified by the outputStyle.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > Beta_
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > Y_
Belos::StatusTestResNorm for specifying general residual norm stopping criteria.
Teuchos::RCP< StatusTestMaxIters< ScalarType, MV, OP > > maxIterTest_
int curDim
The current dimension of the reduction.
RCGSolMgrLAPACKFailure(const std::string &what_arg)
void reset(const ResetType type)
Performs a reset of the solver manager specified by the ResetType. This informs the solver manager th...
RCGSolMgr(const Teuchos::RCP< LinearProblem< ScalarType, MV, OP > > &problem, const Teuchos::RCP< Teuchos::ParameterList > &pl)
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > LUUTAU_
A class for extending the status testing capabilities of Belos via logical combinations.
bool isParameter(const std::string &name) const
Teuchos::RCP< const Teuchos::ParameterList > getCurrentParameters() const
Get a parameter list containing the current parameters for this object.
Details::SolverManagerRequiresRealLapack< ScalarType, MV, OP, scalarTypeIsSupported > base_type
Class which defines basic traits for the operator type.
OrdinalType stride() const
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > D
Teuchos::ScalarTraits< ScalarType > SCT
Parent class to all Belos exceptions.
Definition: BelosTypes.hpp:60
const LinearProblem< ScalarType, MV, OP > & getProblem() const
Return a reference to the linear problem being solved by this solver manager.
OrdinalType numCols() const
RCGSolMgrRecyclingFailure(const std::string &what_arg)
Belos header file which uses auto-configuration information to include necessary C++ headers...
RCGSolMgrLAPACKFailure is thrown when a nonzero value is retuned from an LAPACK call.
SerialDenseMatrix< OrdinalType, ScalarType > & assign(const SerialDenseMatrix< OrdinalType, ScalarType > &Source)
int n
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > Delta
Solutions to local least-squares problems.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > Delta_