Belos Package Browser (Single Doxygen Collection)  Development
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
BelosGCRODRSolMgr.hpp
Go to the documentation of this file.
1 // @HEADER
2 // *****************************************************************************
3 // Belos: Block Linear Solvers Package
4 //
5 // Copyright 2004 NTESS and the Belos contributors.
6 // SPDX-License-Identifier: BSD-3-Clause
7 // *****************************************************************************
8 // @HEADER
9 
10 #ifndef BELOS_GCRODR_SOLMGR_HPP
11 #define BELOS_GCRODR_SOLMGR_HPP
12 
16 
17 #include "BelosConfigDefs.hpp"
19 #include "BelosSolverManager.hpp"
20 #include "BelosLinearProblem.hpp"
21 #include "BelosTypes.hpp"
22 
23 #include "BelosGCRODRIter.hpp"
24 #include "BelosBlockFGmresIter.hpp"
27 #include "BelosStatusTestCombo.hpp"
29 #include "BelosOutputManager.hpp"
30 #include "Teuchos_BLAS.hpp" // includes Teuchos_ConfigDefs.hpp
31 #include "Teuchos_LAPACK.hpp"
32 #include "Teuchos_as.hpp"
33 #ifdef BELOS_TEUCHOS_TIME_MONITOR
34 # include "Teuchos_TimeMonitor.hpp"
35 #endif // BELOS_TEUCHOS_TIME_MONITOR
36 #if defined(HAVE_TEUCHOSCORE_CXX11)
37 # include <type_traits>
38 #endif // defined(HAVE_TEUCHOSCORE_CXX11)
39 
50 namespace Belos {
51 
53 
54 
62  public:
63  GCRODRSolMgrLinearProblemFailure(const std::string& what_arg) : BelosError(what_arg) {}
64  };
65 
73  public:
74  GCRODRSolMgrOrthoFailure(const std::string& what_arg) : BelosError(what_arg) {}
75  };
76 
84  public:
85  GCRODRSolMgrLAPACKFailure(const std::string& what_arg) : BelosError(what_arg) {}
86  };
87 
95  public:
96  GCRODRSolMgrRecyclingFailure(const std::string& what_arg) : BelosError(what_arg) {}
97  };
98 
100 
125  template<class ScalarType, class MV, class OP,
126  const bool lapackSupportsScalarType =
128  class GCRODRSolMgr :
129  public Details::SolverManagerRequiresLapack<ScalarType,MV,OP>
130  {
131  static const bool requiresLapack =
133  typedef Details::SolverManagerRequiresLapack<ScalarType, MV, OP,
135 
136  public:
138  base_type ()
139  {}
142  base_type ()
143  {}
144  virtual ~GCRODRSolMgr () {}
145  };
146 
150  template<class ScalarType, class MV, class OP>
151  class GCRODRSolMgr<ScalarType, MV, OP, true> :
152  public Details::SolverManagerRequiresLapack<ScalarType, MV, OP, true>
153  {
154 
155 #if defined(HAVE_TEUCHOSCORE_CXX11)
156 # if defined(HAVE_TEUCHOS_COMPLEX)
157  #if defined(HAVE_TEUCHOS_LONG_DOUBLE)
158  static_assert (std::is_same<ScalarType, std::complex<float> >::value ||
159  std::is_same<ScalarType, std::complex<double> >::value ||
160  std::is_same<ScalarType, Kokkos::complex<double> >::value ||
161  std::is_same<ScalarType, float>::value ||
162  std::is_same<ScalarType, double>::value ||
163  std::is_same<ScalarType, long double>::value,
164  "Belos::GCRODRSolMgr: ScalarType must be one of the four "
165  "types (S,D,C,Z) supported by LAPACK or long double (largely not impl'd).");
166  #else
167  static_assert (std::is_same<ScalarType, std::complex<float> >::value ||
168  std::is_same<ScalarType, std::complex<double> >::value ||
169  std::is_same<ScalarType, Kokkos::complex<double> >::value ||
170  std::is_same<ScalarType, float>::value ||
171  std::is_same<ScalarType, double>::value,
172  "Belos::GCRODRSolMgr: ScalarType must be one of the four "
173  "types (S,D,C,Z) supported by LAPACK.");
174  #endif
175 # else
176  #if defined(HAVE_TEUCHOS_LONG_DOUBLE)
177  static_assert (std::is_same<ScalarType, float>::value ||
178  std::is_same<ScalarType, double>::value ||
179  std::is_same<ScalarType, long double>::value,
180  "Belos::GCRODRSolMgr: ScalarType must be float, double or long double. "
181  "Complex arithmetic support is currently disabled. To "
182  "enable it, set Teuchos_ENABLE_COMPLEX=ON.");
183  #else
184  static_assert (std::is_same<ScalarType, float>::value ||
185  std::is_same<ScalarType, double>::value,
186  "Belos::GCRODRSolMgr: ScalarType must be float or double. "
187  "Complex arithmetic support is currently disabled. To "
188  "enable it, set Teuchos_ENABLE_COMPLEX=ON.");
189  #endif
190 # endif // defined(HAVE_TEUCHOS_COMPLEX)
191 #endif // defined(HAVE_TEUCHOSCORE_CXX11)
192 
193  private:
200 
201  public:
203 
204 
210  GCRODRSolMgr();
211 
266 
268  virtual ~GCRODRSolMgr() {};
269 
273  }
275 
277 
278 
281  const LinearProblem<ScalarType,MV,OP>& getProblem() const override {
282  return *problem_;
283  }
284 
287  Teuchos::RCP<const Teuchos::ParameterList> getValidParameters() const override;
288 
292  return params_;
293  }
294 
301  return Teuchos::tuple(timerSolve_);
302  }
303 
309  MagnitudeType achievedTol() const override {
310  return achievedTol_;
311  }
312 
314  int getNumIters() const override {
315  return numIters_;
316  }
317 
320  bool isLOADetected() const override { return false; }
321 
323 
325 
326 
328  void setProblem( const Teuchos::RCP<LinearProblem<ScalarType,MV,OP> > &problem ) override {
329  problem_ = problem;
330  }
331 
333  void setParameters( const Teuchos::RCP<Teuchos::ParameterList> &params ) override;
334 
336 
338 
339 
343  void reset( const ResetType type ) override {
344  if ((type & Belos::Problem) && !Teuchos::is_null(problem_)) {
345  bool set = problem_->setProblem();
346  if (!set)
347  throw "Could not set problem.";
348  }
349  else if (type & Belos::RecycleSubspace) {
350  keff = 0;
351  }
352  }
354 
356 
357 
384  ReturnType solve() override;
385 
387 
389 
391  std::string description() const override;
392 
394 
395  private:
396 
397  // Called by all constructors; Contains init instructions common to all constructors
398  void init();
399 
400  // Initialize solver state storage
401  void initializeStateStorage();
402 
403  // Compute updated recycle space given existing recycle space and newly generated Krylov space
404  void buildRecycleSpace2(Teuchos::RCP<GCRODRIter<ScalarType,MV,OP> > gcrodr_iter);
405 
406  // Computes harmonic eigenpairs of projected matrix created during the priming solve.
407  // HH is the projected problem from the initial cycle of Gmres, it is (at least) of dimension m+1 x m.
408  // PP contains the harmonic eigenvectors corresponding to the recycledBlocks eigenvalues of smallest magnitude.
409  // The return value is the number of vectors needed to be stored, recycledBlocks or recycledBlocks+1.
410  int getHarmonicVecs1(int m,
413 
414  // Computes harmonic eigenpairs of projected matrix created during one cycle.
415  // HH is the total block projected problem from the GCRO-DR algorithm, it is (at least) of dimension keff+m+1 x keff+m.
416  // VV is the Krylov vectors from the projected GMRES algorithm, which has (at least) m+1 vectors.
417  // PP contains the harmonic eigenvectors corresponding to the recycledBlocks eigenvalues of smallest magnitude.
418  // The return value is the number of vectors needed to be stored, recycledBlocks or recycledBlocks+1.
419  int getHarmonicVecs2(int keff, int m,
421  const Teuchos::RCP<const MV>& VV,
423 
424  // Sort list of n floating-point numbers and return permutation vector
425  void sort(std::vector<MagnitudeType>& dlist, int n, std::vector<int>& iperm);
426 
427  // Lapack interface
429 
430  // Linear problem.
432 
433  // Output manager.
436 
437  // Status test.
443 
448 
449  // Current parameter list.
451 
452  // Default solver values.
453  static constexpr double orthoKappa_default_ = 0.0;
454  static constexpr int maxRestarts_default_ = 100;
455  static constexpr int maxIters_default_ = 1000;
456  static constexpr int numBlocks_default_ = 50;
457  static constexpr int blockSize_default_ = 1;
458  static constexpr int recycledBlocks_default_ = 5;
459  static constexpr int verbosity_default_ = Belos::Errors;
460  static constexpr int outputStyle_default_ = Belos::General;
461  static constexpr int outputFreq_default_ = -1;
462  static constexpr const char * impResScale_default_ = "Norm of Preconditioned Initial Residual";
463  static constexpr const char * expResScale_default_ = "Norm of Initial Residual";
464  static constexpr const char * label_default_ = "Belos";
465  static constexpr const char * orthoType_default_ = "ICGS";
466 
467  // Current solver values.
468  MagnitudeType convTol_, orthoKappa_, achievedTol_;
469  int maxRestarts_, maxIters_, numIters_;
470  int verbosity_, outputStyle_, outputFreq_;
471  std::string orthoType_;
472  std::string impResScale_, expResScale_;
473 
475  // Solver State Storage
477  //
478  // The number of blocks and recycle blocks (m and k, respectively)
479  int numBlocks_, recycledBlocks_;
480  // Current size of recycled subspace
481  int keff;
482  //
483  // Residual vector
485  //
486  // Search space
488  //
489  // Recycled subspace and its image
491  //
492  // Updated recycle space and its image
494  //
495  // Storage used in constructing harmonic Ritz values/vectors
501  std::vector<ScalarType> tau_;
502  std::vector<ScalarType> work_;
504  std::vector<int> ipiv_;
506 
507  // Timers.
508  std::string label_;
510 
511  // Internal state variables.
512  bool isSet_;
513 
514  // Have we generated or regenerated a recycle space yet this solve?
516  };
517 
518 
519 // Empty Constructor
520 template<class ScalarType, class MV, class OP>
522  achievedTol_(0.0),
523  numIters_(0)
524 {
525  init ();
526 }
527 
528 
529 // Basic Constructor
530 template<class ScalarType, class MV, class OP>
534  achievedTol_(0.0),
535  numIters_(0)
536 {
537  // Initialize local pointers to null, and initialize local variables
538  // to default values.
539  init ();
540 
542  problem == Teuchos::null, std::invalid_argument,
543  "Belos::GCRODRSolMgr constructor: The solver manager's "
544  "constructor needs the linear problem argument 'problem' "
545  "to be non-null.");
546  problem_ = problem;
547 
548  // Set the parameters using the list that was passed in. If null,
549  // we defer initialization until a non-null list is set (by the
550  // client calling setParameters(), or by calling solve() -- in
551  // either case, a null parameter list indicates that default
552  // parameters should be used).
553  if (! pl.is_null ()) {
554  setParameters (pl);
555  }
556 }
557 
558 // Common instructions executed in all constructors
559 template<class ScalarType, class MV, class OP>
561  outputStream_ = Teuchos::rcpFromRef(std::cout);
563  orthoKappa_ = orthoKappa_default_;
564  maxRestarts_ = maxRestarts_default_;
565  maxIters_ = maxIters_default_;
566  numBlocks_ = numBlocks_default_;
567  recycledBlocks_ = recycledBlocks_default_;
568  verbosity_ = verbosity_default_;
569  outputStyle_ = outputStyle_default_;
570  outputFreq_ = outputFreq_default_;
571  orthoType_ = orthoType_default_;
572  impResScale_ = impResScale_default_;
573  expResScale_ = expResScale_default_;
574  label_ = label_default_;
575  isSet_ = false;
576  builtRecycleSpace_ = false;
577  keff = 0;
578  r_ = Teuchos::null;
579  V_ = Teuchos::null;
580  U_ = Teuchos::null;
581  C_ = Teuchos::null;
582  U1_ = Teuchos::null;
583  C1_ = Teuchos::null;
584  PP_ = Teuchos::null;
585  HP_ = Teuchos::null;
586  H2_ = Teuchos::null;
587  R_ = Teuchos::null;
588  H_ = Teuchos::null;
589  B_ = Teuchos::null;
590 }
591 
592 template<class ScalarType, class MV, class OP>
593 void
596 {
597  using Teuchos::isParameterType;
598  using Teuchos::getParameter;
599  using Teuchos::null;
601  using Teuchos::parameterList;
602  using Teuchos::RCP;
603  using Teuchos::rcp;
604  using Teuchos::rcp_dynamic_cast;
605  using Teuchos::rcpFromRef;
609 
610  // The default parameter list contains all parameters that
611  // GCRODRSolMgr understands, and none that it doesn't understand.
612  RCP<const ParameterList> defaultParams = getValidParameters();
613 
614  // Create the internal parameter list if one doesn't already exist.
615  //
616  // (mfh 28 Feb 2011, 10 Mar 2011) At the time this code was written,
617  // ParameterList did not have validators or validateParameters().
618  // This is why the code below carefully validates the parameters one
619  // by one and fills in defaults. This code could be made a lot
620  // shorter by using validators. To do so, we would have to define
621  // appropriate validators for all the parameters. (This would more
622  // or less just move all that validation code out of this routine
623  // into to getValidParameters().)
624  //
625  // For an analogous reason, GCRODRSolMgr defines default parameter
626  // values as class data, as well as in the default ParameterList.
627  // This redundancy could be removed by defining the default
628  // parameter values only in the default ParameterList (which
629  // documents each parameter as well -- handy!).
630  if (params_.is_null()) {
631  params_ = parameterList (*defaultParams);
632  } else {
633  // A common case for setParameters() is for it to be called at the
634  // beginning of the solve() routine. This follows the Belos
635  // pattern of delaying initialization until the last possible
636  // moment (when the user asks Belos to perform the solve). In
637  // this common case, we save ourselves a deep copy of the input
638  // parameter list.
639  if (params_ != params) {
640  // Make a deep copy of the input parameter list. This allows
641  // the caller to modify or change params later, without
642  // affecting the behavior of this solver. This solver will then
643  // only change its internal parameters if setParameters() is
644  // called again.
645  params_ = parameterList (*params);
646  }
647 
648  // Fill in any missing parameters and their default values. Also,
649  // throw an exception if the parameter list has any misspelled or
650  // "extra" parameters. If you don't like this behavior, you'll
651  // want to replace the line of code below with your desired
652  // validation scheme. Note that Teuchos currently only implements
653  // two options:
654  //
655  // 1. validateParameters() requires that params_ has all the
656  // parameters that the default list has, and none that it
657  // doesn't have.
658  //
659  // 2. validateParametersAndSetDefaults() fills in missing
660  // parameters in params_ using the default list, but requires
661  // that any parameter provided in params_ is also in the
662  // default list.
663  //
664  // Here is an easy way to ignore any "extra" or misspelled
665  // parameters: Make a deep copy of the default list, fill in any
666  // "missing" parameters from the _input_ list, and then validate
667  // the input list using the deep copy of the default list. We
668  // show this in code:
669  //
670  // RCP<ParameterList> defaultCopy = parameterList (*getValidParameters ());
671  // defaultCopy->validateParametersAndSetDefaults (params);
672  // params->validateParametersAndSetDefaults (defaultCopy);
673  //
674  // This method is not entirely robust, because the input list may
675  // have incorrect validators set for existing parameters in the
676  // default list. This would then cause "validation" of the
677  // default list to throw an exception. As a result, we've chosen
678  // for now to be intolerant of misspellings and "extra" parameters
679  // in the input list.
680  params_->validateParametersAndSetDefaults (*defaultParams);
681  }
682 
683  // Check for maximum number of restarts.
684  if (params->isParameter ("Maximum Restarts")) {
685  maxRestarts_ = params->get("Maximum Restarts", maxRestarts_default_);
686 
687  // Update parameter in our list.
688  params_->set ("Maximum Restarts", maxRestarts_);
689  }
690 
691  // Check for maximum number of iterations
692  if (params->isParameter ("Maximum Iterations")) {
693  maxIters_ = params->get ("Maximum Iterations", maxIters_default_);
694 
695  // Update parameter in our list and in status test.
696  params_->set ("Maximum Iterations", maxIters_);
697  if (! maxIterTest_.is_null())
698  maxIterTest_->setMaxIters (maxIters_);
699  }
700 
701  // Check for the maximum number of blocks.
702  if (params->isParameter ("Num Blocks")) {
703  numBlocks_ = params->get ("Num Blocks", numBlocks_default_);
704  TEUCHOS_TEST_FOR_EXCEPTION(numBlocks_ <= 0, std::invalid_argument,
705  "Belos::GCRODRSolMgr: The \"Num Blocks\" parameter must "
706  "be strictly positive, but you specified a value of "
707  << numBlocks_ << ".");
708  // Update parameter in our list.
709  params_->set ("Num Blocks", numBlocks_);
710  }
711 
712  // Check for the maximum number of blocks.
713  if (params->isParameter ("Num Recycled Blocks")) {
714  recycledBlocks_ = params->get ("Num Recycled Blocks",
715  recycledBlocks_default_);
716  TEUCHOS_TEST_FOR_EXCEPTION(recycledBlocks_ <= 0, std::invalid_argument,
717  "Belos::GCRODRSolMgr: The \"Num Recycled Blocks\" "
718  "parameter must be strictly positive, but you specified "
719  "a value of " << recycledBlocks_ << ".");
720  TEUCHOS_TEST_FOR_EXCEPTION(recycledBlocks_ >= numBlocks_, std::invalid_argument,
721  "Belos::GCRODRSolMgr: The \"Num Recycled Blocks\" "
722  "parameter must be less than the \"Num Blocks\" "
723  "parameter, but you specified \"Num Recycled Blocks\" "
724  "= " << recycledBlocks_ << " and \"Num Blocks\" = "
725  << numBlocks_ << ".");
726  // Update parameter in our list.
727  params_->set("Num Recycled Blocks", recycledBlocks_);
728  }
729 
730  // Check to see if the timer label changed. If it did, update it in
731  // the parameter list, and create a new timer with that label (if
732  // Belos was compiled with timers enabled).
733  if (params->isParameter ("Timer Label")) {
734  std::string tempLabel = params->get ("Timer Label", label_default_);
735 
736  // Update parameter in our list and solver timer
737  if (tempLabel != label_) {
738  label_ = tempLabel;
739  params_->set ("Timer Label", label_);
740  std::string solveLabel = label_ + ": GCRODRSolMgr total solve time";
741 #ifdef BELOS_TEUCHOS_TIME_MONITOR
742  timerSolve_ = Teuchos::TimeMonitor::getNewCounter (solveLabel);
743 #endif
744  if (ortho_ != Teuchos::null) {
745  ortho_->setLabel( label_ );
746  }
747  }
748  }
749 
750  // Check for a change in verbosity level
751  if (params->isParameter ("Verbosity")) {
752  if (isParameterType<int> (*params, "Verbosity")) {
753  verbosity_ = params->get ("Verbosity", verbosity_default_);
754  } else {
755  verbosity_ = (int) getParameter<Belos::MsgType> (*params, "Verbosity");
756  }
757  // Update parameter in our list.
758  params_->set ("Verbosity", verbosity_);
759  // If the output manager (printer_) is null, then we will
760  // instantiate it later with the correct verbosity.
761  if (! printer_.is_null())
762  printer_->setVerbosity (verbosity_);
763  }
764 
765  // Check for a change in output style
766  if (params->isParameter ("Output Style")) {
767  if (isParameterType<int> (*params, "Output Style")) {
768  outputStyle_ = params->get ("Output Style", outputStyle_default_);
769  } else {
770  outputStyle_ = (int) getParameter<OutputType> (*params, "Output Style");
771  }
772 
773  // Update parameter in our list.
774  params_->set ("Output Style", outputStyle_);
775  // We will (re)instantiate the output status test afresh below.
776  outputTest_ = null;
777  }
778 
779  // Get the output stream for the output manager.
780  //
781  // While storing the output stream in the parameter list (either as
782  // an RCP or as a nonconst reference) is convenient and safe for
783  // programming, it makes it impossible to serialize the parameter
784  // list, read it back in from the serialized representation, and get
785  // the same output stream as before. This is because output streams
786  // may be arbitrary constructed objects.
787  //
788  // In case the user tried reading in the parameter list from a
789  // serialized representation and the output stream can't be read
790  // back in, we set the output stream to point to std::cout. This
791  // ensures reasonable behavior.
792  if (params->isParameter ("Output Stream")) {
793  try {
794  outputStream_ = getParameter<RCP<std::ostream> > (*params, "Output Stream");
795  } catch (InvalidParameter&) {
796  outputStream_ = rcpFromRef (std::cout);
797  }
798  // We assume that a null output stream indicates that the user
799  // doesn't want to print anything, so we replace it with a "black
800  // hole" stream that prints nothing sent to it. (We can't use a
801  // null output stream, since the output manager always sends
802  // things it wants to print to the output stream.)
803  if (outputStream_.is_null()) {
804  outputStream_ = rcp (new Teuchos::oblackholestream);
805  }
806  // Update parameter in our list.
807  params_->set ("Output Stream", outputStream_);
808  // If the output manager (printer_) is null, then we will
809  // instantiate it later with the correct output stream.
810  if (! printer_.is_null()) {
811  printer_->setOStream (outputStream_);
812  }
813  }
814 
815  // frequency level
816  if (verbosity_ & Belos::StatusTestDetails) {
817  if (params->isParameter ("Output Frequency")) {
818  outputFreq_ = params->get ("Output Frequency", outputFreq_default_);
819  }
820 
821  // Update parameter in out list and output status test.
822  params_->set("Output Frequency", outputFreq_);
823  if (! outputTest_.is_null())
824  outputTest_->setOutputFrequency (outputFreq_);
825  }
826 
827  // Create output manager if we need to, using the verbosity level
828  // and output stream that we fetched above. We do this here because
829  // instantiating an OrthoManager using OrthoManagerFactory requires
830  // a valid OutputManager.
831  if (printer_.is_null()) {
832  printer_ = rcp (new OutputManager<ScalarType> (verbosity_, outputStream_));
833  }
834 
835  // Get the orthogonalization manager name ("Orthogonalization").
836  //
837  // Getting default values for the orthogonalization manager
838  // parameters ("Orthogonalization Parameters") requires knowing the
839  // orthogonalization manager name. Save it for later, and also
840  // record whether it's different than before.
842  bool changedOrthoType = false;
843  if (params->isParameter ("Orthogonalization")) {
844  const std::string& tempOrthoType =
845  params->get ("Orthogonalization", orthoType_default_);
846  // Ensure that the specified orthogonalization type is valid.
847  if (! factory.isValidName (tempOrthoType)) {
848  std::ostringstream os;
849  os << "Belos::GCRODRSolMgr: Invalid orthogonalization name \""
850  << tempOrthoType << "\". The following are valid options "
851  << "for the \"Orthogonalization\" name parameter: ";
852  factory.printValidNames (os);
853  throw std::invalid_argument (os.str());
854  }
855  if (tempOrthoType != orthoType_) {
856  changedOrthoType = true;
857  orthoType_ = tempOrthoType;
858  // Update parameter in our list.
859  params_->set ("Orthogonalization", orthoType_);
860  }
861  }
862 
863  // Get any parameters for the orthogonalization ("Orthogonalization
864  // Parameters"). If not supplied, the orthogonalization manager
865  // factory will supply default values.
866  //
867  // NOTE (mfh 12 Jan 2011) For the sake of backwards compatibility,
868  // if params has an "Orthogonalization Constant" parameter and the
869  // DGKS orthogonalization manager is to be used, the value of this
870  // parameter will override DGKS's "depTol" parameter.
871  //
872  // Users must supply the orthogonalization manager parameters as a
873  // sublist (supplying it as an RCP<ParameterList> would make the
874  // resulting parameter list not serializable).
875  RCP<ParameterList> orthoParams;
876  { // The nonmember function sublist() returns an RCP<ParameterList>,
877  // which is what we want here.
878  using Teuchos::sublist;
879  // Abbreviation to avoid typos.
880  const std::string paramName ("Orthogonalization Parameters");
881 
882  try {
883  orthoParams = sublist (params_, paramName, true);
884  } catch (InvalidParameter&) {
885  // We didn't get the parameter list from params, so get a
886  // default parameter list from the OrthoManagerFactory. Modify
887  // params_ so that it has the default parameter list, and set
888  // orthoParams to ensure it's a sublist of params_ (and not just
889  // a copy of one).
890  params_->set (paramName, factory.getDefaultParameters (orthoType_));
891  orthoParams = sublist (params_, paramName, true);
892  }
893  }
894  TEUCHOS_TEST_FOR_EXCEPTION(orthoParams.is_null(), std::logic_error,
895  "Failed to get orthogonalization parameters. "
896  "Please report this bug to the Belos developers.");
897 
898  // Instantiate a new MatOrthoManager subclass instance if necessary.
899  // If not necessary, then tell the existing instance about the new
900  // parameters.
901  if (ortho_.is_null() || changedOrthoType) {
902  // We definitely need to make a new MatOrthoManager, since either
903  // we haven't made one yet, or we've changed orthogonalization
904  // methods. Creating the orthogonalization manager requires that
905  // the OutputManager (printer_) already be initialized.
906  ortho_ = factory.makeMatOrthoManager (orthoType_, null, printer_,
907  label_, orthoParams);
908  } else {
909  // If the MatOrthoManager implements the ParameterListAcceptor
910  // mix-in interface, we can propagate changes to its parameters
911  // without reinstantiating the MatOrthoManager.
912  //
913  // We recommend that all MatOrthoManager subclasses implement
914  // Teuchos::ParameterListAcceptor, but do not (yet) require this.
915  typedef Teuchos::ParameterListAcceptor PLA;
916  RCP<PLA> pla = rcp_dynamic_cast<PLA> (ortho_);
917  if (pla.is_null()) {
918  // Oops, it's not a ParameterListAcceptor. We have to
919  // reinstantiate the MatOrthoManager in order to pass in the
920  // possibly new parameters.
921  ortho_ = factory.makeMatOrthoManager (orthoType_, null, printer_,
922  label_, orthoParams);
923  } else {
924  pla->setParameterList (orthoParams);
925  }
926  }
927 
928  // The DGKS orthogonalization accepts a "Orthogonalization Constant"
929  // parameter (also called kappa in the code, but not in the
930  // parameter list). If its value is provided in the given parameter
931  // list, and its value is positive, use it. Ignore negative values.
932  //
933  // NOTE (mfh 12 Jan 2011) This overrides the "depTol" parameter that
934  // may have been specified in "Orthogonalization Parameters". We
935  // retain this behavior for backwards compatibility.
936  if (params->isParameter ("Orthogonalization Constant")) {
937  MagnitudeType orthoKappa = orthoKappa_default_;
938  if (params->isType<MagnitudeType> ("Orthogonalization Constant")) {
939  orthoKappa = params->get ("Orthogonalization Constant", orthoKappa);
940  }
941  else {
942  orthoKappa = params->get ("Orthogonalization Constant", orthoKappa_default_);
943  }
944 
945  if (orthoKappa > 0) {
946  orthoKappa_ = orthoKappa;
947  // Update parameter in our list.
948  params_->set("Orthogonalization Constant", orthoKappa_);
949  // Only DGKS currently accepts this parameter.
950  if (orthoType_ == "DGKS" && ! ortho_.is_null()) {
951  typedef DGKSOrthoManager<ScalarType, MV, OP> ortho_man_type;
952  // This cast should always succeed; it's a bug
953  // otherwise. (If the cast fails, then orthoType_
954  // doesn't correspond to the OrthoManager subclass
955  // instance that we think we have, so we initialized the
956  // wrong subclass somehow.)
957  rcp_dynamic_cast<ortho_man_type>(ortho_)->setDepTol (orthoKappa_);
958  }
959  }
960  }
961 
962  // Convergence
963  typedef Belos::StatusTestCombo<ScalarType,MV,OP> StatusTestCombo_t;
964  typedef Belos::StatusTestGenResNorm<ScalarType,MV,OP> StatusTestResNorm_t;
965 
966  // Check for convergence tolerance
967  if (params->isParameter("Convergence Tolerance")) {
968  if (params->isType<MagnitudeType> ("Convergence Tolerance")) {
969  convTol_ = params->get ("Convergence Tolerance",
970  static_cast<MagnitudeType> (DefaultSolverParameters::convTol));
971  }
972  else {
973  convTol_ = params->get ("Convergence Tolerance", DefaultSolverParameters::convTol);
974  }
975 
976  // Update parameter in our list and residual tests.
977  params_->set ("Convergence Tolerance", convTol_);
978  if (! impConvTest_.is_null())
979  impConvTest_->setTolerance (convTol_);
980  if (! expConvTest_.is_null())
981  expConvTest_->setTolerance (convTol_);
982  }
983 
984  // Check for a change in scaling, if so we need to build new residual tests.
985  if (params->isParameter ("Implicit Residual Scaling")) {
986  std::string tempImpResScale =
987  getParameter<std::string> (*params, "Implicit Residual Scaling");
988 
989  // Only update the scaling if it's different.
990  if (impResScale_ != tempImpResScale) {
991  ScaleType impResScaleType = convertStringToScaleType (tempImpResScale);
992  impResScale_ = tempImpResScale;
993 
994  // Update parameter in our list and residual tests
995  params_->set("Implicit Residual Scaling", impResScale_);
996  // NOTE (mfh 28 Feb 2011) StatusTestImpResNorm only lets you
997  // call defineScaleForm() once. The code below attempts to call
998  // defineScaleForm(); if the scale form has already been
999  // defined, it constructs a new StatusTestImpResNorm instance.
1000  // StatusTestImpResNorm really should not expose the
1001  // defineScaleForm() method, since it's serving an
1002  // initialization purpose; all initialization should happen in
1003  // the constructor whenever possible. In that case, the code
1004  // below could be simplified into a single (re)instantiation.
1005  if (! impConvTest_.is_null()) {
1006  try {
1007  impConvTest_->defineScaleForm (impResScaleType, Belos::TwoNorm);
1008  }
1009  catch (StatusTestError&) {
1010  // Delete the convergence test so it gets constructed again.
1011  impConvTest_ = null;
1012  convTest_ = null;
1013  }
1014  }
1015  }
1016  }
1017 
1018  if (params->isParameter("Explicit Residual Scaling")) {
1019  std::string tempExpResScale =
1020  getParameter<std::string> (*params, "Explicit Residual Scaling");
1021 
1022  // Only update the scaling if it's different.
1023  if (expResScale_ != tempExpResScale) {
1024  ScaleType expResScaleType = convertStringToScaleType (tempExpResScale);
1025  expResScale_ = tempExpResScale;
1026 
1027  // Update parameter in our list and residual tests
1028  params_->set("Explicit Residual Scaling", expResScale_);
1029  // NOTE (mfh 28 Feb 2011) See note above on the (re)construction
1030  // of StatusTestImpResNorm.
1031  if (! expConvTest_.is_null()) {
1032  try {
1033  expConvTest_->defineScaleForm (expResScaleType, Belos::TwoNorm);
1034  }
1035  catch (StatusTestError&) {
1036  // Delete the convergence test so it gets constructed again.
1037  expConvTest_ = null;
1038  convTest_ = null;
1039  }
1040  }
1041  }
1042  }
1043  //
1044  // Create iteration stopping criteria ("status tests") if we need
1045  // to, by combining three different stopping criteria.
1046  //
1047  // First, construct maximum-number-of-iterations stopping criterion.
1048  if (maxIterTest_.is_null())
1049  maxIterTest_ = rcp (new StatusTestMaxIters<ScalarType,MV,OP> (maxIters_));
1050 
1051  // Implicit residual test, using the native residual to determine if
1052  // convergence was achieved.
1053  if (impConvTest_.is_null()) {
1054  impConvTest_ = rcp (new StatusTestResNorm_t (convTol_));
1055  impConvTest_->defineScaleForm (convertStringToScaleType (impResScale_),
1056  Belos::TwoNorm);
1057  }
1058 
1059  // Explicit residual test once the native residual is below the tolerance
1060  if (expConvTest_.is_null()) {
1061  expConvTest_ = rcp (new StatusTestResNorm_t (convTol_));
1062  expConvTest_->defineResForm (StatusTestResNorm_t::Explicit, Belos::TwoNorm);
1063  expConvTest_->defineScaleForm (convertStringToScaleType (expResScale_),
1064  Belos::TwoNorm);
1065  }
1066  // Convergence test first tests the implicit residual, then the
1067  // explicit residual if the implicit residual test passes.
1068  if (convTest_.is_null()) {
1069  convTest_ = rcp (new StatusTestCombo_t (StatusTestCombo_t::SEQ,
1070  impConvTest_,
1071  expConvTest_));
1072  }
1073  // Construct the complete iteration stopping criterion:
1074  //
1075  // "Stop iterating if the maximum number of iterations has been
1076  // reached, or if the convergence test passes."
1077  sTest_ = rcp (new StatusTestCombo_t (StatusTestCombo_t::OR,
1078  maxIterTest_,
1079  convTest_));
1080  // Create the status test output class.
1081  // This class manages and formats the output from the status test.
1082  StatusTestOutputFactory<ScalarType,MV,OP> stoFactory (outputStyle_);
1083  outputTest_ = stoFactory.create (printer_, sTest_, outputFreq_,
1085 
1086  // Set the solver string for the output test
1087  std::string solverDesc = " GCRODR ";
1088  outputTest_->setSolverDesc( solverDesc );
1089 
1090  // Create the timer if we need to.
1091  if (timerSolve_.is_null()) {
1092  std::string solveLabel = label_ + ": GCRODRSolMgr total solve time";
1093 #ifdef BELOS_TEUCHOS_TIME_MONITOR
1094  timerSolve_ = Teuchos::TimeMonitor::getNewCounter(solveLabel);
1095 #endif
1096  }
1097 
1098  // Inform the solver manager that the current parameters were set.
1099  isSet_ = true;
1100 }
1101 
1102 
1103 template<class ScalarType, class MV, class OP>
1106 {
1107  using Teuchos::ParameterList;
1108  using Teuchos::parameterList;
1109  using Teuchos::RCP;
1110 
1111  static RCP<const ParameterList> validPL;
1112  if (is_null(validPL)) {
1113  RCP<ParameterList> pl = parameterList ();
1114 
1115  // Set all the valid parameters and their default values.
1116  pl->set("Convergence Tolerance", static_cast<MagnitudeType>(DefaultSolverParameters::convTol),
1117  "The relative residual tolerance that needs to be achieved by the\n"
1118  "iterative solver in order for the linear system to be declared converged.");
1119  pl->set("Maximum Restarts", static_cast<int>(maxRestarts_default_),
1120  "The maximum number of cycles allowed for each\n"
1121  "set of RHS solved.");
1122  pl->set("Maximum Iterations", static_cast<int>(maxIters_default_),
1123  "The maximum number of iterations allowed for each\n"
1124  "set of RHS solved.");
1125  // mfh 25 Oct 2010: "Block Size" must be 1 because GCRODR is
1126  // currently not a block method: i.e., it does not work on
1127  // multiple right-hand sides at once.
1128  pl->set("Block Size", static_cast<int>(blockSize_default_),
1129  "Block Size Parameter -- currently must be 1 for GCRODR");
1130  pl->set("Num Blocks", static_cast<int>(numBlocks_default_),
1131  "The maximum number of vectors allowed in the Krylov subspace\n"
1132  "for each set of RHS solved.");
1133  pl->set("Num Recycled Blocks", static_cast<int>(recycledBlocks_default_),
1134  "The maximum number of vectors in the recycled subspace." );
1135  pl->set("Verbosity", static_cast<int>(verbosity_default_),
1136  "What type(s) of solver information should be outputted\n"
1137  "to the output stream.");
1138  pl->set("Output Style", static_cast<int>(outputStyle_default_),
1139  "What style is used for the solver information outputted\n"
1140  "to the output stream.");
1141  pl->set("Output Frequency", static_cast<int>(outputFreq_default_),
1142  "How often convergence information should be outputted\n"
1143  "to the output stream.");
1144  pl->set("Output Stream", Teuchos::rcpFromRef(std::cout),
1145  "A reference-counted pointer to the output stream where all\n"
1146  "solver output is sent.");
1147  pl->set("Implicit Residual Scaling", static_cast<const char *>(impResScale_default_),
1148  "The type of scaling used in the implicit residual convergence test.");
1149  pl->set("Explicit Residual Scaling", static_cast<const char *>(expResScale_default_),
1150  "The type of scaling used in the explicit residual convergence test.");
1151  pl->set("Timer Label", static_cast<const char *>(label_default_),
1152  "The string to use as a prefix for the timer labels.");
1153  {
1155  pl->set("Orthogonalization", static_cast<const char *>(orthoType_default_),
1156  "The type of orthogonalization to use. Valid options: " +
1157  factory.validNamesString());
1158  RCP<const ParameterList> orthoParams =
1159  factory.getDefaultParameters (orthoType_default_);
1160  pl->set ("Orthogonalization Parameters", *orthoParams,
1161  "Parameters specific to the type of orthogonalization used.");
1162  }
1163  pl->set("Orthogonalization Constant",static_cast<MagnitudeType>(orthoKappa_default_),
1164  "When using DGKS orthogonalization: the \"depTol\" constant, used "
1165  "to determine whether another step of classical Gram-Schmidt is "
1166  "necessary. Otherwise ignored.");
1167  validPL = pl;
1168  }
1169  return validPL;
1170 }
1171 
1172 // initializeStateStorage
1173 template<class ScalarType, class MV, class OP>
1175 
1176  ScalarType zero = Teuchos::ScalarTraits<ScalarType>::zero();
1177 
1178  // Check if there is any multivector to clone from.
1179  Teuchos::RCP<const MV> rhsMV = problem_->getRHS();
1180  if (rhsMV == Teuchos::null) {
1181  // Nothing to do
1182  return;
1183  }
1184  else {
1185 
1186  // Initialize the state storage
1187  TEUCHOS_TEST_FOR_EXCEPTION(static_cast<ptrdiff_t>(numBlocks_) > MVT::GetGlobalLength(*rhsMV),std::invalid_argument,
1188  "Belos::GCRODRSolMgr::initializeStateStorage(): Cannot generate a Krylov basis with dimension larger the operator!");
1189 
1190  // If the subspace has not been initialized before, generate it using the RHS from lp_.
1191  if (U_ == Teuchos::null) {
1192  U_ = MVT::Clone( *rhsMV, recycledBlocks_+1 );
1193  }
1194  else {
1195  // Generate U_ by cloning itself ONLY if more space is needed.
1196  if (MVT::GetNumberVecs(*U_) < recycledBlocks_+1) {
1197  Teuchos::RCP<const MV> tmp = U_;
1198  U_ = MVT::Clone( *tmp, recycledBlocks_+1 );
1199  }
1200  }
1201 
1202  // If the subspace has not been initialized before, generate it using the RHS from lp_.
1203  if (C_ == Teuchos::null) {
1204  C_ = MVT::Clone( *rhsMV, recycledBlocks_+1 );
1205  }
1206  else {
1207  // Generate C_ by cloning itself ONLY if more space is needed.
1208  if (MVT::GetNumberVecs(*C_) < recycledBlocks_+1) {
1209  Teuchos::RCP<const MV> tmp = C_;
1210  C_ = MVT::Clone( *tmp, recycledBlocks_+1 );
1211  }
1212  }
1213 
1214  // If the subspace has not been initialized before, generate it using the RHS from lp_.
1215  if (V_ == Teuchos::null) {
1216  V_ = MVT::Clone( *rhsMV, numBlocks_+1 );
1217  }
1218  else {
1219  // Generate V_ by cloning itself ONLY if more space is needed.
1220  if (MVT::GetNumberVecs(*V_) < numBlocks_+1) {
1221  Teuchos::RCP<const MV> tmp = V_;
1222  V_ = MVT::Clone( *tmp, numBlocks_+1 );
1223  }
1224  }
1225 
1226  // If the subspace has not been initialized before, generate it using the RHS from lp_.
1227  if (U1_ == Teuchos::null) {
1228  U1_ = MVT::Clone( *rhsMV, recycledBlocks_+1 );
1229  }
1230  else {
1231  // Generate U1_ by cloning itself ONLY if more space is needed.
1232  if (MVT::GetNumberVecs(*U1_) < recycledBlocks_+1) {
1233  Teuchos::RCP<const MV> tmp = U1_;
1234  U1_ = MVT::Clone( *tmp, recycledBlocks_+1 );
1235  }
1236  }
1237 
1238  // If the subspace has not been initialized before, generate it using the RHS from lp_.
1239  if (C1_ == Teuchos::null) {
1240  C1_ = MVT::Clone( *rhsMV, recycledBlocks_+1 );
1241  }
1242  else {
1243  // Generate C1_ by cloning itself ONLY if more space is needed.
1244  if (MVT::GetNumberVecs(*C1_) < recycledBlocks_+1) {
1245  Teuchos::RCP<const MV> tmp = C1_;
1246  C1_ = MVT::Clone( *tmp, recycledBlocks_+1 );
1247  }
1248  }
1249 
1250  // Generate r_ only if it doesn't exist
1251  if (r_ == Teuchos::null)
1252  r_ = MVT::Clone( *rhsMV, 1 );
1253 
1254  // Size of tau_ will change during computation, so just be sure it starts with appropriate size
1255  tau_.resize(recycledBlocks_+1);
1256 
1257  // Size of work_ will change during computation, so just be sure it starts with appropriate size
1258  work_.resize(recycledBlocks_+1);
1259 
1260  // Size of ipiv_ will change during computation, so just be sure it starts with appropriate size
1261  ipiv_.resize(recycledBlocks_+1);
1262 
1263  // Generate H2_ only if it doesn't exist, otherwise resize it.
1264  if (H2_ == Teuchos::null)
1265  H2_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_+recycledBlocks_+2, numBlocks_+recycledBlocks_+1 ) );
1266  else {
1267  if ( (H2_->numRows() != numBlocks_+recycledBlocks_+2) || (H2_->numCols() != numBlocks_+recycledBlocks_+1) )
1268  H2_->reshape( numBlocks_+recycledBlocks_+2, numBlocks_+recycledBlocks_+1 );
1269  }
1270  H2_->putScalar(zero);
1271 
1272  // Generate R_ only if it doesn't exist, otherwise resize it.
1273  if (R_ == Teuchos::null)
1274  R_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( recycledBlocks_+1, recycledBlocks_+1 ) );
1275  else {
1276  if ( (R_->numRows() != recycledBlocks_+1) || (R_->numCols() != recycledBlocks_+1) )
1277  R_->reshape( recycledBlocks_+1, recycledBlocks_+1 );
1278  }
1279  R_->putScalar(zero);
1280 
1281  // Generate PP_ only if it doesn't exist, otherwise resize it.
1282  if (PP_ == Teuchos::null)
1283  PP_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_+recycledBlocks_+2, recycledBlocks_+1 ) );
1284  else {
1285  if ( (PP_->numRows() != numBlocks_+recycledBlocks_+2) || (PP_->numCols() != recycledBlocks_+1) )
1286  PP_->reshape( numBlocks_+recycledBlocks_+2, recycledBlocks_+1 );
1287  }
1288 
1289  // Generate HP_ only if it doesn't exist, otherwise resize it.
1290  if (HP_ == Teuchos::null)
1291  HP_ = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( numBlocks_+recycledBlocks_+2, numBlocks_+recycledBlocks_+1 ) );
1292  else {
1293  if ( (HP_->numRows() != numBlocks_+recycledBlocks_+2) || (HP_->numCols() != numBlocks_+recycledBlocks_+1) )
1294  HP_->reshape( numBlocks_+recycledBlocks_+2, numBlocks_+recycledBlocks_+1 );
1295  }
1296 
1297  } // end else
1298 }
1299 
1300 
1301 // solve()
1302 template<class ScalarType, class MV, class OP>
1304  using Teuchos::RCP;
1305  using Teuchos::rcp;
1306 
1307  // Set the current parameters if they were not set before.
1308  // NOTE: This may occur if the user generated the solver manager with the default constructor and
1309  // then didn't set any parameters using setParameters().
1310  if (!isSet_) { setParameters( params_ ); }
1311 
1312  ScalarType one = Teuchos::ScalarTraits<ScalarType>::one();
1313  ScalarType zero = Teuchos::ScalarTraits<ScalarType>::zero();
1314  std::vector<int> index(numBlocks_+1);
1315 
1316  TEUCHOS_TEST_FOR_EXCEPTION(problem_ == Teuchos::null,GCRODRSolMgrLinearProblemFailure, "Belos::GCRODRSolMgr::solve(): Linear problem is not a valid object.");
1317 
1318  TEUCHOS_TEST_FOR_EXCEPTION(!problem_->isProblemSet(),GCRODRSolMgrLinearProblemFailure,"Belos::GCRODRSolMgr::solve(): Linear problem is not ready, setProblem() has not been called.");
1319 
1320  // Create indices for the linear systems to be solved.
1321  int numRHS2Solve = MVT::GetNumberVecs( *(problem_->getRHS()) );
1322  std::vector<int> currIdx(1);
1323  currIdx[0] = 0;
1324 
1325  // Inform the linear problem of the current linear system to solve.
1326  problem_->setLSIndex( currIdx );
1327 
1328  // Check the number of blocks and change them is necessary.
1329  ptrdiff_t dim = MVT::GetGlobalLength( *(problem_->getRHS()) );
1330  if (static_cast<ptrdiff_t>(numBlocks_) > dim) {
1331  numBlocks_ = Teuchos::as<int>(dim);
1332  printer_->stream(Warnings) <<
1333  "Warning! Requested Krylov subspace dimension is larger than operator dimension!" << std::endl <<
1334  " The maximum number of blocks allowed for the Krylov subspace will be adjusted to " << numBlocks_ << std::endl;
1335  params_->set("Num Blocks", numBlocks_);
1336  }
1337 
1338  // Assume convergence is achieved, then let any failed convergence set this to false.
1339  bool isConverged = true;
1340 
1341  // Initialize storage for all state variables
1342  initializeStateStorage();
1343 
1345  // Parameter list
1346  Teuchos::ParameterList plist;
1347 
1348  plist.set("Num Blocks",numBlocks_);
1349  plist.set("Recycled Blocks",recycledBlocks_);
1350 
1352  // GCRODR solver
1353 
1354  RCP<GCRODRIter<ScalarType,MV,OP> > gcrodr_iter;
1355  gcrodr_iter = rcp( new GCRODRIter<ScalarType,MV,OP>(problem_,printer_,outputTest_,ortho_,plist) );
1356  // Number of iterations required to generate initial recycle space (if needed)
1357  int prime_iterations = 0;
1358 
1359  // Enter solve() iterations
1360  {
1361 #ifdef BELOS_TEUCHOS_TIME_MONITOR
1362  Teuchos::TimeMonitor slvtimer(*timerSolve_);
1363 #endif
1364 
1365  while ( numRHS2Solve > 0 ) {
1366 
1367  // Set flag indicating recycle space has not been generated this solve
1368  builtRecycleSpace_ = false;
1369 
1370  // Reset the status test.
1371  outputTest_->reset();
1372 
1374  // Initialize recycled subspace for GCRODR
1375 
1376  // If there is a subspace to recycle, recycle it, otherwise generate the initial recycled subspace.
1377  if (keff > 0) {
1379  "Belos::GCRODRSolMgr::solve(): Requested size of recycled subspace is not consistent with the current recycle subspace.");
1380 
1381  printer_->stream(Debug) << " Now solving RHS index " << currIdx[0] << " using recycled subspace of dimension " << keff << std::endl << std::endl;
1382  // Compute image of U_ under the new operator
1383  index.resize(keff);
1384  for (int ii=0; ii<keff; ++ii) { index[ii] = ii; }
1385  RCP<const MV> Utmp = MVT::CloneView( *U_, index );
1386  RCP<MV> Ctmp = MVT::CloneViewNonConst( *C_, index );
1387  problem_->apply( *Utmp, *Ctmp );
1388 
1389  RCP<MV> U1tmp = MVT::CloneViewNonConst( *U1_, index );
1390 
1391  // Orthogonalize this block
1392  // Get a matrix to hold the orthonormalization coefficients.
1394  int rank = ortho_->normalize(*Ctmp, rcp(&Rtmp,false));
1395  // Throw an error if we could not orthogonalize this block
1396  TEUCHOS_TEST_FOR_EXCEPTION(rank != keff,GCRODRSolMgrOrthoFailure,"Belos::GCRODRSolMgr::solve(): Failed to compute orthonormal basis for initial recycled subspace.");
1397 
1398  // U_ = U_*R^{-1}
1399  // First, compute LU factorization of R
1400  int info = 0;
1401  ipiv_.resize(Rtmp.numRows());
1402  lapack.GETRF(Rtmp.numRows(),Rtmp.numCols(),Rtmp.values(),Rtmp.stride(),&ipiv_[0],&info);
1403  TEUCHOS_TEST_FOR_EXCEPTION(info != 0, GCRODRSolMgrLAPACKFailure,"Belos::GCRODRSolMgr::solve(): LAPACK _GETRF failed to compute an LU factorization.");
1404 
1405  // Now, form inv(R)
1406  int lwork = Rtmp.numRows();
1407  work_.resize(lwork);
1408  lapack.GETRI(Rtmp.numRows(),Rtmp.values(),Rtmp.stride(),&ipiv_[0],&work_[0],lwork,&info);
1409  TEUCHOS_TEST_FOR_EXCEPTION(info != 0, GCRODRSolMgrLAPACKFailure,"Belos::GCRODRSolMgr::solve(): LAPACK _GETRI failed to invert triangular matrix.");
1410 
1411  // U_ = U1_; (via a swap)
1412  MVT::MvTimesMatAddMv( one, *Utmp, Rtmp, zero, *U1tmp );
1413  std::swap(U_, U1_);
1414 
1415  // Must reinitialize after swap
1416  index.resize(keff);
1417  for (int ii=0; ii<keff; ++ii) { index[ii] = ii; }
1418  Ctmp = MVT::CloneViewNonConst( *C_, index );
1419  Utmp = MVT::CloneView( *U_, index );
1420 
1421  // Compute C_'*r_
1423  problem_->computeCurrPrecResVec( &*r_ );
1424  MVT::MvTransMv( one, *Ctmp, *r_, Ctr );
1425 
1426  // Update solution ( x += U_*C_'*r_ )
1427  RCP<MV> update = MVT::Clone( *problem_->getCurrLHSVec(), 1 );
1428  MVT::MvInit( *update, 0.0 );
1429  MVT::MvTimesMatAddMv( one, *Utmp, Ctr, one, *update );
1430  problem_->updateSolution( update, true );
1431 
1432  // Update residual norm ( r -= C_*C_'*r_ )
1433  MVT::MvTimesMatAddMv( -one, *Ctmp, Ctr, one, *r_ );
1434 
1435  // We recycled space from previous call
1436  prime_iterations = 0;
1437 
1438  }
1439  else {
1440 
1441  // Do one cycle of Gmres to "prime the pump" if there is no subspace to recycle
1442  printer_->stream(Debug) << " No recycled subspace available for RHS index " << currIdx[0] << std::endl << std::endl;
1443 
1444  Teuchos::ParameterList primeList;
1445 
1446  // Tell the block solver that the block size is one.
1447  primeList.set("Num Blocks",numBlocks_);
1448  primeList.set("Recycled Blocks",0);
1449 
1450  // Create GCRODR iterator object to perform one cycle of GMRES.
1451  RCP<GCRODRIter<ScalarType,MV,OP> > gcrodr_prime_iter;
1452  gcrodr_prime_iter = rcp( new GCRODRIter<ScalarType,MV,OP>(problem_,printer_,outputTest_,ortho_,primeList) );
1453 
1454  // Create the first block in the current Krylov basis (residual).
1455  problem_->computeCurrPrecResVec( &*r_ );
1456  index.resize( 1 ); index[0] = 0;
1457  RCP<MV> v0 = MVT::CloneViewNonConst( *V_, index );
1458  MVT::SetBlock(*r_,index,*v0); // V(:,0) = r
1459 
1460  // Set the new state and initialize the solver.
1462  index.resize( numBlocks_+1 );
1463  for (int ii=0; ii<(numBlocks_+1); ++ii) { index[ii] = ii; }
1464  newstate.V = MVT::CloneViewNonConst( *V_, index );
1465  newstate.U = Teuchos::null;
1466  newstate.C = Teuchos::null;
1467  newstate.H = rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *H2_, numBlocks_+1, numBlocks_, recycledBlocks_+1, recycledBlocks_+1 ) );
1468  newstate.B = Teuchos::null;
1469  newstate.curDim = 0;
1470  gcrodr_prime_iter->initialize(newstate);
1471 
1472  // Perform one cycle of GMRES
1473  bool primeConverged = false;
1474  try {
1475  gcrodr_prime_iter->iterate();
1476 
1477  // Check convergence first
1478  if ( convTest_->getStatus() == Passed ) {
1479  // we have convergence
1480  primeConverged = true;
1481  }
1482  }
1483  catch (const GCRODRIterOrthoFailure &e) {
1484  // Try to recover the most recent least-squares solution
1485  gcrodr_prime_iter->updateLSQR( gcrodr_prime_iter->getCurSubspaceDim() );
1486 
1487  // Check to see if the most recent least-squares solution yielded convergence.
1488  sTest_->checkStatus( &*gcrodr_prime_iter );
1489  if (convTest_->getStatus() == Passed)
1490  primeConverged = true;
1491  }
1492  catch (const StatusTestNaNError& e) {
1493  // A NaN was detected in the solver. Set the solution to zero and return unconverged.
1494  achievedTol_ = MT::one();
1495  Teuchos::RCP<MV> X = problem_->getLHS();
1496  MVT::MvInit( *X, SCT::zero() );
1497  printer_->stream(Warnings) << "Belos::GCRODRSolMgr::solve(): Warning! NaN has been detected!"
1498  << std::endl;
1499  return Unconverged;
1500  }
1501  catch (const std::exception &e) {
1502  printer_->stream(Errors) << "Error! Caught exception in GCRODRIter::iterate() at iteration "
1503  << gcrodr_prime_iter->getNumIters() << std::endl
1504  << e.what() << std::endl;
1505  throw;
1506  }
1507  // Record number of iterations in generating initial recycle spacec
1508  prime_iterations = gcrodr_prime_iter->getNumIters();
1509 
1510  // Update the linear problem.
1511  RCP<MV> update = gcrodr_prime_iter->getCurrentUpdate();
1512  problem_->updateSolution( update, true );
1513 
1514  // Get the state.
1515  newstate = gcrodr_prime_iter->getState();
1516  int p = newstate.curDim;
1517 
1518  // Compute harmonic Ritz vectors
1519  // NOTE: The storage for the harmonic Ritz vectors (PP) is made one column larger
1520  // just in case we split a complex conjugate pair.
1521  // NOTE: Generate a recycled subspace only if we have enough vectors. If we converged
1522  // too early, move on to the next linear system and try to generate a subspace again.
1523  if (recycledBlocks_ < p+1) {
1524  int info = 0;
1525  RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > PPtmp = rcp (new Teuchos::SerialDenseMatrix<int,ScalarType> ( Teuchos::View, *PP_, p, recycledBlocks_+1 ) );
1526  // getHarmonicVecs1 assumes PP has recycledBlocks_+1 columns available
1527  keff = getHarmonicVecs1( p, *newstate.H, *PPtmp );
1528  // Hereafter, only keff columns of PP are needed
1529  PPtmp = rcp (new Teuchos::SerialDenseMatrix<int,ScalarType> ( Teuchos::View, *PP_, p, keff ) );
1530  // Now get views into C, U, V
1531  index.resize(keff);
1532  for (int ii=0; ii<keff; ++ii) { index[ii] = ii; }
1533  RCP<MV> Ctmp = MVT::CloneViewNonConst( *C_, index );
1534  RCP<MV> Utmp = MVT::CloneViewNonConst( *U_, index );
1535  RCP<MV> U1tmp = MVT::CloneViewNonConst( *U1_, index );
1536  index.resize(p);
1537  for (int ii=0; ii < p; ++ii) { index[ii] = ii; }
1538  RCP<const MV> Vtmp = MVT::CloneView( *V_, index );
1539 
1540  // Form U (the subspace to recycle)
1541  // U = newstate.V(:,1:p) * PP;
1542  MVT::MvTimesMatAddMv( one, *Vtmp, *PPtmp, zero, *U1tmp );
1543 
1544  // Form orthonormalized C and adjust U so that C = A*U
1545 
1546  // First, compute [Q, R] = qr(H*P);
1547 
1548  // Step #1: Form HP = H*P
1549  Teuchos::SerialDenseMatrix<int,ScalarType> Htmp( Teuchos::View, *H2_, p+1, p, recycledBlocks_+1,recycledBlocks_+1);
1551  HPtmp.multiply( Teuchos::NO_TRANS, Teuchos::NO_TRANS, one, Htmp, *PPtmp, zero );
1552 
1553  // Step #1.5: Perform workspace size query for QR
1554  // factorization of HP. On input, lwork must be -1.
1555  // _GEQRF will put the workspace size in work_[0].
1556  int lwork = -1;
1557  tau_.resize (keff);
1558  lapack.GEQRF (HPtmp.numRows (), HPtmp.numCols (), HPtmp.values (),
1559  HPtmp.stride (), &tau_[0], &work_[0], lwork, &info);
1561  info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve:"
1562  " LAPACK's _GEQRF failed to compute a workspace size.");
1563 
1564  // Step #2: Compute QR factorization of HP
1565  //
1566  // NOTE (mfh 17 Apr 2014) LAPACK promises that the value of
1567  // work_[0] after the workspace query will fit in int. This
1568  // justifies the cast. We call real() first because
1569  // static_cast from std::complex to int doesn't work.
1570  lwork = std::abs (static_cast<int> (Teuchos::ScalarTraits<ScalarType>::real (work_[0])));
1571  work_.resize (lwork); // Allocate workspace for the QR factorization
1572  lapack.GEQRF (HPtmp.numRows (), HPtmp.numCols (), HPtmp.values (),
1573  HPtmp.stride (), &tau_[0], &work_[0], lwork, &info);
1575  info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve:"
1576  " LAPACK's _GEQRF failed to compute a QR factorization.");
1577 
1578  // Step #3: Explicitly construct Q and R factors
1579  // NOTE: The upper triangular part of HP is copied into R and HP becomes Q.
1581  for (int ii = 0; ii < keff; ++ii) {
1582  for (int jj = ii; jj < keff; ++jj) {
1583  Rtmp(ii,jj) = HPtmp(ii,jj);
1584  }
1585  }
1586  // NOTE (mfh 17 Apr 2014): Teuchos::LAPACK's wrapper for
1587  // UNGQR dispatches to the correct Scalar-specific routine.
1588  // It calls {S,D}ORGQR if Scalar is real, and {C,Z}UNGQR if
1589  // Scalar is complex.
1590  lapack.UNGQR (HPtmp.numRows (), HPtmp.numCols (), HPtmp.numCols (),
1591  HPtmp.values (), HPtmp.stride (), &tau_[0], &work_[0],
1592  lwork, &info);
1594  info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve: "
1595  "LAPACK's _UNGQR failed to construct the Q factor.");
1596 
1597  // Now we have [Q,R] = qr(H*P)
1598 
1599  // Now compute C = V(:,1:p+1) * Q
1600  index.resize (p + 1);
1601  for (int ii = 0; ii < (p+1); ++ii) {
1602  index[ii] = ii;
1603  }
1604  Vtmp = MVT::CloneView( *V_, index ); // need new view into V (p+1 vectors now; needed p above)
1605  MVT::MvTimesMatAddMv( one, *Vtmp, HPtmp, zero, *Ctmp );
1606 
1607  // Finally, compute U = U*R^{-1}.
1608  // This unfortuntely requires me to form R^{-1} explicitly and execute U = U * R^{-1}, as
1609  // backsolve capabilities don't exist in the Belos::MultiVec class
1610 
1611  // Step #1: First, compute LU factorization of R
1612  ipiv_.resize(Rtmp.numRows());
1613  lapack.GETRF(Rtmp.numRows(),Rtmp.numCols(),Rtmp.values(),Rtmp.stride(),&ipiv_[0],&info);
1615  info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve: "
1616  "LAPACK's _GETRF failed to compute an LU factorization.");
1617 
1618  // FIXME (mfh 17 Apr 2014) We have to compute the explicit
1619  // inverse of R here because Belos::MultiVecTraits doesn't
1620  // have a triangular solve (where the triangular matrix is
1621  // globally replicated and the "right-hand side" is the
1622  // distributed MultiVector).
1623 
1624  // Step #2: Form inv(R)
1625  lwork = Rtmp.numRows();
1626  work_.resize(lwork);
1627  lapack.GETRI(Rtmp.numRows(),Rtmp.values(),Rtmp.stride(),&ipiv_[0],&work_[0],lwork,&info);
1629  info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve: "
1630  "LAPACK's _GETRI failed to invert triangular matrix.");
1631 
1632  // Step #3: Let U = U * R^{-1}
1633  MVT::MvTimesMatAddMv( one, *U1tmp, Rtmp, zero, *Utmp );
1634 
1635  printer_->stream(Debug)
1636  << " Generated recycled subspace using RHS index " << currIdx[0]
1637  << " of dimension " << keff << std::endl << std::endl;
1638 
1639  } // if (recycledBlocks_ < p+1)
1640 
1641  // Return to outer loop if the priming solve converged, set the next linear system.
1642  if (primeConverged) {
1643  // Inform the linear problem that we are finished with this block linear system.
1644  problem_->setCurrLS();
1645 
1646  // Update indices for the linear systems to be solved.
1647  numRHS2Solve -= 1;
1648  if (numRHS2Solve > 0) {
1649  currIdx[0]++;
1650  problem_->setLSIndex (currIdx); // Set the next indices
1651  }
1652  else {
1653  currIdx.resize (numRHS2Solve);
1654  }
1655 
1656  continue;
1657  }
1658  } // if (keff > 0) ...
1659 
1660  // Prepare for the Gmres iterations with the recycled subspace.
1661 
1662  // Set the current number of recycled blocks and subspace dimension with the GCRO-DR iteration.
1663  gcrodr_iter->setSize( keff, numBlocks_ );
1664 
1665  // Reset the number of iterations.
1666  gcrodr_iter->resetNumIters(prime_iterations);
1667 
1668  // Reset the number of calls that the status test output knows about.
1669  outputTest_->resetNumCalls();
1670 
1671  // Compute the residual after the priming solve, it will be the first block in the current Krylov basis.
1672  problem_->computeCurrPrecResVec( &*r_ );
1673  index.resize( 1 ); index[0] = 0;
1674  RCP<MV> v0 = MVT::CloneViewNonConst( *V_, index );
1675  MVT::SetBlock(*r_,index,*v0); // V(:,0) = r
1676 
1677  // Set the new state and initialize the solver.
1679  index.resize( numBlocks_+1 );
1680  for (int ii=0; ii<(numBlocks_+1); ++ii) { index[ii] = ii; }
1681  newstate.V = MVT::CloneViewNonConst( *V_, index );
1682  index.resize( keff );
1683  for (int ii=0; ii<keff; ++ii) { index[ii] = ii; }
1684  newstate.C = MVT::CloneViewNonConst( *C_, index );
1685  newstate.U = MVT::CloneViewNonConst( *U_, index );
1686  newstate.B = rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *H2_, keff, numBlocks_, 0, keff ) );
1687  newstate.H = rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *H2_, numBlocks_+1, numBlocks_, keff, keff ) );
1688  newstate.curDim = 0;
1689  gcrodr_iter->initialize(newstate);
1690 
1691  // variables needed for inner loop
1692  int numRestarts = 0;
1693  while(1) {
1694 
1695  // tell gcrodr_iter to iterate
1696  try {
1697  gcrodr_iter->iterate();
1698 
1700  //
1701  // check convergence first
1702  //
1704  if ( convTest_->getStatus() == Passed ) {
1705  // we have convergence
1706  break; // break from while(1){gcrodr_iter->iterate()}
1707  }
1709  //
1710  // check for maximum iterations
1711  //
1713  else if ( maxIterTest_->getStatus() == Passed ) {
1714  // we don't have convergence
1715  isConverged = false;
1716  break; // break from while(1){gcrodr_iter->iterate()}
1717  }
1719  //
1720  // check for restarting, i.e. the subspace is full
1721  //
1723  else if ( gcrodr_iter->getCurSubspaceDim() == gcrodr_iter->getMaxSubspaceDim() ) {
1724 
1725  // Update the recycled subspace even if we have hit the maximum number of restarts.
1726 
1727  // Update the linear problem.
1728  RCP<MV> update = gcrodr_iter->getCurrentUpdate();
1729  problem_->updateSolution( update, true );
1730 
1731  buildRecycleSpace2(gcrodr_iter);
1732 
1733  printer_->stream(Debug)
1734  << " Generated new recycled subspace using RHS index "
1735  << currIdx[0] << " of dimension " << keff << std::endl
1736  << std::endl;
1737 
1738  // NOTE: If we have hit the maximum number of restarts then quit
1739  if (numRestarts >= maxRestarts_) {
1740  isConverged = false;
1741  break; // break from while(1){gcrodr_iter->iterate()}
1742  }
1743  numRestarts++;
1744 
1745  printer_->stream(Debug)
1746  << " Performing restart number " << numRestarts << " of "
1747  << maxRestarts_ << std::endl << std::endl;
1748 
1749  // Create the restart vector (first block in the current Krylov basis)
1750  problem_->computeCurrPrecResVec( &*r_ );
1751  index.resize( 1 ); index[0] = 0;
1752  RCP<MV> v00 = MVT::CloneViewNonConst( *V_, index );
1753  MVT::SetBlock(*r_,index,*v00); // V(:,0) = r
1754 
1755  // Set the new state and initialize the solver.
1756  GCRODRIterState<ScalarType,MV> restartState;
1757  index.resize( numBlocks_+1 );
1758  for (int ii=0; ii<(numBlocks_+1); ++ii) { index[ii] = ii; }
1759  restartState.V = MVT::CloneViewNonConst( *V_, index );
1760  index.resize( keff );
1761  for (int ii=0; ii<keff; ++ii) { index[ii] = ii; }
1762  restartState.U = MVT::CloneViewNonConst( *U_, index );
1763  restartState.C = MVT::CloneViewNonConst( *C_, index );
1764  restartState.B = rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *H2_, keff, numBlocks_, 0, keff ) );
1765  restartState.H = rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>( Teuchos::View, *H2_, numBlocks_+1, numBlocks_, keff, keff ) );
1766  restartState.curDim = 0;
1767  gcrodr_iter->initialize(restartState);
1768 
1769 
1770  } // end of restarting
1771 
1773  //
1774  // we returned from iterate(), but none of our status tests Passed.
1775  // something is wrong, and it is probably our fault.
1776  //
1778 
1779  else {
1781  true, std::logic_error, "Belos::GCRODRSolMgr::solve: "
1782  "Invalid return from GCRODRIter::iterate().");
1783  }
1784  }
1785  catch (const GCRODRIterOrthoFailure &e) {
1786  // Try to recover the most recent least-squares solution
1787  gcrodr_iter->updateLSQR( gcrodr_iter->getCurSubspaceDim() );
1788 
1789  // Check to see if the most recent least-squares solution yielded convergence.
1790  sTest_->checkStatus( &*gcrodr_iter );
1791  if (convTest_->getStatus() != Passed)
1792  isConverged = false;
1793  break;
1794  }
1795  catch (const std::exception& e) {
1796  printer_->stream(Errors)
1797  << "Error! Caught exception in GCRODRIter::iterate() at iteration "
1798  << gcrodr_iter->getNumIters() << std::endl << e.what() << std::endl;
1799  throw;
1800  }
1801  }
1802 
1803  // Compute the current solution.
1804  // Update the linear problem.
1805  RCP<MV> update = gcrodr_iter->getCurrentUpdate();
1806  problem_->updateSolution( update, true );
1807 
1808  // Inform the linear problem that we are finished with this block linear system.
1809  problem_->setCurrLS();
1810 
1811  // If we didn't build a recycle space this solve but ran at least k iterations,
1812  // force build of new recycle space
1813 
1814  if (!builtRecycleSpace_) {
1815  buildRecycleSpace2(gcrodr_iter);
1816  printer_->stream(Debug)
1817  << " Generated new recycled subspace using RHS index " << currIdx[0]
1818  << " of dimension " << keff << std::endl << std::endl;
1819  }
1820 
1821  // Update indices for the linear systems to be solved.
1822  numRHS2Solve -= 1;
1823  if (numRHS2Solve > 0) {
1824  currIdx[0]++;
1825  problem_->setLSIndex (currIdx); // Set the next indices
1826  }
1827  else {
1828  currIdx.resize (numRHS2Solve);
1829  }
1830  } // while (numRHS2Solve > 0)
1831  }
1832 
1833  sTest_->print (printer_->stream (FinalSummary)); // print final summary
1834 
1835  // print timing information
1836 #ifdef BELOS_TEUCHOS_TIME_MONITOR
1837  // Calling summarize() can be expensive, so don't call unless the
1838  // user wants to print out timing details. summarize() will do all
1839  // the work even if it's passed a "black hole" output stream.
1840  if (verbosity_ & TimingDetails)
1841  Teuchos::TimeMonitor::summarize( printer_->stream(TimingDetails) );
1842 #endif // BELOS_TEUCHOS_TIME_MONITOR
1843 
1844  // get iteration information for this solve
1845  numIters_ = maxIterTest_->getNumIters ();
1846 
1847  // Save the convergence test value ("achieved tolerance") for this
1848  // solve. This solver (unlike BlockGmresSolMgr) always has two
1849  // residual norm status tests: an explicit and an implicit test.
1850  // The master convergence test convTest_ is a SEQ combo of the
1851  // implicit resp. explicit tests. If the implicit test never
1852  // passes, then the explicit test won't ever be executed. This
1853  // manifests as expConvTest_->getTestValue()->size() < 1. We deal
1854  // with this case by using the values returned by
1855  // impConvTest_->getTestValue().
1856  {
1857  const std::vector<MagnitudeType>* pTestValues = expConvTest_->getTestValue();
1858  if (pTestValues == NULL || pTestValues->size() < 1) {
1859  pTestValues = impConvTest_->getTestValue();
1860  }
1861  TEUCHOS_TEST_FOR_EXCEPTION(pTestValues == NULL, std::logic_error,
1862  "Belos::GCRODRSolMgr::solve(): The implicit convergence test's getTestValue() "
1863  "method returned NULL. Please report this bug to the Belos developers.");
1864  TEUCHOS_TEST_FOR_EXCEPTION(pTestValues->size() < 1, std::logic_error,
1865  "Belos::GCRODRSolMgr::solve(): The implicit convergence test's getTestValue() "
1866  "method returned a vector of length zero. Please report this bug to the "
1867  "Belos developers.");
1868 
1869  // FIXME (mfh 12 Dec 2011) Does pTestValues really contain the
1870  // achieved tolerances for all vectors in the current solve(), or
1871  // just for the vectors from the last deflation?
1872  achievedTol_ = *std::max_element (pTestValues->begin(), pTestValues->end());
1873  }
1874 
1875  return isConverged ? Converged : Unconverged; // return from solve()
1876 }
1877 
1878 // Given existing recycle space and Krylov space, build new recycle space
1879 template<class ScalarType, class MV, class OP>
1881 
1883  ScalarType zero = Teuchos::ScalarTraits<ScalarType>::zero();
1884 
1885  std::vector<MagnitudeType> d(keff);
1886  std::vector<ScalarType> dscalar(keff);
1887  std::vector<int> index(numBlocks_+1);
1888 
1889  // Get the state
1890  GCRODRIterState<ScalarType,MV> oldState = gcrodr_iter->getState();
1891  int p = oldState.curDim;
1892 
1893  // insufficient new information to update recycle space
1894  if (p<1) return;
1895 
1896  // Take the norm of the recycled vectors
1897  {
1898  index.resize(keff);
1899  for (int ii=0; ii<keff; ++ii) { index[ii] = ii; }
1900  Teuchos::RCP<MV> Utmp = MVT::CloneViewNonConst( *U_, index );
1901  d.resize(keff);
1902  dscalar.resize(keff);
1903  MVT::MvNorm( *Utmp, d );
1904  for (int i=0; i<keff; ++i) {
1905  d[i] = one / d[i];
1906  dscalar[i] = (ScalarType)d[i];
1907  }
1908  MVT::MvScale( *Utmp, dscalar );
1909  }
1910 
1911  // Get view into current "full" upper Hessnburg matrix
1913 
1914  // Insert D into the leading keff x keff block of H2
1915  for (int i=0; i<keff; ++i) {
1916  (*H2tmp)(i,i) = d[i];
1917  }
1918 
1919  // Compute the harmoic Ritz pairs for the generalized eigenproblem
1920  // getHarmonicVecs2 assumes PP has recycledBlocks_+1 columns available
1921  int keff_new;
1922  {
1923  Teuchos::SerialDenseMatrix<int,ScalarType> PPtmp( Teuchos::View, *PP_, p+keff, recycledBlocks_+1 );
1924  keff_new = getHarmonicVecs2( keff, p, *H2tmp, oldState.V, PPtmp );
1925  }
1926 
1927  // Code to form new U, C
1928  // U = [U V(:,1:p)] * P; (in two steps)
1929 
1930  // U(:,1:keff) = matmul(U(:,1:keff_old),PP(1:keff_old,1:keff)) (step 1)
1931  Teuchos::RCP<MV> U1tmp;
1932  {
1933  index.resize( keff );
1934  for (int ii=0; ii<keff; ++ii) { index[ii] = ii; }
1935  Teuchos::RCP<const MV> Utmp = MVT::CloneView( *U_, index );
1936  index.resize( keff_new );
1937  for (int ii=0; ii<keff_new; ++ii) { index[ii] = ii; }
1938  U1tmp = MVT::CloneViewNonConst( *U1_, index );
1939  Teuchos::SerialDenseMatrix<int,ScalarType> PPtmp( Teuchos::View, *PP_, keff, keff_new );
1940  MVT::MvTimesMatAddMv( one, *Utmp, PPtmp, zero, *U1tmp );
1941  }
1942 
1943  // U(:,1:keff) = U(:,1:keff) + matmul(V(:,1:m-k),PP(keff_old+1:m-k+keff_old,1:keff)) (step 2)
1944  {
1945  index.resize(p);
1946  for (int ii=0; ii < p; ii++) { index[ii] = ii; }
1947  Teuchos::RCP<const MV> Vtmp = MVT::CloneView( *V_, index );
1948  Teuchos::SerialDenseMatrix<int,ScalarType> PPtmp( Teuchos::View, *PP_, p, keff_new, keff );
1949  MVT::MvTimesMatAddMv( one, *Vtmp, PPtmp, one, *U1tmp );
1950  }
1951 
1952  // Form HP = H*P
1953  Teuchos::SerialDenseMatrix<int,ScalarType> HPtmp( Teuchos::View, *HP_, p+keff+1, keff_new );
1954  {
1955  Teuchos::SerialDenseMatrix<int,ScalarType> PPtmp( Teuchos::View, *PP_, p+keff, keff_new );
1956  HPtmp.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,one,*H2tmp,PPtmp,zero);
1957  }
1958 
1959  // Workspace size query for QR factorization of HP (the worksize will be placed in work_[0])
1960  int info = 0, lwork = -1;
1961  tau_.resize (keff_new);
1962  lapack.GEQRF (HPtmp.numRows (), HPtmp.numCols (), HPtmp.values (),
1963  HPtmp.stride (), &tau_[0], &work_[0], lwork, &info);
1965  info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve: "
1966  "LAPACK's _GEQRF failed to compute a workspace size.");
1967 
1968  // NOTE (mfh 18 Apr 2014) LAPACK promises that the value of work_[0]
1969  // after the workspace query will fit in int. This justifies the
1970  // cast. We call real() first because static_cast from std::complex
1971  // to int doesn't work.
1972  lwork = std::abs (static_cast<int> (Teuchos::ScalarTraits<ScalarType>::real (work_[0])));
1973  work_.resize (lwork); // Allocate workspace for the QR factorization
1974  lapack.GEQRF (HPtmp.numRows (), HPtmp.numCols (), HPtmp.values (),
1975  HPtmp.stride (), &tau_[0], &work_[0], lwork, &info);
1977  info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve: "
1978  "LAPACK's _GEQRF failed to compute a QR factorization.");
1979 
1980  // Explicitly construct Q and R factors
1981  // NOTE: The upper triangular part of HP is copied into R and HP becomes Q.
1982  Teuchos::SerialDenseMatrix<int,ScalarType> Rtmp( Teuchos::View, *R_, keff_new, keff_new );
1983  for(int i=0;i<keff_new;i++) { for(int j=i;j<keff_new;j++) Rtmp(i,j) = HPtmp(i,j); }
1984 
1985  // NOTE (mfh 18 Apr 2014): Teuchos::LAPACK's wrapper for UNGQR
1986  // dispatches to the correct Scalar-specific routine. It calls
1987  // {S,D}ORGQR if Scalar is real, and {C,Z}UNGQR if Scalar is
1988  // complex.
1989  lapack.UNGQR (HPtmp.numRows (), HPtmp.numCols (), HPtmp.numCols (),
1990  HPtmp.values (), HPtmp.stride (), &tau_[0], &work_[0],
1991  lwork, &info);
1993  info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve: "
1994  "LAPACK's _UNGQR failed to construct the Q factor.");
1995 
1996  // Form orthonormalized C and adjust U accordingly so that C = A*U
1997  // C = [C V] * Q;
1998 
1999  // C(:,1:keff) = matmul(C(:,1:keff_old),QQ(1:keff_old,1:keff))
2000  {
2001  Teuchos::RCP<MV> C1tmp;
2002  {
2003  index.resize(keff);
2004  for (int i=0; i < keff; i++) { index[i] = i; }
2005  Teuchos::RCP<const MV> Ctmp = MVT::CloneView( *C_, index );
2006  index.resize(keff_new);
2007  for (int i=0; i < keff_new; i++) { index[i] = i; }
2008  C1tmp = MVT::CloneViewNonConst( *C1_, index );
2009  Teuchos::SerialDenseMatrix<int,ScalarType> PPtmp( Teuchos::View, *HP_, keff, keff_new );
2010  MVT::MvTimesMatAddMv( one, *Ctmp, PPtmp, zero, *C1tmp );
2011  }
2012  // Now compute C += V(:,1:p+1) * Q
2013  {
2014  index.resize( p+1 );
2015  for (int i=0; i < p+1; ++i) { index[i] = i; }
2016  Teuchos::RCP<const MV> Vtmp = MVT::CloneView( *V_, index );
2017  Teuchos::SerialDenseMatrix<int,ScalarType> PPtmp( Teuchos::View, *HP_, p+1, keff_new, keff, 0 );
2018  MVT::MvTimesMatAddMv( one, *Vtmp, PPtmp, one, *C1tmp );
2019  }
2020  }
2021 
2022  // C_ = C1_; (via a swap)
2023  std::swap(C_, C1_);
2024 
2025  // Finally, compute U_ = U_*R^{-1}
2026  // First, compute LU factorization of R
2027  ipiv_.resize(Rtmp.numRows());
2028  lapack.GETRF(Rtmp.numRows(),Rtmp.numCols(),Rtmp.values(),Rtmp.stride(),&ipiv_[0],&info);
2029  TEUCHOS_TEST_FOR_EXCEPTION(info != 0,GCRODRSolMgrLAPACKFailure,"Belos::GCRODRSolMgr::solve(): LAPACK _GETRF failed to compute an LU factorization.");
2030 
2031  // Now, form inv(R)
2032  lwork = Rtmp.numRows();
2033  work_.resize(lwork);
2034  lapack.GETRI(Rtmp.numRows(),Rtmp.values(),Rtmp.stride(),&ipiv_[0],&work_[0],lwork,&info);
2035  TEUCHOS_TEST_FOR_EXCEPTION(info != 0, GCRODRSolMgrLAPACKFailure,"Belos::GCRODRSolMgr::solve(): LAPACK _GETRI failed to compute an LU factorization.");
2036 
2037  {
2038  index.resize(keff_new);
2039  for (int i=0; i < keff_new; i++) { index[i] = i; }
2040  Teuchos::RCP<MV> Utmp = MVT::CloneViewNonConst( *U_, index );
2041  MVT::MvTimesMatAddMv( one, *U1tmp, Rtmp, zero, *Utmp );
2042  }
2043 
2044  // Set the current number of recycled blocks and subspace dimension with the GCRO-DR iteration.
2045  if (keff != keff_new) {
2046  keff = keff_new;
2047  gcrodr_iter->setSize( keff, numBlocks_ );
2048  // Important to zero this out before next cyle
2049  Teuchos::SerialDenseMatrix<int,ScalarType> b1( Teuchos::View, *H2_, recycledBlocks_+2, 1, 0, recycledBlocks_ );
2050  b1.putScalar(zero);
2051  }
2052 
2053 }
2054 
2055 
2056 // Compute the harmonic eigenpairs of the projected, dense system.
2057 template<class ScalarType, class MV, class OP>
2061  int i, j;
2062  bool xtraVec = false;
2063  ScalarType one = Teuchos::ScalarTraits<ScalarType>::one();
2064 
2065  // Real and imaginary eigenvalue components
2066  std::vector<MagnitudeType> wr(m), wi(m);
2067 
2068  // Real and imaginary (right) eigenvectors; Don't zero out matrix when constructing
2070 
2071  // Magnitude of harmonic Ritz values
2072  std::vector<MagnitudeType> w(m);
2073 
2074  // Sorted order of harmonic Ritz values, also used for GEEV
2075  std::vector<int> iperm(m);
2076 
2077  // Output info
2078  int info = 0;
2079 
2080  // Set flag indicating recycle space has been generated this solve
2081  builtRecycleSpace_ = true;
2082 
2083  // Solve linear system: H_m^{-H}*e_m
2086  e_m[m-1] = one;
2087  lapack.GESV(m, 1, HHt.values(), HHt.stride(), &iperm[0], e_m.values(), e_m.stride(), &info);
2088  TEUCHOS_TEST_FOR_EXCEPTION(info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve(): LAPACK GESV failed to compute a solution.");
2089 
2090  // Compute H_m + d*H_m^{-H}*e_m*e_m^H
2091  ScalarType d = HH(m, m-1) * HH(m, m-1);
2093  for( i=0; i<m; ++i )
2094  harmHH(i, m-1) += d * e_m[i];
2095 
2096  // Revise to do query for optimal workspace first
2097  // Create simple storage for the left eigenvectors, which we don't care about.
2098  const int ldvl = 1;
2099  ScalarType* vl = 0;
2100 
2101  // Size of workspace and workspace for GEEV
2102  int lwork = -1;
2103  std::vector<ScalarType> work(1);
2104  std::vector<MagnitudeType> rwork(2*m);
2105 
2106  // First query GEEV for the optimal workspace size
2107  lapack.GEEV('N', 'V', m, harmHH.values(), harmHH.stride(), &wr[0], &wi[0],
2108  vl, ldvl, vr.values(), vr.stride(), &work[0], lwork, &rwork[0], &info);
2109 
2110  lwork = std::abs (static_cast<int> (Teuchos::ScalarTraits<ScalarType>::real (work[0])));
2111  work.resize( lwork );
2112 
2113  lapack.GEEV('N', 'V', m, harmHH.values(), harmHH.stride(), &wr[0], &wi[0],
2114  vl, ldvl, vr.values(), vr.stride(), &work[0], lwork, &rwork[0], &info);
2115  TEUCHOS_TEST_FOR_EXCEPTION(info != 0, GCRODRSolMgrLAPACKFailure,"Belos::GCRODRSolMgr::solve(): LAPACK GEEV failed to compute eigensolutions.");
2116 
2117  // Construct magnitude of each harmonic Ritz value
2118  for( i=0; i<m; ++i )
2119  w[i] = Teuchos::ScalarTraits<MagnitudeType>::squareroot( wr[i]*wr[i] + wi[i]*wi[i] );
2120 
2121  // Construct magnitude of each harmonic Ritz value
2122  this->sort(w, m, iperm);
2123 
2124  const bool scalarTypeIsComplex = Teuchos::ScalarTraits<ScalarType>::isComplex;
2125 
2126  // Select recycledBlocks_ smallest eigenvectors
2127  for( i=0; i<recycledBlocks_; ++i ) {
2128  for( j=0; j<m; j++ ) {
2129  PP(j,i) = vr(j,iperm[i]);
2130  }
2131  }
2132 
2133  if(!scalarTypeIsComplex) {
2134 
2135  // Determine exact size for PP (i.e., determine if we need to store an additional vector)
2136  if (wi[iperm[recycledBlocks_-1]] != 0.0) {
2137  int countImag = 0;
2138  for ( i=0; i<recycledBlocks_; ++i ) {
2139  if (wi[iperm[i]] != 0.0)
2140  countImag++;
2141  }
2142  // Check to see if this count is even or odd:
2143  if (countImag % 2)
2144  xtraVec = true;
2145  }
2146 
2147  if (xtraVec) { // we need to store one more vector
2148  if (wi[iperm[recycledBlocks_-1]] > 0.0) { // I picked the "real" component
2149  for( j=0; j<m; ++j ) { // so get the "imag" component
2150  PP(j,recycledBlocks_) = vr(j,iperm[recycledBlocks_-1]+1);
2151  }
2152  }
2153  else { // I picked the "imag" component
2154  for( j=0; j<m; ++j ) { // so get the "real" component
2155  PP(j,recycledBlocks_) = vr(j,iperm[recycledBlocks_-1]-1);
2156  }
2157  }
2158  }
2159 
2160  }
2161 
2162  // Return whether we needed to store an additional vector
2163  if (xtraVec) {
2164  return recycledBlocks_+1;
2165  }
2166  else {
2167  return recycledBlocks_;
2168  }
2169 
2170 }
2171 
2172 // Compute the harmonic eigenpairs of the projected, dense system.
2173 template<class ScalarType, class MV, class OP>
2176  const Teuchos::RCP<const MV>& VV,
2178  int i, j;
2179  int m2 = HH.numCols();
2180  bool xtraVec = false;
2181  ScalarType one = Teuchos::ScalarTraits<ScalarType>::one();
2182  ScalarType zero = Teuchos::ScalarTraits<ScalarType>::zero();
2183  std::vector<int> index;
2184 
2185  // Real and imaginary eigenvalue components
2186  std::vector<MagnitudeType> wr(m2), wi(m2);
2187 
2188  // Magnitude of harmonic Ritz values
2189  std::vector<MagnitudeType> w(m2);
2190 
2191  // Real and imaginary (right) eigenvectors; Don't zero out matrix when constructing
2193 
2194  // Sorted order of harmonic Ritz values
2195  std::vector<int> iperm(m2);
2196 
2197  // Set flag indicating recycle space has been generated this solve
2198  builtRecycleSpace_ = true;
2199 
2200  // Form matrices for generalized eigenproblem
2201 
2202  // B = H2' * H2; Don't zero out matrix when constructing
2204  B.multiply(Teuchos::TRANS,Teuchos::NO_TRANS,one,HH,HH,zero);
2205 
2206  // A_tmp = | C'*U 0 |
2207  // | V_{m+1}'*U I |
2208  Teuchos::SerialDenseMatrix<int,ScalarType> A_tmp( keffloc+m+1, keffloc+m );
2209 
2210  // A_tmp(1:keffloc,1:keffloc) = C' * U;
2211  index.resize(keffloc);
2212  for (i=0; i<keffloc; ++i) { index[i] = i; }
2213  Teuchos::RCP<const MV> Ctmp = MVT::CloneView( *C_, index );
2214  Teuchos::RCP<const MV> Utmp = MVT::CloneView( *U_, index );
2215  Teuchos::SerialDenseMatrix<int,ScalarType> A11( Teuchos::View, A_tmp, keffloc, keffloc );
2216  MVT::MvTransMv( one, *Ctmp, *Utmp, A11 );
2217 
2218  // A_tmp(keffloc+1:m-k+keffloc+1,1:keffloc) = V' * U;
2219  Teuchos::SerialDenseMatrix<int,ScalarType> A21( Teuchos::View, A_tmp, m+1, keffloc, keffloc );
2220  index.resize(m+1);
2221  for (i=0; i < m+1; i++) { index[i] = i; }
2222  Teuchos::RCP<const MV> Vp = MVT::CloneView( *VV, index );
2223  MVT::MvTransMv( one, *Vp, *Utmp, A21 );
2224 
2225  // A_tmp(keffloc+1:m-k+keffloc,keffloc+1:m-k+keffloc) = eye(m-k);
2226  for( i=keffloc; i<keffloc+m; i++ ) {
2227  A_tmp(i,i) = one;
2228  }
2229 
2230  // A = H2' * A_tmp;
2232  A.multiply( Teuchos::TRANS, Teuchos::NO_TRANS, one, HH, A_tmp, zero );
2233 
2234  // Compute k smallest harmonic Ritz pairs
2235  // SUBROUTINE DGGEVX( BALANC, JOBVL, JOBVR, SENSE, N, A, LDA, B, LDB,
2236  // ALPHAR, ALPHAI, BETA, VL, LDVL, VR, LDVR, ILO,
2237  // IHI, LSCALE, RSCALE, ABNRM, BBNRM, RCONDE,
2238  // RCONDV, WORK, LWORK, IWORK, BWORK, INFO )
2239  // MLP: 'SCALING' in DGGEVX generates incorrect eigenvalues. Therefore, only permuting
2240  char balanc='P', jobvl='N', jobvr='V', sense='N';
2241  int ld = A.numRows();
2242  int lwork = 6*ld;
2243  int ldvl = ld, ldvr = ld;
2244  int info = 0,ilo = 0,ihi = 0;
2245  MagnitudeType abnrm = 0.0, bbnrm = 0.0;
2246  ScalarType *vl = 0; // This is never referenced by dggevx if jobvl == 'N'
2247  std::vector<ScalarType> beta(ld);
2248  std::vector<ScalarType> work(lwork);
2249  std::vector<MagnitudeType> rwork(lwork);
2250  std::vector<MagnitudeType> lscale(ld), rscale(ld);
2251  std::vector<MagnitudeType> rconde(ld), rcondv(ld);
2252  std::vector<int> iwork(ld+6);
2253  int *bwork = 0; // If sense == 'N', bwork is never referenced
2254  //lapack.GGEVX(balanc, jobvl, jobvr, sense, ld, A.values(), ld, B.values(), ld, &wr[0], &wi[0],
2255  // &beta[0], vl, ldvl, vr.values(), ldvr, &ilo, &ihi, &lscale[0], &rscale[0],
2256  // &abnrm, &bbnrm, &rconde[0], &rcondv[0], &work[0], lwork, &iwork[0], bwork, &info);
2257  lapack.GGEVX(balanc, jobvl, jobvr, sense, ld, A.values(), ld, B.values(), ld, &wr[0], &wi[0],
2258  &beta[0], vl, ldvl, vr.values(), ldvr, &ilo, &ihi, &lscale[0], &rscale[0],
2259  &abnrm, &bbnrm, &rconde[0], &rcondv[0], &work[0], lwork, &rwork[0],
2260  &iwork[0], bwork, &info);
2261  TEUCHOS_TEST_FOR_EXCEPTION(info != 0, GCRODRSolMgrLAPACKFailure, "Belos::GCRODRSolMgr::solve(): LAPACK GGEVX failed to compute eigensolutions.");
2262 
2263  // Construct magnitude of each harmonic Ritz value
2264  // NOTE : Forming alpha/beta *should* be okay here, given assumptions on construction of matrix pencil above
2265  for( i=0; i<ld; i++ ) {
2266  w[i] = Teuchos::ScalarTraits<MagnitudeType>::squareroot (wr[i]*wr[i] + wi[i]*wi[i]) /
2268  }
2269 
2270  // Construct magnitude of each harmonic Ritz value
2271  this->sort(w,ld,iperm);
2272 
2273  const bool scalarTypeIsComplex = Teuchos::ScalarTraits<ScalarType>::isComplex;
2274 
2275  // Select recycledBlocks_ smallest eigenvectors
2276  for( i=0; i<recycledBlocks_; i++ ) {
2277  for( j=0; j<ld; j++ ) {
2278  PP(j,i) = vr(j,iperm[ld-recycledBlocks_+i]);
2279  }
2280  }
2281 
2282  if(!scalarTypeIsComplex) {
2283 
2284  // Determine exact size for PP (i.e., determine if we need to store an additional vector)
2285  if (wi[iperm[ld-recycledBlocks_]] != 0.0) {
2286  int countImag = 0;
2287  for ( i=ld-recycledBlocks_; i<ld; i++ ) {
2288  if (wi[iperm[i]] != 0.0)
2289  countImag++;
2290  }
2291  // Check to see if this count is even or odd:
2292  if (countImag % 2)
2293  xtraVec = true;
2294  }
2295 
2296  if (xtraVec) { // we need to store one more vector
2297  if (wi[iperm[ld-recycledBlocks_]] > 0.0) { // I picked the "real" component
2298  for( j=0; j<ld; j++ ) { // so get the "imag" component
2299  PP(j,recycledBlocks_) = vr(j,iperm[ld-recycledBlocks_]+1);
2300  }
2301  }
2302  else { // I picked the "imag" component
2303  for( j=0; j<ld; j++ ) { // so get the "real" component
2304  PP(j,recycledBlocks_) = vr(j,iperm[ld-recycledBlocks_]-1);
2305  }
2306  }
2307  }
2308 
2309  }
2310 
2311  // Return whether we needed to store an additional vector
2312  if (xtraVec) {
2313  return recycledBlocks_+1;
2314  }
2315  else {
2316  return recycledBlocks_;
2317  }
2318 
2319 }
2320 
2321 
2322 // This method sorts list of n floating-point numbers and return permutation vector
2323 template<class ScalarType, class MV, class OP>
2324 void GCRODRSolMgr<ScalarType,MV,OP,true>::sort(std::vector<MagnitudeType>& dlist, int n, std::vector<int>& iperm) {
2325  int l, r, j, i, flag;
2326  int RR2;
2327  MagnitudeType dRR, dK;
2328 
2329  // Initialize the permutation vector.
2330  for(j=0;j<n;j++)
2331  iperm[j] = j;
2332 
2333  if (n <= 1) return;
2334 
2335  l = n / 2 + 1;
2336  r = n - 1;
2337  l = l - 1;
2338  dRR = dlist[l - 1];
2339  dK = dlist[l - 1];
2340 
2341  RR2 = iperm[l - 1];
2342  while (r != 0) {
2343  j = l;
2344  flag = 1;
2345 
2346  while (flag == 1) {
2347  i = j;
2348  j = j + j;
2349 
2350  if (j > r + 1)
2351  flag = 0;
2352  else {
2353  if (j < r + 1)
2354  if (dlist[j] > dlist[j - 1]) j = j + 1;
2355 
2356  if (dlist[j - 1] > dK) {
2357  dlist[i - 1] = dlist[j - 1];
2358  iperm[i - 1] = iperm[j - 1];
2359  }
2360  else {
2361  flag = 0;
2362  }
2363  }
2364  }
2365  dlist[i - 1] = dRR;
2366  iperm[i - 1] = RR2;
2367 
2368  if (l == 1) {
2369  dRR = dlist [r];
2370  RR2 = iperm[r];
2371  dK = dlist[r];
2372  dlist[r] = dlist[0];
2373  iperm[r] = iperm[0];
2374  r = r - 1;
2375  }
2376  else {
2377  l = l - 1;
2378  dRR = dlist[l - 1];
2379  RR2 = iperm[l - 1];
2380  dK = dlist[l - 1];
2381  }
2382  }
2383  dlist[0] = dRR;
2384  iperm[0] = RR2;
2385 }
2386 
2387 
2388 template<class ScalarType, class MV, class OP>
2390  std::ostringstream out;
2391  out << "Belos::GCRODRSolMgr<...,"<<Teuchos::ScalarTraits<ScalarType>::name()<<">";
2392  out << "{";
2393  out << "Ortho Type: \"" << orthoType_ << "\"";
2394  out << ", Num Blocks: " <<numBlocks_;
2395  out << ", Num Recycle Blocks: " << recycledBlocks_;
2396  out << ", Max Restarts: " << maxRestarts_;
2397  out << "}";
2398  return out.str ();
2399 }
2400 
2401 } // namespace Belos
2402 
2403 #endif /* BELOS_GCRODR_SOLMGR_HPP */
ScaleType convertStringToScaleType(const std::string &scaleType)
Convert the given string to its ScaleType enum value.
Definition: BelosTypes.cpp:74
ScalarType * values() const
Teuchos::RCP< StatusTestMaxIters< ScalarType, MV, OP > > maxIterTest_
Collection of types and exceptions used within the Belos solvers.
Teuchos::ScalarTraits< ScalarType >::magnitudeType MagnitudeType
Teuchos::ScalarTraits< MagnitudeType > MT
Partial specialization for ScalarType types for which Teuchos::LAPACK has a valid implementation...
Belos&#39;s basic output manager for sending information of select verbosity levels to the appropriate ou...
Teuchos::RCP< OutputManager< ScalarType > > printer_
GCRODRSolMgrLAPACKFailure(const std::string &what_arg)
Class which manages the output and verbosity of the Belos solvers.
bool is_null(const boost::shared_ptr< T > &p)
static const bool requiresLapack
ScaleType
The type of scaling to use on the residual norm value.
Definition: BelosTypes.hpp:88
Exception thrown to signal error in a status test during Belos::StatusTest::checkStatus().
T & get(ParameterList &l, const std::string &name)
GCRODRSolMgrLAPACKFailure is thrown when a nonzero value is retuned from an LAPACK call...
static RCP< Time > getNewCounter(const std::string &name)
bool is_null(const std::shared_ptr< T > &p)
int multiply(ETransp transa, ETransp transb, ScalarType alpha, const SerialDenseMatrix< OrdinalType, ScalarType > &A, const SerialDenseMatrix< OrdinalType, ScalarType > &B, ScalarType beta)
static T squareroot(T x)
#define TEUCHOS_TEST_FOR_EXCEPTION(throw_exception_test, Exception, msg)
A factory class for generating StatusTestOutput objects.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > H_
Teuchos::RCP< const Teuchos::ParameterList > getDefaultParameters(const std::string &name) const
Default parameters for the given MatOrthoManager subclass.
An implementation of StatusTestResNorm using a family of residual norms.
GCRODRSolMgrLinearProblemFailure is thrown when the linear problem is not setup (i.e.
GCRODRSolMgrRecyclingFailure(const std::string &what_arg)
This class implements the GCRODR iteration, where a single-std::vector Krylov subspace is constructed...
ParameterList & set(std::string const &name, T &&value, std::string const &docString="", RCP< const ParameterEntryValidator > const &validator=null)
static const double convTol
Default convergence tolerance.
Definition: BelosTypes.hpp:261
Belos::StatusTest class for specifying a maximum number of iterations.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > PP_
int curDim
The current dimension of the reduction.
static std::string name()
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > HP_
A factory class for generating StatusTestOutput objects.
std::string validNamesString() const
List (as a string) of recognized MatOrthoManager names.
Traits class which defines basic operations on multivectors.
Belos::StatusTest for logically combining several status tests.
bool isParameter(const std::string &name) const
A Belos::StatusTest class for specifying a maximum number of iterations.
GCRODRSolMgrOrthoFailure(const std::string &what_arg)
Teuchos::RCP< SolverManager< ScalarType, MV, OP > > clone() const override
clone for Inverted Injection (DII)
ResetType
How to reset the solver.
Definition: BelosTypes.hpp:174
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 > V
The current Krylov basis.
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)
bool is_null(const RCP< T > &p)
bool isValidName(const std::string &name) const
Whether this factory recognizes the MatOrthoManager with the given name.
Teuchos::RCP< StatusTest< ScalarType, MV, OP > > convTest_
Teuchos::RCP< MV > U
The recycled subspace and its projection.
std::ostream & printValidNames(std::ostream &out) const
Print all recognized MatOrthoManager names to the given ostream.
int putScalar(const ScalarType value=Teuchos::ScalarTraits< ScalarType >::zero())
Teuchos::RCP< StatusTestOutput< ScalarType, MV, OP > > outputTest_
A linear system to solve, and its associated information.
GCRODRSolMgrOrthoFailure is thrown when the orthogonalization manager is unable to generate orthonorm...
Class which describes the linear problem to be solved by the iterative solver.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > H2_
GCRODRIterOrthoFailure is thrown when the GCRODRIter object is unable to compute independent directio...
Implementation of the GCRODR (Recycling GMRES) iterative linear solver.
Type traits class that says whether Teuchos::LAPACK has a valid implementation for the given ScalarTy...
ReturnType
Whether the Belos solve converged for all linear systems.
Definition: BelosTypes.hpp:123
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > B_
Belos concrete class for performing the GCRO-DR iteration.
Structure to contain pointers to GCRODRIter state variables.
MagnitudeType achievedTol() const override
Tolerance achieved by the last solve() invocation.
Teuchos::Array< Teuchos::RCP< Teuchos::Time > > getTimers() const
Return the timers for this object.
void reset(const ResetType type) override
Performs a reset of the solver manager specified by the ResetType. This informs the solver manager th...
void setProblem(const Teuchos::RCP< LinearProblem< ScalarType, MV, OP > > &problem) override
Set the linear problem that needs to be solved.
GCRODRSolMgrRecyclingFailure is thrown when any problem occurs in using/creating the recycling subspa...
Teuchos::RCP< StatusTestGenResNorm< ScalarType, MV, OP > > impConvTest_
Teuchos::RCP< LinearProblem< ScalarType, MV, OP > > problem_
static magnitudeType magnitude(T a)
OrdinalType numCols() const
GCRODRSolMgrLinearProblemFailure(const std::string &what_arg)
GCRODRSolMgr(const Teuchos::RCP< LinearProblem< ScalarType, MV, OP > > &problem, const Teuchos::RCP< Teuchos::ParameterList > &pl)
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.
Belos::StatusTestResNorm for specifying general residual norm stopping criteria.
Teuchos::RCP< const Teuchos::ParameterList > getCurrentParameters() const override
Get a parameter list containing the current parameters for this object.
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > B
The projection of the Krylov subspace against the recycled subspace.
OrthoManagerFactory< ScalarType, MV, OP > ortho_factory_type
bool isType(const std::string &name) const
An implementation of the Belos::MatOrthoManager that performs orthogonalization using (potentially) m...
const LinearProblem< ScalarType, MV, OP > & getProblem() const override
Get current linear problem being solved for in this object.
A class for extending the status testing capabilities of Belos via logical combinations.
int getNumIters() const override
Get the iteration count for the most recent call to solve().
Details::SolverManagerRequiresLapack< ScalarType, MV, OP, requiresLapack > base_type
Class which defines basic traits for the operator type.
OperatorTraits< ScalarType, MV, OP > OPT
Teuchos::RCP< MV > C
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > H
The current Hessenberg matrix.
Parent class to all Belos exceptions.
Definition: BelosTypes.hpp:28
Teuchos::RCP< StatusTest< ScalarType, MV, OP > > sTest_
Base class for Belos::SolverManager subclasses which normally can only compile with ScalarType types ...
Teuchos::RCP< MatOrthoManager< ScalarType, MV, OP > > ortho_
Orthogonalization manager.
Belos header file which uses auto-configuration information to include necessary C++ headers...
bool isLOADetected() const override
Return whether a loss of accuracy was detected by this solver during the most current solve...
Teuchos::RCP< Teuchos::ParameterList > params_
Teuchos::RCP< Teuchos::SerialDenseMatrix< int, ScalarType > > R_
OrdinalType stride() const
OrdinalType numRows() const
Belos concrete class for performing the block, flexible GMRES iteration.
Teuchos::RCP< Belos::MatOrthoManager< Scalar, MV, OP > > makeMatOrthoManager(const std::string &ortho, const Teuchos::RCP< const OP > &M, const Teuchos::RCP< OutputManager< Scalar > > &, const std::string &label, const Teuchos::RCP< Teuchos::ParameterList > &params)
Return an instance of the specified MatOrthoManager subclass.