Anasazi Version of the Day
Loading...
Searching...
No Matches
AnasaziLOBPCG.hpp
Go to the documentation of this file.
1// @HEADER
2// *****************************************************************************
3// Anasazi: Block Eigensolvers Package
4//
5// Copyright 2004 NTESS and the Anasazi contributors.
6// SPDX-License-Identifier: BSD-3-Clause
7// *****************************************************************************
8// @HEADER
9
10
15/*
16 LOBPCG contains local storage of up to 10*blockSize_ vectors, representing 10 entities
17 X,H,P,R
18 KX,KH,KP (product of K and the above)
19 MX,MH,MP (product of M and the above, not allocated if we don't have an M matrix)
20 If full orthogonalization is enabled, one extra multivector of blockSize_ vectors is required to
21 compute the local update of X and P.
22
23 A solver is bound to an eigenproblem at declaration.
24 Other solver parameters (e.g., block size, auxiliary vectors) can be changed dynamically.
25
26 The orthogonalization manager is used to project away from the auxiliary vectors.
27 If full orthogonalization is enabled, the orthogonalization manager is also used to construct an M orthonormal basis.
28 The orthogonalization manager is subclass of MatOrthoManager, which LOBPCG assumes to be defined by the M inner product.
29 LOBPCG will not work correctly if the orthomanager uses a different inner product.
30 */
31
32
33#ifndef ANASAZI_LOBPCG_HPP
34#define ANASAZI_LOBPCG_HPP
35
36#include "AnasaziTypes.hpp"
37
38#include "AnasaziEigensolver.hpp"
39#include "AnasaziMultiVecTraits.hpp"
40#include "AnasaziOperatorTraits.hpp"
41#include "Teuchos_ScalarTraits.hpp"
42
43#include "AnasaziMatOrthoManager.hpp"
44#include "AnasaziSolverUtils.hpp"
45
46#include "Teuchos_LAPACK.hpp"
47#include "Teuchos_BLAS.hpp"
48#include "Teuchos_SerialDenseMatrix.hpp"
49#include "Teuchos_ParameterList.hpp"
50#include "Teuchos_TimeMonitor.hpp"
51
72namespace Anasazi {
73
75
76
81 template <class ScalarType, class MultiVector>
82 struct LOBPCGState {
84 Teuchos::RCP<const MultiVector> V;
86 Teuchos::RCP<const MultiVector> KV;
88 Teuchos::RCP<const MultiVector> MV;
89
91 Teuchos::RCP<const MultiVector> X;
93 Teuchos::RCP<const MultiVector> KX;
95 Teuchos::RCP<const MultiVector> MX;
96
98 Teuchos::RCP<const MultiVector> P;
100 Teuchos::RCP<const MultiVector> KP;
102 Teuchos::RCP<const MultiVector> MP;
103
108 Teuchos::RCP<const MultiVector> H;
110 Teuchos::RCP<const MultiVector> KH;
112 Teuchos::RCP<const MultiVector> MH;
113
115 Teuchos::RCP<const MultiVector> R;
116
118 Teuchos::RCP<const std::vector<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType> > T;
119
120 LOBPCGState() :
121 V(Teuchos::null),KV(Teuchos::null),MV(Teuchos::null),
122 X(Teuchos::null),KX(Teuchos::null),MX(Teuchos::null),
123 P(Teuchos::null),KP(Teuchos::null),MP(Teuchos::null),
124 H(Teuchos::null),KH(Teuchos::null),MH(Teuchos::null),
125 R(Teuchos::null),T(Teuchos::null) {};
126 };
127
129
131
132
146 class LOBPCGRitzFailure : public AnasaziError {public:
147 LOBPCGRitzFailure(const std::string& what_arg) : AnasaziError(what_arg)
148 {}};
149
162 class LOBPCGInitFailure : public AnasaziError {public:
163 LOBPCGInitFailure(const std::string& what_arg) : AnasaziError(what_arg)
164 {}};
165
182 class LOBPCGOrthoFailure : public AnasaziError {public:
183 LOBPCGOrthoFailure(const std::string& what_arg) : AnasaziError(what_arg)
184 {}};
185
187
188
189 template <class ScalarType, class MV, class OP>
190 class LOBPCG : public Eigensolver<ScalarType,MV,OP> {
191 public:
192
194
195
203 LOBPCG( const Teuchos::RCP<Eigenproblem<ScalarType,MV,OP> > &problem,
204 const Teuchos::RCP<SortManager<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType> > &sorter,
205 const Teuchos::RCP<OutputManager<ScalarType> > &printer,
206 const Teuchos::RCP<StatusTest<ScalarType,MV,OP> > &tester,
207 const Teuchos::RCP<MatOrthoManager<ScalarType,MV,OP> > &ortho,
208 Teuchos::ParameterList &params
209 );
210
212 virtual ~LOBPCG() {};
213
215
217
218
245 void iterate();
246
272 void initialize(LOBPCGState<ScalarType,MV>& newstate);
273
277 void initialize();
278
298 bool isInitialized() const;
299
310 LOBPCGState<ScalarType,MV> getState() const;
311
313
315
316
318 int getNumIters() const;
319
321 void resetNumIters();
322
330 Teuchos::RCP<const MV> getRitzVectors();
331
337 std::vector<Value<ScalarType> > getRitzValues();
338
346 std::vector<int> getRitzIndex();
347
348
354 std::vector<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType> getResNorms();
355
356
362 std::vector<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType> getRes2Norms();
363
364
372 std::vector<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType> getRitzRes2Norms();
373
374
383 int getCurSubspaceDim() const;
384
388 int getMaxSubspaceDim() const;
389
391
393
394
396 void setStatusTest(Teuchos::RCP<StatusTest<ScalarType,MV,OP> > test);
397
399 Teuchos::RCP<StatusTest<ScalarType,MV,OP> > getStatusTest() const;
400
402 const Eigenproblem<ScalarType,MV,OP>& getProblem() const;
403
404
413 void setBlockSize(int blockSize);
414
415
417 int getBlockSize() const;
418
419
431 void setAuxVecs(const Teuchos::Array<Teuchos::RCP<const MV> > &auxvecs);
432
434 Teuchos::Array<Teuchos::RCP<const MV> > getAuxVecs() const;
435
437
439
440
446 void setFullOrtho(bool fullOrtho);
447
449 bool getFullOrtho() const;
450
452 bool hasP();
453
455
457
458
460 void currentStatus(std::ostream &os);
461
463
464 private:
465 //
466 //
467 //
468 void setupViews();
469 //
470 // Convenience typedefs
471 //
472 typedef SolverUtils<ScalarType,MV,OP> Utils;
473 typedef MultiVecTraits<ScalarType,MV> MVT;
474 typedef OperatorTraits<ScalarType,MV,OP> OPT;
475 typedef Teuchos::ScalarTraits<ScalarType> SCT;
476 typedef typename SCT::magnitudeType MagnitudeType;
477 const MagnitudeType ONE;
478 const MagnitudeType ZERO;
479 const MagnitudeType NANVAL;
480 //
481 // Internal structs
482 //
483 struct CheckList {
484 bool checkX, checkMX, checkKX;
485 bool checkH, checkMH;
486 bool checkP, checkMP, checkKP;
487 bool checkR, checkQ;
488 CheckList() : checkX(false),checkMX(false),checkKX(false),
489 checkH(false),checkMH(false),
490 checkP(false),checkMP(false),checkKP(false),
491 checkR(false),checkQ(false) {};
492 };
493 //
494 // Internal methods
495 //
496 std::string accuracyCheck(const CheckList &chk, const std::string &where) const;
497 //
498 // Classes inputed through constructor that define the eigenproblem to be solved.
499 //
500 const Teuchos::RCP<Eigenproblem<ScalarType,MV,OP> > problem_;
501 const Teuchos::RCP<SortManager<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType> > sm_;
502 const Teuchos::RCP<OutputManager<ScalarType> > om_;
503 Teuchos::RCP<StatusTest<ScalarType,MV,OP> > tester_;
504 const Teuchos::RCP<MatOrthoManager<ScalarType,MV,OP> > orthman_;
505 //
506 // Information obtained from the eigenproblem
507 //
508 Teuchos::RCP<const OP> Op_;
509 Teuchos::RCP<const OP> MOp_;
510 Teuchos::RCP<const OP> Prec_;
511 bool hasM_;
512 //
513 // Internal timers
514 //
515 Teuchos::RCP<Teuchos::Time> timerOp_, timerMOp_, timerPrec_,
516 timerSort_,
517 timerLocalProj_, timerDS_,
518 timerLocalUpdate_, timerCompRes_,
519 timerOrtho_, timerInit_;
520 //
521 // Counters
522 //
523 // Number of operator applications
524 int count_ApplyOp_, count_ApplyM_, count_ApplyPrec_;
525
526 //
527 // Algorithmic parameters.
528 //
529 // blockSize_ is the solver block size
530 int blockSize_;
531 //
532 // fullOrtho_ dictates whether the orthogonalization procedures specified by Hetmaniuk and Lehoucq should
533 // be activated (see citations at the top of this file)
534 bool fullOrtho_;
535
536 //
537 // Current solver state
538 //
539 // initialized_ specifies that the basis vectors have been initialized and the iterate() routine
540 // is capable of running; _initialize is controlled by the initialize() member method
541 // For the implications of the state of initialized_, please see documentation for initialize()
542 bool initialized_;
543 //
544 // nevLocal_ reflects how much of the current basis is valid (0 <= nevLocal_ <= 3*blockSize_)
545 // this tells us how many of the values in theta_ are valid Ritz values
546 int nevLocal_;
547 //
548 // hasP_ tells us whether there is valid data in P (and KP,MP)
549 bool hasP_;
550 //
551 // State Multivecs
552 // V_, KV_ MV_ and R_ are primary pointers to allocated multivectors
553 Teuchos::RCP<MV> V_, KV_, MV_, R_;
554 // the rest are multivector views into V_, KV_ and MV_
555 Teuchos::RCP<MV> X_, KX_, MX_,
556 H_, KH_, MH_,
557 P_, KP_, MP_;
558
559 //
560 // if fullOrtho_ == true, then we must produce the following on every iteration:
561 // [newX newP] = [X H P] [CX;CP]
562 // the structure of [CX;CP] when using full orthogonalization does not allow us to
563 // do this in situ, and R_ does not have enough storage for newX and newP. therefore,
564 // we must allocate additional storage for this.
565 // otherwise, when not using full orthogonalization, the structure
566 // [newX newP] = [X H P] [CX1 0 ]
567 // [CX2 CP2] allows us to work using only R as work space
568 // [CX3 CP3]
569 Teuchos::RCP<MV> tmpmvec_;
570 //
571 // auxiliary vectors
572 Teuchos::Array<Teuchos::RCP<const MV> > auxVecs_;
573 int numAuxVecs_;
574 //
575 // Number of iterations that have been performed.
576 int iter_;
577 //
578 // Current eigenvalues, residual norms
579 std::vector<MagnitudeType> theta_, Rnorms_, R2norms_;
580 //
581 // are the residual norms current with the residual?
582 bool Rnorms_current_, R2norms_current_;
583
584 };
585
586
587
588
590 // Constructor
591 template <class ScalarType, class MV, class OP>
592 LOBPCG<ScalarType,MV,OP>::LOBPCG(
593 const Teuchos::RCP<Eigenproblem<ScalarType,MV,OP> > &problem,
594 const Teuchos::RCP<SortManager<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType> > &sorter,
595 const Teuchos::RCP<OutputManager<ScalarType> > &printer,
596 const Teuchos::RCP<StatusTest<ScalarType,MV,OP> > &tester,
597 const Teuchos::RCP<MatOrthoManager<ScalarType,MV,OP> > &ortho,
598 Teuchos::ParameterList &params
599 ) :
600 ONE(Teuchos::ScalarTraits<MagnitudeType>::one()),
601 ZERO(Teuchos::ScalarTraits<MagnitudeType>::zero()),
602 NANVAL(Teuchos::ScalarTraits<MagnitudeType>::nan()),
603 // problem, tools
604 problem_(problem),
605 sm_(sorter),
606 om_(printer),
607 tester_(tester),
608 orthman_(ortho),
609 // timers, counters
610#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
611 timerOp_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Operation Op*x")),
612 timerMOp_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Operation M*x")),
613 timerPrec_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Operation Prec*x")),
614 timerSort_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Sorting eigenvalues")),
615 timerLocalProj_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Local projection")),
616 timerDS_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Direct solve")),
617 timerLocalUpdate_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Local update")),
618 timerCompRes_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Computing residuals")),
619 timerOrtho_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Orthogonalization")),
620 timerInit_(Teuchos::TimeMonitor::getNewTimer("Anasazi: LOBPCG::Initialization")),
621#endif
622 count_ApplyOp_(0),
623 count_ApplyM_(0),
624 count_ApplyPrec_(0),
625 // internal data
626 blockSize_(0),
627 fullOrtho_(params.get("Full Ortho", true)),
628 initialized_(false),
629 nevLocal_(0),
630 hasP_(false),
631 auxVecs_( Teuchos::Array<Teuchos::RCP<const MV> >(0) ),
632 numAuxVecs_(0),
633 iter_(0),
634 Rnorms_current_(false),
635 R2norms_current_(false)
636 {
637 TEUCHOS_TEST_FOR_EXCEPTION(problem_ == Teuchos::null,std::invalid_argument,
638 "Anasazi::LOBPCG::constructor: user passed null problem pointer.");
639 TEUCHOS_TEST_FOR_EXCEPTION(sm_ == Teuchos::null,std::invalid_argument,
640 "Anasazi::LOBPCG::constructor: user passed null sort manager pointer.");
641 TEUCHOS_TEST_FOR_EXCEPTION(om_ == Teuchos::null,std::invalid_argument,
642 "Anasazi::LOBPCG::constructor: user passed null output manager pointer.");
643 TEUCHOS_TEST_FOR_EXCEPTION(tester_ == Teuchos::null,std::invalid_argument,
644 "Anasazi::LOBPCG::constructor: user passed null status test pointer.");
645 TEUCHOS_TEST_FOR_EXCEPTION(orthman_ == Teuchos::null,std::invalid_argument,
646 "Anasazi::LOBPCG::constructor: user passed null orthogonalization manager pointer.");
647 TEUCHOS_TEST_FOR_EXCEPTION(problem_->isProblemSet() == false, std::invalid_argument,
648 "Anasazi::LOBPCG::constructor: problem is not set.");
649 TEUCHOS_TEST_FOR_EXCEPTION(problem_->isHermitian() == false, std::invalid_argument,
650 "Anasazi::LOBPCG::constructor: problem is not Hermitian; LOBPCG requires Hermitian problem.");
651
652 // get the problem operators
653 Op_ = problem_->getOperator();
654 TEUCHOS_TEST_FOR_EXCEPTION(Op_ == Teuchos::null, std::invalid_argument,
655 "Anasazi::LOBPCG::constructor: problem provides no operator.");
656 MOp_ = problem_->getM();
657 Prec_ = problem_->getPrec();
658 hasM_ = (MOp_ != Teuchos::null);
659
660 // set the block size and allocate data
661 int bs = params.get("Block Size", problem_->getNEV());
662 setBlockSize(bs);
663 }
664
665
667 // Set the block size and make necessary adjustments.
668 template <class ScalarType, class MV, class OP>
669 void LOBPCG<ScalarType,MV,OP>::setBlockSize (int newBS)
670 {
671 // time spent here counts towards timerInit_
672#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
673 Teuchos::TimeMonitor lcltimer( *timerInit_ );
674#endif
675
676 // This routine only allocates space; it doesn't not perform any computation
677 // if size is decreased, take the first newBS vectors of all and leave state as is
678 // otherwise, grow/allocate space and set solver to unitialized
679
680 Teuchos::RCP<const MV> tmp;
681 // grab some Multivector to Clone
682 // in practice, getInitVec() should always provide this, but it is possible to use a
683 // Eigenproblem with nothing in getInitVec() by manually initializing with initialize();
684 // in case of that strange scenario, we will try to Clone from R_ because it is smaller
685 // than V_, and we don't want to keep V_ around longer than necessary
686 if (blockSize_ > 0) {
687 tmp = R_;
688 }
689 else {
690 tmp = problem_->getInitVec();
691 TEUCHOS_TEST_FOR_EXCEPTION(tmp == Teuchos::null,std::logic_error,
692 "Anasazi::LOBPCG::setBlockSize(): eigenproblem did not specify initial vectors to clone from.");
693 }
694
695 TEUCHOS_TEST_FOR_EXCEPTION(newBS <= 0 || static_cast<ptrdiff_t>(newBS) > MVT::GetGlobalLength(*tmp),
696 std::invalid_argument, "Anasazi::LOBPCG::setBlockSize(): block size must be strictly positive.");
697 if (newBS == blockSize_) {
698 // do nothing
699 return;
700 }
701 else if (newBS < blockSize_ && initialized_) {
702 //
703 // shrink vectors
704
705 // release views so we can modify the bases
706 X_ = Teuchos::null;
707 KX_ = Teuchos::null;
708 MX_ = Teuchos::null;
709 H_ = Teuchos::null;
710 KH_ = Teuchos::null;
711 MH_ = Teuchos::null;
712 P_ = Teuchos::null;
713 KP_ = Teuchos::null;
714 MP_ = Teuchos::null;
715
716 // make new indices vectors
717 std::vector<int> newind(newBS), oldind(newBS);
718 for (int i=0; i<newBS; i++) {
719 newind[i] = i;
720 oldind[i] = i;
721 }
722
723 Teuchos::RCP<MV> newV, newMV, newKV, newR;
724 Teuchos::RCP<const MV> src;
725 // allocate R and newV
726 newR = MVT::Clone(*tmp,newBS);
727 newV = MVT::Clone(*tmp,newBS*3);
728 newKV = MVT::Clone(*tmp,newBS*3);
729 if (hasM_) {
730 newMV = MVT::Clone(*tmp,newBS*3);
731 }
732
733 //
734 // if we are initialized, we want to pull the data from V_ into newV:
735 // bs | bs | bs
736 // newV = [newX | **** |newP ]
737 // newKV = [newKX| **** |newKP]
738 // newMV = [newMX| **** |newMP]
739 // where
740 // oldbs | oldbs | oldbs
741 // V_ = [newX *** | ******* | newP ***]
742 // KV_ = [newKX *** | ******* | newKP ***]
743 // MV_ = [newMX *** | ******* | newMP ***]
744 //
745 // we don't care to copy the data corresponding to H
746 // we will not copy the M data if !hasM_, because it doesn't exist
747 //
748
749 // these are shrink operations which preserve their data
750 theta_.resize(3*newBS);
751 Rnorms_.resize(newBS);
752 R2norms_.resize(newBS);
753
754 // copy residual vectors: oldind,newind currently contains [0,...,newBS-1]
755 src = MVT::CloneView(*R_,newind);
756 MVT::SetBlock(*src,newind,*newR);
757 // free old memory and point to new memory
758 R_ = newR;
759
760 // copy in order: newX newKX newMX, then newP newKP newMP
761 // for X: [0,bs-1] <-- [0,bs-1]
762 src = MVT::CloneView(*V_,oldind);
763 MVT::SetBlock(*src,newind,*newV);
764 src = MVT::CloneView(*KV_,oldind);
765 MVT::SetBlock(*src,newind,*newKV);
766 if (hasM_) {
767 src = MVT::CloneView(*MV_,oldind);
768 MVT::SetBlock(*src,newind,*newMV);
769 }
770 // for P: [2*bs, 3*bs-1] <-- [2*oldbs, 2*oldbs+bs-1]
771 for (int i=0; i<newBS; i++) {
772 newind[i] += 2*newBS;
773 oldind[i] += 2*blockSize_;
774 }
775 src = MVT::CloneView(*V_,oldind);
776 MVT::SetBlock(*src,newind,*newV);
777 src = MVT::CloneView(*KV_,oldind);
778 MVT::SetBlock(*src,newind,*newKV);
779 if (hasM_) {
780 src = MVT::CloneView(*MV_,oldind);
781 MVT::SetBlock(*src,newind,*newMV);
782 }
783
784 // release temp view
785 src = Teuchos::null;
786
787 // release old allocations and point at new memory
788 V_ = newV;
789 KV_ = newKV;
790 if (hasM_) {
791 MV_ = newMV;
792 }
793 else {
794 MV_ = V_;
795 }
796 }
797 else {
798 // newBS > blockSize_ or not initialized
799 // this is also the scenario for our initial call to setBlockSize(), in the constructor
800 initialized_ = false;
801 hasP_ = false;
802
803 // release views
804 X_ = Teuchos::null;
805 KX_ = Teuchos::null;
806 MX_ = Teuchos::null;
807 H_ = Teuchos::null;
808 KH_ = Teuchos::null;
809 MH_ = Teuchos::null;
810 P_ = Teuchos::null;
811 KP_ = Teuchos::null;
812 MP_ = Teuchos::null;
813
814 // free allocated storage
815 R_ = Teuchos::null;
816 V_ = Teuchos::null;
817
818 // allocate scalar vectors
819 theta_.resize(3*newBS,NANVAL);
820 Rnorms_.resize(newBS,NANVAL);
821 R2norms_.resize(newBS,NANVAL);
822
823 // clone multivectors off of tmp
824 R_ = MVT::Clone(*tmp,newBS);
825 V_ = MVT::Clone(*tmp,newBS*3);
826 KV_ = MVT::Clone(*tmp,newBS*3);
827 if (hasM_) {
828 MV_ = MVT::Clone(*tmp,newBS*3);
829 }
830 else {
831 MV_ = V_;
832 }
833 }
834
835 // allocate tmp space
836 tmpmvec_ = Teuchos::null;
837 if (fullOrtho_) {
838 tmpmvec_ = MVT::Clone(*tmp,newBS);
839 }
840
841 // set new block size
842 blockSize_ = newBS;
843
844 // setup new views
845 setupViews();
846 }
847
848
850 // Setup views into V,KV,MV
851 template <class ScalarType, class MV, class OP>
852 void LOBPCG<ScalarType,MV,OP>::setupViews()
853 {
854 std::vector<int> ind(blockSize_);
855
856 for (int i=0; i<blockSize_; i++) {
857 ind[i] = i;
858 }
859 X_ = MVT::CloneViewNonConst(*V_,ind);
860 KX_ = MVT::CloneViewNonConst(*KV_,ind);
861 if (hasM_) {
862 MX_ = MVT::CloneViewNonConst(*MV_,ind);
863 }
864 else {
865 MX_ = X_;
866 }
867
868 for (int i=0; i<blockSize_; i++) {
869 ind[i] += blockSize_;
870 }
871 H_ = MVT::CloneViewNonConst(*V_,ind);
872 KH_ = MVT::CloneViewNonConst(*KV_,ind);
873 if (hasM_) {
874 MH_ = MVT::CloneViewNonConst(*MV_,ind);
875 }
876 else {
877 MH_ = H_;
878 }
879
880 for (int i=0; i<blockSize_; i++) {
881 ind[i] += blockSize_;
882 }
883 P_ = MVT::CloneViewNonConst(*V_,ind);
884 KP_ = MVT::CloneViewNonConst(*KV_,ind);
885 if (hasM_) {
886 MP_ = MVT::CloneViewNonConst(*MV_,ind);
887 }
888 else {
889 MP_ = P_;
890 }
891 }
892
893
895 // Set the auxiliary vectors
896 template <class ScalarType, class MV, class OP>
897 void LOBPCG<ScalarType,MV,OP>::setAuxVecs(const Teuchos::Array<Teuchos::RCP<const MV> > &auxvecs) {
898 typedef typename Teuchos::Array<Teuchos::RCP<const MV> >::iterator tarcpmv;
899
900 // set new auxiliary vectors
901 auxVecs_ = auxvecs;
902
903 numAuxVecs_ = 0;
904 for (tarcpmv i=auxVecs_.begin(); i != auxVecs_.end(); i++) {
905 numAuxVecs_ += MVT::GetNumberVecs(**i);
906 }
907
908 // If the solver has been initialized, X and P are not necessarily orthogonal to new auxiliary vectors
909 if (numAuxVecs_ > 0 && initialized_) {
910 initialized_ = false;
911 hasP_ = false;
912 }
913
914 if (om_->isVerbosity( Debug ) ) {
915 // Check almost everything here
916 CheckList chk;
917 chk.checkQ = true;
918 om_->print( Debug, accuracyCheck(chk, ": in setAuxVecs()") );
919 }
920 }
921
922
924 /* Initialize the state of the solver
925 *
926 * POST-CONDITIONS:
927 *
928 * initialized_ == true
929 * X is orthonormal, orthogonal to auxVecs_
930 * KX = Op*X
931 * MX = M*X if hasM_
932 * theta_ contains Ritz values of X
933 * R = KX - MX*diag(theta_)
934 * if hasP() == true,
935 * P orthogonal to auxVecs_
936 * if fullOrtho_ == true,
937 * P orthonormal and orthogonal to X
938 * KP = Op*P
939 * MP = M*P
940 */
941 template <class ScalarType, class MV, class OP>
942 void LOBPCG<ScalarType,MV,OP>::initialize(LOBPCGState<ScalarType,MV>& newstate)
943 {
944 // NOTE: memory has been allocated by setBlockSize(). Use SetBlock below; do not Clone
945 // NOTE: Overall time spent in this routine is counted to timerInit_; portions will also be counted towards other primitives
946
947#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
948 Teuchos::TimeMonitor inittimer( *timerInit_ );
949#endif
950
951 std::vector<int> bsind(blockSize_);
952 for (int i=0; i<blockSize_; i++) bsind[i] = i;
953
954 // in LOBPCG, X (the subspace iterate) is primary
955 // the order of dependence follows like so.
956 // --init-> X
957 // --op apply-> MX,KX
958 // --ritz analysis-> theta
959 // --optional-> P,MP,KP
960 //
961 // if the user specifies all data for a level, we will accept it.
962 // otherwise, we will generate the whole level, and all subsequent levels.
963 //
964 // the data members are ordered based on dependence, and the levels are
965 // partitioned according to the amount of work required to produce the
966 // items in a level.
967 //
968 // inconsitent multivectors widths and lengths will not be tolerated, and
969 // will be treated with exceptions.
970
971 // set up X, KX, MX: get them from "state" if user specified them
972
973 //----------------------------------------
974 // set up X, MX, KX
975 //----------------------------------------
976 if (newstate.X != Teuchos::null) {
977 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength(*newstate.X) != MVT::GetGlobalLength(*X_),
978 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): vector length of newstate.X not correct." );
979 // newstate.X must have blockSize_ vectors; any more will be ignored
980 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetNumberVecs(*newstate.X) < blockSize_,
981 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): newstate.X must have at least block size vectors.");
982
983 // put X data in X_
984 MVT::SetBlock(*newstate.X,bsind,*X_);
985
986 // put MX data in MX_
987 if (hasM_) {
988 if (newstate.MX != Teuchos::null) {
989 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength(*newstate.MX) != MVT::GetGlobalLength(*MX_),
990 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): vector length of newstate.MX not correct." );
991 // newstate.MX must have blockSize_ vectors; any more will be ignored
992 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetNumberVecs(*newstate.MX) < blockSize_,
993 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): newstate.MX must have at least block size vectors.");
994 MVT::SetBlock(*newstate.MX,bsind,*MX_);
995 }
996 else {
997 // user didn't specify MX, compute it
998 {
999#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1000 Teuchos::TimeMonitor lcltimer( *timerMOp_ );
1001#endif
1002 OPT::Apply(*MOp_,*X_,*MX_);
1003 count_ApplyM_ += blockSize_;
1004 }
1005 // we generated MX; we will generate R as well
1006 newstate.R = Teuchos::null;
1007 }
1008 }
1009
1010 // put data in KX
1011 if (newstate.KX != Teuchos::null) {
1012 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength(*newstate.KX) != MVT::GetGlobalLength(*KX_),
1013 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): vector length of newstate.KX not correct." );
1014 // newstate.KX must have blockSize_ vectors; any more will be ignored
1015 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetNumberVecs(*newstate.KX) < blockSize_,
1016 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): newstate.KX must have at least block size vectors.");
1017 MVT::SetBlock(*newstate.KX,bsind,*KX_);
1018 }
1019 else {
1020 // user didn't specify KX, compute it
1021 {
1022#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1023 Teuchos::TimeMonitor lcltimer( *timerOp_ );
1024#endif
1025 OPT::Apply(*Op_,*X_,*KX_);
1026 count_ApplyOp_ += blockSize_;
1027 }
1028 // we generated KX; we will generate R as well
1029 newstate.R = Teuchos::null;
1030 }
1031 }
1032 else {
1033 // user did not specify X
1034 // we will initialize X, compute KX and MX, and compute R
1035 //
1036 // clear state so we won't use any data from it below
1037 newstate.P = Teuchos::null;
1038 newstate.KP = Teuchos::null;
1039 newstate.MP = Teuchos::null;
1040 newstate.R = Teuchos::null;
1041 newstate.T = Teuchos::null;
1042
1043 // generate a basis and projectAndNormalize
1044 Teuchos::RCP<const MV> ivec = problem_->getInitVec();
1045 TEUCHOS_TEST_FOR_EXCEPTION(ivec == Teuchos::null,std::logic_error,
1046 "Anasazi::LOBPCG::initialize(): Eigenproblem did not specify initial vectors to clone from.");
1047
1048 int initSize = MVT::GetNumberVecs(*ivec);
1049 if (initSize > blockSize_) {
1050 // we need only the first blockSize_ vectors from ivec; get a view of them
1051 initSize = blockSize_;
1052 std::vector<int> ind(blockSize_);
1053 for (int i=0; i<blockSize_; i++) ind[i] = i;
1054 ivec = MVT::CloneView(*ivec,ind);
1055 }
1056
1057 // assign ivec to first part of X_
1058 if (initSize > 0) {
1059 std::vector<int> ind(initSize);
1060 for (int i=0; i<initSize; i++) ind[i] = i;
1061 MVT::SetBlock(*ivec,ind,*X_);
1062 }
1063 // fill the rest of X_ with random
1064 if (blockSize_ > initSize) {
1065 std::vector<int> ind(blockSize_ - initSize);
1066 for (int i=0; i<blockSize_ - initSize; i++) ind[i] = initSize + i;
1067 Teuchos::RCP<MV> rX = MVT::CloneViewNonConst(*X_,ind);
1068 MVT::MvRandom(*rX);
1069 rX = Teuchos::null;
1070 }
1071
1072 // put data in MX
1073 if (hasM_) {
1074#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1075 Teuchos::TimeMonitor lcltimer( *timerMOp_ );
1076#endif
1077 OPT::Apply(*MOp_,*X_,*MX_);
1078 count_ApplyM_ += blockSize_;
1079 }
1080
1081 // remove auxVecs from X_ and normalize it
1082 if (numAuxVecs_ > 0) {
1083#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1084 Teuchos::TimeMonitor lcltimer( *timerOrtho_ );
1085#endif
1086 Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > dummy;
1087 int rank = orthman_->projectAndNormalizeMat(*X_,auxVecs_,dummy,Teuchos::null,MX_);
1088 TEUCHOS_TEST_FOR_EXCEPTION(rank != blockSize_, LOBPCGInitFailure,
1089 "Anasazi::LOBPCG::initialize(): Couldn't generate initial basis of full rank.");
1090 }
1091 else {
1092#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1093 Teuchos::TimeMonitor lcltimer( *timerOrtho_ );
1094#endif
1095 int rank = orthman_->normalizeMat(*X_,Teuchos::null,MX_);
1096 TEUCHOS_TEST_FOR_EXCEPTION(rank != blockSize_, LOBPCGInitFailure,
1097 "Anasazi::LOBPCG::initialize(): Couldn't generate initial basis of full rank.");
1098 }
1099
1100 // put data in KX
1101 {
1102#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1103 Teuchos::TimeMonitor lcltimer( *timerOp_ );
1104#endif
1105 OPT::Apply(*Op_,*X_,*KX_);
1106 count_ApplyOp_ += blockSize_;
1107 }
1108 } // end if (newstate.X != Teuchos::null)
1109
1110
1111 //----------------------------------------
1112 // set up Ritz values
1113 //----------------------------------------
1114 theta_.resize(3*blockSize_,NANVAL);
1115 if (newstate.T != Teuchos::null) {
1116 TEUCHOS_TEST_FOR_EXCEPTION( (signed int)(newstate.T->size()) < blockSize_,
1117 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): newstate.T must contain at least block size Ritz values.");
1118 for (int i=0; i<blockSize_; i++) {
1119 theta_[i] = (*newstate.T)[i];
1120 }
1121 nevLocal_ = blockSize_;
1122 }
1123 else {
1124 // get ritz vecs/vals
1125 Teuchos::SerialDenseMatrix<int,ScalarType> KK(blockSize_,blockSize_),
1126 MM(blockSize_,blockSize_),
1127 S(blockSize_,blockSize_);
1128 {
1129#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1130 Teuchos::TimeMonitor lcltimer( *timerLocalProj_ );
1131#endif
1132 // project K
1133 MVT::MvTransMv(ONE,*X_,*KX_,KK);
1134 // project M
1135 MVT::MvTransMv(ONE,*X_,*MX_,MM);
1136 nevLocal_ = blockSize_;
1137 }
1138
1139 // solve the projected problem
1140 {
1141#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1142 Teuchos::TimeMonitor lcltimer( *timerDS_ );
1143#endif
1144 Utils::directSolver(blockSize_, KK, Teuchos::rcpFromRef(MM), S, theta_, nevLocal_, 1);
1145 TEUCHOS_TEST_FOR_EXCEPTION(nevLocal_ != blockSize_,LOBPCGInitFailure,
1146 "Anasazi::LOBPCG::initialize(): Initial Ritz analysis did not produce enough Ritz pairs to initialize algorithm.");
1147 }
1148
1149 // We only have blockSize_ ritz pairs, ergo we do not need to select.
1150 // However, we still require them to be ordered correctly
1151 {
1152#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1153 Teuchos::TimeMonitor lcltimer( *timerSort_ );
1154#endif
1155
1156 std::vector<int> order(blockSize_);
1157 //
1158 // sort the first blockSize_ values in theta_
1159 sm_->sort(theta_, Teuchos::rcpFromRef(order), blockSize_); // don't catch exception
1160 //
1161 // apply the same ordering to the primitive ritz vectors
1162 Utils::permuteVectors(order,S);
1163 }
1164
1165 // update the solution, use R for storage
1166 {
1167#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1168 Teuchos::TimeMonitor lcltimer( *timerLocalUpdate_ );
1169#endif
1170 // X <- X*S
1171 MVT::MvAddMv( ONE, *X_, ZERO, *X_, *R_ );
1172 MVT::MvTimesMatAddMv( ONE, *R_, S, ZERO, *X_ );
1173 // KX <- KX*S
1174 MVT::MvAddMv( ONE, *KX_, ZERO, *KX_, *R_ );
1175 MVT::MvTimesMatAddMv( ONE, *R_, S, ZERO, *KX_ );
1176 if (hasM_) {
1177 // MX <- MX*S
1178 MVT::MvAddMv( ONE, *MX_, ZERO, *MX_, *R_ );
1179 MVT::MvTimesMatAddMv( ONE, *R_, S, ZERO, *MX_ );
1180 }
1181 }
1182 }
1183
1184 //----------------------------------------
1185 // compute R
1186 //----------------------------------------
1187 if (newstate.R != Teuchos::null) {
1188 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength(*newstate.R) != MVT::GetGlobalLength(*R_),
1189 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): vector length of newstate.R not correct." );
1190 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetNumberVecs(*newstate.R) < blockSize_,
1191 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): newstate.R must have blockSize number of vectors." );
1192 MVT::SetBlock(*newstate.R,bsind,*R_);
1193 }
1194 else {
1195#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1196 Teuchos::TimeMonitor lcltimer( *timerCompRes_ );
1197#endif
1198 // form R <- KX - MX*T
1199 MVT::MvAddMv(ZERO,*KX_,ONE,*KX_,*R_);
1200 Teuchos::SerialDenseMatrix<int,ScalarType> T(blockSize_,blockSize_);
1201 for (int i=0; i<blockSize_; i++) T(i,i) = theta_[i];
1202 MVT::MvTimesMatAddMv(-ONE,*MX_,T,ONE,*R_);
1203 }
1204
1205 // R has been updated; mark the norms as out-of-date
1206 Rnorms_current_ = false;
1207 R2norms_current_ = false;
1208
1209 // put data in P,KP,MP: P is not used to set theta
1210 if (newstate.P != Teuchos::null) {
1211 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetNumberVecs(*newstate.P) < blockSize_ ,
1212 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): newstate.P must have blockSize number of vectors." );
1213 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength(*newstate.P) != MVT::GetGlobalLength(*P_),
1214 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): vector length of newstate.P not correct." );
1215 hasP_ = true;
1216
1217 // set P_
1218 MVT::SetBlock(*newstate.P,bsind,*P_);
1219
1220 // set/compute KP_
1221 if (newstate.KP != Teuchos::null) {
1222 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetNumberVecs(*newstate.KP) < blockSize_,
1223 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): newstate.KP must have blockSize number of vectors." );
1224 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength(*newstate.KP) != MVT::GetGlobalLength(*KP_),
1225 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): vector length of newstate.KP not correct." );
1226 MVT::SetBlock(*newstate.KP,bsind,*KP_);
1227 }
1228 else {
1229#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1230 Teuchos::TimeMonitor lcltimer( *timerOp_ );
1231#endif
1232 OPT::Apply(*Op_,*P_,*KP_);
1233 count_ApplyOp_ += blockSize_;
1234 }
1235
1236 // set/compute MP_
1237 if (hasM_) {
1238 if (newstate.MP != Teuchos::null) {
1239 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetNumberVecs(*newstate.MP) < blockSize_,
1240 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): newstate.MP must have blockSize number of vectors." );
1241 TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength(*newstate.MP) != MVT::GetGlobalLength(*MP_),
1242 std::invalid_argument, "Anasazi::LOBPCG::initialize(newstate): vector length of newstate.MP not correct." );
1243 MVT::SetBlock(*newstate.MP,bsind,*MP_);
1244 }
1245 else {
1246#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1247 Teuchos::TimeMonitor lcltimer( *timerMOp_ );
1248#endif
1249 OPT::Apply(*MOp_,*P_,*MP_);
1250 count_ApplyM_ += blockSize_;
1251 }
1252 }
1253 }
1254 else {
1255 hasP_ = false;
1256 }
1257
1258 // finally, we are initialized
1259 initialized_ = true;
1260
1261 if (om_->isVerbosity( Debug ) ) {
1262 // Check almost everything here
1263 CheckList chk;
1264 chk.checkX = true;
1265 chk.checkKX = true;
1266 chk.checkMX = true;
1267 chk.checkP = true;
1268 chk.checkKP = true;
1269 chk.checkMP = true;
1270 chk.checkR = true;
1271 chk.checkQ = true;
1272 om_->print( Debug, accuracyCheck(chk, ": after initialize()") );
1273 }
1274
1275 }
1276
1277 template <class ScalarType, class MV, class OP>
1283
1284
1286 // Instruct the solver to use full orthogonalization
1287 template <class ScalarType, class MV, class OP>
1288 void LOBPCG<ScalarType,MV,OP>::setFullOrtho (bool fullOrtho)
1289 {
1290 if ( fullOrtho_ == true || initialized_ == false || fullOrtho == fullOrtho_ ) {
1291 // state is already orthogonalized or solver is not initialized
1292 fullOrtho_ = fullOrtho;
1293 }
1294 else {
1295 // solver is initialized, state is not fully orthogonalized, and user has requested full orthogonalization
1296 // ergo, we must throw away data in P
1297 fullOrtho_ = true;
1298 hasP_ = false;
1299 }
1300
1301 // the user has called setFullOrtho, so the class has been instantiated
1302 // ergo, the data has already been allocated, i.e., setBlockSize() has been called
1303 // if it is already allocated, it should be the proper size
1304 if (fullOrtho_ && tmpmvec_ == Teuchos::null) {
1305 // allocated the workspace
1306 tmpmvec_ = MVT::Clone(*X_,blockSize_);
1307 }
1308 else if (fullOrtho_==false) {
1309 // free the workspace
1310 tmpmvec_ = Teuchos::null;
1311 }
1312 }
1313
1314
1316 // Perform LOBPCG iterations until the StatusTest tells us to stop.
1317 template <class ScalarType, class MV, class OP>
1318 void LOBPCG<ScalarType,MV,OP>::iterate ()
1319 {
1320 //
1321 // Allocate/initialize data structures
1322 //
1323 if (initialized_ == false) {
1324 initialize();
1325 }
1326
1327 //
1328 // Miscellaneous definitions
1329 const int oneBlock = blockSize_;
1330 const int twoBlocks = 2*blockSize_;
1331 const int threeBlocks = 3*blockSize_;
1332
1333 std::vector<int> indblock1(blockSize_), indblock2(blockSize_), indblock3(blockSize_);
1334 for (int i=0; i<blockSize_; i++) {
1335 indblock1[i] = i;
1336 indblock2[i] = i + blockSize_;
1337 indblock3[i] = i + 2*blockSize_;
1338 }
1339
1340 //
1341 // Define dense projected/local matrices
1342 // KK = Local stiffness matrix (size: 3*blockSize_ x 3*blockSize_)
1343 // MM = Local mass matrix (size: 3*blockSize_ x 3*blockSize_)
1344 // S = Local eigenvectors (size: 3*blockSize_ x 3*blockSize_)
1345 Teuchos::SerialDenseMatrix<int,ScalarType> KK( threeBlocks, threeBlocks ),
1346 MM( threeBlocks, threeBlocks ),
1347 S( threeBlocks, threeBlocks );
1348
1349 while (tester_->checkStatus(this) != Passed) {
1350
1351 // Print information on current status
1352 if (om_->isVerbosity(Debug)) {
1353 currentStatus( om_->stream(Debug) );
1354 }
1355 else if (om_->isVerbosity(IterationDetails)) {
1356 currentStatus( om_->stream(IterationDetails) );
1357 }
1358
1359 // increment iteration counter
1360 iter_++;
1361
1362 // Apply the preconditioner on the residuals: H <- Prec*R
1363 if (Prec_ != Teuchos::null) {
1364#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1365 Teuchos::TimeMonitor lcltimer( *timerPrec_ );
1366#endif
1367 OPT::Apply( *Prec_, *R_, *H_ ); // don't catch the exception
1368 count_ApplyPrec_ += blockSize_;
1369 }
1370 else {
1371 MVT::MvAddMv(ONE,*R_,ZERO,*R_,*H_);
1372 }
1373
1374 // Apply the mass matrix on H
1375 if (hasM_) {
1376#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1377 Teuchos::TimeMonitor lcltimer( *timerMOp_ );
1378#endif
1379 OPT::Apply( *MOp_, *H_, *MH_); // don't catch the exception
1380 count_ApplyM_ += blockSize_;
1381 }
1382
1383 // orthogonalize H against the auxiliary vectors
1384 // optionally: orthogonalize H against X and P ([X P] is already orthonormal)
1385 Teuchos::Array<Teuchos::RCP<const MV> > Q;
1386 Q = auxVecs_;
1387 if (fullOrtho_) {
1388 // X and P are not contiguous, so there is not much point in putting them under
1389 // a single multivector view
1390 Q.push_back(X_);
1391 if (hasP_) {
1392 Q.push_back(P_);
1393 }
1394 }
1395 {
1396#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1397 Teuchos::TimeMonitor lcltimer( *timerOrtho_ );
1398#endif
1399 Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > dummyC =
1400 Teuchos::tuple<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > >(Teuchos::null);
1401 int rank = orthman_->projectAndNormalizeMat(*H_,Q,dummyC,Teuchos::null,MH_);
1402 // our views are currently in place; it is safe to throw an exception
1403 TEUCHOS_TEST_FOR_EXCEPTION(rank != blockSize_,LOBPCGOrthoFailure,
1404 "Anasazi::LOBPCG::iterate(): unable to compute orthonormal basis for H.");
1405 }
1406
1407 if (om_->isVerbosity( Debug ) ) {
1408 CheckList chk;
1409 chk.checkH = true;
1410 chk.checkMH = true;
1411 om_->print( Debug, accuracyCheck(chk, ": after ortho H") );
1412 }
1413 else if (om_->isVerbosity( OrthoDetails ) ) {
1414 CheckList chk;
1415 chk.checkH = true;
1416 chk.checkMH = true;
1417 om_->print( OrthoDetails, accuracyCheck(chk,": after ortho H") );
1418 }
1419
1420 // Apply the stiffness matrix to H
1421 {
1422#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1423 Teuchos::TimeMonitor lcltimer( *timerOp_ );
1424#endif
1425 OPT::Apply( *Op_, *H_, *KH_); // don't catch the exception
1426 count_ApplyOp_ += blockSize_;
1427 }
1428
1429 if (hasP_) {
1430 nevLocal_ = threeBlocks;
1431 }
1432 else {
1433 nevLocal_ = twoBlocks;
1434 }
1435
1436 //
1437 // we need bases: [X H P] and [H P] (only need the latter if fullOrtho == false)
1438 // we need to perform the following operations:
1439 // X' [KX KH KP]
1440 // X' [MX MH MP]
1441 // H' [KH KP]
1442 // H' [MH MP]
1443 // P' [KP]
1444 // P' [MP]
1445 // [X H P] CX
1446 // [X H P] CP if fullOrtho
1447 // [H P] CP if !fullOrtho
1448 //
1449 // since M[X H P] is potentially the same memory as [X H P], and
1450 // because we are not allowed to have overlapping non-const views of
1451 // a multivector, we will now abandon our non-const views in favor of
1452 // const views
1453 //
1454 X_ = Teuchos::null;
1455 KX_ = Teuchos::null;
1456 MX_ = Teuchos::null;
1457 H_ = Teuchos::null;
1458 KH_ = Teuchos::null;
1459 MH_ = Teuchos::null;
1460 P_ = Teuchos::null;
1461 KP_ = Teuchos::null;
1462 MP_ = Teuchos::null;
1463 Teuchos::RCP<const MV> cX, cH, cXHP, cHP, cK_XHP, cK_HP, cM_XHP, cM_HP, cP, cK_P, cM_P;
1464 {
1465 cX = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(V_),indblock1);
1466 cH = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(V_),indblock2);
1467
1468 std::vector<int> indXHP(nevLocal_);
1469 for (int i=0; i<nevLocal_; i++) {
1470 indXHP[i] = i;
1471 }
1472 cXHP = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(V_),indXHP);
1473 cK_XHP = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(KV_),indXHP);
1474 if (hasM_) {
1475 cM_XHP = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(MV_),indXHP);
1476 }
1477 else {
1478 cM_XHP = cXHP;
1479 }
1480
1481 std::vector<int> indHP(nevLocal_-blockSize_);
1482 for (int i=blockSize_; i<nevLocal_; i++) {
1483 indHP[i-blockSize_] = i;
1484 }
1485 cHP = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(V_),indHP);
1486 cK_HP = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(KV_),indHP);
1487 if (hasM_) {
1488 cM_HP = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(MV_),indHP);
1489 }
1490 else {
1491 cM_HP = cHP;
1492 }
1493
1494 if (nevLocal_ == threeBlocks) {
1495 cP = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(V_),indblock3);
1496 cK_P = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(KV_),indblock3);
1497 if (hasM_) {
1498 cM_P = MVT::CloneView(*Teuchos::rcp_implicit_cast<const MV>(MV_),indblock3);
1499 }
1500 else {
1501 cM_P = cP;
1502 }
1503 }
1504 }
1505
1506 //
1507 //----------------------------------------
1508 // Form "local" mass and stiffness matrices
1509 //----------------------------------------
1510 {
1511 // We will form only the block upper triangular part of
1512 // [X H P]' K [X H P] and [X H P]' M [X H P]
1513 // Get the necessary views into KK and MM:
1514 // [--K1--] [--M1--]
1515 // KK = [ -K2-] MM = [ -M2-]
1516 // [ K3] [ M3]
1517 //
1518 // It is okay to declare a zero-area view of a Teuchos::SerialDenseMatrix
1519 //
1520 Teuchos::SerialDenseMatrix<int,ScalarType>
1521 K1(Teuchos::View,KK,blockSize_,nevLocal_ ,0*blockSize_,0*blockSize_),
1522 K2(Teuchos::View,KK,blockSize_,nevLocal_-1*blockSize_,1*blockSize_,1*blockSize_),
1523 K3(Teuchos::View,KK,blockSize_,nevLocal_-2*blockSize_,2*blockSize_,2*blockSize_),
1524 M1(Teuchos::View,MM,blockSize_,nevLocal_ ,0*blockSize_,0*blockSize_),
1525 M2(Teuchos::View,MM,blockSize_,nevLocal_-1*blockSize_,1*blockSize_,1*blockSize_),
1526 M3(Teuchos::View,MM,blockSize_,nevLocal_-2*blockSize_,2*blockSize_,2*blockSize_);
1527 {
1528#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1529 Teuchos::TimeMonitor lcltimer( *timerLocalProj_ );
1530#endif
1531 MVT::MvTransMv( ONE, *cX, *cK_XHP, K1 );
1532 MVT::MvTransMv( ONE, *cX, *cM_XHP, M1 );
1533 MVT::MvTransMv( ONE, *cH, *cK_HP , K2 );
1534 MVT::MvTransMv( ONE, *cH, *cM_HP , M2 );
1535 if (nevLocal_ == threeBlocks) {
1536 MVT::MvTransMv( ONE, *cP, *cK_P, K3 );
1537 MVT::MvTransMv( ONE, *cP, *cM_P, M3 );
1538 }
1539 }
1540 }
1541 // below, we only need bases [X H P] and [H P] and friends
1542 // furthermore, we only need [H P] and friends if fullOrtho == false
1543 // clear the others now
1544 cX = Teuchos::null;
1545 cH = Teuchos::null;
1546 cP = Teuchos::null;
1547 cK_P = Teuchos::null;
1548 cM_P = Teuchos::null;
1549 if (fullOrtho_ == true) {
1550 cHP = Teuchos::null;
1551 cK_HP = Teuchos::null;
1552 cM_HP = Teuchos::null;
1553 }
1554
1555 //
1556 //---------------------------------------------------
1557 // Perform a spectral decomposition of (KK,MM)
1558 //---------------------------------------------------
1559 //
1560 // Get pointers to relevant part of KK, MM and S for Rayleigh-Ritz analysis
1561 Teuchos::SerialDenseMatrix<int,ScalarType> lclKK(Teuchos::View,KK,nevLocal_,nevLocal_),
1562 lclMM(Teuchos::View,MM,nevLocal_,nevLocal_),
1563 lclS(Teuchos::View, S,nevLocal_,nevLocal_);
1564 {
1565#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1566 Teuchos::TimeMonitor lcltimer( *timerDS_ );
1567#endif
1568 int localSize = nevLocal_;
1569 Utils::directSolver(localSize, lclKK, Teuchos::rcpFromRef(lclMM), lclS, theta_, nevLocal_, 0);
1570 // localSize tells directSolver() how big KK,MM are
1571 // however, directSolver() may choose to use only the principle submatrices of KK,MM
1572 // because of loss of MM-orthogonality in the projected eigenvectors
1573 // nevLocal_ tells us how much it used, telling us the effective localSize
1574 // (i.e., how much of KK,MM used by directSolver)
1575 // we will not tolerate any indefiniteness, and will throw an exception if it was
1576 // detected by directSolver
1577 //
1578 if (nevLocal_ != localSize) {
1579 // before throwing the exception, and thereby leaving iterate(), setup the views again
1580 // first, clear the const views
1581 cXHP = Teuchos::null;
1582 cK_XHP = Teuchos::null;
1583 cM_XHP = Teuchos::null;
1584 cHP = Teuchos::null;
1585 cK_HP = Teuchos::null;
1586 cM_HP = Teuchos::null;
1587 setupViews();
1588 }
1589 TEUCHOS_TEST_FOR_EXCEPTION(nevLocal_ != localSize, LOBPCGRitzFailure,
1590 "Anasazi::LOBPCG::iterate(): indefiniteness detected in projected mass matrix." );
1591 }
1592
1593 //
1594 //---------------------------------------------------
1595 // Sort the ritz values using the sort manager
1596 //---------------------------------------------------
1597 Teuchos::LAPACK<int,ScalarType> lapack;
1598 Teuchos::BLAS<int,ScalarType> blas;
1599 {
1600#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1601 Teuchos::TimeMonitor lcltimer( *timerSort_ );
1602#endif
1603
1604 std::vector<int> order(nevLocal_);
1605 //
1606 // Sort the first nevLocal_ values in theta_
1607 sm_->sort(theta_, Teuchos::rcpFromRef(order), nevLocal_); // don't catch exception
1608 //
1609 // Sort the primitive ritz vectors
1610 Utils::permuteVectors(order,lclS);
1611 }
1612
1613 //
1614 //----------------------------------------
1615 // Compute coefficients for X and P under [X H P]
1616 //----------------------------------------
1617 // Before computing X,P, optionally perform orthogonalization per Hetmaniuk,Lehoucq paper
1618 // CX will be the coefficients of [X,H,P] for new X, CP for new P
1619 // The paper suggests orthogonalizing CP against CX and orthonormalizing CP, w.r.t. MM
1620 // Here, we will also orthonormalize CX.
1621 // This is accomplished using the Cholesky factorization of [CX CP]^H lclMM [CX CP]
1622 Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > CX, CP;
1623 if (fullOrtho_) {
1624 // build orthonormal basis for ( 0 ) that is MM orthogonal to ( S11 )
1625 // ( S21 ) ( S21 )
1626 // ( S31 ) ( S31 )
1627 // Do this using Cholesky factorization of ( S11 0 )
1628 // ( S21 S21 )
1629 // ( S31 S31 )
1630 // ( S11 0 )
1631 // Build C = ( S21 S21 )
1632 // ( S31 S31 )
1633 Teuchos::SerialDenseMatrix<int,ScalarType> C(nevLocal_,twoBlocks),
1634 MMC(nevLocal_,twoBlocks),
1635 CMMC(twoBlocks ,twoBlocks);
1636
1637 // first block of rows: ( S11 0 )
1638 for (int j=0; j<oneBlock; j++) {
1639 for (int i=0; i<oneBlock; i++) {
1640 // CX
1641 C(i,j) = lclS(i,j);
1642 // CP
1643 C(i,j+oneBlock) = ZERO;
1644 }
1645 }
1646 // second block of rows: (S21 S21)
1647 for (int j=0; j<oneBlock; j++) {
1648 for (int i=oneBlock; i<twoBlocks; i++) {
1649 // CX
1650 C(i,j) = lclS(i,j);
1651 // CP
1652 C(i,j+oneBlock) = lclS(i,j);
1653 }
1654 }
1655 // third block of rows
1656 if (nevLocal_ == threeBlocks) {
1657 for (int j=0; j<oneBlock; j++) {
1658 for (int i=twoBlocks; i<threeBlocks; i++) {
1659 // CX
1660 C(i,j) = lclS(i,j);
1661 // CP
1662 C(i,j+oneBlock) = lclS(i,j);
1663 }
1664 }
1665 }
1666
1667 // compute MMC = lclMM*C
1668 {
1669 int teuchosret;
1670 teuchosret = MMC.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,ONE,lclMM,C,ZERO);
1671 TEUCHOS_TEST_FOR_EXCEPTION(teuchosret != 0,std::logic_error,
1672 "Anasazi::LOBPCG::iterate(): Logic error calling SerialDenseMatrix::multiply");
1673 // compute CMMC = C^H*MMC == C^H*lclMM*C
1674 teuchosret = CMMC.multiply(Teuchos::CONJ_TRANS,Teuchos::NO_TRANS,ONE,C,MMC,ZERO);
1675 TEUCHOS_TEST_FOR_EXCEPTION(teuchosret != 0,std::logic_error,
1676 "Anasazi::LOBPCG::iterate(): Logic error calling SerialDenseMatrix::multiply");
1677 }
1678
1679 // compute R (cholesky) of CMMC
1680 {
1681 int info;
1682 lapack.POTRF('U',twoBlocks,CMMC.values(),CMMC.stride(),&info);
1683 // our views ARE NOT currently in place; we must reestablish them before throwing an exception
1684 if (info != 0) {
1685 // Check symmetry of H'*K*H, this might be the first place a non-Hermitian operator will cause a problem.
1686 Teuchos::SerialDenseMatrix<int,ScalarType> K22(Teuchos::View,KK,blockSize_,blockSize_,1*blockSize_,1*blockSize_);
1687 Teuchos::SerialDenseMatrix<int,ScalarType> K22_trans( K22, Teuchos::CONJ_TRANS );
1688 K22_trans -= K22;
1689 MagnitudeType symNorm = K22_trans.normOne();
1690 MagnitudeType tol = SCT::magnitude(SCT::squareroot(SCT::eps()));
1691
1692 cXHP = Teuchos::null;
1693 cHP = Teuchos::null;
1694 cK_XHP = Teuchos::null;
1695 cK_HP = Teuchos::null;
1696 cM_XHP = Teuchos::null;
1697 cM_HP = Teuchos::null;
1698 setupViews();
1699
1700 std::string errMsg = "Anasazi::LOBPCG::iterate(): Cholesky factorization failed during full orthogonalization.";
1701 if ( symNorm < tol )
1702 {
1703 TEUCHOS_TEST_FOR_EXCEPTION(info != 0, LOBPCGOrthoFailure, errMsg );
1704 }
1705 else
1706 {
1707 errMsg += " Check the operator A (or K), it may not be Hermitian!";
1708 TEUCHOS_TEST_FOR_EXCEPTION(info != 0, LOBPCGOrthoFailure, errMsg );
1709 }
1710 }
1711 }
1712 // compute C = C inv(R)
1713 blas.TRSM(Teuchos::RIGHT_SIDE,Teuchos::UPPER_TRI,Teuchos::NO_TRANS,Teuchos::NON_UNIT_DIAG,
1714 nevLocal_,twoBlocks,ONE,CMMC.values(),CMMC.stride(),C.values(),C.stride());
1715 // put C(:,0:oneBlock-1) into CX
1716 CX = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>(Teuchos::Copy,C,nevLocal_,oneBlock,0,0) );
1717 // put C(:,oneBlock:twoBlocks-1) into CP
1718 CP = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>(Teuchos::Copy,C,nevLocal_,oneBlock,0,oneBlock) );
1719
1720 // check the results
1721 if (om_->isVerbosity( Debug ) ) {
1722 Teuchos::SerialDenseMatrix<int,ScalarType> tmp1(nevLocal_,oneBlock),
1723 tmp2(oneBlock,oneBlock);
1724 MagnitudeType tmp;
1725 int teuchosret;
1726 std::stringstream os;
1727 os.precision(2);
1728 os.setf(std::ios::scientific, std::ios::floatfield);
1729
1730 os << " Checking Full Ortho: iteration " << iter_ << std::endl;
1731
1732 // check CX^T MM CX == I
1733 // compute tmp1 = MM*CX
1734 teuchosret = tmp1.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,ONE,lclMM,*CX,ZERO);
1735 TEUCHOS_TEST_FOR_EXCEPTION(teuchosret != 0,std::logic_error,
1736 "Anasazi::LOBPCG::iterate(): Logic error calling SerialDenseMatrix::multiply");
1737 // compute tmp2 = CX^H*tmp1 == CX^H*MM*CX
1738 teuchosret = tmp2.multiply(Teuchos::CONJ_TRANS,Teuchos::NO_TRANS,ONE,*CX,tmp1,ZERO);
1739 TEUCHOS_TEST_FOR_EXCEPTION(teuchosret != 0,std::logic_error,
1740 "Anasazi::LOBPCG::iterate(): Logic error calling SerialDenseMatrix::multiply");
1741 // subtrace tmp2 - I == CX^H * MM * CX - I
1742 for (int i=0; i<oneBlock; i++) tmp2(i,i) -= ONE;
1743 tmp = tmp2.normFrobenius();
1744 os << " >> Error in CX^H MM CX == I : " << tmp << std::endl;
1745
1746 // check CP^T MM CP == I
1747 // compute tmp1 = MM*CP
1748 teuchosret = tmp1.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,ONE,lclMM,*CP,ZERO);
1749 TEUCHOS_TEST_FOR_EXCEPTION(teuchosret != 0,std::logic_error,
1750 "Anasazi::LOBPCG::iterate(): Logic error calling SerialDenseMatrix::multiply");
1751 // compute tmp2 = CP^H*tmp1 == CP^H*MM*CP
1752 teuchosret = tmp2.multiply(Teuchos::CONJ_TRANS,Teuchos::NO_TRANS,ONE,*CP,tmp1,ZERO);
1753 TEUCHOS_TEST_FOR_EXCEPTION(teuchosret != 0,std::logic_error,
1754 "Anasazi::LOBPCG::iterate(): Logic error calling SerialDenseMatrix::multiply");
1755 // subtrace tmp2 - I == CP^H * MM * CP - I
1756 for (int i=0; i<oneBlock; i++) tmp2(i,i) -= ONE;
1757 tmp = tmp2.normFrobenius();
1758 os << " >> Error in CP^H MM CP == I : " << tmp << std::endl;
1759
1760 // check CX^T MM CP == 0
1761 // compute tmp1 = MM*CP
1762 teuchosret = tmp1.multiply(Teuchos::NO_TRANS,Teuchos::NO_TRANS,ONE,lclMM,*CP,ZERO);
1763 TEUCHOS_TEST_FOR_EXCEPTION(teuchosret != 0,std::logic_error,"Anasazi::LOBPCG::iterate(): Logic error calling SerialDenseMatrix::multiply");
1764 // compute tmp2 = CX^H*tmp1 == CX^H*MM*CP
1765 teuchosret = tmp2.multiply(Teuchos::CONJ_TRANS,Teuchos::NO_TRANS,ONE,*CX,tmp1,ZERO);
1766 TEUCHOS_TEST_FOR_EXCEPTION(teuchosret != 0,std::logic_error,"Anasazi::LOBPCG::iterate(): Logic error calling SerialDenseMatrix::multiply");
1767 // subtrace tmp2 == CX^H * MM * CP
1768 tmp = tmp2.normFrobenius();
1769 os << " >> Error in CX^H MM CP == 0 : " << tmp << std::endl;
1770
1771 os << std::endl;
1772 om_->print(Debug,os.str());
1773 }
1774 }
1775 else {
1776 // [S11 ... ...]
1777 // S = [S21 ... ...]
1778 // [S31 ... ...]
1779 //
1780 // CX = [S11]
1781 // [S21]
1782 // [S31] -> X = [X H P] CX
1783 //
1784 // CP = [S21] -> P = [H P] CP
1785 // [S31]
1786 //
1787 CX = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>(Teuchos::Copy,lclS,nevLocal_ ,oneBlock,0 ,0) );
1788 CP = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>(Teuchos::Copy,lclS,nevLocal_-oneBlock,oneBlock,oneBlock,0) );
1789 }
1790
1791 //
1792 //----------------------------------------
1793 // Compute new X and new P
1794 //----------------------------------------
1795 // Note: Use R as a temporary work space and (if full ortho) tmpMV as well
1796 {
1797#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1798 Teuchos::TimeMonitor lcltimer( *timerLocalUpdate_ );
1799#endif
1800
1801 // if full ortho, then CX and CP are dense
1802 // we multiply [X H P]*CX into tmpMV
1803 // [X H P]*CP into R
1804 // then put V(:,firstblock) <- tmpMV
1805 // V(:,thirdblock) <- R
1806 //
1807 // if no full ortho, then [H P]*CP doesn't reference first block (X)
1808 // of V, so that we can modify it before computing P
1809 // so we multiply [X H P]*CX into R
1810 // V(:,firstblock) <- R
1811 // multiply [H P]*CP into R
1812 // V(:,thirdblock) <- R
1813 //
1814 // mutatis mutandis for K[XP] and M[XP]
1815 //
1816 // use SetBlock to do the assignments into V_
1817 //
1818 // in either case, views are only allowed to be overlapping
1819 // if they are const, and it should be assume that SetBlock
1820 // creates a view of the associated part
1821 //
1822 // we have from above const-pointers to [KM]XHP, [KM]HP and (if hasP) [KM]P
1823 //
1824 if (fullOrtho_) {
1825 // X,P
1826 MVT::MvTimesMatAddMv(ONE,*cXHP,*CX,ZERO,*tmpmvec_);
1827 MVT::MvTimesMatAddMv(ONE,*cXHP,*CP,ZERO,*R_);
1828 cXHP = Teuchos::null;
1829 MVT::SetBlock(*tmpmvec_,indblock1,*V_);
1830 MVT::SetBlock(*R_ ,indblock3,*V_);
1831 // KX,KP
1832 MVT::MvTimesMatAddMv(ONE,*cK_XHP,*CX,ZERO,*tmpmvec_);
1833 MVT::MvTimesMatAddMv(ONE,*cK_XHP,*CP,ZERO,*R_);
1834 cK_XHP = Teuchos::null;
1835 MVT::SetBlock(*tmpmvec_,indblock1,*KV_);
1836 MVT::SetBlock(*R_ ,indblock3,*KV_);
1837 // MX,MP
1838 if (hasM_) {
1839 MVT::MvTimesMatAddMv(ONE,*cM_XHP,*CX,ZERO,*tmpmvec_);
1840 MVT::MvTimesMatAddMv(ONE,*cM_XHP,*CP,ZERO,*R_);
1841 cM_XHP = Teuchos::null;
1842 MVT::SetBlock(*tmpmvec_,indblock1,*MV_);
1843 MVT::SetBlock(*R_ ,indblock3,*MV_);
1844 }
1845 else {
1846 cM_XHP = Teuchos::null;
1847 }
1848 }
1849 else {
1850 // X,P
1851 MVT::MvTimesMatAddMv(ONE,*cXHP,*CX,ZERO,*R_);
1852 cXHP = Teuchos::null;
1853 MVT::SetBlock(*R_,indblock1,*V_);
1854 MVT::MvTimesMatAddMv(ONE,*cHP,*CP,ZERO,*R_);
1855 cHP = Teuchos::null;
1856 MVT::SetBlock(*R_,indblock3,*V_);
1857 // KX,KP
1858 MVT::MvTimesMatAddMv(ONE,*cK_XHP,*CX,ZERO,*R_);
1859 cK_XHP = Teuchos::null;
1860 MVT::SetBlock(*R_,indblock1,*KV_);
1861 MVT::MvTimesMatAddMv(ONE,*cK_HP,*CP,ZERO,*R_);
1862 cK_HP = Teuchos::null;
1863 MVT::SetBlock(*R_,indblock3,*KV_);
1864 // MX,MP
1865 if (hasM_) {
1866 MVT::MvTimesMatAddMv(ONE,*cM_XHP,*CX,ZERO,*R_);
1867 cM_XHP = Teuchos::null;
1868 MVT::SetBlock(*R_,indblock1,*MV_);
1869 MVT::MvTimesMatAddMv(ONE,*cM_HP,*CP,ZERO,*R_);
1870 cM_HP = Teuchos::null;
1871 MVT::SetBlock(*R_,indblock3,*MV_);
1872 }
1873 else {
1874 cM_XHP = Teuchos::null;
1875 cM_HP = Teuchos::null;
1876 }
1877 }
1878 } // end timing block
1879 // done with coefficient matrices
1880 CX = Teuchos::null;
1881 CP = Teuchos::null;
1882
1883 //
1884 // we now have a P direction
1885 hasP_ = true;
1886
1887 // debugging check: all of our const views should have been cleared by now
1888 // if not, we have a logic error above
1889 TEUCHOS_TEST_FOR_EXCEPTION( cXHP != Teuchos::null || cK_XHP != Teuchos::null || cM_XHP != Teuchos::null
1890 || cHP != Teuchos::null || cK_HP != Teuchos::null || cM_HP != Teuchos::null
1891 || cP != Teuchos::null || cK_P != Teuchos::null || cM_P != Teuchos::null,
1892 std::logic_error,
1893 "Anasazi::BlockKrylovSchur::iterate(): const views were not all cleared! Something went wrong!" );
1894
1895 //
1896 // recreate our const MV views of X,H,P and friends
1897 setupViews();
1898
1899 //
1900 // Compute the new residuals, explicitly
1901 {
1902#ifdef ANASAZI_TEUCHOS_TIME_MONITOR
1903 Teuchos::TimeMonitor lcltimer( *timerCompRes_ );
1904#endif
1905 MVT::MvAddMv( ONE, *KX_, ZERO, *KX_, *R_ );
1906 Teuchos::SerialDenseMatrix<int,ScalarType> T( blockSize_, blockSize_ );
1907 for (int i = 0; i < blockSize_; i++) {
1908 T(i,i) = theta_[i];
1909 }
1910 MVT::MvTimesMatAddMv( -ONE, *MX_, T, ONE, *R_ );
1911 }
1912
1913 // R has been updated; mark the norms as out-of-date
1914 Rnorms_current_ = false;
1915 R2norms_current_ = false;
1916
1917 // When required, monitor some orthogonalities
1918 if (om_->isVerbosity( Debug ) ) {
1919 // Check almost everything here
1920 CheckList chk;
1921 chk.checkX = true;
1922 chk.checkKX = true;
1923 chk.checkMX = true;
1924 chk.checkP = true;
1925 chk.checkKP = true;
1926 chk.checkMP = true;
1927 chk.checkR = true;
1928 om_->print( Debug, accuracyCheck(chk, ": after local update") );
1929 }
1930 else if (om_->isVerbosity( OrthoDetails )) {
1931 CheckList chk;
1932 chk.checkX = true;
1933 chk.checkP = true;
1934 chk.checkR = true;
1935 om_->print( OrthoDetails, accuracyCheck(chk, ": after local update") );
1936 }
1937 } // end while (statusTest == false)
1938 }
1939
1940
1942 // compute/return residual M-norms
1943 template <class ScalarType, class MV, class OP>
1944 std::vector<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType>
1945 LOBPCG<ScalarType,MV,OP>::getResNorms() {
1946 if (Rnorms_current_ == false) {
1947 // Update the residual norms
1948 orthman_->norm(*R_,Rnorms_);
1949 Rnorms_current_ = true;
1950 }
1951 return Rnorms_;
1952 }
1953
1954
1956 // compute/return residual 2-norms
1957 template <class ScalarType, class MV, class OP>
1958 std::vector<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType>
1959 LOBPCG<ScalarType,MV,OP>::getRes2Norms() {
1960 if (R2norms_current_ == false) {
1961 // Update the residual 2-norms
1962 MVT::MvNorm(*R_,R2norms_);
1963 R2norms_current_ = true;
1964 }
1965 return R2norms_;
1966 }
1967
1968
1969
1970
1972 // Check accuracy, orthogonality, and other debugging stuff
1973 //
1974 // bools specify which tests we want to run (instead of running more than we actually care about)
1975 //
1976 // we don't bother checking the following because they are computed explicitly:
1977 // H == Prec*R
1978 // KH == K*H
1979 //
1980 //
1981 // checkX : X orthonormal
1982 // orthogonal to auxvecs
1983 // checkMX: check MX == M*X
1984 // checkKX: check KX == K*X
1985 // checkP : if fullortho P orthonormal and orthogonal to X
1986 // orthogonal to auxvecs
1987 // checkMP: check MP == M*P
1988 // checkKP: check KP == K*P
1989 // checkH : if fullortho H orthonormal and orthogonal to X and P
1990 // orthogonal to auxvecs
1991 // checkMH: check MH == M*H
1992 // checkR : check R orthogonal to X
1993 // checkQ : check that auxiliary vectors are actually orthonormal
1994 //
1995 // TODO:
1996 // add checkTheta
1997 //
1998 template <class ScalarType, class MV, class OP>
1999 std::string LOBPCG<ScalarType,MV,OP>::accuracyCheck( const CheckList &chk, const std::string &where ) const
2000 {
2001 using std::endl;
2002
2003 std::stringstream os;
2004 os.precision(2);
2005 os.setf(std::ios::scientific, std::ios::floatfield);
2006 MagnitudeType tmp;
2007
2008 os << " Debugging checks: iteration " << iter_ << where << endl;
2009
2010 // X and friends
2011 if (chk.checkX && initialized_) {
2012 tmp = orthman_->orthonormError(*X_);
2013 os << " >> Error in X^H M X == I : " << tmp << endl;
2014 for (Array_size_type i=0; i<auxVecs_.size(); i++) {
2015 tmp = orthman_->orthogError(*X_,*auxVecs_[i]);
2016 os << " >> Error in X^H M Q[" << i << "] == 0 : " << tmp << endl;
2017 }
2018 }
2019 if (chk.checkMX && hasM_ && initialized_) {
2020 tmp = Utils::errorEquality(*X_, *MX_, MOp_);
2021 os << " >> Error in MX == M*X : " << tmp << endl;
2022 }
2023 if (chk.checkKX && initialized_) {
2024 tmp = Utils::errorEquality(*X_, *KX_, Op_);
2025 os << " >> Error in KX == K*X : " << tmp << endl;
2026 }
2027
2028 // P and friends
2029 if (chk.checkP && hasP_ && initialized_) {
2030 if (fullOrtho_) {
2031 tmp = orthman_->orthonormError(*P_);
2032 os << " >> Error in P^H M P == I : " << tmp << endl;
2033 tmp = orthman_->orthogError(*P_,*X_);
2034 os << " >> Error in P^H M X == 0 : " << tmp << endl;
2035 }
2036 for (Array_size_type i=0; i<auxVecs_.size(); i++) {
2037 tmp = orthman_->orthogError(*P_,*auxVecs_[i]);
2038 os << " >> Error in P^H M Q[" << i << "] == 0 : " << tmp << endl;
2039 }
2040 }
2041 if (chk.checkMP && hasM_ && hasP_ && initialized_) {
2042 tmp = Utils::errorEquality(*P_, *MP_, MOp_);
2043 os << " >> Error in MP == M*P : " << tmp << endl;
2044 }
2045 if (chk.checkKP && hasP_ && initialized_) {
2046 tmp = Utils::errorEquality(*P_, *KP_, Op_);
2047 os << " >> Error in KP == K*P : " << tmp << endl;
2048 }
2049
2050 // H and friends
2051 if (chk.checkH && initialized_) {
2052 if (fullOrtho_) {
2053 tmp = orthman_->orthonormError(*H_);
2054 os << " >> Error in H^H M H == I : " << tmp << endl;
2055 tmp = orthman_->orthogError(*H_,*X_);
2056 os << " >> Error in H^H M X == 0 : " << tmp << endl;
2057 if (hasP_) {
2058 tmp = orthman_->orthogError(*H_,*P_);
2059 os << " >> Error in H^H M P == 0 : " << tmp << endl;
2060 }
2061 }
2062 for (Array_size_type i=0; i<auxVecs_.size(); i++) {
2063 tmp = orthman_->orthogError(*H_,*auxVecs_[i]);
2064 os << " >> Error in H^H M Q[" << i << "] == 0 : " << tmp << endl;
2065 }
2066 }
2067 if (chk.checkMH && hasM_ && initialized_) {
2068 tmp = Utils::errorEquality(*H_, *MH_, MOp_);
2069 os << " >> Error in MH == M*H : " << tmp << endl;
2070 }
2071
2072 // R: this is not M-orthogonality, but standard euclidean orthogonality
2073 if (chk.checkR && initialized_) {
2074 Teuchos::SerialDenseMatrix<int,ScalarType> xTx(blockSize_,blockSize_);
2075 MVT::MvTransMv(ONE,*X_,*R_,xTx);
2076 tmp = xTx.normFrobenius();
2077 MVT::MvTransMv(ONE,*R_,*R_,xTx);
2078 double normR = xTx.normFrobenius();
2079 os << " >> RelError in X^H R == 0: " << tmp/normR << endl;
2080 }
2081
2082 // Q
2083 if (chk.checkQ) {
2084 for (Array_size_type i=0; i<auxVecs_.size(); i++) {
2085 tmp = orthman_->orthonormError(*auxVecs_[i]);
2086 os << " >> Error in Q[" << i << "]^H M Q[" << i << "] == I : " << tmp << endl;
2087 for (Array_size_type j=i+1; j<auxVecs_.size(); j++) {
2088 tmp = orthman_->orthogError(*auxVecs_[i],*auxVecs_[j]);
2089 os << " >> Error in Q[" << i << "]^H M Q[" << j << "] == 0 : " << tmp << endl;
2090 }
2091 }
2092 }
2093
2094 os << endl;
2095
2096 return os.str();
2097 }
2098
2099
2101 // Print the current status of the solver
2102 template <class ScalarType, class MV, class OP>
2103 void
2104 LOBPCG<ScalarType,MV,OP>::currentStatus(std::ostream &os)
2105 {
2106 using std::endl;
2107
2108 os.setf(std::ios::scientific, std::ios::floatfield);
2109 os.precision(6);
2110 os <<endl;
2111 os <<"================================================================================" << endl;
2112 os << endl;
2113 os <<" LOBPCG Solver Status" << endl;
2114 os << endl;
2115 os <<"The solver is "<<(initialized_ ? "initialized." : "not initialized.") << endl;
2116 os <<"The number of iterations performed is " << iter_ << endl;
2117 os <<"The current block size is " << blockSize_ << endl;
2118 os <<"The number of auxiliary vectors is " << numAuxVecs_ << endl;
2119 os <<"The number of operations Op*x is " << count_ApplyOp_ << endl;
2120 os <<"The number of operations M*x is " << count_ApplyM_ << endl;
2121 os <<"The number of operations Prec*x is " << count_ApplyPrec_ << endl;
2122
2123 os.setf(std::ios_base::right, std::ios_base::adjustfield);
2124
2125 if (initialized_) {
2126 os << endl;
2127 os <<"CURRENT EIGENVALUE ESTIMATES "<<endl;
2128 os << std::setw(20) << "Eigenvalue"
2129 << std::setw(20) << "Residual(M)"
2130 << std::setw(20) << "Residual(2)"
2131 << endl;
2132 os <<"--------------------------------------------------------------------------------"<<endl;
2133 for (int i=0; i<blockSize_; i++) {
2134 os << std::setw(20) << theta_[i];
2135 if (Rnorms_current_) os << std::setw(20) << Rnorms_[i];
2136 else os << std::setw(20) << "not current";
2137 if (R2norms_current_) os << std::setw(20) << R2norms_[i];
2138 else os << std::setw(20) << "not current";
2139 os << endl;
2140 }
2141 }
2142 os <<"================================================================================" << endl;
2143 os << endl;
2144 }
2145
2147 // are we initialized or not?
2148 template <class ScalarType, class MV, class OP>
2149 bool LOBPCG<ScalarType,MV,OP>::isInitialized() const {
2150 return initialized_;
2151 }
2152
2153
2155 // is P valid or not?
2156 template <class ScalarType, class MV, class OP>
2157 bool LOBPCG<ScalarType,MV,OP>::hasP() {
2158 return hasP_;
2159 }
2160
2162 // is full orthogonalization enabled or not?
2163 template <class ScalarType, class MV, class OP>
2164 bool LOBPCG<ScalarType,MV,OP>::getFullOrtho() const {
2165 return(fullOrtho_);
2166 }
2167
2168
2170 // return the current auxilliary vectors
2171 template <class ScalarType, class MV, class OP>
2172 Teuchos::Array<Teuchos::RCP<const MV> > LOBPCG<ScalarType,MV,OP>::getAuxVecs() const {
2173 return auxVecs_;
2174 }
2175
2177 // return the current block size
2178 template <class ScalarType, class MV, class OP>
2179 int LOBPCG<ScalarType,MV,OP>::getBlockSize() const {
2180 return(blockSize_);
2181 }
2182
2184 // return the current eigenproblem
2185 template <class ScalarType, class MV, class OP>
2186 const Eigenproblem<ScalarType,MV,OP>& LOBPCG<ScalarType,MV,OP>::getProblem() const {
2187 return(*problem_);
2188 }
2189
2190
2192 // return the max subspace dimension
2193 template <class ScalarType, class MV, class OP>
2194 int LOBPCG<ScalarType,MV,OP>::getMaxSubspaceDim() const {
2195 return 3*blockSize_;
2196 }
2197
2199 // return the current subspace dimension
2200 template <class ScalarType, class MV, class OP>
2201 int LOBPCG<ScalarType,MV,OP>::getCurSubspaceDim() const {
2202 if (!initialized_) return 0;
2203 return nevLocal_;
2204 }
2205
2206
2208 // return the current ritz residual norms
2209 template <class ScalarType, class MV, class OP>
2210 std::vector<typename Teuchos::ScalarTraits<ScalarType>::magnitudeType>
2211 LOBPCG<ScalarType,MV,OP>::getRitzRes2Norms()
2212 {
2213 return this->getRes2Norms();
2214 }
2215
2216
2218 // return the current compression indices
2219 template <class ScalarType, class MV, class OP>
2220 std::vector<int> LOBPCG<ScalarType,MV,OP>::getRitzIndex() {
2221 std::vector<int> ret(nevLocal_,0);
2222 return ret;
2223 }
2224
2225
2227 // return the current ritz values
2228 template <class ScalarType, class MV, class OP>
2229 std::vector<Value<ScalarType> > LOBPCG<ScalarType,MV,OP>::getRitzValues() {
2230 std::vector<Value<ScalarType> > ret(nevLocal_);
2231 for (int i=0; i<nevLocal_; i++) {
2232 ret[i].realpart = theta_[i];
2233 ret[i].imagpart = ZERO;
2234 }
2235 return ret;
2236 }
2237
2239 // Set a new StatusTest for the solver.
2240 template <class ScalarType, class MV, class OP>
2241 void LOBPCG<ScalarType,MV,OP>::setStatusTest(Teuchos::RCP<StatusTest<ScalarType,MV,OP> > test) {
2242 TEUCHOS_TEST_FOR_EXCEPTION(test == Teuchos::null,std::invalid_argument,
2243 "Anasazi::LOBPCG::setStatusTest() was passed a null StatusTest.");
2244 tester_ = test;
2245 }
2246
2248 // Get the current StatusTest used by the solver.
2249 template <class ScalarType, class MV, class OP>
2250 Teuchos::RCP<StatusTest<ScalarType,MV,OP> > LOBPCG<ScalarType,MV,OP>::getStatusTest() const {
2251 return tester_;
2252 }
2253
2255 // return the current ritz vectors
2256 template <class ScalarType, class MV, class OP>
2257 Teuchos::RCP<const MV> LOBPCG<ScalarType,MV,OP>::getRitzVectors() {
2258 return X_;
2259 }
2260
2261
2263 // reset the iteration counter
2264 template <class ScalarType, class MV, class OP>
2265 void LOBPCG<ScalarType,MV,OP>::resetNumIters() {
2266 iter_=0;
2267 }
2268
2269
2271 // return the number of iterations
2272 template <class ScalarType, class MV, class OP>
2273 int LOBPCG<ScalarType,MV,OP>::getNumIters() const {
2274 return(iter_);
2275 }
2276
2277
2279 // return the state
2280 template <class ScalarType, class MV, class OP>
2281 LOBPCGState<ScalarType,MV> LOBPCG<ScalarType,MV,OP>::getState() const {
2282 LOBPCGState<ScalarType,MV> state;
2283 state.V = V_;
2284 state.KV = KV_;
2285 state.X = X_;
2286 state.KX = KX_;
2287 state.P = P_;
2288 state.KP = KP_;
2289 state.H = H_;
2290 state.KH = KH_;
2291 state.R = R_;
2292 state.T = Teuchos::rcp(new std::vector<MagnitudeType>(theta_));
2293 if (hasM_) {
2294 state.MV = MV_;
2295 state.MX = MX_;
2296 state.MP = MP_;
2297 state.MH = MH_;
2298 }
2299 else {
2300 state.MX = Teuchos::null;
2301 state.MP = Teuchos::null;
2302 state.MH = Teuchos::null;
2303 }
2304 return state;
2305 }
2306
2307} // end Anasazi namespace
2308
2309#endif // ANASAZI_LOBPCG_HPP
AnasaziEigensolver.hpp
Pure virtual base class which describes the basic interface to the iterative eigensolver.
AnasaziMatOrthoManager.hpp
Templated virtual class for providing orthogonalization/orthonormalization methods with matrix-based ...
AnasaziMultiVecTraits.hpp
Declaration of basic traits for the multivector type.
AnasaziOperatorTraits.hpp
Virtual base class which defines basic traits for the operator type.
AnasaziSolverUtils.hpp
Class which provides internal utilities for the Anasazi solvers.
AnasaziTypes.hpp
Types and exceptions used within Anasazi solvers and interfaces.
Anasazi::AnasaziError
An exception class parent to all Anasazi exceptions.
Definition AnasaziTypes.hpp:31
Anasazi::Eigensolver
The Eigensolver is a templated virtual base class that defines the basic interface that any eigensolv...
Definition AnasaziEigensolver.hpp:35
Anasazi::LOBPCGInitFailure
LOBPCGInitFailure is thrown when the LOBPCG solver is unable to generate an initial iterate in the LO...
Definition AnasaziLOBPCG.hpp:162
Anasazi::LOBPCGOrthoFailure
LOBPCGOrthoFailure is thrown when an orthogonalization attempt fails.
Definition AnasaziLOBPCG.hpp:182
Anasazi::LOBPCGRitzFailure
LOBPCGRitzFailure is thrown when the LOBPCG solver is unable to continue a call to LOBPCG::iterate() ...
Definition AnasaziLOBPCG.hpp:146
Anasazi::LOBPCG
This class provides the Locally Optimal Block Preconditioned Conjugate Gradient (LOBPCG) iteration,...
Definition AnasaziLOBPCG.hpp:190
Anasazi::LOBPCG::iterate
void iterate()
This method performs LOBPCG iterations until the status test indicates the need to stop or an error o...
Definition AnasaziLOBPCG.hpp:1318
Anasazi::LOBPCG::getRitzRes2Norms
std::vector< typename Teuchos::ScalarTraits< ScalarType >::magnitudeType > getRitzRes2Norms()
Get the 2-norms of the residuals.
Definition AnasaziLOBPCG.hpp:2211
Anasazi::LOBPCG::setStatusTest
void setStatusTest(Teuchos::RCP< StatusTest< ScalarType, MV, OP > > test)
Set a new StatusTest for the solver.
Definition AnasaziLOBPCG.hpp:2241
Anasazi::LOBPCG::~LOBPCG
virtual ~LOBPCG()
LOBPCG destructor
Definition AnasaziLOBPCG.hpp:212
Anasazi::LOBPCG::resetNumIters
void resetNumIters()
Reset the iteration count.
Definition AnasaziLOBPCG.hpp:2265
Anasazi::LOBPCG::setBlockSize
void setBlockSize(int blockSize)
Set the blocksize to be used by the iterative solver in solving this eigenproblem.
Definition AnasaziLOBPCG.hpp:669
Anasazi::LOBPCG::getResNorms
std::vector< typename Teuchos::ScalarTraits< ScalarType >::magnitudeType > getResNorms()
Get the current residual norms.
Definition AnasaziLOBPCG.hpp:1945
Anasazi::LOBPCG::getRitzVectors
Teuchos::RCP< const MV > getRitzVectors()
Get the Ritz vectors from the previous iteration.
Definition AnasaziLOBPCG.hpp:2257
Anasazi::LOBPCG::initialize
void initialize()
Initialize the solver with the initial vectors from the eigenproblem or random data.
Definition AnasaziLOBPCG.hpp:1278
Anasazi::LOBPCG::setFullOrtho
void setFullOrtho(bool fullOrtho)
Instruct the LOBPCG iteration to use full orthogonality.
Definition AnasaziLOBPCG.hpp:1288
Anasazi::LOBPCG::getProblem
const Eigenproblem< ScalarType, MV, OP > & getProblem() const
Get a constant reference to the eigenvalue problem.
Definition AnasaziLOBPCG.hpp:2186
Anasazi::LOBPCG::currentStatus
void currentStatus(std::ostream &os)
This method requests that the solver print out its current status to screen.
Definition AnasaziLOBPCG.hpp:2104
Anasazi::LOBPCG::getRitzIndex
std::vector< int > getRitzIndex()
Get the index used for extracting Ritz vectors from getRitzVectors().
Definition AnasaziLOBPCG.hpp:2220
Anasazi::LOBPCG::getStatusTest
Teuchos::RCP< StatusTest< ScalarType, MV, OP > > getStatusTest() const
Get the current StatusTest used by the solver.
Definition AnasaziLOBPCG.hpp:2250
Anasazi::LOBPCG::getCurSubspaceDim
int getCurSubspaceDim() const
Get the dimension of the search subspace used to generate the current eigenvectors and eigenvalues.
Definition AnasaziLOBPCG.hpp:2201
Anasazi::LOBPCG::getRitzValues
std::vector< Value< ScalarType > > getRitzValues()
Get the Ritz values from the previous iteration.
Definition AnasaziLOBPCG.hpp:2229
Anasazi::LOBPCG::getRes2Norms
std::vector< typename Teuchos::ScalarTraits< ScalarType >::magnitudeType > getRes2Norms()
Get the current residual 2-norms.
Definition AnasaziLOBPCG.hpp:1959
Anasazi::LOBPCG::LOBPCG
LOBPCG(const Teuchos::RCP< Eigenproblem< ScalarType, MV, OP > > &problem, const Teuchos::RCP< SortManager< typename Teuchos::ScalarTraits< ScalarType >::magnitudeType > > &sorter, const Teuchos::RCP< OutputManager< ScalarType > > &printer, const Teuchos::RCP< StatusTest< ScalarType, MV, OP > > &tester, const Teuchos::RCP< MatOrthoManager< ScalarType, MV, OP > > &ortho, Teuchos::ParameterList &params)
LOBPCG constructor with eigenproblem, solver utilities, and parameter list of solver options.
Definition AnasaziLOBPCG.hpp:592
Anasazi::LOBPCG::setAuxVecs
void setAuxVecs(const Teuchos::Array< Teuchos::RCP< const MV > > &auxvecs)
Set the auxiliary vectors for the solver.
Definition AnasaziLOBPCG.hpp:897
Anasazi::LOBPCG::getBlockSize
int getBlockSize() const
Get the blocksize to be used by the iterative solver in solving this eigenproblem.
Definition AnasaziLOBPCG.hpp:2179
Anasazi::LOBPCG::getNumIters
int getNumIters() const
Get the current iteration count.
Definition AnasaziLOBPCG.hpp:2273
Anasazi::LOBPCG::getMaxSubspaceDim
int getMaxSubspaceDim() const
Get the maximum dimension allocated for the search subspace. For LOBPCG, this always returns 3*getBlo...
Definition AnasaziLOBPCG.hpp:2194
Anasazi::LOBPCG::getFullOrtho
bool getFullOrtho() const
Determine if the LOBPCG iteration is using full orthogonality.
Definition AnasaziLOBPCG.hpp:2164
Anasazi::LOBPCG::getState
LOBPCGState< ScalarType, MV > getState() const
Get the current state of the eigensolver.
Definition AnasaziLOBPCG.hpp:2281
Anasazi::LOBPCG::isInitialized
bool isInitialized() const
Indicates whether the solver has been initialized or not.
Definition AnasaziLOBPCG.hpp:2149
Anasazi::LOBPCG::getAuxVecs
Teuchos::Array< Teuchos::RCP< const MV > > getAuxVecs() const
Get the current auxiliary vectors.
Definition AnasaziLOBPCG.hpp:2172
Anasazi::LOBPCG::hasP
bool hasP()
Indicates whether the search direction given by getState() is valid.
Definition AnasaziLOBPCG.hpp:2157
Anasazi::Operator
Anasazi's templated virtual class for constructing an operator that can interface with the OperatorTr...
Definition AnasaziOperator.hpp:35
Anasazi::Operator::Operator
Operator()
Default constructor.
Definition AnasaziOperator.hpp:40
Anasazi::SortManager
Anasazi's templated pure virtual class for managing the sorting of approximate eigenvalues computed b...
Definition AnasaziSortManager.hpp:47
Anasazi
Namespace Anasazi contains the classes, structs, enums and utilities used by the Anasazi package.
Anasazi::LOBPCGState
Structure to contain pointers to Anasazi state variables.
Definition AnasaziLOBPCG.hpp:82
Anasazi::LOBPCGState::KH
Teuchos::RCP< const MultiVector > KH
The image of the current preconditioned residual vectors under K.
Definition AnasaziLOBPCG.hpp:110
Anasazi::LOBPCGState::X
Teuchos::RCP< const MultiVector > X
The current eigenvectors.
Definition AnasaziLOBPCG.hpp:91
Anasazi::LOBPCGState::KV
Teuchos::RCP< const MultiVector > KV
The image of the current test basis under K.
Definition AnasaziLOBPCG.hpp:86
Anasazi::LOBPCGState::MX
Teuchos::RCP< const MultiVector > MX
The image of the current eigenvectors under M, or Teuchos::null if M was not specified.
Definition AnasaziLOBPCG.hpp:95
Anasazi::LOBPCGState::V
Teuchos::RCP< const MultiVector > V
The current test basis.
Definition AnasaziLOBPCG.hpp:84
Anasazi::LOBPCGState::KX
Teuchos::RCP< const MultiVector > KX
The image of the current eigenvectors under K.
Definition AnasaziLOBPCG.hpp:93
Anasazi::LOBPCGState::R
Teuchos::RCP< const MultiVector > R
The current residual vectors.
Definition AnasaziLOBPCG.hpp:115
Anasazi::LOBPCGState::MP
Teuchos::RCP< const MultiVector > MP
The image of the current search direction under M, or Teuchos::null if M was not specified.
Definition AnasaziLOBPCG.hpp:102
Anasazi::LOBPCGState::T
Teuchos::RCP< const std::vector< typename Teuchos::ScalarTraits< ScalarType >::magnitudeType > > T
The current Ritz values.
Definition AnasaziLOBPCG.hpp:118
Anasazi::LOBPCGState::MV
Teuchos::RCP< const MultiVector > MV
The image of the current test basis under M, or Teuchos::null if M was not specified.
Definition AnasaziLOBPCG.hpp:88
Anasazi::LOBPCGState::H
Teuchos::RCP< const MultiVector > H
The current preconditioned residual vectors.
Definition AnasaziLOBPCG.hpp:108
Anasazi::LOBPCGState::P
Teuchos::RCP< const MultiVector > P
The current search direction.
Definition AnasaziLOBPCG.hpp:98
Anasazi::LOBPCGState::MH
Teuchos::RCP< const MultiVector > MH
The image of the current preconditioned residual vectors under M, or Teuchos::null if M was not speci...
Definition AnasaziLOBPCG.hpp:112
Anasazi::LOBPCGState::KP
Teuchos::RCP< const MultiVector > KP
The image of the current search direction under K.
Definition AnasaziLOBPCG.hpp:100
Generated by doxygen 1.9.8