This file is indexed.

/usr/include/trilinos/BelosICGSOrthoManager.hpp is in libtrilinos-belos-dev 12.10.1-3.

This file is owned by root:root, with mode 0o644.

The actual contents of the file can be viewed below.

   1
   2
   3
   4
   5
   6
   7
   8
   9
  10
  11
  12
  13
  14
  15
  16
  17
  18
  19
  20
  21
  22
  23
  24
  25
  26
  27
  28
  29
  30
  31
  32
  33
  34
  35
  36
  37
  38
  39
  40
  41
  42
  43
  44
  45
  46
  47
  48
  49
  50
  51
  52
  53
  54
  55
  56
  57
  58
  59
  60
  61
  62
  63
  64
  65
  66
  67
  68
  69
  70
  71
  72
  73
  74
  75
  76
  77
  78
  79
  80
  81
  82
  83
  84
  85
  86
  87
  88
  89
  90
  91
  92
  93
  94
  95
  96
  97
  98
  99
 100
 101
 102
 103
 104
 105
 106
 107
 108
 109
 110
 111
 112
 113
 114
 115
 116
 117
 118
 119
 120
 121
 122
 123
 124
 125
 126
 127
 128
 129
 130
 131
 132
 133
 134
 135
 136
 137
 138
 139
 140
 141
 142
 143
 144
 145
 146
 147
 148
 149
 150
 151
 152
 153
 154
 155
 156
 157
 158
 159
 160
 161
 162
 163
 164
 165
 166
 167
 168
 169
 170
 171
 172
 173
 174
 175
 176
 177
 178
 179
 180
 181
 182
 183
 184
 185
 186
 187
 188
 189
 190
 191
 192
 193
 194
 195
 196
 197
 198
 199
 200
 201
 202
 203
 204
 205
 206
 207
 208
 209
 210
 211
 212
 213
 214
 215
 216
 217
 218
 219
 220
 221
 222
 223
 224
 225
 226
 227
 228
 229
 230
 231
 232
 233
 234
 235
 236
 237
 238
 239
 240
 241
 242
 243
 244
 245
 246
 247
 248
 249
 250
 251
 252
 253
 254
 255
 256
 257
 258
 259
 260
 261
 262
 263
 264
 265
 266
 267
 268
 269
 270
 271
 272
 273
 274
 275
 276
 277
 278
 279
 280
 281
 282
 283
 284
 285
 286
 287
 288
 289
 290
 291
 292
 293
 294
 295
 296
 297
 298
 299
 300
 301
 302
 303
 304
 305
 306
 307
 308
 309
 310
 311
 312
 313
 314
 315
 316
 317
 318
 319
 320
 321
 322
 323
 324
 325
 326
 327
 328
 329
 330
 331
 332
 333
 334
 335
 336
 337
 338
 339
 340
 341
 342
 343
 344
 345
 346
 347
 348
 349
 350
 351
 352
 353
 354
 355
 356
 357
 358
 359
 360
 361
 362
 363
 364
 365
 366
 367
 368
 369
 370
 371
 372
 373
 374
 375
 376
 377
 378
 379
 380
 381
 382
 383
 384
 385
 386
 387
 388
 389
 390
 391
 392
 393
 394
 395
 396
 397
 398
 399
 400
 401
 402
 403
 404
 405
 406
 407
 408
 409
 410
 411
 412
 413
 414
 415
 416
 417
 418
 419
 420
 421
 422
 423
 424
 425
 426
 427
 428
 429
 430
 431
 432
 433
 434
 435
 436
 437
 438
 439
 440
 441
 442
 443
 444
 445
 446
 447
 448
 449
 450
 451
 452
 453
 454
 455
 456
 457
 458
 459
 460
 461
 462
 463
 464
 465
 466
 467
 468
 469
 470
 471
 472
 473
 474
 475
 476
 477
 478
 479
 480
 481
 482
 483
 484
 485
 486
 487
 488
 489
 490
 491
 492
 493
 494
 495
 496
 497
 498
 499
 500
 501
 502
 503
 504
 505
 506
 507
 508
 509
 510
 511
 512
 513
 514
 515
 516
 517
 518
 519
 520
 521
 522
 523
 524
 525
 526
 527
 528
 529
 530
 531
 532
 533
 534
 535
 536
 537
 538
 539
 540
 541
 542
 543
 544
 545
 546
 547
 548
 549
 550
 551
 552
 553
 554
 555
 556
 557
 558
 559
 560
 561
 562
 563
 564
 565
 566
 567
 568
 569
 570
 571
 572
 573
 574
 575
 576
 577
 578
 579
 580
 581
 582
 583
 584
 585
 586
 587
 588
 589
 590
 591
 592
 593
 594
 595
 596
 597
 598
 599
 600
 601
 602
 603
 604
 605
 606
 607
 608
 609
 610
 611
 612
 613
 614
 615
 616
 617
 618
 619
 620
 621
 622
 623
 624
 625
 626
 627
 628
 629
 630
 631
 632
 633
 634
 635
 636
 637
 638
 639
 640
 641
 642
 643
 644
 645
 646
 647
 648
 649
 650
 651
 652
 653
 654
 655
 656
 657
 658
 659
 660
 661
 662
 663
 664
 665
 666
 667
 668
 669
 670
 671
 672
 673
 674
 675
 676
 677
 678
 679
 680
 681
 682
 683
 684
 685
 686
 687
 688
 689
 690
 691
 692
 693
 694
 695
 696
 697
 698
 699
 700
 701
 702
 703
 704
 705
 706
 707
 708
 709
 710
 711
 712
 713
 714
 715
 716
 717
 718
 719
 720
 721
 722
 723
 724
 725
 726
 727
 728
 729
 730
 731
 732
 733
 734
 735
 736
 737
 738
 739
 740
 741
 742
 743
 744
 745
 746
 747
 748
 749
 750
 751
 752
 753
 754
 755
 756
 757
 758
 759
 760
 761
 762
 763
 764
 765
 766
 767
 768
 769
 770
 771
 772
 773
 774
 775
 776
 777
 778
 779
 780
 781
 782
 783
 784
 785
 786
 787
 788
 789
 790
 791
 792
 793
 794
 795
 796
 797
 798
 799
 800
 801
 802
 803
 804
 805
 806
 807
 808
 809
 810
 811
 812
 813
 814
 815
 816
 817
 818
 819
 820
 821
 822
 823
 824
 825
 826
 827
 828
 829
 830
 831
 832
 833
 834
 835
 836
 837
 838
 839
 840
 841
 842
 843
 844
 845
 846
 847
 848
 849
 850
 851
 852
 853
 854
 855
 856
 857
 858
 859
 860
 861
 862
 863
 864
 865
 866
 867
 868
 869
 870
 871
 872
 873
 874
 875
 876
 877
 878
 879
 880
 881
 882
 883
 884
 885
 886
 887
 888
 889
 890
 891
 892
 893
 894
 895
 896
 897
 898
 899
 900
 901
 902
 903
 904
 905
 906
 907
 908
 909
 910
 911
 912
 913
 914
 915
 916
 917
 918
 919
 920
 921
 922
 923
 924
 925
 926
 927
 928
 929
 930
 931
 932
 933
 934
 935
 936
 937
 938
 939
 940
 941
 942
 943
 944
 945
 946
 947
 948
 949
 950
 951
 952
 953
 954
 955
 956
 957
 958
 959
 960
 961
 962
 963
 964
 965
 966
 967
 968
 969
 970
 971
 972
 973
 974
 975
 976
 977
 978
 979
 980
 981
 982
 983
 984
 985
 986
 987
 988
 989
 990
 991
 992
 993
 994
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
//@HEADER
// ************************************************************************
//
//                 Belos: Block Linear Solvers Package
//                  Copyright 2004 Sandia Corporation
//
// Under the terms of Contract DE-AC04-94AL85000 with Sandia Corporation,
// the U.S. Government retains certain rights in this software.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// 1. Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// 2. Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
//
// 3. Neither the name of the Corporation nor the names of the
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "AS IS" AND ANY
// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL SANDIA CORPORATION OR THE
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
// Questions? Contact Michael A. Heroux (maherou@sandia.gov)
//
// ************************************************************************
//@HEADER


/*! \file BelosICGSOrthoManager.hpp
  \brief Iterated Classical Gram-Schmidt (ICGS) implementation of the Belos::OrthoManager class
*/

#ifndef BELOS_ICGS_ORTHOMANAGER_HPP
#define BELOS_ICGS_ORTHOMANAGER_HPP

/*!   \class Belos::ICGSOrthoManager
      \brief An implementation of the Belos::MatOrthoManager that performs orthogonalization
      using multiple steps of classical Gram-Schmidt.

      \author Chris Baker, Ulrich Hetmaniuk, Rich Lehoucq, and Heidi Thornquist
*/

// #define ORTHO_DEBUG

#include "BelosConfigDefs.hpp"
#include "BelosMultiVecTraits.hpp"
#include "BelosOperatorTraits.hpp"
#include "BelosMatOrthoManager.hpp"

#include "Teuchos_as.hpp"
#include "Teuchos_ParameterListAcceptorDefaultBase.hpp"
#ifdef BELOS_TEUCHOS_TIME_MONITOR
#include "Teuchos_TimeMonitor.hpp"
#endif // BELOS_TEUCHOS_TIME_MONITOR

namespace Belos {

  template<class ScalarType, class MV, class OP>
  class ICGSOrthoManager :
    public MatOrthoManager<ScalarType,MV,OP>,
    public Teuchos::ParameterListAcceptorDefaultBase
  {
  private:
    typedef typename Teuchos::ScalarTraits<ScalarType>::magnitudeType MagnitudeType;
    typedef typename Teuchos::ScalarTraits<MagnitudeType> MGT;
    typedef Teuchos::ScalarTraits<ScalarType>  SCT;
    typedef MultiVecTraits<ScalarType,MV>      MVT;
    typedef OperatorTraits<ScalarType,MV,OP>   OPT;

  public:
    //! @name Constructor/Destructor
    //@{

    //! Constructor specifying re-orthogonalization tolerance.
    ICGSOrthoManager( const std::string& label = "Belos",
                      Teuchos::RCP<const OP> Op = Teuchos::null,
                      const int max_ortho_steps = 2,
                      const MagnitudeType blk_tol = 10*MGT::squareroot( MGT::eps() ),
                      const MagnitudeType sing_tol = 10*MGT::eps() )
      : MatOrthoManager<ScalarType,MV,OP>(Op),
      max_ortho_steps_( max_ortho_steps ),
      blk_tol_( blk_tol ),
      sing_tol_( sing_tol ),
      label_( label )
    {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
      std::string orthoLabel = label_ + ": Orthogonalization";
      timerOrtho_ = Teuchos::TimeMonitor::getNewCounter(orthoLabel);

      std::string updateLabel = label_ + ": Ortho (Update)";
      timerUpdate_ = Teuchos::TimeMonitor::getNewCounter(updateLabel);

      std::string normLabel = label_ + ": Ortho (Norm)";
      timerNorm_ = Teuchos::TimeMonitor::getNewCounter(normLabel);

      std::string ipLabel = label_ + ": Ortho (Inner Product)";
      timerInnerProd_ = Teuchos::TimeMonitor::getNewCounter(ipLabel);
#endif
    }

    //! Constructor that takes a list of parameters.
    ICGSOrthoManager (const Teuchos::RCP<Teuchos::ParameterList>& plist,
                      const std::string& label = "Belos",
                      Teuchos::RCP<const OP> Op = Teuchos::null) :
      MatOrthoManager<ScalarType,MV,OP>(Op),
      max_ortho_steps_ (2),
      blk_tol_ (10 * MGT::squareroot (MGT::eps())),
      sing_tol_ (10 * MGT::eps()),
      label_ (label)
    {
      setParameterList (plist);

#ifdef BELOS_TEUCHOS_TIME_MONITOR
      std::string orthoLabel = label_ + ": Orthogonalization";
      timerOrtho_ = Teuchos::TimeMonitor::getNewCounter(orthoLabel);

      std::string updateLabel = label_ + ": Ortho (Update)";
      timerUpdate_ = Teuchos::TimeMonitor::getNewCounter(updateLabel);

      std::string normLabel = label_ + ": Ortho (Norm)";
      timerNorm_ = Teuchos::TimeMonitor::getNewCounter(normLabel);

      std::string ipLabel = label_ + ": Ortho (Inner Product)";
      timerInnerProd_ = Teuchos::TimeMonitor::getNewCounter(ipLabel);
#endif
    }

    //! Destructor
    ~ICGSOrthoManager() {}
    //@}

    //! @name Implementation of Teuchos::ParameterListAcceptorDefaultBase interface
    //@{

    void
    setParameterList (const Teuchos::RCP<Teuchos::ParameterList>& plist)
    {
      using Teuchos::Exceptions::InvalidParameterName;
      using Teuchos::ParameterList;
      using Teuchos::parameterList;
      using Teuchos::RCP;

      RCP<const ParameterList> defaultParams = getValidParameters();
      RCP<ParameterList> params;
      if (plist.is_null()) {
        params = parameterList (*defaultParams);
      } else {
        params = plist;
        // Some users might want to specify "blkTol" as "depTol".  Due
        // to this case, we don't invoke
        // validateParametersAndSetDefaults on params.  Instead, we go
        // through the parameter list one parameter at a time and look
        // for alternatives.
      }

      // Using temporary variables and fetching all values before
      // setting the output arguments ensures the strong exception
      // guarantee for this function: if an exception is thrown, no
      // externally visible side effects (in this case, setting the
      // output arguments) have taken place.
      int maxNumOrthogPasses;
      MagnitudeType blkTol;
      MagnitudeType singTol;

      try {
        maxNumOrthogPasses = params->get<int> ("maxNumOrthogPasses");
      } catch (InvalidParameterName&) {
        maxNumOrthogPasses = defaultParams->get<int> ("maxNumOrthogPasses");
        params->set ("maxNumOrthogPasses", maxNumOrthogPasses);
      }

      // Handling of the "blkTol" parameter is a special case.  This
      // is because some users may prefer to call this parameter
      // "depTol" for consistency with DGKS.  However, our default
      // parameter list calls this "blkTol", and we don't want the
      // default list's value to override the user's value.  Thus, we
      // first check the user's parameter list for both names, and
      // only then access the default parameter list.
      try {
        blkTol = params->get<MagnitudeType> ("blkTol");
      } catch (InvalidParameterName&) {
        try {
          blkTol = params->get<MagnitudeType> ("depTol");
          // "depTol" is the wrong name, so remove it and replace with
          // "blkTol".  We'll set "blkTol" below.
          params->remove ("depTol");
        } catch (InvalidParameterName&) {
          blkTol = defaultParams->get<MagnitudeType> ("blkTol");
        }
        params->set ("blkTol", blkTol);
      }

      try {
        singTol = params->get<MagnitudeType> ("singTol");
      } catch (InvalidParameterName&) {
        singTol = defaultParams->get<MagnitudeType> ("singTol");
        params->set ("singTol", singTol);
      }

      max_ortho_steps_ = maxNumOrthogPasses;
      blk_tol_ = blkTol;
      sing_tol_ = singTol;

      setMyParamList (params);
    }

    Teuchos::RCP<const Teuchos::ParameterList>
    getValidParameters () const
    {
      using Teuchos::as;
      using Teuchos::ParameterList;
      using Teuchos::parameterList;
      using Teuchos::RCP;

      if (defaultParams_.is_null()) {
        RCP<ParameterList> params = parameterList ("ICGS");

        // Default parameter values for ICGS orthogonalization.
        // Documentation will be embedded in the parameter list.
        const int defaultMaxNumOrthogPasses = 2;
        const MagnitudeType eps = MGT::eps();
        const MagnitudeType defaultBlkTol =
          as<MagnitudeType> (10) * MGT::squareroot (eps);
        const MagnitudeType defaultSingTol = as<MagnitudeType> (10) * eps;

        params->set ("maxNumOrthogPasses", defaultMaxNumOrthogPasses,
                     "Maximum number of orthogonalization passes (includes the "
                     "first).  Default is 2, since \"twice is enough\" for Krylov "
                     "methods.");
        params->set ("blkTol", defaultBlkTol, "Block reorthogonalization "
                     "threshhold.");
        params->set ("singTol", defaultSingTol, "Singular block detection "
                     "threshold.");
        defaultParams_ = params;
      }
      return defaultParams_;
    }
    //@}

    /// \brief "Fast" but possibly unsafe or less accurate parameters.
    ///
    /// Use this parameter list when you care more about speed than
    /// accuracy of the orthogonalization.
    Teuchos::RCP<const Teuchos::ParameterList>
    getFastParameters () const
    {
      using Teuchos::as;
      using Teuchos::ParameterList;
      using Teuchos::parameterList;
      using Teuchos::RCP;

      RCP<const ParameterList> defaultParams = getValidParameters ();
      // Start with a clone of the default parameters.
      RCP<ParameterList> params = parameterList (*defaultParams);

      const int maxBlkOrtho = 1; // No block reorthogonalization
      const MagnitudeType blkTol = MGT::zero();
      const MagnitudeType singTol = MGT::zero();

      params->set ("maxNumOrthogPasses", maxBlkOrtho);
      params->set ("blkTol", blkTol);
      params->set ("singTol", singTol);

      return params;
    }

    //! @name Accessor routines
    //@{

    //! Set parameter for block re-orthogonalization threshhold.
    void setBlkTol( const MagnitudeType blk_tol ) {
      // Update the parameter list as well.
      Teuchos::RCP<Teuchos::ParameterList> params = getNonconstParameterList();
      if (! params.is_null()) {
        // If it's null, then we haven't called setParameterList()
        // yet.  It's entirely possible to construct the parameter
        // list on demand, so we don't try to create the parameter
        // list here.
        params->set ("blkTol", blk_tol);
      }
      blk_tol_ = blk_tol;
    }

    //! Set parameter for singular block detection.
    void setSingTol( const MagnitudeType sing_tol ) {
      // Update the parameter list as well.
      Teuchos::RCP<Teuchos::ParameterList> params = getNonconstParameterList();
      if (! params.is_null()) {
        // If it's null, then we haven't called setParameterList()
        // yet.  It's entirely possible to construct the parameter
        // list on demand, so we don't try to create the parameter
        // list here.
        params->set ("singTol", sing_tol);
      }
      sing_tol_ = sing_tol;
    }

    //! Return parameter for block re-orthogonalization threshhold.
    MagnitudeType getBlkTol() const { return blk_tol_; }

    //! Return parameter for singular block detection.
    MagnitudeType getSingTol() const { return sing_tol_; }

    //@}


    //! @name Orthogonalization methods
    //@{

    /*! \brief Given a list of (mutually and internally) orthonormal bases \c Q, this method
     * takes a multivector \c X and projects it onto the space orthogonal to the individual <tt>Q[i]</tt>,
     * optionally returning the coefficients of \c X for the individual <tt>Q[i]</tt>. All of this is done with respect
     * to the inner product innerProd().
     *
     * After calling this routine, \c X will be orthogonal to each of the \c <tt>Q[i]</tt>.
     *
     * The method uses either one or two steps of classical Gram-Schmidt. The algebraically
     * equivalent projection matrix is \f$P_Q = I - Q Q^H Op\f$, if \c Op is the matrix specified for
     * use in the inner product. Note, this is not an orthogonal projector.
     *
     @param X [in/out] The multivector to be modified.
       On output, \c X will be orthogonal to <tt>Q[i]</tt> with respect to innerProd().

     @param MX [in/out] The image of \c X under the operator \c Op.
       If \f$ MX != 0\f$: On input, this is expected to be consistent with \c X. On output, this is updated consistent with updates to \c X.
       If \f$ MX == 0\f$ or \f$ Op == 0\f$: \c MX is not referenced.

     @param C [out] The coefficients of \c X in the \c *Q[i], with respect to innerProd(). If <tt>C[i]</tt> is a non-null pointer
       and \c *C[i] matches the dimensions of \c X and \c *Q[i], then the coefficients computed during the orthogonalization
       routine will be stored in the matrix \c *C[i]. If <tt>C[i]</tt> is a non-null pointer whose size does not match the dimensions of
       \c X and \c *Q[i], then a std::invalid_argument std::exception will be thrown. Otherwise, if <tt>C.size() < i</tt> or <tt>C[i]</tt> is a null
       pointer, then the orthogonalization manager will declare storage for the coefficients and the user will not have access to them.

     @param Q [in] A list of multivector bases specifying the subspaces to be orthogonalized against. Each <tt>Q[i]</tt> is assumed to have
     orthonormal columns, and the <tt>Q[i]</tt> are assumed to be mutually orthogonal.
    */
    void project ( MV &X, Teuchos::RCP<MV> MX,
                   Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                   Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const;


    /*! \brief This method calls project(X,Teuchos::null,C,Q); see documentation for that function.
    */
    void project ( MV &X,
                   Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                   Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const {
      project(X,Teuchos::null,C,Q);
    }



    /*! \brief This method takes a multivector \c X and attempts to compute an orthonormal basis for \f$colspan(X)\f$, with respect to innerProd().
     *
     * The method uses classical Gram-Schmidt, so that the coefficient matrix \c B is upper triangular.
     *
     * This routine returns an integer \c rank stating the rank of the computed basis. If \c X does not have full rank and the normalize() routine does
     * not attempt to augment the subspace, then \c rank may be smaller than the number of columns in \c X. In this case, only the first \c rank columns of
     * output \c X and first \c rank rows of \c B will be valid.
     *
     * The method attempts to find a basis with dimension the same as the number of columns in \c X. It does this by augmenting linearly dependant
     * vectors in \c X with random directions. A finite number of these attempts will be made; therefore, it is possible that the dimension of the
     * computed basis is less than the number of vectors in \c X.
     *
     @param X [in/out] The multivector to the modified.
       On output, \c X will have some number of orthonormal columns (with respect to innerProd()).

     @param MX [in/out] The image of \c X under the operator \c Op.
       If \f$ MX != 0\f$: On input, this is expected to be consistent with \c X. On output, this is updated consistent with updates to \c X.
       If \f$ MX == 0\f$ or \f$ Op == 0\f$: \c MX is not referenced.

     @param B [out] The coefficients of the original \c X with respect to the computed basis. The first rows in \c B
            corresponding to the valid columns in \c X will be upper triangular.

     @return Rank of the basis computed by this method.
    */
    int normalize ( MV &X, Teuchos::RCP<MV> MX,
                    Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > B) const;


    /*! \brief This method calls normalize(X,Teuchos::null,B); see documentation for that function.
    */
    int normalize ( MV &X, Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > B ) const {
      return normalize(X,Teuchos::null,B);
    }

  protected:

    /*! \brief Given a set of bases <tt>Q[i]</tt> and a multivector \c X, this method computes an orthonormal basis for \f$colspan(X) - \sum_i colspan(Q[i])\f$.
     *
     *  This routine returns an integer \c rank stating the rank of the computed basis. If the subspace \f$colspan(X) - \sum_i colspan(Q[i])\f$ does not
     *  have dimension as large as the number of columns of \c X and the orthogonalization manager doe not attempt to augment the subspace, then \c rank
     *  may be smaller than the number of columns of \c X. In this case, only the first \c rank columns of output \c X and first \c rank rows of \c B will
     *  be valid.
     *
     * The method attempts to find a basis with dimension the same as the number of columns in \c X. It does this by augmenting linearly dependant
     * vectors with random directions. A finite number of these attempts will be made; therefore, it is possible that the dimension of the
     * computed basis is less than the number of vectors in \c X.
     *
     @param X [in/out] The multivector to the modified.
       On output, the relevant rows of \c X will be orthogonal to the <tt>Q[i]</tt> and will have orthonormal columns (with respect to innerProd()).

     @param MX [in/out] The image of \c X under the operator \c Op.
       If \f$ MX != 0\f$: On input, this is expected to be consistent with \c X. On output, this is updated consistent with updates to \c X.
       If \f$ MX == 0\f$ or \f$ Op == 0\f$: \c MX is not referenced.

     @param C [out] The coefficients of the original \c X in the \c
     *Q[i], with respect to innerProd(). If <tt>C[i]</tt> is a
     non-null pointer and \c *C[i] matches the dimensions of \c X and
     \c *Q[i], then the coefficients computed during the
     orthogonalization routine will be stored in the matrix \c
     *C[i]. If <tt>C[i]</tt> is a non-null pointer whose size does not
     match the dimensions of \c X and \c *Q[i], then *C[i] will first
     be resized to the correct size.  This will destroy the original
     contents of the matrix.  (This is a change from previous
     behavior, in which a std::invalid_argument exception was thrown
     if *C[i] was of the wrong size.)  Otherwise, if <tt>C.size() <
     i<\tt> or <tt>C[i]</tt> is a null pointer, then the
     orthogonalization manager will declare storage for the
     coefficients and the user will not have access to them.

     @param B [out] The coefficients of the original \c X with respect to the computed basis. The first rows in \c B
            corresponding to the valid columns in \c X will be upper triangular.

     @param Q [in] A list of multivector bases specifying the subspaces to be orthogonalized against. Each <tt>Q[i]</tt> is assumed to have
     orthonormal columns, and the <tt>Q[i]</tt> are assumed to be mutually orthogonal.

     @return Rank of the basis computed by this method.
    */
    virtual int
    projectAndNormalizeWithMxImpl (MV &X,
                                   Teuchos::RCP<MV> MX,
                                   Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                                   Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > B,
                                   Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const;

  public:
    //@}
    //! @name Error methods
    //@{

    /*! \brief This method computes the error in orthonormality of a multivector, measured
     * as the Frobenius norm of the difference <tt>innerProd(X,Y) - I</tt>.
     */
    typename Teuchos::ScalarTraits<ScalarType>::magnitudeType
    orthonormError(const MV &X) const {
      return orthonormError(X,Teuchos::null);
    }

    /*! \brief This method computes the error in orthonormality of a multivector, measured
     * as the Frobenius norm of the difference <tt>innerProd(X,Y) - I</tt>.
     *  The method has the option of exploiting a caller-provided \c MX.
     */
    typename Teuchos::ScalarTraits<ScalarType>::magnitudeType
    orthonormError(const MV &X, Teuchos::RCP<const MV> MX) const;

    /*! \brief This method computes the error in orthogonality of two multivectors, measured
     * as the Frobenius norm of <tt>innerProd(X,Y)</tt>.
     */
    typename Teuchos::ScalarTraits<ScalarType>::magnitudeType
    orthogError(const MV &X1, const MV &X2) const {
      return orthogError(X1,Teuchos::null,X2);
    }

    /*! \brief This method computes the error in orthogonality of two multivectors, measured
     * as the Frobenius norm of <tt>innerProd(X,Y)</tt>.
     *  The method has the option of exploiting a caller-provided \c MX.
     */
    typename Teuchos::ScalarTraits<ScalarType>::magnitudeType
    orthogError(const MV &X1, Teuchos::RCP<const MV> MX1, const MV &X2) const;

    //@}

    //! @name Label methods
    //@{

    /*! \brief This method sets the label used by the timers in the orthogonalization manager.
     */
    void setLabel(const std::string& label);

    /*! \brief This method returns the label being used by the timers in the orthogonalization manager.
     */
    const std::string& getLabel() const { return label_; }

    //@}

  private:
    //! Max number of (re)orthogonalization steps, including the first.
    int max_ortho_steps_;
    //! Block reorthogonalization threshold.
    MagnitudeType blk_tol_;
    //! Singular block detection threshold.
    MagnitudeType sing_tol_;

    //! Label for timers.
    std::string label_;
#ifdef BELOS_TEUCHOS_TIME_MONITOR
    Teuchos::RCP<Teuchos::Time> timerOrtho_, timerUpdate_,
      timerNorm_, timerScale_, timerInnerProd_;
#endif // BELOS_TEUCHOS_TIME_MONITOR

    //! Default parameter list.
    mutable Teuchos::RCP<Teuchos::ParameterList> defaultParams_;

    //! Routine to find an orthonormal basis for X
    int findBasis(MV &X, Teuchos::RCP<MV> MX,
                  Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > C,
                  bool completeBasis, int howMany = -1 ) const;

    //! Routine to compute the block orthogonalization
    bool blkOrtho1 ( MV &X, Teuchos::RCP<MV> MX,
                     Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                     Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const;

    //! Routine to compute the block orthogonalization
    bool blkOrtho ( MV &X, Teuchos::RCP<MV> MX,
                    Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                    Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const;

    /// Project X against QQ and normalize X, one vector at a time
    ///
    /// \note QQ is called QQ, rather than Q, because we convert it
    ///   internally from an ArrayView to an Array (named Q inside).
    ///   This is because the C++ compiler doesn't know how to do type
    ///   inference (Array has a constructor that takes an ArrayView
    ///   input).  This routine wants an Array rather than an
    ///   ArrayView internally, because it likes to add (via
    ///   push_back()) and remove (via resize()) elements to the Q
    ///   array.  Remember that Arrays can be passed by value, just
    ///   like std::vector objects, so this routine can add whatever
    ///   it likes to the Q array without changing it from the
    ///   caller's perspective.
    int blkOrthoSing ( MV &X, Teuchos::RCP<MV> MX,
                       Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                       Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > B,
                       Teuchos::ArrayView<Teuchos::RCP<const MV> > QQ) const;
  };

  //////////////////////////////////////////////////////////////////////////////////////////////////
  // Set the label for this orthogonalization manager and create new timers if it's changed
  template<class ScalarType, class MV, class OP>
  void ICGSOrthoManager<ScalarType,MV,OP>::setLabel(const std::string& label)
  {
    if (label != label_) {
      label_ = label;
      std::string orthoLabel = label_ + ": Orthogonalization";
#ifdef BELOS_TEUCHOS_TIME_MONITOR
      timerOrtho_ = Teuchos::TimeMonitor::getNewCounter(orthoLabel);
#endif

      std::string updateLabel = label_ + ": Ortho (Update)";
#ifdef BELOS_TEUCHOS_TIME_MONITOR
      timerUpdate_ = Teuchos::TimeMonitor::getNewCounter(updateLabel);
#endif

      std::string normLabel = label_ + ": Ortho (Norm)";
#ifdef BELOS_TEUCHOS_TIME_MONITOR
      timerNorm_ = Teuchos::TimeMonitor::getNewCounter(normLabel);
#endif

      std::string ipLabel = label_ + ": Ortho (Inner Product)";
#ifdef BELOS_TEUCHOS_TIME_MONITOR
      timerInnerProd_ = Teuchos::TimeMonitor::getNewCounter(ipLabel);
#endif
    }
  }

  //////////////////////////////////////////////////////////////////////////////////////////////////
  // Compute the distance from orthonormality
  template<class ScalarType, class MV, class OP>
  typename Teuchos::ScalarTraits<ScalarType>::magnitudeType
  ICGSOrthoManager<ScalarType,MV,OP>::orthonormError(const MV &X, Teuchos::RCP<const MV> MX) const {
    const ScalarType ONE = SCT::one();
    int rank = MVT::GetNumberVecs(X);
    Teuchos::SerialDenseMatrix<int,ScalarType> xTx(rank,rank);
    MatOrthoManager<ScalarType,MV,OP>::innerProd(X,X,MX,xTx);
    for (int i=0; i<rank; i++) {
      xTx(i,i) -= ONE;
    }
    return xTx.normFrobenius();
  }

  //////////////////////////////////////////////////////////////////////////////////////////////////
  // Compute the distance from orthogonality
  template<class ScalarType, class MV, class OP>
  typename Teuchos::ScalarTraits<ScalarType>::magnitudeType
  ICGSOrthoManager<ScalarType,MV,OP>::orthogError(const MV &X1, Teuchos::RCP<const MV> MX1, const MV &X2) const {
    int r1 = MVT::GetNumberVecs(X1);
    int r2  = MVT::GetNumberVecs(X2);
    Teuchos::SerialDenseMatrix<int,ScalarType> xTx(r2,r1);
    MatOrthoManager<ScalarType,MV,OP>::innerProd(X2,X1,MX1,xTx);
    return xTx.normFrobenius();
  }

  //////////////////////////////////////////////////////////////////////////////////////////////////
  // Find an Op-orthonormal basis for span(X) - span(W)
  template<class ScalarType, class MV, class OP>
  int
  ICGSOrthoManager<ScalarType, MV, OP>::
  projectAndNormalizeWithMxImpl (MV &X,
                                 Teuchos::RCP<MV> MX,
                                 Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                                 Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > B,
                                 Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const
  {
    using Teuchos::Array;
    using Teuchos::null;
    using Teuchos::is_null;
    using Teuchos::RCP;
    using Teuchos::rcp;
    using Teuchos::SerialDenseMatrix;
    typedef SerialDenseMatrix< int, ScalarType > serial_dense_matrix_type;
    typedef typename Array< RCP< const MV > >::size_type size_type;

#ifdef BELOS_TEUCHOS_TIME_MONITOR
    Teuchos::TimeMonitor orthotimer(*timerOrtho_);
#endif

    ScalarType    ONE  = SCT::one();
    //ScalarType    ZERO  = SCT::zero();

    int nq = Q.size();
    int xc = MVT::GetNumberVecs( X );
    ptrdiff_t xr = MVT::GetGlobalLength( X );
    int rank = xc;

    // If the user doesn't want to store the normalization
    // coefficients, allocate some local memory for them.  This will
    // go away at the end of this method.
    if (is_null (B)) {
      B = rcp (new serial_dense_matrix_type (xc, xc));
    }
    // Likewise, if the user doesn't want to store the projection
    // coefficients, allocate some local memory for them.  Also make
    // sure that all the entries of C are the right size.  We're going
    // to overwrite them anyway, so we don't have to worry about the
    // contents (other than to resize them if they are the wrong
    // size).
    if (C.size() < nq)
      C.resize (nq);
    for (size_type k = 0; k < nq; ++k)
      {
        const int numRows = MVT::GetNumberVecs (*Q[k]);
        const int numCols = xc; // Number of vectors in X

        if (is_null (C[k]))
          C[k] = rcp (new serial_dense_matrix_type (numRows, numCols));
        else if (C[k]->numRows() != numRows || C[k]->numCols() != numCols)
        {
          int err = C[k]->reshape (numRows, numCols);
          TEUCHOS_TEST_FOR_EXCEPTION(err != 0, std::runtime_error,
              "IMGS orthogonalization: failed to reshape "
              "C[" << k << "] (the array of block "
              "coefficients resulting from projecting X "
              "against Q[1:" << nq << "]).");
        }
      }

    /******   DO NOT MODIFY *MX IF _hasOp == false   ******/
    if (this->_hasOp) {
      if (MX == Teuchos::null) {
        // we need to allocate space for MX
        MX = MVT::Clone(X,MVT::GetNumberVecs(X));
        OPT::Apply(*(this->_Op),X,*MX);
      }
    }
    else {
      // Op == I  -->  MX = X (ignore it if the user passed it in)
      MX = Teuchos::rcp( &X, false );
    }

    int mxc = MVT::GetNumberVecs( *MX );
    ptrdiff_t mxr = MVT::GetGlobalLength( *MX );

    // short-circuit
    TEUCHOS_TEST_FOR_EXCEPTION( xc == 0 || xr == 0, std::invalid_argument, "Belos::ICGSOrthoManager::projectAndNormalize(): X must be non-empty" );

    int numbas = 0;
    for (int i=0; i<nq; i++) {
      numbas += MVT::GetNumberVecs( *Q[i] );
    }

    // check size of B
    TEUCHOS_TEST_FOR_EXCEPTION( B->numRows() != xc || B->numCols() != xc, std::invalid_argument,
                        "Belos::ICGSOrthoManager::projectAndNormalize(): Size of X must be consistant with size of B" );
    // check size of X and MX
    TEUCHOS_TEST_FOR_EXCEPTION( xc<0 || xr<0 || mxc<0 || mxr<0, std::invalid_argument,
                        "Belos::ICGSOrthoManager::projectAndNormalize(): MVT returned negative dimensions for X,MX" );
    // check size of X w.r.t. MX
    TEUCHOS_TEST_FOR_EXCEPTION( xc!=mxc || xr!=mxr, std::invalid_argument,
                        "Belos::ICGSOrthoManager::projectAndNormalize(): Size of X must be consistant with size of MX" );
    // check feasibility
    //TEUCHOS_TEST_FOR_EXCEPTION( numbas+xc > xr, std::invalid_argument,
    //                    "Belos::ICGSOrthoManager::projectAndNormalize(): Orthogonality constraints not feasible" );

    // Some flags for checking dependency returns from the internal orthogonalization methods
    bool dep_flg = false;

    // Make a temporary copy of X and MX, just in case a block dependency is detected.
    Teuchos::RCP<MV> tmpX, tmpMX;
    tmpX = MVT::CloneCopy(X);
    if (this->_hasOp) {
      tmpMX = MVT::CloneCopy(*MX);
    }

    if (xc == 1) {

      // Use the cheaper block orthogonalization.
      // NOTE: Don't check for dependencies because the update has one vector.
      dep_flg = blkOrtho1( X, MX, C, Q );

      // Normalize the new block X
      {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
        Teuchos::TimeMonitor normTimer( *timerNorm_ );
#endif
        if ( B == Teuchos::null ) {
          B = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>(xc,xc) );
        }
        std::vector<ScalarType> diag(xc);
        MVT::MvDot( X, *MX, diag );
        (*B)(0,0) = SCT::squareroot(SCT::magnitude(diag[0]));
        rank = 1;
        MVT::MvScale( X, ONE/(*B)(0,0) );
        if (this->_hasOp) {
          // Update MXj.
          MVT::MvScale( *MX, ONE/(*B)(0,0) );
        }
      }

    }
    else {

      // Use the cheaper block orthogonalization.
      dep_flg = blkOrtho( X, MX, C, Q );

      // If a dependency has been detected in this block, then perform
      // the more expensive nonblock (single vector at a time)
      // orthogonalization.
      if (dep_flg) {
        rank = blkOrthoSing( *tmpX, tmpMX, C, B, Q );

        // Copy tmpX back into X.
        MVT::Assign( *tmpX, X );
        if (this->_hasOp) {
          MVT::Assign( *tmpMX, *MX );
        }
      }
      else {
        // There is no dependency, so orthonormalize new block X
        rank = findBasis( X, MX, B, false );
        if (rank < xc) {
          // A dependency was found during orthonormalization of X,
          // rerun orthogonalization using more expensive nonblock
          // orthogonalization.
          rank = blkOrthoSing( *tmpX, tmpMX, C, B, Q );

          // Copy tmpX back into X.
          MVT::Assign( *tmpX, X );
          if (this->_hasOp) {
            MVT::Assign( *tmpMX, *MX );
          }
        }
      }
    } // if (xc == 1) {

    // this should not raise an std::exception; but our post-conditions oblige us to check
    TEUCHOS_TEST_FOR_EXCEPTION( rank > xc || rank < 0, std::logic_error,
                        "Belos::ICGSOrthoManager::projectAndNormalize(): Debug error in rank variable." );

    // Return the rank of X.
    return rank;
  }



  //////////////////////////////////////////////////////////////////////////////////////////////////
  // Find an Op-orthonormal basis for span(X), with rank numvectors(X)
  template<class ScalarType, class MV, class OP>
  int ICGSOrthoManager<ScalarType, MV, OP>::normalize(
                                MV &X, Teuchos::RCP<MV> MX,
                                Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > B ) const {

#ifdef BELOS_TEUCHOS_TIME_MONITOR
    Teuchos::TimeMonitor orthotimer(*timerOrtho_);
#endif

    // call findBasis, with the instruction to try to generate a basis of rank numvecs(X)
    return findBasis(X, MX, B, true);
  }



  //////////////////////////////////////////////////////////////////////////////////////////////////
  template<class ScalarType, class MV, class OP>
  void ICGSOrthoManager<ScalarType, MV, OP>::project(
                          MV &X, Teuchos::RCP<MV> MX,
                          Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                          Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const {
    // For the inner product defined by the operator Op or the identity (Op == 0)
    //   -> Orthogonalize X against each Q[i]
    // Modify MX accordingly
    //
    // Note that when Op is 0, MX is not referenced
    //
    // Parameter variables
    //
    // X  : Vectors to be transformed
    //
    // MX : Image of the block of vectors X by the mass matrix
    //
    // Q  : Bases to orthogonalize against. These are assumed orthonormal, mutually and independently.
    //

#ifdef BELOS_TEUCHOS_TIME_MONITOR
    Teuchos::TimeMonitor orthotimer(*timerOrtho_);
#endif

    int xc = MVT::GetNumberVecs( X );
    ptrdiff_t xr = MVT::GetGlobalLength( X );
    int nq = Q.size();
    std::vector<int> qcs(nq);
    // short-circuit
    if (nq == 0 || xc == 0 || xr == 0) {
      return;
    }
    ptrdiff_t qr = MVT::GetGlobalLength ( *Q[0] );
    // if we don't have enough C, expand it with null references
    // if we have too many, resize to throw away the latter ones
    // if we have exactly as many as we have Q, this call has no effect
    C.resize(nq);


    /******   DO NOT MODIFY *MX IF _hasOp == false   ******/
    if (this->_hasOp) {
      if (MX == Teuchos::null) {
        // we need to allocate space for MX
        MX = MVT::Clone(X,MVT::GetNumberVecs(X));
        OPT::Apply(*(this->_Op),X,*MX);
      }
    }
    else {
      // Op == I  -->  MX = X (ignore it if the user passed it in)
      MX = Teuchos::rcp( &X, false );
    }
    int mxc = MVT::GetNumberVecs( *MX );
    ptrdiff_t mxr = MVT::GetGlobalLength( *MX );

    // check size of X and Q w.r.t. common sense
    TEUCHOS_TEST_FOR_EXCEPTION( xc<0 || xr<0 || mxc<0 || mxr<0, std::invalid_argument,
                        "Belos::ICGSOrthoManager::project(): MVT returned negative dimensions for X,MX" );
    // check size of X w.r.t. MX and Q
    TEUCHOS_TEST_FOR_EXCEPTION( xc!=mxc || xr!=mxr || xr!=qr, std::invalid_argument,
                        "Belos::ICGSOrthoManager::project(): Size of X not consistant with MX,Q" );

    // tally up size of all Q and check/allocate C
    int baslen = 0;
    for (int i=0; i<nq; i++) {
      TEUCHOS_TEST_FOR_EXCEPTION( MVT::GetGlobalLength( *Q[i] ) != qr, std::invalid_argument,
                          "Belos::ICGSOrthoManager::project(): Q lengths not mutually consistant" );
      qcs[i] = MVT::GetNumberVecs( *Q[i] );
      TEUCHOS_TEST_FOR_EXCEPTION( qr < qcs[i], std::invalid_argument,
                          "Belos::ICGSOrthoManager::project(): Q has less rows than columns" );
      baslen += qcs[i];

      // check size of C[i]
      if ( C[i] == Teuchos::null ) {
        C[i] = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>(qcs[i],xc) );
      }
      else {
        TEUCHOS_TEST_FOR_EXCEPTION( C[i]->numRows() != qcs[i] || C[i]->numCols() != xc , std::invalid_argument,
                           "Belos::ICGSOrthoManager::project(): Size of Q not consistant with size of C" );
      }
    }

    // Use the cheaper block orthogonalization, don't check for rank deficiency.
    blkOrtho( X, MX, C, Q );

  }

  //////////////////////////////////////////////////////////////////////////////////////////////////
  // Find an Op-orthonormal basis for span(X), with the option of extending the subspace so that
  // the rank is numvectors(X)
  template<class ScalarType, class MV, class OP>
  int
  ICGSOrthoManager<ScalarType, MV, OP>::
  findBasis (MV &X,
             Teuchos::RCP<MV> MX,
             Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > B,
             bool completeBasis,
             int howMany) const
  {
    // For the inner product defined by the operator Op or the identity (Op == 0)
    //   -> Orthonormalize X
    // Modify MX accordingly
    //
    // Note that when Op is 0, MX is not referenced
    //
    // Parameter variables
    //
    // X  : Vectors to be orthonormalized
    // MX : Image of the multivector X under the operator Op
    // Op  : Pointer to the operator for the inner product
    //
    using Teuchos::as;
#ifdef BELOS_TEUCHOS_TIME_MONITOR
    Teuchos::TimeMonitor normTimer (*timerNorm_);
#endif // BELOS_TEUCHOS_TIME_MONITOR

    const ScalarType ONE = SCT::one ();
    const MagnitudeType ZERO = SCT::magnitude (SCT::zero ());

    const int xc = MVT::GetNumberVecs (X);
    const ptrdiff_t xr = MVT::GetGlobalLength (X);

    if (howMany == -1) {
      howMany = xc;
    }

    /*******************************************************
     *  If _hasOp == false, we will not reference MX below *
     *******************************************************/

    // if Op==null, MX == X (via pointer)
    // Otherwise, either the user passed in MX or we will allocated and compute it
    if (this->_hasOp) {
      if (MX == Teuchos::null) {
        // we need to allocate space for MX
        MX = MVT::Clone(X,xc);
        OPT::Apply(*(this->_Op),X,*MX);
      }
    }

    /* if the user doesn't want to store the coefficienets,
     * allocate some local memory for them
     */
    if ( B == Teuchos::null ) {
      B = Teuchos::rcp( new Teuchos::SerialDenseMatrix<int,ScalarType>(xc,xc) );
    }

    const int mxc = (this->_hasOp) ? MVT::GetNumberVecs( *MX ) : xc;
    const ptrdiff_t mxr = (this->_hasOp) ? MVT::GetGlobalLength( *MX ) : xr;

    // check size of C, B
    TEUCHOS_TEST_FOR_EXCEPTION( xc == 0 || xr == 0, std::invalid_argument,
                        "Belos::ICGSOrthoManager::findBasis(): X must be non-empty" );
    TEUCHOS_TEST_FOR_EXCEPTION( B->numRows() != xc || B->numCols() != xc, std::invalid_argument,
                        "Belos::ICGSOrthoManager::findBasis(): Size of X not consistant with size of B" );
    TEUCHOS_TEST_FOR_EXCEPTION( xc != mxc || xr != mxr, std::invalid_argument,
                        "Belos::ICGSOrthoManager::findBasis(): Size of X not consistant with size of MX" );
    TEUCHOS_TEST_FOR_EXCEPTION( as<ptrdiff_t> (xc) > xr, std::invalid_argument,
                        "Belos::ICGSOrthoManager::findBasis(): Size of X not feasible for normalization" );
    TEUCHOS_TEST_FOR_EXCEPTION( howMany < 0 || howMany > xc, std::invalid_argument,
                        "Belos::ICGSOrthoManager::findBasis(): Invalid howMany parameter" );

    /* xstart is which column we are starting the process with, based on howMany
     * columns before xstart are assumed to be Op-orthonormal already
     */
    int xstart = xc - howMany;

    for (int j = xstart; j < xc; j++) {

      // numX is
      // * number of currently orthonormal columns of X
      // * the index of the current column of X
      int numX = j;
      bool addVec = false;

      // Get a view of the vector currently being worked on.
      std::vector<int> index(1);
      index[0] = numX;
      Teuchos::RCP<MV> Xj = MVT::CloneViewNonConst( X, index );
      Teuchos::RCP<MV> MXj;
      if (this->_hasOp) {
        // MXj is a view of the current vector in MX
        MXj = MVT::CloneViewNonConst( *MX, index );
      }
      else {
        // MXj is a pointer to Xj, and MUST NOT be modified
        MXj = Xj;
      }

      // Get a view of the previous vectors.
      std::vector<int> prev_idx( numX );
      Teuchos::RCP<const MV> prevX, prevMX;

      if (numX > 0) {
        for (int i=0; i<numX; i++) {
          prev_idx[i] = i;
        }
        prevX = MVT::CloneView( X, prev_idx );
        if (this->_hasOp) {
          prevMX = MVT::CloneView( *MX, prev_idx );
        }
      }

      // Make storage for these Gram-Schmidt iterations.
      Teuchos::SerialDenseMatrix<int,ScalarType> product(numX, 1);
      std::vector<ScalarType> oldDot( 1 ), newDot( 1 );
      //
      // Save old MXj vector and compute Op-norm
      //
      Teuchos::RCP<MV> oldMXj = MVT::CloneCopy( *MXj );
      MVT::MvDot( *Xj, *MXj, oldDot );
      // Xj^H Op Xj should be real and positive, by the hermitian positive definiteness of Op
      TEUCHOS_TEST_FOR_EXCEPTION( SCT::real(oldDot[0]) < ZERO, OrthoError,
          "Belos::ICGSOrthoManager::findBasis(): Negative definiteness discovered in inner product" );

      if (numX > 0) {

        Teuchos::SerialDenseMatrix<int,ScalarType> P2(numX,1);

        for (int i=0; i<max_ortho_steps_; ++i) {

          // product <- prevX^T MXj
          {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
            Teuchos::TimeMonitor innerProdTimer( *timerInnerProd_ );
#endif
            MatOrthoManager<ScalarType,MV,OP>::innerProd(*prevX,*Xj,MXj,P2);
          }

          // Xj <- Xj - prevX prevX^T MXj
          //     = Xj - prevX product
          {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
            Teuchos::TimeMonitor updateTimer( *timerUpdate_ );
#endif
            MVT::MvTimesMatAddMv( -ONE, *prevX, P2, ONE, *Xj );
          }

          // Update MXj
          if (this->_hasOp) {
            // MXj <- Op*Xj_new
            //      = Op*(Xj_old - prevX prevX^T MXj)
            //      = MXj - prevMX product
#ifdef BELOS_TEUCHOS_TIME_MONITOR
            Teuchos::TimeMonitor updateTimer( *timerUpdate_ );
#endif
            MVT::MvTimesMatAddMv( -ONE, *prevMX, P2, ONE, *MXj );
          }

          // Set coefficients
          if ( i==0 )
            product = P2;
          else
            product += P2;
        }

      } // if (numX > 0)

      // Compute Op-norm with old MXj
      MVT::MvDot( *Xj, *oldMXj, newDot );

      // Check to see if the new vector is dependent.
      if (completeBasis) {
        //
        // We need a complete basis, so add random vectors if necessary
        //
        if ( SCT::magnitude(newDot[0]) < SCT::magnitude(sing_tol_*oldDot[0]) ) {

          // Add a random vector and orthogonalize it against previous vectors in block.
          addVec = true;
#ifdef ORTHO_DEBUG
          std::cout << "Belos::ICGSOrthoManager::findBasis() --> Random for column " << numX << std::endl;
#endif
          //
          Teuchos::RCP<MV> tempXj = MVT::Clone( X, 1 );
          Teuchos::RCP<MV> tempMXj;
          MVT::MvRandom( *tempXj );
          if (this->_hasOp) {
            tempMXj = MVT::Clone( X, 1 );
            OPT::Apply( *(this->_Op), *tempXj, *tempMXj );
          }
          else {
            tempMXj = tempXj;
          }
          MVT::MvDot( *tempXj, *tempMXj, oldDot );
          //
          for (int num_orth=0; num_orth<max_ortho_steps_; num_orth++){
            {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
              Teuchos::TimeMonitor innerProdTimer( *timerInnerProd_ );
#endif
              MatOrthoManager<ScalarType,MV,OP>::innerProd(*prevX,*tempXj,tempMXj,product);
            }
            {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
              Teuchos::TimeMonitor updateTimer( *timerUpdate_ );
#endif
              MVT::MvTimesMatAddMv( -ONE, *prevX, product, ONE, *tempXj );
            }
            if (this->_hasOp) {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
              Teuchos::TimeMonitor updateTimer( *timerUpdate_ );
#endif
              MVT::MvTimesMatAddMv( -ONE, *prevMX, product, ONE, *tempMXj );
            }
          }
          // Compute new Op-norm
          MVT::MvDot( *tempXj, *tempMXj, newDot );
          //
          if ( SCT::magnitude(newDot[0]) >= SCT::magnitude(oldDot[0]*sing_tol_) ) {
            // Copy vector into current column of _basisvecs
            MVT::Assign( *tempXj, *Xj );
            if (this->_hasOp) {
              MVT::Assign( *tempMXj, *MXj );
            }
          }
          else {
            return numX;
          }
        }
      }
      else {
        //
        // We only need to detect dependencies.
        //
        if ( SCT::magnitude(newDot[0]) < SCT::magnitude(oldDot[0]*blk_tol_) ) {
          return numX;
        }
      }

      // If we haven't left this method yet, then we can normalize the new vector Xj.
      // Normalize Xj.
      // Xj <- Xj / std::sqrt(newDot)
      ScalarType diag = SCT::squareroot(SCT::magnitude(newDot[0]));
      {
        MVT::MvScale( *Xj, ONE/diag );
        if (this->_hasOp) {
          // Update MXj.
          MVT::MvScale( *MXj, ONE/diag );
        }
      }

      // If we've added a random vector, enter a zero in the j'th diagonal element.
      if (addVec) {
        (*B)(j,j) = ZERO;
      }
      else {
        (*B)(j,j) = diag;
      }

      // Save the coefficients, if we are working on the original vector and not a randomly generated one
      if (!addVec) {
        for (int i=0; i<numX; i++) {
          (*B)(i,j) = product(i,0);
        }
      }

    } // for (j = 0; j < xc; ++j)

    return xc;
  }

  //////////////////////////////////////////////////////////////////////////////////////////////////
  // Routine to compute the block orthogonalization
  template<class ScalarType, class MV, class OP>
  bool
  ICGSOrthoManager<ScalarType, MV, OP>::blkOrtho1 ( MV &X, Teuchos::RCP<MV> MX,
                                                    Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                                                    Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const
  {
    int nq = Q.size();
    int xc = MVT::GetNumberVecs( X );
    const ScalarType ONE  = SCT::one();

    std::vector<int> qcs( nq );
    for (int i=0; i<nq; i++) {
      qcs[i] = MVT::GetNumberVecs( *Q[i] );
    }

    // Perform the Gram-Schmidt transformation for a block of vectors

    Teuchos::Array<Teuchos::RCP<MV> > MQ(nq);
    // Define the product Q^T * (Op*X)
    for (int i=0; i<nq; i++) {
      // Multiply Q' with MX
      {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
        Teuchos::TimeMonitor innerProdTimer( *timerInnerProd_ );
#endif
        MatOrthoManager<ScalarType,MV,OP>::innerProd(*Q[i],X,MX,*C[i]);
      }
      // Multiply by Q and subtract the result in X
      {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
        Teuchos::TimeMonitor updateTimer( *timerUpdate_ );
#endif
        MVT::MvTimesMatAddMv( -ONE, *Q[i], *C[i], ONE, X );
      }

      // Update MX, with the least number of applications of Op as possible
      if (this->_hasOp) {
        if (xc <= qcs[i]) {
          OPT::Apply( *(this->_Op), X, *MX);
        }
        else {
          // this will possibly be used again below; don't delete it
          MQ[i] = MVT::Clone( *Q[i], qcs[i] );
          OPT::Apply( *(this->_Op), *Q[i], *MQ[i] );
          MVT::MvTimesMatAddMv( -ONE, *MQ[i], *C[i], ONE, *MX );
        }
      }
    }

    // Do as many steps of classical Gram-Schmidt as required by max_ortho_steps_
    for (int j = 1; j < max_ortho_steps_; ++j) {

      for (int i=0; i<nq; i++) {
        Teuchos::SerialDenseMatrix<int,ScalarType> C2(C[i]->numRows(),C[i]->numCols());

        // Apply another step of classical Gram-Schmidt
        {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
          Teuchos::TimeMonitor innerProdTimer( *timerInnerProd_ );
#endif
          MatOrthoManager<ScalarType,MV,OP>::innerProd(*Q[i],X,MX,C2);
        }
        *C[i] += C2;
        {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
          Teuchos::TimeMonitor updateTimer( *timerUpdate_ );
#endif
          MVT::MvTimesMatAddMv( -ONE, *Q[i], C2, ONE, X );
        }

        // Update MX, with the least number of applications of Op as possible
        if (this->_hasOp) {
          if (MQ[i].get()) {
            // MQ was allocated and computed above; use it
            MVT::MvTimesMatAddMv( -ONE, *MQ[i], C2, ONE, *MX );
          }
          else if (xc <= qcs[i]) {
            // MQ was not allocated and computed above; it was cheaper to use X before and it still is
            OPT::Apply( *(this->_Op), X, *MX);
          }
        }
      } // for (int i=0; i<nq; i++)
    } // for (int j = 0; j < max_ortho_steps; ++j)

    return false;
  }

  //////////////////////////////////////////////////////////////////////////////////////////////////
  // Routine to compute the block orthogonalization
  template<class ScalarType, class MV, class OP>
  bool
  ICGSOrthoManager<ScalarType, MV, OP>::blkOrtho ( MV &X, Teuchos::RCP<MV> MX,
                                                   Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                                                   Teuchos::ArrayView<Teuchos::RCP<const MV> > Q) const
  {
    int nq = Q.size();
    int xc = MVT::GetNumberVecs( X );
    bool dep_flg = false;
    const ScalarType ONE  = SCT::one();

    std::vector<int> qcs( nq );
    for (int i=0; i<nq; i++) {
      qcs[i] = MVT::GetNumberVecs( *Q[i] );
    }

    // Perform the Gram-Schmidt transformation for a block of vectors

    // Compute the initial Op-norms
    std::vector<ScalarType> oldDot( xc );
    MVT::MvDot( X, *MX, oldDot );

    Teuchos::Array<Teuchos::RCP<MV> > MQ(nq);
    // Define the product Q^T * (Op*X)
    for (int i=0; i<nq; i++) {
      // Multiply Q' with MX
      {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
        Teuchos::TimeMonitor innerProdTimer( *timerInnerProd_ );
#endif
        MatOrthoManager<ScalarType,MV,OP>::innerProd(*Q[i],X,MX,*C[i]);
      }
      // Multiply by Q and subtract the result in X
      {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
        Teuchos::TimeMonitor updateTimer( *timerUpdate_ );
#endif
        MVT::MvTimesMatAddMv( -ONE, *Q[i], *C[i], ONE, X );
      }
      // Update MX, with the least number of applications of Op as possible
      if (this->_hasOp) {
        if (xc <= qcs[i]) {
          OPT::Apply( *(this->_Op), X, *MX);
        }
        else {
          // this will possibly be used again below; don't delete it
          MQ[i] = MVT::Clone( *Q[i], qcs[i] );
          OPT::Apply( *(this->_Op), *Q[i], *MQ[i] );
          MVT::MvTimesMatAddMv( -ONE, *MQ[i], *C[i], ONE, *MX );
        }
      }
    }

    // Do as many steps of classical Gram-Schmidt as required by max_ortho_steps_
    for (int j = 1; j < max_ortho_steps_; ++j) {

      for (int i=0; i<nq; i++) {
        Teuchos::SerialDenseMatrix<int,ScalarType> C2(C[i]->numRows(),C[i]->numCols());

        // Apply another step of classical Gram-Schmidt
        {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
          Teuchos::TimeMonitor innerProdTimer( *timerInnerProd_ );
#endif
          MatOrthoManager<ScalarType,MV,OP>::innerProd(*Q[i],X,MX,C2);
        }
        *C[i] += C2;
        {
#ifdef BELOS_TEUCHOS_TIME_MONITOR
          Teuchos::TimeMonitor updateTimer( *timerUpdate_ );
#endif
          MVT::MvTimesMatAddMv( -ONE, *Q[i], C2, ONE, X );
        }

        // Update MX, with the least number of applications of Op as possible
        if (this->_hasOp) {
          if (MQ[i].get()) {
            // MQ was allocated and computed above; use it
            MVT::MvTimesMatAddMv( -ONE, *MQ[i], C2, ONE, *MX );
          }
          else if (xc <= qcs[i]) {
            // MQ was not allocated and computed above; it was cheaper to use X before and it still is
            OPT::Apply( *(this->_Op), X, *MX);
          }
        }
      } // for (int i=0; i<nq; i++)
    } // for (int j = 0; j < max_ortho_steps; ++j)

    // Compute new Op-norms
    std::vector<ScalarType> newDot(xc);
    MVT::MvDot( X, *MX, newDot );

    // Check to make sure the new block of vectors are not dependent on previous vectors
    for (int i=0; i<xc; i++){
      if (SCT::magnitude(newDot[i]) < SCT::magnitude(oldDot[i] * blk_tol_)) {
        dep_flg = true;
        break;
      }
    } // end for (i=0;...)

    return dep_flg;
  }

  template<class ScalarType, class MV, class OP>
  int
  ICGSOrthoManager<ScalarType, MV, OP>::blkOrthoSing ( MV &X, Teuchos::RCP<MV> MX,
                                                       Teuchos::Array<Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > > C,
                                                       Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > B,
                                                       Teuchos::ArrayView<Teuchos::RCP<const MV> > QQ) const
  {
    Teuchos::Array<Teuchos::RCP<const MV> > Q (QQ);

    const ScalarType ONE  = SCT::one();
    const ScalarType ZERO  = SCT::zero();

    int nq = Q.size();
    int xc = MVT::GetNumberVecs( X );
    std::vector<int> indX( 1 );
    std::vector<ScalarType> oldDot( 1 ), newDot( 1 );

    std::vector<int> qcs( nq );
    for (int i=0; i<nq; i++) {
      qcs[i] = MVT::GetNumberVecs( *Q[i] );
    }

    // Create pointers for the previous vectors of X that have already been orthonormalized.
    Teuchos::RCP<const MV> lastQ;
    Teuchos::RCP<MV> Xj, MXj;
    Teuchos::RCP<Teuchos::SerialDenseMatrix<int,ScalarType> > lastC;

    // Perform the Gram-Schmidt transformation for each vector in the block of vectors.
    for (int j=0; j<xc; j++) {

      bool dep_flg = false;

      // Get a view of the previously orthogonalized vectors and B, add it to the arrays.
      if (j > 0) {
        std::vector<int> index( j );
        for (int ind=0; ind<j; ind++) {
          index[ind] = ind;
        }
        lastQ = MVT::CloneView( X, index );

        // Add these views to the Q and C arrays.
        Q.push_back( lastQ );
        C.push_back( B );
        qcs.push_back( MVT::GetNumberVecs( *lastQ ) );
      }

      // Get a view of the current vector in X to orthogonalize.
      indX[0] = j;
      Xj = MVT::CloneViewNonConst( X, indX );
      if (this->_hasOp) {
        MXj = MVT::CloneViewNonConst( *MX, indX );
      }
      else {
        MXj = Xj;
      }

      // Compute the initial Op-norms
      MVT::MvDot( *Xj, *MXj, oldDot );

      Teuchos::Array<Teuchos::RCP<MV> > MQ(Q.size());
      // Define the product Q^T * (Op*X)
      for (int i=0; i<Q.size(); i++) {

        // Get a view of the current serial dense matrix
        Teuchos::SerialDenseMatrix<int,ScalarType> tempC( Teuchos::View, *C[i], qcs[i], 1, 0, j );

        // Multiply Q' with MX
        MatOrthoManager<ScalarType,MV,OP>::innerProd(*Q[i],*Xj,MXj,tempC);
        // Multiply by Q and subtract the result in Xj
        MVT::MvTimesMatAddMv( -ONE, *Q[i], tempC, ONE, *Xj );

        // Update MXj, with the least number of applications of Op as possible
        if (this->_hasOp) {
          if (xc <= qcs[i]) {
            OPT::Apply( *(this->_Op), *Xj, *MXj);
          }
          else {
            // this will possibly be used again below; don't delete it
            MQ[i] = MVT::Clone( *Q[i], qcs[i] );
            OPT::Apply( *(this->_Op), *Q[i], *MQ[i] );
            MVT::MvTimesMatAddMv( -ONE, *MQ[i], tempC, ONE, *MXj );
          }
        }
      }

      // Do any additional steps of classical Gram-Schmidt orthogonalization
      for (int num_ortho_steps=1; num_ortho_steps < max_ortho_steps_; ++num_ortho_steps) {

        for (int i=0; i<Q.size(); i++) {
          Teuchos::SerialDenseMatrix<int,ScalarType> tempC( Teuchos::View, *C[i], qcs[i], 1, 0, j );
          Teuchos::SerialDenseMatrix<int,ScalarType> C2( qcs[i], 1 );

          // Apply another step of classical Gram-Schmidt
          MatOrthoManager<ScalarType,MV,OP>::innerProd(*Q[i],*Xj,MXj,C2);
          tempC += C2;
          MVT::MvTimesMatAddMv( -ONE, *Q[i], C2, ONE, *Xj );

          // Update MXj, with the least number of applications of Op as possible
          if (this->_hasOp) {
            if (MQ[i].get()) {
              // MQ was allocated and computed above; use it
              MVT::MvTimesMatAddMv( -ONE, *MQ[i], C2, ONE, *MXj );
            }
            else if (xc <= qcs[i]) {
              // MQ was not allocated and computed above; it was cheaper to use X before and it still is
              OPT::Apply( *(this->_Op), *Xj, *MXj);
            }
          }
        } // for (int i=0; i<Q.size(); i++)

      } // for (int num_ortho_steps=1; num_ortho_steps < max_ortho_steps_; ++num_ortho_steps)

      // Check for linear dependence.
      if (SCT::magnitude(newDot[0]) < SCT::magnitude(oldDot[0]*sing_tol_)) {
        dep_flg = true;
      }

      // Normalize the new vector if it's not dependent
      if (!dep_flg) {
        ScalarType diag = SCT::squareroot(SCT::magnitude(newDot[0]));

        MVT::MvScale( *Xj, ONE/diag );
        if (this->_hasOp) {
          // Update MXj.
          MVT::MvScale( *MXj, ONE/diag );
        }

        // Enter value on diagonal of B.
        (*B)(j,j) = diag;
      }
      else {
        // Create a random vector and orthogonalize it against all previous columns of Q.
        Teuchos::RCP<MV> tempXj = MVT::Clone( X, 1 );
        Teuchos::RCP<MV> tempMXj;
        MVT::MvRandom( *tempXj );
        if (this->_hasOp) {
          tempMXj = MVT::Clone( X, 1 );
          OPT::Apply( *(this->_Op), *tempXj, *tempMXj );
        }
        else {
          tempMXj = tempXj;
        }
        MVT::MvDot( *tempXj, *tempMXj, oldDot );
        //
        for (int num_orth=0; num_orth<max_ortho_steps_; num_orth++) {

          for (int i=0; i<Q.size(); i++) {
            Teuchos::SerialDenseMatrix<int,ScalarType> product( qcs[i], 1 );

            // Apply another step of classical Gram-Schmidt
            MatOrthoManager<ScalarType,MV,OP>::innerProd(*Q[i],*tempXj,tempMXj,product);
            MVT::MvTimesMatAddMv( -ONE, *Q[i], product, ONE, *tempXj );

            // Update MXj, with the least number of applications of Op as possible
            if (this->_hasOp) {
              if (MQ[i].get()) {
                // MQ was allocated and computed above; use it
                MVT::MvTimesMatAddMv( -ONE, *MQ[i], product, ONE, *tempMXj );
              }
              else if (xc <= qcs[i]) {
                // MQ was not allocated and computed above; it was cheaper to use X before and it still is
                OPT::Apply( *(this->_Op), *tempXj, *tempMXj);
              }
            }
          } // for (int i=0; i<nq; i++)

        }

        // Compute the Op-norms after the correction step.
        MVT::MvDot( *tempXj, *tempMXj, newDot );

        // Copy vector into current column of Xj
        if ( SCT::magnitude(newDot[0]) >= SCT::magnitude(oldDot[0]*sing_tol_) ) {
          ScalarType diag = SCT::squareroot(SCT::magnitude(newDot[0]));

          // Enter value on diagonal of B.
          (*B)(j,j) = ZERO;

          // Copy vector into current column of _basisvecs
          MVT::MvAddMv( ONE/diag, *tempXj, ZERO, *tempXj, *Xj );
          if (this->_hasOp) {
            MVT::MvAddMv( ONE/diag, *tempMXj, ZERO, *tempMXj, *MXj );
          }
        }
        else {
          return j;
        }
      } // if (!dep_flg)

      // Remove the vectors from array
      if (j > 0) {
        Q.resize( nq );
        C.resize( nq );
        qcs.resize( nq );
      }

    } // for (int j=0; j<xc; j++)

    return xc;
  }

} // namespace Belos

#endif // BELOS_ICGS_ORTHOMANAGER_HPP