ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/OpenMD/trunk/src/parallel/ForceMatrixDecomposition.cpp
(Generate patch)

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1593 by gezelter, Fri Jul 15 21:35:14 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1929 by gezelter, Mon Aug 19 13:12:00 2013 UTC

# Line 35 | Line 35
35   *                                                                      
36   * [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).            
37   * [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).          
38 < * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
39 < * [4]  Vardeman & Gezelter, in progress (2009).                        
38 > * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008).          
39 > * [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40 > * [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41   */
42   #include "parallel/ForceMatrixDecomposition.hpp"
43   #include "math/SquareMatrix3.hpp"
# Line 53 | Line 54 | namespace OpenMD {
54      // surrounding cells (not just the 14 upper triangular blocks that
55      // are used when the processor can see all pairs)
56   #ifdef IS_MPI
57 <    cellOffsets_.push_back( Vector3i(-1, 0, 0) );
57 <    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
58 <    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
59 <    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
60 <    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
61 <    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
57 >    cellOffsets_.clear();
58      cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59      cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60 <    cellOffsets_.push_back( Vector3i( 1,-1,-1) );
60 >    cellOffsets_.push_back( Vector3i( 1,-1,-1) );                          
61 >    cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62 >    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63      cellOffsets_.push_back( Vector3i( 1, 0,-1) );
66 –    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 –    cellOffsets_.push_back( Vector3i( 0, 1,-1) );
64      cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );      
66 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );      
71 +    cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72 +    cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73 +    cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74 +    cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75 +    cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76 +    cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77 +    cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78 +    cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80 +    cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81 +    cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82 +    cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83 +    cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84 +    cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85   #endif    
86    }
87  
# Line 79 | Line 95 | namespace OpenMD {
95      storageLayout_ = sman_->getStorageLayout();
96      ff_ = info_->getForceField();
97      nLocal_ = snap_->getNumberOfAtoms();
98 <    
98 >  
99      nGroups_ = info_->getNLocalCutoffGroups();
100      // gather the information for atomtype IDs (atids):
101      idents = info_->getIdentArray();
102 +    regions = info_->getRegions();
103      AtomLocalToGlobal = info_->getGlobalAtomIndices();
104      cgLocalToGlobal = info_->getGlobalGroupIndices();
105      vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
# Line 93 | Line 110 | namespace OpenMD {
110      PairList* oneTwo = info_->getOneTwoInteractions();
111      PairList* oneThree = info_->getOneThreeInteractions();
112      PairList* oneFour = info_->getOneFourInteractions();
113 <
113 >    
114 >    if (needVelocities_)
115 >      snap_->cgData.setStorageLayout(DataStorage::dslPosition |
116 >                                     DataStorage::dslVelocity);
117 >    else
118 >      snap_->cgData.setStorageLayout(DataStorage::dslPosition);
119 >    
120   #ifdef IS_MPI
121  
122      MPI::Intracomm row = rowComm.getComm();
# Line 129 | Line 152 | namespace OpenMD {
152      cgRowData.resize(nGroupsInRow_);
153      cgRowData.setStorageLayout(DataStorage::dslPosition);
154      cgColData.resize(nGroupsInCol_);
155 <    cgColData.setStorageLayout(DataStorage::dslPosition);
156 <        
155 >    if (needVelocities_)
156 >      // we only need column velocities if we need them.
157 >      cgColData.setStorageLayout(DataStorage::dslPosition |
158 >                                 DataStorage::dslVelocity);
159 >    else    
160 >      cgColData.setStorageLayout(DataStorage::dslPosition);
161 >      
162      identsRow.resize(nAtomsInRow_);
163      identsCol.resize(nAtomsInCol_);
164      
165      AtomPlanIntRow->gather(idents, identsRow);
166      AtomPlanIntColumn->gather(idents, identsCol);
167 +
168 +    regionsRow.resize(nAtomsInRow_);
169 +    regionsCol.resize(nAtomsInCol_);
170      
171 +    AtomPlanIntRow->gather(regions, regionsRow);
172 +    AtomPlanIntColumn->gather(regions, regionsCol);
173 +    
174      // allocate memory for the parallel objects
175      atypesRow.resize(nAtomsInRow_);
176      atypesCol.resize(nAtomsInCol_);
# Line 149 | Line 183 | namespace OpenMD {
183      pot_row.resize(nAtomsInRow_);
184      pot_col.resize(nAtomsInCol_);
185  
186 +    expot_row.resize(nAtomsInRow_);
187 +    expot_col.resize(nAtomsInCol_);
188 +
189      AtomRowToGlobal.resize(nAtomsInRow_);
190      AtomColToGlobal.resize(nAtomsInCol_);
191      AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
192      AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
193  
157 –    cerr << "Atoms in Local:\n";
158 –    for (int i = 0; i < AtomLocalToGlobal.size(); i++) {
159 –      cerr << "i =\t" << i << "\t localAt =\t" << AtomLocalToGlobal[i] << "\n";
160 –    }
161 –    cerr << "Atoms in Row:\n";
162 –    for (int i = 0; i < AtomRowToGlobal.size(); i++) {
163 –      cerr << "i =\t" << i << "\t rowAt =\t" << AtomRowToGlobal[i] << "\n";
164 –    }
165 –    cerr << "Atoms in Col:\n";
166 –    for (int i = 0; i < AtomColToGlobal.size(); i++) {
167 –      cerr << "i =\t" << i << "\t colAt =\t" << AtomColToGlobal[i] << "\n";
168 –    }
169 –
194      cgRowToGlobal.resize(nGroupsInRow_);
195      cgColToGlobal.resize(nGroupsInCol_);
196      cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
197      cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
198  
175 –    cerr << "Gruops in Local:\n";
176 –    for (int i = 0; i < cgLocalToGlobal.size(); i++) {
177 –      cerr << "i =\t" << i << "\t localCG =\t" << cgLocalToGlobal[i] << "\n";
178 –    }
179 –    cerr << "Groups in Row:\n";
180 –    for (int i = 0; i < cgRowToGlobal.size(); i++) {
181 –      cerr << "i =\t" << i << "\t rowCG =\t" << cgRowToGlobal[i] << "\n";
182 –    }
183 –    cerr << "Groups in Col:\n";
184 –    for (int i = 0; i < cgColToGlobal.size(); i++) {
185 –      cerr << "i =\t" << i << "\t colCG =\t" << cgColToGlobal[i] << "\n";
186 –    }
187 –
188 –
199      massFactorsRow.resize(nAtomsInRow_);
200      massFactorsCol.resize(nAtomsInCol_);
201      AtomPlanRealRow->gather(massFactors, massFactorsRow);
# Line 245 | Line 255 | namespace OpenMD {
255        }      
256      }
257  
258 < #endif
249 <
250 <    // allocate memory for the parallel objects
251 <    atypesLocal.resize(nLocal_);
252 <
253 <    for (int i = 0; i < nLocal_; i++)
254 <      atypesLocal[i] = ff_->getAtomType(idents[i]);
255 <
256 <    groupList_.clear();
257 <    groupList_.resize(nGroups_);
258 <    for (int i = 0; i < nGroups_; i++) {
259 <      int gid = cgLocalToGlobal[i];
260 <      for (int j = 0; j < nLocal_; j++) {
261 <        int aid = AtomLocalToGlobal[j];
262 <        if (globalGroupMembership[aid] == gid) {
263 <          groupList_[i].push_back(j);
264 <        }
265 <      }      
266 <    }
267 <
258 > #else
259      excludesForAtom.clear();
260      excludesForAtom.resize(nLocal_);
261      toposForAtom.clear();
# Line 297 | Line 288 | namespace OpenMD {
288          }
289        }      
290      }
291 <    
291 > #endif
292 >
293 >    // allocate memory for the parallel objects
294 >    atypesLocal.resize(nLocal_);
295 >
296 >    for (int i = 0; i < nLocal_; i++)
297 >      atypesLocal[i] = ff_->getAtomType(idents[i]);
298 >
299 >    groupList_.clear();
300 >    groupList_.resize(nGroups_);
301 >    for (int i = 0; i < nGroups_; i++) {
302 >      int gid = cgLocalToGlobal[i];
303 >      for (int j = 0; j < nLocal_; j++) {
304 >        int aid = AtomLocalToGlobal[j];
305 >        if (globalGroupMembership[aid] == gid) {
306 >          groupList_[i].push_back(j);
307 >        }
308 >      }      
309 >    }
310 >
311 >
312      createGtypeCutoffMap();
313  
314    }
315    
316    void ForceMatrixDecomposition::createGtypeCutoffMap() {
317      
318 +    GrCut.clear();
319 +    GrCutSq.clear();
320 +    GrlistSq.clear();
321 +
322      RealType tol = 1e-6;
323      largestRcut_ = 0.0;
309 –    RealType rc;
324      int atid;
325      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
326      
# Line 391 | Line 405 | namespace OpenMD {
405        }
406        
407        bool gTypeFound = false;
408 <      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
408 >      for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
409          if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
410            groupToGtype[cg1] = gt;
411            gTypeFound = true;
# Line 416 | Line 430 | namespace OpenMD {
430      
431      RealType tradRcut = groupMax;
432  
433 <    for (int i = 0; i < gTypeCutoffs.size();  i++) {
434 <      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
433 >    GrCut.resize( gTypeCutoffs.size() );
434 >    GrCutSq.resize( gTypeCutoffs.size() );
435 >    GrlistSq.resize( gTypeCutoffs.size() );
436 >
437 >
438 >    for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
439 >      GrCut[i].resize( gTypeCutoffs.size() , 0.0);
440 >      GrCutSq[i].resize( gTypeCutoffs.size(), 0.0 );
441 >      GrlistSq[i].resize( gTypeCutoffs.size(), 0.0 );
442 >
443 >      for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {      
444          RealType thisRcut;
445          switch(cutoffPolicy_) {
446          case TRADITIONAL:
# Line 439 | Line 462 | namespace OpenMD {
462            break;
463          }
464  
465 <        pair<int,int> key = make_pair(i,j);
443 <        gTypeCutoffMap[key].first = thisRcut;
465 >        GrCut[i][j] = thisRcut;
466          if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
467 <        gTypeCutoffMap[key].second = thisRcut*thisRcut;
468 <        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
467 >        GrCutSq[i][j] = thisRcut * thisRcut;
468 >        GrlistSq[i][j] = pow(thisRcut + skinThickness_, 2);
469 >
470 >        // pair<int,int> key = make_pair(i,j);
471 >        // gTypeCutoffMap[key].first = thisRcut;
472 >        // gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
473          // sanity check
474          
475          if (userChoseCutoff_) {
476 <          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
476 >          if (abs(GrCut[i][j] - userCutoff_) > 0.0001) {
477              sprintf(painCave.errMsg,
478                      "ForceMatrixDecomposition::createGtypeCutoffMap "
479                      "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
# Line 460 | Line 486 | namespace OpenMD {
486      }
487    }
488  
489 <
464 <  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
489 >  void ForceMatrixDecomposition::getGroupCutoffs(int &cg1, int &cg2, RealType &rcut, RealType &rcutsq, RealType &rlistsq) {
490      int i, j;  
491   #ifdef IS_MPI
492      i = groupRowToGtype[cg1];
# Line 470 | Line 495 | namespace OpenMD {
495      i = groupToGtype[cg1];
496      j = groupToGtype[cg2];
497   #endif    
498 <    return gTypeCutoffMap[make_pair(i,j)];
498 >    rcut = GrCut[i][j];
499 >    rcutsq = GrCutSq[i][j];
500 >    rlistsq = GrlistSq[i][j];
501 >    return;
502 >    //return gTypeCutoffMap[make_pair(i,j)];
503    }
504  
505    int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
506 <    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
506 >    for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
507        if (toposForAtom[atom1][j] == atom2)
508          return topoDist[atom1][j];
509 <    }
509 >    }                                          
510      return 0;
511    }
512  
513    void ForceMatrixDecomposition::zeroWorkArrays() {
514      pairwisePot = 0.0;
515      embeddingPot = 0.0;
516 +    excludedPot = 0.0;
517 +    excludedSelfPot = 0.0;
518  
519   #ifdef IS_MPI
520      if (storageLayout_ & DataStorage::dslForce) {
# Line 502 | Line 533 | namespace OpenMD {
533      fill(pot_col.begin(), pot_col.end(),
534           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
535  
536 +    fill(expot_row.begin(), expot_row.end(),
537 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
538 +
539 +    fill(expot_col.begin(), expot_col.end(),
540 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
541 +
542      if (storageLayout_ & DataStorage::dslParticlePot) {    
543        fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
544             0.0);
# Line 535 | Line 572 | namespace OpenMD {
572             atomColData.skippedCharge.end(), 0.0);
573      }
574  
575 +    if (storageLayout_ & DataStorage::dslFlucQForce) {      
576 +      fill(atomRowData.flucQFrc.begin(),
577 +           atomRowData.flucQFrc.end(), 0.0);
578 +      fill(atomColData.flucQFrc.begin(),
579 +           atomColData.flucQFrc.end(), 0.0);
580 +    }
581 +
582 +    if (storageLayout_ & DataStorage::dslElectricField) {    
583 +      fill(atomRowData.electricField.begin(),
584 +           atomRowData.electricField.end(), V3Zero);
585 +      fill(atomColData.electricField.begin(),
586 +           atomColData.electricField.end(), V3Zero);
587 +    }
588 +
589   #endif
590      // even in parallel, we need to zero out the local arrays:
591  
# Line 547 | Line 598 | namespace OpenMD {
598        fill(snap_->atomData.density.begin(),
599             snap_->atomData.density.end(), 0.0);
600      }
601 +
602      if (storageLayout_ & DataStorage::dslFunctional) {
603        fill(snap_->atomData.functional.begin(),
604             snap_->atomData.functional.end(), 0.0);
605      }
606 +
607      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
608        fill(snap_->atomData.functionalDerivative.begin(),
609             snap_->atomData.functionalDerivative.end(), 0.0);
610      }
611 +
612      if (storageLayout_ & DataStorage::dslSkippedCharge) {      
613        fill(snap_->atomData.skippedCharge.begin(),
614             snap_->atomData.skippedCharge.end(), 0.0);
615      }
616 <    
616 >
617 >    if (storageLayout_ & DataStorage::dslElectricField) {      
618 >      fill(snap_->atomData.electricField.begin(),
619 >           snap_->atomData.electricField.end(), V3Zero);
620 >    }
621    }
622  
623  
# Line 576 | Line 634 | namespace OpenMD {
634      
635      // gather up the cutoff group positions
636  
579 –    cerr  << "before gather\n";
580 –    for (int i = 0; i < snap_->cgData.position.size(); i++) {
581 –      cerr << "cgpos = " << snap_->cgData.position[i] << "\n";
582 –    }
583 –
637      cgPlanVectorRow->gather(snap_->cgData.position,
638                              cgRowData.position);
639  
587 –    cerr  << "after gather\n";
588 –    for (int i = 0; i < cgRowData.position.size(); i++) {
589 –      cerr << "cgRpos = " << cgRowData.position[i] << "\n";
590 –    }
591 –
640      cgPlanVectorColumn->gather(snap_->cgData.position,
641                                 cgColData.position);
642 <    for (int i = 0; i < cgColData.position.size(); i++) {
643 <      cerr << "cgCpos = " << cgColData.position[i] << "\n";
642 >
643 >
644 >
645 >    if (needVelocities_) {
646 >      // gather up the atomic velocities
647 >      AtomPlanVectorColumn->gather(snap_->atomData.velocity,
648 >                                   atomColData.velocity);
649 >      
650 >      cgPlanVectorColumn->gather(snap_->cgData.velocity,
651 >                                 cgColData.velocity);
652      }
653  
654      
# Line 603 | Line 659 | namespace OpenMD {
659        AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
660                                     atomColData.aMat);
661      }
662 <    
663 <    // if needed, gather the atomic eletrostatic frames
664 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
665 <      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
666 <                                atomRowData.electroFrame);
667 <      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
668 <                                   atomColData.electroFrame);
662 >
663 >    // if needed, gather the atomic eletrostatic information
664 >    if (storageLayout_ & DataStorage::dslDipole) {
665 >      AtomPlanVectorRow->gather(snap_->atomData.dipole,
666 >                                atomRowData.dipole);
667 >      AtomPlanVectorColumn->gather(snap_->atomData.dipole,
668 >                                   atomColData.dipole);
669      }
670  
671 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
672 +      AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
673 +                                atomRowData.quadrupole);
674 +      AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
675 +                                   atomColData.quadrupole);
676 +    }
677 +        
678 +    // if needed, gather the atomic fluctuating charge values
679 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {
680 +      AtomPlanRealRow->gather(snap_->atomData.flucQPos,
681 +                              atomRowData.flucQPos);
682 +      AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
683 +                                 atomColData.flucQPos);
684 +    }
685 +
686   #endif      
687    }
688    
# Line 634 | Line 705 | namespace OpenMD {
705        for (int i = 0; i < n; i++)
706          snap_->atomData.density[i] += rho_tmp[i];
707      }
708 +
709 +    // this isn't necessary if we don't have polarizable atoms, but
710 +    // we'll leave it here for now.
711 +    if (storageLayout_ & DataStorage::dslElectricField) {
712 +      
713 +      AtomPlanVectorRow->scatter(atomRowData.electricField,
714 +                                 snap_->atomData.electricField);
715 +      
716 +      int n = snap_->atomData.electricField.size();
717 +      vector<Vector3d> field_tmp(n, V3Zero);
718 +      AtomPlanVectorColumn->scatter(atomColData.electricField,
719 +                                    field_tmp);
720 +      for (int i = 0; i < n; i++)
721 +        snap_->atomData.electricField[i] += field_tmp[i];
722 +    }
723   #endif
724    }
725  
# Line 708 | Line 794 | namespace OpenMD {
794        }
795        
796        AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
797 <      for (int i = 0; i < ns; i++)
797 >      for (int i = 0; i < ns; i++)
798          snap_->atomData.skippedCharge[i] += skch_tmp[i];
799 +            
800      }
801      
802 +    if (storageLayout_ & DataStorage::dslFlucQForce) {
803 +
804 +      int nq = snap_->atomData.flucQFrc.size();
805 +      vector<RealType> fqfrc_tmp(nq, 0.0);
806 +
807 +      AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
808 +      for (int i = 0; i < nq; i++) {
809 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
810 +        fqfrc_tmp[i] = 0.0;
811 +      }
812 +      
813 +      AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
814 +      for (int i = 0; i < nq; i++)
815 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
816 +            
817 +    }
818 +
819 +    if (storageLayout_ & DataStorage::dslElectricField) {
820 +
821 +      int nef = snap_->atomData.electricField.size();
822 +      vector<Vector3d> efield_tmp(nef, V3Zero);
823 +
824 +      AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
825 +      for (int i = 0; i < nef; i++) {
826 +        snap_->atomData.electricField[i] += efield_tmp[i];
827 +        efield_tmp[i] = 0.0;
828 +      }
829 +      
830 +      AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
831 +      for (int i = 0; i < nef; i++)
832 +        snap_->atomData.electricField[i] += efield_tmp[i];
833 +    }
834 +
835 +
836      nLocal_ = snap_->getNumberOfAtoms();
837  
838      vector<potVec> pot_temp(nLocal_,
839                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
840 +    vector<potVec> expot_temp(nLocal_,
841 +                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
842  
843      // scatter/gather pot_row into the members of my column
844            
845      AtomPlanPotRow->scatter(pot_row, pot_temp);
846 +    AtomPlanPotRow->scatter(expot_row, expot_temp);
847  
848 <    for (int ii = 0;  ii < pot_temp.size(); ii++ )
848 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
849        pairwisePot += pot_temp[ii];
850 <    
850 >
851 >    for (int ii = 0;  ii < expot_temp.size(); ii++ )
852 >      excludedPot += expot_temp[ii];
853 >        
854 >    if (storageLayout_ & DataStorage::dslParticlePot) {
855 >      // This is the pairwise contribution to the particle pot.  The
856 >      // embedding contribution is added in each of the low level
857 >      // non-bonded routines.  In single processor, this is done in
858 >      // unpackInteractionData, not in collectData.
859 >      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
860 >        for (int i = 0; i < nLocal_; i++) {
861 >          // factor of two is because the total potential terms are divided
862 >          // by 2 in parallel due to row/ column scatter      
863 >          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
864 >        }
865 >      }
866 >    }
867 >
868      fill(pot_temp.begin(), pot_temp.end(),
869           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
870 +    fill(expot_temp.begin(), expot_temp.end(),
871 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
872        
873      AtomPlanPotColumn->scatter(pot_col, pot_temp);    
874 +    AtomPlanPotColumn->scatter(expot_col, expot_temp);    
875      
876      for (int ii = 0;  ii < pot_temp.size(); ii++ )
877        pairwisePot += pot_temp[ii];    
878 +
879 +    for (int ii = 0;  ii < expot_temp.size(); ii++ )
880 +      excludedPot += expot_temp[ii];    
881 +
882 +    if (storageLayout_ & DataStorage::dslParticlePot) {
883 +      // This is the pairwise contribution to the particle pot.  The
884 +      // embedding contribution is added in each of the low level
885 +      // non-bonded routines.  In single processor, this is done in
886 +      // unpackInteractionData, not in collectData.
887 +      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
888 +        for (int i = 0; i < nLocal_; i++) {
889 +          // factor of two is because the total potential terms are divided
890 +          // by 2 in parallel due to row/ column scatter      
891 +          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
892 +        }
893 +      }
894 +    }
895 +    
896 +    if (storageLayout_ & DataStorage::dslParticlePot) {
897 +      int npp = snap_->atomData.particlePot.size();
898 +      vector<RealType> ppot_temp(npp, 0.0);
899 +
900 +      // This is the direct or embedding contribution to the particle
901 +      // pot.
902 +      
903 +      AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
904 +      for (int i = 0; i < npp; i++) {
905 +        snap_->atomData.particlePot[i] += ppot_temp[i];
906 +      }
907 +
908 +      fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
909 +      
910 +      AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
911 +      for (int i = 0; i < npp; i++) {
912 +        snap_->atomData.particlePot[i] += ppot_temp[i];
913 +      }
914 +    }
915 +
916 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
917 +      RealType ploc1 = pairwisePot[ii];
918 +      RealType ploc2 = 0.0;
919 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
920 +      pairwisePot[ii] = ploc2;
921 +    }
922 +
923 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
924 +      RealType ploc1 = excludedPot[ii];
925 +      RealType ploc2 = 0.0;
926 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
927 +      excludedPot[ii] = ploc2;
928 +    }
929 +
930 +    // Here be dragons.
931 +    MPI::Intracomm col = colComm.getComm();
932 +
933 +    col.Allreduce(MPI::IN_PLACE,
934 +                  &snap_->frameData.conductiveHeatFlux[0], 3,
935 +                  MPI::REALTYPE, MPI::SUM);
936 +
937 +
938   #endif
939  
736 –    cerr << "pairwisePot = " <<  pairwisePot << "\n";
940    }
941  
942 <  int ForceMatrixDecomposition::getNAtomsInRow() {  
942 >  /**
943 >   * Collects information obtained during the post-pair (and embedding
944 >   * functional) loops onto local data structures.
945 >   */
946 >  void ForceMatrixDecomposition::collectSelfData() {
947 >    snap_ = sman_->getCurrentSnapshot();
948 >    storageLayout_ = sman_->getStorageLayout();
949 >
950   #ifdef IS_MPI
951 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
952 +      RealType ploc1 = embeddingPot[ii];
953 +      RealType ploc2 = 0.0;
954 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
955 +      embeddingPot[ii] = ploc2;
956 +    }    
957 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
958 +      RealType ploc1 = excludedSelfPot[ii];
959 +      RealType ploc2 = 0.0;
960 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
961 +      excludedSelfPot[ii] = ploc2;
962 +    }    
963 + #endif
964 +    
965 +  }
966 +
967 +
968 +
969 +  int& ForceMatrixDecomposition::getNAtomsInRow() {  
970 + #ifdef IS_MPI
971      return nAtomsInRow_;
972   #else
973      return nLocal_;
# Line 747 | Line 977 | namespace OpenMD {
977    /**
978     * returns the list of atoms belonging to this group.  
979     */
980 <  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
980 >  vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
981   #ifdef IS_MPI
982      return groupListRow_[cg1];
983   #else
# Line 755 | Line 985 | namespace OpenMD {
985   #endif
986    }
987  
988 <  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
988 >  vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
989   #ifdef IS_MPI
990      return groupListCol_[cg2];
991   #else
# Line 768 | Line 998 | namespace OpenMD {
998      
999   #ifdef IS_MPI
1000      d = cgColData.position[cg2] - cgRowData.position[cg1];
771 –    cerr << "cg1 = " << cg1 << "\tcg1p = " << cgRowData.position[cg1] << "\n";
772 –    cerr << "cg2 = " << cg2 << "\tcg2p = " << cgColData.position[cg2] << "\n";
1001   #else
1002      d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
775 –    cerr << "cg1 = " << cg1 << "\tcg1p = " << snap_->cgData.position[cg1] << "\n";
776 –    cerr << "cg2 = " << cg2 << "\tcg2p = " << snap_->cgData.position[cg2] << "\n";
1003   #endif
1004      
1005 <    snap_->wrapVector(d);
1005 >    if (usePeriodicBoundaryConditions_) {
1006 >      snap_->wrapVector(d);
1007 >    }
1008      return d;    
1009    }
1010  
1011 +  Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
1012 + #ifdef IS_MPI
1013 +    return cgColData.velocity[cg2];
1014 + #else
1015 +    return snap_->cgData.velocity[cg2];
1016 + #endif
1017 +  }
1018  
1019 +  Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
1020 + #ifdef IS_MPI
1021 +    return atomColData.velocity[atom2];
1022 + #else
1023 +    return snap_->atomData.velocity[atom2];
1024 + #endif
1025 +  }
1026 +
1027 +
1028    Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1029  
1030      Vector3d d;
# Line 790 | Line 1034 | namespace OpenMD {
1034   #else
1035      d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
1036   #endif
1037 <
1038 <    snap_->wrapVector(d);
1037 >    if (usePeriodicBoundaryConditions_) {
1038 >      snap_->wrapVector(d);
1039 >    }
1040      return d;    
1041    }
1042    
# Line 803 | Line 1048 | namespace OpenMD {
1048   #else
1049      d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1050   #endif
1051 <    
1052 <    snap_->wrapVector(d);
1051 >    if (usePeriodicBoundaryConditions_) {
1052 >      snap_->wrapVector(d);
1053 >    }
1054      return d;    
1055    }
1056  
1057 <  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1057 >  RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1058   #ifdef IS_MPI
1059      return massFactorsRow[atom1];
1060   #else
# Line 816 | Line 1062 | namespace OpenMD {
1062   #endif
1063    }
1064  
1065 <  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1065 >  RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1066   #ifdef IS_MPI
1067      return massFactorsCol[atom2];
1068   #else
# Line 833 | Line 1079 | namespace OpenMD {
1079   #else
1080      d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1081   #endif
1082 <
1083 <    snap_->wrapVector(d);
1082 >    if (usePeriodicBoundaryConditions_) {
1083 >      snap_->wrapVector(d);
1084 >    }
1085      return d;    
1086    }
1087  
1088 <  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1088 >  vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1089      return excludesForAtom[atom1];
1090    }
1091  
# Line 846 | Line 1093 | namespace OpenMD {
1093     * We need to exclude some overcounted interactions that result from
1094     * the parallel decomposition.
1095     */
1096 <  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1096 >  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1097      int unique_id_1, unique_id_2;
1098 <    
852 <
853 <    cerr << "sap with atom1, atom2 =\t" << atom1 << "\t" << atom2 << "\n";
1098 >        
1099   #ifdef IS_MPI
1100      // in MPI, we have to look up the unique IDs for each atom
1101      unique_id_1 = AtomRowToGlobal[atom1];
1102      unique_id_2 = AtomColToGlobal[atom2];
1103 <
1104 <    cerr << "sap with uid1, uid2 =\t" << unique_id_1 << "\t" << unique_id_2 << "\n";
1105 <    // this situation should only arise in MPI simulations
1103 >    // group1 = cgRowToGlobal[cg1];
1104 >    // group2 = cgColToGlobal[cg2];
1105 > #else
1106 >    unique_id_1 = AtomLocalToGlobal[atom1];
1107 >    unique_id_2 = AtomLocalToGlobal[atom2];
1108 >    int group1 = cgLocalToGlobal[cg1];
1109 >    int group2 = cgLocalToGlobal[cg2];
1110 > #endif  
1111 >
1112      if (unique_id_1 == unique_id_2) return true;
1113 <    
1113 >
1114 > #ifdef IS_MPI
1115      // this prevents us from doing the pair on multiple processors
1116      if (unique_id_1 < unique_id_2) {
1117        if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1118      } else {
1119 <      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1119 >      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1120      }
1121 + #endif    
1122 +
1123 + #ifndef IS_MPI
1124 +    if (group1 == group2) {
1125 +      if (unique_id_1 < unique_id_2) return true;
1126 +    }
1127   #endif
1128 +    
1129      return false;
1130    }
1131  
# Line 880 | Line 1139 | namespace OpenMD {
1139     * field) must still be handled for these pairs.
1140     */
1141    bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1142 <    int unique_id_2;
1143 < #ifdef IS_MPI
1144 <    // in MPI, we have to look up the unique IDs for the row atom.
886 <    unique_id_2 = AtomColToGlobal[atom2];
887 < #else
888 <    // in the normal loop, the atom numbers are unique
889 <    unique_id_2 = atom2;
890 < #endif
1142 >
1143 >    // excludesForAtom was constructed to use row/column indices in the MPI
1144 >    // version, and to use local IDs in the non-MPI version:
1145      
1146      for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1147           i != excludesForAtom[atom1].end(); ++i) {
1148 <      if ( (*i) == unique_id_2 ) return true;
1148 >      if ( (*i) == atom2 ) return true;
1149      }
1150  
1151      return false;
# Line 922 | Line 1176 | namespace OpenMD {
1176    
1177   #ifdef IS_MPI
1178      idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1179 +    idat.atid1 = identsRow[atom1];
1180 +    idat.atid2 = identsCol[atom2];
1181 +
1182 +    if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0)
1183 +      idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1184 +      
1185      //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1186      //                         ff_->getAtomType(identsCol[atom2]) );
1187      
# Line 930 | Line 1190 | namespace OpenMD {
1190        idat.A2 = &(atomColData.aMat[atom2]);
1191      }
1192      
933 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
934 –      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
935 –      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
936 –    }
937 –
1193      if (storageLayout_ & DataStorage::dslTorque) {
1194        idat.t1 = &(atomRowData.torque[atom1]);
1195        idat.t2 = &(atomColData.torque[atom2]);
1196      }
1197  
1198 +    if (storageLayout_ & DataStorage::dslDipole) {
1199 +      idat.dipole1 = &(atomRowData.dipole[atom1]);
1200 +      idat.dipole2 = &(atomColData.dipole[atom2]);
1201 +    }
1202 +
1203 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1204 +      idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1205 +      idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1206 +    }
1207 +
1208      if (storageLayout_ & DataStorage::dslDensity) {
1209        idat.rho1 = &(atomRowData.density[atom1]);
1210        idat.rho2 = &(atomColData.density[atom2]);
# Line 965 | Line 1230 | namespace OpenMD {
1230        idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1231      }
1232  
1233 < #else
1233 >    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1234 >      idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1235 >      idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1236 >    }
1237  
1238 + #else
1239 +    
1240      idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1241 <    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1242 <    //                         ff_->getAtomType(idents[atom2]) );
1241 >    idat.atid1 = idents[atom1];
1242 >    idat.atid2 = idents[atom2];
1243  
1244 +    if (regions[atom1] >= 0 && regions[atom2] >= 0)
1245 +      idat.sameRegion = (regions[atom1] == regions[atom2]);
1246 +
1247      if (storageLayout_ & DataStorage::dslAmat) {
1248        idat.A1 = &(snap_->atomData.aMat[atom1]);
1249        idat.A2 = &(snap_->atomData.aMat[atom2]);
1250      }
1251  
979 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
980 –      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
981 –      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
982 –    }
983 –
1252      if (storageLayout_ & DataStorage::dslTorque) {
1253        idat.t1 = &(snap_->atomData.torque[atom1]);
1254        idat.t2 = &(snap_->atomData.torque[atom2]);
1255      }
1256  
1257 +    if (storageLayout_ & DataStorage::dslDipole) {
1258 +      idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1259 +      idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1260 +    }
1261 +
1262 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1263 +      idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1264 +      idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1265 +    }
1266 +
1267      if (storageLayout_ & DataStorage::dslDensity) {    
1268        idat.rho1 = &(snap_->atomData.density[atom1]);
1269        idat.rho2 = &(snap_->atomData.density[atom2]);
# Line 1010 | Line 1288 | namespace OpenMD {
1288        idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1289        idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1290      }
1291 +
1292 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1293 +      idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1294 +      idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1295 +    }
1296 +
1297   #endif
1298    }
1299  
1300    
1301    void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1302   #ifdef IS_MPI
1303 <    pot_row[atom1] += 0.5 *  *(idat.pot);
1304 <    pot_col[atom2] += 0.5 *  *(idat.pot);
1303 >    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1304 >    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1305 >    expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1306 >    expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1307  
1308      atomRowData.force[atom1] += *(idat.f1);
1309      atomColData.force[atom2] -= *(idat.f1);
1310 +
1311 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1312 +      atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1313 +      atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1314 +    }
1315 +
1316 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1317 +      atomRowData.electricField[atom1] += *(idat.eField1);
1318 +      atomColData.electricField[atom2] += *(idat.eField2);
1319 +    }
1320 +
1321   #else
1322      pairwisePot += *(idat.pot);
1323 +    excludedPot += *(idat.excludedPot);
1324  
1325      snap_->atomData.force[atom1] += *(idat.f1);
1326      snap_->atomData.force[atom2] -= *(idat.f1);
1327 +
1328 +    if (idat.doParticlePot) {
1329 +      // This is the pairwise contribution to the particle pot.  The
1330 +      // embedding contribution is added in each of the low level
1331 +      // non-bonded routines.  In parallel, this calculation is done
1332 +      // in collectData, not in unpackInteractionData.
1333 +      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1334 +      snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1335 +    }
1336 +    
1337 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1338 +      snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1339 +      snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1340 +    }
1341 +
1342 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1343 +      snap_->atomData.electricField[atom1] += *(idat.eField1);
1344 +      snap_->atomData.electricField[atom2] += *(idat.eField2);
1345 +    }
1346 +
1347   #endif
1348      
1349    }
# Line 1036 | Line 1354 | namespace OpenMD {
1354     * first element of pair is row-indexed CutoffGroup
1355     * second element of pair is column-indexed CutoffGroup
1356     */
1357 <  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1358 <      
1359 <    vector<pair<int, int> > neighborList;
1357 >  void ForceMatrixDecomposition::buildNeighborList(vector<pair<int,int> >& neighborList) {
1358 >    
1359 >    neighborList.clear();
1360      groupCutoffs cuts;
1361      bool doAllPairs = false;
1362  
1363 +    RealType rList_ = (largestRcut_ + skinThickness_);
1364 +    RealType rcut, rcutsq, rlistsq;
1365 +    Snapshot* snap_ = sman_->getCurrentSnapshot();
1366 +    Mat3x3d box;
1367 +    Mat3x3d invBox;
1368 +
1369 +    Vector3d rs, scaled, dr;
1370 +    Vector3i whichCell;
1371 +    int cellIndex;
1372 +
1373   #ifdef IS_MPI
1374      cellListRow_.clear();
1375      cellListCol_.clear();
1376   #else
1377      cellList_.clear();
1378   #endif
1379 <
1380 <    RealType rList_ = (largestRcut_ + skinThickness_);
1381 <    RealType rl2 = rList_ * rList_;
1382 <    Snapshot* snap_ = sman_->getCurrentSnapshot();
1383 <    Mat3x3d Hmat = snap_->getHmat();
1384 <    Vector3d Hx = Hmat.getColumn(0);
1385 <    Vector3d Hy = Hmat.getColumn(1);
1386 <    Vector3d Hz = Hmat.getColumn(2);
1387 <
1388 <    nCells_.x() = (int) ( Hx.length() )/ rList_;
1389 <    nCells_.y() = (int) ( Hy.length() )/ rList_;
1390 <    nCells_.z() = (int) ( Hz.length() )/ rList_;
1391 <
1379 >    
1380 >    if (!usePeriodicBoundaryConditions_) {
1381 >      box = snap_->getBoundingBox();
1382 >      invBox = snap_->getInvBoundingBox();
1383 >    } else {
1384 >      box = snap_->getHmat();
1385 >      invBox = snap_->getInvHmat();
1386 >    }
1387 >    
1388 >    Vector3d boxX = box.getColumn(0);
1389 >    Vector3d boxY = box.getColumn(1);
1390 >    Vector3d boxZ = box.getColumn(2);
1391 >    
1392 >    nCells_.x() = (int) ( boxX.length() )/ rList_;
1393 >    nCells_.y() = (int) ( boxY.length() )/ rList_;
1394 >    nCells_.z() = (int) ( boxZ.length() )/ rList_;
1395 >    
1396      // handle small boxes where the cell offsets can end up repeating cells
1397      
1398      if (nCells_.x() < 3) doAllPairs = true;
1399      if (nCells_.y() < 3) doAllPairs = true;
1400      if (nCells_.z() < 3) doAllPairs = true;
1401 <
1070 <    Mat3x3d invHmat = snap_->getInvHmat();
1071 <    Vector3d rs, scaled, dr;
1072 <    Vector3i whichCell;
1073 <    int cellIndex;
1401 >    
1402      int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1403 <
1403 >    
1404   #ifdef IS_MPI
1405      cellListRow_.resize(nCtot);
1406      cellListCol_.resize(nCtot);
1407   #else
1408      cellList_.resize(nCtot);
1409   #endif
1410 <
1410 >    
1411      if (!doAllPairs) {
1412   #ifdef IS_MPI
1413 <
1413 >      
1414        for (int i = 0; i < nGroupsInRow_; i++) {
1415          rs = cgRowData.position[i];
1416          
1417          // scaled positions relative to the box vectors
1418 <        scaled = invHmat * rs;
1418 >        scaled = invBox * rs;
1419          
1420          // wrap the vector back into the unit box by subtracting integer box
1421          // numbers
1422          for (int j = 0; j < 3; j++) {
1423            scaled[j] -= roundMe(scaled[j]);
1424            scaled[j] += 0.5;
1425 +          // Handle the special case when an object is exactly on the
1426 +          // boundary (a scaled coordinate of 1.0 is the same as
1427 +          // scaled coordinate of 0.0)
1428 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1429          }
1430          
1431          // find xyz-indices of cell that cutoffGroup is in.
# Line 1111 | Line 1443 | namespace OpenMD {
1443          rs = cgColData.position[i];
1444          
1445          // scaled positions relative to the box vectors
1446 <        scaled = invHmat * rs;
1446 >        scaled = invBox * rs;
1447          
1448          // wrap the vector back into the unit box by subtracting integer box
1449          // numbers
1450          for (int j = 0; j < 3; j++) {
1451            scaled[j] -= roundMe(scaled[j]);
1452            scaled[j] += 0.5;
1453 +          // Handle the special case when an object is exactly on the
1454 +          // boundary (a scaled coordinate of 1.0 is the same as
1455 +          // scaled coordinate of 0.0)
1456 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1457          }
1458          
1459          // find xyz-indices of cell that cutoffGroup is in.
# Line 1131 | Line 1467 | namespace OpenMD {
1467          // add this cutoff group to the list of groups in this cell;
1468          cellListCol_[cellIndex].push_back(i);
1469        }
1470 +      
1471   #else
1472        for (int i = 0; i < nGroups_; i++) {
1473          rs = snap_->cgData.position[i];
1474          
1475          // scaled positions relative to the box vectors
1476 <        scaled = invHmat * rs;
1476 >        scaled = invBox * rs;
1477          
1478          // wrap the vector back into the unit box by subtracting integer box
1479          // numbers
1480          for (int j = 0; j < 3; j++) {
1481            scaled[j] -= roundMe(scaled[j]);
1482            scaled[j] += 0.5;
1483 +          // Handle the special case when an object is exactly on the
1484 +          // boundary (a scaled coordinate of 1.0 is the same as
1485 +          // scaled coordinate of 0.0)
1486 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1487          }
1488          
1489          // find xyz-indices of cell that cutoffGroup is in.
# Line 1156 | Line 1497 | namespace OpenMD {
1497          // add this cutoff group to the list of groups in this cell;
1498          cellList_[cellIndex].push_back(i);
1499        }
1500 +
1501   #endif
1502  
1503        for (int m1z = 0; m1z < nCells_.z(); m1z++) {
# Line 1168 | Line 1510 | namespace OpenMD {
1510                   os != cellOffsets_.end(); ++os) {
1511                
1512                Vector3i m2v = m1v + (*os);
1513 <              
1513 >            
1514 >
1515                if (m2v.x() >= nCells_.x()) {
1516                  m2v.x() = 0;          
1517                } else if (m2v.x() < 0) {
# Line 1186 | Line 1529 | namespace OpenMD {
1529                } else if (m2v.z() < 0) {
1530                  m2v.z() = nCells_.z() - 1;
1531                }
1532 <              
1532 >
1533                int m2 = Vlinear (m2v, nCells_);
1534                
1535   #ifdef IS_MPI
# Line 1195 | Line 1538 | namespace OpenMD {
1538                  for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1539                       j2 != cellListCol_[m2].end(); ++j2) {
1540                    
1541 <                  // In parallel, we need to visit *all* pairs of row &
1542 <                  // column indicies and will truncate later on.
1541 >                  // In parallel, we need to visit *all* pairs of row
1542 >                  // & column indicies and will divide labor in the
1543 >                  // force evaluation later.
1544                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1545 <                  snap_->wrapVector(dr);
1546 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1547 <                  if (dr.lengthSquare() < cuts.third) {
1545 >                  if (usePeriodicBoundaryConditions_) {
1546 >                    snap_->wrapVector(dr);
1547 >                  }
1548 >                  getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq );
1549 >                  if (dr.lengthSquare() < rlistsq) {
1550                      neighborList.push_back(make_pair((*j1), (*j2)));
1551                    }                  
1552                  }
1553                }
1554   #else
1209 –              
1555                for (vector<int>::iterator j1 = cellList_[m1].begin();
1556                     j1 != cellList_[m1].end(); ++j1) {
1557                  for (vector<int>::iterator j2 = cellList_[m2].begin();
1558                       j2 != cellList_[m2].end(); ++j2) {
1559 <                  
1559 >    
1560                    // Always do this if we're in different cells or if
1561 <                  // we're in the same cell and the global index of the
1562 <                  // j2 cutoff group is less than the j1 cutoff group
1563 <                  
1564 <                  if (m2 != m1 || (*j2) < (*j1)) {
1561 >                  // we're in the same cell and the global index of
1562 >                  // the j2 cutoff group is greater than or equal to
1563 >                  // the j1 cutoff group.  Note that Rappaport's code
1564 >                  // has a "less than" conditional here, but that
1565 >                  // deals with atom-by-atom computation.  OpenMD
1566 >                  // allows atoms within a single cutoff group to
1567 >                  // interact with each other.
1568 >
1569 >                  if (m2 != m1 || (*j2) >= (*j1) ) {
1570 >
1571                      dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1572 <                    snap_->wrapVector(dr);
1573 <                    cuts = getGroupCutoffs( (*j1), (*j2) );
1574 <                    if (dr.lengthSquare() < cuts.third) {
1572 >                    if (usePeriodicBoundaryConditions_) {
1573 >                      snap_->wrapVector(dr);
1574 >                    }
1575 >                    getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq );
1576 >                    if (dr.lengthSquare() < rlistsq) {
1577                        neighborList.push_back(make_pair((*j1), (*j2)));
1578                      }
1579                    }
# Line 1235 | Line 1588 | namespace OpenMD {
1588        // branch to do all cutoff group pairs
1589   #ifdef IS_MPI
1590        for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1591 <        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1591 >        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1592            dr = cgColData.position[j2] - cgRowData.position[j1];
1593 <          snap_->wrapVector(dr);
1594 <          cuts = getGroupCutoffs( j1, j2 );
1595 <          if (dr.lengthSquare() < cuts.third) {
1593 >          if (usePeriodicBoundaryConditions_) {
1594 >            snap_->wrapVector(dr);
1595 >          }
1596 >          getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq);
1597 >          if (dr.lengthSquare() < rlistsq) {
1598              neighborList.push_back(make_pair(j1, j2));
1599            }
1600          }
1601 <      }
1601 >      }      
1602   #else
1603 <      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1604 <        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1603 >      // include all groups here.
1604 >      for (int j1 = 0; j1 < nGroups_; j1++) {
1605 >        // include self group interactions j2 == j1
1606 >        for (int j2 = j1; j2 < nGroups_; j2++) {
1607            dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1608 <          snap_->wrapVector(dr);
1609 <          cuts = getGroupCutoffs( j1, j2 );
1610 <          if (dr.lengthSquare() < cuts.third) {
1608 >          if (usePeriodicBoundaryConditions_) {
1609 >            snap_->wrapVector(dr);
1610 >          }
1611 >          getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq );
1612 >          if (dr.lengthSquare() < rlistsq) {
1613              neighborList.push_back(make_pair(j1, j2));
1614            }
1615 <        }
1616 <      }        
1615 >        }    
1616 >      }
1617   #endif
1618      }
1619        
# Line 1263 | Line 1622 | namespace OpenMD {
1622      saved_CG_positions_.clear();
1623      for (int i = 0; i < nGroups_; i++)
1624        saved_CG_positions_.push_back(snap_->cgData.position[i]);
1266 –    
1267 –    return neighborList;
1625    }
1626   } //end namespace OpenMD

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines