--- branches/development/src/parallel/ForceMatrixDecomposition.cpp	2011/05/25 16:20:37	1568
+++ branches/development/src/parallel/ForceMatrixDecomposition.cpp	2011/07/12 20:33:14	1592
@@ -42,6 +42,7 @@
 #include "math/SquareMatrix3.hpp"
 #include "nonbonded/NonBondedInteraction.hpp"
 #include "brains/SnapshotManager.hpp"
+#include "brains/PairList.hpp"
 
 using namespace std;
 namespace OpenMD {
@@ -54,20 +55,36 @@ namespace OpenMD {
   void ForceMatrixDecomposition::distributeInitialData() {
     snap_ = sman_->getCurrentSnapshot();
     storageLayout_ = sman_->getStorageLayout();
+    ff_ = info_->getForceField();
     nLocal_ = snap_->getNumberOfAtoms();
-    nGroups_ = snap_->getNumberOfCutoffGroups();
+    
+    nGroups_ = info_->getNLocalCutoffGroups();
+    // gather the information for atomtype IDs (atids):
+    idents = info_->getIdentArray();
+    AtomLocalToGlobal = info_->getGlobalAtomIndices();
+    cgLocalToGlobal = info_->getGlobalGroupIndices();
+    vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
 
+    massFactors = info_->getMassFactors();
+
+    PairList* excludes = info_->getExcludedInteractions();
+    PairList* oneTwo = info_->getOneTwoInteractions();
+    PairList* oneThree = info_->getOneThreeInteractions();
+    PairList* oneFour = info_->getOneFourInteractions();
+
 #ifdef IS_MPI
  
     AtomCommIntRow = new Communicator<Row,int>(nLocal_);
     AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
     AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
     AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
+    AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
 
     AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
     AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
     AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
     AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
+    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
 
     cgCommIntRow = new Communicator<Row,int>(nGroups_);
     cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
@@ -88,39 +105,412 @@ namespace OpenMD {
     cgRowData.setStorageLayout(DataStorage::dslPosition);
     cgColData.resize(nGroupsInCol_);
     cgColData.setStorageLayout(DataStorage::dslPosition);
+        
+    identsRow.resize(nAtomsInRow_);
+    identsCol.resize(nAtomsInCol_);
     
-    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES, 
-                                      vector<RealType> (nAtomsInRow_, 0.0));
-    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
-                                      vector<RealType> (nAtomsInCol_, 0.0));
+    AtomCommIntRow->gather(idents, identsRow);
+    AtomCommIntColumn->gather(idents, identsCol);
+    
+    // allocate memory for the parallel objects
+    atypesRow.resize(nAtomsInRow_);
+    atypesCol.resize(nAtomsInCol_);
 
+    for (int i = 0; i < nAtomsInRow_; i++) 
+      atypesRow[i] = ff_->getAtomType(identsRow[i]);
+    for (int i = 0; i < nAtomsInCol_; i++) 
+      atypesCol[i] = ff_->getAtomType(identsCol[i]);         
 
-    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
-    
-    // gather the information for atomtype IDs (atids):
-    vector<int> identsLocal = info_->getIdentArray();
-    identsRow.reserve(nAtomsInRow_);
-    identsCol.reserve(nAtomsInCol_);
-    
-    AtomCommIntRow->gather(identsLocal, identsRow);
-    AtomCommIntColumn->gather(identsLocal, identsCol);
-    
-    AtomLocalToGlobal = info_->getGlobalAtomIndices();
+    pot_row.resize(nAtomsInRow_);
+    pot_col.resize(nAtomsInCol_);
+
+    AtomRowToGlobal.resize(nAtomsInRow_);
+    AtomColToGlobal.resize(nAtomsInCol_);
     AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
     AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
     
-    cgLocalToGlobal = info_->getGlobalGroupIndices();
+    cgRowToGlobal.resize(nGroupsInRow_);
+    cgColToGlobal.resize(nGroupsInCol_);
     cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
     cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
 
-    // still need:
-    // topoDist
-    // exclude
+    massFactorsRow.resize(nAtomsInRow_);
+    massFactorsCol.resize(nAtomsInCol_);
+    AtomCommRealRow->gather(massFactors, massFactorsRow);
+    AtomCommRealColumn->gather(massFactors, massFactorsCol);
+
+    groupListRow_.clear();
+    groupListRow_.resize(nGroupsInRow_);
+    for (int i = 0; i < nGroupsInRow_; i++) {
+      int gid = cgRowToGlobal[i];
+      for (int j = 0; j < nAtomsInRow_; j++) {
+        int aid = AtomRowToGlobal[j];
+        if (globalGroupMembership[aid] == gid)
+          groupListRow_[i].push_back(j);
+      }      
+    }
+
+    groupListCol_.clear();
+    groupListCol_.resize(nGroupsInCol_);
+    for (int i = 0; i < nGroupsInCol_; i++) {
+      int gid = cgColToGlobal[i];
+      for (int j = 0; j < nAtomsInCol_; j++) {
+        int aid = AtomColToGlobal[j];
+        if (globalGroupMembership[aid] == gid)
+          groupListCol_[i].push_back(j);
+      }      
+    }
+
+    excludesForAtom.clear();
+    excludesForAtom.resize(nAtomsInRow_);
+    toposForAtom.clear();
+    toposForAtom.resize(nAtomsInRow_);
+    topoDist.clear();
+    topoDist.resize(nAtomsInRow_);
+    for (int i = 0; i < nAtomsInRow_; i++) {
+      int iglob = AtomRowToGlobal[i];
+
+      for (int j = 0; j < nAtomsInCol_; j++) {
+        int jglob = AtomColToGlobal[j];
+
+        if (excludes->hasPair(iglob, jglob)) 
+          excludesForAtom[i].push_back(j);       
+        
+        if (oneTwo->hasPair(iglob, jglob)) {
+          toposForAtom[i].push_back(j);
+          topoDist[i].push_back(1);
+        } else {
+          if (oneThree->hasPair(iglob, jglob)) {
+            toposForAtom[i].push_back(j);
+            topoDist[i].push_back(2);
+          } else {
+            if (oneFour->hasPair(iglob, jglob)) {
+              toposForAtom[i].push_back(j);
+              topoDist[i].push_back(3);
+            }
+          }
+        }
+      }      
+    }
+
+#endif
+
+    // allocate memory for the parallel objects
+    atypesLocal.resize(nLocal_);
+
+    for (int i = 0; i < nLocal_; i++) 
+      atypesLocal[i] = ff_->getAtomType(idents[i]);
+
+    groupList_.clear();
+    groupList_.resize(nGroups_);
+    for (int i = 0; i < nGroups_; i++) {
+      int gid = cgLocalToGlobal[i];
+      for (int j = 0; j < nLocal_; j++) {
+        int aid = AtomLocalToGlobal[j];
+        if (globalGroupMembership[aid] == gid) {
+          groupList_[i].push_back(j);
+        }
+      }      
+    }
+
+    excludesForAtom.clear();
+    excludesForAtom.resize(nLocal_);
+    toposForAtom.clear();
+    toposForAtom.resize(nLocal_);
+    topoDist.clear();
+    topoDist.resize(nLocal_);
+
+    for (int i = 0; i < nLocal_; i++) {
+      int iglob = AtomLocalToGlobal[i];
+
+      for (int j = 0; j < nLocal_; j++) {
+        int jglob = AtomLocalToGlobal[j];
+
+        if (excludes->hasPair(iglob, jglob)) 
+          excludesForAtom[i].push_back(j);              
+        
+        if (oneTwo->hasPair(iglob, jglob)) {
+          toposForAtom[i].push_back(j);
+          topoDist[i].push_back(1);
+        } else {
+          if (oneThree->hasPair(iglob, jglob)) {
+            toposForAtom[i].push_back(j);
+            topoDist[i].push_back(2);
+          } else {
+            if (oneFour->hasPair(iglob, jglob)) {
+              toposForAtom[i].push_back(j);
+              topoDist[i].push_back(3);
+            }
+          }
+        }
+      }      
+    }
+    
+    createGtypeCutoffMap();
+
+  }
+   
+  void ForceMatrixDecomposition::createGtypeCutoffMap() {
+    
+    RealType tol = 1e-6;
+    largestRcut_ = 0.0;
+    RealType rc;
+    int atid;
+    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
+    
+    map<int, RealType> atypeCutoff;
+      
+    for (set<AtomType*>::iterator at = atypes.begin(); 
+         at != atypes.end(); ++at){
+      atid = (*at)->getIdent();
+      if (userChoseCutoff_) 
+        atypeCutoff[atid] = userCutoff_;
+      else
+        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
+    }
+    
+    vector<RealType> gTypeCutoffs;
+    // first we do a single loop over the cutoff groups to find the
+    // largest cutoff for any atypes present in this group.
+#ifdef IS_MPI
+    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
+    groupRowToGtype.resize(nGroupsInRow_);
+    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
+      vector<int> atomListRow = getAtomsInGroupRow(cg1);
+      for (vector<int>::iterator ia = atomListRow.begin(); 
+           ia != atomListRow.end(); ++ia) {            
+        int atom1 = (*ia);
+        atid = identsRow[atom1];
+        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
+          groupCutoffRow[cg1] = atypeCutoff[atid];
+        }
+      }
+
+      bool gTypeFound = false;
+      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
+        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
+          groupRowToGtype[cg1] = gt;
+          gTypeFound = true;
+        } 
+      }
+      if (!gTypeFound) {
+        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
+        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
+      }
+      
+    }
+    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
+    groupColToGtype.resize(nGroupsInCol_);
+    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
+      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
+      for (vector<int>::iterator jb = atomListCol.begin(); 
+           jb != atomListCol.end(); ++jb) {            
+        int atom2 = (*jb);
+        atid = identsCol[atom2];
+        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
+          groupCutoffCol[cg2] = atypeCutoff[atid];
+        }
+      }
+      bool gTypeFound = false;
+      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
+        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
+          groupColToGtype[cg2] = gt;
+          gTypeFound = true;
+        } 
+      }
+      if (!gTypeFound) {
+        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
+        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
+      }
+    }
+#else
+
+    vector<RealType> groupCutoff(nGroups_, 0.0);
+    groupToGtype.resize(nGroups_);
+    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
+      groupCutoff[cg1] = 0.0;
+      vector<int> atomList = getAtomsInGroupRow(cg1);
+      for (vector<int>::iterator ia = atomList.begin(); 
+           ia != atomList.end(); ++ia) {            
+        int atom1 = (*ia);
+        atid = idents[atom1];
+        if (atypeCutoff[atid] > groupCutoff[cg1]) 
+          groupCutoff[cg1] = atypeCutoff[atid];
+      }
+      
+      bool gTypeFound = false;
+      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
+        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
+          groupToGtype[cg1] = gt;
+          gTypeFound = true;
+        } 
+      }
+      if (!gTypeFound) {      
+        gTypeCutoffs.push_back( groupCutoff[cg1] );
+        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
+      }      
+    }
 #endif
+
+    // Now we find the maximum group cutoff value present in the simulation
+
+    RealType groupMax = *max_element(gTypeCutoffs.begin(), 
+                                     gTypeCutoffs.end());
+
+#ifdef IS_MPI
+    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, 
+                              MPI::MAX);
+#endif
+    
+    RealType tradRcut = groupMax;
+
+    for (int i = 0; i < gTypeCutoffs.size();  i++) {
+      for (int j = 0; j < gTypeCutoffs.size();  j++) {       
+        RealType thisRcut;
+        switch(cutoffPolicy_) {
+        case TRADITIONAL:
+          thisRcut = tradRcut;
+          break;
+        case MIX:
+          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
+          break;
+        case MAX:
+          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
+          break;
+        default:
+          sprintf(painCave.errMsg,
+                  "ForceMatrixDecomposition::createGtypeCutoffMap " 
+                  "hit an unknown cutoff policy!\n");
+          painCave.severity = OPENMD_ERROR;
+          painCave.isFatal = 1;
+          simError();
+          break;
+        }
+
+        pair<int,int> key = make_pair(i,j);
+        gTypeCutoffMap[key].first = thisRcut;
+        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
+        gTypeCutoffMap[key].second = thisRcut*thisRcut;
+        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
+        // sanity check
+        
+        if (userChoseCutoff_) {
+          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
+            sprintf(painCave.errMsg,
+                    "ForceMatrixDecomposition::createGtypeCutoffMap " 
+                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
+            painCave.severity = OPENMD_ERROR;
+            painCave.isFatal = 1;
+            simError();            
+          }
+        }
+      }
+    }
   }
+
+
+  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
+    int i, j;   
+#ifdef IS_MPI
+    i = groupRowToGtype[cg1];
+    j = groupColToGtype[cg2];
+#else
+    i = groupToGtype[cg1];
+    j = groupToGtype[cg2];
+#endif    
+    return gTypeCutoffMap[make_pair(i,j)];
+  }
+
+  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
+    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
+      if (toposForAtom[atom1][j] == atom2) 
+        return topoDist[atom1][j];
+    }
+    return 0;
+  }
+
+  void ForceMatrixDecomposition::zeroWorkArrays() {
+    pairwisePot = 0.0;
+    embeddingPot = 0.0;
+
+#ifdef IS_MPI
+    if (storageLayout_ & DataStorage::dslForce) {
+      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
+      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
+    }
+
+    if (storageLayout_ & DataStorage::dslTorque) {
+      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
+      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
+    }
     
+    fill(pot_row.begin(), pot_row.end(), 
+         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
 
+    fill(pot_col.begin(), pot_col.end(), 
+         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));   
 
+    if (storageLayout_ & DataStorage::dslParticlePot) {    
+      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
+           0.0);
+      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
+           0.0);
+    }
+
+    if (storageLayout_ & DataStorage::dslDensity) {      
+      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
+      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
+    }
+
+    if (storageLayout_ & DataStorage::dslFunctional) {   
+      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
+           0.0);
+      fill(atomColData.functional.begin(), atomColData.functional.end(),
+           0.0);
+    }
+
+    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
+      fill(atomRowData.functionalDerivative.begin(), 
+           atomRowData.functionalDerivative.end(), 0.0);
+      fill(atomColData.functionalDerivative.begin(), 
+           atomColData.functionalDerivative.end(), 0.0);
+    }
+
+    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
+      fill(atomRowData.skippedCharge.begin(), 
+           atomRowData.skippedCharge.end(), 0.0);
+      fill(atomColData.skippedCharge.begin(), 
+           atomColData.skippedCharge.end(), 0.0);
+    }
+
+#endif
+    // even in parallel, we need to zero out the local arrays:
+
+    if (storageLayout_ & DataStorage::dslParticlePot) {      
+      fill(snap_->atomData.particlePot.begin(), 
+           snap_->atomData.particlePot.end(), 0.0);
+    }
+    
+    if (storageLayout_ & DataStorage::dslDensity) {      
+      fill(snap_->atomData.density.begin(), 
+           snap_->atomData.density.end(), 0.0);
+    }
+    if (storageLayout_ & DataStorage::dslFunctional) {
+      fill(snap_->atomData.functional.begin(), 
+           snap_->atomData.functional.end(), 0.0);
+    }
+    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
+      fill(snap_->atomData.functionalDerivative.begin(), 
+           snap_->atomData.functionalDerivative.end(), 0.0);
+    }
+    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
+      fill(snap_->atomData.skippedCharge.begin(), 
+           snap_->atomData.skippedCharge.end(), 0.0);
+    }
+    
+  }
+
+
   void ForceMatrixDecomposition::distributeData()  {
     snap_ = sman_->getCurrentSnapshot();
     storageLayout_ = sman_->getStorageLayout();
@@ -153,9 +543,13 @@ namespace OpenMD {
       AtomCommMatrixColumn->gather(snap_->atomData.electroFrame, 
                                    atomColData.electroFrame);
     }
+
 #endif      
   }
   
+  /* collects information obtained during the pre-pair loop onto local
+   * data structures.
+   */
   void ForceMatrixDecomposition::collectIntermediateData() {
     snap_ = sman_->getCurrentSnapshot();
     storageLayout_ = sman_->getStorageLayout();
@@ -167,14 +561,18 @@ namespace OpenMD {
                                snap_->atomData.density);
       
       int n = snap_->atomData.density.size();
-      std::vector<RealType> rho_tmp(n, 0.0);
+      vector<RealType> rho_tmp(n, 0.0);
       AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
       for (int i = 0; i < n; i++)
         snap_->atomData.density[i] += rho_tmp[i];
     }
 #endif
   }
-  
+
+  /*
+   * redistributes information obtained during the pre-pair loop out to 
+   * row and column-indexed data structures
+   */
   void ForceMatrixDecomposition::distributeIntermediateData() {
     snap_ = sman_->getCurrentSnapshot();
     storageLayout_ = sman_->getStorageLayout();
@@ -212,38 +610,88 @@ namespace OpenMD {
     AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
     for (int i = 0; i < n; i++)
       snap_->atomData.force[i] += frc_tmp[i];
-    
-    
+        
     if (storageLayout_ & DataStorage::dslTorque) {
 
-      int nt = snap_->atomData.force.size();
+      int nt = snap_->atomData.torque.size();
       vector<Vector3d> trq_tmp(nt, V3Zero);
 
       AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
-      for (int i = 0; i < n; i++) {
+      for (int i = 0; i < nt; i++) {
         snap_->atomData.torque[i] += trq_tmp[i];
         trq_tmp[i] = 0.0;
       }
       
       AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
-      for (int i = 0; i < n; i++)
+      for (int i = 0; i < nt; i++)
         snap_->atomData.torque[i] += trq_tmp[i];
     }
+
+    if (storageLayout_ & DataStorage::dslSkippedCharge) {
+
+      int ns = snap_->atomData.skippedCharge.size();
+      vector<RealType> skch_tmp(ns, 0.0);
+
+      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
+      for (int i = 0; i < ns; i++) {
+        snap_->atomData.skippedCharge[i] += skch_tmp[i];
+        skch_tmp[i] = 0.0;
+      }
+      
+      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
+      for (int i = 0; i < ns; i++)
+        snap_->atomData.skippedCharge[i] += skch_tmp[i];
+    }
     
     nLocal_ = snap_->getNumberOfAtoms();
 
-    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES, 
-                                       vector<RealType> (nLocal_, 0.0));
+    vector<potVec> pot_temp(nLocal_, 
+                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
+
+    // scatter/gather pot_row into the members of my column
+          
+    AtomCommPotRow->scatter(pot_row, pot_temp);
+
+    for (int ii = 0;  ii < pot_temp.size(); ii++ )
+      pairwisePot += pot_temp[ii];
     
-    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
-      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
-      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
-        pot_local[i] += pot_temp[i][ii];
-      }
-    }
+    fill(pot_temp.begin(), pot_temp.end(), 
+         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
+      
+    AtomCommPotColumn->scatter(pot_col, pot_temp);    
+    
+    for (int ii = 0;  ii < pot_temp.size(); ii++ )
+      pairwisePot += pot_temp[ii];    
 #endif
+
   }
 
+  int ForceMatrixDecomposition::getNAtomsInRow() {   
+#ifdef IS_MPI
+    return nAtomsInRow_;
+#else
+    return nLocal_;
+#endif
+  }
+
+  /**
+   * returns the list of atoms belonging to this group.  
+   */
+  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
+#ifdef IS_MPI
+    return groupListRow_[cg1];
+#else 
+    return groupList_[cg1];
+#endif
+  }
+
+  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
+#ifdef IS_MPI
+    return groupListCol_[cg2];
+#else 
+    return groupList_[cg2];
+#endif
+  }
   
   Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
     Vector3d d;
@@ -285,6 +733,23 @@ namespace OpenMD {
     snap_->wrapVector(d);
     return d;    
   }
+
+  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
+#ifdef IS_MPI
+    return massFactorsRow[atom1];
+#else
+    return massFactors[atom1];
+#endif
+  }
+
+  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
+#ifdef IS_MPI
+    return massFactorsCol[atom2];
+#else
+    return massFactors[atom2];
+#endif
+
+  }
     
   Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
     Vector3d d;
@@ -299,6 +764,64 @@ namespace OpenMD {
     return d;    
   }
 
+  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
+    return excludesForAtom[atom1];
+  }
+
+  /**
+   * We need to exclude some overcounted interactions that result from
+   * the parallel decomposition.
+   */
+  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
+    int unique_id_1, unique_id_2;
+
+#ifdef IS_MPI
+    // in MPI, we have to look up the unique IDs for each atom
+    unique_id_1 = AtomRowToGlobal[atom1];
+    unique_id_2 = AtomColToGlobal[atom2];
+
+    // this situation should only arise in MPI simulations
+    if (unique_id_1 == unique_id_2) return true;
+    
+    // this prevents us from doing the pair on multiple processors
+    if (unique_id_1 < unique_id_2) {
+      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
+    } else {
+      if ((unique_id_1 + unique_id_2) % 2 == 1) return true; 
+    }
+#endif
+    return false;
+  }
+
+  /**
+   * We need to handle the interactions for atoms who are involved in
+   * the same rigid body as well as some short range interactions
+   * (bonds, bends, torsions) differently from other interactions.
+   * We'll still visit the pairwise routines, but with a flag that
+   * tells those routines to exclude the pair from direct long range
+   * interactions.  Some indirect interactions (notably reaction
+   * field) must still be handled for these pairs.
+   */
+  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
+    int unique_id_2;
+    
+#ifdef IS_MPI
+    // in MPI, we have to look up the unique IDs for the row atom.
+    unique_id_2 = AtomColToGlobal[atom2];
+#else
+    // in the normal loop, the atom numbers are unique
+    unique_id_2 = atom2;
+#endif
+    
+    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
+         i != excludesForAtom[atom1].end(); ++i) {
+      if ( (*i) == unique_id_2 ) return true;
+    }
+
+    return false;
+  }
+
+
   void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
 #ifdef IS_MPI
     atomRowData.force[atom1] += fg;
@@ -316,10 +839,16 @@ namespace OpenMD {
   }
 
     // filling interaction blocks with pointers
-  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
-    InteractionData idat;
+  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat, 
+                                                     int atom1, int atom2) {
 
+    idat.excluded = excludeAtomPair(atom1, atom2);
+   
 #ifdef IS_MPI
+    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
+    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]), 
+    //                         ff_->getAtomType(identsCol[atom2]) );
+    
     if (storageLayout_ & DataStorage::dslAmat) {
       idat.A1 = &(atomRowData.aMat[atom1]);
       idat.A2 = &(atomColData.aMat[atom2]);
@@ -340,11 +869,32 @@ namespace OpenMD {
       idat.rho2 = &(atomColData.density[atom2]);
     }
 
+    if (storageLayout_ & DataStorage::dslFunctional) {
+      idat.frho1 = &(atomRowData.functional[atom1]);
+      idat.frho2 = &(atomColData.functional[atom2]);
+    }
+
     if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
       idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
       idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
     }
+
+    if (storageLayout_ & DataStorage::dslParticlePot) {
+      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
+      idat.particlePot2 = &(atomColData.particlePot[atom2]);
+    }
+
+    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
+      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
+      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
+    }
+
 #else
+
+    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
+    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]), 
+    //                         ff_->getAtomType(idents[atom2]) );
+
     if (storageLayout_ & DataStorage::dslAmat) {
       idat.A1 = &(snap_->atomData.aMat[atom1]);
       idat.A2 = &(snap_->atomData.aMat[atom2]);
@@ -360,55 +910,50 @@ namespace OpenMD {
       idat.t2 = &(snap_->atomData.torque[atom2]);
     }
 
-    if (storageLayout_ & DataStorage::dslDensity) {
+    if (storageLayout_ & DataStorage::dslDensity) {     
       idat.rho1 = &(snap_->atomData.density[atom1]);
       idat.rho2 = &(snap_->atomData.density[atom2]);
+    }
+
+    if (storageLayout_ & DataStorage::dslFunctional) {
+      idat.frho1 = &(snap_->atomData.functional[atom1]);
+      idat.frho2 = &(snap_->atomData.functional[atom2]);
     }
 
     if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
       idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
       idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
     }
+
+    if (storageLayout_ & DataStorage::dslParticlePot) {
+      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
+      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
+    }
+
+    if (storageLayout_ & DataStorage::dslSkippedCharge) {
+      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
+      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
+    }
 #endif
-    return idat;
   }
 
-  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
-
-    InteractionData idat;
+  
+  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
 #ifdef IS_MPI
-    if (storageLayout_ & DataStorage::dslElectroFrame) {
-      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
-      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
-    }
-    if (storageLayout_ & DataStorage::dslTorque) {
-      idat.t1 = &(atomRowData.torque[atom1]);
-      idat.t2 = &(atomColData.torque[atom2]);
-    }
-    if (storageLayout_ & DataStorage::dslForce) {
-      idat.t1 = &(atomRowData.force[atom1]);
-      idat.t2 = &(atomColData.force[atom2]);
-    }
+    pot_row[atom1] += 0.5 *  *(idat.pot);
+    pot_col[atom2] += 0.5 *  *(idat.pot);
+
+    atomRowData.force[atom1] += *(idat.f1);
+    atomColData.force[atom2] -= *(idat.f1);
 #else
-    if (storageLayout_ & DataStorage::dslElectroFrame) {
-      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
-      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
-    }
-    if (storageLayout_ & DataStorage::dslTorque) {
-      idat.t1 = &(snap_->atomData.torque[atom1]);
-      idat.t2 = &(snap_->atomData.torque[atom2]);
-    }
-    if (storageLayout_ & DataStorage::dslForce) {
-      idat.t1 = &(snap_->atomData.force[atom1]);
-      idat.t2 = &(snap_->atomData.force[atom2]);
-    }
+    pairwisePot += *(idat.pot);
+
+    snap_->atomData.force[atom1] += *(idat.f1);
+    snap_->atomData.force[atom2] -= *(idat.f1);
 #endif
     
   }
 
-
-
-
   /*
    * buildNeighborList
    *
@@ -418,6 +963,9 @@ namespace OpenMD {
   vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
       
     vector<pair<int, int> > neighborList;
+    groupCutoffs cuts;
+    bool doAllPairs = false;
+
 #ifdef IS_MPI
     cellListRow_.clear();
     cellListCol_.clear();
@@ -425,10 +973,7 @@ namespace OpenMD {
     cellList_.clear();
 #endif
 
-    // dangerous to not do error checking.
-    RealType rCut_;
- 
-    RealType rList_ = (rCut_ + skinThickness_);
+    RealType rList_ = (largestRcut_ + skinThickness_);
     RealType rl2 = rList_ * rList_;
     Snapshot* snap_ = sman_->getCurrentSnapshot();
     Mat3x3d Hmat = snap_->getHmat();
@@ -440,156 +985,214 @@ namespace OpenMD {
     nCells_.y() = (int) ( Hy.length() )/ rList_;
     nCells_.z() = (int) ( Hz.length() )/ rList_;
 
+    // handle small boxes where the cell offsets can end up repeating cells
+    
+    if (nCells_.x() < 3) doAllPairs = true;
+    if (nCells_.y() < 3) doAllPairs = true;
+    if (nCells_.z() < 3) doAllPairs = true;
+
     Mat3x3d invHmat = snap_->getInvHmat();
     Vector3d rs, scaled, dr;
     Vector3i whichCell;
     int cellIndex;
+    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
 
 #ifdef IS_MPI
-    for (int i = 0; i < nGroupsInRow_; i++) {
-      rs = cgRowData.position[i];
-      // scaled positions relative to the box vectors
-      scaled = invHmat * rs;
-      // wrap the vector back into the unit box by subtracting integer box 
-      // numbers
-      for (int j = 0; j < 3; j++) 
-        scaled[j] -= roundMe(scaled[j]);
-     
-      // find xyz-indices of cell that cutoffGroup is in.
-      whichCell.x() = nCells_.x() * scaled.x();
-      whichCell.y() = nCells_.y() * scaled.y();
-      whichCell.z() = nCells_.z() * scaled.z();
+    cellListRow_.resize(nCtot);
+    cellListCol_.resize(nCtot);
+#else
+    cellList_.resize(nCtot);
+#endif
 
-      // find single index of this cell:
-      cellIndex = Vlinear(whichCell, nCells_);
-      // add this cutoff group to the list of groups in this cell;
-      cellListRow_[cellIndex].push_back(i);
-    }
+    if (!doAllPairs) {
+#ifdef IS_MPI
 
-    for (int i = 0; i < nGroupsInCol_; i++) {
-      rs = cgColData.position[i];
-      // scaled positions relative to the box vectors
-      scaled = invHmat * rs;
-      // wrap the vector back into the unit box by subtracting integer box 
-      // numbers
-      for (int j = 0; j < 3; j++) 
-        scaled[j] -= roundMe(scaled[j]);
-
-      // find xyz-indices of cell that cutoffGroup is in.
-      whichCell.x() = nCells_.x() * scaled.x();
-      whichCell.y() = nCells_.y() * scaled.y();
-      whichCell.z() = nCells_.z() * scaled.z();
-
-      // find single index of this cell:
-      cellIndex = Vlinear(whichCell, nCells_);
-      // add this cutoff group to the list of groups in this cell;
-      cellListCol_[cellIndex].push_back(i);
-    }
+      for (int i = 0; i < nGroupsInRow_; i++) {
+        rs = cgRowData.position[i];
+        
+        // scaled positions relative to the box vectors
+        scaled = invHmat * rs;
+        
+        // wrap the vector back into the unit box by subtracting integer box 
+        // numbers
+        for (int j = 0; j < 3; j++) {
+          scaled[j] -= roundMe(scaled[j]);
+          scaled[j] += 0.5;
+        }
+        
+        // find xyz-indices of cell that cutoffGroup is in.
+        whichCell.x() = nCells_.x() * scaled.x();
+        whichCell.y() = nCells_.y() * scaled.y();
+        whichCell.z() = nCells_.z() * scaled.z();
+        
+        // find single index of this cell:
+        cellIndex = Vlinear(whichCell, nCells_);
+        
+        // add this cutoff group to the list of groups in this cell;
+        cellListRow_[cellIndex].push_back(i);
+      }
+      
+      for (int i = 0; i < nGroupsInCol_; i++) {
+        rs = cgColData.position[i];
+        
+        // scaled positions relative to the box vectors
+        scaled = invHmat * rs;
+        
+        // wrap the vector back into the unit box by subtracting integer box 
+        // numbers
+        for (int j = 0; j < 3; j++) {
+          scaled[j] -= roundMe(scaled[j]);
+          scaled[j] += 0.5;
+        }
+        
+        // find xyz-indices of cell that cutoffGroup is in.
+        whichCell.x() = nCells_.x() * scaled.x();
+        whichCell.y() = nCells_.y() * scaled.y();
+        whichCell.z() = nCells_.z() * scaled.z();
+        
+        // find single index of this cell:
+        cellIndex = Vlinear(whichCell, nCells_);
+        
+        // add this cutoff group to the list of groups in this cell;
+        cellListCol_[cellIndex].push_back(i);
+      }
 #else
-    for (int i = 0; i < nGroups_; i++) {
-      rs = snap_->cgData.position[i];
-      // scaled positions relative to the box vectors
-      scaled = invHmat * rs;
-      // wrap the vector back into the unit box by subtracting integer box 
-      // numbers
-      for (int j = 0; j < 3; j++) 
-        scaled[j] -= roundMe(scaled[j]);
-
-      // find xyz-indices of cell that cutoffGroup is in.
-      whichCell.x() = nCells_.x() * scaled.x();
-      whichCell.y() = nCells_.y() * scaled.y();
-      whichCell.z() = nCells_.z() * scaled.z();
-
-      // find single index of this cell:
-      cellIndex = Vlinear(whichCell, nCells_);
-      // add this cutoff group to the list of groups in this cell;
-      cellList_[cellIndex].push_back(i);
-    }
+      for (int i = 0; i < nGroups_; i++) {
+        rs = snap_->cgData.position[i];
+        
+        // scaled positions relative to the box vectors
+        scaled = invHmat * rs;
+        
+        // wrap the vector back into the unit box by subtracting integer box 
+        // numbers
+        for (int j = 0; j < 3; j++) {
+          scaled[j] -= roundMe(scaled[j]);
+          scaled[j] += 0.5;
+        }
+        
+        // find xyz-indices of cell that cutoffGroup is in.
+        whichCell.x() = nCells_.x() * scaled.x();
+        whichCell.y() = nCells_.y() * scaled.y();
+        whichCell.z() = nCells_.z() * scaled.z();
+        
+        // find single index of this cell:
+        cellIndex = Vlinear(whichCell, nCells_);      
+        
+        // add this cutoff group to the list of groups in this cell;
+        cellList_[cellIndex].push_back(i);
+      }
 #endif
 
-
-
-    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
-      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
-        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
-          Vector3i m1v(m1x, m1y, m1z);
-          int m1 = Vlinear(m1v, nCells_);
-
-          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
-               os != cellOffsets_.end(); ++os) {
+      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
+        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
+          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
+            Vector3i m1v(m1x, m1y, m1z);
+            int m1 = Vlinear(m1v, nCells_);
             
-            Vector3i m2v = m1v + (*os);
-            
-            if (m2v.x() >= nCells_.x()) {
-              m2v.x() = 0;           
-            } else if (m2v.x() < 0) {
-              m2v.x() = nCells_.x() - 1; 
-            }
-            
-            if (m2v.y() >= nCells_.y()) {
-              m2v.y() = 0;           
-            } else if (m2v.y() < 0) {
-              m2v.y() = nCells_.y() - 1; 
-            }
-            
-            if (m2v.z() >= nCells_.z()) {
-              m2v.z() = 0;           
-            } else if (m2v.z() < 0) {
-              m2v.z() = nCells_.z() - 1; 
-            }
-            
-            int m2 = Vlinear (m2v, nCells_);
-
+            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
+                 os != cellOffsets_.end(); ++os) {
+              
+              Vector3i m2v = m1v + (*os);
+              
+              if (m2v.x() >= nCells_.x()) {
+                m2v.x() = 0;           
+              } else if (m2v.x() < 0) {
+                m2v.x() = nCells_.x() - 1; 
+              }
+              
+              if (m2v.y() >= nCells_.y()) {
+                m2v.y() = 0;           
+              } else if (m2v.y() < 0) {
+                m2v.y() = nCells_.y() - 1; 
+              }
+              
+              if (m2v.z() >= nCells_.z()) {
+                m2v.z() = 0;           
+              } else if (m2v.z() < 0) {
+                m2v.z() = nCells_.z() - 1; 
+              }
+              
+              int m2 = Vlinear (m2v, nCells_);
+              
 #ifdef IS_MPI
-            for (vector<int>::iterator j1 = cellListRow_[m1].begin(); 
-                 j1 != cellListRow_[m1].end(); ++j1) {
-              for (vector<int>::iterator j2 = cellListCol_[m2].begin(); 
-                   j2 != cellListCol_[m2].end(); ++j2) {
-                               
-                // Always do this if we're in different cells or if
-                // we're in the same cell and the global index of the
-                // j2 cutoff group is less than the j1 cutoff group
-
-                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
-                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
-                  snap_->wrapVector(dr);
-                  if (dr.lengthSquare() < rl2) {
-                    neighborList.push_back(make_pair((*j1), (*j2)));
+              for (vector<int>::iterator j1 = cellListRow_[m1].begin(); 
+                   j1 != cellListRow_[m1].end(); ++j1) {
+                for (vector<int>::iterator j2 = cellListCol_[m2].begin(); 
+                     j2 != cellListCol_[m2].end(); ++j2) {
+                  
+                  // Always do this if we're in different cells or if
+                  // we're in the same cell and the global index of the
+                  // j2 cutoff group is less than the j1 cutoff group
+                  
+                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
+                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
+                    snap_->wrapVector(dr);
+                    cuts = getGroupCutoffs( (*j1), (*j2) );
+                    if (dr.lengthSquare() < cuts.third) {
+                      neighborList.push_back(make_pair((*j1), (*j2)));
+                    }
                   }
                 }
               }
-            }
 #else
-            for (vector<int>::iterator j1 = cellList_[m1].begin(); 
-                 j1 != cellList_[m1].end(); ++j1) {
-              for (vector<int>::iterator j2 = cellList_[m2].begin(); 
-                   j2 != cellList_[m2].end(); ++j2) {
-                               
-                // Always do this if we're in different cells or if
-                // we're in the same cell and the global index of the
-                // j2 cutoff group is less than the j1 cutoff group
-
-                if (m2 != m1 || (*j2) < (*j1)) {
-                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
-                  snap_->wrapVector(dr);
-                  if (dr.lengthSquare() < rl2) {
-                    neighborList.push_back(make_pair((*j1), (*j2)));
+              
+              for (vector<int>::iterator j1 = cellList_[m1].begin(); 
+                   j1 != cellList_[m1].end(); ++j1) {
+                for (vector<int>::iterator j2 = cellList_[m2].begin(); 
+                     j2 != cellList_[m2].end(); ++j2) {
+                  
+                  // Always do this if we're in different cells or if
+                  // we're in the same cell and the global index of the
+                  // j2 cutoff group is less than the j1 cutoff group
+                  
+                  if (m2 != m1 || (*j2) < (*j1)) {
+                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
+                    snap_->wrapVector(dr);
+                    cuts = getGroupCutoffs( (*j1), (*j2) );
+                    if (dr.lengthSquare() < cuts.third) {
+                      neighborList.push_back(make_pair((*j1), (*j2)));
+                    }
                   }
                 }
               }
-            }
 #endif
+            }
           }
         }
       }
+    } else {
+      // branch to do all cutoff group pairs
+#ifdef IS_MPI
+      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
+        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {       
+          dr = cgColData.position[j2] - cgRowData.position[j1];
+          snap_->wrapVector(dr);
+          cuts = getGroupCutoffs( j1, j2 );
+          if (dr.lengthSquare() < cuts.third) {
+            neighborList.push_back(make_pair(j1, j2));
+          }
+        }
+      }
+#else
+      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
+        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
+          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
+          snap_->wrapVector(dr);
+          cuts = getGroupCutoffs( j1, j2 );
+          if (dr.lengthSquare() < cuts.third) {
+            neighborList.push_back(make_pair(j1, j2));
+          }
+        }
+      }        
+#endif
     }
-
+      
     // save the local cutoff group positions for the check that is
     // done on each loop:
     saved_CG_positions_.clear();
     for (int i = 0; i < nGroups_; i++)
       saved_CG_positions_.push_back(snap_->cgData.position[i]);
-
+    
     return neighborList;
   }
 } //end namespace OpenMD