--- branches/development/src/parallel/ForceMatrixDecomposition.cpp	2011/07/10 16:05:34	1589
+++ branches/development/src/parallel/ForceMatrixDecomposition.cpp	2012/06/05 17:51:31	1736
@@ -36,7 +36,8 @@
  * [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).             
  * [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).          
  * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
- * [4]  Vardeman & Gezelter, in progress (2009).                        
+ * [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
+ * [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
  */
 #include "parallel/ForceMatrixDecomposition.hpp"
 #include "math/SquareMatrix3.hpp"
@@ -47,17 +48,54 @@ namespace OpenMD {
 using namespace std;
 namespace OpenMD {
 
+  ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
+
+    // In a parallel computation, row and colum scans must visit all
+    // surrounding cells (not just the 14 upper triangular blocks that
+    // are used when the processor can see all pairs)
+#ifdef IS_MPI
+    cellOffsets_.clear();
+    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
+    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
+    cellOffsets_.push_back( Vector3i( 1,-1,-1) );                          
+    cellOffsets_.push_back( Vector3i(-1, 0,-1) );
+    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
+    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
+    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
+    cellOffsets_.push_back( Vector3i( 0, 1,-1) );      
+    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
+    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
+    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
+    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
+    cellOffsets_.push_back( Vector3i(-1, 0, 0) );       
+    cellOffsets_.push_back( Vector3i( 0, 0, 0) );
+    cellOffsets_.push_back( Vector3i( 1, 0, 0) );
+    cellOffsets_.push_back( Vector3i(-1, 1, 0) );
+    cellOffsets_.push_back( Vector3i( 0, 1, 0) );
+    cellOffsets_.push_back( Vector3i( 1, 1, 0) );
+    cellOffsets_.push_back( Vector3i(-1,-1, 1) );
+    cellOffsets_.push_back( Vector3i( 0,-1, 1) );
+    cellOffsets_.push_back( Vector3i( 1,-1, 1) );
+    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
+    cellOffsets_.push_back( Vector3i( 0, 0, 1) );
+    cellOffsets_.push_back( Vector3i( 1, 0, 1) );
+    cellOffsets_.push_back( Vector3i(-1, 1, 1) );
+    cellOffsets_.push_back( Vector3i( 0, 1, 1) );
+    cellOffsets_.push_back( Vector3i( 1, 1, 1) );
+#endif    
+  }
+
+
   /**
    * distributeInitialData is essentially a copy of the older fortran 
    * SimulationSetup
    */
-  
   void ForceMatrixDecomposition::distributeInitialData() {
     snap_ = sman_->getCurrentSnapshot();
     storageLayout_ = sman_->getStorageLayout();
     ff_ = info_->getForceField();
     nLocal_ = snap_->getNumberOfAtoms();
-    
+   
     nGroups_ = info_->getNLocalCutoffGroups();
     // gather the information for atomtype IDs (atids):
     idents = info_->getIdentArray();
@@ -71,31 +109,40 @@ namespace OpenMD {
     PairList* oneTwo = info_->getOneTwoInteractions();
     PairList* oneThree = info_->getOneThreeInteractions();
     PairList* oneFour = info_->getOneFourInteractions();
-
+    
+    if (needVelocities_) 
+      snap_->cgData.setStorageLayout(DataStorage::dslPosition | 
+                                     DataStorage::dslVelocity);
+    else 
+      snap_->cgData.setStorageLayout(DataStorage::dslPosition);
+    
 #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_);
+    MPI::Intracomm row = rowComm.getComm();
+    MPI::Intracomm col = colComm.getComm();
 
-    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_);
+    AtomPlanIntRow = new Plan<int>(row, nLocal_);
+    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
+    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
+    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
+    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
 
-    cgCommIntRow = new Communicator<Row,int>(nGroups_);
-    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
-    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
-    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
+    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
+    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
+    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
+    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
+    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
 
-    nAtomsInRow_ = AtomCommIntRow->getSize();
-    nAtomsInCol_ = AtomCommIntColumn->getSize();
-    nGroupsInRow_ = cgCommIntRow->getSize();
-    nGroupsInCol_ = cgCommIntColumn->getSize();
+    cgPlanIntRow = new Plan<int>(row, nGroups_);
+    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
+    cgPlanIntColumn = new Plan<int>(col, nGroups_);
+    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
 
+    nAtomsInRow_ = AtomPlanIntRow->getSize();
+    nAtomsInCol_ = AtomPlanIntColumn->getSize();
+    nGroupsInRow_ = cgPlanIntRow->getSize();
+    nGroupsInCol_ = cgPlanIntColumn->getSize();
+
     // Modify the data storage objects with the correct layouts and sizes:
     atomRowData.resize(nAtomsInRow_);
     atomRowData.setStorageLayout(storageLayout_);
@@ -104,33 +151,46 @@ namespace OpenMD {
     cgRowData.resize(nGroupsInRow_);
     cgRowData.setStorageLayout(DataStorage::dslPosition);
     cgColData.resize(nGroupsInCol_);
-    cgColData.setStorageLayout(DataStorage::dslPosition);
-        
+    if (needVelocities_)
+      // we only need column velocities if we need them.
+      cgColData.setStorageLayout(DataStorage::dslPosition |
+                                 DataStorage::dslVelocity);
+    else     
+      cgColData.setStorageLayout(DataStorage::dslPosition);
+      
     identsRow.resize(nAtomsInRow_);
     identsCol.resize(nAtomsInCol_);
     
-    AtomCommIntRow->gather(idents, identsRow);
-    AtomCommIntColumn->gather(idents, identsCol);
+    AtomPlanIntRow->gather(idents, identsRow);
+    AtomPlanIntColumn->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]);         
+
+    pot_row.resize(nAtomsInRow_);
+    pot_col.resize(nAtomsInCol_);
+
     AtomRowToGlobal.resize(nAtomsInRow_);
     AtomColToGlobal.resize(nAtomsInCol_);
+    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
+    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
+
     cgRowToGlobal.resize(nGroupsInRow_);
     cgColToGlobal.resize(nGroupsInCol_);
+    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
+    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
+
     massFactorsRow.resize(nAtomsInRow_);
     massFactorsCol.resize(nAtomsInCol_);
-    pot_row.resize(nAtomsInRow_);
-    pot_col.resize(nAtomsInCol_);
+    AtomPlanRealRow->gather(massFactors, massFactorsRow);
+    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
 
-    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
-    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
-    
-    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
-    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
-
-    AtomCommRealRow->gather(massFactors, massFactorsRow);
-    AtomCommRealColumn->gather(massFactors, massFactorsCol);
-
     groupListRow_.clear();
     groupListRow_.resize(nGroupsInRow_);
     for (int i = 0; i < nGroupsInRow_; i++) {
@@ -185,20 +245,7 @@ namespace OpenMD {
       }      
     }
 
-#endif
-
-    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);
-        }
-      }      
-    }
-
+#else
     excludesForAtom.clear();
     excludesForAtom.resize(nLocal_);
     toposForAtom.clear();
@@ -231,7 +278,27 @@ namespace OpenMD {
         }
       }      
     }
-    
+#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);
+        }
+      }      
+    }
+
+
     createGtypeCutoffMap();
 
   }
@@ -239,9 +306,11 @@ namespace OpenMD {
   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(); 
@@ -249,10 +318,10 @@ namespace OpenMD {
       atid = (*at)->getIdent();
       if (userChoseCutoff_) 
         atypeCutoff[atid] = userCutoff_;
-      else 
+      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.
@@ -312,19 +381,16 @@ namespace OpenMD {
     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]) {
+        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) {
@@ -332,7 +398,7 @@ namespace OpenMD {
           gTypeFound = true;
         } 
       }
-      if (!gTypeFound) {
+      if (!gTypeFound) {      
         gTypeCutoffs.push_back( groupCutoff[cg1] );
         groupToGtype[cg1] = gTypeCutoffs.size() - 1;
       }      
@@ -341,10 +407,12 @@ namespace OpenMD {
 
     // Now we find the maximum group cutoff value present in the simulation
 
-    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
+    RealType groupMax = *max_element(gTypeCutoffs.begin(), 
+                                     gTypeCutoffs.end());
 
 #ifdef IS_MPI
-    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
+    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, 
+                              MPI::MAX);
 #endif
     
     RealType tradRcut = groupMax;
@@ -374,13 +442,9 @@ namespace OpenMD {
 
         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_) {
@@ -440,8 +504,10 @@ namespace OpenMD {
          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);
+      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
+           0.0);
+      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
+           0.0);
     }
 
     if (storageLayout_ & DataStorage::dslDensity) {      
@@ -450,8 +516,10 @@ namespace OpenMD {
     }
 
     if (storageLayout_ & DataStorage::dslFunctional) {   
-      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
-      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
+      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
+           0.0);
+      fill(atomColData.functional.begin(), atomColData.functional.end(),
+           0.0);
     }
 
     if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
@@ -468,8 +536,30 @@ namespace OpenMD {
            atomColData.skippedCharge.end(), 0.0);
     }
 
-#else
-    
+    if (storageLayout_ & DataStorage::dslFlucQForce) {      
+      fill(atomRowData.flucQFrc.begin(), 
+           atomRowData.flucQFrc.end(), 0.0);
+      fill(atomColData.flucQFrc.begin(), 
+           atomColData.flucQFrc.end(), 0.0);
+    }
+
+    if (storageLayout_ & DataStorage::dslElectricField) {    
+      fill(atomRowData.electricField.begin(), 
+           atomRowData.electricField.end(), V3Zero);
+      fill(atomColData.electricField.begin(), 
+           atomColData.electricField.end(), V3Zero);
+    }
+
+    if (storageLayout_ & DataStorage::dslFlucQForce) {    
+      fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
+           0.0);
+      fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.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);
@@ -479,20 +569,26 @@ namespace OpenMD {
       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);
     }
-#endif
-    
+
+    if (storageLayout_ & DataStorage::dslElectricField) {      
+      fill(snap_->atomData.electricField.begin(), 
+           snap_->atomData.electricField.end(), V3Zero);
+    }
   }
 
 
@@ -502,32 +598,55 @@ namespace OpenMD {
 #ifdef IS_MPI
     
     // gather up the atomic positions
-    AtomCommVectorRow->gather(snap_->atomData.position, 
+    AtomPlanVectorRow->gather(snap_->atomData.position, 
                               atomRowData.position);
-    AtomCommVectorColumn->gather(snap_->atomData.position, 
+    AtomPlanVectorColumn->gather(snap_->atomData.position, 
                                  atomColData.position);
     
     // gather up the cutoff group positions
-    cgCommVectorRow->gather(snap_->cgData.position, 
+
+    cgPlanVectorRow->gather(snap_->cgData.position, 
                             cgRowData.position);
-    cgCommVectorColumn->gather(snap_->cgData.position, 
+
+    cgPlanVectorColumn->gather(snap_->cgData.position, 
                                cgColData.position);
+
+
+
+    if (needVelocities_) {
+      // gather up the atomic velocities
+      AtomPlanVectorColumn->gather(snap_->atomData.velocity, 
+                                   atomColData.velocity);
+      
+      cgPlanVectorColumn->gather(snap_->cgData.velocity, 
+                                 cgColData.velocity);
+    }
+
     
     // if needed, gather the atomic rotation matrices
     if (storageLayout_ & DataStorage::dslAmat) {
-      AtomCommMatrixRow->gather(snap_->atomData.aMat, 
+      AtomPlanMatrixRow->gather(snap_->atomData.aMat, 
                                 atomRowData.aMat);
-      AtomCommMatrixColumn->gather(snap_->atomData.aMat, 
+      AtomPlanMatrixColumn->gather(snap_->atomData.aMat, 
                                    atomColData.aMat);
     }
     
     // if needed, gather the atomic eletrostatic frames
     if (storageLayout_ & DataStorage::dslElectroFrame) {
-      AtomCommMatrixRow->gather(snap_->atomData.electroFrame, 
+      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame, 
                                 atomRowData.electroFrame);
-      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame, 
+      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame, 
                                    atomColData.electroFrame);
     }
+
+    // if needed, gather the atomic fluctuating charge values
+    if (storageLayout_ & DataStorage::dslFlucQPosition) {
+      AtomPlanRealRow->gather(snap_->atomData.flucQPos, 
+                              atomRowData.flucQPos);
+      AtomPlanRealColumn->gather(snap_->atomData.flucQPos, 
+                                 atomColData.flucQPos);
+    }
+
 #endif      
   }
   
@@ -541,15 +660,27 @@ namespace OpenMD {
     
     if (storageLayout_ & DataStorage::dslDensity) {
       
-      AtomCommRealRow->scatter(atomRowData.density, 
+      AtomPlanRealRow->scatter(atomRowData.density, 
                                snap_->atomData.density);
       
       int n = snap_->atomData.density.size();
       vector<RealType> rho_tmp(n, 0.0);
-      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
+      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
       for (int i = 0; i < n; i++)
         snap_->atomData.density[i] += rho_tmp[i];
     }
+
+    if (storageLayout_ & DataStorage::dslElectricField) {
+      
+      AtomPlanVectorRow->scatter(atomRowData.electricField, 
+                                 snap_->atomData.electricField);
+      
+      int n = snap_->atomData.electricField.size();
+      vector<Vector3d> field_tmp(n, V3Zero);
+      AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp);
+      for (int i = 0; i < n; i++)
+        snap_->atomData.electricField[i] += field_tmp[i];
+    }
 #endif
   }
 
@@ -562,16 +693,16 @@ namespace OpenMD {
     storageLayout_ = sman_->getStorageLayout();
 #ifdef IS_MPI
     if (storageLayout_ & DataStorage::dslFunctional) {
-      AtomCommRealRow->gather(snap_->atomData.functional, 
+      AtomPlanRealRow->gather(snap_->atomData.functional, 
                               atomRowData.functional);
-      AtomCommRealColumn->gather(snap_->atomData.functional, 
+      AtomPlanRealColumn->gather(snap_->atomData.functional, 
                                  atomColData.functional);
     }
     
     if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
-      AtomCommRealRow->gather(snap_->atomData.functionalDerivative, 
+      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative, 
                               atomRowData.functionalDerivative);
-      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative, 
+      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative, 
                                  atomColData.functionalDerivative);
     }
 #endif
@@ -585,29 +716,29 @@ namespace OpenMD {
     int n = snap_->atomData.force.size();
     vector<Vector3d> frc_tmp(n, V3Zero);
     
-    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
+    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
     for (int i = 0; i < n; i++) {
       snap_->atomData.force[i] += frc_tmp[i];
       frc_tmp[i] = 0.0;
     }
     
-    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
-    for (int i = 0; i < n; i++)
+    AtomPlanVectorColumn->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.torque.size();
       vector<Vector3d> trq_tmp(nt, V3Zero);
 
-      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
+      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
       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);
+      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
       for (int i = 0; i < nt; i++)
         snap_->atomData.torque[i] += trq_tmp[i];
     }
@@ -617,17 +748,35 @@ namespace OpenMD {
       int ns = snap_->atomData.skippedCharge.size();
       vector<RealType> skch_tmp(ns, 0.0);
 
-      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
+      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
       for (int i = 0; i < ns; i++) {
-        snap_->atomData.skippedCharge[i] = skch_tmp[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++)
+      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
+      for (int i = 0; i < ns; i++) 
         snap_->atomData.skippedCharge[i] += skch_tmp[i];
+            
     }
     
+    if (storageLayout_ & DataStorage::dslFlucQForce) {
+
+      int nq = snap_->atomData.flucQFrc.size();
+      vector<RealType> fqfrc_tmp(nq, 0.0);
+
+      AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
+      for (int i = 0; i < nq; i++) {
+        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
+        fqfrc_tmp[i] = 0.0;
+      }
+      
+      AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
+      for (int i = 0; i < nq; i++) 
+        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
+            
+    }
+
     nLocal_ = snap_->getNumberOfAtoms();
 
     vector<potVec> pot_temp(nLocal_, 
@@ -635,18 +784,89 @@ namespace OpenMD {
 
     // scatter/gather pot_row into the members of my column
           
-    AtomCommPotRow->scatter(pot_row, pot_temp);
+    AtomPlanPotRow->scatter(pot_row, pot_temp);
 
     for (int ii = 0;  ii < pot_temp.size(); ii++ )
       pairwisePot += pot_temp[ii];
-    
+        
+    if (storageLayout_ & DataStorage::dslParticlePot) {
+      // This is the pairwise contribution to the particle pot.  The
+      // embedding contribution is added in each of the low level
+      // non-bonded routines.  In single processor, this is done in
+      // unpackInteractionData, not in collectData.
+      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
+        for (int i = 0; i < nLocal_; i++) {
+          // factor of two is because the total potential terms are divided
+          // by 2 in parallel due to row/ column scatter       
+          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
+        }
+      }
+    }
+
     fill(pot_temp.begin(), pot_temp.end(), 
          Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
       
-    AtomCommPotColumn->scatter(pot_col, pot_temp);    
+    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
     
     for (int ii = 0;  ii < pot_temp.size(); ii++ )
       pairwisePot += pot_temp[ii];    
+
+    if (storageLayout_ & DataStorage::dslParticlePot) {
+      // This is the pairwise contribution to the particle pot.  The
+      // embedding contribution is added in each of the low level
+      // non-bonded routines.  In single processor, this is done in
+      // unpackInteractionData, not in collectData.
+      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
+        for (int i = 0; i < nLocal_; i++) {
+          // factor of two is because the total potential terms are divided
+          // by 2 in parallel due to row/ column scatter       
+          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
+        }
+      }
+    }
+    
+    if (storageLayout_ & DataStorage::dslParticlePot) {
+      int npp = snap_->atomData.particlePot.size();
+      vector<RealType> ppot_temp(npp, 0.0);
+
+      // This is the direct or embedding contribution to the particle
+      // pot.
+      
+      AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
+      for (int i = 0; i < npp; i++) {
+        snap_->atomData.particlePot[i] += ppot_temp[i];
+      }
+
+      fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
+      
+      AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
+      for (int i = 0; i < npp; i++) {
+        snap_->atomData.particlePot[i] += ppot_temp[i];
+      }
+    }
+
+    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
+      RealType ploc1 = pairwisePot[ii];
+      RealType ploc2 = 0.0;
+      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
+      pairwisePot[ii] = ploc2;
+    }
+
+    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
+      RealType ploc1 = embeddingPot[ii];
+      RealType ploc2 = 0.0;
+      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
+      embeddingPot[ii] = ploc2;
+    }
+    
+    // Here be dragons.
+    MPI::Intracomm col = colComm.getComm();
+
+    col.Allreduce(MPI::IN_PLACE, 
+                  &snap_->frameData.conductiveHeatFlux[0], 3, 
+                  MPI::REALTYPE, MPI::SUM);
+
+
 #endif
 
   }
@@ -691,7 +911,23 @@ namespace OpenMD {
     return d;    
   }
 
+  Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
+#ifdef IS_MPI
+    return cgColData.velocity[cg2];
+#else
+    return snap_->cgData.velocity[cg2];
+#endif
+  }
 
+  Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
+#ifdef IS_MPI
+    return atomColData.velocity[atom2];
+#else
+    return snap_->atomData.velocity[atom2];
+#endif
+  }
+
+
   Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
 
     Vector3d d;
@@ -759,22 +995,27 @@ namespace OpenMD {
    */
   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];
+#else
+    unique_id_1 = AtomLocalToGlobal[atom1];
+    unique_id_2 = AtomLocalToGlobal[atom2];
+#endif   
 
-    // this situation should only arise in MPI simulations
     if (unique_id_1 == unique_id_2) return true;
-    
+
+#ifdef IS_MPI
     // 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; 
+      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
     }
 #endif
+    
     return false;
   }
 
@@ -788,19 +1029,13 @@ namespace OpenMD {
    * field) must still be handled for these pairs.
    */
   bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
-    int unique_id_2;
+
+    // excludesForAtom was constructed to use row/column indices in the MPI
+    // version, and to use local IDs in the non-MPI version:
     
-#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;
+      if ( (*i) == atom2 ) return true;
     }
 
     return false;
@@ -830,10 +1065,10 @@ namespace OpenMD {
     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]) );
     
-    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]);
@@ -874,11 +1109,22 @@ namespace OpenMD {
       idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
     }
 
+    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
+      idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
+      idat.flucQ2 = &(atomColData.flucQPos[atom2]);
+    }
+
 #else
+    
 
-    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]), 
-                             ff_->getAtomType(idents[atom2]) );
+    // cerr << "atoms = " << atom1 << " " << atom2 << "\n";
+    // cerr << "pos1 = " << snap_->atomData.position[atom1] << "\n";
+    // cerr << "pos2 = " << snap_->atomData.position[atom2] << "\n";
 
+    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]);
@@ -918,22 +1164,59 @@ namespace OpenMD {
       idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
       idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
     }
+
+    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
+      idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
+      idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
+    }
+
 #endif
   }
 
   
   void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
 #ifdef IS_MPI
-    pot_row[atom1] += 0.5 *  *(idat.pot);
-    pot_col[atom2] += 0.5 *  *(idat.pot);
+    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
+    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
 
     atomRowData.force[atom1] += *(idat.f1);
     atomColData.force[atom2] -= *(idat.f1);
+
+    if (storageLayout_ & DataStorage::dslFlucQForce) {              
+      atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
+      atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
+    }
+
+    if (storageLayout_ & DataStorage::dslElectricField) {              
+      atomRowData.electricField[atom1] += *(idat.eField1);
+      atomColData.electricField[atom2] += *(idat.eField2);
+    }
+
 #else
     pairwisePot += *(idat.pot);
 
     snap_->atomData.force[atom1] += *(idat.f1);
     snap_->atomData.force[atom2] -= *(idat.f1);
+
+    if (idat.doParticlePot) {
+      // This is the pairwise contribution to the particle pot.  The
+      // embedding contribution is added in each of the low level
+      // non-bonded routines.  In parallel, this calculation is done
+      // in collectData, not in unpackInteractionData.
+      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
+      snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
+    }
+    
+    if (storageLayout_ & DataStorage::dslFlucQForce) {              
+      snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
+      snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
+    }
+
+    if (storageLayout_ & DataStorage::dslElectricField) {              
+      snap_->atomData.electricField[atom1] += *(idat.eField1);
+      snap_->atomData.electricField[atom2] += *(idat.eField2);
+    }
+
 #endif
     
   }
@@ -1015,7 +1298,6 @@ namespace OpenMD {
         // 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];
         
@@ -1040,6 +1322,7 @@ namespace OpenMD {
         // 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];
@@ -1060,11 +1343,12 @@ namespace OpenMD {
         whichCell.z() = nCells_.z() * scaled.z();
         
         // find single index of this cell:
-        cellIndex = Vlinear(whichCell, nCells_);      
+        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++) {
@@ -1077,7 +1361,8 @@ namespace OpenMD {
                  os != cellOffsets_.end(); ++os) {
               
               Vector3i m2v = m1v + (*os);
-              
+             
+
               if (m2v.x() >= nCells_.x()) {
                 m2v.x() = 0;           
               } else if (m2v.x() < 0) {
@@ -1095,7 +1380,7 @@ namespace OpenMD {
               } else if (m2v.z() < 0) {
                 m2v.z() = nCells_.z() - 1; 
               }
-              
+
               int m2 = Vlinear (m2v, nCells_);
               
 #ifdef IS_MPI
@@ -1104,32 +1389,36 @@ namespace OpenMD {
                 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)));
-                    }
-                  }
+                  // In parallel, we need to visit *all* pairs of row
+                  // & column indicies and will divide labor in the
+                  // force evaluation later.
+                  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)) {
+                  // we're in the same cell and the global index of
+                  // the j2 cutoff group is greater than or equal to
+                  // the j1 cutoff group.  Note that Rappaport's code
+                  // has a "less than" conditional here, but that
+                  // deals with atom-by-atom computation.  OpenMD
+                  // allows atoms within a single cutoff group to
+                  // interact with each other.
+
+
+
+                  if (m2 != m1 || (*j2) >= (*j1) ) {
+
                     dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
                     snap_->wrapVector(dr);
                     cuts = getGroupCutoffs( (*j1), (*j2) );
@@ -1148,7 +1437,7 @@ namespace OpenMD {
       // branch to do all cutoff group pairs
 #ifdef IS_MPI
       for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
-        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {       
+        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
           dr = cgColData.position[j2] - cgRowData.position[j1];
           snap_->wrapVector(dr);
           cuts = getGroupCutoffs( j1, j2 );
@@ -1156,18 +1445,20 @@ namespace OpenMD {
             neighborList.push_back(make_pair(j1, j2));
           }
         }
-      }
+      }      
 #else
-      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
-        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
+      // include all groups here.
+      for (int j1 = 0; j1 < nGroups_; j1++) {
+        // include self group interactions j2 == j1
+        for (int j2 = j1; 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
     }