--- branches/development/src/parallel/ForceDecomposition.cpp	2011/01/17 21:34:36	1540
+++ branches/development/src/parallel/ForceMatrixDecomposition.cpp	2011/05/27 16:45:44	1571
@@ -38,140 +38,790 @@
  * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
  * [4]  Vardeman & Gezelter, in progress (2009).                        
  */
-#include "parallel/ForceDecomposition.hpp"
-#include "parallel/Communicator.hpp"
+#include "parallel/ForceMatrixDecomposition.hpp"
 #include "math/SquareMatrix3.hpp"
+#include "nonbonded/NonBondedInteraction.hpp"
+#include "brains/SnapshotManager.hpp"
+#include "brains/PairList.hpp"
 
+using namespace std;
 namespace OpenMD {
 
-  void ForceDecomposition::distributeInitialData() {
+  /**
+   * 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_ = snap_->getNumberOfCutoffGroups();
+
+    // gather the information for atomtype IDs (atids):
+    identsLocal = info_->getIdentArray();
+    AtomLocalToGlobal = info_->getGlobalAtomIndices();
+    cgLocalToGlobal = info_->getGlobalGroupIndices();
+    vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
+    vector<RealType> massFactorsLocal = info_->getMassFactors();
+    PairList excludes = info_->getExcludedInteractions();
+    PairList oneTwo = info_->getOneTwoInteractions();
+    PairList oneThree = info_->getOneThreeInteractions();
+    PairList oneFour = info_->getOneFourInteractions();
+    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
+
 #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_);
 
-    int nAtoms;
-    int nGroups;
+    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
+    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
+    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
+    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
 
-    AtomCommRealI = new Communicator<Row,RealType>(nAtoms);
-    AtomCommVectorI = new Communicator<Row,Vector3d>(nAtoms);
-    AtomCommMatrixI = new Communicator<Row,Mat3x3d>(nAtoms);
+    cgCommIntRow = new Communicator<Row,int>(nGroups_);
+    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
+    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
+    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
 
-    AtomCommRealJ = new Communicator<Column,RealType>(nAtoms);
-    AtomCommVectorJ = new Communicator<Column,Vector3d>(nAtoms);
-    AtomCommMatrixJ = new Communicator<Column,Mat3x3d>(nAtoms);
+    nAtomsInRow_ = AtomCommIntRow->getSize();
+    nAtomsInCol_ = AtomCommIntColumn->getSize();
+    nGroupsInRow_ = cgCommIntRow->getSize();
+    nGroupsInCol_ = cgCommIntColumn->getSize();
 
-    cgCommVectorI = new Communicator<Row,Vector3d>(nGroups);
-    cgCommVectorJ = new Communicator<Column,Vector3d>(nGroups);
-    // more to come
-#endif
-  }
+    // Modify the data storage objects with the correct layouts and sizes:
+    atomRowData.resize(nAtomsInRow_);
+    atomRowData.setStorageLayout(storageLayout_);
+    atomColData.resize(nAtomsInCol_);
+    atomColData.setStorageLayout(storageLayout_);
+    cgRowData.resize(nGroupsInRow_);
+    cgRowData.setStorageLayout(DataStorage::dslPosition);
+    cgColData.resize(nGroupsInCol_);
+    cgColData.setStorageLayout(DataStorage::dslPosition);
     
+    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));
+    
+    identsRow.reserve(nAtomsInRow_);
+    identsCol.reserve(nAtomsInCol_);
+    
+    AtomCommIntRow->gather(identsLocal, identsRow);
+    AtomCommIntColumn->gather(identsLocal, identsCol);
+    
+    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
+    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
+    
+    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
+    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
 
+    AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
+    AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
 
-  void ForceDecomposition::distributeData()  {
+    groupListRow_.clear();
+    groupListRow_.reserve(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_.reserve(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);
+      }      
+    }
+
+    skipsForRowAtom.clear();
+    skipsForRowAtom.reserve(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)) 
+          skipsForRowAtom[i].push_back(j);       
+      }      
+    }
+
+    toposForRowAtom.clear();
+    toposForRowAtom.reserve(nAtomsInRow_);
+    for (int i = 0; i < nAtomsInRow_; i++) {
+      int iglob = AtomRowToGlobal[i];
+      int nTopos = 0;
+      for (int j = 0; j < nAtomsInCol_; j++) {
+        int jglob = AtomColToGlobal[j];        
+        if (oneTwo.hasPair(iglob, jglob)) {
+          toposForRowAtom[i].push_back(j);
+          topoDistRow[i][nTopos] = 1;
+          nTopos++;
+        }
+        if (oneThree.hasPair(iglob, jglob)) {
+          toposForRowAtom[i].push_back(j);
+          topoDistRow[i][nTopos] = 2;
+          nTopos++;
+        }
+        if (oneFour.hasPair(iglob, jglob)) {
+          toposForRowAtom[i].push_back(j);
+          topoDistRow[i][nTopos] = 3;
+          nTopos++;
+        }
+      }      
+    }
+
+#endif
+
+    groupList_.clear();
+    groupList_.reserve(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);
+      }      
+    }
+
+    skipsForLocalAtom.clear();
+    skipsForLocalAtom.reserve(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)) 
+          skipsForLocalAtom[i].push_back(j);       
+      }      
+    }
+
+    toposForLocalAtom.clear();
+    toposForLocalAtom.reserve(nLocal_);
+    for (int i = 0; i < nLocal_; i++) {
+      int iglob = AtomLocalToGlobal[i];
+      int nTopos = 0;
+      for (int j = 0; j < nLocal_; j++) {
+        int jglob = AtomLocalToGlobal[j];        
+        if (oneTwo.hasPair(iglob, jglob)) {
+          toposForLocalAtom[i].push_back(j);
+          topoDistLocal[i][nTopos] = 1;
+          nTopos++;
+        }
+        if (oneThree.hasPair(iglob, jglob)) {
+          toposForLocalAtom[i].push_back(j);
+          topoDistLocal[i][nTopos] = 2;
+          nTopos++;
+        }
+        if (oneFour.hasPair(iglob, jglob)) {
+          toposForLocalAtom[i].push_back(j);
+          topoDistLocal[i][nTopos] = 3;
+          nTopos++;
+        }
+      }      
+    }
+  }
+   
+  void ForceMatrixDecomposition::distributeData()  {
+    snap_ = sman_->getCurrentSnapshot();
+    storageLayout_ = sman_->getStorageLayout();
 #ifdef IS_MPI
-    Snapshot* snap = sman_->getCurrentSnapshot();
     
     // gather up the atomic positions
-    AtomCommVectorI->gather(snap->atomData.position, 
-                            snap->atomIData.position);
-    AtomCommVectorJ->gather(snap->atomData.position, 
-                            snap->atomJData.position);
+    AtomCommVectorRow->gather(snap_->atomData.position, 
+                              atomRowData.position);
+    AtomCommVectorColumn->gather(snap_->atomData.position, 
+                                 atomColData.position);
     
     // gather up the cutoff group positions
-    cgCommVectorI->gather(snap->cgData.position, 
-                          snap->cgIData.position);
-    cgCommVectorJ->gather(snap->cgData.position, 
-                          snap->cgJData.position);
+    cgCommVectorRow->gather(snap_->cgData.position, 
+                            cgRowData.position);
+    cgCommVectorColumn->gather(snap_->cgData.position, 
+                               cgColData.position);
     
     // if needed, gather the atomic rotation matrices
-    if (snap->atomData.getStorageLayout() & DataStorage::dslAmat) {
-      AtomCommMatrixI->gather(snap->atomData.aMat, 
-                              snap->atomIData.aMat);
-      AtomCommMatrixJ->gather(snap->atomData.aMat, 
-                              snap->atomJData.aMat);
+    if (storageLayout_ & DataStorage::dslAmat) {
+      AtomCommMatrixRow->gather(snap_->atomData.aMat, 
+                                atomRowData.aMat);
+      AtomCommMatrixColumn->gather(snap_->atomData.aMat, 
+                                   atomColData.aMat);
     }
     
     // if needed, gather the atomic eletrostatic frames
-    if (snap->atomData.getStorageLayout() & DataStorage::dslElectroFrame) {
-      AtomCommMatrixI->gather(snap->atomData.electroFrame, 
-                              snap->atomIData.electroFrame);
-      AtomCommMatrixJ->gather(snap->atomData.electroFrame, 
-                              snap->atomJData.electroFrame);
+    if (storageLayout_ & DataStorage::dslElectroFrame) {
+      AtomCommMatrixRow->gather(snap_->atomData.electroFrame, 
+                                atomRowData.electroFrame);
+      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame, 
+                                   atomColData.electroFrame);
     }
 #endif      
   }
   
-  void ForceDecomposition::collectIntermediateData() {
+  void ForceMatrixDecomposition::collectIntermediateData() {
+    snap_ = sman_->getCurrentSnapshot();
+    storageLayout_ = sman_->getStorageLayout();
 #ifdef IS_MPI
-    Snapshot* snap = sman_->getCurrentSnapshot();
     
-    if (snap->atomData.getStorageLayout() & DataStorage::dslDensity) {
-      AtomCommRealI->scatter(snap->atomIData.density, 
-                             snap->atomData.density);
-      std::vector<RealType> rho_tmp;
-      int n = snap->getNumberOfAtoms();
-      rho_tmp.reserve( n );
-      AtomCommRealJ->scatter(snap->atomJData.density, rho_tmp);
+    if (storageLayout_ & DataStorage::dslDensity) {
+      
+      AtomCommRealRow->scatter(atomRowData.density, 
+                               snap_->atomData.density);
+      
+      int n = snap_->atomData.density.size();
+      std::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];
+        snap_->atomData.density[i] += rho_tmp[i];
     }
 #endif
   }
   
-  void ForceDecomposition::distributeIntermediateData() {
+  void ForceMatrixDecomposition::distributeIntermediateData() {
+    snap_ = sman_->getCurrentSnapshot();
+    storageLayout_ = sman_->getStorageLayout();
 #ifdef IS_MPI
-    Snapshot* snap = sman_->getCurrentSnapshot();
-    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctional) {
-      AtomCommRealI->gather(snap->atomData.functional, 
-                            snap->atomIData.functional);
-      AtomCommRealJ->gather(snap->atomData.functional, 
-                            snap->atomJData.functional);
+    if (storageLayout_ & DataStorage::dslFunctional) {
+      AtomCommRealRow->gather(snap_->atomData.functional, 
+                              atomRowData.functional);
+      AtomCommRealColumn->gather(snap_->atomData.functional, 
+                                 atomColData.functional);
     }
     
-    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctionalDerivative) {
-      AtomCommRealI->gather(snap->atomData.functionalDerivative, 
-                            snap->atomIData.functionalDerivative);
-      AtomCommRealJ->gather(snap->atomData.functionalDerivative, 
-                            snap->atomJData.functionalDerivative);
+    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
+      AtomCommRealRow->gather(snap_->atomData.functionalDerivative, 
+                              atomRowData.functionalDerivative);
+      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative, 
+                                 atomColData.functionalDerivative);
     }
 #endif
   }
   
   
-  void ForceDecomposition::collectData() {
-#ifdef IS_MPI
-    Snapshot* snap = sman_->getCurrentSnapshot();
-    int n = snap->getNumberOfAtoms();
-
-    std::vector<Vector3d> frc_tmp;
-    frc_tmp.reserve( n );
+  void ForceMatrixDecomposition::collectData() {
+    snap_ = sman_->getCurrentSnapshot();
+    storageLayout_ = sman_->getStorageLayout();
+#ifdef IS_MPI    
+    int n = snap_->atomData.force.size();
+    vector<Vector3d> frc_tmp(n, V3Zero);
     
-    AtomCommVectorI->scatter(snap->atomIData.force, frc_tmp);
-    for (int i = 0; i < n; i++)
-      snap->atomData.force[i] += frc_tmp[i];
+    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
+    for (int i = 0; i < n; i++) {
+      snap_->atomData.force[i] += frc_tmp[i];
+      frc_tmp[i] = 0.0;
+    }
     
-    AtomCommVectorJ->scatter(snap->atomJData.force, frc_tmp);
+    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
     for (int i = 0; i < n; i++)
-      snap->atomData.force[i] += frc_tmp[i];
+      snap_->atomData.force[i] += frc_tmp[i];
     
     
-    if (snap->atomData.getStorageLayout() & DataStorage::dslTorque) {
-      std::vector<Vector3d> trq_tmp;
-      trq_tmp.reserve( n );
+    if (storageLayout_ & DataStorage::dslTorque) {
+
+      int nt = snap_->atomData.force.size();
+      vector<Vector3d> trq_tmp(nt, V3Zero);
+
+      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
+      for (int i = 0; i < n; i++) {
+        snap_->atomData.torque[i] += trq_tmp[i];
+        trq_tmp[i] = 0.0;
+      }
       
-      AtomCommVectorI->scatter(snap->atomIData.torque, trq_tmp);
+      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
       for (int i = 0; i < n; i++)
-        snap->atomData.torque[i] += trq_tmp[i];
-      
-      AtomCommVectorJ->scatter(snap->atomJData.torque, trq_tmp);
-      for (int i = 0; i < n; i++)
-        snap->atomData.torque[i] += trq_tmp[i];
+        snap_->atomData.torque[i] += trq_tmp[i];
     }
     
-    // Still need pot!
+    nLocal_ = snap_->getNumberOfAtoms();
+
+    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES, 
+                                       vector<RealType> (nLocal_, 0.0));
     
+    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];
+      }
+    }
+#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;
+    
+#ifdef IS_MPI
+    d = cgColData.position[cg2] - cgRowData.position[cg1];
+#else
+    d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
+#endif
+    
+    snap_->wrapVector(d);
+    return d;    
+  }
+
+
+  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
+
+    Vector3d d;
+    
+#ifdef IS_MPI
+    d = cgRowData.position[cg1] - atomRowData.position[atom1];
+#else
+    d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
+#endif
+
+    snap_->wrapVector(d);
+    return d;    
+  }
+  
+  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
+    Vector3d d;
+    
+#ifdef IS_MPI
+    d = cgColData.position[cg2] - atomColData.position[atom2];
+#else
+    d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
+#endif
+    
+    snap_->wrapVector(d);
+    return d;    
+  }
+
+  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
+#ifdef IS_MPI
+    return massFactorsRow[atom1];
+#else
+    return massFactorsLocal[atom1];
+#endif
+  }
+
+  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
+#ifdef IS_MPI
+    return massFactorsCol[atom2];
+#else
+    return massFactorsLocal[atom2];
+#endif
+
+  }
+    
+  Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
+    Vector3d d;
+    
+#ifdef IS_MPI
+    d = atomColData.position[atom2] - atomRowData.position[atom1];
+#else
+    d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
+#endif
+
+    snap_->wrapVector(d);
+    return d;    
+  }
+
+  vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
+#ifdef IS_MPI
+    return skipsForRowAtom[atom1];
+#else
+    return skipsForLocalAtom[atom1];
+#endif
+  }
+
+  /**
+   * there are a number of reasons to skip a pair or a particle mostly
+   * we do this to exclude atoms who are involved in short range
+   * interactions (bonds, bends, torsions), but we also 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; 
+    }
+#else
+    // in the normal loop, the atom numbers are unique
+    unique_id_1 = atom1;
+    unique_id_2 = atom2;
+#endif
+    
+#ifdef IS_MPI
+    for (vector<int>::iterator i = skipsForRowAtom[atom1].begin();
+         i != skipsForRowAtom[atom1].end(); ++i) {
+      if ( (*i) == unique_id_2 ) return true;
+    }    
+#else
+    for (vector<int>::iterator i = skipsForLocalAtom[atom1].begin();
+         i != skipsForLocalAtom[atom1].end(); ++i) {
+      if ( (*i) == unique_id_2 ) return true;
+    }    
+#endif
+  }
+
+  int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
+    
+#ifdef IS_MPI
+    for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
+      if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
+    }
+#else
+    for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
+      if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
+    }
+#endif
+
+    // zero is default for unconnected (i.e. normal) pair interactions
+    return 0;
+  }
+
+  void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
+#ifdef IS_MPI
+    atomRowData.force[atom1] += fg;
+#else
+    snap_->atomData.force[atom1] += fg;
+#endif
+  }
+
+  void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
+#ifdef IS_MPI
+    atomColData.force[atom2] += fg;
+#else
+    snap_->atomData.force[atom2] += fg;
+#endif
+  }
+
+    // filling interaction blocks with pointers
+  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
+    InteractionData idat;
+
+#ifdef IS_MPI
+    
+    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]);
+    }
+    
+    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::dslDensity) {
+      idat.rho1 = &(atomRowData.density[atom1]);
+      idat.rho2 = &(atomColData.density[atom2]);
+    }
+
+    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
+      idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
+      idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
+    }
+
+#else
+
+    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]), 
+                             ff_->getAtomType(identsLocal[atom2]) );
+
+    if (storageLayout_ & DataStorage::dslAmat) {
+      idat.A1 = &(snap_->atomData.aMat[atom1]);
+      idat.A2 = &(snap_->atomData.aMat[atom2]);
+    }
+
+    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::dslDensity) {
+      idat.rho1 = &(snap_->atomData.density[atom1]);
+      idat.rho2 = &(snap_->atomData.density[atom2]);
+    }
+
+    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
+      idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
+      idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
+    }
+#endif
+    return idat;
+  }
+
+  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
+
+    InteractionData idat;
+#ifdef IS_MPI
+    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]), 
+                             ff_->getAtomType(identsCol[atom2]) );
+
+    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]);
+    }
+#else
+    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]), 
+                             ff_->getAtomType(identsLocal[atom2]) );
+
+    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]);
+    }
+#endif    
+  }
+
+  /*
+   * buildNeighborList
+   *
+   * first element of pair is row-indexed CutoffGroup
+   * second element of pair is column-indexed CutoffGroup
+   */
+  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
+      
+    vector<pair<int, int> > neighborList;
+#ifdef IS_MPI
+    cellListRow_.clear();
+    cellListCol_.clear();
+#else
+    cellList_.clear();
+#endif
+
+    // dangerous to not do error checking.
+    RealType rCut_;
+ 
+    RealType rList_ = (rCut_ + skinThickness_);
+    RealType rl2 = rList_ * rList_;
+    Snapshot* snap_ = sman_->getCurrentSnapshot();
+    Mat3x3d Hmat = snap_->getHmat();
+    Vector3d Hx = Hmat.getColumn(0);
+    Vector3d Hy = Hmat.getColumn(1);
+    Vector3d Hz = Hmat.getColumn(2);
+
+    nCells_.x() = (int) ( Hx.length() )/ rList_;
+    nCells_.y() = (int) ( Hy.length() )/ rList_;
+    nCells_.z() = (int) ( Hz.length() )/ rList_;
+
+    Mat3x3d invHmat = snap_->getInvHmat();
+    Vector3d rs, scaled, dr;
+    Vector3i whichCell;
+    int cellIndex;
+
+#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();
+
+      // 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]);
+
+      // 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);
+    }
+#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) {
+            
+            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)));
+                  }
+                }
+              }
+            }
+#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)));
+                  }
+                }
+              }
+            }
+#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