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

Comparing trunk/src/brains/SimInfo.cpp (file contents):
Revision 705 by chrisfen, Tue Nov 1 19:14:27 2005 UTC vs.
Revision 1969 by gezelter, Wed Feb 26 14:14:50 2014 UTC

# Line 6 | Line 6
6   * redistribute this software in source and binary code form, provided
7   * that the following conditions are met:
8   *
9 < * 1. Acknowledgement of the program authors must be made in any
10 < *    publication of scientific results based in part on use of the
11 < *    program.  An acceptable form of acknowledgement is citation of
12 < *    the article in which the program was described (Matthew
13 < *    A. Meineke, Charles F. Vardeman II, Teng Lin, Christopher
14 < *    J. Fennell and J. Daniel Gezelter, "OOPSE: An Object-Oriented
15 < *    Parallel Simulation Engine for Molecular Dynamics,"
16 < *    J. Comput. Chem. 26, pp. 252-271 (2005))
17 < *
18 < * 2. Redistributions of source code must retain the above copyright
9 > * 1. Redistributions of source code must retain the above copyright
10   *    notice, this list of conditions and the following disclaimer.
11   *
12 < * 3. Redistributions in binary form must reproduce the above copyright
12 > * 2. Redistributions in binary form must reproduce the above copyright
13   *    notice, this list of conditions and the following disclaimer in the
14   *    documentation and/or other materials provided with the
15   *    distribution.
# Line 37 | Line 28
28   * arising out of the use of or inability to use software, even if the
29   * University of Notre Dame has been advised of the possibility of
30   * such damages.
31 + *
32 + * SUPPORT OPEN SCIENCE!  If you use OpenMD or its source code in your
33 + * research, please cite the appropriate papers when you publish your
34 + * work.  Good starting points are:
35 + *                                                                      
36 + * [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).            
37 + * [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).          
38 + * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008).          
39 + * [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40 + * [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41   */
42  
43   /**
# Line 46 | Line 47
47   * @version 1.0
48   */
49  
50 + #ifdef IS_MPI
51 + #include <mpi.h>
52 + #endif
53   #include <algorithm>
54   #include <set>
55 + #include <map>
56  
57   #include "brains/SimInfo.hpp"
58   #include "math/Vector3.hpp"
59   #include "primitives/Molecule.hpp"
60 < #include "UseTheForce/fCutoffPolicy.h"
56 < #include "UseTheForce/DarkSide/fElectrostaticSummationMethod.h"
57 < #include "UseTheForce/doForces_interface.h"
58 < #include "UseTheForce/DarkSide/electrostatic_interface.h"
59 < #include "UseTheForce/notifyCutoffs_interface.h"
60 > #include "primitives/StuntDouble.hpp"
61   #include "utils/MemoryUtils.hpp"
62   #include "utils/simError.h"
63   #include "selection/SelectionManager.hpp"
64 + #include "io/ForceFieldOptions.hpp"
65 + #include "brains/ForceField.hpp"
66 + #include "nonbonded/SwitchingFunction.hpp"
67  
68 < #ifdef IS_MPI
69 < #include "UseTheForce/mpiComponentPlan.h"
70 < #include "UseTheForce/DarkSide/simParallel_interface.h"
71 < #endif
72 <
73 < namespace oopse {
70 <
71 <  SimInfo::SimInfo(MakeStamps* stamps, std::vector<std::pair<MoleculeStamp*, int> >& molStampPairs,
72 <                   ForceField* ff, Globals* simParams) :
73 <    stamps_(stamps), forceField_(ff), simParams_(simParams),
74 <    ndf_(0), ndfRaw_(0), ndfTrans_(0), nZconstraint_(0),
68 > using namespace std;
69 > namespace OpenMD {
70 >  
71 >  SimInfo::SimInfo(ForceField* ff, Globals* simParams) :
72 >    forceField_(ff), simParams_(simParams),
73 >    ndf_(0), fdf_local(0), ndfRaw_(0), ndfTrans_(0), nZconstraint_(0),
74      nGlobalMols_(0), nGlobalAtoms_(0), nGlobalCutoffGroups_(0),
75 <    nGlobalIntegrableObjects_(0), nGlobalRigidBodies_(0),
76 <    nAtoms_(0), nBonds_(0),  nBends_(0), nTorsions_(0), nRigidBodies_(0),
77 <    nIntegrableObjects_(0),  nCutoffGroups_(0), nConstraints_(0),
78 <    sman_(NULL), fortranInitialized_(false) {
79 <
80 <            
81 <      std::vector<std::pair<MoleculeStamp*, int> >::iterator i;
83 <      MoleculeStamp* molStamp;
84 <      int nMolWithSameStamp;
85 <      int nCutoffAtoms = 0; // number of atoms belong to cutoff groups
86 <      int nGroups = 0;      //total cutoff groups defined in meta-data file
87 <      CutoffGroupStamp* cgStamp;    
88 <      RigidBodyStamp* rbStamp;
89 <      int nRigidAtoms = 0;
75 >    nGlobalIntegrableObjects_(0), nGlobalRigidBodies_(0),
76 >    nGlobalFluctuatingCharges_(0), nGlobalBonds_(0), nGlobalBends_(0),
77 >    nGlobalTorsions_(0), nGlobalInversions_(0), nAtoms_(0), nBonds_(0),
78 >    nBends_(0), nTorsions_(0), nInversions_(0), nRigidBodies_(0),
79 >    nIntegrableObjects_(0), nCutoffGroups_(0), nConstraints_(0),
80 >    nFluctuatingCharges_(0), sman_(NULL), topologyDone_(false),
81 >    calcBoxDipole_(false), useAtomicVirial_(true) {    
82      
83 <      for (i = molStampPairs.begin(); i !=molStampPairs.end(); ++i) {
84 <        molStamp = i->first;
85 <        nMolWithSameStamp = i->second;
86 <        
87 <        addMoleculeStamp(molStamp, nMolWithSameStamp);
88 <
89 <        //calculate atoms in molecules
90 <        nGlobalAtoms_ += molStamp->getNAtoms() *nMolWithSameStamp;  
91 <
92 <
93 <        //calculate atoms in cutoff groups
94 <        int nAtomsInGroups = 0;
95 <        int nCutoffGroupsInStamp = molStamp->getNCutoffGroups();
96 <        
97 <        for (int j=0; j < nCutoffGroupsInStamp; j++) {
98 <          cgStamp = molStamp->getCutoffGroup(j);
99 <          nAtomsInGroups += cgStamp->getNMembers();
100 <        }
109 <
110 <        nGroups += nCutoffGroupsInStamp * nMolWithSameStamp;
111 <
112 <        nCutoffAtoms += nAtomsInGroups * nMolWithSameStamp;            
113 <
114 <        //calculate atoms in rigid bodies
115 <        int nAtomsInRigidBodies = 0;
116 <        int nRigidBodiesInStamp = molStamp->getNRigidBodies();
117 <        
118 <        for (int j=0; j < nRigidBodiesInStamp; j++) {
119 <          rbStamp = molStamp->getRigidBody(j);
120 <          nAtomsInRigidBodies += rbStamp->getNMembers();
121 <        }
122 <
123 <        nGlobalRigidBodies_ += nRigidBodiesInStamp * nMolWithSameStamp;
124 <        nRigidAtoms += nAtomsInRigidBodies * nMolWithSameStamp;            
125 <        
83 >    MoleculeStamp* molStamp;
84 >    int nMolWithSameStamp;
85 >    int nCutoffAtoms = 0; // number of atoms belong to cutoff groups
86 >    int nGroups = 0;       //total cutoff groups defined in meta-data file
87 >    CutoffGroupStamp* cgStamp;    
88 >    RigidBodyStamp* rbStamp;
89 >    int nRigidAtoms = 0;
90 >    
91 >    vector<Component*> components = simParams->getComponents();
92 >    
93 >    for (vector<Component*>::iterator i = components.begin();
94 >         i !=components.end(); ++i) {
95 >      molStamp = (*i)->getMoleculeStamp();
96 >      if ( (*i)->haveRegion() ) {        
97 >        molStamp->setRegion( (*i)->getRegion() );
98 >      } else {
99 >        // set the region to a disallowed value:
100 >        molStamp->setRegion( -1 );
101        }
102  
103 <      //every free atom (atom does not belong to cutoff groups) is a cutoff
104 <      //group therefore the total number of cutoff groups in the system is
105 <      //equal to the total number of atoms minus number of atoms belong to
106 <      //cutoff group defined in meta-data file plus the number of cutoff
107 <      //groups defined in meta-data file
108 <      nGlobalCutoffGroups_ = nGlobalAtoms_ - nCutoffAtoms + nGroups;
109 <
110 <      //every free atom (atom does not belong to rigid bodies) is an
111 <      //integrable object therefore the total number of integrable objects
112 <      //in the system is equal to the total number of atoms minus number of
113 <      //atoms belong to rigid body defined in meta-data file plus the number
114 <      //of rigid bodies defined in meta-data file
115 <      nGlobalIntegrableObjects_ = nGlobalAtoms_ - nRigidAtoms
116 <                                                + nGlobalRigidBodies_;
117 <  
118 <      nGlobalMols_ = molStampIds_.size();
119 <
120 < #ifdef IS_MPI    
121 <      molToProcMap_.resize(nGlobalMols_);
122 < #endif
123 <
103 >      nMolWithSameStamp = (*i)->getNMol();
104 >      
105 >      addMoleculeStamp(molStamp, nMolWithSameStamp);
106 >      
107 >      //calculate atoms in molecules
108 >      nGlobalAtoms_ += molStamp->getNAtoms() * nMolWithSameStamp;
109 >      nGlobalBonds_ += molStamp->getNBonds() * nMolWithSameStamp;
110 >      nGlobalBends_ += molStamp->getNBends() * nMolWithSameStamp;
111 >      nGlobalTorsions_ += molStamp->getNTorsions() * nMolWithSameStamp;
112 >      nGlobalInversions_ += molStamp->getNInversions() * nMolWithSameStamp;
113 >      
114 >      //calculate atoms in cutoff groups
115 >      int nAtomsInGroups = 0;
116 >      int nCutoffGroupsInStamp = molStamp->getNCutoffGroups();
117 >      
118 >      for (int j=0; j < nCutoffGroupsInStamp; j++) {
119 >        cgStamp = molStamp->getCutoffGroupStamp(j);
120 >        nAtomsInGroups += cgStamp->getNMembers();
121 >      }
122 >      
123 >      nGroups += nCutoffGroupsInStamp * nMolWithSameStamp;
124 >      
125 >      nCutoffAtoms += nAtomsInGroups * nMolWithSameStamp;            
126 >      
127 >      //calculate atoms in rigid bodies
128 >      int nAtomsInRigidBodies = 0;
129 >      int nRigidBodiesInStamp = molStamp->getNRigidBodies();
130 >      
131 >      for (int j=0; j < nRigidBodiesInStamp; j++) {
132 >        rbStamp = molStamp->getRigidBodyStamp(j);
133 >        nAtomsInRigidBodies += rbStamp->getNMembers();
134 >      }
135 >      
136 >      nGlobalRigidBodies_ += nRigidBodiesInStamp * nMolWithSameStamp;
137 >      nRigidAtoms += nAtomsInRigidBodies * nMolWithSameStamp;            
138 >      
139      }
140 +    
141 +    //every free atom (atom does not belong to cutoff groups) is a cutoff
142 +    //group therefore the total number of cutoff groups in the system is
143 +    //equal to the total number of atoms minus number of atoms belong to
144 +    //cutoff group defined in meta-data file plus the number of cutoff
145 +    //groups defined in meta-data file
146  
147 +    nGlobalCutoffGroups_ = nGlobalAtoms_ - nCutoffAtoms + nGroups;
148 +    
149 +    //every free atom (atom does not belong to rigid bodies) is an
150 +    //integrable object therefore the total number of integrable objects
151 +    //in the system is equal to the total number of atoms minus number of
152 +    //atoms belong to rigid body defined in meta-data file plus the number
153 +    //of rigid bodies defined in meta-data file
154 +    nGlobalIntegrableObjects_ = nGlobalAtoms_ - nRigidAtoms
155 +      + nGlobalRigidBodies_;
156 +    
157 +    nGlobalMols_ = molStampIds_.size();
158 +    molToProcMap_.resize(nGlobalMols_);
159 +  }
160 +  
161    SimInfo::~SimInfo() {
162 <    std::map<int, Molecule*>::iterator i;
162 >    map<int, Molecule*>::iterator i;
163      for (i = molecules_.begin(); i != molecules_.end(); ++i) {
164        delete i->second;
165      }
166      molecules_.clear();
167        
158    delete stamps_;
168      delete sman_;
169      delete simParams_;
170      delete forceField_;
171    }
172  
164  int SimInfo::getNGlobalConstraints() {
165    int nGlobalConstraints;
166 #ifdef IS_MPI
167    MPI_Allreduce(&nConstraints_, &nGlobalConstraints, 1, MPI_INT, MPI_SUM,
168                  MPI_COMM_WORLD);    
169 #else
170    nGlobalConstraints =  nConstraints_;
171 #endif
172    return nGlobalConstraints;
173  }
173  
174    bool SimInfo::addMolecule(Molecule* mol) {
175      MoleculeIterator i;
176 <
176 >    
177      i = molecules_.find(mol->getGlobalIndex());
178      if (i == molecules_.end() ) {
179 <
180 <      molecules_.insert(std::make_pair(mol->getGlobalIndex(), mol));
181 <        
179 >      
180 >      molecules_.insert(make_pair(mol->getGlobalIndex(), mol));
181 >      
182        nAtoms_ += mol->getNAtoms();
183        nBonds_ += mol->getNBonds();
184        nBends_ += mol->getNBends();
185        nTorsions_ += mol->getNTorsions();
186 +      nInversions_ += mol->getNInversions();
187        nRigidBodies_ += mol->getNRigidBodies();
188        nIntegrableObjects_ += mol->getNIntegrableObjects();
189        nCutoffGroups_ += mol->getNCutoffGroups();
190        nConstraints_ += mol->getNConstraintPairs();
191 <
192 <      addExcludePairs(mol);
193 <        
191 >      
192 >      addInteractionPairs(mol);
193 >      
194        return true;
195      } else {
196        return false;
197      }
198    }
199 <
199 >  
200    bool SimInfo::removeMolecule(Molecule* mol) {
201      MoleculeIterator i;
202      i = molecules_.find(mol->getGlobalIndex());
# Line 209 | Line 209 | namespace oopse {
209        nBonds_ -= mol->getNBonds();
210        nBends_ -= mol->getNBends();
211        nTorsions_ -= mol->getNTorsions();
212 +      nInversions_ -= mol->getNInversions();
213        nRigidBodies_ -= mol->getNRigidBodies();
214        nIntegrableObjects_ -= mol->getNIntegrableObjects();
215        nCutoffGroups_ -= mol->getNCutoffGroups();
216        nConstraints_ -= mol->getNConstraintPairs();
217  
218 <      removeExcludePairs(mol);
218 >      removeInteractionPairs(mol);
219        molecules_.erase(mol->getGlobalIndex());
220  
221        delete mol;
# Line 223 | Line 224 | namespace oopse {
224      } else {
225        return false;
226      }
226
227
227    }    
228  
229          
# Line 240 | Line 239 | namespace oopse {
239  
240  
241    void SimInfo::calcNdf() {
242 <    int ndf_local;
242 >    int ndf_local, nfq_local;
243      MoleculeIterator i;
244 <    std::vector<StuntDouble*>::iterator j;
244 >    vector<StuntDouble*>::iterator j;
245 >    vector<Atom*>::iterator k;
246 >
247      Molecule* mol;
248 <    StuntDouble* integrableObject;
248 >    StuntDouble* sd;
249 >    Atom* atom;
250  
251      ndf_local = 0;
252 +    nfq_local = 0;
253      
254      for (mol = beginMolecule(i); mol != NULL; mol = nextMolecule(i)) {
252      for (integrableObject = mol->beginIntegrableObject(j); integrableObject != NULL;
253           integrableObject = mol->nextIntegrableObject(j)) {
255  
256 +      for (sd = mol->beginIntegrableObject(j); sd != NULL;
257 +           sd = mol->nextIntegrableObject(j)) {
258 +
259          ndf_local += 3;
260  
261 <        if (integrableObject->isDirectional()) {
262 <          if (integrableObject->isLinear()) {
261 >        if (sd->isDirectional()) {
262 >          if (sd->isLinear()) {
263              ndf_local += 2;
264            } else {
265              ndf_local += 3;
266            }
267          }
268 <            
269 <      }//end for (integrableObject)
270 <    }// end for (mol)
268 >      }
269 >
270 >      for (atom = mol->beginFluctuatingCharge(k); atom != NULL;
271 >           atom = mol->nextFluctuatingCharge(k)) {
272 >        if (atom->isFluctuatingCharge()) {
273 >          nfq_local++;
274 >        }
275 >      }
276 >    }
277      
278 +    ndfLocal_ = ndf_local;
279 +
280      // n_constraints is local, so subtract them on each processor
281      ndf_local -= nConstraints_;
282  
283   #ifdef IS_MPI
284 <    MPI_Allreduce(&ndf_local,&ndf_,1,MPI_INT,MPI_SUM, MPI_COMM_WORLD);
284 >    MPI_Allreduce(&ndf_local, &ndf_, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
285 >    MPI_Allreduce(&nfq_local, &nGlobalFluctuatingCharges_, 1,
286 >      MPI_INT, MPI_SUM, MPI_COMM_WORLD);
287 >    // MPI::COMM_WORLD.Allreduce(&ndf_local, &ndf_, 1, MPI::INT,MPI::SUM);
288 >    // MPI::COMM_WORLD.Allreduce(&nfq_local, &nGlobalFluctuatingCharges_, 1,
289 >    //                           MPI::INT, MPI::SUM);
290   #else
291      ndf_ = ndf_local;
292 +    nGlobalFluctuatingCharges_ = nfq_local;
293   #endif
294  
295      // nZconstraints_ is global, as are the 3 COM translations for the
# Line 280 | Line 298 | namespace oopse {
298  
299    }
300  
301 +  int SimInfo::getFdf() {
302 + #ifdef IS_MPI
303 +    MPI_Allreduce(&fdf_local, &fdf_, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
304 +    // MPI::COMM_WORLD.Allreduce(&fdf_local, &fdf_, 1, MPI::INT, MPI::SUM);
305 + #else
306 +    fdf_ = fdf_local;
307 + #endif
308 +    return fdf_;
309 +  }
310 +  
311 +  unsigned int SimInfo::getNLocalCutoffGroups(){
312 +    int nLocalCutoffAtoms = 0;
313 +    Molecule* mol;
314 +    MoleculeIterator mi;
315 +    CutoffGroup* cg;
316 +    Molecule::CutoffGroupIterator ci;
317 +    
318 +    for (mol = beginMolecule(mi); mol != NULL; mol  = nextMolecule(mi)) {
319 +      
320 +      for (cg = mol->beginCutoffGroup(ci); cg != NULL;
321 +           cg = mol->nextCutoffGroup(ci)) {
322 +        nLocalCutoffAtoms += cg->getNumAtom();
323 +        
324 +      }        
325 +    }
326 +    
327 +    return nAtoms_ - nLocalCutoffAtoms + nCutoffGroups_;
328 +  }
329 +    
330    void SimInfo::calcNdfRaw() {
331      int ndfRaw_local;
332  
333      MoleculeIterator i;
334 <    std::vector<StuntDouble*>::iterator j;
334 >    vector<StuntDouble*>::iterator j;
335      Molecule* mol;
336 <    StuntDouble* integrableObject;
336 >    StuntDouble* sd;
337  
338      // Raw degrees of freedom that we have to set
339      ndfRaw_local = 0;
340      
341      for (mol = beginMolecule(i); mol != NULL; mol = nextMolecule(i)) {
295      for (integrableObject = mol->beginIntegrableObject(j); integrableObject != NULL;
296           integrableObject = mol->nextIntegrableObject(j)) {
342  
343 +      for (sd = mol->beginIntegrableObject(j); sd != NULL;
344 +           sd = mol->nextIntegrableObject(j)) {
345 +
346          ndfRaw_local += 3;
347  
348 <        if (integrableObject->isDirectional()) {
349 <          if (integrableObject->isLinear()) {
348 >        if (sd->isDirectional()) {
349 >          if (sd->isLinear()) {
350              ndfRaw_local += 2;
351            } else {
352              ndfRaw_local += 3;
# Line 309 | Line 357 | namespace oopse {
357      }
358      
359   #ifdef IS_MPI
360 <    MPI_Allreduce(&ndfRaw_local,&ndfRaw_,1,MPI_INT,MPI_SUM, MPI_COMM_WORLD);
360 >    MPI_Allreduce(&ndfRaw_local, &ndfRaw_, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
361 >    // MPI::COMM_WORLD.Allreduce(&ndfRaw_local, &ndfRaw_, 1, MPI::INT, MPI::SUM);
362   #else
363      ndfRaw_ = ndfRaw_local;
364   #endif
# Line 322 | Line 371 | namespace oopse {
371  
372  
373   #ifdef IS_MPI
374 <    MPI_Allreduce(&ndfTrans_local,&ndfTrans_,1,MPI_INT,MPI_SUM, MPI_COMM_WORLD);
374 >    MPI_Allreduce(&ndfTrans_local, &ndfTrans_, 1,
375 >      MPI_INT, MPI_SUM, MPI_COMM_WORLD);
376 >    // MPI::COMM_WORLD.Allreduce(&ndfTrans_local, &ndfTrans_, 1,
377 >    //                           MPI::INT, MPI::SUM);
378   #else
379      ndfTrans_ = ndfTrans_local;
380   #endif
# Line 331 | Line 383 | namespace oopse {
383  
384    }
385  
386 <  void SimInfo::addExcludePairs(Molecule* mol) {
387 <    std::vector<Bond*>::iterator bondIter;
388 <    std::vector<Bend*>::iterator bendIter;
389 <    std::vector<Torsion*>::iterator torsionIter;
386 >  void SimInfo::addInteractionPairs(Molecule* mol) {
387 >    ForceFieldOptions& options_ = forceField_->getForceFieldOptions();
388 >    vector<Bond*>::iterator bondIter;
389 >    vector<Bend*>::iterator bendIter;
390 >    vector<Torsion*>::iterator torsionIter;
391 >    vector<Inversion*>::iterator inversionIter;
392      Bond* bond;
393      Bend* bend;
394      Torsion* torsion;
395 +    Inversion* inversion;
396      int a;
397      int b;
398      int c;
399      int d;
400 +
401 +    // atomGroups can be used to add special interaction maps between
402 +    // groups of atoms that are in two separate rigid bodies.
403 +    // However, most site-site interactions between two rigid bodies
404 +    // are probably not special, just the ones between the physically
405 +    // bonded atoms.  Interactions *within* a single rigid body should
406 +    // always be excluded.  These are done at the bottom of this
407 +    // function.
408 +
409 +    map<int, set<int> > atomGroups;
410 +    Molecule::RigidBodyIterator rbIter;
411 +    RigidBody* rb;
412 +    Molecule::IntegrableObjectIterator ii;
413 +    StuntDouble* sd;
414      
415 <    for (bond= mol->beginBond(bondIter); bond != NULL; bond = mol->nextBond(bondIter)) {
415 >    for (sd = mol->beginIntegrableObject(ii); sd != NULL;
416 >         sd = mol->nextIntegrableObject(ii)) {
417 >      
418 >      if (sd->isRigidBody()) {
419 >        rb = static_cast<RigidBody*>(sd);
420 >        vector<Atom*> atoms = rb->getAtoms();
421 >        set<int> rigidAtoms;
422 >        for (int i = 0; i < static_cast<int>(atoms.size()); ++i) {
423 >          rigidAtoms.insert(atoms[i]->getGlobalIndex());
424 >        }
425 >        for (int i = 0; i < static_cast<int>(atoms.size()); ++i) {
426 >          atomGroups.insert(map<int, set<int> >::value_type(atoms[i]->getGlobalIndex(), rigidAtoms));
427 >        }      
428 >      } else {
429 >        set<int> oneAtomSet;
430 >        oneAtomSet.insert(sd->getGlobalIndex());
431 >        atomGroups.insert(map<int, set<int> >::value_type(sd->getGlobalIndex(), oneAtomSet));        
432 >      }
433 >    }  
434 >
435 >          
436 >    for (bond= mol->beginBond(bondIter); bond != NULL;
437 >         bond = mol->nextBond(bondIter)) {
438 >
439        a = bond->getAtomA()->getGlobalIndex();
440 <      b = bond->getAtomB()->getGlobalIndex();        
441 <      exclude_.addPair(a, b);
440 >      b = bond->getAtomB()->getGlobalIndex();  
441 >
442 >      if (options_.havevdw12scale() || options_.haveelectrostatic12scale()) {
443 >        oneTwoInteractions_.addPair(a, b);
444 >      } else {
445 >        excludedInteractions_.addPair(a, b);
446 >      }
447      }
448  
449 <    for (bend= mol->beginBend(bendIter); bend != NULL; bend = mol->nextBend(bendIter)) {
449 >    for (bend= mol->beginBend(bendIter); bend != NULL;
450 >         bend = mol->nextBend(bendIter)) {
451 >
452        a = bend->getAtomA()->getGlobalIndex();
453        b = bend->getAtomB()->getGlobalIndex();        
454        c = bend->getAtomC()->getGlobalIndex();
455 +      
456 +      if (options_.havevdw12scale() || options_.haveelectrostatic12scale()) {
457 +        oneTwoInteractions_.addPair(a, b);      
458 +        oneTwoInteractions_.addPair(b, c);
459 +      } else {
460 +        excludedInteractions_.addPair(a, b);
461 +        excludedInteractions_.addPair(b, c);
462 +      }
463  
464 <      exclude_.addPair(a, b);
465 <      exclude_.addPair(a, c);
466 <      exclude_.addPair(b, c);        
464 >      if (options_.havevdw13scale() || options_.haveelectrostatic13scale()) {
465 >        oneThreeInteractions_.addPair(a, c);      
466 >      } else {
467 >        excludedInteractions_.addPair(a, c);
468 >      }
469      }
470  
471 <    for (torsion= mol->beginTorsion(torsionIter); torsion != NULL; torsion = mol->nextTorsion(torsionIter)) {
471 >    for (torsion= mol->beginTorsion(torsionIter); torsion != NULL;
472 >         torsion = mol->nextTorsion(torsionIter)) {
473 >
474        a = torsion->getAtomA()->getGlobalIndex();
475        b = torsion->getAtomB()->getGlobalIndex();        
476        c = torsion->getAtomC()->getGlobalIndex();        
477 <      d = torsion->getAtomD()->getGlobalIndex();        
477 >      d = torsion->getAtomD()->getGlobalIndex();      
478  
479 <      exclude_.addPair(a, b);
480 <      exclude_.addPair(a, c);
481 <      exclude_.addPair(a, d);
482 <      exclude_.addPair(b, c);
483 <      exclude_.addPair(b, d);
484 <      exclude_.addPair(c, d);        
479 >      if (options_.havevdw12scale() || options_.haveelectrostatic12scale()) {
480 >        oneTwoInteractions_.addPair(a, b);      
481 >        oneTwoInteractions_.addPair(b, c);
482 >        oneTwoInteractions_.addPair(c, d);
483 >      } else {
484 >        excludedInteractions_.addPair(a, b);
485 >        excludedInteractions_.addPair(b, c);
486 >        excludedInteractions_.addPair(c, d);
487 >      }
488 >
489 >      if (options_.havevdw13scale() || options_.haveelectrostatic13scale()) {
490 >        oneThreeInteractions_.addPair(a, c);      
491 >        oneThreeInteractions_.addPair(b, d);      
492 >      } else {
493 >        excludedInteractions_.addPair(a, c);
494 >        excludedInteractions_.addPair(b, d);
495 >      }
496 >
497 >      if (options_.havevdw14scale() || options_.haveelectrostatic14scale()) {
498 >        oneFourInteractions_.addPair(a, d);      
499 >      } else {
500 >        excludedInteractions_.addPair(a, d);
501 >      }
502      }
503  
504 <    Molecule::RigidBodyIterator rbIter;
505 <    RigidBody* rb;
506 <    for (rb = mol->beginRigidBody(rbIter); rb != NULL; rb = mol->nextRigidBody(rbIter)) {
507 <      std::vector<Atom*> atoms = rb->getAtoms();
508 <      for (int i = 0; i < atoms.size() -1 ; ++i) {
509 <        for (int j = i + 1; j < atoms.size(); ++j) {
504 >    for (inversion= mol->beginInversion(inversionIter); inversion != NULL;
505 >         inversion = mol->nextInversion(inversionIter)) {
506 >
507 >      a = inversion->getAtomA()->getGlobalIndex();
508 >      b = inversion->getAtomB()->getGlobalIndex();        
509 >      c = inversion->getAtomC()->getGlobalIndex();        
510 >      d = inversion->getAtomD()->getGlobalIndex();        
511 >
512 >      if (options_.havevdw12scale() || options_.haveelectrostatic12scale()) {
513 >        oneTwoInteractions_.addPair(a, b);      
514 >        oneTwoInteractions_.addPair(a, c);
515 >        oneTwoInteractions_.addPair(a, d);
516 >      } else {
517 >        excludedInteractions_.addPair(a, b);
518 >        excludedInteractions_.addPair(a, c);
519 >        excludedInteractions_.addPair(a, d);
520 >      }
521 >
522 >      if (options_.havevdw13scale() || options_.haveelectrostatic13scale()) {
523 >        oneThreeInteractions_.addPair(b, c);    
524 >        oneThreeInteractions_.addPair(b, d);    
525 >        oneThreeInteractions_.addPair(c, d);      
526 >      } else {
527 >        excludedInteractions_.addPair(b, c);
528 >        excludedInteractions_.addPair(b, d);
529 >        excludedInteractions_.addPair(c, d);
530 >      }
531 >    }
532 >
533 >    for (rb = mol->beginRigidBody(rbIter); rb != NULL;
534 >         rb = mol->nextRigidBody(rbIter)) {
535 >      vector<Atom*> atoms = rb->getAtoms();
536 >      for (int i = 0; i < static_cast<int>(atoms.size()) -1 ; ++i) {
537 >        for (int j = i + 1; j < static_cast<int>(atoms.size()); ++j) {
538            a = atoms[i]->getGlobalIndex();
539            b = atoms[j]->getGlobalIndex();
540 <          exclude_.addPair(a, b);
540 >          excludedInteractions_.addPair(a, b);
541          }
542        }
543      }        
544  
545    }
546  
547 <  void SimInfo::removeExcludePairs(Molecule* mol) {
548 <    std::vector<Bond*>::iterator bondIter;
549 <    std::vector<Bend*>::iterator bendIter;
550 <    std::vector<Torsion*>::iterator torsionIter;
547 >  void SimInfo::removeInteractionPairs(Molecule* mol) {
548 >    ForceFieldOptions& options_ = forceField_->getForceFieldOptions();
549 >    vector<Bond*>::iterator bondIter;
550 >    vector<Bend*>::iterator bendIter;
551 >    vector<Torsion*>::iterator torsionIter;
552 >    vector<Inversion*>::iterator inversionIter;
553      Bond* bond;
554      Bend* bend;
555      Torsion* torsion;
556 +    Inversion* inversion;
557      int a;
558      int b;
559      int c;
560      int d;
561 +
562 +    map<int, set<int> > atomGroups;
563 +    Molecule::RigidBodyIterator rbIter;
564 +    RigidBody* rb;
565 +    Molecule::IntegrableObjectIterator ii;
566 +    StuntDouble* sd;
567      
568 <    for (bond= mol->beginBond(bondIter); bond != NULL; bond = mol->nextBond(bondIter)) {
568 >    for (sd = mol->beginIntegrableObject(ii); sd != NULL;
569 >         sd = mol->nextIntegrableObject(ii)) {
570 >      
571 >      if (sd->isRigidBody()) {
572 >        rb = static_cast<RigidBody*>(sd);
573 >        vector<Atom*> atoms = rb->getAtoms();
574 >        set<int> rigidAtoms;
575 >        for (int i = 0; i < static_cast<int>(atoms.size()); ++i) {
576 >          rigidAtoms.insert(atoms[i]->getGlobalIndex());
577 >        }
578 >        for (int i = 0; i < static_cast<int>(atoms.size()); ++i) {
579 >          atomGroups.insert(map<int, set<int> >::value_type(atoms[i]->getGlobalIndex(), rigidAtoms));
580 >        }      
581 >      } else {
582 >        set<int> oneAtomSet;
583 >        oneAtomSet.insert(sd->getGlobalIndex());
584 >        atomGroups.insert(map<int, set<int> >::value_type(sd->getGlobalIndex(), oneAtomSet));        
585 >      }
586 >    }  
587 >
588 >    for (bond= mol->beginBond(bondIter); bond != NULL;
589 >         bond = mol->nextBond(bondIter)) {
590 >      
591        a = bond->getAtomA()->getGlobalIndex();
592 <      b = bond->getAtomB()->getGlobalIndex();        
593 <      exclude_.removePair(a, b);
592 >      b = bond->getAtomB()->getGlobalIndex();  
593 >    
594 >      if (options_.havevdw12scale() || options_.haveelectrostatic12scale()) {
595 >        oneTwoInteractions_.removePair(a, b);
596 >      } else {
597 >        excludedInteractions_.removePair(a, b);
598 >      }
599      }
600  
601 <    for (bend= mol->beginBend(bendIter); bend != NULL; bend = mol->nextBend(bendIter)) {
601 >    for (bend= mol->beginBend(bendIter); bend != NULL;
602 >         bend = mol->nextBend(bendIter)) {
603 >
604        a = bend->getAtomA()->getGlobalIndex();
605        b = bend->getAtomB()->getGlobalIndex();        
606        c = bend->getAtomC()->getGlobalIndex();
607 +      
608 +      if (options_.havevdw12scale() || options_.haveelectrostatic12scale()) {
609 +        oneTwoInteractions_.removePair(a, b);      
610 +        oneTwoInteractions_.removePair(b, c);
611 +      } else {
612 +        excludedInteractions_.removePair(a, b);
613 +        excludedInteractions_.removePair(b, c);
614 +      }
615  
616 <      exclude_.removePair(a, b);
617 <      exclude_.removePair(a, c);
618 <      exclude_.removePair(b, c);        
616 >      if (options_.havevdw13scale() || options_.haveelectrostatic13scale()) {
617 >        oneThreeInteractions_.removePair(a, c);      
618 >      } else {
619 >        excludedInteractions_.removePair(a, c);
620 >      }
621      }
622  
623 <    for (torsion= mol->beginTorsion(torsionIter); torsion != NULL; torsion = mol->nextTorsion(torsionIter)) {
623 >    for (torsion= mol->beginTorsion(torsionIter); torsion != NULL;
624 >         torsion = mol->nextTorsion(torsionIter)) {
625 >
626        a = torsion->getAtomA()->getGlobalIndex();
627        b = torsion->getAtomB()->getGlobalIndex();        
628        c = torsion->getAtomC()->getGlobalIndex();        
629 <      d = torsion->getAtomD()->getGlobalIndex();        
629 >      d = torsion->getAtomD()->getGlobalIndex();      
630 >  
631 >      if (options_.havevdw12scale() || options_.haveelectrostatic12scale()) {
632 >        oneTwoInteractions_.removePair(a, b);      
633 >        oneTwoInteractions_.removePair(b, c);
634 >        oneTwoInteractions_.removePair(c, d);
635 >      } else {
636 >        excludedInteractions_.removePair(a, b);
637 >        excludedInteractions_.removePair(b, c);
638 >        excludedInteractions_.removePair(c, d);
639 >      }
640  
641 <      exclude_.removePair(a, b);
642 <      exclude_.removePair(a, c);
643 <      exclude_.removePair(a, d);
644 <      exclude_.removePair(b, c);
645 <      exclude_.removePair(b, d);
646 <      exclude_.removePair(c, d);        
641 >      if (options_.havevdw13scale() || options_.haveelectrostatic13scale()) {
642 >        oneThreeInteractions_.removePair(a, c);      
643 >        oneThreeInteractions_.removePair(b, d);      
644 >      } else {
645 >        excludedInteractions_.removePair(a, c);
646 >        excludedInteractions_.removePair(b, d);
647 >      }
648 >
649 >      if (options_.havevdw14scale() || options_.haveelectrostatic14scale()) {
650 >        oneFourInteractions_.removePair(a, d);      
651 >      } else {
652 >        excludedInteractions_.removePair(a, d);
653 >      }
654      }
655  
656 <    Molecule::RigidBodyIterator rbIter;
657 <    RigidBody* rb;
658 <    for (rb = mol->beginRigidBody(rbIter); rb != NULL; rb = mol->nextRigidBody(rbIter)) {
659 <      std::vector<Atom*> atoms = rb->getAtoms();
660 <      for (int i = 0; i < atoms.size() -1 ; ++i) {
661 <        for (int j = i + 1; j < atoms.size(); ++j) {
656 >    for (inversion= mol->beginInversion(inversionIter); inversion != NULL;
657 >         inversion = mol->nextInversion(inversionIter)) {
658 >
659 >      a = inversion->getAtomA()->getGlobalIndex();
660 >      b = inversion->getAtomB()->getGlobalIndex();        
661 >      c = inversion->getAtomC()->getGlobalIndex();        
662 >      d = inversion->getAtomD()->getGlobalIndex();        
663 >
664 >      if (options_.havevdw12scale() || options_.haveelectrostatic12scale()) {
665 >        oneTwoInteractions_.removePair(a, b);      
666 >        oneTwoInteractions_.removePair(a, c);
667 >        oneTwoInteractions_.removePair(a, d);
668 >      } else {
669 >        excludedInteractions_.removePair(a, b);
670 >        excludedInteractions_.removePair(a, c);
671 >        excludedInteractions_.removePair(a, d);
672 >      }
673 >
674 >      if (options_.havevdw13scale() || options_.haveelectrostatic13scale()) {
675 >        oneThreeInteractions_.removePair(b, c);    
676 >        oneThreeInteractions_.removePair(b, d);    
677 >        oneThreeInteractions_.removePair(c, d);      
678 >      } else {
679 >        excludedInteractions_.removePair(b, c);
680 >        excludedInteractions_.removePair(b, d);
681 >        excludedInteractions_.removePair(c, d);
682 >      }
683 >    }
684 >
685 >    for (rb = mol->beginRigidBody(rbIter); rb != NULL;
686 >         rb = mol->nextRigidBody(rbIter)) {
687 >      vector<Atom*> atoms = rb->getAtoms();
688 >      for (int i = 0; i < static_cast<int>(atoms.size()) -1 ; ++i) {
689 >        for (int j = i + 1; j < static_cast<int>(atoms.size()); ++j) {
690            a = atoms[i]->getGlobalIndex();
691            b = atoms[j]->getGlobalIndex();
692 <          exclude_.removePair(a, b);
692 >          excludedInteractions_.removePair(a, b);
693          }
694        }
695      }        
696 <
696 >    
697    }
698 <
699 <
698 >  
699 >  
700    void SimInfo::addMoleculeStamp(MoleculeStamp* molStamp, int nmol) {
701      int curStampId;
702 <
702 >    
703      //index from 0
704      curStampId = moleculeStamps_.size();
705  
# Line 456 | Line 707 | namespace oopse {
707      molStampIds_.insert(molStampIds_.end(), nmol, curStampId);
708    }
709  
459  void SimInfo::update() {
710  
711 <    setupSimType();
712 <
713 < #ifdef IS_MPI
714 <    setupFortranParallel();
715 < #endif
716 <
717 <    setupFortranSim();
718 <
719 <    //setup fortran force field
470 <    /** @deprecate */    
471 <    int isError = 0;
472 <    
473 <    setupElectrostaticSummationMethod( isError );
474 <
475 <    if(isError){
476 <      sprintf( painCave.errMsg,
477 <               "ForceField error: There was an error initializing the forceField in fortran.\n" );
478 <      painCave.isFatal = 1;
479 <      simError();
480 <    }
481 <  
482 <    
483 <    setupCutoff();
484 <
711 >  /**
712 >   * update
713 >   *
714 >   *  Performs the global checks and variable settings after the
715 >   *  objects have been created.
716 >   *
717 >   */
718 >  void SimInfo::update() {  
719 >    setupSimVariables();
720      calcNdf();
721      calcNdfRaw();
722      calcNdfTrans();
488
489    fortranInitialized_ = true;
723    }
724 <
725 <  std::set<AtomType*> SimInfo::getUniqueAtomTypes() {
724 >  
725 >  /**
726 >   * getSimulatedAtomTypes
727 >   *
728 >   * Returns an STL set of AtomType* that are actually present in this
729 >   * simulation.  Must query all processors to assemble this information.
730 >   *
731 >   */
732 >  set<AtomType*> SimInfo::getSimulatedAtomTypes() {
733      SimInfo::MoleculeIterator mi;
734      Molecule* mol;
735      Molecule::AtomIterator ai;
736      Atom* atom;
737 <    std::set<AtomType*> atomTypes;
738 <
737 >    set<AtomType*> atomTypes;
738 >    
739      for(mol = beginMolecule(mi); mol != NULL; mol = nextMolecule(mi)) {
740 <
741 <      for(atom = mol->beginAtom(ai); atom != NULL; atom = mol->nextAtom(ai)) {
740 >      for(atom = mol->beginAtom(ai); atom != NULL;
741 >          atom = mol->nextAtom(ai)) {
742          atomTypes.insert(atom->getAtomType());
743 <      }
744 <        
743 >      }      
744 >    }    
745 >    
746 > #ifdef IS_MPI
747 >
748 >    // loop over the found atom types on this processor, and add their
749 >    // numerical idents to a vector:
750 >    
751 >    vector<int> foundTypes;
752 >    set<AtomType*>::iterator i;
753 >    for (i = atomTypes.begin(); i != atomTypes.end(); ++i)
754 >      foundTypes.push_back( (*i)->getIdent() );
755 >
756 >    // count_local holds the number of found types on this processor
757 >    int count_local = foundTypes.size();
758 >
759 >    int nproc;
760 >    MPI_Comm_size( MPI_COMM_WORLD, &nproc);
761 >    // int nproc = MPI::COMM_WORLD.Get_size();
762 >
763 >    // we need arrays to hold the counts and displacement vectors for
764 >    // all processors
765 >    vector<int> counts(nproc, 0);
766 >    vector<int> disps(nproc, 0);
767 >
768 >    // fill the counts array
769 >    MPI_Allgather(&count_local, 1, MPI_INT, &counts[0],
770 >                  1, MPI_INT, MPI_COMM_WORLD);
771 >    // MPI::COMM_WORLD.Allgather(&count_local, 1, MPI::INT, &counts[0],
772 >    //                           1, MPI::INT);
773 >  
774 >    // use the processor counts to compute the displacement array
775 >    disps[0] = 0;    
776 >    int totalCount = counts[0];
777 >    for (int iproc = 1; iproc < nproc; iproc++) {
778 >      disps[iproc] = disps[iproc-1] + counts[iproc-1];
779 >      totalCount += counts[iproc];
780      }
781  
782 <    return atomTypes;        
783 <  }
782 >    // we need a (possibly redundant) set of all found types:
783 >    vector<int> ftGlobal(totalCount);
784 >    
785 >    // now spray out the foundTypes to all the other processors:    
786 >    MPI_Allgatherv(&foundTypes[0], count_local, MPI_INT,
787 >                   &ftGlobal[0], &counts[0], &disps[0],
788 >                   MPI_INT, MPI_COMM_WORLD);
789 >    // MPI::COMM_WORLD.Allgatherv(&foundTypes[0], count_local, MPI::INT,
790 >    //                            &ftGlobal[0], &counts[0], &disps[0],
791 >    //                            MPI::INT);
792  
793 <  void SimInfo::setupSimType() {
794 <    std::set<AtomType*>::iterator i;
795 <    std::set<AtomType*> atomTypes;
796 <    atomTypes = getUniqueAtomTypes();
793 >    vector<int>::iterator j;
794 >
795 >    // foundIdents is a stl set, so inserting an already found ident
796 >    // will have no effect.
797 >    set<int> foundIdents;
798 >
799 >    for (j = ftGlobal.begin(); j != ftGlobal.end(); ++j)
800 >      foundIdents.insert((*j));
801      
802 <    int useLennardJones = 0;
803 <    int useElectrostatic = 0;
804 <    int useEAM = 0;
805 <    int useCharge = 0;
806 <    int useDirectional = 0;
807 <    int useDipole = 0;
808 <    int useGayBerne = 0;
522 <    int useSticky = 0;
523 <    int useStickyPower = 0;
524 <    int useShape = 0;
525 <    int useFLARB = 0; //it is not in AtomType yet
526 <    int useDirectionalAtom = 0;    
527 <    int useElectrostatics = 0;
528 <    //usePBC and useRF are from simParams
529 <    int usePBC = simParams_->getUsePeriodicBoundaryConditions();
530 <    int useRF;
531 <    int useDW;
532 <    std::string myMethod;
802 >    // now iterate over the foundIdents and get the actual atom types
803 >    // that correspond to these:
804 >    set<int>::iterator it;
805 >    for (it = foundIdents.begin(); it != foundIdents.end(); ++it)
806 >      atomTypes.insert( forceField_->getAtomType((*it)) );
807 >
808 > #endif
809  
810 <    // set the useRF logical
811 <    useRF = 0;
536 <    useDW = 0;
810 >    return atomTypes;        
811 >  }
812  
813  
814 <    if (simParams_->haveElectrostaticSummationMethod()) {
815 <      std::string myMethod = simParams_->getElectrostaticSummationMethod();
816 <      toUpper(myMethod);
817 <      if (myMethod == "REACTION_FIELD") {
543 <        useRF=1;
544 <      } else {
545 <        if (myMethod == "DAMPED_WOLF") {
546 <          useDW = 1;
547 <        }
548 <      }
549 <    }
814 >  int getGlobalCountOfType(AtomType* atype) {
815 >    /*
816 >    set<AtomType*> atypes = getSimulatedAtomTypes();
817 >    map<AtomType*, int> counts_;
818  
819 +    for(mol = beginMolecule(mi); mol != NULL; mol = nextMolecule(mi)) {
820 +      for(atom = mol->beginAtom(ai); atom != NULL;
821 +          atom = mol->nextAtom(ai)) {
822 +        atom->getAtomType();
823 +      }      
824 +    }    
825 +    */
826 +    return 0;
827 +  }
828 +
829 +  void SimInfo::setupSimVariables() {
830 +    useAtomicVirial_ = simParams_->getUseAtomicVirial();
831 +    // we only call setAccumulateBoxDipole if the accumulateBoxDipole
832 +    // parameter is true
833 +    calcBoxDipole_ = false;
834 +    if ( simParams_->haveAccumulateBoxDipole() )
835 +      if ( simParams_->getAccumulateBoxDipole() ) {
836 +        calcBoxDipole_ = true;      
837 +      }
838 +    
839 +    set<AtomType*>::iterator i;
840 +    set<AtomType*> atomTypes;
841 +    atomTypes = getSimulatedAtomTypes();    
842 +    bool usesElectrostatic = false;
843 +    bool usesMetallic = false;
844 +    bool usesDirectional = false;
845 +    bool usesFluctuatingCharges =  false;
846      //loop over all of the atom types
847      for (i = atomTypes.begin(); i != atomTypes.end(); ++i) {
848 <      useLennardJones |= (*i)->isLennardJones();
849 <      useElectrostatic |= (*i)->isElectrostatic();
850 <      useEAM |= (*i)->isEAM();
851 <      useCharge |= (*i)->isCharge();
557 <      useDirectional |= (*i)->isDirectional();
558 <      useDipole |= (*i)->isDipole();
559 <      useGayBerne |= (*i)->isGayBerne();
560 <      useSticky |= (*i)->isSticky();
561 <      useStickyPower |= (*i)->isStickyPower();
562 <      useShape |= (*i)->isShape();
848 >      usesElectrostatic |= (*i)->isElectrostatic();
849 >      usesMetallic |= (*i)->isMetal();
850 >      usesDirectional |= (*i)->isDirectional();
851 >      usesFluctuatingCharges |= (*i)->isFluctuatingCharge();
852      }
853  
854 <    if (useSticky || useStickyPower || useDipole || useGayBerne || useShape) {
566 <      useDirectionalAtom = 1;
567 <    }
568 <
569 <    if (useCharge || useDipole) {
570 <      useElectrostatics = 1;
571 <    }
572 <
573 < #ifdef IS_MPI    
854 > #ifdef IS_MPI
855      int temp;
856  
857 <    temp = usePBC;
858 <    MPI_Allreduce(&temp, &usePBC, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
857 >    temp = usesDirectional;
858 >    MPI_Allreduce(MPI_IN_PLACE, &temp, 1, MPI_INT,  MPI_LOR, MPI_COMM_WORLD);
859 >    usesDirectionalAtoms_ = (temp == 0) ? false : true;
860  
861 <    temp = useDirectionalAtom;
862 <    MPI_Allreduce(&temp, &useDirectionalAtom, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
861 >    // MPI::COMM_WORLD.Allreduce(&temp, &usesDirectionalAtoms_, 1, MPI::BOOL,
862 >    //                           MPI::LOR);
863 >    
864 >    temp = usesMetallic;
865 >    MPI_Allreduce(MPI_IN_PLACE, &temp, 1, MPI_INT,  MPI_LOR, MPI_COMM_WORLD);
866 >    usesMetallicAtoms_ = (temp == 0) ? false : true;
867  
868 <    temp = useLennardJones;
869 <    MPI_Allreduce(&temp, &useLennardJones, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
868 >    // MPI::COMM_WORLD.Allreduce(&temp, &usesMetallicAtoms_, 1, MPI::BOOL,
869 >    //                           MPI::LOR);
870 >    
871 >    temp = usesElectrostatic;
872 >    MPI_Allreduce(MPI_IN_PLACE, &temp, 1, MPI_INT,  MPI_LOR, MPI_COMM_WORLD);
873 >    usesElectrostaticAtoms_ = (temp == 0) ? false : true;
874  
875 <    temp = useElectrostatics;
876 <    MPI_Allreduce(&temp, &useElectrostatics, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
875 >    // MPI::COMM_WORLD.Allreduce(&temp, &usesElectrostaticAtoms_, 1, MPI::BOOL,
876 >    //                           MPI::LOR);
877  
878 <    temp = useCharge;
879 <    MPI_Allreduce(&temp, &useCharge, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
878 >    temp = usesFluctuatingCharges;
879 >    MPI_Allreduce(MPI_IN_PLACE, &temp, 1, MPI_INT,  MPI_LOR, MPI_COMM_WORLD);
880 >    usesFluctuatingCharges_ = (temp == 0) ? false : true;
881  
882 <    temp = useDipole;
883 <    MPI_Allreduce(&temp, &useDipole, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
882 >    // MPI::COMM_WORLD.Allreduce(&temp, &usesFluctuatingCharges_, 1, MPI::BOOL,
883 >    //                           MPI::LOR);
884  
885 <    temp = useSticky;
595 <    MPI_Allreduce(&temp, &useSticky, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
885 > #else
886  
887 <    temp = useStickyPower;
888 <    MPI_Allreduce(&temp, &useStickyPower, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
887 >    usesDirectionalAtoms_ = usesDirectional;
888 >    usesMetallicAtoms_ = usesMetallic;
889 >    usesElectrostaticAtoms_ = usesElectrostatic;
890 >    usesFluctuatingCharges_ = usesFluctuatingCharges;
891 >
892 > #endif
893      
894 <    temp = useGayBerne;
895 <    MPI_Allreduce(&temp, &useGayBerne, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
894 >    requiresPrepair_ = usesMetallicAtoms_ ? true : false;
895 >    requiresSkipCorrection_ = usesElectrostaticAtoms_ ? true : false;
896 >    requiresSelfCorrection_ = usesElectrostaticAtoms_ ? true : false;    
897 >  }
898  
603    temp = useEAM;
604    MPI_Allreduce(&temp, &useEAM, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
899  
900 <    temp = useShape;
901 <    MPI_Allreduce(&temp, &useShape, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);  
900 >  vector<int> SimInfo::getGlobalAtomIndices() {
901 >    SimInfo::MoleculeIterator mi;
902 >    Molecule* mol;
903 >    Molecule::AtomIterator ai;
904 >    Atom* atom;
905  
906 <    temp = useFLARB;
907 <    MPI_Allreduce(&temp, &useFLARB, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
906 >    vector<int> GlobalAtomIndices(getNAtoms(), 0);
907 >    
908 >    for (mol = beginMolecule(mi); mol != NULL; mol  = nextMolecule(mi)) {
909 >      
910 >      for (atom = mol->beginAtom(ai); atom != NULL; atom = mol->nextAtom(ai)) {
911 >        GlobalAtomIndices[atom->getLocalIndex()] = atom->getGlobalIndex();
912 >      }
913 >    }
914 >    return GlobalAtomIndices;
915 >  }
916  
612    temp = useRF;
613    MPI_Allreduce(&temp, &useRF, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
917  
918 <    temp = useDW;
919 <    MPI_Allreduce(&temp, &useDW, 1, MPI_INT, MPI_LOR, MPI_COMM_WORLD);    
918 >  vector<int> SimInfo::getGlobalGroupIndices() {
919 >    SimInfo::MoleculeIterator mi;
920 >    Molecule* mol;
921 >    Molecule::CutoffGroupIterator ci;
922 >    CutoffGroup* cg;
923  
924 < #endif
925 <
926 <    fInfo_.SIM_uses_PBC = usePBC;    
621 <    fInfo_.SIM_uses_DirectionalAtoms = useDirectionalAtom;
622 <    fInfo_.SIM_uses_LennardJones = useLennardJones;
623 <    fInfo_.SIM_uses_Electrostatics = useElectrostatics;    
624 <    fInfo_.SIM_uses_Charges = useCharge;
625 <    fInfo_.SIM_uses_Dipoles = useDipole;
626 <    fInfo_.SIM_uses_Sticky = useSticky;
627 <    fInfo_.SIM_uses_StickyPower = useStickyPower;
628 <    fInfo_.SIM_uses_GayBerne = useGayBerne;
629 <    fInfo_.SIM_uses_EAM = useEAM;
630 <    fInfo_.SIM_uses_Shapes = useShape;
631 <    fInfo_.SIM_uses_FLARB = useFLARB;
632 <    fInfo_.SIM_uses_RF = useRF;
633 <    fInfo_.SIM_uses_DampedWolf = useDW;
634 <
635 <    if( myMethod == "REACTION_FIELD") {
924 >    vector<int> GlobalGroupIndices;
925 >    
926 >    for (mol = beginMolecule(mi); mol != NULL; mol  = nextMolecule(mi)) {
927        
928 <      if (simParams_->haveDielectric()) {
929 <        fInfo_.dielect = simParams_->getDielectric();
930 <      } else {
931 <        sprintf(painCave.errMsg,
932 <                "SimSetup Error: No Dielectric constant was set.\n"
933 <                "\tYou are trying to use Reaction Field without"
643 <                "\tsetting a dielectric constant!\n");
644 <        painCave.isFatal = 1;
645 <        simError();
646 <      }      
928 >      //local index of cutoff group is trivial, it only depends on the
929 >      //order of travesing
930 >      for (cg = mol->beginCutoffGroup(ci); cg != NULL;
931 >           cg = mol->nextCutoffGroup(ci)) {
932 >        GlobalGroupIndices.push_back(cg->getGlobalIndex());
933 >      }        
934      }
935 <
935 >    return GlobalGroupIndices;
936    }
937  
651  void SimInfo::setupFortranSim() {
652    int isError;
653    int nExclude;
654    std::vector<int> fortranGlobalGroupMembership;
655    
656    nExclude = exclude_.getSize();
657    isError = 0;
938  
939 <    //globalGroupMembership_ is filled by SimCreator    
660 <    for (int i = 0; i < nGlobalAtoms_; i++) {
661 <      fortranGlobalGroupMembership.push_back(globalGroupMembership_[i] + 1);
662 <    }
939 >  void SimInfo::prepareTopology() {
940  
941      //calculate mass ratio of cutoff group
665    std::vector<double> mfact;
942      SimInfo::MoleculeIterator mi;
943      Molecule* mol;
944      Molecule::CutoffGroupIterator ci;
945      CutoffGroup* cg;
946      Molecule::AtomIterator ai;
947      Atom* atom;
948 <    double totalMass;
948 >    RealType totalMass;
949  
950 <    //to avoid memory reallocation, reserve enough space for mfact
951 <    mfact.reserve(getNCutoffGroups());
950 >    /**
951 >     * The mass factor is the relative mass of an atom to the total
952 >     * mass of the cutoff group it belongs to.  By default, all atoms
953 >     * are their own cutoff groups, and therefore have mass factors of
954 >     * 1.  We need some special handling for massless atoms, which
955 >     * will be treated as carrying the entire mass of the cutoff
956 >     * group.
957 >     */
958 >    massFactors_.clear();
959 >    massFactors_.resize(getNAtoms(), 1.0);
960      
961      for(mol = beginMolecule(mi); mol != NULL; mol = nextMolecule(mi)) {        
962 <      for (cg = mol->beginCutoffGroup(ci); cg != NULL; cg = mol->nextCutoffGroup(ci)) {
962 >      for (cg = mol->beginCutoffGroup(ci); cg != NULL;
963 >           cg = mol->nextCutoffGroup(ci)) {
964  
965          totalMass = cg->getMass();
966          for(atom = cg->beginAtom(ai); atom != NULL; atom = cg->nextAtom(ai)) {
967            // Check for massless groups - set mfact to 1 if true
968 <          if (totalMass != 0)
969 <            mfact.push_back(atom->getMass()/totalMass);
968 >          if (totalMass != 0)
969 >            massFactors_[atom->getLocalIndex()] = atom->getMass()/totalMass;
970            else
971 <            mfact.push_back( 1.0 );
971 >            massFactors_[atom->getLocalIndex()] = 1.0;
972          }
688
973        }      
974      }
975  
976 <    //fill ident array of local atoms (it is actually ident of AtomType, it is so confusing !!!)
693 <    std::vector<int> identArray;
976 >    // Build the identArray_ and regions_
977  
978 <    //to avoid memory reallocation, reserve enough space identArray
979 <    identArray.reserve(getNAtoms());
980 <    
981 <    for(mol = beginMolecule(mi); mol != NULL; mol = nextMolecule(mi)) {        
978 >    identArray_.clear();
979 >    identArray_.reserve(getNAtoms());  
980 >    regions_.clear();
981 >    regions_.reserve(getNAtoms());
982 >
983 >    for(mol = beginMolecule(mi); mol != NULL; mol = nextMolecule(mi)) {      
984 >      int reg = mol->getRegion();      
985        for(atom = mol->beginAtom(ai); atom != NULL; atom = mol->nextAtom(ai)) {
986 <        identArray.push_back(atom->getIdent());
986 >        identArray_.push_back(atom->getIdent());
987 >        regions_.push_back(reg);
988        }
989      }    
990 <
991 <    //fill molMembershipArray
705 <    //molMembershipArray is filled by SimCreator    
706 <    std::vector<int> molMembershipArray(nGlobalAtoms_);
707 <    for (int i = 0; i < nGlobalAtoms_; i++) {
708 <      molMembershipArray[i] = globalMolMembership_[i] + 1;
709 <    }
710 <    
711 <    //setup fortran simulation
712 <    int nGlobalExcludes = 0;
713 <    int* globalExcludes = NULL;
714 <    int* excludeList = exclude_.getExcludeList();
715 <    setFortranSim( &fInfo_, &nGlobalAtoms_, &nAtoms_, &identArray[0], &nExclude, excludeList ,
716 <                   &nGlobalExcludes, globalExcludes, &molMembershipArray[0],
717 <                   &mfact[0], &nCutoffGroups_, &fortranGlobalGroupMembership[0], &isError);
718 <
719 <    if( isError ){
720 <
721 <      sprintf( painCave.errMsg,
722 <               "There was an error setting the simulation information in fortran.\n" );
723 <      painCave.isFatal = 1;
724 <      painCave.severity = OOPSE_ERROR;
725 <      simError();
726 <    }
727 <
728 < #ifdef IS_MPI
729 <    sprintf( checkPointMsg,
730 <             "succesfully sent the simulation information to fortran.\n");
731 <    MPIcheckPoint();
732 < #endif // is_mpi
733 <  }
734 <
735 <
736 < #ifdef IS_MPI
737 <  void SimInfo::setupFortranParallel() {
738 <    
739 <    //SimInfo is responsible for creating localToGlobalAtomIndex and localToGlobalGroupIndex
740 <    std::vector<int> localToGlobalAtomIndex(getNAtoms(), 0);
741 <    std::vector<int> localToGlobalCutoffGroupIndex;
742 <    SimInfo::MoleculeIterator mi;
743 <    Molecule::AtomIterator ai;
744 <    Molecule::CutoffGroupIterator ci;
745 <    Molecule* mol;
746 <    Atom* atom;
747 <    CutoffGroup* cg;
748 <    mpiSimData parallelData;
749 <    int isError;
750 <
751 <    for (mol = beginMolecule(mi); mol != NULL; mol  = nextMolecule(mi)) {
752 <
753 <      //local index(index in DataStorge) of atom is important
754 <      for (atom = mol->beginAtom(ai); atom != NULL; atom = mol->nextAtom(ai)) {
755 <        localToGlobalAtomIndex[atom->getLocalIndex()] = atom->getGlobalIndex() + 1;
756 <      }
757 <
758 <      //local index of cutoff group is trivial, it only depends on the order of travesing
759 <      for (cg = mol->beginCutoffGroup(ci); cg != NULL; cg = mol->nextCutoffGroup(ci)) {
760 <        localToGlobalCutoffGroupIndex.push_back(cg->getGlobalIndex() + 1);
761 <      }        
762 <        
763 <    }
764 <
765 <    //fill up mpiSimData struct
766 <    parallelData.nMolGlobal = getNGlobalMolecules();
767 <    parallelData.nMolLocal = getNMolecules();
768 <    parallelData.nAtomsGlobal = getNGlobalAtoms();
769 <    parallelData.nAtomsLocal = getNAtoms();
770 <    parallelData.nGroupsGlobal = getNGlobalCutoffGroups();
771 <    parallelData.nGroupsLocal = getNCutoffGroups();
772 <    parallelData.myNode = worldRank;
773 <    MPI_Comm_size(MPI_COMM_WORLD, &(parallelData.nProcessors));
774 <
775 <    //pass mpiSimData struct and index arrays to fortran
776 <    setFsimParallel(&parallelData, &(parallelData.nAtomsLocal),
777 <                    &localToGlobalAtomIndex[0],  &(parallelData.nGroupsLocal),
778 <                    &localToGlobalCutoffGroupIndex[0], &isError);
779 <
780 <    if (isError) {
781 <      sprintf(painCave.errMsg,
782 <              "mpiRefresh errror: fortran didn't like something we gave it.\n");
783 <      painCave.isFatal = 1;
784 <      simError();
785 <    }
786 <
787 <    sprintf(checkPointMsg, " mpiRefresh successful.\n");
788 <    MPIcheckPoint();
789 <
790 <
791 <  }
792 <
793 < #endif
794 <
795 <  double SimInfo::calcMaxCutoffRadius() {
796 <
797 <
798 <    std::set<AtomType*> atomTypes;
799 <    std::set<AtomType*>::iterator i;
800 <    std::vector<double> cutoffRadius;
801 <
802 <    //get the unique atom types
803 <    atomTypes = getUniqueAtomTypes();
804 <
805 <    //query the max cutoff radius among these atom types
806 <    for (i = atomTypes.begin(); i != atomTypes.end(); ++i) {
807 <      cutoffRadius.push_back(forceField_->getRcutFromAtomType(*i));
808 <    }
809 <
810 <    double maxCutoffRadius = *(std::max_element(cutoffRadius.begin(), cutoffRadius.end()));
811 < #ifdef IS_MPI
812 <    //pick the max cutoff radius among the processors
813 < #endif
814 <
815 <    return maxCutoffRadius;
816 <  }
817 <
818 <  void SimInfo::getCutoff(double& rcut, double& rsw) {
819 <    
820 <    if (fInfo_.SIM_uses_Charges | fInfo_.SIM_uses_Dipoles | fInfo_.SIM_uses_RF) {
821 <        
822 <      if (!simParams_->haveCutoffRadius()){
823 <        sprintf(painCave.errMsg,
824 <                "SimCreator Warning: No value was set for the cutoffRadius.\n"
825 <                "\tOOPSE will use a default value of 15.0 angstroms"
826 <                "\tfor the cutoffRadius.\n");
827 <        painCave.isFatal = 0;
828 <        simError();
829 <        rcut = 15.0;
830 <      } else{
831 <        rcut = simParams_->getCutoffRadius();
832 <      }
833 <
834 <      if (!simParams_->haveSwitchingRadius()){
835 <        sprintf(painCave.errMsg,
836 <                "SimCreator Warning: No value was set for switchingRadius.\n"
837 <                "\tOOPSE will use a default value of\n"
838 <                "\t0.85 * cutoffRadius for the switchingRadius\n");
839 <        painCave.isFatal = 0;
840 <        simError();
841 <        rsw = 0.85 * rcut;
842 <      } else{
843 <        rsw = simParams_->getSwitchingRadius();
844 <      }
845 <
846 <    } else {
847 <      // if charge, dipole or reaction field is not used and the cutofff radius is not specified in
848 <      //meta-data file, the maximum cutoff radius calculated from forcefiled will be used
849 <        
850 <      if (simParams_->haveCutoffRadius()) {
851 <        rcut = simParams_->getCutoffRadius();
852 <      } else {
853 <        //set cutoff radius to the maximum cutoff radius based on atom types in the whole system
854 <        rcut = calcMaxCutoffRadius();
855 <      }
856 <
857 <      if (simParams_->haveSwitchingRadius()) {
858 <        rsw  = simParams_->getSwitchingRadius();
859 <      } else {
860 <        rsw = rcut;
861 <      }
862 <    
863 <    }
990 >      
991 >    topologyDone_ = true;
992    }
993  
866  void SimInfo::setupCutoff() {    
867    getCutoff(rcut_, rsw_);    
868    double rnblist = rcut_ + 1; // skin of neighbor list
869
870    //Pass these cutoff radius etc. to fortran. This function should be called once and only once
871    
872    int cp =  TRADITIONAL_CUTOFF_POLICY;
873    if (simParams_->haveCutoffPolicy()) {
874      std::string myPolicy = simParams_->getCutoffPolicy();
875      toUpper(myPolicy);
876      if (myPolicy == "MIX") {
877        cp = MIX_CUTOFF_POLICY;
878      } else {
879        if (myPolicy == "MAX") {
880          cp = MAX_CUTOFF_POLICY;
881        } else {
882          if (myPolicy == "TRADITIONAL") {            
883            cp = TRADITIONAL_CUTOFF_POLICY;
884          } else {
885            // throw error        
886            sprintf( painCave.errMsg,
887                     "SimInfo error: Unknown cutoffPolicy. (Input file specified %s .)\n\tcutoffPolicy must be one of: \"Mix\", \"Max\", or \"Traditional\".", myPolicy.c_str() );
888            painCave.isFatal = 1;
889            simError();
890          }    
891        }          
892      }
893    }
894
895
896    if (simParams_->haveSkinThickness()) {
897      double skinThickness = simParams_->getSkinThickness();
898    }
899
900    notifyFortranCutoffs(&rcut_, &rsw_, &rnblist, &cp);
901    // also send cutoff notification to electrostatics
902    setElectrostaticCutoffRadius(&rcut_, &rsw_);
903  }
904
905  void SimInfo::setupElectrostaticSummationMethod( int isError ) {    
906    
907    int errorOut;
908    int esm =  NONE;
909    double alphaVal;
910    double dielectric;
911
912    errorOut = isError;
913    alphaVal = simParams_->getDampingAlpha();
914    dielectric = simParams_->getDielectric();
915
916    if (simParams_->haveElectrostaticSummationMethod()) {
917      std::string myMethod = simParams_->getElectrostaticSummationMethod();
918      toUpper(myMethod);
919      if (myMethod == "NONE") {
920        esm = NONE;
921      } else {
922        if (myMethod == "UNDAMPED_WOLF") {
923          esm = UNDAMPED_WOLF;
924        } else {
925          if (myMethod == "DAMPED_WOLF") {            
926            esm = DAMPED_WOLF;
927            if (!simParams_->haveDampingAlpha()) {
928              //throw error
929              sprintf( painCave.errMsg,
930                       "SimInfo warning: dampingAlpha was not specified in the input file. A default value of %f (1/ang) will be used for the Damped Wolf Method.", alphaVal);
931              painCave.isFatal = 0;
932              simError();
933            }
934          } else {
935            if (myMethod == "REACTION_FIELD") {      
936              esm = REACTION_FIELD;
937            } else {
938              // throw error        
939              sprintf( painCave.errMsg,
940                       "SimInfo error: Unknown electrostaticSummationMethod. (Input file specified %s .)\n\telectrostaticSummationMethod must be one of: \"none\", \"undamped_wolf\", \"damped_wolf\", or \"reaction_field\".", myMethod.c_str() );
941              painCave.isFatal = 1;
942              simError();
943            }    
944          }          
945        }
946      }
947    }
948    // let's pass some summation method variables to fortran
949    setElectrostaticSummationMethod( &esm );
950    setDampedWolfAlpha( &alphaVal );
951    setReactionFieldDielectric( &dielectric );
952    initFortranFF( &esm, &errorOut );
953  }
954
994    void SimInfo::addProperty(GenericData* genData) {
995      properties_.addProperty(genData);  
996    }
997  
998 <  void SimInfo::removeProperty(const std::string& propName) {
998 >  void SimInfo::removeProperty(const string& propName) {
999      properties_.removeProperty(propName);  
1000    }
1001  
# Line 964 | Line 1003 | namespace oopse {
1003      properties_.clearProperties();
1004    }
1005  
1006 <  std::vector<std::string> SimInfo::getPropertyNames() {
1006 >  vector<string> SimInfo::getPropertyNames() {
1007      return properties_.getPropertyNames();  
1008    }
1009        
1010 <  std::vector<GenericData*> SimInfo::getProperties() {
1010 >  vector<GenericData*> SimInfo::getProperties() {
1011      return properties_.getProperties();
1012    }
1013  
1014 <  GenericData* SimInfo::getPropertyByName(const std::string& propName) {
1014 >  GenericData* SimInfo::getPropertyByName(const string& propName) {
1015      return properties_.getPropertyByName(propName);
1016    }
1017  
# Line 983 | Line 1022 | namespace oopse {
1022      delete sman_;
1023      sman_ = sman;
1024  
986    Molecule* mol;
987    RigidBody* rb;
988    Atom* atom;
1025      SimInfo::MoleculeIterator mi;
1026 +    Molecule::AtomIterator ai;
1027      Molecule::RigidBodyIterator rbIter;
1028 <    Molecule::AtomIterator atomIter;;
1028 >    Molecule::CutoffGroupIterator cgIter;
1029 >    Molecule::BondIterator bondIter;
1030 >    Molecule::BendIterator bendIter;
1031 >    Molecule::TorsionIterator torsionIter;
1032 >    Molecule::InversionIterator inversionIter;
1033  
1034 +    Molecule* mol;
1035 +    Atom* atom;
1036 +    RigidBody* rb;
1037 +    CutoffGroup* cg;
1038 +    Bond* bond;
1039 +    Bend* bend;
1040 +    Torsion* torsion;
1041 +    Inversion* inversion;    
1042 +
1043      for (mol = beginMolecule(mi); mol != NULL; mol = nextMolecule(mi)) {
1044          
1045 <      for (atom = mol->beginAtom(atomIter); atom != NULL; atom = mol->nextAtom(atomIter)) {
1045 >      for (atom = mol->beginAtom(ai); atom != NULL;
1046 >           atom = mol->nextAtom(ai)) {
1047          atom->setSnapshotManager(sman_);
1048 <      }
1049 <        
1050 <      for (rb = mol->beginRigidBody(rbIter); rb != NULL; rb = mol->nextRigidBody(rbIter)) {
1048 >      }        
1049 >      for (rb = mol->beginRigidBody(rbIter); rb != NULL;
1050 >           rb = mol->nextRigidBody(rbIter)) {
1051          rb->setSnapshotManager(sman_);
1052        }
1053 <    }    
1054 <    
1053 >      for (cg = mol->beginCutoffGroup(cgIter); cg != NULL;
1054 >           cg = mol->nextCutoffGroup(cgIter)) {
1055 >        cg->setSnapshotManager(sman_);
1056 >      }
1057 >      for (bond = mol->beginBond(bondIter); bond != NULL;
1058 >           bond = mol->nextBond(bondIter)) {
1059 >        bond->setSnapshotManager(sman_);
1060 >      }
1061 >      for (bend = mol->beginBend(bendIter); bend != NULL;
1062 >           bend = mol->nextBend(bendIter)) {
1063 >        bend->setSnapshotManager(sman_);
1064 >      }
1065 >      for (torsion = mol->beginTorsion(torsionIter); torsion != NULL;
1066 >           torsion = mol->nextTorsion(torsionIter)) {
1067 >        torsion->setSnapshotManager(sman_);
1068 >      }
1069 >      for (inversion = mol->beginInversion(inversionIter); inversion != NULL;
1070 >           inversion = mol->nextInversion(inversionIter)) {
1071 >        inversion->setSnapshotManager(sman_);
1072 >      }
1073 >    }
1074    }
1075  
1006  Vector3d SimInfo::getComVel(){
1007    SimInfo::MoleculeIterator i;
1008    Molecule* mol;
1076  
1077 <    Vector3d comVel(0.0);
1011 <    double totalMass = 0.0;
1012 <    
1013 <
1014 <    for (mol = beginMolecule(i); mol != NULL; mol = nextMolecule(i)) {
1015 <      double mass = mol->getMass();
1016 <      totalMass += mass;
1017 <      comVel += mass * mol->getComVel();
1018 <    }  
1077 >  ostream& operator <<(ostream& o, SimInfo& info) {
1078  
1020 #ifdef IS_MPI
1021    double tmpMass = totalMass;
1022    Vector3d tmpComVel(comVel);    
1023    MPI_Allreduce(&tmpMass,&totalMass,1,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1024    MPI_Allreduce(tmpComVel.getArrayPointer(), comVel.getArrayPointer(),3,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1025 #endif
1026
1027    comVel /= totalMass;
1028
1029    return comVel;
1030  }
1031
1032  Vector3d SimInfo::getCom(){
1033    SimInfo::MoleculeIterator i;
1034    Molecule* mol;
1035
1036    Vector3d com(0.0);
1037    double totalMass = 0.0;
1038    
1039    for (mol = beginMolecule(i); mol != NULL; mol = nextMolecule(i)) {
1040      double mass = mol->getMass();
1041      totalMass += mass;
1042      com += mass * mol->getCom();
1043    }  
1044
1045 #ifdef IS_MPI
1046    double tmpMass = totalMass;
1047    Vector3d tmpCom(com);    
1048    MPI_Allreduce(&tmpMass,&totalMass,1,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1049    MPI_Allreduce(tmpCom.getArrayPointer(), com.getArrayPointer(),3,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1050 #endif
1051
1052    com /= totalMass;
1053
1054    return com;
1055
1056  }        
1057
1058  std::ostream& operator <<(std::ostream& o, SimInfo& info) {
1059
1079      return o;
1080    }
1081    
1082 <  
1083 <   /*
1084 <   Returns center of mass and center of mass velocity in one function call.
1085 <   */
1086 <  
1087 <   void SimInfo::getComAll(Vector3d &com, Vector3d &comVel){
1088 <      SimInfo::MoleculeIterator i;
1089 <      Molecule* mol;
1090 <      
1091 <    
1092 <      double totalMass = 0.0;
1093 <    
1082 >  
1083 >  StuntDouble* SimInfo::getIOIndexToIntegrableObject(int index) {
1084 >    if (index >= int(IOIndexToIntegrableObject.size())) {
1085 >      sprintf(painCave.errMsg,
1086 >              "SimInfo::getIOIndexToIntegrableObject Error: Integrable Object\n"
1087 >              "\tindex exceeds number of known objects!\n");
1088 >      painCave.isFatal = 1;
1089 >      simError();
1090 >      return NULL;
1091 >    } else
1092 >      return IOIndexToIntegrableObject.at(index);
1093 >  }
1094 >  
1095 >  void SimInfo::setIOIndexToIntegrableObject(const vector<StuntDouble*>& v) {
1096 >    IOIndexToIntegrableObject= v;
1097 >  }
1098  
1099 <      for (mol = beginMolecule(i); mol != NULL; mol = nextMolecule(i)) {
1100 <         double mass = mol->getMass();
1078 <         totalMass += mass;
1079 <         com += mass * mol->getCom();
1080 <         comVel += mass * mol->getComVel();          
1081 <      }  
1082 <      
1099 >  int SimInfo::getNGlobalConstraints() {
1100 >    int nGlobalConstraints;
1101   #ifdef IS_MPI
1102 <      double tmpMass = totalMass;
1103 <      Vector3d tmpCom(com);  
1104 <      Vector3d tmpComVel(comVel);
1105 <      MPI_Allreduce(&tmpMass,&totalMass,1,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1106 <      MPI_Allreduce(tmpCom.getArrayPointer(), com.getArrayPointer(),3,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1107 <      MPI_Allreduce(tmpComVel.getArrayPointer(), comVel.getArrayPointer(),3,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1102 >    MPI_Allreduce(&nConstraints_, &nGlobalConstraints, 1,  
1103 >                              MPI_INT, MPI_SUM, MPI_COMM_WORLD);
1104 >    // MPI::COMM_WORLD.Allreduce(&nConstraints_, &nGlobalConstraints, 1,
1105 >    //                           MPI::INT, MPI::SUM);
1106 > #else
1107 >    nGlobalConstraints =  nConstraints_;
1108   #endif
1109 <      
1110 <      com /= totalMass;
1093 <      comVel /= totalMass;
1094 <   }        
1095 <  
1096 <   /*
1097 <   Return intertia tensor for entire system and angular momentum Vector.
1109 >    return nGlobalConstraints;
1110 >  }
1111  
1112 + }//end namespace OpenMD
1113  
1100       [  Ixx -Ixy  -Ixz ]
1101  J =| -Iyx  Iyy  -Iyz |
1102       [ -Izx -Iyz   Izz ]
1103    */
1104
1105   void SimInfo::getInertiaTensor(Mat3x3d &inertiaTensor, Vector3d &angularMomentum){
1106      
1107
1108      double xx = 0.0;
1109      double yy = 0.0;
1110      double zz = 0.0;
1111      double xy = 0.0;
1112      double xz = 0.0;
1113      double yz = 0.0;
1114      Vector3d com(0.0);
1115      Vector3d comVel(0.0);
1116      
1117      getComAll(com, comVel);
1118      
1119      SimInfo::MoleculeIterator i;
1120      Molecule* mol;
1121      
1122      Vector3d thisq(0.0);
1123      Vector3d thisv(0.0);
1124
1125      double thisMass = 0.0;
1126    
1127      
1128      
1129  
1130      for (mol = beginMolecule(i); mol != NULL; mol = nextMolecule(i)) {
1131        
1132         thisq = mol->getCom()-com;
1133         thisv = mol->getComVel()-comVel;
1134         thisMass = mol->getMass();
1135         // Compute moment of intertia coefficients.
1136         xx += thisq[0]*thisq[0]*thisMass;
1137         yy += thisq[1]*thisq[1]*thisMass;
1138         zz += thisq[2]*thisq[2]*thisMass;
1139        
1140         // compute products of intertia
1141         xy += thisq[0]*thisq[1]*thisMass;
1142         xz += thisq[0]*thisq[2]*thisMass;
1143         yz += thisq[1]*thisq[2]*thisMass;
1144            
1145         angularMomentum += cross( thisq, thisv ) * thisMass;
1146            
1147      }  
1148      
1149      
1150      inertiaTensor(0,0) = yy + zz;
1151      inertiaTensor(0,1) = -xy;
1152      inertiaTensor(0,2) = -xz;
1153      inertiaTensor(1,0) = -xy;
1154      inertiaTensor(1,1) = xx + zz;
1155      inertiaTensor(1,2) = -yz;
1156      inertiaTensor(2,0) = -xz;
1157      inertiaTensor(2,1) = -yz;
1158      inertiaTensor(2,2) = xx + yy;
1159      
1160 #ifdef IS_MPI
1161      Mat3x3d tmpI(inertiaTensor);
1162      Vector3d tmpAngMom;
1163      MPI_Allreduce(tmpI.getArrayPointer(), inertiaTensor.getArrayPointer(),9,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1164      MPI_Allreduce(tmpAngMom.getArrayPointer(), angularMomentum.getArrayPointer(),3,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1165 #endif
1166              
1167      return;
1168   }
1169
1170   //Returns the angular momentum of the system
1171   Vector3d SimInfo::getAngularMomentum(){
1172      
1173      Vector3d com(0.0);
1174      Vector3d comVel(0.0);
1175      Vector3d angularMomentum(0.0);
1176      
1177      getComAll(com,comVel);
1178      
1179      SimInfo::MoleculeIterator i;
1180      Molecule* mol;
1181      
1182      Vector3d thisr(0.0);
1183      Vector3d thisp(0.0);
1184      
1185      double thisMass;
1186      
1187      for (mol = beginMolecule(i); mol != NULL; mol = nextMolecule(i)) {        
1188        thisMass = mol->getMass();
1189        thisr = mol->getCom()-com;
1190        thisp = (mol->getComVel()-comVel)*thisMass;
1191        
1192        angularMomentum += cross( thisr, thisp );
1193        
1194      }  
1195      
1196 #ifdef IS_MPI
1197      Vector3d tmpAngMom;
1198      MPI_Allreduce(tmpAngMom.getArrayPointer(), angularMomentum.getArrayPointer(),3,MPI_DOUBLE,MPI_SUM, MPI_COMM_WORLD);
1199 #endif
1200      
1201      return angularMomentum;
1202   }
1203  
1204  
1205 }//end namespace oopse
1206

Comparing trunk/src/brains/SimInfo.cpp (property svn:keywords):
Revision 705 by chrisfen, Tue Nov 1 19:14:27 2005 UTC vs.
Revision 1969 by gezelter, Wed Feb 26 14:14:50 2014 UTC

# Line 0 | Line 1
1 + Author Id Revision Date

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines