| 277 |  | // do some sanity checking | 
| 278 |  |  | 
| 279 |  | int selectionCount = seleMan_.getSelectionCount(); | 
| 280 | + |  | 
| 281 |  | int nIntegrable = info->getNGlobalIntegrableObjects(); | 
| 282 |  |  | 
| 283 |  | if (selectionCount > nIntegrable) { | 
| 914 |  |  | 
| 915 |  | if ((c > 0.81) && (c < 1.21)) {//restrict scaling coefficients | 
| 916 |  | c = sqrt(c); | 
| 917 | < | //std::cerr << "cold slab scaling coefficient: " << c << endl; | 
| 917 | < | //now convert to hotBin coefficient | 
| 917 | > |  | 
| 918 |  | RealType w = 0.0; | 
| 919 |  | if (rnemdFluxType_ ==  rnemdFullKE) { | 
| 920 |  | x = 1.0 + px * (1.0 - c); | 
| 952 |  | } | 
| 953 |  | } | 
| 954 |  | w = sqrt(w); | 
| 955 | – | // std::cerr << "xh= " << x << "\tyh= " << y << "\tzh= " << z | 
| 956 | – | //           << "\twh= " << w << endl; | 
| 955 |  | for (sdi = hotBin.begin(); sdi != hotBin.end(); sdi++) { | 
| 956 |  | if (rnemdFluxType_ == rnemdFullKE) { | 
| 957 |  | vel = (*sdi)->getVel(); | 
| 1260 |  |  | 
| 1261 |  | if (inA) { | 
| 1262 |  | hotBin.push_back(sd); | 
| 1265 | – | //std::cerr << "before, velocity = " << vel << endl; | 
| 1263 |  | Ph += mass * vel; | 
| 1267 | – | //std::cerr << "after, velocity = " << vel << endl; | 
| 1264 |  | Mh += mass; | 
| 1265 |  | Kh += mass * vel.lengthSquare(); | 
| 1266 |  | if (rnemdFluxType_ == rnemdFullKE) { | 
| 1308 |  |  | 
| 1309 |  | Kh *= 0.5; | 
| 1310 |  | Kc *= 0.5; | 
| 1315 | – |  | 
| 1316 | – | // std::cerr << "Mh= " << Mh << "\tKh= " << Kh << "\tMc= " << Mc | 
| 1317 | – | //        << "\tKc= " << Kc << endl; | 
| 1318 | – | // std::cerr << "Ph= " << Ph << "\tPc= " << Pc << endl; | 
| 1311 |  |  | 
| 1312 |  | #ifdef IS_MPI | 
| 1313 |  | MPI::COMM_WORLD.Allreduce(MPI::IN_PLACE, &Ph[0], 3, MPI::REALTYPE, MPI::SUM); | 
| 1339 |  | if (hDenominator > 0.0) { | 
| 1340 |  | RealType h = sqrt(hNumerator / hDenominator); | 
| 1341 |  | if ((h > 0.9) && (h < 1.1)) { | 
| 1342 | < | // std::cerr << "cold slab scaling coefficient: " << c << "\n"; | 
| 1351 | < | // std::cerr << "hot slab scaling coefficient: " << h <<  "\n"; | 
| 1342 | > |  | 
| 1343 |  | vector<StuntDouble*>::iterator sdi; | 
| 1344 |  | Vector3d vel; | 
| 1345 |  | for (sdi = coldBin.begin(); sdi != coldBin.end(); sdi++) { | 
| 1414 |  |  | 
| 1415 |  | seleMan_.setSelectionSet(evaluator_.evaluate()); | 
| 1416 |  |  | 
| 1417 | < | int selei; | 
| 1417 | > | int selei(0); | 
| 1418 |  | StuntDouble* sd; | 
| 1419 |  | int idx; | 
| 1420 |  |  | 
| 1440 |  | sd != NULL; | 
| 1441 |  | sd = mol->nextIntegrableObject(iiter)) | 
| 1442 |  | */ | 
| 1443 | + |  | 
| 1444 |  | for (sd = seleMan_.beginSelected(selei); sd != NULL; | 
| 1445 |  | sd = seleMan_.nextSelected(selei)) { | 
| 1446 | < |  | 
| 1446 | > |  | 
| 1447 |  | idx = sd->getLocalIndex(); | 
| 1448 |  |  | 
| 1449 |  | Vector3d pos = sd->getPos(); | 
| 1460 |  | // The modulo operator is used to wrap the case when we are | 
| 1461 |  | // beyond the end of the bins back to the beginning. | 
| 1462 |  | int binNo = int(nBins_ * (pos.z() / hmat(2,2) + 0.5)) % nBins_; | 
| 1463 | < |  | 
| 1463 | > |  | 
| 1464 |  | RealType mass = sd->getMass(); | 
| 1465 |  | Vector3d vel = sd->getVel(); | 
| 1466 |  |  | 
| 1491 |  | } | 
| 1492 |  | } | 
| 1493 |  |  | 
| 1502 | – |  | 
| 1494 |  | #ifdef IS_MPI | 
| 1495 |  | MPI::COMM_WORLD.Allreduce(MPI::IN_PLACE, &binCount[0], | 
| 1496 |  | nBins_, MPI::INT, MPI::SUM); | 
| 1523 |  |  | 
| 1524 |  | temp = 2.0 * binKE[i] / (binDOF[i] * PhysicalConstants::kb * | 
| 1525 |  | PhysicalConstants::energyConvert); | 
| 1526 | < |  | 
| 1526 | > |  | 
| 1527 |  | for (unsigned int j = 0; j < outputMask_.size(); ++j) { | 
| 1528 |  | if(outputMask_[j]) { | 
| 1529 |  | switch(j) { |