40 |
|
* [5] Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011). |
41 |
|
*/ |
42 |
|
|
43 |
+ |
#ifdef IS_MPI |
44 |
+ |
#include <mpi.h> |
45 |
+ |
#endif |
46 |
+ |
|
47 |
|
#include <stdio.h> |
48 |
|
#include <string.h> |
49 |
|
|
61 |
|
#include "math/erfc.hpp" |
62 |
|
#include "math/SquareMatrix.hpp" |
63 |
|
#include "primitives/Molecule.hpp" |
64 |
+ |
#include "flucq/FluctuatingChargeForces.hpp" |
65 |
|
|
61 |
– |
|
66 |
|
namespace OpenMD { |
67 |
|
|
68 |
|
Electrostatic::Electrostatic(): name_("Electrostatic"), initialized_(false), |
71 |
|
haveDampingAlpha_(false), |
72 |
|
haveDielectric_(false), |
73 |
|
haveElectroSplines_(false) |
74 |
< |
{} |
74 |
> |
{ |
75 |
> |
flucQ_ = new FluctuatingChargeForces(info_); |
76 |
> |
} |
77 |
|
|
78 |
+ |
void Electrostatic::setForceField(ForceField *ff) { |
79 |
+ |
forceField_ = ff; |
80 |
+ |
flucQ_->setForceField(forceField_); |
81 |
+ |
} |
82 |
+ |
|
83 |
+ |
void Electrostatic::setSimulatedAtomTypes(set<AtomType*> &simtypes) { |
84 |
+ |
simTypes_ = simtypes; |
85 |
+ |
flucQ_->setSimulatedAtomTypes(simTypes_); |
86 |
+ |
} |
87 |
+ |
|
88 |
|
void Electrostatic::initialize() { |
89 |
|
|
90 |
|
Globals* simParams_ = info_->getSimParams(); |
305 |
|
db0c_4 = 3.0*b2c - 6.0*r2*b3c + r2*r2*b4c; |
306 |
|
db0c_5 = -15.0*r*b3c + 10.0*r2*r*b4c - r2*r2*r*b5c; |
307 |
|
|
308 |
< |
if (summationMethod_ == esm_EWALD_FULL) { |
293 |
< |
selfMult1_ *= 2.0; |
294 |
< |
selfMult2_ *= 2.0; |
295 |
< |
selfMult4_ *= 2.0; |
296 |
< |
} else { |
308 |
> |
if (summationMethod_ != esm_EWALD_FULL) { |
309 |
|
selfMult1_ -= b0c; |
310 |
|
selfMult2_ += (db0c_2 + 2.0*db0c_1*ric) / 3.0; |
311 |
|
selfMult4_ -= (db0c_4 + 4.0*db0c_3*ric) / 15.0; |
768 |
|
Tb.zero(); // Torque on site b |
769 |
|
Ea.zero(); // Electric field at site a |
770 |
|
Eb.zero(); // Electric field at site b |
771 |
+ |
Pa = 0.0; // Site potential at site a |
772 |
+ |
Pb = 0.0; // Site potential at site b |
773 |
|
dUdCa = 0.0; // fluctuating charge force at site a |
774 |
|
dUdCb = 0.0; // fluctuating charge force at site a |
775 |
|
|
782 |
|
// Excluded potential that is still computed for fluctuating charges |
783 |
|
excluded_Pot= 0.0; |
784 |
|
|
771 |
– |
|
785 |
|
// some variables we'll need independent of electrostatic type: |
786 |
|
|
787 |
|
ri = 1.0 / *(idat.rij); |
844 |
|
if (idat.excluded) { |
845 |
|
*(idat.skippedCharge2) += C_a; |
846 |
|
} else { |
847 |
< |
// only do the field if we're not excluded: |
847 |
> |
// only do the field and site potentials if we're not excluded: |
848 |
|
Eb -= C_a * pre11_ * dv01 * rhat; |
849 |
+ |
Pb += C_a * pre11_ * v01; |
850 |
|
} |
851 |
|
} |
852 |
|
|
854 |
|
D_a = *(idat.dipole1); |
855 |
|
rdDa = dot(rhat, D_a); |
856 |
|
rxDa = cross(rhat, D_a); |
857 |
< |
if (!idat.excluded) |
857 |
> |
if (!idat.excluded) { |
858 |
|
Eb -= pre12_ * ((dv11-v11or) * rdDa * rhat + v11or * D_a); |
859 |
+ |
Pb += pre12_ * v11 * rdDa; |
860 |
+ |
} |
861 |
+ |
|
862 |
|
} |
863 |
|
|
864 |
|
if (a_is_Quadrupole) { |
868 |
|
rQa = rhat * Q_a; |
869 |
|
rdQar = dot(rhat, Qar); |
870 |
|
rxQar = cross(rhat, Qar); |
871 |
< |
if (!idat.excluded) |
871 |
> |
if (!idat.excluded) { |
872 |
|
Eb -= pre14_ * (trQa * rhat * dv21 + 2.0 * Qar * v22or |
873 |
|
+ rdQar * rhat * (dv22 - 2.0*v22or)); |
874 |
+ |
Pb += pre14_ * (v21 * trQa + v22 * rdQar); |
875 |
+ |
} |
876 |
|
} |
877 |
|
|
878 |
|
if (b_is_Charge) { |
886 |
|
} else { |
887 |
|
// only do the field if we're not excluded: |
888 |
|
Ea += C_b * pre11_ * dv01 * rhat; |
889 |
+ |
Pa += C_b * pre11_ * v01; |
890 |
+ |
|
891 |
|
} |
892 |
|
} |
893 |
|
|
895 |
|
D_b = *(idat.dipole2); |
896 |
|
rdDb = dot(rhat, D_b); |
897 |
|
rxDb = cross(rhat, D_b); |
898 |
< |
if (!idat.excluded) |
898 |
> |
if (!idat.excluded) { |
899 |
|
Ea += pre12_ * ((dv11-v11or) * rdDb * rhat + v11or * D_b); |
900 |
+ |
Pa += pre12_ * v11 * rdDb; |
901 |
+ |
} |
902 |
|
} |
903 |
|
|
904 |
|
if (b_is_Quadrupole) { |
908 |
|
rQb = rhat * Q_b; |
909 |
|
rdQbr = dot(rhat, Qbr); |
910 |
|
rxQbr = cross(rhat, Qbr); |
911 |
< |
if (!idat.excluded) |
911 |
> |
if (!idat.excluded) { |
912 |
|
Ea += pre14_ * (trQb * rhat * dv21 + 2.0 * Qbr * v22or |
913 |
|
+ rdQbr * rhat * (dv22 - 2.0*v22or)); |
914 |
+ |
Pa += pre14_ * (v21 * trQb + v22 * rdQbr); |
915 |
+ |
} |
916 |
|
} |
917 |
< |
|
917 |
> |
|
918 |
> |
|
919 |
|
if ((a_is_Fluctuating || b_is_Fluctuating) && idat.excluded) { |
920 |
|
J = Jij[FQtids[idat.atid1]][FQtids[idat.atid2]]; |
921 |
|
} |
922 |
< |
|
922 |
> |
|
923 |
|
if (a_is_Charge) { |
924 |
|
|
925 |
|
if (b_is_Charge) { |
926 |
|
pref = pre11_ * *(idat.electroMult); |
927 |
|
U += C_a * C_b * pref * v01; |
928 |
|
F += C_a * C_b * pref * dv01 * rhat; |
929 |
< |
|
929 |
> |
|
930 |
|
// If this is an excluded pair, there are still indirect |
931 |
|
// interactions via the reaction field we must worry about: |
932 |
|
|
935 |
|
indirect_Pot += rfContrib; |
936 |
|
indirect_F += rfContrib * 2.0 * ri * rhat; |
937 |
|
} |
938 |
< |
|
938 |
> |
|
939 |
|
// Fluctuating charge forces are handled via Coulomb integrals |
940 |
|
// for excluded pairs (i.e. those connected via bonds) and |
941 |
|
// with the standard charge-charge interaction otherwise. |
942 |
|
|
943 |
< |
if (idat.excluded) { |
943 |
> |
if (idat.excluded) { |
944 |
|
if (a_is_Fluctuating || b_is_Fluctuating) { |
945 |
|
coulInt = J->getValueAt( *(idat.rij) ); |
946 |
< |
if (a_is_Fluctuating) dUdCa += coulInt * C_b; |
947 |
< |
if (b_is_Fluctuating) dUdCb += coulInt * C_a; |
948 |
< |
excluded_Pot += C_a * C_b * coulInt; |
923 |
< |
} |
946 |
> |
if (a_is_Fluctuating) dUdCa += C_b * coulInt; |
947 |
> |
if (b_is_Fluctuating) dUdCb += C_a * coulInt; |
948 |
> |
} |
949 |
|
} else { |
950 |
|
if (a_is_Fluctuating) dUdCa += C_b * pref * v01; |
951 |
< |
if (a_is_Fluctuating) dUdCb += C_a * pref * v01; |
952 |
< |
} |
951 |
> |
if (b_is_Fluctuating) dUdCb += C_a * pref * v01; |
952 |
> |
} |
953 |
|
} |
954 |
|
|
955 |
|
if (b_is_Dipole) { |
1015 |
|
F -= pref * (rdDa * rdDb) * (dv22 - 2.0*v22or) * rhat; |
1016 |
|
Ta += pref * ( v21 * DaxDb - v22 * rdDb * rxDa); |
1017 |
|
Tb += pref * (-v21 * DaxDb - v22 * rdDa * rxDb); |
993 |
– |
|
1018 |
|
// Even if we excluded this pair from direct interactions, we |
1019 |
|
// still have the reaction-field-mediated dipole-dipole |
1020 |
|
// interaction: |
1074 |
|
trQaQb = QaQb.trace(); |
1075 |
|
rQaQb = rhat * QaQb; |
1076 |
|
QaQbr = QaQb * rhat; |
1077 |
< |
QaxQb = cross(Q_a, Q_b); |
1077 |
> |
QaxQb = mCross(Q_a, Q_b); |
1078 |
|
rQaQbr = dot(rQa, Qbr); |
1079 |
|
rQaxQbr = cross(rQa, Qbr); |
1080 |
|
|
1105 |
|
// + 4.0 * cross(rhat, QbQar) |
1106 |
|
|
1107 |
|
Tb += pref * 2.0 * cross(rhat,Qbr) * rdQar * v43; |
1084 |
– |
|
1108 |
|
} |
1109 |
|
} |
1110 |
|
|
1113 |
|
*(idat.eField2) += Eb * *(idat.electroMult); |
1114 |
|
} |
1115 |
|
|
1116 |
+ |
if (idat.doSitePotential) { |
1117 |
+ |
*(idat.sPot1) += Pa * *(idat.electroMult); |
1118 |
+ |
*(idat.sPot2) += Pb * *(idat.electroMult); |
1119 |
+ |
} |
1120 |
+ |
|
1121 |
|
if (a_is_Fluctuating) *(idat.dVdFQ1) += dUdCa * *(idat.sw); |
1122 |
|
if (b_is_Fluctuating) *(idat.dVdFQ2) += dUdCb * *(idat.sw); |
1123 |
|
|
1172 |
|
|
1173 |
|
if (i_is_Fluctuating) { |
1174 |
|
C_a += *(sdat.flucQ); |
1175 |
< |
// dVdFQ is really a force, so this is negative the derivative |
1176 |
< |
*(sdat.dVdFQ) -= *(sdat.flucQ) * data.hardness + data.electronegativity; |
1177 |
< |
(*(sdat.excludedPot))[ELECTROSTATIC_FAMILY] += (*sdat.flucQ) * |
1178 |
< |
(*(sdat.flucQ) * data.hardness * 0.5 + data.electronegativity); |
1175 |
> |
|
1176 |
> |
flucQ_->getSelfInteraction(sdat.atid, *(sdat.flucQ), |
1177 |
> |
(*(sdat.excludedPot))[ELECTROSTATIC_FAMILY], |
1178 |
> |
*(sdat.flucQfrc) ); |
1179 |
> |
|
1180 |
|
} |
1181 |
|
|
1182 |
|
switch (summationMethod_) { |
1227 |
|
} |
1228 |
|
|
1229 |
|
|
1230 |
< |
void Electrostatic::ReciprocalSpaceSum(potVec& pot) { |
1230 |
> |
void Electrostatic::ReciprocalSpaceSum(RealType& pot) { |
1231 |
|
|
1232 |
|
RealType kPot = 0.0; |
1233 |
|
RealType kVir = 0.0; |
1252 |
|
Vector3d box = hmat.diagonals(); |
1253 |
|
RealType boxMax = box.max(); |
1254 |
|
|
1226 |
– |
cerr << "da = " << dampingAlpha_ << " rc = " << cutoffRadius_ << "\n"; |
1227 |
– |
cerr << "boxMax = " << boxMax << "\n"; |
1255 |
|
//int kMax = int(2.0 * M_PI / (pow(dampingAlpha_,2)*cutoffRadius_ * boxMax) ); |
1256 |
< |
int kMax = 5; |
1230 |
< |
cerr << "kMax = " << kMax << "\n"; |
1256 |
> |
int kMax = 7; |
1257 |
|
int kSqMax = kMax*kMax + 2; |
1258 |
|
|
1259 |
|
int kLimit = kMax+1; |
1273 |
|
|
1274 |
|
// Calculate and store exponential factors |
1275 |
|
|
1276 |
< |
vector<vector<Vector3d> > eCos; |
1277 |
< |
vector<vector<Vector3d> > eSin; |
1276 |
> |
vector<vector<RealType> > elc; |
1277 |
> |
vector<vector<RealType> > emc; |
1278 |
> |
vector<vector<RealType> > enc; |
1279 |
> |
vector<vector<RealType> > els; |
1280 |
> |
vector<vector<RealType> > ems; |
1281 |
> |
vector<vector<RealType> > ens; |
1282 |
|
|
1283 |
|
int nMax = info_->getNAtoms(); |
1284 |
|
|
1285 |
< |
eCos.resize(kLimit+1); |
1286 |
< |
eSin.resize(kLimit+1); |
1285 |
> |
elc.resize(kLimit+1); |
1286 |
> |
emc.resize(kLimit+1); |
1287 |
> |
enc.resize(kLimit+1); |
1288 |
> |
els.resize(kLimit+1); |
1289 |
> |
ems.resize(kLimit+1); |
1290 |
> |
ens.resize(kLimit+1); |
1291 |
> |
|
1292 |
|
for (int j = 0; j < kLimit+1; j++) { |
1293 |
< |
eCos[j].resize(nMax); |
1294 |
< |
eSin[j].resize(nMax); |
1293 |
> |
elc[j].resize(nMax); |
1294 |
> |
emc[j].resize(nMax); |
1295 |
> |
enc[j].resize(nMax); |
1296 |
> |
els[j].resize(nMax); |
1297 |
> |
ems[j].resize(nMax); |
1298 |
> |
ens[j].resize(nMax); |
1299 |
|
} |
1300 |
|
|
1301 |
|
Vector3d t( 2.0 * M_PI ); |
1302 |
|
t.Vdiv(t, box); |
1303 |
< |
|
1303 |
> |
|
1304 |
|
SimInfo::MoleculeIterator mi; |
1305 |
|
Molecule::AtomIterator ai; |
1306 |
|
int i; |
1307 |
|
Vector3d r; |
1308 |
|
Vector3d tt; |
1270 |
– |
Vector3d w; |
1271 |
– |
Vector3d u; |
1272 |
– |
Vector3d a; |
1273 |
– |
Vector3d b; |
1309 |
|
|
1310 |
|
for (Molecule* mol = info_->beginMolecule(mi); mol != NULL; |
1311 |
|
mol = info_->nextMolecule(mi)) { |
1316 |
|
r = atom->getPos(); |
1317 |
|
info_->getSnapshotManager()->getCurrentSnapshot()->wrapVector(r); |
1318 |
|
|
1284 |
– |
// Shift so that all coordinates are in the range [0,L]: |
1285 |
– |
|
1286 |
– |
r += box/2.0; |
1287 |
– |
|
1319 |
|
tt.Vmul(t, r); |
1320 |
|
|
1321 |
< |
//cerr << "tt = " << tt << "\n"; |
1321 |
> |
elc[1][i] = 1.0; |
1322 |
> |
emc[1][i] = 1.0; |
1323 |
> |
enc[1][i] = 1.0; |
1324 |
> |
els[1][i] = 0.0; |
1325 |
> |
ems[1][i] = 0.0; |
1326 |
> |
ens[1][i] = 0.0; |
1327 |
> |
|
1328 |
> |
elc[2][i] = cos(tt.x()); |
1329 |
> |
emc[2][i] = cos(tt.y()); |
1330 |
> |
enc[2][i] = cos(tt.z()); |
1331 |
> |
els[2][i] = sin(tt.x()); |
1332 |
> |
ems[2][i] = sin(tt.y()); |
1333 |
> |
ens[2][i] = sin(tt.z()); |
1334 |
|
|
1292 |
– |
eCos[1][i] = Vector3d(1.0, 1.0, 1.0); |
1293 |
– |
eSin[1][i] = Vector3d(0.0, 0.0, 0.0); |
1294 |
– |
eCos[2][i] = Vector3d(cos(tt.x()), cos(tt.y()), cos(tt.z())); |
1295 |
– |
eSin[2][i] = Vector3d(sin(tt.x()), sin(tt.y()), sin(tt.z())); |
1296 |
– |
u = eCos[2][i]; |
1297 |
– |
w = eSin[2][i]; |
1298 |
– |
|
1335 |
|
for(int l = 3; l <= kLimit; l++) { |
1336 |
< |
a.Vmul(eCos[l-1][i], u); |
1337 |
< |
b.Vmul(eSin[l-1][i], w); |
1338 |
< |
eCos[l][i] = a - b; |
1339 |
< |
a.Vmul(eSin[l-1][i], u); |
1340 |
< |
b.Vmul(eCos[l-1][i], w); |
1341 |
< |
eSin[l][i] = a + b; |
1336 |
> |
elc[l][i]=elc[l-1][i]*elc[2][i]-els[l-1][i]*els[2][i]; |
1337 |
> |
emc[l][i]=emc[l-1][i]*emc[2][i]-ems[l-1][i]*ems[2][i]; |
1338 |
> |
enc[l][i]=enc[l-1][i]*enc[2][i]-ens[l-1][i]*ens[2][i]; |
1339 |
> |
els[l][i]=els[l-1][i]*elc[2][i]+elc[l-1][i]*els[2][i]; |
1340 |
> |
ems[l][i]=ems[l-1][i]*emc[2][i]+emc[l-1][i]*ems[2][i]; |
1341 |
> |
ens[l][i]=ens[l-1][i]*enc[2][i]+enc[l-1][i]*ens[2][i]; |
1342 |
|
} |
1343 |
|
} |
1344 |
|
} |
1382 |
|
std::vector<RealType> qks(nMax, 0.0); |
1383 |
|
std::vector<Vector3d> dxk(nMax, V3Zero); |
1384 |
|
std::vector<Vector3d> qxk(nMax, V3Zero); |
1385 |
< |
|
1385 |
> |
RealType rl, rm, rn; |
1386 |
> |
Vector3d kVec; |
1387 |
> |
Vector3d Qk; |
1388 |
> |
Mat3x3d k2; |
1389 |
> |
RealType ckcs, ckss, dkcs, dkss, qkcs, qkss; |
1390 |
> |
int atid; |
1391 |
> |
ElectrostaticAtomData data; |
1392 |
> |
RealType C, dk, qk; |
1393 |
> |
Vector3d D; |
1394 |
> |
Mat3x3d Q; |
1395 |
> |
|
1396 |
|
int mMin = kLimit; |
1397 |
|
int nMin = kLimit + 1; |
1398 |
|
for (int l = 1; l <= kLimit; l++) { |
1399 |
< |
int ll =l - 1; |
1400 |
< |
RealType rl = xcl * float(ll); |
1399 |
> |
int ll = l - 1; |
1400 |
> |
rl = xcl * float(ll); |
1401 |
|
for (int mmm = mMin; mmm <= kLim2; mmm++) { |
1402 |
|
int mm = mmm - kLimit; |
1403 |
|
int m = abs(mm) + 1; |
1404 |
< |
RealType rm = ycl * float(mm); |
1404 |
> |
rm = ycl * float(mm); |
1405 |
|
// Set temporary products of exponential terms |
1406 |
|
for (Molecule* mol = info_->beginMolecule(mi); mol != NULL; |
1407 |
|
mol = info_->nextMolecule(mi)) { |
1410 |
|
|
1411 |
|
i = atom->getLocalIndex(); |
1412 |
|
if(mm < 0) { |
1413 |
< |
clm[i] = eCos[l][i].x()*eCos[m][i].y() |
1414 |
< |
+ eSin[l][i].x()*eSin[m][i].y(); |
1369 |
< |
slm[i] = eSin[l][i].x()*eCos[m][i].y() |
1370 |
< |
- eSin[m][i].y()*eCos[l][i].x(); |
1413 |
> |
clm[i]=elc[l][i]*emc[m][i]+els[l][i]*ems[m][i]; |
1414 |
> |
slm[i]=els[l][i]*emc[m][i]-ems[m][i]*elc[l][i]; |
1415 |
|
} else { |
1416 |
< |
clm[i] = eCos[l][i].x()*eCos[m][i].y() |
1417 |
< |
- eSin[l][i].x()*eSin[m][i].y(); |
1374 |
< |
slm[i] = eSin[l][i].x()*eCos[m][i].y() |
1375 |
< |
+ eSin[m][i].y()*eCos[l][i].x(); |
1416 |
> |
clm[i]=elc[l][i]*emc[m][i]-els[l][i]*ems[m][i]; |
1417 |
> |
slm[i]=els[l][i]*emc[m][i]+ems[m][i]*elc[l][i]; |
1418 |
|
} |
1419 |
|
} |
1420 |
|
} |
1421 |
|
for (int nnn = nMin; nnn <= kLim2; nnn++) { |
1422 |
|
int nn = nnn - kLimit; |
1423 |
|
int n = abs(nn) + 1; |
1424 |
< |
RealType rn = zcl * float(nn); |
1424 |
> |
rn = zcl * float(nn); |
1425 |
|
// Test on magnitude of k vector: |
1426 |
|
int kk=ll*ll + mm*mm + nn*nn; |
1427 |
|
if(kk <= kSqLim) { |
1428 |
< |
Vector3d kVec = Vector3d(rl, rm, rn); |
1429 |
< |
Mat3x3d k2 = outProduct(kVec, kVec); |
1428 |
> |
kVec = Vector3d(rl, rm, rn); |
1429 |
> |
k2 = outProduct(kVec, kVec); |
1430 |
|
// Calculate exp(ikr) terms |
1431 |
|
for (Molecule* mol = info_->beginMolecule(mi); mol != NULL; |
1432 |
|
mol = info_->nextMolecule(mi)) { |
1435 |
|
i = atom->getLocalIndex(); |
1436 |
|
|
1437 |
|
if (nn < 0) { |
1438 |
< |
ckr[i]=clm[i]*eCos[n][i].z()+slm[i]*eSin[n][i].z(); |
1439 |
< |
skr[i]=slm[i]*eCos[n][i].z()-clm[i]*eSin[n][i].z(); |
1438 |
> |
ckr[i]=clm[i]*enc[n][i]+slm[i]*ens[n][i]; |
1439 |
> |
skr[i]=slm[i]*enc[n][i]-clm[i]*ens[n][i]; |
1440 |
> |
|
1441 |
|
} else { |
1442 |
< |
ckr[i]=clm[i]*eCos[n][i].z()-slm[i]*eSin[n][i].z(); |
1443 |
< |
skr[i]=slm[i]*eCos[n][i].z()+clm[i]*eSin[n][i].z(); |
1442 |
> |
ckr[i]=clm[i]*enc[n][i]-slm[i]*ens[n][i]; |
1443 |
> |
skr[i]=slm[i]*enc[n][i]+clm[i]*ens[n][i]; |
1444 |
|
} |
1445 |
|
} |
1446 |
|
} |
1453 |
|
atom = mol->nextAtom(ai)) { |
1454 |
|
i = atom->getLocalIndex(); |
1455 |
|
int atid = atom->getAtomType()->getIdent(); |
1456 |
< |
ElectrostaticAtomData data = ElectrostaticMap[Etids[atid]]; |
1456 |
> |
data = ElectrostaticMap[Etids[atid]]; |
1457 |
|
|
1458 |
|
if (data.is_Charge) { |
1459 |
< |
RealType C = data.fixedCharge; |
1459 |
> |
C = data.fixedCharge; |
1460 |
|
if (atom->isFluctuatingCharge()) C += atom->getFlucQPos(); |
1461 |
|
ckc[i] = C * ckr[i]; |
1462 |
|
cks[i] = C * skr[i]; |
1463 |
|
} |
1464 |
|
|
1465 |
|
if (data.is_Dipole) { |
1466 |
< |
Vector3d D = atom->getDipole() * mPoleConverter; |
1467 |
< |
RealType dk = dot(kVec, D); |
1468 |
< |
dxk[i] = cross(kVec, D); |
1466 |
> |
D = atom->getDipole() * mPoleConverter; |
1467 |
> |
dk = dot(D, kVec); |
1468 |
> |
dxk[i] = cross(D, kVec); |
1469 |
|
dkc[i] = dk * ckr[i]; |
1470 |
|
dks[i] = dk * skr[i]; |
1471 |
|
} |
1472 |
|
if (data.is_Quadrupole) { |
1473 |
< |
Mat3x3d Q = atom->getQuadrupole(); |
1474 |
< |
Q *= mPoleConverter; |
1475 |
< |
RealType qk = -( Q * k2 ).trace(); |
1476 |
< |
qxk[i] = -2.0 * cross(k2, Q); |
1473 |
> |
Q = atom->getQuadrupole() * mPoleConverter; |
1474 |
> |
Qk = Q * kVec; |
1475 |
> |
qk = dot(kVec, Qk); |
1476 |
> |
qxk[i] = -cross(kVec, Qk); |
1477 |
|
qkc[i] = qk * ckr[i]; |
1478 |
|
qks[i] = qk * skr[i]; |
1479 |
|
} |
1482 |
|
|
1483 |
|
// calculate vector sums |
1484 |
|
|
1485 |
< |
RealType ckcs = std::accumulate(ckc.begin(),ckc.end(),0.0); |
1486 |
< |
RealType ckss = std::accumulate(cks.begin(),cks.end(),0.0); |
1487 |
< |
RealType dkcs = std::accumulate(dkc.begin(),dkc.end(),0.0); |
1488 |
< |
RealType dkss = std::accumulate(dks.begin(),dks.end(),0.0); |
1489 |
< |
RealType qkcs = std::accumulate(qkc.begin(),qkc.end(),0.0); |
1490 |
< |
RealType qkss = std::accumulate(qks.begin(),qks.end(),0.0); |
1448 |
< |
|
1485 |
> |
ckcs = std::accumulate(ckc.begin(),ckc.end(),0.0); |
1486 |
> |
ckss = std::accumulate(cks.begin(),cks.end(),0.0); |
1487 |
> |
dkcs = std::accumulate(dkc.begin(),dkc.end(),0.0); |
1488 |
> |
dkss = std::accumulate(dks.begin(),dks.end(),0.0); |
1489 |
> |
qkcs = std::accumulate(qkc.begin(),qkc.end(),0.0); |
1490 |
> |
qkss = std::accumulate(qks.begin(),qks.end(),0.0); |
1491 |
|
|
1492 |
|
#ifdef IS_MPI |
1493 |
< |
MPI::COMM_WORLD.Allreduce(MPI::IN_PLACE, &ckcs, 1, MPI::REALTYPE, |
1494 |
< |
MPI::SUM); |
1495 |
< |
MPI::COMM_WORLD.Allreduce(MPI::IN_PLACE, &ckss, 1, MPI::REALTYPE, |
1496 |
< |
MPI::SUM); |
1497 |
< |
MPI::COMM_WORLD.Allreduce(MPI::IN_PLACE, &dkcs, 1, MPI::REALTYPE, |
1498 |
< |
MPI::SUM); |
1499 |
< |
MPI::COMM_WORLD.Allreduce(MPI::IN_PLACE, &dkss, 1, MPI::REALTYPE, |
1500 |
< |
MPI::SUM); |
1501 |
< |
MPI::COMM_WORLD.Allreduce(MPI::IN_PLACE, &qkcs, 1, MPI::REALTYPE, |
1502 |
< |
MPI::SUM); |
1503 |
< |
MPI::COMM_WORLD.Allreduce(MPI::IN_PLACE, &qkss, 1, MPI::REALTYPE, |
1504 |
< |
MPI::SUM); |
1493 |
> |
MPI_Allreduce(MPI_IN_PLACE, &ckcs, 1, MPI_REALTYPE, |
1494 |
> |
MPI_SUM, MPI_COMM_WORLD); |
1495 |
> |
MPI_Allreduce(MPI_IN_PLACE, &ckss, 1, MPI_REALTYPE, |
1496 |
> |
MPI_SUM, MPI_COMM_WORLD); |
1497 |
> |
MPI_Allreduce(MPI_IN_PLACE, &dkcs, 1, MPI_REALTYPE, |
1498 |
> |
MPI_SUM, MPI_COMM_WORLD); |
1499 |
> |
MPI_Allreduce(MPI_IN_PLACE, &dkss, 1, MPI_REALTYPE, |
1500 |
> |
MPI_SUM, MPI_COMM_WORLD); |
1501 |
> |
MPI_Allreduce(MPI_IN_PLACE, &qkcs, 1, MPI_REALTYPE, |
1502 |
> |
MPI_SUM, MPI_COMM_WORLD); |
1503 |
> |
MPI_Allreduce(MPI_IN_PLACE, &qkss, 1, MPI_REALTYPE, |
1504 |
> |
MPI_SUM, MPI_COMM_WORLD); |
1505 |
|
#endif |
1506 |
|
|
1507 |
< |
// Accumulate potential energy and virial contribution: |
1507 |
> |
// Accumulate potential energy and virial contribution: |
1508 |
|
|
1509 |
< |
kPot += 2.0 * rvol * AK[kk]*((ckss+dkcs-qkss)*(ckss+dkcs-qkss) |
1510 |
< |
+ (ckcs-dkss-qkcs)*(ckcs-dkss-qkss)); |
1509 |
> |
kPot += 2.0 * rvol * AK[kk]*((ckss+dkcs-qkss)*(ckss+dkcs-qkss) |
1510 |
> |
+ (ckcs-dkss-qkcs)*(ckcs-dkss-qkcs)); |
1511 |
|
|
1512 |
< |
kVir -= 2.0 * rvol * AK[kk]*(ckcs*ckcs+ckss*ckss |
1513 |
< |
+4.0*(ckss*dkcs-ckcs*dkss) |
1514 |
< |
+3.0*(dkcs*dkcs+dkss*dkss) |
1515 |
< |
-6.0*(ckss*qkss+ckcs*qkcs) |
1516 |
< |
+8.0*(dkss*qkcs-dkcs*qkss) |
1517 |
< |
+5.0*(qkss*qkss+qkcs*qkcs)); |
1512 |
> |
kVir += 2.0 * rvol * AK[kk]*(ckcs*ckcs+ckss*ckss |
1513 |
> |
+4.0*(ckss*dkcs-ckcs*dkss) |
1514 |
> |
+3.0*(dkcs*dkcs+dkss*dkss) |
1515 |
> |
-6.0*(ckss*qkss+ckcs*qkcs) |
1516 |
> |
+8.0*(dkss*qkcs-dkcs*qkss) |
1517 |
> |
+5.0*(qkss*qkss+qkcs*qkcs)); |
1518 |
|
|
1519 |
|
// Calculate force and torque for each site: |
1520 |
|
|
1524 |
|
atom = mol->nextAtom(ai)) { |
1525 |
|
|
1526 |
|
i = atom->getLocalIndex(); |
1527 |
< |
int atid = atom->getAtomType()->getIdent(); |
1528 |
< |
ElectrostaticAtomData data = ElectrostaticMap[Etids[atid]]; |
1529 |
< |
|
1527 |
> |
atid = atom->getAtomType()->getIdent(); |
1528 |
> |
data = ElectrostaticMap[Etids[atid]]; |
1529 |
> |
|
1530 |
|
RealType qfrc = AK[kk]*((cks[i]+dkc[i]-qks[i])*(ckcs-dkss-qkcs) |
1531 |
< |
- (ckc[i]-dks[i]-qkc[i])*(ckss+dkcs-qkss)); |
1531 |
> |
- (ckc[i]-dks[i]-qkc[i])*(ckss+dkcs-qkss)); |
1532 |
|
RealType qtrq1 = AK[kk]*(skr[i]*(ckcs-dkss-qkcs) |
1533 |
|
-ckr[i]*(ckss+dkcs-qkss)); |
1534 |
< |
RealType qtrq2 = 2.0*AK[kk]*(ckr[i]*(ckcs-dkss-qkcs)+ |
1535 |
< |
skr[i]*(ckss+dkcs-qkss)); |
1536 |
< |
|
1495 |
< |
|
1534 |
> |
RealType qtrq2 = 2.0*AK[kk]*(ckr[i]*(ckcs-dkss-qkcs) |
1535 |
> |
+skr[i]*(ckss+dkcs-qkss)); |
1536 |
> |
|
1537 |
|
atom->addFrc( 4.0 * rvol * qfrc * kVec ); |
1538 |
< |
|
1538 |
> |
|
1539 |
> |
if (atom->isFluctuatingCharge()) { |
1540 |
> |
atom->addFlucQFrc( - 2.0 * rvol * qtrq2 ); |
1541 |
> |
} |
1542 |
> |
|
1543 |
|
if (data.is_Dipole) { |
1544 |
|
atom->addTrq( 4.0 * rvol * qtrq1 * dxk[i] ); |
1545 |
|
} |
1550 |
|
} |
1551 |
|
} |
1552 |
|
} |
1553 |
+ |
nMin = 1; |
1554 |
|
} |
1555 |
+ |
mMin = 1; |
1556 |
|
} |
1557 |
< |
cerr << "kPot = " << kPot << "\n"; |
1511 |
< |
pot[ELECTROSTATIC_FAMILY] += kPot; |
1557 |
> |
pot += kPot; |
1558 |
|
} |
1559 |
|
} |