| 1 |
gezelter |
2 |
#include <math.h> |
| 2 |
|
|
|
| 3 |
tim |
3 |
#include "primitives/Atom.hpp" |
| 4 |
|
|
#include "primitives/SRI.hpp" |
| 5 |
|
|
#include "primitives/AbstractClasses.hpp" |
| 6 |
|
|
#include "brains/SimInfo.hpp" |
| 7 |
|
|
#include "UseTheForce/ForceFields.hpp" |
| 8 |
|
|
#include "brains/Thermo.hpp" |
| 9 |
|
|
#include "io/ReadWrite.hpp" |
| 10 |
|
|
#include "integrators/Integrator.hpp" |
| 11 |
|
|
#include "utils/simError.h" |
| 12 |
gezelter |
2 |
|
| 13 |
|
|
#ifdef IS_MPI |
| 14 |
tim |
3 |
#include "brains/mpiSimulation.hpp" |
| 15 |
gezelter |
2 |
#endif |
| 16 |
|
|
|
| 17 |
|
|
|
| 18 |
|
|
// Basic isotropic thermostating and barostating via the Melchionna |
| 19 |
|
|
// modification of the Hoover algorithm: |
| 20 |
|
|
// |
| 21 |
|
|
// Melchionna, S., Ciccotti, G., and Holian, B. L., 1993, |
| 22 |
|
|
// Molec. Phys., 78, 533. |
| 23 |
|
|
// |
| 24 |
|
|
// and |
| 25 |
|
|
// |
| 26 |
|
|
// Hoover, W. G., 1986, Phys. Rev. A, 34, 2499. |
| 27 |
|
|
|
| 28 |
|
|
template<typename T> NPT<T>::NPT ( SimInfo *theInfo, ForceFields* the_ff): |
| 29 |
|
|
T( theInfo, the_ff ) |
| 30 |
|
|
{ |
| 31 |
|
|
GenericData* data; |
| 32 |
tim |
132 |
DoubleGenericData * chiValue; |
| 33 |
|
|
DoubleGenericData * integralOfChidtValue; |
| 34 |
gezelter |
2 |
|
| 35 |
|
|
chiValue = NULL; |
| 36 |
|
|
integralOfChidtValue = NULL; |
| 37 |
|
|
|
| 38 |
|
|
chi = 0.0; |
| 39 |
|
|
integralOfChidt = 0.0; |
| 40 |
|
|
have_tau_thermostat = 0; |
| 41 |
|
|
have_tau_barostat = 0; |
| 42 |
|
|
have_target_temp = 0; |
| 43 |
|
|
have_target_pressure = 0; |
| 44 |
|
|
have_chi_tolerance = 0; |
| 45 |
|
|
have_eta_tolerance = 0; |
| 46 |
|
|
have_pos_iter_tolerance = 0; |
| 47 |
|
|
|
| 48 |
|
|
// retrieve chi and integralOfChidt from simInfo |
| 49 |
|
|
data = info->getProperty(CHIVALUE_ID); |
| 50 |
|
|
if(data){ |
| 51 |
tim |
132 |
chiValue = dynamic_cast<DoubleGenericData*>(data); |
| 52 |
gezelter |
2 |
} |
| 53 |
|
|
|
| 54 |
|
|
data = info->getProperty(INTEGRALOFCHIDT_ID); |
| 55 |
|
|
if(data){ |
| 56 |
tim |
132 |
integralOfChidtValue = dynamic_cast<DoubleGenericData*>(data); |
| 57 |
gezelter |
2 |
} |
| 58 |
|
|
|
| 59 |
|
|
// chi and integralOfChidt should appear by pair |
| 60 |
|
|
if(chiValue && integralOfChidtValue){ |
| 61 |
|
|
chi = chiValue->getData(); |
| 62 |
|
|
integralOfChidt = integralOfChidtValue->getData(); |
| 63 |
|
|
} |
| 64 |
|
|
|
| 65 |
|
|
oldPos = new double[3*integrableObjects.size()]; |
| 66 |
|
|
oldVel = new double[3*integrableObjects.size()]; |
| 67 |
|
|
oldJi = new double[3*integrableObjects.size()]; |
| 68 |
|
|
|
| 69 |
|
|
} |
| 70 |
|
|
|
| 71 |
|
|
template<typename T> NPT<T>::~NPT() { |
| 72 |
|
|
delete[] oldPos; |
| 73 |
|
|
delete[] oldVel; |
| 74 |
|
|
delete[] oldJi; |
| 75 |
|
|
} |
| 76 |
|
|
|
| 77 |
|
|
template<typename T> void NPT<T>::moveA() { |
| 78 |
|
|
|
| 79 |
|
|
//new version of NPT |
| 80 |
|
|
int i, j, k; |
| 81 |
|
|
double Tb[3], ji[3]; |
| 82 |
|
|
double mass; |
| 83 |
|
|
double vel[3], pos[3], frc[3]; |
| 84 |
|
|
double sc[3]; |
| 85 |
|
|
double COM[3]; |
| 86 |
|
|
|
| 87 |
|
|
instaTemp = tStats->getTemperature(); |
| 88 |
|
|
tStats->getPressureTensor( press ); |
| 89 |
|
|
instaPress = p_convert * (press[0][0] + press[1][1] + press[2][2]) / 3.0; |
| 90 |
|
|
instaVol = tStats->getVolume(); |
| 91 |
|
|
|
| 92 |
|
|
tStats->getCOM(COM); |
| 93 |
|
|
|
| 94 |
|
|
//evolve velocity half step |
| 95 |
|
|
|
| 96 |
|
|
calcVelScale(); |
| 97 |
|
|
|
| 98 |
|
|
for( i=0; i<integrableObjects.size(); i++ ){ |
| 99 |
|
|
|
| 100 |
|
|
integrableObjects[i]->getVel( vel ); |
| 101 |
|
|
integrableObjects[i]->getFrc( frc ); |
| 102 |
|
|
|
| 103 |
|
|
mass = integrableObjects[i]->getMass(); |
| 104 |
|
|
|
| 105 |
|
|
getVelScaleA( sc, vel ); |
| 106 |
|
|
|
| 107 |
|
|
for (j=0; j < 3; j++) { |
| 108 |
|
|
|
| 109 |
|
|
// velocity half step (use chi from previous step here): |
| 110 |
|
|
vel[j] += dt2 * ((frc[j] / mass ) * eConvert - sc[j]); |
| 111 |
|
|
|
| 112 |
|
|
} |
| 113 |
|
|
|
| 114 |
|
|
integrableObjects[i]->setVel( vel ); |
| 115 |
|
|
|
| 116 |
|
|
if( integrableObjects[i]->isDirectional() ){ |
| 117 |
|
|
|
| 118 |
|
|
// get and convert the torque to body frame |
| 119 |
|
|
|
| 120 |
|
|
integrableObjects[i]->getTrq( Tb ); |
| 121 |
|
|
integrableObjects[i]->lab2Body( Tb ); |
| 122 |
|
|
|
| 123 |
|
|
// get the angular momentum, and propagate a half step |
| 124 |
|
|
|
| 125 |
|
|
integrableObjects[i]->getJ( ji ); |
| 126 |
|
|
|
| 127 |
|
|
for (j=0; j < 3; j++) |
| 128 |
|
|
ji[j] += dt2 * (Tb[j] * eConvert - ji[j]*chi); |
| 129 |
|
|
|
| 130 |
|
|
this->rotationPropagation( integrableObjects[i], ji ); |
| 131 |
|
|
|
| 132 |
|
|
integrableObjects[i]->setJ( ji ); |
| 133 |
|
|
} |
| 134 |
|
|
} |
| 135 |
|
|
|
| 136 |
|
|
// evolve chi and eta half step |
| 137 |
|
|
|
| 138 |
|
|
evolveChiA(); |
| 139 |
|
|
evolveEtaA(); |
| 140 |
|
|
|
| 141 |
|
|
//calculate the integral of chidt |
| 142 |
|
|
integralOfChidt += dt2*chi; |
| 143 |
|
|
|
| 144 |
|
|
//save the old positions |
| 145 |
|
|
for(i = 0; i < integrableObjects.size(); i++){ |
| 146 |
|
|
integrableObjects[i]->getPos(pos); |
| 147 |
|
|
for(j = 0; j < 3; j++) |
| 148 |
|
|
oldPos[i*3 + j] = pos[j]; |
| 149 |
|
|
} |
| 150 |
|
|
|
| 151 |
|
|
//the first estimation of r(t+dt) is equal to r(t) |
| 152 |
|
|
|
| 153 |
|
|
for(k = 0; k < 5; k ++){ |
| 154 |
|
|
|
| 155 |
|
|
for(i =0 ; i < integrableObjects.size(); i++){ |
| 156 |
|
|
|
| 157 |
|
|
integrableObjects[i]->getVel(vel); |
| 158 |
|
|
integrableObjects[i]->getPos(pos); |
| 159 |
|
|
|
| 160 |
|
|
this->getPosScale( pos, COM, i, sc ); |
| 161 |
|
|
|
| 162 |
|
|
for(j = 0; j < 3; j++) |
| 163 |
|
|
pos[j] = oldPos[i*3 + j] + dt*(vel[j] + sc[j]); |
| 164 |
|
|
|
| 165 |
|
|
integrableObjects[i]->setPos( pos ); |
| 166 |
|
|
} |
| 167 |
|
|
|
| 168 |
|
|
if(nConstrained) |
| 169 |
|
|
constrainA(); |
| 170 |
|
|
} |
| 171 |
|
|
|
| 172 |
|
|
|
| 173 |
|
|
// Scale the box after all the positions have been moved: |
| 174 |
|
|
|
| 175 |
|
|
this->scaleSimBox(); |
| 176 |
|
|
} |
| 177 |
|
|
|
| 178 |
|
|
template<typename T> void NPT<T>::moveB( void ){ |
| 179 |
|
|
|
| 180 |
|
|
//new version of NPT |
| 181 |
|
|
int i, j, k; |
| 182 |
|
|
double Tb[3], ji[3], sc[3]; |
| 183 |
|
|
double vel[3], frc[3]; |
| 184 |
|
|
double mass; |
| 185 |
|
|
|
| 186 |
|
|
// Set things up for the iteration: |
| 187 |
|
|
|
| 188 |
|
|
for( i=0; i<integrableObjects.size(); i++ ){ |
| 189 |
|
|
|
| 190 |
|
|
integrableObjects[i]->getVel( vel ); |
| 191 |
|
|
|
| 192 |
|
|
for (j=0; j < 3; j++) |
| 193 |
|
|
oldVel[3*i + j] = vel[j]; |
| 194 |
|
|
|
| 195 |
|
|
if( integrableObjects[i]->isDirectional() ){ |
| 196 |
|
|
|
| 197 |
|
|
integrableObjects[i]->getJ( ji ); |
| 198 |
|
|
|
| 199 |
|
|
for (j=0; j < 3; j++) |
| 200 |
|
|
oldJi[3*i + j] = ji[j]; |
| 201 |
|
|
|
| 202 |
|
|
} |
| 203 |
|
|
} |
| 204 |
|
|
|
| 205 |
|
|
// do the iteration: |
| 206 |
|
|
|
| 207 |
|
|
instaVol = tStats->getVolume(); |
| 208 |
|
|
|
| 209 |
|
|
for (k=0; k < 4; k++) { |
| 210 |
|
|
|
| 211 |
|
|
instaTemp = tStats->getTemperature(); |
| 212 |
|
|
instaPress = tStats->getPressure(); |
| 213 |
|
|
|
| 214 |
|
|
// evolve chi another half step using the temperature at t + dt/2 |
| 215 |
|
|
|
| 216 |
|
|
this->evolveChiB(); |
| 217 |
|
|
this->evolveEtaB(); |
| 218 |
|
|
this->calcVelScale(); |
| 219 |
|
|
|
| 220 |
|
|
for( i=0; i<integrableObjects.size(); i++ ){ |
| 221 |
|
|
|
| 222 |
|
|
integrableObjects[i]->getFrc( frc ); |
| 223 |
|
|
integrableObjects[i]->getVel(vel); |
| 224 |
|
|
|
| 225 |
|
|
mass = integrableObjects[i]->getMass(); |
| 226 |
|
|
|
| 227 |
|
|
getVelScaleB( sc, i ); |
| 228 |
|
|
|
| 229 |
|
|
// velocity half step |
| 230 |
|
|
for (j=0; j < 3; j++) |
| 231 |
|
|
vel[j] = oldVel[3*i+j] + dt2 * ((frc[j] / mass ) * eConvert - sc[j]); |
| 232 |
|
|
|
| 233 |
|
|
integrableObjects[i]->setVel( vel ); |
| 234 |
|
|
|
| 235 |
|
|
if( integrableObjects[i]->isDirectional() ){ |
| 236 |
|
|
|
| 237 |
|
|
// get and convert the torque to body frame |
| 238 |
|
|
|
| 239 |
|
|
integrableObjects[i]->getTrq( Tb ); |
| 240 |
|
|
integrableObjects[i]->lab2Body( Tb ); |
| 241 |
|
|
|
| 242 |
|
|
for (j=0; j < 3; j++) |
| 243 |
|
|
ji[j] = oldJi[3*i + j] + dt2 * (Tb[j] * eConvert - oldJi[3*i+j]*chi); |
| 244 |
|
|
|
| 245 |
|
|
integrableObjects[i]->setJ( ji ); |
| 246 |
|
|
} |
| 247 |
|
|
} |
| 248 |
|
|
|
| 249 |
|
|
if(nConstrained) |
| 250 |
|
|
constrainB(); |
| 251 |
|
|
|
| 252 |
|
|
if ( this->chiConverged() && this->etaConverged() ) break; |
| 253 |
|
|
} |
| 254 |
|
|
|
| 255 |
|
|
//calculate integral of chida |
| 256 |
|
|
integralOfChidt += dt2*chi; |
| 257 |
|
|
|
| 258 |
|
|
|
| 259 |
|
|
} |
| 260 |
|
|
|
| 261 |
|
|
template<typename T> void NPT<T>::resetIntegrator() { |
| 262 |
|
|
chi = 0.0; |
| 263 |
|
|
T::resetIntegrator(); |
| 264 |
|
|
} |
| 265 |
|
|
|
| 266 |
|
|
template<typename T> void NPT<T>::evolveChiA() { |
| 267 |
|
|
chi += dt2 * ( instaTemp / targetTemp - 1.0) / tt2; |
| 268 |
|
|
oldChi = chi; |
| 269 |
|
|
} |
| 270 |
|
|
|
| 271 |
|
|
template<typename T> void NPT<T>::evolveChiB() { |
| 272 |
|
|
|
| 273 |
|
|
prevChi = chi; |
| 274 |
|
|
chi = oldChi + dt2 * ( instaTemp / targetTemp - 1.0) / tt2; |
| 275 |
|
|
} |
| 276 |
|
|
|
| 277 |
|
|
template<typename T> bool NPT<T>::chiConverged() { |
| 278 |
|
|
|
| 279 |
|
|
return ( fabs( prevChi - chi ) <= chiTolerance ); |
| 280 |
|
|
} |
| 281 |
|
|
|
| 282 |
|
|
template<typename T> int NPT<T>::readyCheck() { |
| 283 |
|
|
|
| 284 |
|
|
//check parent's readyCheck() first |
| 285 |
|
|
if (T::readyCheck() == -1) |
| 286 |
|
|
return -1; |
| 287 |
|
|
|
| 288 |
|
|
// First check to see if we have a target temperature. |
| 289 |
|
|
// Not having one is fatal. |
| 290 |
|
|
|
| 291 |
|
|
if (!have_target_temp) { |
| 292 |
|
|
sprintf( painCave.errMsg, |
| 293 |
|
|
"NPT error: You can't use the NPT integrator\n" |
| 294 |
|
|
" without a targetTemp!\n" |
| 295 |
|
|
); |
| 296 |
|
|
painCave.isFatal = 1; |
| 297 |
|
|
simError(); |
| 298 |
|
|
return -1; |
| 299 |
|
|
} |
| 300 |
|
|
|
| 301 |
|
|
if (!have_target_pressure) { |
| 302 |
|
|
sprintf( painCave.errMsg, |
| 303 |
|
|
"NPT error: You can't use the NPT integrator\n" |
| 304 |
|
|
" without a targetPressure!\n" |
| 305 |
|
|
); |
| 306 |
|
|
painCave.isFatal = 1; |
| 307 |
|
|
simError(); |
| 308 |
|
|
return -1; |
| 309 |
|
|
} |
| 310 |
|
|
|
| 311 |
|
|
// We must set tauThermostat. |
| 312 |
|
|
|
| 313 |
|
|
if (!have_tau_thermostat) { |
| 314 |
|
|
sprintf( painCave.errMsg, |
| 315 |
|
|
"NPT error: If you use the NPT\n" |
| 316 |
|
|
" integrator, you must set tauThermostat.\n"); |
| 317 |
|
|
painCave.isFatal = 1; |
| 318 |
|
|
simError(); |
| 319 |
|
|
return -1; |
| 320 |
|
|
} |
| 321 |
|
|
|
| 322 |
|
|
// We must set tauBarostat. |
| 323 |
|
|
|
| 324 |
|
|
if (!have_tau_barostat) { |
| 325 |
|
|
sprintf( painCave.errMsg, |
| 326 |
|
|
"If you use the NPT integrator, you must set tauBarostat.\n"); |
| 327 |
|
|
painCave.severity = OOPSE_ERROR; |
| 328 |
|
|
painCave.isFatal = 1; |
| 329 |
|
|
simError(); |
| 330 |
|
|
return -1; |
| 331 |
|
|
} |
| 332 |
|
|
|
| 333 |
|
|
if (!have_chi_tolerance) { |
| 334 |
|
|
sprintf( painCave.errMsg, |
| 335 |
|
|
"Setting chi tolerance to 1e-6 in NPT integrator\n"); |
| 336 |
|
|
chiTolerance = 1e-6; |
| 337 |
|
|
have_chi_tolerance = 1; |
| 338 |
|
|
painCave.severity = OOPSE_INFO; |
| 339 |
|
|
painCave.isFatal = 0; |
| 340 |
|
|
simError(); |
| 341 |
|
|
} |
| 342 |
|
|
|
| 343 |
|
|
if (!have_eta_tolerance) { |
| 344 |
|
|
sprintf( painCave.errMsg, |
| 345 |
|
|
"Setting eta tolerance to 1e-6 in NPT integrator"); |
| 346 |
|
|
etaTolerance = 1e-6; |
| 347 |
|
|
have_eta_tolerance = 1; |
| 348 |
|
|
painCave.severity = OOPSE_INFO; |
| 349 |
|
|
painCave.isFatal = 0; |
| 350 |
|
|
simError(); |
| 351 |
|
|
} |
| 352 |
|
|
|
| 353 |
|
|
// We need NkBT a lot, so just set it here: This is the RAW number |
| 354 |
|
|
// of integrableObjects, so no subtraction or addition of constraints or |
| 355 |
|
|
// orientational degrees of freedom: |
| 356 |
|
|
|
| 357 |
|
|
NkBT = (double)(info->getTotIntegrableObjects()) * kB * targetTemp; |
| 358 |
|
|
|
| 359 |
|
|
// fkBT is used because the thermostat operates on more degrees of freedom |
| 360 |
|
|
// than the barostat (when there are particles with orientational degrees |
| 361 |
|
|
// of freedom). |
| 362 |
|
|
|
| 363 |
|
|
fkBT = (double)(info->getNDF()) * kB * targetTemp; |
| 364 |
|
|
|
| 365 |
|
|
tt2 = tauThermostat * tauThermostat; |
| 366 |
|
|
tb2 = tauBarostat * tauBarostat; |
| 367 |
|
|
|
| 368 |
|
|
return 1; |
| 369 |
|
|
} |