# | Line 1 | Line 1 | |
---|---|---|
1 | #include <iostream> | |
2 | < | #include <cstdlib> |
3 | < | #include <cmath> |
4 | < | |
2 | > | #include <stdlib.h> |
3 | > | #include <math.h> |
4 | > | #include "Rattle.hpp" |
5 | > | #include "Roll.hpp" |
6 | #ifdef IS_MPI | |
7 | #include "mpiSimulation.hpp" | |
8 | #include <unistd.h> | |
9 | #endif //is_mpi | |
10 | ||
11 | + | #ifdef PROFILE |
12 | + | #include "mdProfile.hpp" |
13 | + | #endif // profile |
14 | + | |
15 | #include "Integrator.hpp" | |
16 | #include "simError.h" | |
17 | ||
18 | ||
19 | < | Integrator::Integrator( SimInfo *theInfo, ForceFields* the_ff ){ |
20 | < | |
19 | > | template<typename T> Integrator<T>::Integrator(SimInfo* theInfo, |
20 | > | ForceFields* the_ff){ |
21 | info = theInfo; | |
22 | myFF = the_ff; | |
23 | isFirst = 1; | |
# | Line 21 | Line 26 | Integrator::Integrator( SimInfo *theInfo, ForceFields* | |
26 | nMols = info->n_mol; | |
27 | ||
28 | // give a little love back to the SimInfo object | |
24 | – | |
25 | – | if( info->the_integrator != NULL ) delete info->the_integrator; |
26 | – | info->the_integrator = this; |
29 | ||
30 | + | if (info->the_integrator != NULL){ |
31 | + | delete info->the_integrator; |
32 | + | } |
33 | + | |
34 | nAtoms = info->n_atoms; | |
35 | + | integrableObjects = info->integrableObjects; |
36 | ||
37 | < | std::cerr << "integ nAtoms = " << nAtoms << "\n"; |
37 | > | consFramework = new RollFramework(info); |
38 | ||
39 | < | // check for constraints |
39 | > | if(consFramework == NULL){ |
40 | > | sprintf(painCave.errMsg, |
41 | > | "Integrator::Intergrator() Error: Memory allocation error for RattleFramework" ); |
42 | > | painCave.isFatal = 1; |
43 | > | simError(); |
44 | > | } |
45 | ||
46 | < | constrainedA = NULL; |
47 | < | constrainedB = NULL; |
46 | > | /* |
47 | > | // check for constraints |
48 | > | |
49 | > | constrainedA = NULL; |
50 | > | constrainedB = NULL; |
51 | constrainedDsqr = NULL; | |
52 | < | moving = NULL; |
53 | < | moved = NULL; |
54 | < | oldPos = NULL; |
55 | < | |
52 | > | moving = NULL; |
53 | > | moved = NULL; |
54 | > | oldPos = NULL; |
55 | > | |
56 | nConstrained = 0; | |
57 | ||
58 | checkConstraints(); | |
59 | + | */ |
60 | } | |
61 | ||
62 | < | Integrator::~Integrator() { |
63 | < | |
64 | < | if( nConstrained ){ |
62 | > | template<typename T> Integrator<T>::~Integrator(){ |
63 | > | if (consFramework != NULL) |
64 | > | delete consFramework; |
65 | > | /* |
66 | > | if (nConstrained){ |
67 | delete[] constrainedA; | |
68 | delete[] constrainedB; | |
69 | delete[] constrainedDsqr; | |
# | Line 53 | Line 71 | Integrator::~Integrator() { | |
71 | delete[] moved; | |
72 | delete[] oldPos; | |
73 | } | |
74 | < | |
74 | > | */ |
75 | } | |
76 | ||
77 | < | void Integrator::checkConstraints( void ){ |
78 | < | |
61 | < | |
77 | > | /* |
78 | > | template<typename T> void Integrator<T>::checkConstraints(void){ |
79 | isConstrained = 0; | |
80 | ||
81 | < | Constraint *temp_con; |
82 | < | Constraint *dummy_plug; |
81 | > | Constraint* temp_con; |
82 | > | Constraint* dummy_plug; |
83 | temp_con = new Constraint[info->n_SRI]; | |
84 | nConstrained = 0; | |
85 | int constrained = 0; | |
86 | < | |
86 | > | |
87 | SRI** theArray; | |
88 | < | for(int i = 0; i < nMols; i++){ |
89 | < | |
90 | < | theArray = (SRI**) molecules[i].getMyBonds(); |
91 | < | for(int j=0; j<molecules[i].getNBonds(); j++){ |
75 | < | |
88 | > | for (int i = 0; i < nMols; i++){ |
89 | > | |
90 | > | theArray = (SRI * *) molecules[i].getMyBonds(); |
91 | > | for (int j = 0; j < molecules[i].getNBonds(); j++){ |
92 | constrained = theArray[j]->is_constrained(); | |
93 | ||
94 | < | std::cerr << "Is the folowing bond constrained \n"; |
95 | < | theArray[j]->printMe(); |
96 | < | |
97 | < | if(constrained){ |
98 | < | |
83 | < | std::cerr << "Yes\n"; |
94 | > | if (constrained){ |
95 | > | dummy_plug = theArray[j]->get_constraint(); |
96 | > | temp_con[nConstrained].set_a(dummy_plug->get_a()); |
97 | > | temp_con[nConstrained].set_b(dummy_plug->get_b()); |
98 | > | temp_con[nConstrained].set_dsqr(dummy_plug->get_dsqr()); |
99 | ||
100 | < | dummy_plug = theArray[j]->get_constraint(); |
101 | < | temp_con[nConstrained].set_a( dummy_plug->get_a() ); |
102 | < | temp_con[nConstrained].set_b( dummy_plug->get_b() ); |
88 | < | temp_con[nConstrained].set_dsqr( dummy_plug->get_dsqr() ); |
89 | < | |
90 | < | nConstrained++; |
91 | < | constrained = 0; |
92 | < | } |
93 | < | else std::cerr << "No.\n"; |
100 | > | nConstrained++; |
101 | > | constrained = 0; |
102 | > | } |
103 | } | |
104 | ||
105 | < | theArray = (SRI**) molecules[i].getMyBends(); |
106 | < | for(int j=0; j<molecules[i].getNBends(); j++){ |
98 | < | |
105 | > | theArray = (SRI * *) molecules[i].getMyBends(); |
106 | > | for (int j = 0; j < molecules[i].getNBends(); j++){ |
107 | constrained = theArray[j]->is_constrained(); | |
108 | < | |
109 | < | if(constrained){ |
110 | < | |
111 | < | dummy_plug = theArray[j]->get_constraint(); |
112 | < | temp_con[nConstrained].set_a( dummy_plug->get_a() ); |
113 | < | temp_con[nConstrained].set_b( dummy_plug->get_b() ); |
114 | < | temp_con[nConstrained].set_dsqr( dummy_plug->get_dsqr() ); |
115 | < | |
116 | < | nConstrained++; |
109 | < | constrained = 0; |
108 | > | |
109 | > | if (constrained){ |
110 | > | dummy_plug = theArray[j]->get_constraint(); |
111 | > | temp_con[nConstrained].set_a(dummy_plug->get_a()); |
112 | > | temp_con[nConstrained].set_b(Dummy_plug->get_b()); |
113 | > | temp_con[nConstrained].set_dsqr(dummy_plug->get_dsqr()); |
114 | > | |
115 | > | nConstrained++; |
116 | > | constrained = 0; |
117 | } | |
118 | } | |
119 | ||
120 | < | theArray = (SRI**) molecules[i].getMyTorsions(); |
121 | < | for(int j=0; j<molecules[i].getNTorsions(); j++){ |
115 | < | |
120 | > | theArray = (SRI * *) molecules[i].getMyTorsions(); |
121 | > | for (int j = 0; j < molecules[i].getNTorsions(); j++){ |
122 | constrained = theArray[j]->is_constrained(); | |
123 | < | |
124 | < | if(constrained){ |
125 | < | |
126 | < | dummy_plug = theArray[j]->get_constraint(); |
127 | < | temp_con[nConstrained].set_a( dummy_plug->get_a() ); |
128 | < | temp_con[nConstrained].set_b( dummy_plug->get_b() ); |
129 | < | temp_con[nConstrained].set_dsqr( dummy_plug->get_dsqr() ); |
130 | < | |
131 | < | nConstrained++; |
126 | < | constrained = 0; |
123 | > | |
124 | > | if (constrained){ |
125 | > | dummy_plug = theArray[j]->get_constraint(); |
126 | > | temp_con[nConstrained].set_a(dummy_plug->get_a()); |
127 | > | temp_con[nConstrained].set_b(dummy_plug->get_b()); |
128 | > | temp_con[nConstrained].set_dsqr(dummy_plug->get_dsqr()); |
129 | > | |
130 | > | nConstrained++; |
131 | > | constrained = 0; |
132 | } | |
133 | } | |
134 | } | |
135 | ||
136 | < | if(nConstrained > 0){ |
137 | < | |
136 | > | |
137 | > | if (nConstrained > 0){ |
138 | isConstrained = 1; | |
139 | ||
140 | < | if(constrainedA != NULL ) delete[] constrainedA; |
141 | < | if(constrainedB != NULL ) delete[] constrainedB; |
142 | < | if(constrainedDsqr != NULL ) delete[] constrainedDsqr; |
143 | < | |
144 | < | constrainedA = new int[nConstrained]; |
145 | < | constrainedB = new int[nConstrained]; |
140 | > | if (constrainedA != NULL) |
141 | > | delete[] constrainedA; |
142 | > | if (constrainedB != NULL) |
143 | > | delete[] constrainedB; |
144 | > | if (constrainedDsqr != NULL) |
145 | > | delete[] constrainedDsqr; |
146 | > | |
147 | > | constrainedA = new int[nConstrained]; |
148 | > | constrainedB = new int[nConstrained]; |
149 | constrainedDsqr = new double[nConstrained]; | |
150 | < | |
151 | < | for( int i = 0; i < nConstrained; i++){ |
144 | < | |
150 | > | |
151 | > | for (int i = 0; i < nConstrained; i++){ |
152 | constrainedA[i] = temp_con[i].get_a(); | |
153 | constrainedB[i] = temp_con[i].get_b(); | |
154 | constrainedDsqr[i] = temp_con[i].get_dsqr(); | |
148 | – | |
155 | } | |
156 | ||
157 | < | |
158 | < | // save oldAtoms to check for lode balanceing later on. |
159 | < | |
157 | > | |
158 | > | // save oldAtoms to check for lode balancing later on. |
159 | > | |
160 | oldAtoms = nAtoms; | |
161 | < | |
161 | > | |
162 | moving = new int[nAtoms]; | |
163 | < | moved = new int[nAtoms]; |
163 | > | moved = new int[nAtoms]; |
164 | ||
165 | < | oldPos = new double[nAtoms*3]; |
165 | > | oldPos = new double[nAtoms * 3]; |
166 | } | |
167 | < | |
167 | > | |
168 | delete[] temp_con; | |
169 | } | |
170 | + | */ |
171 | ||
172 | + | template<typename T> void Integrator<T>::integrate(void){ |
173 | ||
174 | < | void Integrator::integrate( void ){ |
175 | < | |
176 | < | int i, j; // loop counters |
169 | < | |
170 | < | double runTime = info->run_time; |
171 | < | double sampleTime = info->sampleTime; |
172 | < | double statusTime = info->statusTime; |
174 | > | double runTime = info->run_time; |
175 | > | double sampleTime = info->sampleTime; |
176 | > | double statusTime = info->statusTime; |
177 | double thermalTime = info->thermalTime; | |
178 | + | double resetTime = info->resetTime; |
179 | ||
180 | + | double difference; |
181 | double currSample; | |
182 | double currThermal; | |
183 | double currStatus; | |
184 | < | double currTime; |
184 | > | double currReset; |
185 | ||
186 | int calcPot, calcStress; | |
181 | – | int isError; |
187 | ||
188 | < | tStats = new Thermo( info ); |
189 | < | statOut = new StatWriter( info ); |
190 | < | dumpOut = new DumpWriter( info ); |
188 | > | tStats = new Thermo(info); |
189 | > | statOut = new StatWriter(info); |
190 | > | dumpOut = new DumpWriter(info); |
191 | ||
192 | atoms = info->atoms; | |
188 | – | DirectionalAtom* dAtom; |
193 | ||
194 | dt = info->dt; | |
195 | dt2 = 0.5 * dt; | |
196 | ||
197 | < | // initialize the forces before the first step |
197 | > | readyCheck(); |
198 | ||
199 | < | myFF->doForces(1,1); |
199 | > | // remove center of mass drift velocity (in case we passed in a configuration |
200 | > | // that was drifting |
201 | > | tStats->removeCOMdrift(); |
202 | > | //tStats->removeAngularMomentum(); |
203 | ||
204 | < | if( info->setTemp ){ |
205 | < | |
206 | < | tStats->velocitize(); |
204 | > | // initialize the retraints if necessary |
205 | > | if (info->useSolidThermInt && !info->useLiquidThermInt) { |
206 | > | myFF->initRestraints(); |
207 | } | |
208 | + | |
209 | + | // initialize the forces before the first step |
210 | + | |
211 | + | calcForce(1, 1); |
212 | + | |
213 | + | //execute constraint algorithm to make sure at the very beginning the system is constrained |
214 | + | //consFramework->doPreConstraint(); |
215 | + | //consFramework->doConstrainA(); |
216 | + | //calcForce(1, 1); |
217 | + | //consFramework->doConstrainB(); |
218 | ||
219 | < | dumpOut->writeDump( 0.0 ); |
220 | < | statOut->writeStat( 0.0 ); |
221 | < | |
219 | > | if (info->setTemp){ |
220 | > | thermalize(); |
221 | > | } |
222 | > | |
223 | calcPot = 0; | |
224 | calcStress = 0; | |
225 | < | currSample = sampleTime; |
226 | < | currThermal = thermalTime; |
227 | < | currStatus = statusTime; |
228 | < | currTime = 0.0;; |
225 | > | currSample = sampleTime + info->getTime(); |
226 | > | currThermal = thermalTime+ info->getTime(); |
227 | > | currStatus = statusTime + info->getTime(); |
228 | > | currReset = resetTime + info->getTime(); |
229 | ||
230 | + | dumpOut->writeDump(info->getTime()); |
231 | + | statOut->writeStat(info->getTime()); |
232 | ||
213 | – | readyCheck(); |
233 | ||
234 | #ifdef IS_MPI | |
235 | < | strcpy( checkPointMsg, |
217 | < | "The integrator is ready to go." ); |
235 | > | strcpy(checkPointMsg, "The integrator is ready to go."); |
236 | MPIcheckPoint(); | |
237 | #endif // is_mpi | |
238 | ||
239 | < | while( currTime < runTime ){ |
240 | < | |
241 | < | if( (currTime+dt) >= currStatus ){ |
239 | > | while (info->getTime() < runTime && !stopIntegrator()){ |
240 | > | difference = info->getTime() + dt - currStatus; |
241 | > | if (difference > 0 || fabs(difference) < 1e-4 ){ |
242 | calcPot = 1; | |
243 | calcStress = 1; | |
244 | } | |
245 | ||
246 | < | std::cerr << currTime << "\n"; |
246 | > | #ifdef PROFILE |
247 | > | startProfile( pro1 ); |
248 | > | #endif |
249 | > | |
250 | > | integrateStep(calcPot, calcStress); |
251 | ||
252 | < | integrateStep( calcPot, calcStress ); |
253 | < | |
232 | < | currTime += dt; |
252 | > | #ifdef PROFILE |
253 | > | endProfile( pro1 ); |
254 | ||
255 | < | if( info->setTemp ){ |
256 | < | if( currTime >= currThermal ){ |
257 | < | tStats->velocitize(); |
258 | < | currThermal += thermalTime; |
255 | > | startProfile( pro2 ); |
256 | > | #endif // profile |
257 | > | |
258 | > | info->incrTime(dt); |
259 | > | |
260 | > | if (info->setTemp){ |
261 | > | if (info->getTime() >= currThermal){ |
262 | > | thermalize(); |
263 | > | currThermal += thermalTime; |
264 | } | |
265 | } | |
266 | ||
267 | < | if( currTime >= currSample ){ |
268 | < | dumpOut->writeDump( currTime ); |
267 | > | if (info->getTime() >= currSample){ |
268 | > | dumpOut->writeDump(info->getTime()); |
269 | currSample += sampleTime; | |
270 | } | |
271 | ||
272 | < | if( currTime >= currStatus ){ |
273 | < | statOut->writeStat( currTime ); |
274 | < | calcPot = 0; |
272 | > | if (info->getTime() >= currStatus){ |
273 | > | statOut->writeStat(info->getTime()); |
274 | > | calcPot = 0; |
275 | calcStress = 0; | |
276 | currStatus += statusTime; | |
277 | < | } |
277 | > | } |
278 | ||
279 | + | if (info->resetIntegrator){ |
280 | + | if (info->getTime() >= currReset){ |
281 | + | this->resetIntegrator(); |
282 | + | currReset += resetTime; |
283 | + | } |
284 | + | } |
285 | + | |
286 | + | #ifdef PROFILE |
287 | + | endProfile( pro2 ); |
288 | + | #endif //profile |
289 | + | |
290 | #ifdef IS_MPI | |
291 | < | strcpy( checkPointMsg, |
255 | < | "successfully took a time step." ); |
291 | > | strcpy(checkPointMsg, "successfully took a time step."); |
292 | MPIcheckPoint(); | |
293 | #endif // is_mpi | |
258 | – | |
294 | } | |
295 | ||
296 | < | dumpOut->writeFinal(currTime); |
296 | > | // dump out a file containing the omega values for the final configuration |
297 | > | if (info->useSolidThermInt && !info->useLiquidThermInt) |
298 | > | myFF->dumpzAngle(); |
299 | > | |
300 | ||
301 | delete dumpOut; | |
302 | delete statOut; | |
303 | } | |
304 | ||
305 | < | void Integrator::integrateStep( int calcPot, int calcStress ){ |
305 | > | template<typename T> void Integrator<T>::integrateStep(int calcPot, |
306 | > | int calcStress){ |
307 | > | // Position full step, and velocity half step |
308 | ||
309 | + | #ifdef PROFILE |
310 | + | startProfile(pro3); |
311 | + | #endif //profile |
312 | ||
313 | < | |
314 | < | // Position full step, and velocity half step |
313 | > | //save old state (position, velocity etc) |
314 | > | consFramework->doPreConstraint(); |
315 | ||
316 | < | preMove(); |
316 | > | #ifdef PROFILE |
317 | > | endProfile(pro3); |
318 | > | |
319 | > | startProfile(pro4); |
320 | > | #endif // profile |
321 | > | |
322 | moveA(); | |
275 | – | if( nConstrained ) constrainA(); |
323 | ||
324 | + | #ifdef PROFILE |
325 | + | endProfile(pro4); |
326 | + | |
327 | + | startProfile(pro5); |
328 | + | #endif//profile |
329 | + | |
330 | + | |
331 | + | #ifdef IS_MPI |
332 | + | strcpy(checkPointMsg, "Succesful moveA\n"); |
333 | + | MPIcheckPoint(); |
334 | + | #endif // is_mpi |
335 | + | |
336 | // calc forces | |
337 | + | calcForce(calcPot, calcStress); |
338 | ||
339 | < | myFF->doForces(calcPot,calcStress); |
339 | > | #ifdef IS_MPI |
340 | > | strcpy(checkPointMsg, "Succesful doForces\n"); |
341 | > | MPIcheckPoint(); |
342 | > | #endif // is_mpi |
343 | ||
344 | + | #ifdef PROFILE |
345 | + | endProfile( pro5 ); |
346 | + | |
347 | + | startProfile( pro6 ); |
348 | + | #endif //profile |
349 | + | |
350 | + | consFramework->doPreConstraint(); |
351 | + | |
352 | // finish the velocity half step | |
353 | < | |
353 | > | |
354 | moveB(); | |
355 | < | if( nConstrained ) constrainB(); |
356 | < | |
355 | > | |
356 | > | #ifdef PROFILE |
357 | > | endProfile(pro6); |
358 | > | #endif // profile |
359 | > | |
360 | > | #ifdef IS_MPI |
361 | > | strcpy(checkPointMsg, "Succesful moveB\n"); |
362 | > | MPIcheckPoint(); |
363 | > | #endif // is_mpi |
364 | } | |
365 | ||
366 | ||
367 | < | void Integrator::moveA( void ){ |
368 | < | |
291 | < | int i, j; |
367 | > | template<typename T> void Integrator<T>::moveA(void){ |
368 | > | size_t i, j; |
369 | DirectionalAtom* dAtom; | |
370 | double Tb[3], ji[3]; | |
294 | – | double A[3][3], I[3][3]; |
295 | – | double angle; |
371 | double vel[3], pos[3], frc[3]; | |
372 | double mass; | |
373 | + | double omega; |
374 | + | |
375 | + | for (i = 0; i < integrableObjects.size() ; i++){ |
376 | + | integrableObjects[i]->getVel(vel); |
377 | + | integrableObjects[i]->getPos(pos); |
378 | + | integrableObjects[i]->getFrc(frc); |
379 | + | |
380 | + | mass = integrableObjects[i]->getMass(); |
381 | ||
382 | < | for( i=0; i<nAtoms; i++ ){ |
300 | < | |
301 | < | atoms[i]->getVel( vel ); |
302 | < | atoms[i]->getPos( pos ); |
303 | < | atoms[i]->getFrc( frc ); |
304 | < | |
305 | < | mass = atoms[i]->getMass(); |
306 | < | |
307 | < | for (j=0; j < 3; j++) { |
382 | > | for (j = 0; j < 3; j++){ |
383 | // velocity half step | |
384 | < | vel[j] += ( dt2 * frc[j] / mass ) * eConvert; |
384 | > | vel[j] += (dt2 * frc[j] / mass) * eConvert; |
385 | // position whole step | |
386 | pos[j] += dt * vel[j]; | |
387 | } | |
388 | ||
389 | < | atoms[i]->setVel( vel ); |
390 | < | atoms[i]->setPos( pos ); |
389 | > | integrableObjects[i]->setVel(vel); |
390 | > | integrableObjects[i]->setPos(pos); |
391 | ||
392 | < | if( atoms[i]->isDirectional() ){ |
392 | > | if (integrableObjects[i]->isDirectional()){ |
393 | ||
319 | – | dAtom = (DirectionalAtom *)atoms[i]; |
320 | – | |
394 | // get and convert the torque to body frame | |
322 | – | |
323 | – | dAtom->getTrq( Tb ); |
324 | – | dAtom->lab2Body( Tb ); |
395 | ||
396 | + | integrableObjects[i]->getTrq(Tb); |
397 | + | integrableObjects[i]->lab2Body(Tb); |
398 | + | |
399 | // get the angular momentum, and propagate a half step | |
400 | ||
401 | < | dAtom->getJ( ji ); |
401 | > | integrableObjects[i]->getJ(ji); |
402 | ||
403 | < | for (j=0; j < 3; j++) |
403 | > | for (j = 0; j < 3; j++) |
404 | ji[j] += (dt2 * Tb[j]) * eConvert; | |
332 | – | |
333 | – | // use the angular velocities to propagate the rotation matrix a |
334 | – | // full time step |
405 | ||
406 | < | dAtom->getA(A); |
337 | < | dAtom->getI(I); |
338 | < | |
339 | < | // rotate about the x-axis |
340 | < | angle = dt2 * ji[0] / I[0][0]; |
341 | < | this->rotate( 1, 2, angle, ji, A ); |
406 | > | this->rotationPropagation( integrableObjects[i], ji ); |
407 | ||
408 | < | // rotate about the y-axis |
344 | < | angle = dt2 * ji[1] / I[1][1]; |
345 | < | this->rotate( 2, 0, angle, ji, A ); |
346 | < | |
347 | < | // rotate about the z-axis |
348 | < | angle = dt * ji[2] / I[2][2]; |
349 | < | this->rotate( 0, 1, angle, ji, A); |
350 | < | |
351 | < | // rotate about the y-axis |
352 | < | angle = dt2 * ji[1] / I[1][1]; |
353 | < | this->rotate( 2, 0, angle, ji, A ); |
354 | < | |
355 | < | // rotate about the x-axis |
356 | < | angle = dt2 * ji[0] / I[0][0]; |
357 | < | this->rotate( 1, 2, angle, ji, A ); |
358 | < | |
408 | > | integrableObjects[i]->setJ(ji); |
409 | ||
410 | < | dAtom->setJ( ji ); |
361 | < | dAtom->setA( A ); |
362 | < | |
363 | < | } |
410 | > | } |
411 | } | |
412 | + | |
413 | + | consFramework->doConstrainA(); |
414 | } | |
415 | ||
416 | ||
417 | < | void Integrator::moveB( void ){ |
417 | > | template<typename T> void Integrator<T>::moveB(void){ |
418 | int i, j; | |
370 | – | DirectionalAtom* dAtom; |
419 | double Tb[3], ji[3]; | |
420 | double vel[3], frc[3]; | |
421 | double mass; | |
422 | ||
423 | < | for( i=0; i<nAtoms; i++ ){ |
424 | < | |
425 | < | atoms[i]->getVel( vel ); |
378 | < | atoms[i]->getFrc( frc ); |
423 | > | for (i = 0; i < integrableObjects.size(); i++){ |
424 | > | integrableObjects[i]->getVel(vel); |
425 | > | integrableObjects[i]->getFrc(frc); |
426 | ||
427 | < | mass = atoms[i]->getMass(); |
427 | > | mass = integrableObjects[i]->getMass(); |
428 | ||
429 | // velocity half step | |
430 | < | for (j=0; j < 3; j++) |
431 | < | vel[j] += ( dt2 * frc[j] / mass ) * eConvert; |
385 | < | |
386 | < | atoms[i]->setVel( vel ); |
387 | < | |
388 | < | if( atoms[i]->isDirectional() ){ |
430 | > | for (j = 0; j < 3; j++) |
431 | > | vel[j] += (dt2 * frc[j] / mass) * eConvert; |
432 | ||
433 | < | dAtom = (DirectionalAtom *)atoms[i]; |
433 | > | integrableObjects[i]->setVel(vel); |
434 | ||
435 | < | // get and convert the torque to body frame |
435 | > | if (integrableObjects[i]->isDirectional()){ |
436 | ||
437 | < | dAtom->getTrq( Tb ); |
395 | < | dAtom->lab2Body( Tb ); |
437 | > | // get and convert the torque to body frame |
438 | ||
439 | + | integrableObjects[i]->getTrq(Tb); |
440 | + | integrableObjects[i]->lab2Body(Tb); |
441 | + | |
442 | // get the angular momentum, and propagate a half step | |
443 | ||
444 | < | dAtom->getJ( ji ); |
444 | > | integrableObjects[i]->getJ(ji); |
445 | ||
446 | < | for (j=0; j < 3; j++) |
446 | > | for (j = 0; j < 3; j++) |
447 | ji[j] += (dt2 * Tb[j]) * eConvert; | |
403 | – | |
448 | ||
449 | < | dAtom->setJ( ji ); |
449 | > | |
450 | > | integrableObjects[i]->setJ(ji); |
451 | } | |
452 | + | |
453 | } | |
454 | + | |
455 | + | consFramework->doConstrainB(); |
456 | } | |
457 | ||
458 | < | void Integrator::preMove( void ){ |
458 | > | /* |
459 | > | template<typename T> void Integrator<T>::preMove(void){ |
460 | int i, j; | |
461 | double pos[3]; | |
462 | ||
463 | < | if( nConstrained ){ |
463 | > | if (nConstrained){ |
464 | > | for (i = 0; i < nAtoms; i++){ |
465 | > | atoms[i]->getPos(pos); |
466 | ||
467 | < | for(i=0; i < nAtoms; i++) { |
468 | < | |
418 | < | atoms[i]->getPos( pos ); |
419 | < | |
420 | < | for (j = 0; j < 3; j++) { |
421 | < | oldPos[3*i + j] = pos[j]; |
467 | > | for (j = 0; j < 3; j++){ |
468 | > | oldPos[3 * i + j] = pos[j]; |
469 | } | |
423 | – | |
470 | } | |
471 | < | } |
471 | > | } |
472 | } | |
473 | ||
474 | < | void Integrator::constrainA(){ |
475 | < | |
430 | < | int i,j,k; |
474 | > | template<typename T> void Integrator<T>::constrainA(){ |
475 | > | int i, j; |
476 | int done; | |
477 | double posA[3], posB[3]; | |
478 | double velA[3], velB[3]; | |
# | Line 442 | Line 487 | void Integrator::constrainA(){ | |
487 | double gab; | |
488 | int iteration; | |
489 | ||
490 | < | for( i=0; i<nAtoms; i++){ |
490 | > | for (i = 0; i < nAtoms; i++){ |
491 | moving[i] = 0; | |
492 | < | moved[i] = 1; |
492 | > | moved[i] = 1; |
493 | } | |
494 | ||
495 | iteration = 0; | |
496 | done = 0; | |
497 | < | while( !done && (iteration < maxIteration )){ |
453 | < | |
497 | > | while (!done && (iteration < maxIteration)){ |
498 | done = 1; | |
499 | < | for(i=0; i<nConstrained; i++){ |
456 | < | |
499 | > | for (i = 0; i < nConstrained; i++){ |
500 | a = constrainedA[i]; | |
501 | b = constrainedB[i]; | |
459 | – | |
460 | – | ax = (a*3) + 0; |
461 | – | ay = (a*3) + 1; |
462 | – | az = (a*3) + 2; |
502 | ||
503 | < | bx = (b*3) + 0; |
504 | < | by = (b*3) + 1; |
505 | < | bz = (b*3) + 2; |
503 | > | ax = (a * 3) + 0; |
504 | > | ay = (a * 3) + 1; |
505 | > | az = (a * 3) + 2; |
506 | ||
507 | < | if( moved[a] || moved[b] ){ |
508 | < | |
509 | < | atoms[a]->getPos( posA ); |
510 | < | atoms[b]->getPos( posB ); |
511 | < | |
512 | < | for (j = 0; j < 3; j++ ) |
507 | > | bx = (b * 3) + 0; |
508 | > | by = (b * 3) + 1; |
509 | > | bz = (b * 3) + 2; |
510 | > | |
511 | > | if (moved[a] || moved[b]){ |
512 | > | atoms[a]->getPos(posA); |
513 | > | atoms[b]->getPos(posB); |
514 | > | |
515 | > | for (j = 0; j < 3; j++) |
516 | pab[j] = posA[j] - posB[j]; | |
475 | – | |
476 | – | //periodic boundary condition |
517 | ||
518 | < | info->wrapVector( pab ); |
518 | > | //periodic boundary condition |
519 | ||
520 | < | pabsq = pab[0] * pab[0] + pab[1] * pab[1] + pab[2] * pab[2]; |
520 | > | info->wrapVector(pab); |
521 | ||
522 | < | rabsq = constrainedDsqr[i]; |
483 | < | diffsq = rabsq - pabsq; |
522 | > | pabsq = pab[0] * pab[0] + pab[1] * pab[1] + pab[2] * pab[2]; |
523 | ||
524 | < | // the original rattle code from alan tidesley |
525 | < | if (fabs(diffsq) > (tol*rabsq*2)) { |
487 | < | rab[0] = oldPos[ax] - oldPos[bx]; |
488 | < | rab[1] = oldPos[ay] - oldPos[by]; |
489 | < | rab[2] = oldPos[az] - oldPos[bz]; |
524 | > | rabsq = constrainedDsqr[i]; |
525 | > | diffsq = rabsq - pabsq; |
526 | ||
527 | < | info->wrapVector( rab ); |
527 | > | // the original rattle code from alan tidesley |
528 | > | if (fabs(diffsq) > (tol * rabsq * 2)){ |
529 | > | rab[0] = oldPos[ax] - oldPos[bx]; |
530 | > | rab[1] = oldPos[ay] - oldPos[by]; |
531 | > | rab[2] = oldPos[az] - oldPos[bz]; |
532 | ||
533 | < | rpab = rab[0] * pab[0] + rab[1] * pab[1] + rab[2] * pab[2]; |
533 | > | info->wrapVector(rab); |
534 | ||
535 | < | rpabsq = rpab * rpab; |
535 | > | rpab = rab[0] * pab[0] + rab[1] * pab[1] + rab[2] * pab[2]; |
536 | ||
537 | + | rpabsq = rpab * rpab; |
538 | ||
498 | – | if (rpabsq < (rabsq * -diffsq)){ |
539 | ||
540 | + | if (rpabsq < (rabsq * -diffsq)){ |
541 | #ifdef IS_MPI | |
542 | < | a = atoms[a]->getGlobalIndex(); |
543 | < | b = atoms[b]->getGlobalIndex(); |
542 | > | a = atoms[a]->getGlobalIndex(); |
543 | > | b = atoms[b]->getGlobalIndex(); |
544 | #endif //is_mpi | |
545 | < | sprintf( painCave.errMsg, |
546 | < | "Constraint failure in constrainA at atom %d and %d.\n", |
547 | < | a, b ); |
548 | < | painCave.isFatal = 1; |
549 | < | simError(); |
550 | < | } |
545 | > | sprintf(painCave.errMsg, |
546 | > | "Constraint failure in constrainA at atom %d and %d.\n", a, |
547 | > | b); |
548 | > | painCave.isFatal = 1; |
549 | > | simError(); |
550 | > | } |
551 | ||
552 | < | rma = 1.0 / atoms[a]->getMass(); |
553 | < | rmb = 1.0 / atoms[b]->getMass(); |
552 | > | rma = 1.0 / atoms[a]->getMass(); |
553 | > | rmb = 1.0 / atoms[b]->getMass(); |
554 | ||
555 | < | gab = diffsq / ( 2.0 * ( rma + rmb ) * rpab ); |
555 | > | gab = diffsq / (2.0 * (rma + rmb) * rpab); |
556 | ||
557 | dx = rab[0] * gab; | |
558 | dy = rab[1] * gab; | |
559 | dz = rab[2] * gab; | |
560 | ||
561 | < | posA[0] += rma * dx; |
562 | < | posA[1] += rma * dy; |
563 | < | posA[2] += rma * dz; |
561 | > | posA[0] += rma * dx; |
562 | > | posA[1] += rma * dy; |
563 | > | posA[2] += rma * dz; |
564 | ||
565 | < | atoms[a]->setPos( posA ); |
565 | > | atoms[a]->setPos(posA); |
566 | ||
567 | < | posB[0] -= rmb * dx; |
568 | < | posB[1] -= rmb * dy; |
569 | < | posB[2] -= rmb * dz; |
567 | > | posB[0] -= rmb * dx; |
568 | > | posB[1] -= rmb * dy; |
569 | > | posB[2] -= rmb * dz; |
570 | ||
571 | < | atoms[b]->setPos( posB ); |
571 | > | atoms[b]->setPos(posB); |
572 | ||
573 | dx = dx / dt; | |
574 | dy = dy / dt; | |
575 | dz = dz / dt; | |
576 | ||
577 | < | atoms[a]->getVel( velA ); |
577 | > | atoms[a]->getVel(velA); |
578 | ||
579 | < | velA[0] += rma * dx; |
580 | < | velA[1] += rma * dy; |
581 | < | velA[2] += rma * dz; |
579 | > | velA[0] += rma * dx; |
580 | > | velA[1] += rma * dy; |
581 | > | velA[2] += rma * dz; |
582 | ||
583 | < | atoms[a]->setVel( velA ); |
583 | > | atoms[a]->setVel(velA); |
584 | ||
585 | < | atoms[b]->getVel( velB ); |
585 | > | atoms[b]->getVel(velB); |
586 | ||
587 | < | velB[0] -= rmb * dx; |
588 | < | velB[1] -= rmb * dy; |
589 | < | velB[2] -= rmb * dz; |
587 | > | velB[0] -= rmb * dx; |
588 | > | velB[1] -= rmb * dy; |
589 | > | velB[2] -= rmb * dz; |
590 | ||
591 | < | atoms[b]->setVel( velB ); |
591 | > | atoms[b]->setVel(velB); |
592 | ||
593 | < | moving[a] = 1; |
594 | < | moving[b] = 1; |
595 | < | done = 0; |
596 | < | } |
593 | > | moving[a] = 1; |
594 | > | moving[b] = 1; |
595 | > | done = 0; |
596 | > | } |
597 | } | |
598 | } | |
599 | < | |
600 | < | for(i=0; i<nAtoms; i++){ |
560 | < | |
599 | > | |
600 | > | for (i = 0; i < nAtoms; i++){ |
601 | moved[i] = moving[i]; | |
602 | moving[i] = 0; | |
603 | } | |
# | Line 565 | Line 605 | void Integrator::constrainA(){ | |
605 | iteration++; | |
606 | } | |
607 | ||
608 | < | if( !done ){ |
609 | < | |
610 | < | sprintf( painCave.errMsg, |
611 | < | "Constraint failure in constrainA, too many iterations: %d\n", |
572 | < | iteration ); |
608 | > | if (!done){ |
609 | > | sprintf(painCave.errMsg, |
610 | > | "Constraint failure in constrainA, too many iterations: %d\n", |
611 | > | iteration); |
612 | painCave.isFatal = 1; | |
613 | simError(); | |
614 | } | |
615 | ||
616 | } | |
617 | ||
618 | < | void Integrator::constrainB( void ){ |
619 | < | |
581 | < | int i,j,k; |
618 | > | template<typename T> void Integrator<T>::constrainB(void){ |
619 | > | int i, j; |
620 | int done; | |
621 | double posA[3], posB[3]; | |
622 | double velA[3], velB[3]; | |
# | Line 587 | Line 625 | void Integrator::constrainB( void ){ | |
625 | int a, b, ax, ay, az, bx, by, bz; | |
626 | double rma, rmb; | |
627 | double dx, dy, dz; | |
628 | < | double rabsq, pabsq, rvab; |
591 | < | double diffsq; |
628 | > | double rvab; |
629 | double gab; | |
630 | int iteration; | |
631 | ||
632 | < | for(i=0; i<nAtoms; i++){ |
632 | > | for (i = 0; i < nAtoms; i++){ |
633 | moving[i] = 0; | |
634 | moved[i] = 1; | |
635 | } | |
636 | ||
637 | done = 0; | |
638 | iteration = 0; | |
639 | < | while( !done && (iteration < maxIteration ) ){ |
603 | < | |
639 | > | while (!done && (iteration < maxIteration)){ |
640 | done = 1; | |
641 | ||
642 | < | for(i=0; i<nConstrained; i++){ |
607 | < | |
642 | > | for (i = 0; i < nConstrained; i++){ |
643 | a = constrainedA[i]; | |
644 | b = constrainedB[i]; | |
645 | ||
646 | < | ax = (a*3) + 0; |
647 | < | ay = (a*3) + 1; |
648 | < | az = (a*3) + 2; |
646 | > | ax = (a * 3) + 0; |
647 | > | ay = (a * 3) + 1; |
648 | > | az = (a * 3) + 2; |
649 | ||
650 | < | bx = (b*3) + 0; |
651 | < | by = (b*3) + 1; |
652 | < | bz = (b*3) + 2; |
650 | > | bx = (b * 3) + 0; |
651 | > | by = (b * 3) + 1; |
652 | > | bz = (b * 3) + 2; |
653 | ||
654 | < | if( moved[a] || moved[b] ){ |
654 | > | if (moved[a] || moved[b]){ |
655 | > | atoms[a]->getVel(velA); |
656 | > | atoms[b]->getVel(velB); |
657 | ||
658 | < | atoms[a]->getVel( velA ); |
659 | < | atoms[b]->getVel( velB ); |
660 | < | |
624 | < | vxab = velA[0] - velB[0]; |
625 | < | vyab = velA[1] - velB[1]; |
626 | < | vzab = velA[2] - velB[2]; |
658 | > | vxab = velA[0] - velB[0]; |
659 | > | vyab = velA[1] - velB[1]; |
660 | > | vzab = velA[2] - velB[2]; |
661 | ||
662 | < | atoms[a]->getPos( posA ); |
663 | < | atoms[b]->getPos( posB ); |
662 | > | atoms[a]->getPos(posA); |
663 | > | atoms[b]->getPos(posB); |
664 | ||
665 | < | for (j = 0; j < 3; j++) |
665 | > | for (j = 0; j < 3; j++) |
666 | rab[j] = posA[j] - posB[j]; | |
633 | – | |
634 | – | info->wrapVector( rab ); |
635 | – | |
636 | – | rma = 1.0 / atoms[a]->getMass(); |
637 | – | rmb = 1.0 / atoms[b]->getMass(); |
667 | ||
668 | < | rvab = rab[0] * vxab + rab[1] * vyab + rab[2] * vzab; |
640 | < | |
641 | < | gab = -rvab / ( ( rma + rmb ) * constrainedDsqr[i] ); |
668 | > | info->wrapVector(rab); |
669 | ||
670 | < | if (fabs(gab) > tol) { |
671 | < | |
645 | < | dx = rab[0] * gab; |
646 | < | dy = rab[1] * gab; |
647 | < | dz = rab[2] * gab; |
648 | < | |
649 | < | velA[0] += rma * dx; |
650 | < | velA[1] += rma * dy; |
651 | < | velA[2] += rma * dz; |
670 | > | rma = 1.0 / atoms[a]->getMass(); |
671 | > | rmb = 1.0 / atoms[b]->getMass(); |
672 | ||
673 | < | atoms[a]->setVel( velA ); |
673 | > | rvab = rab[0] * vxab + rab[1] * vyab + rab[2] * vzab; |
674 | ||
675 | < | velB[0] -= rmb * dx; |
656 | < | velB[1] -= rmb * dy; |
657 | < | velB[2] -= rmb * dz; |
675 | > | gab = -rvab / ((rma + rmb) * constrainedDsqr[i]); |
676 | ||
677 | < | atoms[b]->setVel( velB ); |
678 | < | |
679 | < | moving[a] = 1; |
680 | < | moving[b] = 1; |
681 | < | done = 0; |
682 | < | } |
677 | > | if (fabs(gab) > tol){ |
678 | > | dx = rab[0] * gab; |
679 | > | dy = rab[1] * gab; |
680 | > | dz = rab[2] * gab; |
681 | > | |
682 | > | velA[0] += rma * dx; |
683 | > | velA[1] += rma * dy; |
684 | > | velA[2] += rma * dz; |
685 | > | |
686 | > | atoms[a]->setVel(velA); |
687 | > | |
688 | > | velB[0] -= rmb * dx; |
689 | > | velB[1] -= rmb * dy; |
690 | > | velB[2] -= rmb * dz; |
691 | > | |
692 | > | atoms[b]->setVel(velB); |
693 | > | |
694 | > | moving[a] = 1; |
695 | > | moving[b] = 1; |
696 | > | done = 0; |
697 | > | } |
698 | } | |
699 | } | |
700 | ||
701 | < | for(i=0; i<nAtoms; i++){ |
701 | > | for (i = 0; i < nAtoms; i++){ |
702 | moved[i] = moving[i]; | |
703 | moving[i] = 0; | |
704 | } | |
705 | < | |
705 | > | |
706 | iteration++; | |
707 | } | |
675 | – | |
676 | – | if( !done ){ |
708 | ||
709 | < | |
710 | < | sprintf( painCave.errMsg, |
711 | < | "Constraint failure in constrainB, too many iterations: %d\n", |
712 | < | iteration ); |
709 | > | if (!done){ |
710 | > | sprintf(painCave.errMsg, |
711 | > | "Constraint failure in constrainB, too many iterations: %d\n", |
712 | > | iteration); |
713 | painCave.isFatal = 1; | |
714 | simError(); | |
715 | < | } |
685 | < | |
715 | > | } |
716 | } | |
717 | + | */ |
718 | + | template<typename T> void Integrator<T>::rotationPropagation |
719 | + | ( StuntDouble* sd, double ji[3] ){ |
720 | ||
721 | < | void Integrator::rotate( int axes1, int axes2, double angle, double ji[3], |
722 | < | double A[3][3] ){ |
721 | > | double angle; |
722 | > | double A[3][3], I[3][3]; |
723 | > | int i, j, k; |
724 | ||
725 | < | int i,j,k; |
725 | > | // use the angular velocities to propagate the rotation matrix a |
726 | > | // full time step |
727 | > | |
728 | > | sd->getA(A); |
729 | > | sd->getI(I); |
730 | > | |
731 | > | if (sd->isLinear()) { |
732 | > | i = sd->linearAxis(); |
733 | > | j = (i+1)%3; |
734 | > | k = (i+2)%3; |
735 | > | |
736 | > | angle = dt2 * ji[j] / I[j][j]; |
737 | > | this->rotate( k, i, angle, ji, A ); |
738 | > | |
739 | > | angle = dt * ji[k] / I[k][k]; |
740 | > | this->rotate( i, j, angle, ji, A); |
741 | > | |
742 | > | angle = dt2 * ji[j] / I[j][j]; |
743 | > | this->rotate( k, i, angle, ji, A ); |
744 | > | |
745 | > | } else { |
746 | > | // rotate about the x-axis |
747 | > | angle = dt2 * ji[0] / I[0][0]; |
748 | > | this->rotate( 1, 2, angle, ji, A ); |
749 | > | |
750 | > | // rotate about the y-axis |
751 | > | angle = dt2 * ji[1] / I[1][1]; |
752 | > | this->rotate( 2, 0, angle, ji, A ); |
753 | > | |
754 | > | // rotate about the z-axis |
755 | > | angle = dt * ji[2] / I[2][2]; |
756 | > | sd->addZangle(angle); |
757 | > | this->rotate( 0, 1, angle, ji, A); |
758 | > | |
759 | > | // rotate about the y-axis |
760 | > | angle = dt2 * ji[1] / I[1][1]; |
761 | > | this->rotate( 2, 0, angle, ji, A ); |
762 | > | |
763 | > | // rotate about the x-axis |
764 | > | angle = dt2 * ji[0] / I[0][0]; |
765 | > | this->rotate( 1, 2, angle, ji, A ); |
766 | > | |
767 | > | } |
768 | > | sd->setA( A ); |
769 | > | } |
770 | > | |
771 | > | template<typename T> void Integrator<T>::rotate(int axes1, int axes2, |
772 | > | double angle, double ji[3], |
773 | > | double A[3][3]){ |
774 | > | int i, j, k; |
775 | double sinAngle; | |
776 | double cosAngle; | |
777 | double angleSqr; | |
# | Line 700 | Line 783 | void Integrator::rotate( int axes1, int axes2, double | |
783 | ||
784 | // initialize the tempA | |
785 | ||
786 | < | for(i=0; i<3; i++){ |
787 | < | for(j=0; j<3; j++){ |
786 | > | for (i = 0; i < 3; i++){ |
787 | > | for (j = 0; j < 3; j++){ |
788 | tempA[j][i] = A[i][j]; | |
789 | } | |
790 | } | |
791 | ||
792 | // initialize the tempJ | |
793 | ||
794 | < | for( i=0; i<3; i++) tempJ[i] = ji[i]; |
795 | < | |
794 | > | for (i = 0; i < 3; i++) |
795 | > | tempJ[i] = ji[i]; |
796 | > | |
797 | // initalize rot as a unit matrix | |
798 | ||
799 | rot[0][0] = 1.0; | |
# | Line 719 | Line 803 | void Integrator::rotate( int axes1, int axes2, double | |
803 | rot[1][0] = 0.0; | |
804 | rot[1][1] = 1.0; | |
805 | rot[1][2] = 0.0; | |
806 | < | |
806 | > | |
807 | rot[2][0] = 0.0; | |
808 | rot[2][1] = 0.0; | |
809 | rot[2][2] = 1.0; | |
810 | < | |
810 | > | |
811 | // use a small angle aproximation for sin and cosine | |
812 | ||
813 | < | angleSqr = angle * angle; |
813 | > | angleSqr = angle * angle; |
814 | angleSqrOver4 = angleSqr / 4.0; | |
815 | top = 1.0 - angleSqrOver4; | |
816 | bottom = 1.0 + angleSqrOver4; | |
# | Line 739 | Line 823 | void Integrator::rotate( int axes1, int axes2, double | |
823 | ||
824 | rot[axes1][axes2] = sinAngle; | |
825 | rot[axes2][axes1] = -sinAngle; | |
826 | < | |
826 | > | |
827 | // rotate the momentum acoording to: ji[] = rot[][] * ji[] | |
828 | < | |
829 | < | for(i=0; i<3; i++){ |
828 | > | |
829 | > | for (i = 0; i < 3; i++){ |
830 | ji[i] = 0.0; | |
831 | < | for(k=0; k<3; k++){ |
831 | > | for (k = 0; k < 3; k++){ |
832 | ji[i] += rot[i][k] * tempJ[k]; | |
833 | } | |
834 | } | |
835 | ||
836 | < | // rotate the Rotation matrix acording to: |
836 | > | // rotate the Rotation matrix acording to: |
837 | // A[][] = A[][] * transpose(rot[][]) | |
838 | ||
839 | ||
# | Line 757 | Line 841 | void Integrator::rotate( int axes1, int axes2, double | |
841 | // calculation as: | |
842 | // transpose(A[][]) = transpose(A[][]) * transpose(rot[][]) | |
843 | ||
844 | < | for(i=0; i<3; i++){ |
845 | < | for(j=0; j<3; j++){ |
844 | > | for (i = 0; i < 3; i++){ |
845 | > | for (j = 0; j < 3; j++){ |
846 | A[j][i] = 0.0; | |
847 | < | for(k=0; k<3; k++){ |
848 | < | A[j][i] += tempA[i][k] * rot[j][k]; |
847 | > | for (k = 0; k < 3; k++){ |
848 | > | A[j][i] += tempA[i][k] * rot[j][k]; |
849 | } | |
850 | } | |
851 | } | |
852 | } | |
853 | + | |
854 | + | template<typename T> void Integrator<T>::calcForce(int calcPot, int calcStress){ |
855 | + | myFF->doForces(calcPot, calcStress); |
856 | + | } |
857 | + | |
858 | + | template<typename T> void Integrator<T>::thermalize(){ |
859 | + | tStats->velocitize(); |
860 | + | } |
861 | + | |
862 | + | template<typename T> double Integrator<T>::getConservedQuantity(void){ |
863 | + | return tStats->getTotalE(); |
864 | + | } |
865 | + | template<typename T> string Integrator<T>::getAdditionalParameters(void){ |
866 | + | //By default, return a null string |
867 | + | //The reason we use string instead of char* is that if we use char*, we will |
868 | + | //return a pointer point to local variable which might cause problem |
869 | + | return string(); |
870 | + | } |
871 | + | |
872 | + | |
873 | + | template<typename T> void Integrator<T>::printQuaternion(StuntDouble* sd){ |
874 | + | Mat4x4d S; |
875 | + | double I[3][3]; |
876 | + | Vector4d j4; |
877 | + | Vector3d j; |
878 | + | Vector3d tempJ; |
879 | + | Vector4d qdot; |
880 | + | Vector4d omega4; |
881 | + | Mat4x4d I4; |
882 | + | Quaternion q; |
883 | + | double I0; |
884 | + | Vector4d p_qua; |
885 | + | |
886 | + | if (sd->isDirectional()){ |
887 | + | sd->getQ(q.vec); |
888 | + | sd->getI(I); |
889 | + | sd->getJ(j.vec); |
890 | + | |
891 | + | //omega4[0] = 0.0; |
892 | + | //omega4[1] = j[0]/I[0][0]; |
893 | + | //omega4[2] = j[1]/I[1][1]; |
894 | + | //omega4[3] = j[2]/I[2][2]; |
895 | + | |
896 | + | //S = getS(q); |
897 | + | //qdot = 0.5 * S * omega4; |
898 | + | |
899 | + | //I0 = (qdot[1] * q[1] * I[0][0] + qdot[2] * q[2] * I[1][1] + qdot[3] * q[3] * I[2][2])/(qdot[1] * q[1]+ qdot[2] * q[2] + qdot[3] * q[3]); |
900 | + | |
901 | + | //I4.element[0][0] = I0; |
902 | + | //I4.element[1][1] = I[0][0]; |
903 | + | //I4.element[2][2] = I[1][1]; |
904 | + | //I4.element[3][3] = I[2][2]; |
905 | + | |
906 | + | S = getS(q); |
907 | + | j4[0] = 0.0; |
908 | + | j4[1] = j[0]; |
909 | + | j4[2] = j[1]; |
910 | + | j4[3] = j[2]; |
911 | + | |
912 | + | p_qua = 2 * S * j4; |
913 | + | |
914 | + | j4 = 0.5 * S.transpose() * p_qua; |
915 | + | //cout << "q0^2 + q1^2 + q2^2 + q3^2 = " << q[0]*q[0] + q[1]*q[1] + q[2]*q[2] + q[3]*q[3] << endl; |
916 | + | //cout << "q0*q0dot + q1*q1dot + q2 *q2dot + q3*q3dot = " <<q[0]*qdot[0] + q[1]*qdot[1] + q[2]*qdot[2] + q[3]*qdot[3] << endl; |
917 | + | //cout << "q1*q1dot* Ixx + q2*q2dot* Iyy + q3 *q3dot* Izz = " << qdot[1] * q[1] * I[0][0] + qdot[2] * q[2] * I[1][1] + qdot[3] * q[3] * I[2][2] << endl; |
918 | + | //cout << "q1*q1dot + q2 *q2dot + q3*q3dot = " << qdot[1] * q[1]+ qdot[2] * q[2] + qdot[3] * q[3] << endl; |
919 | + | //cout << "I0 = " << I0 << endl; |
920 | + | cout << "p_qua[0] = " << p_qua[0] << endl; |
921 | + | } |
922 | + | } |
923 | + | |
924 | + | template<typename T> Mat4x4d Integrator<T>::getS(const Quaternion& q){ |
925 | + | Mat4x4d result; |
926 | + | |
927 | + | result.element[0][0] = q.x; |
928 | + | result.element[0][1] = -q.y; |
929 | + | result.element[0][2] = -q.z; |
930 | + | result.element[0][3] = -q.w; |
931 | + | |
932 | + | result.element[1][0] = q.y; |
933 | + | result.element[1][1] = q.x; |
934 | + | result.element[1][2] = -q.w; |
935 | + | result.element[1][3] = q.z; |
936 | + | |
937 | + | result.element[2][0] = q.z; |
938 | + | result.element[2][1] = q.w; |
939 | + | result.element[2][2] = q.x; |
940 | + | result.element[2][3] = -q.y; |
941 | + | |
942 | + | result.element[3][0] = q.w; |
943 | + | result.element[3][1] = -q.z; |
944 | + | result.element[3][2] = q.y; |
945 | + | result.element[3][3] = q.x; |
946 | + | |
947 | + | return result; |
948 | + | } |
949 | + |
– | Removed lines |
+ | Added lines |
< | Changed lines |
> | Changed lines |