9 |
|
#include "simError.h" |
10 |
|
|
11 |
|
|
12 |
< |
// Basic isotropic thermostating and barostating via the Melchionna |
12 |
> |
// Basic non-isotropic thermostating and barostating via the Melchionna |
13 |
|
// modification of the Hoover algorithm: |
14 |
|
// |
15 |
|
// Melchionna, S., Ciccotti, G., and Holian, B. L., 1993, |
19 |
|
// |
20 |
|
// Hoover, W. G., 1986, Phys. Rev. A, 34, 2499. |
21 |
|
|
22 |
< |
NPTi::NPTi ( SimInfo *theInfo, ForceFields* the_ff): |
22 |
> |
NPTf::NPTf ( SimInfo *theInfo, ForceFields* the_ff): |
23 |
|
Integrator( theInfo, the_ff ) |
24 |
|
{ |
25 |
|
int i; |
31 |
|
have_target_pressure = 0; |
32 |
|
} |
33 |
|
|
34 |
< |
void NPTi::moveA() { |
34 |
> |
void NPTf::moveA() { |
35 |
|
|
36 |
|
int i,j,k; |
37 |
|
int atomIndex, aMatIndex; |
39 |
|
double Tb[3]; |
40 |
|
double ji[3]; |
41 |
|
double rj[3]; |
42 |
+ |
double ident[3][3], eta1[3][3], eta2[3][3], hmnew[3][3]; |
43 |
+ |
double hm[9]; |
44 |
+ |
double vx, vy, vz; |
45 |
+ |
double scx, scy, scz; |
46 |
|
double instaTemp, instaPress, instaVol; |
47 |
|
double tt2, tb2; |
48 |
|
double angle; |
49 |
+ |
double press[9]; |
50 |
|
|
51 |
|
tt2 = tauThermostat * tauThermostat; |
52 |
|
tb2 = tauBarostat * tauBarostat; |
53 |
|
|
54 |
|
instaTemp = tStats->getTemperature(); |
55 |
< |
instaPress = tStats->getPressure(); |
55 |
> |
tStats->getPressureTensor(press); |
56 |
|
instaVol = tStats->getVolume(); |
57 |
|
|
58 |
|
// first evolve chi a half step |
59 |
|
|
60 |
|
chi += dt2 * ( instaTemp / targetTemp - 1.0) / tt2; |
61 |
|
|
62 |
< |
for (i = 0; i < 9; i++) { |
63 |
< |
eta[i] += dt2 * ( instaVol * (sigma[i] - targetPressure*identMat[i])) |
64 |
< |
/ (NkBT*tb2)); |
65 |
< |
} |
66 |
< |
|
62 |
> |
eta[0] += dt2 * instaVol * (press[0] - targetPressure/p_convert) / |
63 |
> |
(NkBT*tb2); |
64 |
> |
eta[1] += dt2 * instaVol * press[1] / (NkBT*tb2); |
65 |
> |
eta[2] += dt2 * instaVol * press[2] / (NkBT*tb2); |
66 |
> |
eta[3] += dt2 * instaVol * press[3] / (NkBT*tb2); |
67 |
> |
eta[4] += dt2 * instaVol * (press[4] - targetPressure/p_convert) / |
68 |
> |
(NkBT*tb2); |
69 |
> |
eta[5] += dt2 * instaVol * press[5] / (NkBT*tb2); |
70 |
> |
eta[6] += dt2 * instaVol * press[6] / (NkBT*tb2); |
71 |
> |
eta[7] += dt2 * instaVol * press[7] / (NkBT*tb2); |
72 |
> |
eta[8] += dt2 * instaVol * (press[8] - targetPressure/p_convert) / |
73 |
> |
(NkBT*tb2); |
74 |
> |
|
75 |
|
for( i=0; i<nAtoms; i++ ){ |
76 |
|
atomIndex = i * 3; |
77 |
|
aMatIndex = i * 9; |
78 |
|
|
79 |
|
// velocity half step |
80 |
< |
for( j=atomIndex; j<(atomIndex+3); j++ ) |
81 |
< |
vel[j] += dt2 * ((frc[j]/atoms[i]->getMass())*eConvert |
82 |
< |
- vel[j]*(chi+eta)); |
80 |
> |
|
81 |
> |
vx = vel[atomIndex]; |
82 |
> |
vy = vel[atomIndex+1]; |
83 |
> |
vz = vel[atomIndex+2]; |
84 |
> |
|
85 |
> |
scx = (chi + eta[0])*vx + eta[1]*vy + eta[2]*vz; |
86 |
> |
scy = eta[3]*vx + (chi + eta[4])*vy + eta[5]*vz; |
87 |
> |
scz = eta[6]*vx + eta[7]*vy + (chi + eta[8])*vz; |
88 |
> |
|
89 |
> |
vx += dt2 * ((frc[atomIndex] /atoms[i]->getMass())*eConvert - scx); |
90 |
> |
vy += dt2 * ((frc[atomIndex+1]/atoms[i]->getMass())*eConvert - scy); |
91 |
> |
vz += dt2 * ((frc[atomIndex+2]/atoms[i]->getMass())*eConvert - scz); |
92 |
|
|
93 |
+ |
vel[atomIndex] = vx; |
94 |
+ |
vel[atomIndex+1] = vy; |
95 |
+ |
vel[atomIndex+2] = vz; |
96 |
+ |
|
97 |
|
// position whole step |
98 |
|
|
99 |
< |
for( j=atomIndex; j<(atomIndex+3); j=j+3 ) { |
100 |
< |
rj[0] = pos[j]; |
101 |
< |
rj[1] = pos[j+1]; |
76 |
< |
rj[2] = pos[j+2]; |
99 |
> |
rj[0] = pos[atomIndex]; |
100 |
> |
rj[1] = pos[atomIndex+1]; |
101 |
> |
rj[2] = pos[atomIndex+2]; |
102 |
|
|
103 |
< |
info->wrapVector(rj); |
103 |
> |
info->wrapVector(rj); |
104 |
|
|
105 |
< |
pos[j] += dt * (vel[j] + eta*rj[0]); |
106 |
< |
pos[j+1] += dt * (vel[j+1] + eta*rj[1]); |
107 |
< |
pos[j+2] += dt * (vel[j+2] + eta*rj[2]); |
83 |
< |
} |
105 |
> |
scx = eta[0]*rj[0] + eta[1]*rj[1] + eta[2]*rj[2]; |
106 |
> |
scy = eta[3]*rj[0] + eta[4]*rj[1] + eta[5]*rj[2]; |
107 |
> |
scz = eta[6]*rj[0] + eta[7]*rj[1] + eta[8]*rj[2]; |
108 |
|
|
109 |
< |
// Scale the box after all the positions have been moved: |
110 |
< |
|
111 |
< |
info->scaleBox(exp(dt*eta)); |
109 |
> |
pos[atomIndex] += dt * (vel[atomIndex] + scx); |
110 |
> |
pos[atomIndex+1] += dt * (vel[atomIndex+1] + scy); |
111 |
> |
pos[atomIndex+2] += dt * (vel[atomIndex+2] + scz); |
112 |
|
|
113 |
|
if( atoms[i]->isDirectional() ){ |
114 |
|
|
161 |
|
} |
162 |
|
|
163 |
|
} |
164 |
+ |
|
165 |
+ |
// Scale the box after all the positions have been moved: |
166 |
+ |
|
167 |
+ |
// Use a taylor expansion for eta products: Hmat = Hmat . exp(dt * etaMat) |
168 |
+ |
// Hmat = Hmat . ( Ident + dt * etaMat + dt^2 * etaMat*etaMat / 2) |
169 |
+ |
|
170 |
+ |
|
171 |
+ |
for(i=0; i<3; i++){ |
172 |
+ |
for(j=0; j<3; j++){ |
173 |
+ |
ident[i][j] = 0.0; |
174 |
+ |
eta1[i][j] = eta[3*i+j]; |
175 |
+ |
eta2[i][j] = 0.0; |
176 |
+ |
for(k=0; k<3; k++){ |
177 |
+ |
eta2[i][j] += eta[3*i+k] * eta[3*k+j]; |
178 |
+ |
} |
179 |
+ |
} |
180 |
+ |
ident[i][i] = 1.0; |
181 |
+ |
} |
182 |
+ |
|
183 |
+ |
|
184 |
+ |
info->getBoxM(hm); |
185 |
+ |
|
186 |
+ |
for(i=0; i<3; i++){ |
187 |
+ |
for(j=0; j<3; j++){ |
188 |
+ |
hmnew[i][j] = 0.0; |
189 |
+ |
for(k=0; k<3; k++){ |
190 |
+ |
// remember that hmat has transpose ordering for Fortran compat: |
191 |
+ |
hmnew[i][j] += hm[3*k+i] * (ident[k][j] |
192 |
+ |
+ dt * eta1[k][j] |
193 |
+ |
+ 0.5 * dt * dt * eta2[k][j]); |
194 |
+ |
} |
195 |
+ |
} |
196 |
+ |
} |
197 |
+ |
|
198 |
+ |
for (i = 0; i < 3; i++) { |
199 |
+ |
for (j = 0; j < 3; j++) { |
200 |
+ |
// remember that hmat has transpose ordering for Fortran compat: |
201 |
+ |
hm[3*j + i] = hmnew[i][j]; |
202 |
+ |
} |
203 |
+ |
} |
204 |
+ |
|
205 |
+ |
info->setBoxM(hm); |
206 |
+ |
|
207 |
|
} |
208 |
|
|
209 |
< |
void NPTi::moveB( void ){ |
209 |
> |
void NPTf::moveB( void ){ |
210 |
|
int i,j,k; |
211 |
|
int atomIndex; |
212 |
|
DirectionalAtom* dAtom; |
213 |
|
double Tb[3]; |
214 |
|
double ji[3]; |
215 |
< |
double instaTemp, instaPress, instaVol; |
215 |
> |
double press[9]; |
216 |
> |
double instaTemp, instaVol; |
217 |
|
double tt2, tb2; |
218 |
+ |
double vx, vy, vz; |
219 |
+ |
double scx, scy, scz; |
220 |
+ |
const double p_convert = 1.63882576e8; |
221 |
|
|
222 |
|
tt2 = tauThermostat * tauThermostat; |
223 |
|
tb2 = tauBarostat * tauBarostat; |
224 |
|
|
225 |
|
instaTemp = tStats->getTemperature(); |
226 |
< |
instaPress = tStats->getPressure(); |
226 |
> |
tStats->getPressureTensor(press); |
227 |
|
instaVol = tStats->getVolume(); |
228 |
< |
|
228 |
> |
|
229 |
> |
// first evolve chi a half step |
230 |
> |
|
231 |
|
chi += dt2 * ( instaTemp / targetTemp - 1.0) / tt2; |
159 |
– |
eta += dt2 * ( instaVol * (instaPress - targetPressure) / (NkBT*tb2)); |
232 |
|
|
233 |
+ |
eta[0] += dt2 * instaVol * (press[0] - targetPressure/p_convert) / |
234 |
+ |
(NkBT*tb2); |
235 |
+ |
eta[1] += dt2 * instaVol * press[1] / (NkBT*tb2); |
236 |
+ |
eta[2] += dt2 * instaVol * press[2] / (NkBT*tb2); |
237 |
+ |
eta[3] += dt2 * instaVol * press[3] / (NkBT*tb2); |
238 |
+ |
eta[4] += dt2 * instaVol * (press[4] - targetPressure/p_convert) / |
239 |
+ |
(NkBT*tb2); |
240 |
+ |
eta[5] += dt2 * instaVol * press[5] / (NkBT*tb2); |
241 |
+ |
eta[6] += dt2 * instaVol * press[6] / (NkBT*tb2); |
242 |
+ |
eta[7] += dt2 * instaVol * press[7] / (NkBT*tb2); |
243 |
+ |
eta[8] += dt2 * instaVol * (press[8] - targetPressure/p_convert) / |
244 |
+ |
(NkBT*tb2); |
245 |
+ |
|
246 |
|
for( i=0; i<nAtoms; i++ ){ |
247 |
|
atomIndex = i * 3; |
248 |
< |
|
248 |
> |
|
249 |
|
// velocity half step |
165 |
– |
for( j=atomIndex; j<(atomIndex+3); j++ ) |
166 |
– |
for( j=atomIndex; j<(atomIndex+3); j++ ) |
167 |
– |
vel[j] += dt2 * ((frc[j]/atoms[i]->getMass())*eConvert |
168 |
– |
- vel[j]*(chi+eta)); |
250 |
|
|
251 |
+ |
vx = vel[atomIndex]; |
252 |
+ |
vy = vel[atomIndex+1]; |
253 |
+ |
vz = vel[atomIndex+2]; |
254 |
+ |
|
255 |
+ |
scx = (chi + eta[0])*vx + eta[1]*vy + eta[2]*vz; |
256 |
+ |
scy = eta[3]*vx + (chi + eta[4])*vy + eta[5]*vz; |
257 |
+ |
scz = eta[6]*vx + eta[7]*vy + (chi + eta[8])*vz; |
258 |
+ |
|
259 |
+ |
vx += dt2 * ((frc[atomIndex] /atoms[i]->getMass())*eConvert - scx); |
260 |
+ |
vy += dt2 * ((frc[atomIndex+1]/atoms[i]->getMass())*eConvert - scy); |
261 |
+ |
vz += dt2 * ((frc[atomIndex+2]/atoms[i]->getMass())*eConvert - scz); |
262 |
+ |
|
263 |
+ |
vel[atomIndex] = vx; |
264 |
+ |
vel[atomIndex+1] = vy; |
265 |
+ |
vel[atomIndex+2] = vz; |
266 |
+ |
|
267 |
|
if( atoms[i]->isDirectional() ){ |
268 |
|
|
269 |
|
dAtom = (DirectionalAtom *)atoms[i]; |
294 |
|
} |
295 |
|
} |
296 |
|
|
297 |
< |
int NPTi::readyCheck() { |
297 |
> |
int NPTf::readyCheck() { |
298 |
|
|
299 |
|
// First check to see if we have a target temperature. |
300 |
|
// Not having one is fatal. |
301 |
|
|
302 |
|
if (!have_target_temp) { |
303 |
|
sprintf( painCave.errMsg, |
304 |
< |
"NPTi error: You can't use the NPTi integrator\n" |
304 |
> |
"NPTf error: You can't use the NPTf integrator\n" |
305 |
|
" without a targetTemp!\n" |
306 |
|
); |
307 |
|
painCave.isFatal = 1; |
311 |
|
|
312 |
|
if (!have_target_pressure) { |
313 |
|
sprintf( painCave.errMsg, |
314 |
< |
"NPTi error: You can't use the NPTi integrator\n" |
314 |
> |
"NPTf error: You can't use the NPTf integrator\n" |
315 |
|
" without a targetPressure!\n" |
316 |
|
); |
317 |
|
painCave.isFatal = 1; |
323 |
|
|
324 |
|
if (!have_tau_thermostat) { |
325 |
|
sprintf( painCave.errMsg, |
326 |
< |
"NPTi error: If you use the NPTi\n" |
326 |
> |
"NPTf error: If you use the NPTf\n" |
327 |
|
" integrator, you must set tauThermostat.\n"); |
328 |
|
painCave.isFatal = 1; |
329 |
|
simError(); |
334 |
|
|
335 |
|
if (!have_tau_barostat) { |
336 |
|
sprintf( painCave.errMsg, |
337 |
< |
"NPTi error: If you use the NPTi\n" |
337 |
> |
"NPTf error: If you use the NPTf\n" |
338 |
|
" integrator, you must set tauBarostat.\n"); |
339 |
|
painCave.isFatal = 1; |
340 |
|
simError(); |