ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/group/trunk/OOPSE/libmdtools/Integrator.cpp
Revision: 1284
Committed: Mon Jun 21 18:52:21 2004 UTC (20 years ago) by tim
File size: 19066 byte(s)
Log Message:
roll in progress

File Contents

# Content
1 #include <iostream>
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 template<typename T> Integrator<T>::Integrator(SimInfo* theInfo,
20 ForceFields* the_ff){
21 info = theInfo;
22 myFF = the_ff;
23 isFirst = 1;
24
25 molecules = info->molecules;
26 nMols = info->n_mol;
27
28 // give a little love back to the SimInfo object
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 consFramework = new RattleFramework(info);
38
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 /*
47 // check for constraints
48
49 constrainedA = NULL;
50 constrainedB = NULL;
51 constrainedDsqr = NULL;
52 moving = NULL;
53 moved = NULL;
54 oldPos = NULL;
55
56 nConstrained = 0;
57
58 checkConstraints();
59 */
60 }
61
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;
70 delete[] moving;
71 delete[] moved;
72 delete[] oldPos;
73 }
74 */
75 }
76
77 /*
78 template<typename T> void Integrator<T>::checkConstraints(void){
79 isConstrained = 0;
80
81 Constraint* temp_con;
82 Constraint* dummy_plug;
83 temp_con = new Constraint[info->n_SRI];
84 nConstrained = 0;
85 int constrained = 0;
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++){
92 constrained = theArray[j]->is_constrained();
93
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 nConstrained++;
101 constrained = 0;
102 }
103 }
104
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 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++){
122 constrained = theArray[j]->is_constrained();
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
137 if (nConstrained > 0){
138 isConstrained = 1;
139
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++){
152 constrainedA[i] = temp_con[i].get_a();
153 constrainedB[i] = temp_con[i].get_b();
154 constrainedDsqr[i] = temp_con[i].get_dsqr();
155 }
156
157
158 // save oldAtoms to check for lode balancing later on.
159
160 oldAtoms = nAtoms;
161
162 moving = new int[nAtoms];
163 moved = new int[nAtoms];
164
165 oldPos = new double[nAtoms * 3];
166 }
167
168 delete[] temp_con;
169 }
170 */
171
172 template<typename T> void Integrator<T>::integrate(void){
173
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 currReset;
185
186 int calcPot, calcStress;
187
188 tStats = new Thermo(info);
189 statOut = new StatWriter(info);
190 dumpOut = new DumpWriter(info);
191
192 atoms = info->atoms;
193
194 dt = info->dt;
195 dt2 = 0.5 * dt;
196
197 readyCheck();
198
199 // remove center of mass drift velocity (in case we passed in a configuration
200 // that was drifting
201 tStats->removeCOMdrift();
202
203 // initialize the retraints if necessary
204 if (info->useSolidThermInt && !info->useLiquidThermInt) {
205 myFF->initRestraints();
206 }
207
208 // initialize the forces before the first step
209
210 calcForce(1, 1);
211
212 //execute constraint algorithm to make sure at the very beginning the system is constrained
213 //consFramework->doPreConstraint();
214 //consFramework->doConstrainA();
215 //calcForce(1, 1);
216 //consFramework->doConstrainB();
217
218 if (info->setTemp){
219 thermalize();
220 }
221
222 calcPot = 0;
223 calcStress = 0;
224 currSample = sampleTime + info->getTime();
225 currThermal = thermalTime+ info->getTime();
226 currStatus = statusTime + info->getTime();
227 currReset = resetTime + info->getTime();
228
229 dumpOut->writeDump(info->getTime());
230 statOut->writeStat(info->getTime());
231
232
233 #ifdef IS_MPI
234 strcpy(checkPointMsg, "The integrator is ready to go.");
235 MPIcheckPoint();
236 #endif // is_mpi
237
238 while (info->getTime() < runTime && !stopIntegrator()){
239 difference = info->getTime() + dt - currStatus;
240 if (difference > 0 || fabs(difference) < 1e-4 ){
241 calcPot = 1;
242 calcStress = 1;
243 }
244
245 #ifdef PROFILE
246 startProfile( pro1 );
247 #endif
248
249 integrateStep(calcPot, calcStress);
250
251 #ifdef PROFILE
252 endProfile( pro1 );
253
254 startProfile( pro2 );
255 #endif // profile
256
257 info->incrTime(dt);
258
259 if (info->setTemp){
260 if (info->getTime() >= currThermal){
261 thermalize();
262 currThermal += thermalTime;
263 }
264 }
265
266 if (info->getTime() >= currSample){
267 dumpOut->writeDump(info->getTime());
268 currSample += sampleTime;
269 }
270
271 if (info->getTime() >= currStatus){
272 statOut->writeStat(info->getTime());
273 calcPot = 0;
274 calcStress = 0;
275 currStatus += statusTime;
276 }
277
278 if (info->resetIntegrator){
279 if (info->getTime() >= currReset){
280 this->resetIntegrator();
281 currReset += resetTime;
282 }
283 }
284
285 #ifdef PROFILE
286 endProfile( pro2 );
287 #endif //profile
288
289 #ifdef IS_MPI
290 strcpy(checkPointMsg, "successfully took a time step.");
291 MPIcheckPoint();
292 #endif // is_mpi
293 }
294
295 // dump out a file containing the omega values for the final configuration
296 if (info->useSolidThermInt && !info->useLiquidThermInt)
297 myFF->dumpzAngle();
298
299
300 delete dumpOut;
301 delete statOut;
302 }
303
304 template<typename T> void Integrator<T>::integrateStep(int calcPot,
305 int calcStress){
306 // Position full step, and velocity half step
307
308 #ifdef PROFILE
309 startProfile(pro3);
310 #endif //profile
311
312 //save old state (position, velocity etc)
313 consFramework->doPreConstraint();
314
315 #ifdef PROFILE
316 endProfile(pro3);
317
318 startProfile(pro4);
319 #endif // profile
320
321 moveA();
322
323 #ifdef PROFILE
324 endProfile(pro4);
325
326 startProfile(pro5);
327 #endif//profile
328
329
330 #ifdef IS_MPI
331 strcpy(checkPointMsg, "Succesful moveA\n");
332 MPIcheckPoint();
333 #endif // is_mpi
334
335 // calc forces
336 calcForce(calcPot, calcStress);
337
338 #ifdef IS_MPI
339 strcpy(checkPointMsg, "Succesful doForces\n");
340 MPIcheckPoint();
341 #endif // is_mpi
342
343 #ifdef PROFILE
344 endProfile( pro5 );
345
346 startProfile( pro6 );
347 #endif //profile
348
349 // finish the velocity half step
350
351 moveB();
352
353 #ifdef PROFILE
354 endProfile(pro6);
355 #endif // profile
356
357 #ifdef IS_MPI
358 strcpy(checkPointMsg, "Succesful moveB\n");
359 MPIcheckPoint();
360 #endif // is_mpi
361 }
362
363
364 template<typename T> void Integrator<T>::moveA(void){
365 size_t i, j;
366 DirectionalAtom* dAtom;
367 double Tb[3], ji[3];
368 double vel[3], pos[3], frc[3];
369 double mass;
370 double omega;
371
372 for (i = 0; i < integrableObjects.size() ; i++){
373 integrableObjects[i]->getVel(vel);
374 integrableObjects[i]->getPos(pos);
375 integrableObjects[i]->getFrc(frc);
376
377 mass = integrableObjects[i]->getMass();
378
379 for (j = 0; j < 3; j++){
380 // velocity half step
381 vel[j] += (dt2 * frc[j] / mass) * eConvert;
382 // position whole step
383 pos[j] += dt * vel[j];
384 }
385
386 integrableObjects[i]->setVel(vel);
387 integrableObjects[i]->setPos(pos);
388
389 if (integrableObjects[i]->isDirectional()){
390
391 // get and convert the torque to body frame
392
393 integrableObjects[i]->getTrq(Tb);
394 integrableObjects[i]->lab2Body(Tb);
395
396 // get the angular momentum, and propagate a half step
397
398 integrableObjects[i]->getJ(ji);
399
400 for (j = 0; j < 3; j++)
401 ji[j] += (dt2 * Tb[j]) * eConvert;
402
403 this->rotationPropagation( integrableObjects[i], ji );
404
405 integrableObjects[i]->setJ(ji);
406 }
407 }
408
409 consFramework->doConstrainA();
410 }
411
412
413 template<typename T> void Integrator<T>::moveB(void){
414 int i, j;
415 double Tb[3], ji[3];
416 double vel[3], frc[3];
417 double mass;
418
419 for (i = 0; i < integrableObjects.size(); i++){
420 integrableObjects[i]->getVel(vel);
421 integrableObjects[i]->getFrc(frc);
422
423 mass = integrableObjects[i]->getMass();
424
425 // velocity half step
426 for (j = 0; j < 3; j++)
427 vel[j] += (dt2 * frc[j] / mass) * eConvert;
428
429 integrableObjects[i]->setVel(vel);
430
431 if (integrableObjects[i]->isDirectional()){
432
433 // get and convert the torque to body frame
434
435 integrableObjects[i]->getTrq(Tb);
436 integrableObjects[i]->lab2Body(Tb);
437
438 // get the angular momentum, and propagate a half step
439
440 integrableObjects[i]->getJ(ji);
441
442 for (j = 0; j < 3; j++)
443 ji[j] += (dt2 * Tb[j]) * eConvert;
444
445
446 integrableObjects[i]->setJ(ji);
447 }
448 }
449
450 consFramework->doConstrainB();
451 }
452
453 /*
454 template<typename T> void Integrator<T>::preMove(void){
455 int i, j;
456 double pos[3];
457
458 if (nConstrained){
459 for (i = 0; i < nAtoms; i++){
460 atoms[i]->getPos(pos);
461
462 for (j = 0; j < 3; j++){
463 oldPos[3 * i + j] = pos[j];
464 }
465 }
466 }
467 }
468
469 template<typename T> void Integrator<T>::constrainA(){
470 int i, j;
471 int done;
472 double posA[3], posB[3];
473 double velA[3], velB[3];
474 double pab[3];
475 double rab[3];
476 int a, b, ax, ay, az, bx, by, bz;
477 double rma, rmb;
478 double dx, dy, dz;
479 double rpab;
480 double rabsq, pabsq, rpabsq;
481 double diffsq;
482 double gab;
483 int iteration;
484
485 for (i = 0; i < nAtoms; i++){
486 moving[i] = 0;
487 moved[i] = 1;
488 }
489
490 iteration = 0;
491 done = 0;
492 while (!done && (iteration < maxIteration)){
493 done = 1;
494 for (i = 0; i < nConstrained; i++){
495 a = constrainedA[i];
496 b = constrainedB[i];
497
498 ax = (a * 3) + 0;
499 ay = (a * 3) + 1;
500 az = (a * 3) + 2;
501
502 bx = (b * 3) + 0;
503 by = (b * 3) + 1;
504 bz = (b * 3) + 2;
505
506 if (moved[a] || moved[b]){
507 atoms[a]->getPos(posA);
508 atoms[b]->getPos(posB);
509
510 for (j = 0; j < 3; j++)
511 pab[j] = posA[j] - posB[j];
512
513 //periodic boundary condition
514
515 info->wrapVector(pab);
516
517 pabsq = pab[0] * pab[0] + pab[1] * pab[1] + pab[2] * pab[2];
518
519 rabsq = constrainedDsqr[i];
520 diffsq = rabsq - pabsq;
521
522 // the original rattle code from alan tidesley
523 if (fabs(diffsq) > (tol * rabsq * 2)){
524 rab[0] = oldPos[ax] - oldPos[bx];
525 rab[1] = oldPos[ay] - oldPos[by];
526 rab[2] = oldPos[az] - oldPos[bz];
527
528 info->wrapVector(rab);
529
530 rpab = rab[0] * pab[0] + rab[1] * pab[1] + rab[2] * pab[2];
531
532 rpabsq = rpab * rpab;
533
534
535 if (rpabsq < (rabsq * -diffsq)){
536 #ifdef IS_MPI
537 a = atoms[a]->getGlobalIndex();
538 b = atoms[b]->getGlobalIndex();
539 #endif //is_mpi
540 sprintf(painCave.errMsg,
541 "Constraint failure in constrainA at atom %d and %d.\n", a,
542 b);
543 painCave.isFatal = 1;
544 simError();
545 }
546
547 rma = 1.0 / atoms[a]->getMass();
548 rmb = 1.0 / atoms[b]->getMass();
549
550 gab = diffsq / (2.0 * (rma + rmb) * rpab);
551
552 dx = rab[0] * gab;
553 dy = rab[1] * gab;
554 dz = rab[2] * gab;
555
556 posA[0] += rma * dx;
557 posA[1] += rma * dy;
558 posA[2] += rma * dz;
559
560 atoms[a]->setPos(posA);
561
562 posB[0] -= rmb * dx;
563 posB[1] -= rmb * dy;
564 posB[2] -= rmb * dz;
565
566 atoms[b]->setPos(posB);
567
568 dx = dx / dt;
569 dy = dy / dt;
570 dz = dz / dt;
571
572 atoms[a]->getVel(velA);
573
574 velA[0] += rma * dx;
575 velA[1] += rma * dy;
576 velA[2] += rma * dz;
577
578 atoms[a]->setVel(velA);
579
580 atoms[b]->getVel(velB);
581
582 velB[0] -= rmb * dx;
583 velB[1] -= rmb * dy;
584 velB[2] -= rmb * dz;
585
586 atoms[b]->setVel(velB);
587
588 moving[a] = 1;
589 moving[b] = 1;
590 done = 0;
591 }
592 }
593 }
594
595 for (i = 0; i < nAtoms; i++){
596 moved[i] = moving[i];
597 moving[i] = 0;
598 }
599
600 iteration++;
601 }
602
603 if (!done){
604 sprintf(painCave.errMsg,
605 "Constraint failure in constrainA, too many iterations: %d\n",
606 iteration);
607 painCave.isFatal = 1;
608 simError();
609 }
610
611 }
612
613 template<typename T> void Integrator<T>::constrainB(void){
614 int i, j;
615 int done;
616 double posA[3], posB[3];
617 double velA[3], velB[3];
618 double vxab, vyab, vzab;
619 double rab[3];
620 int a, b, ax, ay, az, bx, by, bz;
621 double rma, rmb;
622 double dx, dy, dz;
623 double rvab;
624 double gab;
625 int iteration;
626
627 for (i = 0; i < nAtoms; i++){
628 moving[i] = 0;
629 moved[i] = 1;
630 }
631
632 done = 0;
633 iteration = 0;
634 while (!done && (iteration < maxIteration)){
635 done = 1;
636
637 for (i = 0; i < nConstrained; i++){
638 a = constrainedA[i];
639 b = constrainedB[i];
640
641 ax = (a * 3) + 0;
642 ay = (a * 3) + 1;
643 az = (a * 3) + 2;
644
645 bx = (b * 3) + 0;
646 by = (b * 3) + 1;
647 bz = (b * 3) + 2;
648
649 if (moved[a] || moved[b]){
650 atoms[a]->getVel(velA);
651 atoms[b]->getVel(velB);
652
653 vxab = velA[0] - velB[0];
654 vyab = velA[1] - velB[1];
655 vzab = velA[2] - velB[2];
656
657 atoms[a]->getPos(posA);
658 atoms[b]->getPos(posB);
659
660 for (j = 0; j < 3; j++)
661 rab[j] = posA[j] - posB[j];
662
663 info->wrapVector(rab);
664
665 rma = 1.0 / atoms[a]->getMass();
666 rmb = 1.0 / atoms[b]->getMass();
667
668 rvab = rab[0] * vxab + rab[1] * vyab + rab[2] * vzab;
669
670 gab = -rvab / ((rma + rmb) * constrainedDsqr[i]);
671
672 if (fabs(gab) > tol){
673 dx = rab[0] * gab;
674 dy = rab[1] * gab;
675 dz = rab[2] * gab;
676
677 velA[0] += rma * dx;
678 velA[1] += rma * dy;
679 velA[2] += rma * dz;
680
681 atoms[a]->setVel(velA);
682
683 velB[0] -= rmb * dx;
684 velB[1] -= rmb * dy;
685 velB[2] -= rmb * dz;
686
687 atoms[b]->setVel(velB);
688
689 moving[a] = 1;
690 moving[b] = 1;
691 done = 0;
692 }
693 }
694 }
695
696 for (i = 0; i < nAtoms; i++){
697 moved[i] = moving[i];
698 moving[i] = 0;
699 }
700
701 iteration++;
702 }
703
704 if (!done){
705 sprintf(painCave.errMsg,
706 "Constraint failure in constrainB, too many iterations: %d\n",
707 iteration);
708 painCave.isFatal = 1;
709 simError();
710 }
711 }
712 */
713 template<typename T> void Integrator<T>::rotationPropagation
714 ( StuntDouble* sd, double ji[3] ){
715
716 double angle;
717 double A[3][3], I[3][3];
718 int i, j, k;
719
720 // use the angular velocities to propagate the rotation matrix a
721 // full time step
722
723 sd->getA(A);
724 sd->getI(I);
725
726 if (sd->isLinear()) {
727 i = sd->linearAxis();
728 j = (i+1)%3;
729 k = (i+2)%3;
730
731 angle = dt2 * ji[j] / I[j][j];
732 this->rotate( k, i, angle, ji, A );
733
734 angle = dt * ji[k] / I[k][k];
735 this->rotate( i, j, angle, ji, A);
736
737 angle = dt2 * ji[j] / I[j][j];
738 this->rotate( k, i, angle, ji, A );
739
740 } else {
741 // rotate about the x-axis
742 angle = dt2 * ji[0] / I[0][0];
743 this->rotate( 1, 2, angle, ji, A );
744
745 // rotate about the y-axis
746 angle = dt2 * ji[1] / I[1][1];
747 this->rotate( 2, 0, angle, ji, A );
748
749 // rotate about the z-axis
750 angle = dt * ji[2] / I[2][2];
751 sd->addZangle(angle);
752 this->rotate( 0, 1, angle, ji, A);
753
754 // rotate about the y-axis
755 angle = dt2 * ji[1] / I[1][1];
756 this->rotate( 2, 0, angle, ji, A );
757
758 // rotate about the x-axis
759 angle = dt2 * ji[0] / I[0][0];
760 this->rotate( 1, 2, angle, ji, A );
761
762 }
763 sd->setA( A );
764 }
765
766 template<typename T> void Integrator<T>::rotate(int axes1, int axes2,
767 double angle, double ji[3],
768 double A[3][3]){
769 int i, j, k;
770 double sinAngle;
771 double cosAngle;
772 double angleSqr;
773 double angleSqrOver4;
774 double top, bottom;
775 double rot[3][3];
776 double tempA[3][3];
777 double tempJ[3];
778
779 // initialize the tempA
780
781 for (i = 0; i < 3; i++){
782 for (j = 0; j < 3; j++){
783 tempA[j][i] = A[i][j];
784 }
785 }
786
787 // initialize the tempJ
788
789 for (i = 0; i < 3; i++)
790 tempJ[i] = ji[i];
791
792 // initalize rot as a unit matrix
793
794 rot[0][0] = 1.0;
795 rot[0][1] = 0.0;
796 rot[0][2] = 0.0;
797
798 rot[1][0] = 0.0;
799 rot[1][1] = 1.0;
800 rot[1][2] = 0.0;
801
802 rot[2][0] = 0.0;
803 rot[2][1] = 0.0;
804 rot[2][2] = 1.0;
805
806 // use a small angle aproximation for sin and cosine
807
808 angleSqr = angle * angle;
809 angleSqrOver4 = angleSqr / 4.0;
810 top = 1.0 - angleSqrOver4;
811 bottom = 1.0 + angleSqrOver4;
812
813 cosAngle = top / bottom;
814 sinAngle = angle / bottom;
815
816 rot[axes1][axes1] = cosAngle;
817 rot[axes2][axes2] = cosAngle;
818
819 rot[axes1][axes2] = sinAngle;
820 rot[axes2][axes1] = -sinAngle;
821
822 // rotate the momentum acoording to: ji[] = rot[][] * ji[]
823
824 for (i = 0; i < 3; i++){
825 ji[i] = 0.0;
826 for (k = 0; k < 3; k++){
827 ji[i] += rot[i][k] * tempJ[k];
828 }
829 }
830
831 // rotate the Rotation matrix acording to:
832 // A[][] = A[][] * transpose(rot[][])
833
834
835 // NOte for as yet unknown reason, we are performing the
836 // calculation as:
837 // transpose(A[][]) = transpose(A[][]) * transpose(rot[][])
838
839 for (i = 0; i < 3; i++){
840 for (j = 0; j < 3; j++){
841 A[j][i] = 0.0;
842 for (k = 0; k < 3; k++){
843 A[j][i] += tempA[i][k] * rot[j][k];
844 }
845 }
846 }
847 }
848
849 template<typename T> void Integrator<T>::calcForce(int calcPot, int calcStress){
850 myFF->doForces(calcPot, calcStress);
851 }
852
853 template<typename T> void Integrator<T>::thermalize(){
854 tStats->velocitize();
855 }
856
857 template<typename T> double Integrator<T>::getConservedQuantity(void){
858 return tStats->getTotalE();
859 }
860 template<typename T> string Integrator<T>::getAdditionalParameters(void){
861 //By default, return a null string
862 //The reason we use string instead of char* is that if we use char*, we will
863 //return a pointer point to local variable which might cause problem
864 return string();
865 }