1 #define PETSCTS_DLL 2 3 /* 4 * Code for Timestepping with Runge Kutta 5 * 6 * Written by 7 * Asbjorn Hoiland Aarrestad 8 * asbjorn@aarrestad.com 9 * http://asbjorn.aarrestad.com/ 10 * 11 */ 12 #include "private/tsimpl.h" /*I "petscts.h" I*/ 13 #include "time.h" 14 15 typedef struct { 16 Vec y1,y2; /* work wectors for the two rk permuations */ 17 PetscInt nok,nnok; /* counters for ok and not ok steps */ 18 PetscReal maxerror; /* variable to tell the maxerror allowed */ 19 PetscReal ferror; /* variable to tell (global maxerror)/(total time) */ 20 PetscReal tolerance; /* initial value set for maxerror by user */ 21 Vec tmp,tmp_y,*k; /* two temp vectors and the k vectors for rk */ 22 PetscScalar a[7][6]; /* rk scalars */ 23 PetscScalar b1[7],b2[7]; /* rk scalars */ 24 PetscReal c[7]; /* rk scalars */ 25 PetscInt p,s; /* variables to tell the size of the runge-kutta solver */ 26 } TS_RK; 27 28 EXTERN_C_BEGIN 29 #undef __FUNCT__ 30 #define __FUNCT__ "TSRKSetTolerance_RK" 31 PetscErrorCode PETSCTS_DLLEXPORT TSRKSetTolerance_RK(TS ts,PetscReal aabs) 32 { 33 TS_RK *rk = (TS_RK*)ts->data; 34 35 PetscFunctionBegin; 36 rk->tolerance = aabs; 37 PetscFunctionReturn(0); 38 } 39 EXTERN_C_END 40 41 #undef __FUNCT__ 42 #define __FUNCT__ "TSRKSetTolerance" 43 /*@ 44 TSRKSetTolerance - Sets the total error the RK explicit time integrators 45 will allow over the given time interval. 46 47 Logically Collective on TS 48 49 Input parameters: 50 + ts - the time-step context 51 - aabs - the absolute tolerance 52 53 Level: intermediate 54 55 .keywords: RK, tolerance 56 57 .seealso: TSSundialsSetTolerance() 58 59 @*/ 60 PetscErrorCode PETSCTS_DLLEXPORT TSRKSetTolerance(TS ts,PetscReal aabs) 61 { 62 PetscErrorCode ierr; 63 64 PetscFunctionBegin; 65 PetscValidLogicalCollectiveReal(ts,aabs,2); 66 ierr = PetscTryMethod(ts,"TSRKSetTolerance_C",(TS,PetscReal),(ts,aabs));CHKERRQ(ierr); 67 PetscFunctionReturn(0); 68 } 69 70 71 #undef __FUNCT__ 72 #define __FUNCT__ "TSSetUp_RK" 73 static PetscErrorCode TSSetUp_RK(TS ts) 74 { 75 TS_RK *rk = (TS_RK*)ts->data; 76 PetscErrorCode ierr; 77 78 PetscFunctionBegin; 79 rk->nok = 0; 80 rk->nnok = 0; 81 rk->maxerror = rk->tolerance; 82 83 /* fixing maxerror: global vs local */ 84 rk->ferror = rk->maxerror / (ts->max_time - ts->ptime); 85 86 /* 34.0/45.0 gives double precision division */ 87 /* defining variables needed for Runge-Kutta computing */ 88 /* when changing below, please remember to change a, b1, b2 and c above! */ 89 /* Found in table on page 171: Dormand-Prince 5(4) */ 90 91 /* are these right? */ 92 rk->p=6; 93 rk->s=7; 94 95 rk->a[1][0]=1.0/5.0; 96 rk->a[2][0]=3.0/40.0; 97 rk->a[2][1]=9.0/40.0; 98 rk->a[3][0]=44.0/45.0; 99 rk->a[3][1]=-56.0/15.0; 100 rk->a[3][2]=32.0/9.0; 101 rk->a[4][0]=19372.0/6561.0; 102 rk->a[4][1]=-25360.0/2187.0; 103 rk->a[4][2]=64448.0/6561.0; 104 rk->a[4][3]=-212.0/729.0; 105 rk->a[5][0]=9017.0/3168.0; 106 rk->a[5][1]=-355.0/33.0; 107 rk->a[5][2]=46732.0/5247.0; 108 rk->a[5][3]=49.0/176.0; 109 rk->a[5][4]=-5103.0/18656.0; 110 rk->a[6][0]=35.0/384.0; 111 rk->a[6][1]=0.0; 112 rk->a[6][2]=500.0/1113.0; 113 rk->a[6][3]=125.0/192.0; 114 rk->a[6][4]=-2187.0/6784.0; 115 rk->a[6][5]=11.0/84.0; 116 117 118 rk->c[0]=0.0; 119 rk->c[1]=1.0/5.0; 120 rk->c[2]=3.0/10.0; 121 rk->c[3]=4.0/5.0; 122 rk->c[4]=8.0/9.0; 123 rk->c[5]=1.0; 124 rk->c[6]=1.0; 125 126 rk->b1[0]=35.0/384.0; 127 rk->b1[1]=0.0; 128 rk->b1[2]=500.0/1113.0; 129 rk->b1[3]=125.0/192.0; 130 rk->b1[4]=-2187.0/6784.0; 131 rk->b1[5]=11.0/84.0; 132 rk->b1[6]=0.0; 133 134 rk->b2[0]=5179.0/57600.0; 135 rk->b2[1]=0.0; 136 rk->b2[2]=7571.0/16695.0; 137 rk->b2[3]=393.0/640.0; 138 rk->b2[4]=-92097.0/339200.0; 139 rk->b2[5]=187.0/2100.0; 140 rk->b2[6]=1.0/40.0; 141 142 143 /* Found in table on page 170: Fehlberg 4(5) */ 144 /* 145 rk->p=5; 146 rk->s=6; 147 148 rk->a[1][0]=1.0/4.0; 149 rk->a[2][0]=3.0/32.0; 150 rk->a[2][1]=9.0/32.0; 151 rk->a[3][0]=1932.0/2197.0; 152 rk->a[3][1]=-7200.0/2197.0; 153 rk->a[3][2]=7296.0/2197.0; 154 rk->a[4][0]=439.0/216.0; 155 rk->a[4][1]=-8.0; 156 rk->a[4][2]=3680.0/513.0; 157 rk->a[4][3]=-845.0/4104.0; 158 rk->a[5][0]=-8.0/27.0; 159 rk->a[5][1]=2.0; 160 rk->a[5][2]=-3544.0/2565.0; 161 rk->a[5][3]=1859.0/4104.0; 162 rk->a[5][4]=-11.0/40.0; 163 164 rk->c[0]=0.0; 165 rk->c[1]=1.0/4.0; 166 rk->c[2]=3.0/8.0; 167 rk->c[3]=12.0/13.0; 168 rk->c[4]=1.0; 169 rk->c[5]=1.0/2.0; 170 171 rk->b1[0]=25.0/216.0; 172 rk->b1[1]=0.0; 173 rk->b1[2]=1408.0/2565.0; 174 rk->b1[3]=2197.0/4104.0; 175 rk->b1[4]=-1.0/5.0; 176 rk->b1[5]=0.0; 177 178 rk->b2[0]=16.0/135.0; 179 rk->b2[1]=0.0; 180 rk->b2[2]=6656.0/12825.0; 181 rk->b2[3]=28561.0/56430.0; 182 rk->b2[4]=-9.0/50.0; 183 rk->b2[5]=2.0/55.0; 184 */ 185 /* Found in table on page 169: Merson 4("5") */ 186 /* 187 rk->p=4; 188 rk->s=5; 189 rk->a[1][0] = 1.0/3.0; 190 rk->a[2][0] = 1.0/6.0; 191 rk->a[2][1] = 1.0/6.0; 192 rk->a[3][0] = 1.0/8.0; 193 rk->a[3][1] = 0.0; 194 rk->a[3][2] = 3.0/8.0; 195 rk->a[4][0] = 1.0/2.0; 196 rk->a[4][1] = 0.0; 197 rk->a[4][2] = -3.0/2.0; 198 rk->a[4][3] = 2.0; 199 200 rk->c[0] = 0.0; 201 rk->c[1] = 1.0/3.0; 202 rk->c[2] = 1.0/3.0; 203 rk->c[3] = 0.5; 204 rk->c[4] = 1.0; 205 206 rk->b1[0] = 1.0/2.0; 207 rk->b1[1] = 0.0; 208 rk->b1[2] = -3.0/2.0; 209 rk->b1[3] = 2.0; 210 rk->b1[4] = 0.0; 211 212 rk->b2[0] = 1.0/6.0; 213 rk->b2[1] = 0.0; 214 rk->b2[2] = 0.0; 215 rk->b2[3] = 2.0/3.0; 216 rk->b2[4] = 1.0/6.0; 217 */ 218 219 /* making b2 -> e=b1-b2 */ 220 /* 221 for(i=0;i<rk->s;i++){ 222 rk->b2[i] = (rk->b1[i]) - (rk->b2[i]); 223 } 224 */ 225 rk->b2[0]=71.0/57600.0; 226 rk->b2[1]=0.0; 227 rk->b2[2]=-71.0/16695.0; 228 rk->b2[3]=71.0/1920.0; 229 rk->b2[4]=-17253.0/339200.0; 230 rk->b2[5]=22.0/525.0; 231 rk->b2[6]=-1.0/40.0; 232 233 /* initializing vectors */ 234 ierr = VecDuplicate(ts->vec_sol,&rk->y1);CHKERRQ(ierr); 235 ierr = VecDuplicate(ts->vec_sol,&rk->y2);CHKERRQ(ierr); 236 ierr = VecDuplicate(rk->y1,&rk->tmp);CHKERRQ(ierr); 237 ierr = VecDuplicate(rk->y1,&rk->tmp_y);CHKERRQ(ierr); 238 ierr = VecDuplicateVecs(rk->y1,rk->s,&rk->k);CHKERRQ(ierr); 239 240 PetscFunctionReturn(0); 241 } 242 243 /*------------------------------------------------------------*/ 244 #undef __FUNCT__ 245 #define __FUNCT__ "TSRKqs" 246 PetscErrorCode TSRKqs(TS ts,PetscReal t,PetscReal h) 247 { 248 TS_RK *rk = (TS_RK*)ts->data; 249 PetscErrorCode ierr; 250 PetscInt j,l; 251 PetscReal tmp_t=t; 252 PetscScalar hh=h; 253 254 PetscFunctionBegin; 255 /* k[0]=0 */ 256 ierr = VecSet(rk->k[0],0.0);CHKERRQ(ierr); 257 258 /* k[0] = derivs(t,y1) */ 259 ierr = TSComputeRHSFunction(ts,t,rk->y1,rk->k[0]);CHKERRQ(ierr); 260 /* looping over runge-kutta variables */ 261 /* building the k - array of vectors */ 262 for(j = 1 ; j < rk->s ; j++){ 263 264 /* rk->tmp = 0 */ 265 ierr = VecSet(rk->tmp,0.0);CHKERRQ(ierr); 266 267 for(l=0;l<j;l++){ 268 /* tmp += a(j,l)*k[l] */ 269 ierr = VecAXPY(rk->tmp,rk->a[j][l],rk->k[l]);CHKERRQ(ierr); 270 } 271 272 /* ierr = VecView(rk->tmp,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr); */ 273 274 /* k[j] = derivs(t+c(j)*h,y1+h*tmp,k(j)) */ 275 /* I need the following helpers: 276 PetscScalar tmp_t=t+c(j)*h 277 Vec tmp_y=h*tmp+y1 278 */ 279 280 tmp_t = t + rk->c[j] * h; 281 282 /* tmp_y = h * tmp + y1 */ 283 ierr = VecWAXPY(rk->tmp_y,hh,rk->tmp,rk->y1);CHKERRQ(ierr); 284 285 /* rk->k[j]=0 */ 286 ierr = VecSet(rk->k[j],0.0);CHKERRQ(ierr); 287 ierr = TSComputeRHSFunction(ts,tmp_t,rk->tmp_y,rk->k[j]);CHKERRQ(ierr); 288 } 289 290 /* tmp=0 and tmp_y=0 */ 291 ierr = VecSet(rk->tmp,0.0);CHKERRQ(ierr); 292 ierr = VecSet(rk->tmp_y,0.0);CHKERRQ(ierr); 293 294 for(j = 0 ; j < rk->s ; j++){ 295 /* tmp=b1[j]*k[j]+tmp */ 296 ierr = VecAXPY(rk->tmp,rk->b1[j],rk->k[j]);CHKERRQ(ierr); 297 /* tmp_y=b2[j]*k[j]+tmp_y */ 298 ierr = VecAXPY(rk->tmp_y,rk->b2[j],rk->k[j]);CHKERRQ(ierr); 299 } 300 301 /* y2 = hh * tmp_y */ 302 ierr = VecSet(rk->y2,0.0);CHKERRQ(ierr); 303 ierr = VecAXPY(rk->y2,hh,rk->tmp_y);CHKERRQ(ierr); 304 /* y1 = hh*tmp + y1 */ 305 ierr = VecAXPY(rk->y1,hh,rk->tmp);CHKERRQ(ierr); 306 /* Finding difference between y1 and y2 */ 307 PetscFunctionReturn(0); 308 } 309 310 #undef __FUNCT__ 311 #define __FUNCT__ "TSStep_RK" 312 static PetscErrorCode TSStep_RK(TS ts,PetscInt *steps,PetscReal *ptime) 313 { 314 TS_RK *rk = (TS_RK*)ts->data; 315 PetscErrorCode ierr; 316 PetscReal norm=0.0,dt_fac=0.0,fac = 0.0/*,ttmp=0.0*/; 317 PetscInt i, max_steps = ts->max_steps; 318 319 PetscFunctionBegin; 320 ierr=VecCopy(ts->vec_sol,rk->y1);CHKERRQ(ierr); 321 *steps = -ts->steps; 322 /* trying to save the vector */ 323 ierr = TSMonitor(ts,ts->steps,ts->ptime,ts->vec_sol);CHKERRQ(ierr); 324 /* while loop to get from start to stop */ 325 for (i = 0; i < max_steps; i++) { 326 ierr = TSPreStep(ts);CHKERRQ(ierr); /* Note that this is called once per STEP, not once per STAGE. */ 327 /* calling rkqs */ 328 /* 329 -- input 330 ts - pointer to ts 331 ts->ptime - current time 332 ts->time_step - try this timestep 333 y1 - solution for this step 334 335 --output 336 y1 - suggested solution 337 y2 - check solution (runge - kutta second permutation) 338 */ 339 ierr = TSRKqs(ts,ts->ptime,ts->time_step);CHKERRQ(ierr); 340 /* counting steps */ 341 ts->steps++; 342 /* checking for maxerror */ 343 /* comparing difference to maxerror */ 344 ierr = VecNorm(rk->y2,NORM_2,&norm);CHKERRQ(ierr); 345 /* modifying maxerror to satisfy this timestep */ 346 rk->maxerror = rk->ferror * ts->time_step; 347 /* ierr = PetscPrintf(PETSC_COMM_WORLD,"norm err: %f maxerror: %f dt: %f",norm,rk->maxerror,ts->time_step);CHKERRQ(ierr); */ 348 349 /* handling ok and not ok */ 350 if (norm < rk->maxerror){ 351 /* if ok: */ 352 ierr=VecCopy(rk->y1,ts->vec_sol);CHKERRQ(ierr); /* saves the suggested solution to current solution */ 353 ts->ptime += ts->time_step; /* storing the new current time */ 354 rk->nok++; 355 fac=5.0; 356 /* trying to save the vector */ 357 ierr = TSPostStep(ts);CHKERRQ(ierr); 358 ierr = TSMonitor(ts,ts->steps,ts->ptime,ts->vec_sol);CHKERRQ(ierr); 359 if (ts->ptime >= ts->max_time) break; 360 } else{ 361 /* if not OK */ 362 rk->nnok++; 363 fac=1.0; 364 ierr=VecCopy(ts->vec_sol,rk->y1);CHKERRQ(ierr); /* restores old solution */ 365 } 366 367 /*Computing next stepsize. See page 167 in Solving ODE 1 368 * 369 * h_new = h * min( facmax , max( facmin , fac * (tol/err)^(1/(p+1)) ) ) 370 * facmax set above 371 * facmin 372 */ 373 dt_fac = exp(log((rk->maxerror) / norm) / ((rk->p) + 1) ) * 0.9 ; 374 375 if (dt_fac > fac){ 376 /*ierr = PetscPrintf(PETSC_COMM_WORLD,"changing fac %f\n",fac);*/ 377 dt_fac = fac; 378 } 379 380 /* computing new ts->time_step */ 381 ts->time_step = ts->time_step * dt_fac; 382 383 if (ts->ptime+ts->time_step > ts->max_time){ 384 ts->time_step = ts->max_time - ts->ptime; 385 } 386 387 if (ts->time_step < 1e-14){ 388 ierr = PetscPrintf(PETSC_COMM_WORLD,"Very small steps: %f\n",ts->time_step);CHKERRQ(ierr); 389 ts->time_step = 1e-14; 390 } 391 392 /* trying to purify h */ 393 /* (did not give any visible result) */ 394 /* ttmp = ts->ptime + ts->time_step; 395 ts->time_step = ttmp - ts->ptime; */ 396 397 } 398 399 ierr=VecCopy(rk->y1,ts->vec_sol);CHKERRQ(ierr); 400 *steps += ts->steps; 401 *ptime = ts->ptime; 402 PetscFunctionReturn(0); 403 } 404 405 /*------------------------------------------------------------*/ 406 #undef __FUNCT__ 407 #define __FUNCT__ "TSDestroy_RK" 408 static PetscErrorCode TSDestroy_RK(TS ts) 409 { 410 TS_RK *rk = (TS_RK*)ts->data; 411 PetscErrorCode ierr; 412 PetscInt i; 413 414 /* REMEMBER TO DESTROY ALL */ 415 416 PetscFunctionBegin; 417 if (rk->y1) {ierr = VecDestroy(rk->y1);CHKERRQ(ierr);} 418 if (rk->y2) {ierr = VecDestroy(rk->y2);CHKERRQ(ierr);} 419 if (rk->tmp) {ierr = VecDestroy(rk->tmp);CHKERRQ(ierr);} 420 if (rk->tmp_y) {ierr = VecDestroy(rk->tmp_y);CHKERRQ(ierr);} 421 for(i=0;i<rk->s;i++){ 422 if (rk->k[i]) {ierr = VecDestroy(rk->k[i]);CHKERRQ(ierr);} 423 } 424 ierr = PetscFree(rk);CHKERRQ(ierr); 425 PetscFunctionReturn(0); 426 } 427 /*------------------------------------------------------------*/ 428 429 #undef __FUNCT__ 430 #define __FUNCT__ "TSSetFromOptions_RK" 431 static PetscErrorCode TSSetFromOptions_RK(TS ts) 432 { 433 TS_RK *rk = (TS_RK*)ts->data; 434 PetscErrorCode ierr; 435 436 PetscFunctionBegin; 437 ierr = PetscOptionsHead("RK ODE solver options");CHKERRQ(ierr); 438 ierr = PetscOptionsReal("-ts_rk_tol","Tolerance for convergence","TSRKSetTolerance",rk->tolerance,&rk->tolerance,PETSC_NULL);CHKERRQ(ierr); 439 ierr = PetscOptionsTail();CHKERRQ(ierr); 440 PetscFunctionReturn(0); 441 } 442 443 #undef __FUNCT__ 444 #define __FUNCT__ "TSView_RK" 445 static PetscErrorCode TSView_RK(TS ts,PetscViewer viewer) 446 { 447 TS_RK *rk = (TS_RK*)ts->data; 448 PetscErrorCode ierr; 449 450 PetscFunctionBegin; 451 ierr = PetscPrintf(PETSC_COMM_WORLD," number of ok steps: %D\n",rk->nok);CHKERRQ(ierr); 452 ierr = PetscPrintf(PETSC_COMM_WORLD," number of rejected steps: %D\n",rk->nnok);CHKERRQ(ierr); 453 PetscFunctionReturn(0); 454 } 455 456 /* ------------------------------------------------------------ */ 457 /*MC 458 TSRK - ODE solver using the explicit Runge-Kutta methods 459 460 Options Database: 461 . -ts_rk_tol <tol> Tolerance for convergence 462 463 Contributed by: Asbjorn Hoiland Aarrestad, asbjorn@aarrestad.com, http://asbjorn.aarrestad.com/ 464 465 Level: beginner 466 467 .seealso: TSCreate(), TS, TSSetType(), TSEULER, TSRKSetTolerance() 468 469 M*/ 470 471 EXTERN_C_BEGIN 472 #undef __FUNCT__ 473 #define __FUNCT__ "TSCreate_RK" 474 PetscErrorCode PETSCTS_DLLEXPORT TSCreate_RK(TS ts) 475 { 476 TS_RK *rk; 477 PetscErrorCode ierr; 478 479 PetscFunctionBegin; 480 ts->ops->setup = TSSetUp_RK; 481 ts->ops->step = TSStep_RK; 482 ts->ops->destroy = TSDestroy_RK; 483 ts->ops->setfromoptions = TSSetFromOptions_RK; 484 ts->ops->view = TSView_RK; 485 486 ierr = PetscNewLog(ts,TS_RK,&rk);CHKERRQ(ierr); 487 ts->data = (void*)rk; 488 489 ierr = PetscObjectComposeFunctionDynamic((PetscObject)ts,"TSRKSetTolerance_C","TSRKSetTolerance_RK",TSRKSetTolerance_RK);CHKERRQ(ierr); 490 491 PetscFunctionReturn(0); 492 } 493 EXTERN_C_END 494 495 496 497 498