1 /* 2 Code for timestepping with implicit Theta method 3 */ 4 #include <petsc/private/tsimpl.h> /*I "petscts.h" I*/ 5 #include <petscsnes.h> 6 #include <petscdm.h> 7 #include <petscmat.h> 8 9 typedef struct { 10 /* context for time stepping */ 11 PetscReal stage_time; 12 Vec X0,X,Xdot; /* Storage for stages and time derivative */ 13 Vec affine; /* Affine vector needed for residual at beginning of step in endpoint formulation */ 14 PetscReal Theta; 15 PetscReal ptime; 16 PetscReal time_step; 17 PetscInt order; 18 PetscBool endpoint; 19 PetscBool extrapolate; 20 TSStepStatus status; 21 Vec VecCostIntegral0; /* Backup for roll-backs due to events */ 22 23 /* context for sensitivity analysis */ 24 PetscInt num_tlm; /* Total number of tangent linear equations */ 25 Vec *VecsDeltaLam; /* Increment of the adjoint sensitivity w.r.t IC at stage */ 26 Vec *VecsDeltaMu; /* Increment of the adjoint sensitivity w.r.t P at stage */ 27 Vec *VecsSensiTemp; /* Vector to be multiplied with Jacobian transpose */ 28 Mat MatDeltaFwdSensip; /* Increment of the forward sensitivity at stage */ 29 Vec VecDeltaFwdSensipCol; /* Working vector for holding one column of the sensitivity matrix */ 30 Mat MatFwdSensip0; /* backup for roll-backs due to events */ 31 Vec VecIntegralSensipTemp; /* Working vector for forward integral sensitivity */ 32 Vec *VecsIntegralSensip0; /* backup for roll-backs due to events */ 33 Vec *VecsDeltaLam2; /* Increment of the 2nd-order adjoint sensitivity w.r.t IC at stage */ 34 Vec *VecsDeltaMu2; /* Increment of the 2nd-order adjoint sensitivity w.r.t P at stage */ 35 Vec *VecsSensi2Temp; /* Working vectors that holds the residual for the second-order adjoint */ 36 Vec *VecsAffine; /* Working vectors to store residuals */ 37 /* context for error estimation */ 38 Vec vec_sol_prev; 39 Vec vec_lte_work; 40 } TS_Theta; 41 42 static PetscErrorCode TSThetaGetX0AndXdot(TS ts,DM dm,Vec *X0,Vec *Xdot) 43 { 44 TS_Theta *th = (TS_Theta*)ts->data; 45 PetscErrorCode ierr; 46 47 PetscFunctionBegin; 48 if (X0) { 49 if (dm && dm != ts->dm) { 50 ierr = DMGetNamedGlobalVector(dm,"TSTheta_X0",X0);CHKERRQ(ierr); 51 } else *X0 = ts->vec_sol; 52 } 53 if (Xdot) { 54 if (dm && dm != ts->dm) { 55 ierr = DMGetNamedGlobalVector(dm,"TSTheta_Xdot",Xdot);CHKERRQ(ierr); 56 } else *Xdot = th->Xdot; 57 } 58 PetscFunctionReturn(0); 59 } 60 61 static PetscErrorCode TSThetaRestoreX0AndXdot(TS ts,DM dm,Vec *X0,Vec *Xdot) 62 { 63 PetscErrorCode ierr; 64 65 PetscFunctionBegin; 66 if (X0) { 67 if (dm && dm != ts->dm) { 68 ierr = DMRestoreNamedGlobalVector(dm,"TSTheta_X0",X0);CHKERRQ(ierr); 69 } 70 } 71 if (Xdot) { 72 if (dm && dm != ts->dm) { 73 ierr = DMRestoreNamedGlobalVector(dm,"TSTheta_Xdot",Xdot);CHKERRQ(ierr); 74 } 75 } 76 PetscFunctionReturn(0); 77 } 78 79 static PetscErrorCode DMCoarsenHook_TSTheta(DM fine,DM coarse,void *ctx) 80 { 81 PetscFunctionBegin; 82 PetscFunctionReturn(0); 83 } 84 85 static PetscErrorCode DMRestrictHook_TSTheta(DM fine,Mat restrct,Vec rscale,Mat inject,DM coarse,void *ctx) 86 { 87 TS ts = (TS)ctx; 88 PetscErrorCode ierr; 89 Vec X0,Xdot,X0_c,Xdot_c; 90 91 PetscFunctionBegin; 92 ierr = TSThetaGetX0AndXdot(ts,fine,&X0,&Xdot);CHKERRQ(ierr); 93 ierr = TSThetaGetX0AndXdot(ts,coarse,&X0_c,&Xdot_c);CHKERRQ(ierr); 94 ierr = MatRestrict(restrct,X0,X0_c);CHKERRQ(ierr); 95 ierr = MatRestrict(restrct,Xdot,Xdot_c);CHKERRQ(ierr); 96 ierr = VecPointwiseMult(X0_c,rscale,X0_c);CHKERRQ(ierr); 97 ierr = VecPointwiseMult(Xdot_c,rscale,Xdot_c);CHKERRQ(ierr); 98 ierr = TSThetaRestoreX0AndXdot(ts,fine,&X0,&Xdot);CHKERRQ(ierr); 99 ierr = TSThetaRestoreX0AndXdot(ts,coarse,&X0_c,&Xdot_c);CHKERRQ(ierr); 100 PetscFunctionReturn(0); 101 } 102 103 static PetscErrorCode DMSubDomainHook_TSTheta(DM dm,DM subdm,void *ctx) 104 { 105 PetscFunctionBegin; 106 PetscFunctionReturn(0); 107 } 108 109 static PetscErrorCode DMSubDomainRestrictHook_TSTheta(DM dm,VecScatter gscat,VecScatter lscat,DM subdm,void *ctx) 110 { 111 TS ts = (TS)ctx; 112 PetscErrorCode ierr; 113 Vec X0,Xdot,X0_sub,Xdot_sub; 114 115 PetscFunctionBegin; 116 ierr = TSThetaGetX0AndXdot(ts,dm,&X0,&Xdot);CHKERRQ(ierr); 117 ierr = TSThetaGetX0AndXdot(ts,subdm,&X0_sub,&Xdot_sub);CHKERRQ(ierr); 118 119 ierr = VecScatterBegin(gscat,X0,X0_sub,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 120 ierr = VecScatterEnd(gscat,X0,X0_sub,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 121 122 ierr = VecScatterBegin(gscat,Xdot,Xdot_sub,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 123 ierr = VecScatterEnd(gscat,Xdot,Xdot_sub,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr); 124 125 ierr = TSThetaRestoreX0AndXdot(ts,dm,&X0,&Xdot);CHKERRQ(ierr); 126 ierr = TSThetaRestoreX0AndXdot(ts,subdm,&X0_sub,&Xdot_sub);CHKERRQ(ierr); 127 PetscFunctionReturn(0); 128 } 129 130 static PetscErrorCode TSThetaEvaluateCostIntegral(TS ts) 131 { 132 TS_Theta *th = (TS_Theta*)ts->data; 133 PetscErrorCode ierr; 134 135 PetscFunctionBegin; 136 if (th->endpoint) { 137 /* Evolve ts->vec_costintegral to compute integrals */ 138 if (th->Theta!=1.0) { 139 ierr = TSComputeCostIntegrand(ts,th->ptime,th->X0,ts->vec_costintegrand);CHKERRQ(ierr); 140 ierr = VecAXPY(ts->vec_costintegral,th->time_step*(1.0-th->Theta),ts->vec_costintegrand);CHKERRQ(ierr); 141 } 142 ierr = TSComputeCostIntegrand(ts,ts->ptime,ts->vec_sol,ts->vec_costintegrand);CHKERRQ(ierr); 143 ierr = VecAXPY(ts->vec_costintegral,th->time_step*th->Theta,ts->vec_costintegrand);CHKERRQ(ierr); 144 } else { 145 ierr = TSComputeCostIntegrand(ts,th->stage_time,th->X,ts->vec_costintegrand);CHKERRQ(ierr); 146 ierr = VecAXPY(ts->vec_costintegral,th->time_step,ts->vec_costintegrand);CHKERRQ(ierr); 147 } 148 PetscFunctionReturn(0); 149 } 150 151 static PetscErrorCode TSForwardCostIntegral_Theta(TS ts) 152 { 153 TS_Theta *th = (TS_Theta*)ts->data; 154 PetscErrorCode ierr; 155 156 PetscFunctionBegin; 157 /* backup cost integral */ 158 ierr = VecCopy(ts->vec_costintegral,th->VecCostIntegral0);CHKERRQ(ierr); 159 ierr = TSThetaEvaluateCostIntegral(ts);CHKERRQ(ierr); 160 PetscFunctionReturn(0); 161 } 162 163 static PetscErrorCode TSAdjointCostIntegral_Theta(TS ts) 164 { 165 PetscErrorCode ierr; 166 167 PetscFunctionBegin; 168 ierr = TSThetaEvaluateCostIntegral(ts);CHKERRQ(ierr); 169 PetscFunctionReturn(0); 170 } 171 172 static PetscErrorCode TSTheta_SNESSolve(TS ts,Vec b,Vec x) 173 { 174 PetscInt nits,lits; 175 PetscErrorCode ierr; 176 177 PetscFunctionBegin; 178 ierr = SNESSolve(ts->snes,b,x);CHKERRQ(ierr); 179 ierr = SNESGetIterationNumber(ts->snes,&nits);CHKERRQ(ierr); 180 ierr = SNESGetLinearSolveIterations(ts->snes,&lits);CHKERRQ(ierr); 181 ts->snes_its += nits; ts->ksp_its += lits; 182 PetscFunctionReturn(0); 183 } 184 185 static PetscErrorCode TSStep_Theta(TS ts) 186 { 187 TS_Theta *th = (TS_Theta*)ts->data; 188 PetscInt rejections = 0; 189 PetscBool stageok,accept = PETSC_TRUE; 190 PetscReal next_time_step = ts->time_step; 191 PetscErrorCode ierr; 192 193 PetscFunctionBegin; 194 if (!ts->steprollback) { 195 if (th->vec_sol_prev) { ierr = VecCopy(th->X0,th->vec_sol_prev);CHKERRQ(ierr); } 196 ierr = VecCopy(ts->vec_sol,th->X0);CHKERRQ(ierr); 197 } 198 199 th->status = TS_STEP_INCOMPLETE; 200 while (!ts->reason && th->status != TS_STEP_COMPLETE) { 201 202 PetscReal shift = 1/(th->Theta*ts->time_step); 203 th->stage_time = ts->ptime + (th->endpoint ? (PetscReal)1 : th->Theta)*ts->time_step; 204 205 ierr = VecCopy(th->X0,th->X);CHKERRQ(ierr); 206 if (th->extrapolate && !ts->steprestart) { 207 ierr = VecAXPY(th->X,1/shift,th->Xdot);CHKERRQ(ierr); 208 } 209 if (th->endpoint) { /* This formulation assumes linear time-independent mass matrix */ 210 if (!th->affine) {ierr = VecDuplicate(ts->vec_sol,&th->affine);CHKERRQ(ierr);} 211 ierr = VecZeroEntries(th->Xdot);CHKERRQ(ierr); 212 ierr = TSComputeIFunction(ts,ts->ptime,th->X0,th->Xdot,th->affine,PETSC_FALSE);CHKERRQ(ierr); 213 ierr = VecScale(th->affine,(th->Theta-1)/th->Theta);CHKERRQ(ierr); 214 } else if (th->affine) { /* Just in case th->endpoint is changed between calls to TSStep_Theta() */ 215 ierr = VecZeroEntries(th->affine);CHKERRQ(ierr); 216 } 217 ierr = TSPreStage(ts,th->stage_time);CHKERRQ(ierr); 218 ierr = TSTheta_SNESSolve(ts,th->affine,th->X);CHKERRQ(ierr); 219 ierr = TSPostStage(ts,th->stage_time,0,&th->X);CHKERRQ(ierr); 220 ierr = TSAdaptCheckStage(ts->adapt,ts,th->stage_time,th->X,&stageok);CHKERRQ(ierr); 221 if (!stageok) goto reject_step; 222 223 th->status = TS_STEP_PENDING; 224 if (th->endpoint) { 225 ierr = VecCopy(th->X,ts->vec_sol);CHKERRQ(ierr); 226 } else { 227 ierr = VecAXPBYPCZ(th->Xdot,-shift,shift,0,th->X0,th->X);CHKERRQ(ierr); 228 ierr = VecAXPY(ts->vec_sol,ts->time_step,th->Xdot);CHKERRQ(ierr); 229 } 230 ierr = TSAdaptChoose(ts->adapt,ts,ts->time_step,NULL,&next_time_step,&accept);CHKERRQ(ierr); 231 th->status = accept ? TS_STEP_COMPLETE : TS_STEP_INCOMPLETE; 232 if (!accept) { 233 ierr = VecCopy(th->X0,ts->vec_sol);CHKERRQ(ierr); 234 ts->time_step = next_time_step; 235 goto reject_step; 236 } 237 238 if (ts->forward_solve || ts->costintegralfwd) { /* Save the info for the later use in cost integral evaluation */ 239 th->ptime = ts->ptime; 240 th->time_step = ts->time_step; 241 } 242 243 ts->ptime += ts->time_step; 244 ts->time_step = next_time_step; 245 break; 246 247 reject_step: 248 ts->reject++; accept = PETSC_FALSE; 249 if (!ts->reason && ++rejections > ts->max_reject && ts->max_reject >= 0) { 250 ts->reason = TS_DIVERGED_STEP_REJECTED; 251 ierr = PetscInfo2(ts,"Step=%D, step rejections %D greater than current TS allowed, stopping solve\n",ts->steps,rejections);CHKERRQ(ierr); 252 } 253 } 254 PetscFunctionReturn(0); 255 } 256 257 static PetscErrorCode TSAdjointStep_Theta(TS ts) 258 { 259 TS_Theta *th = (TS_Theta*)ts->data; 260 Vec *VecsDeltaLam = th->VecsDeltaLam,*VecsDeltaMu = th->VecsDeltaMu,*VecsSensiTemp = th->VecsSensiTemp; 261 Vec *VecsDeltaLam2 = th->VecsDeltaLam2,*VecsDeltaMu2 = th->VecsDeltaMu2,*VecsSensi2Temp = th->VecsSensi2Temp; 262 PetscInt nadj; 263 Mat J,Jp; 264 KSP ksp; 265 PetscReal shift; 266 PetscScalar *xarr; 267 PetscErrorCode ierr; 268 269 PetscFunctionBegin; 270 th->status = TS_STEP_INCOMPLETE; 271 ierr = SNESGetKSP(ts->snes,&ksp);CHKERRQ(ierr); 272 ierr = TSGetIJacobian(ts,&J,&Jp,NULL,NULL);CHKERRQ(ierr); 273 274 /* If endpoint=1, th->ptime and th->X0 will be used; if endpoint=0, th->stage_time and th->X will be used. */ 275 th->stage_time = th->endpoint ? ts->ptime : (ts->ptime+(1.-th->Theta)*ts->time_step); /* time_step is negative*/ 276 th->ptime = ts->ptime + ts->time_step; 277 th->time_step = -ts->time_step; 278 279 /* Build RHS for first-order adjoint */ 280 if (ts->vec_costintegral) { /* Cost function has an integral term */ 281 if (th->endpoint) { 282 ierr = TSComputeDRDUFunction(ts,th->stage_time,ts->vec_sol,ts->vecs_drdu);CHKERRQ(ierr); 283 } else { 284 ierr = TSComputeDRDUFunction(ts,th->stage_time,th->X,ts->vecs_drdu);CHKERRQ(ierr); 285 } 286 } 287 for (nadj=0; nadj<ts->numcost; nadj++) { 288 ierr = VecCopy(ts->vecs_sensi[nadj],VecsSensiTemp[nadj]);CHKERRQ(ierr); 289 ierr = VecScale(VecsSensiTemp[nadj],1./(th->Theta*th->time_step));CHKERRQ(ierr); 290 if (ts->vec_costintegral) { 291 ierr = VecAXPY(VecsSensiTemp[nadj],1.,ts->vecs_drdu[nadj]);CHKERRQ(ierr); 292 } 293 } 294 295 /* Build LHS for first-order adjoint */ 296 shift = 1./(th->Theta*th->time_step); 297 if (th->endpoint) { 298 ierr = TSComputeIJacobian(ts,th->stage_time,ts->vec_sol,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); 299 } else { 300 ierr = TSComputeIJacobian(ts,th->stage_time,th->X,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); 301 } 302 ierr = KSPSetOperators(ksp,J,Jp);CHKERRQ(ierr); 303 304 /* Solve stage equation LHS*lambda_s = RHS for first-order adjoint */ 305 for (nadj=0; nadj<ts->numcost; nadj++) { 306 ierr = KSPSolveTranspose(ksp,VecsSensiTemp[nadj],VecsDeltaLam[nadj]);CHKERRQ(ierr); 307 } 308 309 if (ts->vecs_sensi2) { /* U_{n+1} */ 310 if (!th->endpoint) SETERRQ(PetscObjectComm((PetscObject)ts),PETSC_ERR_SUP,"Operation not implemented in TS_Theta"); 311 /* Get w1 at t_{n+1} from TLM matrix */ 312 ierr = MatDenseGetColumn(ts->mat_sensip,0,&xarr);CHKERRQ(ierr); 313 ierr = VecPlaceArray(ts->vec_sensip_col,xarr);CHKERRQ(ierr); 314 /* lambda_s^T F_UU w_1 */ 315 ierr = TSComputeIHessianProductFunction1(ts,th->stage_time,ts->vec_sol,VecsDeltaLam,ts->vec_sensip_col,ts->vecs_fuu);CHKERRQ(ierr); 316 ierr = VecResetArray(ts->vec_sensip_col);CHKERRQ(ierr); 317 ierr = MatDenseRestoreColumn(ts->mat_sensip,&xarr);CHKERRQ(ierr); 318 if (ts->vecs_fup) { 319 /* lambda_s^T F_UP w_2 */ 320 ierr = TSComputeIHessianProductFunction2(ts,th->stage_time,ts->vec_sol,VecsDeltaLam,ts->vec_dir,ts->vecs_fup);CHKERRQ(ierr); 321 } 322 for (nadj=0; nadj<ts->numcost; nadj++) { /* compute the residual */ 323 ierr = VecCopy(ts->vecs_sensi2[nadj],VecsSensi2Temp[nadj]);CHKERRQ(ierr); 324 ierr = VecScale(VecsSensi2Temp[nadj],1./shift);CHKERRQ(ierr); 325 ierr = VecAXPY(VecsSensi2Temp[nadj],1.,ts->vecs_fuu[nadj]);CHKERRQ(ierr); 326 ierr = VecAXPY(VecsSensi2Temp[nadj],1.,ts->vecs_fuu[nadj]);CHKERRQ(ierr); 327 if (ts->vecs_fup) { 328 ierr = VecAXPY(VecsSensi2Temp[nadj],1.,ts->vecs_fup[nadj]);CHKERRQ(ierr); 329 } 330 if (ts->vec_costintegral) { 331 ierr = VecAXPY(VecsSensi2Temp[nadj],1.,ts->vecs_drdu[nadj]);CHKERRQ(ierr); 332 } 333 } 334 /* Solve stage equation LHS X = RHS for second-order adjoint */ 335 for (nadj=0; nadj<ts->numcost; nadj++) { 336 ierr = KSPSolveTranspose(ksp,VecsSensiTemp[nadj],VecsDeltaLam2[nadj]);CHKERRQ(ierr); 337 } 338 } 339 340 /* Update sensitivities, and evaluate integrals if there is any */ 341 if(th->endpoint) { /* two-stage case */ 342 if (th->Theta != 1.) { /* general case */ 343 shift = 1./((th->Theta-1.)*th->time_step); 344 ierr = TSComputeIJacobian(ts,th->ptime,th->X0,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); 345 if (ts->vec_costintegral) { /* R_U at t_n */ 346 ierr = TSComputeDRDUFunction(ts,th->ptime,th->X0,ts->vecs_drdu);CHKERRQ(ierr); 347 } 348 for (nadj=0; nadj<ts->numcost; nadj++) { 349 ierr = MatMultTranspose(J,VecsDeltaLam[nadj],ts->vecs_sensi[nadj]);CHKERRQ(ierr); 350 ierr = VecScale(ts->vecs_sensi[nadj],1./shift);CHKERRQ(ierr); 351 if (ts->vec_costintegral) { 352 ierr = VecAXPY(ts->vecs_sensi[nadj],-1./shift,ts->vecs_drdu[nadj]);CHKERRQ(ierr); 353 } 354 } 355 if (ts->vecs_sensi2) { /* second-order */ 356 /* Get w1 at t_n from TLM matrix */ 357 ierr = MatDenseGetColumn(th->MatFwdSensip0,0,&xarr);CHKERRQ(ierr); 358 ierr = VecPlaceArray(ts->vec_sensip_col,xarr);CHKERRQ(ierr); 359 /* lambda_s^T F_UU w_1 */ 360 ierr = TSComputeIHessianProductFunction1(ts,th->ptime,th->X0,VecsDeltaLam,ts->vec_sensip_col,ts->vecs_fuu);CHKERRQ(ierr); 361 ierr = VecResetArray(ts->vec_sensip_col);CHKERRQ(ierr); 362 ierr = MatDenseRestoreColumn(ts->mat_sensip,&xarr);CHKERRQ(ierr); 363 if (ts->vecs_fup) { 364 /* lambda_s^T F_UU w_2 */ 365 ierr = TSComputeIHessianProductFunction2(ts,th->ptime,th->X0,VecsDeltaLam,ts->vec_dir,ts->vecs_fup);CHKERRQ(ierr); 366 } 367 for (nadj=0; nadj<ts->numcost; nadj++) { 368 /* M^T Lambda_s + h(1-theta) F_U^T Lambda_s + h(1-theta) R_U */ 369 ierr = MatMultTranspose(J,VecsDeltaLam2[nadj],ts->vecs_sensi2[nadj]);CHKERRQ(ierr); 370 ierr = VecScale(ts->vecs_sensi2[nadj],1./shift);CHKERRQ(ierr); 371 ierr = VecAXPY(ts->vecs_sensi2[nadj],-1./shift,ts->vecs_fuu[nadj]);CHKERRQ(ierr); 372 ierr = VecAXPY(ts->vecs_sensi2[nadj],-1./shift,ts->vecs_fuu[nadj]);CHKERRQ(ierr); 373 if (ts->vecs_fup) { 374 ierr = VecAXPY(ts->vecs_sensi2[nadj],-1./shift,ts->vecs_fup[nadj]);CHKERRQ(ierr); 375 } 376 if (ts->vec_costintegral) { 377 ierr = VecAXPY(ts->vecs_sensi2[nadj],-1./shift,ts->vecs_drdu[nadj]);CHKERRQ(ierr); 378 } 379 } 380 } 381 } else { /* backward Euler */ 382 shift = 0.0; 383 ierr = TSComputeIJacobian(ts,th->stage_time,ts->vec_sol,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); /* get -f_u */ 384 for (nadj=0; nadj<ts->numcost; nadj++) { 385 ierr = MatMultTranspose(J,VecsDeltaLam[nadj],VecsSensiTemp[nadj]);CHKERRQ(ierr); 386 ierr = VecAXPY(ts->vecs_sensi[nadj],-th->time_step,VecsSensiTemp[nadj]);CHKERRQ(ierr); 387 if (ts->vec_costintegral) { /* wrong? */ 388 ierr = VecAXPY(ts->vecs_sensi[nadj],th->time_step,ts->vecs_drdu[nadj]);CHKERRQ(ierr); 389 } 390 } 391 if (ts->vecs_sensi2) { 392 for (nadj=0; nadj<ts->numcost; nadj++) { 393 ierr = MatMultTranspose(J,VecsDeltaLam[nadj],VecsSensi2Temp[nadj]);CHKERRQ(ierr); 394 ierr = VecAXPY(ts->vecs_sensi2[nadj],-th->time_step,VecsSensi2Temp[nadj]);CHKERRQ(ierr); 395 } 396 } 397 } 398 399 if (ts->vecs_sensip) { /* sensitivities wrt parameters */ 400 /* U_{n+1} */ 401 ierr = TSComputeRHSJacobianP(ts,th->stage_time,ts->vec_sol,ts->Jacp);CHKERRQ(ierr); 402 if (ts->vec_costintegral) { 403 ierr = TSComputeDRDPFunction(ts,th->stage_time,ts->vec_sol,ts->vecs_drdp);CHKERRQ(ierr); 404 } 405 for (nadj=0; nadj<ts->numcost; nadj++) { 406 ierr = MatMultTranspose(ts->Jacp,VecsDeltaLam[nadj],VecsDeltaMu[nadj]);CHKERRQ(ierr); 407 ierr = VecAXPY(ts->vecs_sensip[nadj],th->time_step*th->Theta,VecsDeltaMu[nadj]);CHKERRQ(ierr); 408 } 409 if (ts->vecs_sensip2) { /* second-order */ 410 /* lambda_s^T F_PU w_1 */ 411 ierr = TSComputeIHessianProductFunction3(ts,th->stage_time,ts->vec_sol,VecsDeltaLam,ts->vec_sensip_col,ts->vecs_fpu);CHKERRQ(ierr); 412 /* lambda_s^T F_PP w_2 */ 413 ierr = TSComputeIHessianProductFunction4(ts,th->stage_time,ts->vec_sol,VecsDeltaLam,ts->vec_dir,ts->vecs_fpp);CHKERRQ(ierr); 414 for (nadj=0; nadj<ts->numcost; nadj++) { 415 ierr = MatMultTranspose(ts->Jacp,VecsDeltaLam2[nadj],VecsDeltaMu2[nadj]);CHKERRQ(ierr); 416 ierr = VecAXPY(ts->vecs_sensip2[nadj],th->time_step*th->Theta,VecsDeltaMu2[nadj]);CHKERRQ(ierr); 417 if (ts->vecs_fpu) { 418 ierr = VecAXPY(ts->vecs_sensi2[nadj],th->time_step*th->Theta,ts->vecs_fpu[nadj]);CHKERRQ(ierr); 419 } 420 if (ts->vecs_fpp) { 421 ierr = VecAXPY(ts->vecs_sensi2[nadj],th->time_step*th->Theta,ts->vecs_fpp[nadj]);CHKERRQ(ierr); 422 } 423 if (ts->vec_costintegral) { 424 ierr = VecAXPY(ts->vecs_sensip2[nadj],th->time_step*th->Theta,ts->vecs_drdp[nadj]);CHKERRQ(ierr); 425 } 426 } 427 } 428 429 /* U_s */ 430 if (th->Theta!=1.) { 431 ierr = TSComputeRHSJacobianP(ts,th->ptime,th->X0,ts->Jacp);CHKERRQ(ierr); 432 if (ts->vec_costintegral) { 433 ierr = TSComputeDRDPFunction(ts,th->ptime,th->X0,ts->vecs_drdp);CHKERRQ(ierr); 434 } 435 for (nadj=0; nadj<ts->numcost; nadj++) { 436 ierr = MatMultTranspose(ts->Jacp,VecsDeltaLam[nadj],VecsDeltaMu[nadj]);CHKERRQ(ierr); 437 ierr = VecAXPY(ts->vecs_sensip[nadj],th->time_step*(1.-th->Theta),VecsDeltaMu[nadj]);CHKERRQ(ierr); 438 if (ts->vecs_sensip2) { /* second-order */ 439 /* lambda_s^T F_PU w_1 */ 440 ierr = TSComputeIHessianProductFunction3(ts,th->stage_time,ts->vec_sol,VecsDeltaLam,ts->vec_sensip_col,ts->vecs_fpu);CHKERRQ(ierr); 441 /* lambda_s^T F_PP w_2 */ 442 ierr = TSComputeIHessianProductFunction4(ts,th->stage_time,ts->vec_sol,VecsDeltaLam,ts->vec_dir,ts->vecs_fpp);CHKERRQ(ierr); 443 for (nadj=0; nadj<ts->numcost; nadj++) { 444 ierr = MatMultTranspose(ts->Jacp,VecsDeltaLam2[nadj],VecsDeltaMu2[nadj]);CHKERRQ(ierr); 445 ierr = VecAXPY(ts->vecs_sensip2[nadj],th->time_step*(1.-th->Theta),VecsDeltaMu2[nadj]);CHKERRQ(ierr); 446 if (ts->vecs_fpu) { 447 ierr = VecAXPY(ts->vecs_sensi2[nadj],th->time_step*(1.-th->Theta),ts->vecs_fpu[nadj]);CHKERRQ(ierr); 448 } 449 if (ts->vecs_fpp) { 450 ierr = VecAXPY(ts->vecs_sensi2[nadj],th->time_step*(1.-th->Theta),ts->vecs_fpp[nadj]);CHKERRQ(ierr); 451 } 452 if (ts->vec_costintegral) { 453 ierr = VecAXPY(ts->vecs_sensip2[nadj],th->time_step*(1.-th->Theta),ts->vecs_drdp[nadj]);CHKERRQ(ierr); 454 } 455 } 456 } 457 } 458 } 459 } 460 } else { /* one-stage case */ 461 shift = 0.0; 462 ierr = TSComputeIJacobian(ts,th->stage_time,th->X,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); /* get -f_y */ 463 if (ts->vec_costintegral) { 464 ierr = TSComputeDRDUFunction(ts,th->stage_time,th->X,ts->vecs_drdu);CHKERRQ(ierr); 465 } 466 for (nadj=0; nadj<ts->numcost; nadj++) { 467 ierr = MatMultTranspose(J,VecsDeltaLam[nadj],VecsSensiTemp[nadj]);CHKERRQ(ierr); 468 ierr = VecAXPY(ts->vecs_sensi[nadj],-th->time_step,VecsSensiTemp[nadj]);CHKERRQ(ierr); 469 if (ts->vec_costintegral) { 470 ierr = VecAXPY(ts->vecs_sensi[nadj],th->time_step,ts->vecs_drdu[nadj]);CHKERRQ(ierr); 471 } 472 } 473 if (ts->vecs_sensip) { 474 ierr = TSComputeRHSJacobianP(ts,th->stage_time,th->X,ts->Jacp);CHKERRQ(ierr); 475 for (nadj=0; nadj<ts->numcost; nadj++) { 476 ierr = MatMultTranspose(ts->Jacp,VecsDeltaLam[nadj],VecsDeltaMu[nadj]);CHKERRQ(ierr); 477 ierr = VecAXPY(ts->vecs_sensip[nadj],th->time_step,VecsDeltaMu[nadj]);CHKERRQ(ierr); 478 } 479 if (ts->vec_costintegral) { 480 ierr = TSComputeDRDPFunction(ts,th->stage_time,th->X,ts->vecs_drdp);CHKERRQ(ierr); 481 for (nadj=0; nadj<ts->numcost; nadj++) { 482 ierr = VecAXPY(ts->vecs_sensip[nadj],th->time_step,ts->vecs_drdp[nadj]);CHKERRQ(ierr); 483 } 484 } 485 } 486 } 487 488 th->status = TS_STEP_COMPLETE; 489 PetscFunctionReturn(0); 490 } 491 492 static PetscErrorCode TSInterpolate_Theta(TS ts,PetscReal t,Vec X) 493 { 494 TS_Theta *th = (TS_Theta*)ts->data; 495 PetscReal dt = t - ts->ptime; 496 PetscErrorCode ierr; 497 498 PetscFunctionBegin; 499 ierr = VecCopy(ts->vec_sol,th->X);CHKERRQ(ierr); 500 if (th->endpoint) dt *= th->Theta; 501 ierr = VecWAXPY(X,dt,th->Xdot,th->X);CHKERRQ(ierr); 502 PetscFunctionReturn(0); 503 } 504 505 static PetscErrorCode TSEvaluateWLTE_Theta(TS ts,NormType wnormtype,PetscInt *order,PetscReal *wlte) 506 { 507 TS_Theta *th = (TS_Theta*)ts->data; 508 Vec X = ts->vec_sol; /* X = solution */ 509 Vec Y = th->vec_lte_work; /* Y = X + LTE */ 510 PetscReal wltea,wlter; 511 PetscErrorCode ierr; 512 513 PetscFunctionBegin; 514 if (!th->vec_sol_prev) {*wlte = -1; PetscFunctionReturn(0);} 515 /* Cannot compute LTE in first step or in restart after event */ 516 if (ts->steprestart) {*wlte = -1; PetscFunctionReturn(0);} 517 /* Compute LTE using backward differences with non-constant time step */ 518 { 519 PetscReal h = ts->time_step, h_prev = ts->ptime - ts->ptime_prev; 520 PetscReal a = 1 + h_prev/h; 521 PetscScalar scal[3]; Vec vecs[3]; 522 scal[0] = +1/a; scal[1] = -1/(a-1); scal[2] = +1/(a*(a-1)); 523 vecs[0] = X; vecs[1] = th->X0; vecs[2] = th->vec_sol_prev; 524 ierr = VecCopy(X,Y);CHKERRQ(ierr); 525 ierr = VecMAXPY(Y,3,scal,vecs);CHKERRQ(ierr); 526 ierr = TSErrorWeightedNorm(ts,X,Y,wnormtype,wlte,&wltea,&wlter);CHKERRQ(ierr); 527 } 528 if (order) *order = 2; 529 PetscFunctionReturn(0); 530 } 531 532 static PetscErrorCode TSRollBack_Theta(TS ts) 533 { 534 TS_Theta *th = (TS_Theta*)ts->data; 535 PetscInt ncost; 536 PetscErrorCode ierr; 537 538 PetscFunctionBegin; 539 ierr = VecCopy(th->X0,ts->vec_sol);CHKERRQ(ierr); 540 if (ts->vec_costintegral && ts->costintegralfwd) { 541 ierr = VecCopy(th->VecCostIntegral0,ts->vec_costintegral);CHKERRQ(ierr); 542 } 543 th->status = TS_STEP_INCOMPLETE; 544 if (ts->mat_sensip) { 545 ierr = MatCopy(th->MatFwdSensip0,ts->mat_sensip,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 546 } 547 if (ts->vecs_integral_sensip) { 548 for (ncost=0;ncost<ts->numcost;ncost++) { 549 ierr = VecCopy(th->VecsIntegralSensip0[ncost],ts->vecs_integral_sensip[ncost]);CHKERRQ(ierr); 550 } 551 } 552 PetscFunctionReturn(0); 553 } 554 555 static PetscErrorCode TSForwardStep_Theta(TS ts) 556 { 557 TS_Theta *th = (TS_Theta*)ts->data; 558 Mat MatDeltaFwdSensip = th->MatDeltaFwdSensip; 559 Vec VecDeltaFwdSensipCol = th->VecDeltaFwdSensipCol; 560 PetscInt ncost,ntlm; 561 KSP ksp; 562 Mat J,Jp; 563 PetscReal shift; 564 PetscScalar *barr,*xarr; 565 PetscErrorCode ierr; 566 567 PetscFunctionBegin; 568 ierr = MatCopy(ts->mat_sensip,th->MatFwdSensip0,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 569 570 for (ncost=0; ncost<ts->numcost; ncost++) { 571 if (ts->vecs_integral_sensip) { 572 ierr = VecCopy(ts->vecs_integral_sensip[ncost],th->VecsIntegralSensip0[ncost]);CHKERRQ(ierr); 573 } 574 } 575 576 ierr = SNESGetKSP(ts->snes,&ksp);CHKERRQ(ierr); 577 ierr = TSGetIJacobian(ts,&J,&Jp,NULL,NULL);CHKERRQ(ierr); 578 579 /* Build RHS */ 580 if (th->endpoint) { /* 2-stage method*/ 581 shift = 1./((th->Theta-1.)*th->time_step); 582 ierr = TSComputeIJacobian(ts,th->ptime,th->X0,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); 583 ierr = MatMatMult(J,ts->mat_sensip,MAT_REUSE_MATRIX,PETSC_DEFAULT,&MatDeltaFwdSensip);CHKERRQ(ierr); 584 ierr = MatScale(MatDeltaFwdSensip,(th->Theta-1.)/th->Theta);CHKERRQ(ierr); 585 586 /* Add the f_p forcing terms */ 587 ierr = TSComputeRHSJacobianP(ts,th->ptime,th->X0,ts->Jacp);CHKERRQ(ierr); 588 ierr = MatAXPY(MatDeltaFwdSensip,(1.-th->Theta)/th->Theta,ts->Jacp,SUBSET_NONZERO_PATTERN);CHKERRQ(ierr); 589 ierr = TSComputeRHSJacobianP(ts,th->stage_time,ts->vec_sol,ts->Jacp);CHKERRQ(ierr); 590 ierr = MatAXPY(MatDeltaFwdSensip,1.,ts->Jacp,SUBSET_NONZERO_PATTERN);CHKERRQ(ierr); 591 } else { /* 1-stage method */ 592 shift = 0.0; 593 ierr = TSComputeIJacobian(ts,th->stage_time,th->X,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); 594 ierr = MatMatMult(J,ts->mat_sensip,MAT_REUSE_MATRIX,PETSC_DEFAULT,&MatDeltaFwdSensip);CHKERRQ(ierr); 595 ierr = MatScale(MatDeltaFwdSensip,-1.);CHKERRQ(ierr); 596 597 /* Add the f_p forcing terms */ 598 ierr = TSComputeRHSJacobianP(ts,th->stage_time,th->X,ts->Jacp);CHKERRQ(ierr); 599 ierr = MatAXPY(MatDeltaFwdSensip,1.,ts->Jacp,SUBSET_NONZERO_PATTERN);CHKERRQ(ierr); 600 } 601 602 /* Build LHS */ 603 shift = 1/(th->Theta*th->time_step); 604 if (th->endpoint) { 605 ierr = TSComputeIJacobian(ts,th->stage_time,ts->vec_sol,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); 606 } else { 607 ierr = TSComputeIJacobian(ts,th->stage_time,th->X,th->Xdot,shift,J,Jp,PETSC_FALSE);CHKERRQ(ierr); 608 } 609 ierr = KSPSetOperators(ksp,J,Jp);CHKERRQ(ierr); 610 611 /* 612 Evaluate the first stage of integral gradients with the 2-stage method: 613 drdu|t_n*S(t_n) + drdp|t_n 614 This is done before the linear solve because the sensitivity variable S(t_n) will be propagated to S(t_{n+1}) 615 */ 616 if (th->endpoint) { /* 2-stage method only */ 617 if (ts->vecs_integral_sensip) { 618 ierr = TSComputeDRDUFunction(ts,th->ptime,th->X0,ts->vecs_drdu);CHKERRQ(ierr); 619 ierr = TSComputeDRDPFunction(ts,th->ptime,th->X0,ts->vecs_drdp);CHKERRQ(ierr); 620 for (ncost=0; ncost<ts->numcost; ncost++) { 621 ierr = MatMultTranspose(ts->mat_sensip,ts->vecs_drdu[ncost],th->VecIntegralSensipTemp);CHKERRQ(ierr); 622 ierr = VecAXPY(th->VecIntegralSensipTemp,1,ts->vecs_drdp[ncost]);CHKERRQ(ierr); 623 ierr = VecAXPY(ts->vecs_integral_sensip[ncost],th->time_step*(1.-th->Theta),th->VecIntegralSensipTemp);CHKERRQ(ierr); 624 } 625 } 626 } 627 628 /* Solve the tangent linear equation for forward sensitivities to parameters */ 629 for (ntlm=0; ntlm<th->num_tlm; ntlm++) { 630 KSPConvergedReason kspreason; 631 ierr = MatDenseGetColumn(MatDeltaFwdSensip,ntlm,&barr);CHKERRQ(ierr); 632 ierr = VecPlaceArray(VecDeltaFwdSensipCol,barr);CHKERRQ(ierr); 633 if (th->endpoint) { 634 ierr = MatDenseGetColumn(ts->mat_sensip,ntlm,&xarr);CHKERRQ(ierr); 635 ierr = VecPlaceArray(ts->vec_sensip_col,xarr);CHKERRQ(ierr); 636 ierr = KSPSolve(ksp,VecDeltaFwdSensipCol,ts->vec_sensip_col);CHKERRQ(ierr); 637 ierr = VecResetArray(ts->vec_sensip_col);CHKERRQ(ierr); 638 ierr = MatDenseRestoreColumn(ts->mat_sensip,&xarr);CHKERRQ(ierr); 639 } else { 640 ierr = KSPSolve(ksp,VecDeltaFwdSensipCol,VecDeltaFwdSensipCol);CHKERRQ(ierr); 641 } 642 ierr = KSPGetConvergedReason(ksp,&kspreason);CHKERRQ(ierr); 643 if (kspreason < 0) { 644 ts->reason = TSFORWARD_DIVERGED_LINEAR_SOLVE; 645 ierr = PetscInfo2(ts,"Step=%D, %Dth tangent linear solve, linear solve fails, stopping tangent linear solve\n",ts->steps,ntlm);CHKERRQ(ierr); 646 } 647 ierr = VecResetArray(VecDeltaFwdSensipCol);CHKERRQ(ierr); 648 ierr = MatDenseRestoreColumn(MatDeltaFwdSensip,&barr);CHKERRQ(ierr); 649 } 650 651 652 /* 653 Evaluate the second stage of integral gradients with the 2-stage method: 654 drdu|t_{n+1}*S(t_{n+1}) + drdp|t_{n+1} 655 */ 656 if (ts->vecs_integral_sensip) { 657 if (!th->endpoint) { 658 ierr = MatAXPY(ts->mat_sensip,1,MatDeltaFwdSensip,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 659 ierr = TSComputeDRDUFunction(ts,th->stage_time,th->X,ts->vecs_drdu);CHKERRQ(ierr); 660 ierr = TSComputeDRDPFunction(ts,th->stage_time,th->X,ts->vecs_drdp);CHKERRQ(ierr); 661 for (ncost=0; ncost<ts->numcost; ncost++) { 662 ierr = MatMultTranspose(ts->mat_sensip,ts->vecs_drdu[ncost],th->VecIntegralSensipTemp);CHKERRQ(ierr); 663 ierr = VecAXPY(th->VecIntegralSensipTemp,1,ts->vecs_drdp[ncost]);CHKERRQ(ierr); 664 ierr = VecAXPY(ts->vecs_integral_sensip[ncost],th->time_step,th->VecIntegralSensipTemp);CHKERRQ(ierr); 665 } 666 ierr = MatAXPY(ts->mat_sensip,(1.-th->Theta)/th->Theta,MatDeltaFwdSensip,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 667 } else { 668 ierr = TSComputeDRDUFunction(ts,th->stage_time,ts->vec_sol,ts->vecs_drdu);CHKERRQ(ierr); 669 ierr = TSComputeDRDPFunction(ts,th->stage_time,ts->vec_sol,ts->vecs_drdp);CHKERRQ(ierr); 670 for (ncost=0; ncost<ts->numcost; ncost++) { 671 ierr = MatMultTranspose(ts->mat_sensip,ts->vecs_drdu[ncost],th->VecIntegralSensipTemp);CHKERRQ(ierr); 672 ierr = VecAXPY(th->VecIntegralSensipTemp,1,ts->vecs_drdp[ncost]);CHKERRQ(ierr); 673 ierr = VecAXPY(ts->vecs_integral_sensip[ncost],th->time_step*th->Theta,th->VecIntegralSensipTemp);CHKERRQ(ierr); 674 } 675 } 676 } else { 677 if (!th->endpoint) { 678 ierr = MatAXPY(ts->mat_sensip,1./th->Theta,MatDeltaFwdSensip,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 679 } 680 } 681 PetscFunctionReturn(0); 682 } 683 684 /*------------------------------------------------------------*/ 685 static PetscErrorCode TSReset_Theta(TS ts) 686 { 687 TS_Theta *th = (TS_Theta*)ts->data; 688 PetscErrorCode ierr; 689 690 PetscFunctionBegin; 691 ierr = VecDestroy(&th->X);CHKERRQ(ierr); 692 ierr = VecDestroy(&th->Xdot);CHKERRQ(ierr); 693 ierr = VecDestroy(&th->X0);CHKERRQ(ierr); 694 ierr = VecDestroy(&th->affine);CHKERRQ(ierr); 695 696 ierr = VecDestroy(&th->vec_sol_prev);CHKERRQ(ierr); 697 ierr = VecDestroy(&th->vec_lte_work);CHKERRQ(ierr); 698 699 ierr = VecDestroy(&th->VecCostIntegral0);CHKERRQ(ierr); 700 if (ts->forward_solve) { 701 if (ts->vecs_integral_sensip) { 702 ierr = VecDestroy(&th->VecIntegralSensipTemp);CHKERRQ(ierr); 703 ierr = VecDestroyVecs(ts->numcost,&th->VecsIntegralSensip0);CHKERRQ(ierr); 704 } 705 ierr = VecDestroy(&th->VecDeltaFwdSensipCol);CHKERRQ(ierr); 706 ierr = MatDestroy(&th->MatDeltaFwdSensip);CHKERRQ(ierr); 707 ierr = MatDestroy(&th->MatFwdSensip0);CHKERRQ(ierr); 708 } 709 ierr = VecDestroyVecs(ts->numcost,&th->VecsDeltaLam);CHKERRQ(ierr); 710 ierr = VecDestroyVecs(ts->numcost,&th->VecsDeltaMu);CHKERRQ(ierr); 711 ierr = VecDestroyVecs(ts->numcost,&th->VecsDeltaLam2);CHKERRQ(ierr); 712 ierr = VecDestroyVecs(ts->numcost,&th->VecsDeltaMu2);CHKERRQ(ierr); 713 ierr = VecDestroyVecs(ts->numcost,&th->VecsSensiTemp);CHKERRQ(ierr); 714 ierr = VecDestroyVecs(ts->numcost,&th->VecsSensi2Temp);CHKERRQ(ierr); 715 716 PetscFunctionReturn(0); 717 } 718 719 static PetscErrorCode TSDestroy_Theta(TS ts) 720 { 721 PetscErrorCode ierr; 722 723 PetscFunctionBegin; 724 ierr = TSReset_Theta(ts);CHKERRQ(ierr); 725 if (ts->dm) { 726 ierr = DMCoarsenHookRemove(ts->dm,DMCoarsenHook_TSTheta,DMRestrictHook_TSTheta,ts);CHKERRQ(ierr); 727 ierr = DMSubDomainHookRemove(ts->dm,DMSubDomainHook_TSTheta,DMSubDomainRestrictHook_TSTheta,ts);CHKERRQ(ierr); 728 } 729 ierr = PetscFree(ts->data);CHKERRQ(ierr); 730 ierr = PetscObjectComposeFunction((PetscObject)ts,"TSThetaGetTheta_C",NULL);CHKERRQ(ierr); 731 ierr = PetscObjectComposeFunction((PetscObject)ts,"TSThetaSetTheta_C",NULL);CHKERRQ(ierr); 732 ierr = PetscObjectComposeFunction((PetscObject)ts,"TSThetaGetEndpoint_C",NULL);CHKERRQ(ierr); 733 ierr = PetscObjectComposeFunction((PetscObject)ts,"TSThetaSetEndpoint_C",NULL);CHKERRQ(ierr); 734 PetscFunctionReturn(0); 735 } 736 737 /* 738 This defines the nonlinear equation that is to be solved with SNES 739 G(U) = F[t0+Theta*dt, U, (U-U0)*shift] = 0 740 */ 741 static PetscErrorCode SNESTSFormFunction_Theta(SNES snes,Vec x,Vec y,TS ts) 742 { 743 TS_Theta *th = (TS_Theta*)ts->data; 744 PetscErrorCode ierr; 745 Vec X0,Xdot; 746 DM dm,dmsave; 747 PetscReal shift = 1/(th->Theta*ts->time_step); 748 749 PetscFunctionBegin; 750 ierr = SNESGetDM(snes,&dm);CHKERRQ(ierr); 751 /* When using the endpoint variant, this is actually 1/Theta * Xdot */ 752 ierr = TSThetaGetX0AndXdot(ts,dm,&X0,&Xdot);CHKERRQ(ierr); 753 ierr = VecAXPBYPCZ(Xdot,-shift,shift,0,X0,x);CHKERRQ(ierr); 754 755 /* DM monkey-business allows user code to call TSGetDM() inside of functions evaluated on levels of FAS */ 756 dmsave = ts->dm; 757 ts->dm = dm; 758 ierr = TSComputeIFunction(ts,th->stage_time,x,Xdot,y,PETSC_FALSE);CHKERRQ(ierr); 759 ts->dm = dmsave; 760 ierr = TSThetaRestoreX0AndXdot(ts,dm,&X0,&Xdot);CHKERRQ(ierr); 761 PetscFunctionReturn(0); 762 } 763 764 static PetscErrorCode SNESTSFormJacobian_Theta(SNES snes,Vec x,Mat A,Mat B,TS ts) 765 { 766 TS_Theta *th = (TS_Theta*)ts->data; 767 PetscErrorCode ierr; 768 Vec Xdot; 769 DM dm,dmsave; 770 PetscReal shift = 1/(th->Theta*ts->time_step); 771 772 PetscFunctionBegin; 773 ierr = SNESGetDM(snes,&dm);CHKERRQ(ierr); 774 /* Xdot has already been computed in SNESTSFormFunction_Theta (SNES guarantees this) */ 775 ierr = TSThetaGetX0AndXdot(ts,dm,NULL,&Xdot);CHKERRQ(ierr); 776 777 dmsave = ts->dm; 778 ts->dm = dm; 779 ierr = TSComputeIJacobian(ts,th->stage_time,x,Xdot,shift,A,B,PETSC_FALSE);CHKERRQ(ierr); 780 ts->dm = dmsave; 781 ierr = TSThetaRestoreX0AndXdot(ts,dm,NULL,&Xdot);CHKERRQ(ierr); 782 PetscFunctionReturn(0); 783 } 784 785 static PetscErrorCode TSForwardSetUp_Theta(TS ts) 786 { 787 TS_Theta *th = (TS_Theta*)ts->data; 788 PetscErrorCode ierr; 789 790 PetscFunctionBegin; 791 /* combine sensitivities to parameters and sensitivities to initial values into one array */ 792 th->num_tlm = ts->num_parameters; 793 ierr = MatDuplicate(ts->mat_sensip,MAT_DO_NOT_COPY_VALUES,&th->MatDeltaFwdSensip);CHKERRQ(ierr); 794 if (ts->vecs_integral_sensip) { 795 ierr = VecDuplicate(ts->vecs_integral_sensip[0],&th->VecIntegralSensipTemp);CHKERRQ(ierr); 796 } 797 /* backup sensitivity results for roll-backs */ 798 ierr = MatDuplicate(ts->mat_sensip,MAT_DO_NOT_COPY_VALUES,&th->MatFwdSensip0);CHKERRQ(ierr); 799 800 if (ts->vecs_integral_sensip) { 801 ierr = VecDuplicateVecs(ts->vecs_integral_sensip[0],ts->numcost,&th->VecsIntegralSensip0);CHKERRQ(ierr); 802 } 803 ierr = VecDuplicate(ts->vec_sol,&th->VecDeltaFwdSensipCol);CHKERRQ(ierr); 804 ierr = VecDuplicate(ts->vec_sol,&ts->vec_sensip_col);CHKERRQ(ierr); 805 PetscFunctionReturn(0); 806 } 807 808 static PetscErrorCode TSSetUp_Theta(TS ts) 809 { 810 TS_Theta *th = (TS_Theta*)ts->data; 811 PetscBool match; 812 PetscErrorCode ierr; 813 814 PetscFunctionBegin; 815 if (!th->VecCostIntegral0 && ts->vec_costintegral && ts->costintegralfwd) { /* back up cost integral */ 816 ierr = VecDuplicate(ts->vec_costintegral,&th->VecCostIntegral0);CHKERRQ(ierr); 817 } 818 if (!th->X) { 819 ierr = VecDuplicate(ts->vec_sol,&th->X);CHKERRQ(ierr); 820 } 821 if (!th->Xdot) { 822 ierr = VecDuplicate(ts->vec_sol,&th->Xdot);CHKERRQ(ierr); 823 } 824 if (!th->X0) { 825 ierr = VecDuplicate(ts->vec_sol,&th->X0);CHKERRQ(ierr); 826 } 827 if (th->endpoint) { 828 ierr = VecDuplicate(ts->vec_sol,&th->affine);CHKERRQ(ierr); 829 } 830 831 th->order = (th->Theta == 0.5) ? 2 : 1; 832 833 ierr = TSGetDM(ts,&ts->dm);CHKERRQ(ierr); 834 ierr = DMCoarsenHookAdd(ts->dm,DMCoarsenHook_TSTheta,DMRestrictHook_TSTheta,ts);CHKERRQ(ierr); 835 ierr = DMSubDomainHookAdd(ts->dm,DMSubDomainHook_TSTheta,DMSubDomainRestrictHook_TSTheta,ts);CHKERRQ(ierr); 836 837 ierr = TSGetAdapt(ts,&ts->adapt);CHKERRQ(ierr); 838 ierr = TSAdaptCandidatesClear(ts->adapt);CHKERRQ(ierr); 839 ierr = PetscObjectTypeCompare((PetscObject)ts->adapt,TSADAPTNONE,&match);CHKERRQ(ierr); 840 if (!match) { 841 ierr = VecDuplicate(ts->vec_sol,&th->vec_sol_prev);CHKERRQ(ierr); 842 ierr = VecDuplicate(ts->vec_sol,&th->vec_lte_work);CHKERRQ(ierr); 843 } 844 ierr = TSGetSNES(ts,&ts->snes);CHKERRQ(ierr); 845 PetscFunctionReturn(0); 846 } 847 848 /*------------------------------------------------------------*/ 849 850 static PetscErrorCode TSAdjointSetUp_Theta(TS ts) 851 { 852 TS_Theta *th = (TS_Theta*)ts->data; 853 PetscErrorCode ierr; 854 855 PetscFunctionBegin; 856 ierr = VecDuplicateVecs(ts->vecs_sensi[0],ts->numcost,&th->VecsDeltaLam);CHKERRQ(ierr); 857 ierr = VecDuplicateVecs(ts->vecs_sensi[0],ts->numcost,&th->VecsSensiTemp);CHKERRQ(ierr); 858 if (ts->vecs_sensip) { 859 ierr = VecDuplicateVecs(ts->vecs_sensip[0],ts->numcost,&th->VecsDeltaMu);CHKERRQ(ierr); 860 } 861 if (ts->vecs_sensi2) { 862 ierr = VecDuplicateVecs(ts->vecs_sensi[0],ts->numcost,&th->VecsDeltaLam2);CHKERRQ(ierr); 863 ierr = VecDuplicateVecs(ts->vecs_sensi2[0],ts->numcost,&th->VecsSensi2Temp);CHKERRQ(ierr); 864 } 865 if (ts->vecs_sensip2) { 866 ierr = VecDuplicateVecs(ts->vecs_sensi2[0],ts->numcost,&th->VecsDeltaMu2);CHKERRQ(ierr); 867 } 868 PetscFunctionReturn(0); 869 } 870 871 static PetscErrorCode TSSetFromOptions_Theta(PetscOptionItems *PetscOptionsObject,TS ts) 872 { 873 TS_Theta *th = (TS_Theta*)ts->data; 874 PetscErrorCode ierr; 875 876 PetscFunctionBegin; 877 ierr = PetscOptionsHead(PetscOptionsObject,"Theta ODE solver options");CHKERRQ(ierr); 878 { 879 ierr = PetscOptionsReal("-ts_theta_theta","Location of stage (0<Theta<=1)","TSThetaSetTheta",th->Theta,&th->Theta,NULL);CHKERRQ(ierr); 880 ierr = PetscOptionsBool("-ts_theta_endpoint","Use the endpoint instead of midpoint form of the Theta method","TSThetaSetEndpoint",th->endpoint,&th->endpoint,NULL);CHKERRQ(ierr); 881 ierr = PetscOptionsBool("-ts_theta_initial_guess_extrapolate","Extrapolate stage initial guess from previous solution (sometimes unstable)","TSThetaSetExtrapolate",th->extrapolate,&th->extrapolate,NULL);CHKERRQ(ierr); 882 } 883 ierr = PetscOptionsTail();CHKERRQ(ierr); 884 PetscFunctionReturn(0); 885 } 886 887 static PetscErrorCode TSView_Theta(TS ts,PetscViewer viewer) 888 { 889 TS_Theta *th = (TS_Theta*)ts->data; 890 PetscBool iascii; 891 PetscErrorCode ierr; 892 893 PetscFunctionBegin; 894 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);CHKERRQ(ierr); 895 if (iascii) { 896 ierr = PetscViewerASCIIPrintf(viewer," Theta=%g\n",(double)th->Theta);CHKERRQ(ierr); 897 ierr = PetscViewerASCIIPrintf(viewer," Extrapolation=%s\n",th->extrapolate ? "yes" : "no");CHKERRQ(ierr); 898 } 899 PetscFunctionReturn(0); 900 } 901 902 static PetscErrorCode TSThetaGetTheta_Theta(TS ts,PetscReal *theta) 903 { 904 TS_Theta *th = (TS_Theta*)ts->data; 905 906 PetscFunctionBegin; 907 *theta = th->Theta; 908 PetscFunctionReturn(0); 909 } 910 911 static PetscErrorCode TSThetaSetTheta_Theta(TS ts,PetscReal theta) 912 { 913 TS_Theta *th = (TS_Theta*)ts->data; 914 915 PetscFunctionBegin; 916 if (theta <= 0 || 1 < theta) SETERRQ1(PetscObjectComm((PetscObject)ts),PETSC_ERR_ARG_OUTOFRANGE,"Theta %g not in range (0,1]",(double)theta); 917 th->Theta = theta; 918 th->order = (th->Theta == 0.5) ? 2 : 1; 919 PetscFunctionReturn(0); 920 } 921 922 static PetscErrorCode TSThetaGetEndpoint_Theta(TS ts,PetscBool *endpoint) 923 { 924 TS_Theta *th = (TS_Theta*)ts->data; 925 926 PetscFunctionBegin; 927 *endpoint = th->endpoint; 928 PetscFunctionReturn(0); 929 } 930 931 static PetscErrorCode TSThetaSetEndpoint_Theta(TS ts,PetscBool flg) 932 { 933 TS_Theta *th = (TS_Theta*)ts->data; 934 935 PetscFunctionBegin; 936 th->endpoint = flg; 937 PetscFunctionReturn(0); 938 } 939 940 #if defined(PETSC_HAVE_COMPLEX) 941 static PetscErrorCode TSComputeLinearStability_Theta(TS ts,PetscReal xr,PetscReal xi,PetscReal *yr,PetscReal *yi) 942 { 943 PetscComplex z = xr + xi*PETSC_i,f; 944 TS_Theta *th = (TS_Theta*)ts->data; 945 const PetscReal one = 1.0; 946 947 PetscFunctionBegin; 948 f = (one + (one - th->Theta)*z)/(one - th->Theta*z); 949 *yr = PetscRealPartComplex(f); 950 *yi = PetscImaginaryPartComplex(f); 951 PetscFunctionReturn(0); 952 } 953 #endif 954 955 static PetscErrorCode TSGetStages_Theta(TS ts,PetscInt *ns,Vec **Y) 956 { 957 TS_Theta *th = (TS_Theta*)ts->data; 958 959 PetscFunctionBegin; 960 if (ns) *ns = 1; 961 if (Y) *Y = th->endpoint ? &(th->X0) : &(th->X); 962 PetscFunctionReturn(0); 963 } 964 965 /* ------------------------------------------------------------ */ 966 /*MC 967 TSTHETA - DAE solver using the implicit Theta method 968 969 Level: beginner 970 971 Options Database: 972 + -ts_theta_theta <Theta> - Location of stage (0<Theta<=1) 973 . -ts_theta_endpoint <flag> - Use the endpoint (like Crank-Nicholson) instead of midpoint form of the Theta method 974 - -ts_theta_initial_guess_extrapolate <flg> - Extrapolate stage initial guess from previous solution (sometimes unstable) 975 976 Notes: 977 $ -ts_type theta -ts_theta_theta 1.0 corresponds to backward Euler (TSBEULER) 978 $ -ts_type theta -ts_theta_theta 0.5 corresponds to the implicit midpoint rule 979 $ -ts_type theta -ts_theta_theta 0.5 -ts_theta_endpoint corresponds to Crank-Nicholson (TSCN) 980 981 This method can be applied to DAE. 982 983 This method is cast as a 1-stage implicit Runge-Kutta method. 984 985 .vb 986 Theta | Theta 987 ------------- 988 | 1 989 .ve 990 991 For the default Theta=0.5, this is also known as the implicit midpoint rule. 992 993 When the endpoint variant is chosen, the method becomes a 2-stage method with first stage explicit: 994 995 .vb 996 0 | 0 0 997 1 | 1-Theta Theta 998 ------------------- 999 | 1-Theta Theta 1000 .ve 1001 1002 For the default Theta=0.5, this is the trapezoid rule (also known as Crank-Nicolson, see TSCN). 1003 1004 To apply a diagonally implicit RK method to DAE, the stage formula 1005 1006 $ Y_i = X + h sum_j a_ij Y'_j 1007 1008 is interpreted as a formula for Y'_i in terms of Y_i and known values (Y'_j, j<i) 1009 1010 .seealso: TSCreate(), TS, TSSetType(), TSCN, TSBEULER, TSThetaSetTheta(), TSThetaSetEndpoint() 1011 1012 M*/ 1013 PETSC_EXTERN PetscErrorCode TSCreate_Theta(TS ts) 1014 { 1015 TS_Theta *th; 1016 PetscErrorCode ierr; 1017 1018 PetscFunctionBegin; 1019 ts->ops->reset = TSReset_Theta; 1020 ts->ops->destroy = TSDestroy_Theta; 1021 ts->ops->view = TSView_Theta; 1022 ts->ops->setup = TSSetUp_Theta; 1023 ts->ops->adjointsetup = TSAdjointSetUp_Theta; 1024 ts->ops->step = TSStep_Theta; 1025 ts->ops->interpolate = TSInterpolate_Theta; 1026 ts->ops->evaluatewlte = TSEvaluateWLTE_Theta; 1027 ts->ops->rollback = TSRollBack_Theta; 1028 ts->ops->setfromoptions = TSSetFromOptions_Theta; 1029 ts->ops->snesfunction = SNESTSFormFunction_Theta; 1030 ts->ops->snesjacobian = SNESTSFormJacobian_Theta; 1031 #if defined(PETSC_HAVE_COMPLEX) 1032 ts->ops->linearstability = TSComputeLinearStability_Theta; 1033 #endif 1034 ts->ops->getstages = TSGetStages_Theta; 1035 ts->ops->adjointstep = TSAdjointStep_Theta; 1036 ts->ops->adjointintegral = TSAdjointCostIntegral_Theta; 1037 ts->ops->forwardintegral = TSForwardCostIntegral_Theta; 1038 ts->default_adapt_type = TSADAPTNONE; 1039 ts->ops->forwardsetup = TSForwardSetUp_Theta; 1040 ts->ops->forwardstep = TSForwardStep_Theta; 1041 1042 ts->usessnes = PETSC_TRUE; 1043 1044 ierr = PetscNewLog(ts,&th);CHKERRQ(ierr); 1045 ts->data = (void*)th; 1046 1047 th->VecsDeltaLam = NULL; 1048 th->VecsDeltaMu = NULL; 1049 th->VecsSensiTemp = NULL; 1050 th->VecsSensi2Temp = NULL; 1051 1052 th->extrapolate = PETSC_FALSE; 1053 th->Theta = 0.5; 1054 th->order = 2; 1055 ierr = PetscObjectComposeFunction((PetscObject)ts,"TSThetaGetTheta_C",TSThetaGetTheta_Theta);CHKERRQ(ierr); 1056 ierr = PetscObjectComposeFunction((PetscObject)ts,"TSThetaSetTheta_C",TSThetaSetTheta_Theta);CHKERRQ(ierr); 1057 ierr = PetscObjectComposeFunction((PetscObject)ts,"TSThetaGetEndpoint_C",TSThetaGetEndpoint_Theta);CHKERRQ(ierr); 1058 ierr = PetscObjectComposeFunction((PetscObject)ts,"TSThetaSetEndpoint_C",TSThetaSetEndpoint_Theta);CHKERRQ(ierr); 1059 PetscFunctionReturn(0); 1060 } 1061 1062 /*@ 1063 TSThetaGetTheta - Get the abscissa of the stage in (0,1]. 1064 1065 Not Collective 1066 1067 Input Parameter: 1068 . ts - timestepping context 1069 1070 Output Parameter: 1071 . theta - stage abscissa 1072 1073 Note: 1074 Use of this function is normally only required to hack TSTHETA to use a modified integration scheme. 1075 1076 Level: Advanced 1077 1078 .seealso: TSThetaSetTheta() 1079 @*/ 1080 PetscErrorCode TSThetaGetTheta(TS ts,PetscReal *theta) 1081 { 1082 PetscErrorCode ierr; 1083 1084 PetscFunctionBegin; 1085 PetscValidHeaderSpecific(ts,TS_CLASSID,1); 1086 PetscValidPointer(theta,2); 1087 ierr = PetscUseMethod(ts,"TSThetaGetTheta_C",(TS,PetscReal*),(ts,theta));CHKERRQ(ierr); 1088 PetscFunctionReturn(0); 1089 } 1090 1091 /*@ 1092 TSThetaSetTheta - Set the abscissa of the stage in (0,1]. 1093 1094 Not Collective 1095 1096 Input Parameter: 1097 + ts - timestepping context 1098 - theta - stage abscissa 1099 1100 Options Database: 1101 . -ts_theta_theta <theta> 1102 1103 Level: Intermediate 1104 1105 .seealso: TSThetaGetTheta() 1106 @*/ 1107 PetscErrorCode TSThetaSetTheta(TS ts,PetscReal theta) 1108 { 1109 PetscErrorCode ierr; 1110 1111 PetscFunctionBegin; 1112 PetscValidHeaderSpecific(ts,TS_CLASSID,1); 1113 ierr = PetscTryMethod(ts,"TSThetaSetTheta_C",(TS,PetscReal),(ts,theta));CHKERRQ(ierr); 1114 PetscFunctionReturn(0); 1115 } 1116 1117 /*@ 1118 TSThetaGetEndpoint - Gets whether to use the endpoint variant of the method (e.g. trapezoid/Crank-Nicolson instead of midpoint rule). 1119 1120 Not Collective 1121 1122 Input Parameter: 1123 . ts - timestepping context 1124 1125 Output Parameter: 1126 . endpoint - PETSC_TRUE when using the endpoint variant 1127 1128 Level: Advanced 1129 1130 .seealso: TSThetaSetEndpoint(), TSTHETA, TSCN 1131 @*/ 1132 PetscErrorCode TSThetaGetEndpoint(TS ts,PetscBool *endpoint) 1133 { 1134 PetscErrorCode ierr; 1135 1136 PetscFunctionBegin; 1137 PetscValidHeaderSpecific(ts,TS_CLASSID,1); 1138 PetscValidPointer(endpoint,2); 1139 ierr = PetscUseMethod(ts,"TSThetaGetEndpoint_C",(TS,PetscBool*),(ts,endpoint));CHKERRQ(ierr); 1140 PetscFunctionReturn(0); 1141 } 1142 1143 /*@ 1144 TSThetaSetEndpoint - Sets whether to use the endpoint variant of the method (e.g. trapezoid/Crank-Nicolson instead of midpoint rule). 1145 1146 Not Collective 1147 1148 Input Parameter: 1149 + ts - timestepping context 1150 - flg - PETSC_TRUE to use the endpoint variant 1151 1152 Options Database: 1153 . -ts_theta_endpoint <flg> 1154 1155 Level: Intermediate 1156 1157 .seealso: TSTHETA, TSCN 1158 @*/ 1159 PetscErrorCode TSThetaSetEndpoint(TS ts,PetscBool flg) 1160 { 1161 PetscErrorCode ierr; 1162 1163 PetscFunctionBegin; 1164 PetscValidHeaderSpecific(ts,TS_CLASSID,1); 1165 ierr = PetscTryMethod(ts,"TSThetaSetEndpoint_C",(TS,PetscBool),(ts,flg));CHKERRQ(ierr); 1166 PetscFunctionReturn(0); 1167 } 1168 1169 /* 1170 * TSBEULER and TSCN are straightforward specializations of TSTHETA. 1171 * The creation functions for these specializations are below. 1172 */ 1173 1174 static PetscErrorCode TSSetUp_BEuler(TS ts) 1175 { 1176 TS_Theta *th = (TS_Theta*)ts->data; 1177 PetscErrorCode ierr; 1178 1179 PetscFunctionBegin; 1180 if (th->Theta != 1.0) SETERRQ(PetscObjectComm((PetscObject)ts),PETSC_ERR_OPT_OVERWRITE,"Can not change the default value (1) of theta when using backward Euler\n"); 1181 if (th->endpoint) SETERRQ(PetscObjectComm((PetscObject)ts),PETSC_ERR_OPT_OVERWRITE,"Can not change to the endpoint form of the Theta methods when using backward Euler\n"); 1182 ierr = TSSetUp_Theta(ts);CHKERRQ(ierr); 1183 PetscFunctionReturn(0); 1184 } 1185 1186 static PetscErrorCode TSView_BEuler(TS ts,PetscViewer viewer) 1187 { 1188 PetscFunctionBegin; 1189 PetscFunctionReturn(0); 1190 } 1191 1192 /*MC 1193 TSBEULER - ODE solver using the implicit backward Euler method 1194 1195 Level: beginner 1196 1197 Notes: 1198 TSBEULER is equivalent to TSTHETA with Theta=1.0 1199 1200 $ -ts_type theta -ts_theta_theta 1.0 1201 1202 .seealso: TSCreate(), TS, TSSetType(), TSEULER, TSCN, TSTHETA 1203 1204 M*/ 1205 PETSC_EXTERN PetscErrorCode TSCreate_BEuler(TS ts) 1206 { 1207 PetscErrorCode ierr; 1208 1209 PetscFunctionBegin; 1210 ierr = TSCreate_Theta(ts);CHKERRQ(ierr); 1211 ierr = TSThetaSetTheta(ts,1.0);CHKERRQ(ierr); 1212 ierr = TSThetaSetEndpoint(ts,PETSC_FALSE);CHKERRQ(ierr); 1213 ts->ops->setup = TSSetUp_BEuler; 1214 ts->ops->view = TSView_BEuler; 1215 PetscFunctionReturn(0); 1216 } 1217 1218 static PetscErrorCode TSSetUp_CN(TS ts) 1219 { 1220 TS_Theta *th = (TS_Theta*)ts->data; 1221 PetscErrorCode ierr; 1222 1223 PetscFunctionBegin; 1224 if (th->Theta != 0.5) SETERRQ(PetscObjectComm((PetscObject)ts),PETSC_ERR_OPT_OVERWRITE,"Can not change the default value (0.5) of theta when using Crank-Nicolson\n"); 1225 if (!th->endpoint) SETERRQ(PetscObjectComm((PetscObject)ts),PETSC_ERR_OPT_OVERWRITE,"Can not change to the midpoint form of the Theta methods when using Crank-Nicolson\n"); 1226 ierr = TSSetUp_Theta(ts);CHKERRQ(ierr); 1227 PetscFunctionReturn(0); 1228 } 1229 1230 static PetscErrorCode TSView_CN(TS ts,PetscViewer viewer) 1231 { 1232 PetscFunctionBegin; 1233 PetscFunctionReturn(0); 1234 } 1235 1236 /*MC 1237 TSCN - ODE solver using the implicit Crank-Nicolson method. 1238 1239 Level: beginner 1240 1241 Notes: 1242 TSCN is equivalent to TSTHETA with Theta=0.5 and the "endpoint" option set. I.e. 1243 1244 $ -ts_type theta -ts_theta_theta 0.5 -ts_theta_endpoint 1245 1246 .seealso: TSCreate(), TS, TSSetType(), TSBEULER, TSTHETA 1247 1248 M*/ 1249 PETSC_EXTERN PetscErrorCode TSCreate_CN(TS ts) 1250 { 1251 PetscErrorCode ierr; 1252 1253 PetscFunctionBegin; 1254 ierr = TSCreate_Theta(ts);CHKERRQ(ierr); 1255 ierr = TSThetaSetTheta(ts,0.5);CHKERRQ(ierr); 1256 ierr = TSThetaSetEndpoint(ts,PETSC_TRUE);CHKERRQ(ierr); 1257 ts->ops->setup = TSSetUp_CN; 1258 ts->ops->view = TSView_CN; 1259 PetscFunctionReturn(0); 1260 } 1261