1 #include <petsctaolinesearch.h> 2 #include <../src/tao/bound/impls/bnk/bnk.h> 3 4 #include <petscksp.h> 5 6 /* Routine for BFGS preconditioner */ 7 8 PetscErrorCode MatLMVMSolveShell(PC pc, Vec b, Vec x) 9 { 10 PetscErrorCode ierr; 11 Mat M; 12 13 PetscFunctionBegin; 14 PetscValidHeaderSpecific(pc,PC_CLASSID,1); 15 PetscValidHeaderSpecific(b,VEC_CLASSID,2); 16 PetscValidHeaderSpecific(x,VEC_CLASSID,3); 17 ierr = PCShellGetContext(pc,(void**)&M);CHKERRQ(ierr); 18 ierr = MatLMVMSolve(M, b, x);CHKERRQ(ierr); 19 PetscFunctionReturn(0); 20 } 21 22 PetscErrorCode TaoBNKInitialize(Tao tao) 23 { 24 PetscErrorCode ierr; 25 TAO_BNK *bnk = (TAO_BNK *)tao->data; 26 KSPType ksp_type; 27 PC pc; 28 29 PetscReal fmin, ftrial, prered, actred, kappa, sigma; 30 PetscReal tau, tau_1, tau_2, tau_max, tau_min, max_radius; 31 PetscReal delta, step = 1.0; 32 33 PetscInt n,N,needH = 1; 34 35 PetscInt i_max = 5; 36 PetscInt j_max = 1; 37 PetscInt i, j; 38 39 PetscFunctionBegin; 40 /* Number of times ksp stopped because of these reasons */ 41 bnk->ksp_atol = 0; 42 bnk->ksp_rtol = 0; 43 bnk->ksp_dtol = 0; 44 bnk->ksp_ctol = 0; 45 bnk->ksp_negc = 0; 46 bnk->ksp_iter = 0; 47 bnk->ksp_othr = 0; 48 49 /* Initialize trust-region radius when using nash, stcg, or gltr 50 Command automatically ignored for other methods 51 Will be reset during the first iteration 52 */ 53 ierr = KSPGetType(tao->ksp,&ksp_type);CHKERRQ(ierr); 54 ierr = PetscStrcmp(ksp_type,KSPCGNASH,&bnk->is_nash);CHKERRQ(ierr); 55 ierr = PetscStrcmp(ksp_type,KSPCGSTCG,&bnk->is_stcg);CHKERRQ(ierr); 56 ierr = PetscStrcmp(ksp_type,KSPCGGLTR,&bnk->is_gltr);CHKERRQ(ierr); 57 58 ierr = KSPCGSetRadius(tao->ksp,bnk->max_radius);CHKERRQ(ierr); 59 60 if (bnk->is_nash || bnk->is_stcg || bnk->is_gltr) { 61 if (tao->trust0 < 0.0) SETERRQ(PETSC_COMM_SELF,1,"Initial radius negative"); 62 tao->trust = tao->trust0; 63 tao->trust = PetscMax(tao->trust, bnk->min_radius); 64 tao->trust = PetscMin(tao->trust, bnk->max_radius); 65 } 66 67 /* Get vectors we will need */ 68 if (BNK_PC_BFGS == bnk->pc_type && !bnk->M) { 69 ierr = VecGetLocalSize(tao->solution,&n);CHKERRQ(ierr); 70 ierr = VecGetSize(tao->solution,&N);CHKERRQ(ierr); 71 ierr = MatCreateLMVM(((PetscObject)tao)->comm,n,N,&bnk->M);CHKERRQ(ierr); 72 ierr = MatLMVMAllocateVectors(bnk->M,tao->solution);CHKERRQ(ierr); 73 } 74 75 /* create vectors for the limited memory preconditioner */ 76 if ((BNK_PC_BFGS == bnk->pc_type) && (BFGS_SCALE_BFGS != bnk->bfgs_scale_type)) { 77 if (!bnk->Diag) { 78 ierr = VecDuplicate(tao->solution,&bnk->Diag);CHKERRQ(ierr); 79 } 80 } 81 82 /* Modify the preconditioner to use the bfgs approximation */ 83 ierr = KSPGetPC(tao->ksp, &pc);CHKERRQ(ierr); 84 switch(bnk->pc_type) { 85 case BNK_PC_NONE: 86 ierr = PCSetType(pc, PCNONE);CHKERRQ(ierr); 87 ierr = PCSetFromOptions(pc);CHKERRQ(ierr); 88 break; 89 90 case BNK_PC_AHESS: 91 ierr = PCSetType(pc, PCJACOBI);CHKERRQ(ierr); 92 ierr = PCSetFromOptions(pc);CHKERRQ(ierr); 93 ierr = PCJacobiSetUseAbs(pc,PETSC_TRUE);CHKERRQ(ierr); 94 break; 95 96 case BNK_PC_BFGS: 97 ierr = PCSetType(pc, PCSHELL);CHKERRQ(ierr); 98 ierr = PCSetFromOptions(pc);CHKERRQ(ierr); 99 ierr = PCShellSetName(pc, "bfgs");CHKERRQ(ierr); 100 ierr = PCShellSetContext(pc, bnk->M);CHKERRQ(ierr); 101 ierr = PCShellSetApply(pc, MatLMVMSolveShell);CHKERRQ(ierr); 102 break; 103 104 default: 105 /* Use the pc method set by pc_type */ 106 break; 107 } 108 109 /* Initialize trust-region radius. The initialization is only performed 110 when we are using Nash, Steihaug-Toint or the Generalized Lanczos method. */ 111 if (bnk->is_nash || bnk->is_stcg || bnk->is_gltr) { 112 switch(bnk->init_type) { 113 case BNK_INIT_CONSTANT: 114 /* Use the initial radius specified */ 115 break; 116 117 case BNK_INIT_INTERPOLATION: 118 /* Use the initial radius specified */ 119 max_radius = 0.0; 120 121 for (j = 0; j < j_max; ++j) { 122 fmin = bnk->f; 123 sigma = 0.0; 124 125 if (needH) { 126 ierr = TaoComputeHessian(tao, tao->solution,tao->hessian,tao->hessian_pre);CHKERRQ(ierr); 127 needH = 0; 128 } 129 130 for (i = 0; i < i_max; ++i) { 131 ierr = VecCopy(tao->solution,bnk->W);CHKERRQ(ierr); 132 ierr = VecAXPY(bnk->W,-tao->trust/bnk->gnorm,tao->gradient);CHKERRQ(ierr); 133 ierr = TaoComputeObjective(tao, bnk->W, &ftrial);CHKERRQ(ierr); 134 if (PetscIsInfOrNanReal(ftrial)) { 135 tau = bnk->gamma1_i; 136 } else { 137 if (ftrial < fmin) { 138 fmin = ftrial; 139 sigma = -tao->trust / bnk->gnorm; 140 } 141 142 ierr = MatMult(tao->hessian, tao->gradient, tao->stepdirection);CHKERRQ(ierr); 143 ierr = VecDot(tao->gradient, tao->stepdirection, &prered);CHKERRQ(ierr); 144 145 prered = tao->trust * (bnk->gnorm - 0.5 * tao->trust * prered / (bnk->gnorm * bnk->gnorm)); 146 actred = bnk->f - ftrial; 147 if ((PetscAbsScalar(actred) <= bnk->epsilon) && (PetscAbsScalar(prered) <= bnk->epsilon)) { 148 kappa = 1.0; 149 } else { 150 kappa = actred / prered; 151 } 152 153 tau_1 = bnk->theta_i * bnk->gnorm * tao->trust / (bnk->theta_i * bnk->gnorm * tao->trust + (1.0 - bnk->theta_i) * prered - actred); 154 tau_2 = bnk->theta_i * bnk->gnorm * tao->trust / (bnk->theta_i * bnk->gnorm * tao->trust - (1.0 + bnk->theta_i) * prered + actred); 155 tau_min = PetscMin(tau_1, tau_2); 156 tau_max = PetscMax(tau_1, tau_2); 157 158 if (PetscAbsScalar(kappa - 1.0) <= bnk->mu1_i) { 159 /* Great agreement */ 160 max_radius = PetscMax(max_radius, tao->trust); 161 162 if (tau_max < 1.0) { 163 tau = bnk->gamma3_i; 164 } else if (tau_max > bnk->gamma4_i) { 165 tau = bnk->gamma4_i; 166 } else if (tau_1 >= 1.0 && tau_1 <= bnk->gamma4_i && tau_2 < 1.0) { 167 tau = tau_1; 168 } else if (tau_2 >= 1.0 && tau_2 <= bnk->gamma4_i && tau_1 < 1.0) { 169 tau = tau_2; 170 } else { 171 tau = tau_max; 172 } 173 } else if (PetscAbsScalar(kappa - 1.0) <= bnk->mu2_i) { 174 /* Good agreement */ 175 max_radius = PetscMax(max_radius, tao->trust); 176 177 if (tau_max < bnk->gamma2_i) { 178 tau = bnk->gamma2_i; 179 } else if (tau_max > bnk->gamma3_i) { 180 tau = bnk->gamma3_i; 181 } else { 182 tau = tau_max; 183 } 184 } else { 185 /* Not good agreement */ 186 if (tau_min > 1.0) { 187 tau = bnk->gamma2_i; 188 } else if (tau_max < bnk->gamma1_i) { 189 tau = bnk->gamma1_i; 190 } else if ((tau_min < bnk->gamma1_i) && (tau_max >= 1.0)) { 191 tau = bnk->gamma1_i; 192 } else if ((tau_1 >= bnk->gamma1_i) && (tau_1 < 1.0) && ((tau_2 < bnk->gamma1_i) || (tau_2 >= 1.0))) { 193 tau = tau_1; 194 } else if ((tau_2 >= bnk->gamma1_i) && (tau_2 < 1.0) && ((tau_1 < bnk->gamma1_i) || (tau_2 >= 1.0))) { 195 tau = tau_2; 196 } else { 197 tau = tau_max; 198 } 199 } 200 } 201 tao->trust = tau * tao->trust; 202 } 203 204 if (fmin < bnk->f) { 205 bnk->f = fmin; 206 ierr = VecAXPY(tao->solution,sigma,tao->gradient);CHKERRQ(ierr); 207 ierr = TaoComputeGradient(tao,tao->solution,tao->gradient);CHKERRQ(ierr); 208 209 ierr = TaoGradientNorm(tao, tao->gradient,NORM_2,&bnk->gnorm);CHKERRQ(ierr); 210 if (PetscIsInfOrNanReal(bnk->gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute gradient generated Inf or NaN"); 211 needH = 1; 212 213 ierr = TaoLogConvergenceHistory(tao,bnk->f,bnk->gnorm,0.0,tao->ksp_its);CHKERRQ(ierr); 214 ierr = TaoMonitor(tao,tao->niter,bnk->f,bnk->gnorm,0.0,step);CHKERRQ(ierr); 215 ierr = (*tao->ops->convergencetest)(tao,tao->cnvP);CHKERRQ(ierr); 216 if (tao->reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0); 217 } 218 } 219 tao->trust = PetscMax(tao->trust, max_radius); 220 221 /* Modify the radius if it is too large or small */ 222 tao->trust = PetscMax(tao->trust, bnk->min_radius); 223 tao->trust = PetscMin(tao->trust, bnk->max_radius); 224 break; 225 226 default: 227 /* Norm of the first direction will initialize radius */ 228 tao->trust = 0.0; 229 break; 230 } 231 } 232 233 /* Set initial scaling for the BFGS preconditioner 234 This step is done after computing the initial trust-region radius 235 since the function value may have decreased */ 236 if (BNK_PC_BFGS == bnk->pc_type) { 237 if (bnk->f != 0.0) { 238 delta = 2.0 * PetscAbsScalar(bnk->f) / (bnk->gnorm*bnk->gnorm); 239 } else { 240 delta = 2.0 / (bnk->gnorm*bnk->gnorm); 241 } 242 ierr = MatLMVMSetDelta(bnk->M,delta);CHKERRQ(ierr); 243 } 244 245 /* Set counter for gradient/reset steps*/ 246 bnk->newt = 0; 247 bnk->bfgs = 0; 248 bnk->sgrad = 0; 249 bnk->grad = 0; 250 PetscFunctionReturn(0); 251 } 252 253 PetscErrorCode TaoBNKComputeStep(Tao tao, PetscInt *stepType) 254 { 255 PetscErrorCode ierr; 256 TAO_BNK *bnk = (TAO_BNK *)tao->data; 257 KSPConvergedReason ksp_reason; 258 259 PetscReal gdx, delta, e_min; 260 261 PetscInt bfgsUpdates = 0; 262 PetscInt kspits; 263 PetscInt needH = 1; 264 265 PetscFunctionBegin; 266 /* Compute the Hessian */ 267 if (needH) { 268 ierr = TaoComputeHessian(tao,tao->solution,tao->hessian,tao->hessian_pre);CHKERRQ(ierr); 269 } 270 271 if ((BNK_PC_BFGS == bnk->pc_type) && (BFGS_SCALE_AHESS == bnk->bfgs_scale_type)) { 272 /* Obtain diagonal for the bfgs preconditioner */ 273 ierr = MatGetDiagonal(tao->hessian, bnk->Diag);CHKERRQ(ierr); 274 ierr = VecAbs(bnk->Diag);CHKERRQ(ierr); 275 ierr = VecReciprocal(bnk->Diag);CHKERRQ(ierr); 276 ierr = MatLMVMSetScale(bnk->M,bnk->Diag);CHKERRQ(ierr); 277 } 278 279 /* Shift the Hessian matrix */ 280 bnk->pert = bnk->sval; 281 if (bnk->pert > 0) { 282 ierr = MatShift(tao->hessian, bnk->pert);CHKERRQ(ierr); 283 if (tao->hessian != tao->hessian_pre) { 284 ierr = MatShift(tao->hessian_pre, bnk->pert);CHKERRQ(ierr); 285 } 286 } 287 288 if (BNK_PC_BFGS == bnk->pc_type) { 289 if (BFGS_SCALE_PHESS == bnk->bfgs_scale_type) { 290 /* Obtain diagonal for the bfgs preconditioner */ 291 ierr = MatGetDiagonal(tao->hessian, bnk->Diag);CHKERRQ(ierr); 292 ierr = VecAbs(bnk->Diag);CHKERRQ(ierr); 293 ierr = VecReciprocal(bnk->Diag);CHKERRQ(ierr); 294 ierr = MatLMVMSetScale(bnk->M,bnk->Diag);CHKERRQ(ierr); 295 } 296 /* Update the limited memory preconditioner and get existing # of updates */ 297 ierr = MatLMVMUpdate(bnk->M, tao->solution, tao->gradient);CHKERRQ(ierr); 298 ierr = MatLMVMGetUpdates(bnk->M, &bfgsUpdates);CHKERRQ(ierr); 299 } 300 301 /* Solve the Newton system of equations */ 302 ierr = KSPSetOperators(tao->ksp,tao->hessian,tao->hessian_pre);CHKERRQ(ierr); 303 if (bnk->is_nash || bnk->is_stcg || bnk->is_gltr) { 304 ierr = KSPCGSetRadius(tao->ksp,bnk->max_radius);CHKERRQ(ierr); 305 ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr); 306 ierr = KSPGetIterationNumber(tao->ksp,&kspits);CHKERRQ(ierr); 307 tao->ksp_its+=kspits; 308 tao->ksp_tot_its+=kspits; 309 ierr = KSPCGGetNormD(tao->ksp,&bnk->dnorm);CHKERRQ(ierr); 310 311 if (0.0 == tao->trust) { 312 /* Radius was uninitialized; use the norm of the direction */ 313 if (bnk->dnorm > 0.0) { 314 tao->trust = bnk->dnorm; 315 316 /* Modify the radius if it is too large or small */ 317 tao->trust = PetscMax(tao->trust, bnk->min_radius); 318 tao->trust = PetscMin(tao->trust, bnk->max_radius); 319 } else { 320 /* The direction was bad; set radius to default value and re-solve 321 the trust-region subproblem to get a direction */ 322 tao->trust = tao->trust0; 323 324 /* Modify the radius if it is too large or small */ 325 tao->trust = PetscMax(tao->trust, bnk->min_radius); 326 tao->trust = PetscMin(tao->trust, bnk->max_radius); 327 328 ierr = KSPCGSetRadius(tao->ksp,bnk->max_radius);CHKERRQ(ierr); 329 ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr); 330 ierr = KSPGetIterationNumber(tao->ksp,&kspits);CHKERRQ(ierr); 331 tao->ksp_its+=kspits; 332 tao->ksp_tot_its+=kspits; 333 ierr = KSPCGGetNormD(tao->ksp,&bnk->dnorm);CHKERRQ(ierr); 334 335 if (bnk->dnorm == 0.0) SETERRQ(PETSC_COMM_SELF,1, "Initial direction zero"); 336 } 337 } 338 } else { 339 ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr); 340 ierr = KSPGetIterationNumber(tao->ksp, &kspits);CHKERRQ(ierr); 341 tao->ksp_its += kspits; 342 tao->ksp_tot_its+=kspits; 343 } 344 ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr); 345 ierr = KSPGetConvergedReason(tao->ksp, &ksp_reason);CHKERRQ(ierr); 346 if ((KSP_DIVERGED_INDEFINITE_PC == ksp_reason) && (BNK_PC_BFGS == bnk->pc_type) && (bfgsUpdates > 1)) { 347 /* Preconditioner is numerically indefinite; reset the 348 approximate if using BFGS preconditioning. */ 349 350 if (bnk->f != 0.0) { 351 delta = 2.0 * PetscAbsScalar(bnk->f) / (bnk->gnorm*bnk->gnorm); 352 } else { 353 delta = 2.0 / (bnk->gnorm*bnk->gnorm); 354 } 355 ierr = MatLMVMSetDelta(bnk->M,delta);CHKERRQ(ierr); 356 ierr = MatLMVMReset(bnk->M);CHKERRQ(ierr); 357 ierr = MatLMVMUpdate(bnk->M, tao->solution, tao->gradient);CHKERRQ(ierr); 358 bfgsUpdates = 1; 359 } 360 361 if (KSP_CONVERGED_ATOL == ksp_reason) { 362 ++bnk->ksp_atol; 363 } else if (KSP_CONVERGED_RTOL == ksp_reason) { 364 ++bnk->ksp_rtol; 365 } else if (KSP_CONVERGED_CG_CONSTRAINED == ksp_reason) { 366 ++bnk->ksp_ctol; 367 } else if (KSP_CONVERGED_CG_NEG_CURVE == ksp_reason) { 368 ++bnk->ksp_negc; 369 } else if (KSP_DIVERGED_DTOL == ksp_reason) { 370 ++bnk->ksp_dtol; 371 } else if (KSP_DIVERGED_ITS == ksp_reason) { 372 ++bnk->ksp_iter; 373 } else { 374 ++bnk->ksp_othr; 375 } 376 377 /* Check for success (descent direction) */ 378 ierr = VecDot(tao->stepdirection, tao->gradient, &gdx);CHKERRQ(ierr); 379 if ((gdx >= 0.0) || PetscIsInfOrNanReal(gdx)) { 380 /* Newton step is not descent or direction produced Inf or NaN 381 Update the perturbation for next time */ 382 if (bnk->pert <= 0.0) { 383 /* Initialize the perturbation */ 384 bnk->pert = PetscMin(bnk->imax, PetscMax(bnk->imin, bnk->imfac * bnk->gnorm)); 385 if (bnk->is_gltr) { 386 ierr = KSPCGGLTRGetMinEig(tao->ksp,&e_min);CHKERRQ(ierr); 387 bnk->pert = PetscMax(bnk->pert, -e_min); 388 } 389 } else { 390 /* Increase the perturbation */ 391 bnk->pert = PetscMin(bnk->pmax, PetscMax(bnk->pgfac * bnk->pert, bnk->pmgfac * bnk->gnorm)); 392 } 393 394 if (BNK_PC_BFGS != bnk->pc_type) { 395 /* We don't have the bfgs matrix around and updated 396 Must use gradient direction in this case */ 397 ierr = VecCopy(tao->gradient, tao->stepdirection);CHKERRQ(ierr); 398 ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr); 399 ++bnk->grad; 400 *stepType = BNK_GRADIENT; 401 } else { 402 /* Attempt to use the BFGS direction */ 403 ierr = MatLMVMSolve(bnk->M, tao->gradient, tao->stepdirection);CHKERRQ(ierr); 404 ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr); 405 406 /* Check for success (descent direction) */ 407 ierr = VecDot(tao->gradient, tao->stepdirection, &gdx);CHKERRQ(ierr); 408 if ((gdx >= 0) || PetscIsInfOrNanReal(gdx)) { 409 /* BFGS direction is not descent or direction produced not a number 410 We can assert bfgsUpdates > 1 in this case because 411 the first solve produces the scaled gradient direction, 412 which is guaranteed to be descent */ 413 414 /* Use steepest descent direction (scaled) */ 415 416 if (bnk->f != 0.0) { 417 delta = 2.0 * PetscAbsScalar(bnk->f) / (bnk->gnorm*bnk->gnorm); 418 } else { 419 delta = 2.0 / (bnk->gnorm*bnk->gnorm); 420 } 421 ierr = MatLMVMSetDelta(bnk->M, delta);CHKERRQ(ierr); 422 ierr = MatLMVMReset(bnk->M);CHKERRQ(ierr); 423 ierr = MatLMVMUpdate(bnk->M, tao->solution, tao->gradient);CHKERRQ(ierr); 424 ierr = MatLMVMSolve(bnk->M, tao->gradient, tao->stepdirection);CHKERRQ(ierr); 425 ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr); 426 427 bfgsUpdates = 1; 428 ++bnk->sgrad; 429 *stepType = BNK_SCALED_GRADIENT; 430 } else { 431 if (1 == bfgsUpdates) { 432 /* The first BFGS direction is always the scaled gradient */ 433 ++bnk->sgrad; 434 *stepType = BNK_SCALED_GRADIENT; 435 } else { 436 ++bnk->bfgs; 437 *stepType = BNK_BFGS; 438 } 439 } 440 } 441 } else { 442 /* Computed Newton step is descent */ 443 switch (ksp_reason) { 444 case KSP_DIVERGED_NANORINF: 445 case KSP_DIVERGED_BREAKDOWN: 446 case KSP_DIVERGED_INDEFINITE_MAT: 447 case KSP_DIVERGED_INDEFINITE_PC: 448 case KSP_CONVERGED_CG_NEG_CURVE: 449 /* Matrix or preconditioner is indefinite; increase perturbation */ 450 if (bnk->pert <= 0.0) { 451 /* Initialize the perturbation */ 452 bnk->pert = PetscMin(bnk->imax, PetscMax(bnk->imin, bnk->imfac * bnk->gnorm)); 453 if (bnk->is_gltr) { 454 ierr = KSPCGGLTRGetMinEig(tao->ksp, &e_min);CHKERRQ(ierr); 455 bnk->pert = PetscMax(bnk->pert, -e_min); 456 } 457 } else { 458 /* Increase the perturbation */ 459 bnk->pert = PetscMin(bnk->pmax, PetscMax(bnk->pgfac * bnk->pert, bnk->pmgfac * bnk->gnorm)); 460 } 461 break; 462 463 default: 464 /* Newton step computation is good; decrease perturbation */ 465 bnk->pert = PetscMin(bnk->psfac * bnk->pert, bnk->pmsfac * bnk->gnorm); 466 if (bnk->pert < bnk->pmin) { 467 bnk->pert = 0.0; 468 } 469 break; 470 } 471 472 ++bnk->newt; 473 stepType = BNK_NEWTON; 474 } 475 PetscFunctionReturn(0); 476 } 477 478 PetscErrorCode TaoBNKUpdateTrustRadius(Tao tao, PetscReal fold, PetscReal fnew, PetscInt stepType, PetscBool *accept) 479 { 480 TAO_BNK *bnk = (TAO_BNK *)tao->data; 481 PetscErrorCode ierr; 482 483 PetscReal step, prered, actred, kappa; 484 PetscReal gdx, tau_1, tau_2, tau_min, tau_max; 485 486 PetscFunctionBegin; 487 /* Update trust region radius */ 488 *accept = PETSC_FALSE; 489 switch(bnk->update_type) { 490 case BNK_UPDATE_STEP: 491 if (stepType == BNK_NEWTON) { 492 ierr = TaoLineSearchGetStepLength(tao->linesearch, &step);CHKERRQ(ierr); 493 if (step < bnk->nu1) { 494 /* Very bad step taken; reduce radius */ 495 tao->trust = bnk->omega1 * PetscMin(bnk->dnorm, tao->trust); 496 } else if (step < bnk->nu2) { 497 /* Reasonably bad step taken; reduce radius */ 498 tao->trust = bnk->omega2 * PetscMin(bnk->dnorm, tao->trust); 499 } else if (step < bnk->nu3) { 500 /* Reasonable step was taken; leave radius alone */ 501 if (bnk->omega3 < 1.0) { 502 tao->trust = bnk->omega3 * PetscMin(bnk->dnorm, tao->trust); 503 } else if (bnk->omega3 > 1.0) { 504 tao->trust = PetscMax(bnk->omega3 * bnk->dnorm, tao->trust); 505 } 506 } else if (step < bnk->nu4) { 507 /* Full step taken; increase the radius */ 508 tao->trust = PetscMax(bnk->omega4 * bnk->dnorm, tao->trust); 509 } else { 510 /* More than full step taken; increase the radius */ 511 tao->trust = PetscMax(bnk->omega5 * bnk->dnorm, tao->trust); 512 } 513 } else { 514 /* Newton step was not good; reduce the radius */ 515 tao->trust = bnk->omega1 * PetscMin(bnk->dnorm, tao->trust); 516 } 517 break; 518 519 case BNK_UPDATE_REDUCTION: 520 if (stepType == BNK_NEWTON) { 521 /* Get predicted reduction */ 522 ierr = KSPCGGetObjFcn(tao->ksp,&prered);CHKERRQ(ierr); 523 if (prered >= 0.0) { 524 /* The predicted reduction has the wrong sign. This cannot */ 525 /* happen in infinite precision arithmetic. Step should */ 526 /* be rejected! */ 527 tao->trust = bnk->alpha1 * PetscMin(tao->trust, bnk->dnorm); 528 } else { 529 if (PetscIsInfOrNanReal(fnew)) { 530 tao->trust = bnk->alpha1 * PetscMin(tao->trust, bnk->dnorm); 531 } else { 532 /* Compute and actual reduction */ 533 actred = fold - fnew; 534 prered = -prered; 535 if ((PetscAbsScalar(actred) <= bnk->epsilon) && 536 (PetscAbsScalar(prered) <= bnk->epsilon)) { 537 kappa = 1.0; 538 } else { 539 kappa = actred / prered; 540 } 541 /* Accept of reject the step and update radius */ 542 if (kappa < bnk->eta1) { 543 /* Very bad step, rejected */ 544 tao->trust = bnk->alpha1 * PetscMin(tao->trust, bnk->dnorm); 545 } else { 546 /* Accept step here */ 547 *accept = PETSC_TRUE; 548 /* Adjust trust radius */ 549 if (kappa < bnk->eta2) { 550 /* Marginal bad step */ 551 tao->trust = bnk->alpha2 * PetscMin(tao->trust, bnk->dnorm); 552 } else if (kappa < bnk->eta3) { 553 /* Reasonable step */ 554 if (bnk->alpha3 < 1.0) { 555 tao->trust = bnk->alpha3 * PetscMin(bnk->dnorm, tao->trust); 556 } else if (bnk->alpha3 > 1.0) { 557 tao->trust = PetscMax(bnk->alpha3 * bnk->dnorm, tao->trust); 558 } 559 } else if (kappa < bnk->eta4) { 560 /* Good step */ 561 tao->trust = PetscMax(bnk->alpha4 * bnk->dnorm, tao->trust); 562 } else { 563 /* Very good step */ 564 tao->trust = PetscMax(bnk->alpha5 * bnk->dnorm, tao->trust); 565 } 566 } 567 } 568 } 569 } else { 570 /* Newton step was not good; reduce the radius */ 571 tao->trust = bnk->alpha1 * PetscMin(bnk->dnorm, tao->trust); 572 } 573 break; 574 575 default: 576 if (stepType == BNK_NEWTON) { 577 ierr = KSPCGGetObjFcn(tao->ksp,&prered);CHKERRQ(ierr); 578 if (prered >= 0.0) { 579 /* The predicted reduction has the wrong sign. This cannot */ 580 /* happen in infinite precision arithmetic. Step should */ 581 /* be rejected! */ 582 tao->trust = bnk->gamma1 * PetscMin(tao->trust, bnk->dnorm); 583 } else { 584 if (PetscIsInfOrNanReal(fnew)) { 585 tao->trust = bnk->gamma1 * PetscMin(tao->trust, bnk->dnorm); 586 } else { 587 actred = fold - fnew; 588 prered = -prered; 589 if ((PetscAbsScalar(actred) <= bnk->epsilon) && (PetscAbsScalar(prered) <= bnk->epsilon)) { 590 kappa = 1.0; 591 } else { 592 kappa = actred / prered; 593 } 594 595 ierr = VecDot(tao->gradient, tao->stepdirection, &gdx);CHKERRQ(ierr); 596 tau_1 = bnk->theta * gdx / (bnk->theta * gdx - (1.0 - bnk->theta) * prered + actred); 597 tau_2 = bnk->theta * gdx / (bnk->theta * gdx + (1.0 + bnk->theta) * prered - actred); 598 tau_min = PetscMin(tau_1, tau_2); 599 tau_max = PetscMax(tau_1, tau_2); 600 601 if (kappa >= 1.0 - bnk->mu1) { 602 /* Great agreement */ 603 *accept = PETSC_TRUE; 604 if (tau_max < 1.0) { 605 tao->trust = PetscMax(tao->trust, bnk->gamma3 * bnk->dnorm); 606 } else if (tau_max > bnk->gamma4) { 607 tao->trust = PetscMax(tao->trust, bnk->gamma4 * bnk->dnorm); 608 } else { 609 tao->trust = PetscMax(tao->trust, tau_max * bnk->dnorm); 610 } 611 } else if (kappa >= 1.0 - bnk->mu2) { 612 /* Good agreement */ 613 *accept = PETSC_TRUE; 614 if (tau_max < bnk->gamma2) { 615 tao->trust = bnk->gamma2 * PetscMin(tao->trust, bnk->dnorm); 616 } else if (tau_max > bnk->gamma3) { 617 tao->trust = PetscMax(tao->trust, bnk->gamma3 * bnk->dnorm); 618 } else if (tau_max < 1.0) { 619 tao->trust = tau_max * PetscMin(tao->trust, bnk->dnorm); 620 } else { 621 tao->trust = PetscMax(tao->trust, tau_max * bnk->dnorm); 622 } 623 } else { 624 /* Not good agreement */ 625 if (tau_min > 1.0) { 626 tao->trust = bnk->gamma2 * PetscMin(tao->trust, bnk->dnorm); 627 } else if (tau_max < bnk->gamma1) { 628 tao->trust = bnk->gamma1 * PetscMin(tao->trust, bnk->dnorm); 629 } else if ((tau_min < bnk->gamma1) && (tau_max >= 1.0)) { 630 tao->trust = bnk->gamma1 * PetscMin(tao->trust, bnk->dnorm); 631 } else if ((tau_1 >= bnk->gamma1) && (tau_1 < 1.0) && ((tau_2 < bnk->gamma1) || (tau_2 >= 1.0))) { 632 tao->trust = tau_1 * PetscMin(tao->trust, bnk->dnorm); 633 } else if ((tau_2 >= bnk->gamma1) && (tau_2 < 1.0) && ((tau_1 < bnk->gamma1) || (tau_2 >= 1.0))) { 634 tao->trust = tau_2 * PetscMin(tao->trust, bnk->dnorm); 635 } else { 636 tao->trust = tau_max * PetscMin(tao->trust, bnk->dnorm); 637 } 638 } 639 } 640 } 641 } else { 642 /* Newton step was not good; reduce the radius */ 643 tao->trust = bnk->gamma1 * PetscMin(bnk->dnorm, tao->trust); 644 } 645 /* The radius may have been increased; modify if it is too large */ 646 tao->trust = PetscMin(tao->trust, bnk->max_radius); 647 } 648 PetscFunctionReturn(0); 649 } 650 651 /* ---------------------------------------------------------- */ 652 static PetscErrorCode TaoSetUp_BNK(Tao tao) 653 { 654 TAO_BNK *bnk = (TAO_BNK *)tao->data; 655 PetscErrorCode ierr; 656 657 PetscFunctionBegin; 658 if (!tao->gradient) {ierr = VecDuplicate(tao->solution,&tao->gradient);CHKERRQ(ierr);} 659 if (!tao->stepdirection) {ierr = VecDuplicate(tao->solution,&tao->stepdirection);CHKERRQ(ierr);} 660 if (!bnk->W) {ierr = VecDuplicate(tao->solution,&bnk->W);CHKERRQ(ierr);} 661 if (!bnk->Xold) {ierr = VecDuplicate(tao->solution,&bnk->Xold);CHKERRQ(ierr);} 662 if (!bnk->Gold) {ierr = VecDuplicate(tao->solution,&bnk->Gold);CHKERRQ(ierr);} 663 bnk->Diag = 0; 664 bnk->M = 0; 665 PetscFunctionReturn(0); 666 } 667 668 /*------------------------------------------------------------*/ 669 static PetscErrorCode TaoDestroy_BNK(Tao tao) 670 { 671 TAO_BNK *bnk = (TAO_BNK *)tao->data; 672 PetscErrorCode ierr; 673 674 PetscFunctionBegin; 675 if (tao->setupcalled) { 676 ierr = VecDestroy(&bnk->W);CHKERRQ(ierr); 677 ierr = VecDestroy(&bnk->Xold);CHKERRQ(ierr); 678 ierr = VecDestroy(&bnk->Gold);CHKERRQ(ierr); 679 } 680 ierr = VecDestroy(&bnk->Diag);CHKERRQ(ierr); 681 ierr = MatDestroy(&bnk->M);CHKERRQ(ierr); 682 ierr = PetscFree(tao->data);CHKERRQ(ierr); 683 PetscFunctionReturn(0); 684 } 685 686 /*------------------------------------------------------------*/ 687 static PetscErrorCode TaoSetFromOptions_BNK(PetscOptionItems *PetscOptionsObject,Tao tao) 688 { 689 TAO_BNK *bnk = (TAO_BNK *)tao->data; 690 PetscErrorCode ierr; 691 692 PetscFunctionBegin; 693 ierr = PetscOptionsHead(PetscOptionsObject,"Newton line search method for unconstrained optimization");CHKERRQ(ierr); 694 ierr = PetscOptionsEList("-tao_BNK_pc_type", "pc type", "", BNK_PC, BNK_PC_TYPES, BNK_PC[bnk->pc_type], &bnk->pc_type, 0);CHKERRQ(ierr); 695 ierr = PetscOptionsEList("-tao_BNK_bfgs_scale_type", "bfgs scale type", "", BFGS_SCALE, BFGS_SCALE_TYPES, BFGS_SCALE[bnk->bfgs_scale_type], &bnk->bfgs_scale_type, 0);CHKERRQ(ierr); 696 ierr = PetscOptionsEList("-tao_BNK_init_type", "radius initialization type", "", BNK_INIT, BNK_INIT_TYPES, BNK_INIT[bnk->init_type], &bnk->init_type, 0);CHKERRQ(ierr); 697 ierr = PetscOptionsEList("-tao_BNK_update_type", "radius update type", "", BNK_UPDATE, BNK_UPDATE_TYPES, BNK_UPDATE[bnk->update_type], &bnk->update_type, 0);CHKERRQ(ierr); 698 ierr = PetscOptionsReal("-tao_BNK_sval", "perturbation starting value", "", bnk->sval, &bnk->sval,NULL);CHKERRQ(ierr); 699 ierr = PetscOptionsReal("-tao_BNK_imin", "minimum initial perturbation", "", bnk->imin, &bnk->imin,NULL);CHKERRQ(ierr); 700 ierr = PetscOptionsReal("-tao_BNK_imax", "maximum initial perturbation", "", bnk->imax, &bnk->imax,NULL);CHKERRQ(ierr); 701 ierr = PetscOptionsReal("-tao_BNK_imfac", "initial merit factor", "", bnk->imfac, &bnk->imfac,NULL);CHKERRQ(ierr); 702 ierr = PetscOptionsReal("-tao_BNK_pmin", "minimum perturbation", "", bnk->pmin, &bnk->pmin,NULL);CHKERRQ(ierr); 703 ierr = PetscOptionsReal("-tao_BNK_pmax", "maximum perturbation", "", bnk->pmax, &bnk->pmax,NULL);CHKERRQ(ierr); 704 ierr = PetscOptionsReal("-tao_BNK_pgfac", "growth factor", "", bnk->pgfac, &bnk->pgfac,NULL);CHKERRQ(ierr); 705 ierr = PetscOptionsReal("-tao_BNK_psfac", "shrink factor", "", bnk->psfac, &bnk->psfac,NULL);CHKERRQ(ierr); 706 ierr = PetscOptionsReal("-tao_BNK_pmgfac", "merit growth factor", "", bnk->pmgfac, &bnk->pmgfac,NULL);CHKERRQ(ierr); 707 ierr = PetscOptionsReal("-tao_BNK_pmsfac", "merit shrink factor", "", bnk->pmsfac, &bnk->pmsfac,NULL);CHKERRQ(ierr); 708 ierr = PetscOptionsReal("-tao_BNK_eta1", "poor steplength; reduce radius", "", bnk->eta1, &bnk->eta1,NULL);CHKERRQ(ierr); 709 ierr = PetscOptionsReal("-tao_BNK_eta2", "reasonable steplength; leave radius alone", "", bnk->eta2, &bnk->eta2,NULL);CHKERRQ(ierr); 710 ierr = PetscOptionsReal("-tao_BNK_eta3", "good steplength; increase radius", "", bnk->eta3, &bnk->eta3,NULL);CHKERRQ(ierr); 711 ierr = PetscOptionsReal("-tao_BNK_eta4", "excellent steplength; greatly increase radius", "", bnk->eta4, &bnk->eta4,NULL);CHKERRQ(ierr); 712 ierr = PetscOptionsReal("-tao_BNK_alpha1", "", "", bnk->alpha1, &bnk->alpha1,NULL);CHKERRQ(ierr); 713 ierr = PetscOptionsReal("-tao_BNK_alpha2", "", "", bnk->alpha2, &bnk->alpha2,NULL);CHKERRQ(ierr); 714 ierr = PetscOptionsReal("-tao_BNK_alpha3", "", "", bnk->alpha3, &bnk->alpha3,NULL);CHKERRQ(ierr); 715 ierr = PetscOptionsReal("-tao_BNK_alpha4", "", "", bnk->alpha4, &bnk->alpha4,NULL);CHKERRQ(ierr); 716 ierr = PetscOptionsReal("-tao_BNK_alpha5", "", "", bnk->alpha5, &bnk->alpha5,NULL);CHKERRQ(ierr); 717 ierr = PetscOptionsReal("-tao_BNK_nu1", "poor steplength; reduce radius", "", bnk->nu1, &bnk->nu1,NULL);CHKERRQ(ierr); 718 ierr = PetscOptionsReal("-tao_BNK_nu2", "reasonable steplength; leave radius alone", "", bnk->nu2, &bnk->nu2,NULL);CHKERRQ(ierr); 719 ierr = PetscOptionsReal("-tao_BNK_nu3", "good steplength; increase radius", "", bnk->nu3, &bnk->nu3,NULL);CHKERRQ(ierr); 720 ierr = PetscOptionsReal("-tao_BNK_nu4", "excellent steplength; greatly increase radius", "", bnk->nu4, &bnk->nu4,NULL);CHKERRQ(ierr); 721 ierr = PetscOptionsReal("-tao_BNK_omega1", "", "", bnk->omega1, &bnk->omega1,NULL);CHKERRQ(ierr); 722 ierr = PetscOptionsReal("-tao_BNK_omega2", "", "", bnk->omega2, &bnk->omega2,NULL);CHKERRQ(ierr); 723 ierr = PetscOptionsReal("-tao_BNK_omega3", "", "", bnk->omega3, &bnk->omega3,NULL);CHKERRQ(ierr); 724 ierr = PetscOptionsReal("-tao_BNK_omega4", "", "", bnk->omega4, &bnk->omega4,NULL);CHKERRQ(ierr); 725 ierr = PetscOptionsReal("-tao_BNK_omega5", "", "", bnk->omega5, &bnk->omega5,NULL);CHKERRQ(ierr); 726 ierr = PetscOptionsReal("-tao_BNK_mu1_i", "", "", bnk->mu1_i, &bnk->mu1_i,NULL);CHKERRQ(ierr); 727 ierr = PetscOptionsReal("-tao_BNK_mu2_i", "", "", bnk->mu2_i, &bnk->mu2_i,NULL);CHKERRQ(ierr); 728 ierr = PetscOptionsReal("-tao_BNK_gamma1_i", "", "", bnk->gamma1_i, &bnk->gamma1_i,NULL);CHKERRQ(ierr); 729 ierr = PetscOptionsReal("-tao_BNK_gamma2_i", "", "", bnk->gamma2_i, &bnk->gamma2_i,NULL);CHKERRQ(ierr); 730 ierr = PetscOptionsReal("-tao_BNK_gamma3_i", "", "", bnk->gamma3_i, &bnk->gamma3_i,NULL);CHKERRQ(ierr); 731 ierr = PetscOptionsReal("-tao_BNK_gamma4_i", "", "", bnk->gamma4_i, &bnk->gamma4_i,NULL);CHKERRQ(ierr); 732 ierr = PetscOptionsReal("-tao_BNK_theta_i", "", "", bnk->theta_i, &bnk->theta_i,NULL);CHKERRQ(ierr); 733 ierr = PetscOptionsReal("-tao_BNK_mu1", "", "", bnk->mu1, &bnk->mu1,NULL);CHKERRQ(ierr); 734 ierr = PetscOptionsReal("-tao_BNK_mu2", "", "", bnk->mu2, &bnk->mu2,NULL);CHKERRQ(ierr); 735 ierr = PetscOptionsReal("-tao_BNK_gamma1", "", "", bnk->gamma1, &bnk->gamma1,NULL);CHKERRQ(ierr); 736 ierr = PetscOptionsReal("-tao_BNK_gamma2", "", "", bnk->gamma2, &bnk->gamma2,NULL);CHKERRQ(ierr); 737 ierr = PetscOptionsReal("-tao_BNK_gamma3", "", "", bnk->gamma3, &bnk->gamma3,NULL);CHKERRQ(ierr); 738 ierr = PetscOptionsReal("-tao_BNK_gamma4", "", "", bnk->gamma4, &bnk->gamma4,NULL);CHKERRQ(ierr); 739 ierr = PetscOptionsReal("-tao_BNK_theta", "", "", bnk->theta, &bnk->theta,NULL);CHKERRQ(ierr); 740 ierr = PetscOptionsReal("-tao_BNK_min_radius", "lower bound on initial radius", "", bnk->min_radius, &bnk->min_radius,NULL);CHKERRQ(ierr); 741 ierr = PetscOptionsReal("-tao_BNK_max_radius", "upper bound on radius", "", bnk->max_radius, &bnk->max_radius,NULL);CHKERRQ(ierr); 742 ierr = PetscOptionsReal("-tao_BNK_epsilon", "tolerance used when computing actual and predicted reduction", "", bnk->epsilon, &bnk->epsilon,NULL);CHKERRQ(ierr); 743 ierr = PetscOptionsTail();CHKERRQ(ierr); 744 ierr = TaoLineSearchSetFromOptions(tao->linesearch);CHKERRQ(ierr); 745 ierr = KSPSetFromOptions(tao->ksp);CHKERRQ(ierr); 746 PetscFunctionReturn(0); 747 } 748 749 /*------------------------------------------------------------*/ 750 static PetscErrorCode TaoView_BNK(Tao tao, PetscViewer viewer) 751 { 752 TAO_BNK *bnk = (TAO_BNK *)tao->data; 753 PetscInt nrejects; 754 PetscBool isascii; 755 PetscErrorCode ierr; 756 757 PetscFunctionBegin; 758 ierr = PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&isascii);CHKERRQ(ierr); 759 if (isascii) { 760 ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr); 761 if (BNK_PC_BFGS == bnk->pc_type && bnk->M) { 762 ierr = MatLMVMGetRejects(bnk->M,&nrejects);CHKERRQ(ierr); 763 ierr = PetscViewerASCIIPrintf(viewer, "Rejected matrix updates: %D\n",nrejects);CHKERRQ(ierr); 764 } 765 ierr = PetscViewerASCIIPrintf(viewer, "Newton steps: %D\n", bnk->newt);CHKERRQ(ierr); 766 ierr = PetscViewerASCIIPrintf(viewer, "BFGS steps: %D\n", bnk->bfgs);CHKERRQ(ierr); 767 ierr = PetscViewerASCIIPrintf(viewer, "Scaled gradient steps: %D\n", bnk->sgrad);CHKERRQ(ierr); 768 ierr = PetscViewerASCIIPrintf(viewer, "Gradient steps: %D\n", bnk->grad);CHKERRQ(ierr); 769 ierr = PetscViewerASCIIPrintf(viewer, "KSP termination reasons:\n");CHKERRQ(ierr); 770 ierr = PetscViewerASCIIPrintf(viewer, " atol: %D\n", bnk->ksp_atol);CHKERRQ(ierr); 771 ierr = PetscViewerASCIIPrintf(viewer, " rtol: %D\n", bnk->ksp_rtol);CHKERRQ(ierr); 772 ierr = PetscViewerASCIIPrintf(viewer, " ctol: %D\n", bnk->ksp_ctol);CHKERRQ(ierr); 773 ierr = PetscViewerASCIIPrintf(viewer, " negc: %D\n", bnk->ksp_negc);CHKERRQ(ierr); 774 ierr = PetscViewerASCIIPrintf(viewer, " dtol: %D\n", bnk->ksp_dtol);CHKERRQ(ierr); 775 ierr = PetscViewerASCIIPrintf(viewer, " iter: %D\n", bnk->ksp_iter);CHKERRQ(ierr); 776 ierr = PetscViewerASCIIPrintf(viewer, " othr: %D\n", bnk->ksp_othr);CHKERRQ(ierr); 777 ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr); 778 } 779 PetscFunctionReturn(0); 780 } 781 782 /* ---------------------------------------------------------- */ 783 /*MC 784 TAOBNK - Shared base-type for Bounded Newton-Krylov type algorithms. 785 At each iteration, the BNK method solves the symmetric 786 system of equations to obtain the step diretion dk: 787 Hk dk = -gk 788 at which point the step can be globalized either through trust-region 789 methods, or a line search, or a heuristic mixture of both. 790 791 Options Database Keys: 792 + -tao_BNK_pc_type - "none","ahess","bfgs","petsc" 793 . -tao_BNK_bfgs_scale_type - "ahess","phess","bfgs" 794 . -tao_BNK_init_type - "constant","direction","interpolation" 795 . -tao_BNK_update_type - "step","direction","interpolation" 796 . -tao_BNK_sval - perturbation starting value 797 . -tao_BNK_imin - minimum initial perturbation 798 . -tao_BNK_imax - maximum initial perturbation 799 . -tao_BNK_pmin - minimum perturbation 800 . -tao_BNK_pmax - maximum perturbation 801 . -tao_BNK_pgfac - growth factor 802 . -tao_BNK_psfac - shrink factor 803 . -tao_BNK_imfac - initial merit factor 804 . -tao_BNK_pmgfac - merit growth factor 805 . -tao_BNK_pmsfac - merit shrink factor 806 . -tao_BNK_eta1 - poor steplength; reduce radius 807 . -tao_BNK_eta2 - reasonable steplength; leave radius 808 . -tao_BNK_eta3 - good steplength; increase readius 809 . -tao_BNK_eta4 - excellent steplength; greatly increase radius 810 . -tao_BNK_alpha1 - alpha1 reduction 811 . -tao_BNK_alpha2 - alpha2 reduction 812 . -tao_BNK_alpha3 - alpha3 reduction 813 . -tao_BNK_alpha4 - alpha4 reduction 814 . -tao_BNK_alpha - alpha5 reduction 815 . -tao_BNK_mu1 - mu1 interpolation update 816 . -tao_BNK_mu2 - mu2 interpolation update 817 . -tao_BNK_gamma1 - gamma1 interpolation update 818 . -tao_BNK_gamma2 - gamma2 interpolation update 819 . -tao_BNK_gamma3 - gamma3 interpolation update 820 . -tao_BNK_gamma4 - gamma4 interpolation update 821 . -tao_BNK_theta - theta interpolation update 822 . -tao_BNK_omega1 - omega1 step update 823 . -tao_BNK_omega2 - omega2 step update 824 . -tao_BNK_omega3 - omega3 step update 825 . -tao_BNK_omega4 - omega4 step update 826 . -tao_BNK_omega5 - omega5 step update 827 . -tao_BNK_mu1_i - mu1 interpolation init factor 828 . -tao_BNK_mu2_i - mu2 interpolation init factor 829 . -tao_BNK_gamma1_i - gamma1 interpolation init factor 830 . -tao_BNK_gamma2_i - gamma2 interpolation init factor 831 . -tao_BNK_gamma3_i - gamma3 interpolation init factor 832 . -tao_BNK_gamma4_i - gamma4 interpolation init factor 833 - -tao_BNK_theta_i - theta interpolation init factor 834 835 Level: beginner 836 M*/ 837 838 PetscErrorCode TaoCreate_BNK(Tao tao) 839 { 840 TAO_BNK *bnk; 841 const char *morethuente_type = TAOLINESEARCHMT; 842 PetscErrorCode ierr; 843 844 PetscFunctionBegin; 845 ierr = PetscNewLog(tao,&bnk);CHKERRQ(ierr); 846 847 tao->ops->setup = TaoSetUp_BNK; 848 tao->ops->view = TaoView_BNK; 849 tao->ops->setfromoptions = TaoSetFromOptions_BNK; 850 tao->ops->destroy = TaoDestroy_BNK; 851 852 /* Override default settings (unless already changed) */ 853 if (!tao->max_it_changed) tao->max_it = 50; 854 if (!tao->trust0_changed) tao->trust0 = 100.0; 855 856 tao->data = (void*)bnk; 857 858 bnk->sval = 0.0; 859 bnk->imin = 1.0e-4; 860 bnk->imax = 1.0e+2; 861 bnk->imfac = 1.0e-1; 862 863 bnk->pmin = 1.0e-12; 864 bnk->pmax = 1.0e+2; 865 bnk->pgfac = 1.0e+1; 866 bnk->psfac = 4.0e-1; 867 bnk->pmgfac = 1.0e-1; 868 bnk->pmsfac = 1.0e-1; 869 870 /* Default values for trust-region radius update based on steplength */ 871 bnk->nu1 = 0.25; 872 bnk->nu2 = 0.50; 873 bnk->nu3 = 1.00; 874 bnk->nu4 = 1.25; 875 876 bnk->omega1 = 0.25; 877 bnk->omega2 = 0.50; 878 bnk->omega3 = 1.00; 879 bnk->omega4 = 2.00; 880 bnk->omega5 = 4.00; 881 882 /* Default values for trust-region radius update based on reduction */ 883 bnk->eta1 = 1.0e-4; 884 bnk->eta2 = 0.25; 885 bnk->eta3 = 0.50; 886 bnk->eta4 = 0.90; 887 888 bnk->alpha1 = 0.25; 889 bnk->alpha2 = 0.50; 890 bnk->alpha3 = 1.00; 891 bnk->alpha4 = 2.00; 892 bnk->alpha5 = 4.00; 893 894 /* Default values for trust-region radius update based on interpolation */ 895 bnk->mu1 = 0.10; 896 bnk->mu2 = 0.50; 897 898 bnk->gamma1 = 0.25; 899 bnk->gamma2 = 0.50; 900 bnk->gamma3 = 2.00; 901 bnk->gamma4 = 4.00; 902 903 bnk->theta = 0.05; 904 905 /* Default values for trust region initialization based on interpolation */ 906 bnk->mu1_i = 0.35; 907 bnk->mu2_i = 0.50; 908 909 bnk->gamma1_i = 0.0625; 910 bnk->gamma2_i = 0.5; 911 bnk->gamma3_i = 2.0; 912 bnk->gamma4_i = 5.0; 913 914 bnk->theta_i = 0.25; 915 916 /* Remaining parameters */ 917 bnk->min_radius = 1.0e-10; 918 bnk->max_radius = 1.0e10; 919 bnk->epsilon = 1.0e-6; 920 921 bnk->pc_type = BNK_PC_BFGS; 922 bnk->bfgs_scale_type = BFGS_SCALE_PHESS; 923 bnk->init_type = BNK_INIT_INTERPOLATION; 924 bnk->update_type = BNK_UPDATE_STEP; 925 926 ierr = TaoLineSearchCreate(((PetscObject)tao)->comm,&tao->linesearch);CHKERRQ(ierr); 927 ierr = PetscObjectIncrementTabLevel((PetscObject)tao->linesearch, (PetscObject)tao, 1);CHKERRQ(ierr); 928 ierr = TaoLineSearchSetType(tao->linesearch,morethuente_type);CHKERRQ(ierr); 929 ierr = TaoLineSearchUseTaoRoutines(tao->linesearch,tao);CHKERRQ(ierr); 930 ierr = TaoLineSearchSetOptionsPrefix(tao->linesearch,tao->hdr.prefix);CHKERRQ(ierr); 931 932 /* Set linear solver to default for symmetric matrices */ 933 ierr = KSPCreate(((PetscObject)tao)->comm,&tao->ksp);CHKERRQ(ierr); 934 ierr = PetscObjectIncrementTabLevel((PetscObject)tao->ksp, (PetscObject)tao, 1);CHKERRQ(ierr); 935 ierr = KSPSetOptionsPrefix(tao->ksp,tao->hdr.prefix);CHKERRQ(ierr); 936 ierr = KSPSetType(tao->ksp,KSPCGSTCG);CHKERRQ(ierr); 937 PetscFunctionReturn(0); 938 } 939