1 #include <petsctaolinesearch.h> /*I "petsctaolinesearch.h" I*/ 2 #include <../src/tao/unconstrained/impls/lmvm/lmvm.h> 3 #include <../src/tao/bound/impls/blmvm/blmvm.h> 4 5 /*------------------------------------------------------------*/ 6 static PetscErrorCode TaoSolve_BLMVM(Tao tao) 7 { 8 PetscErrorCode ierr; 9 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 10 TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING; 11 PetscReal f, fold, gdx, gnorm, gnorm2; 12 PetscReal stepsize = 1.0,delta; 13 14 PetscFunctionBegin; 15 /* Project initial point onto bounds */ 16 ierr = TaoComputeVariableBounds(tao);CHKERRQ(ierr); 17 ierr = VecMedian(tao->XL,tao->solution,tao->XU,tao->solution);CHKERRQ(ierr); 18 ierr = TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);CHKERRQ(ierr); 19 20 21 /* Check convergence criteria */ 22 ierr = TaoComputeObjectiveAndGradient(tao, tao->solution,&f,blmP->unprojected_gradient);CHKERRQ(ierr); 23 ierr = VecBoundGradientProjection(blmP->unprojected_gradient,tao->solution, tao->XL,tao->XU,tao->gradient);CHKERRQ(ierr); 24 25 ierr = TaoGradientNorm(tao, tao->gradient,NORM_2,&gnorm);CHKERRQ(ierr); 26 if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PetscObjectComm((PetscObject)tao),PETSC_ERR_USER, "User provided compute function generated Inf or NaN"); 27 28 tao->reason = TAO_CONTINUE_ITERATING; 29 ierr = TaoLogConvergenceHistory(tao,f,gnorm,0.0,tao->ksp_its);CHKERRQ(ierr); 30 ierr = TaoMonitor(tao,tao->niter,f,gnorm,0.0,stepsize);CHKERRQ(ierr); 31 ierr = (*tao->ops->convergencetest)(tao,tao->cnvP);CHKERRQ(ierr); 32 if (tao->reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0); 33 34 /* Set counter for gradient/reset steps */ 35 if (!blmP->recycle) { 36 blmP->grad = 0; 37 blmP->reset = 0; 38 ierr = MatLMVMReset(blmP->M, PETSC_FALSE);CHKERRQ(ierr); 39 } 40 41 /* Have not converged; continue with Newton method */ 42 while (tao->reason == TAO_CONTINUE_ITERATING) { 43 /* Call general purpose update function */ 44 if (tao->ops->update) { 45 ierr = (*tao->ops->update)(tao, tao->niter, tao->user_update);CHKERRQ(ierr); 46 } 47 /* Compute direction */ 48 gnorm2 = gnorm*gnorm; 49 if (gnorm2 == 0.0) gnorm2 = PETSC_MACHINE_EPSILON; 50 if (f == 0.0) { 51 delta = 2.0 / gnorm2; 52 } else { 53 delta = 2.0 * PetscAbsScalar(f) / gnorm2; 54 } 55 ierr = MatLMVMSymBroydenSetDelta(blmP->M, delta);CHKERRQ(ierr); 56 ierr = MatLMVMUpdate(blmP->M, tao->solution, tao->gradient);CHKERRQ(ierr); 57 ierr = MatSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 58 ierr = VecBoundGradientProjection(tao->stepdirection,tao->solution,tao->XL,tao->XU,tao->gradient);CHKERRQ(ierr); 59 60 /* Check for success (descent direction) */ 61 ierr = VecDot(blmP->unprojected_gradient, tao->gradient, &gdx);CHKERRQ(ierr); 62 if (gdx <= 0) { 63 /* Step is not descent or solve was not successful 64 Use steepest descent direction (scaled) */ 65 ++blmP->grad; 66 67 ierr = MatLMVMReset(blmP->M, PETSC_FALSE);CHKERRQ(ierr); 68 ierr = MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);CHKERRQ(ierr); 69 ierr = MatSolve(blmP->M,blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 70 } 71 ierr = VecScale(tao->stepdirection,-1.0);CHKERRQ(ierr); 72 73 /* Perform the linesearch */ 74 fold = f; 75 ierr = VecCopy(tao->solution, blmP->Xold);CHKERRQ(ierr); 76 ierr = VecCopy(blmP->unprojected_gradient, blmP->Gold);CHKERRQ(ierr); 77 ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr); 78 ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);CHKERRQ(ierr); 79 ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr); 80 81 if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) { 82 /* Linesearch failed 83 Reset factors and use scaled (projected) gradient step */ 84 ++blmP->reset; 85 86 f = fold; 87 ierr = VecCopy(blmP->Xold, tao->solution);CHKERRQ(ierr); 88 ierr = VecCopy(blmP->Gold, blmP->unprojected_gradient);CHKERRQ(ierr); 89 90 ierr = MatLMVMReset(blmP->M, PETSC_FALSE);CHKERRQ(ierr); 91 ierr = MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);CHKERRQ(ierr); 92 ierr = MatSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 93 ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr); 94 95 /* This may be incorrect; linesearch has values for stepmax and stepmin 96 that should be reset. */ 97 ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr); 98 ierr = TaoLineSearchApply(tao->linesearch,tao->solution,&f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);CHKERRQ(ierr); 99 ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr); 100 101 if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) { 102 tao->reason = TAO_DIVERGED_LS_FAILURE; 103 break; 104 } 105 } 106 107 /* Check for converged */ 108 ierr = VecBoundGradientProjection(blmP->unprojected_gradient, tao->solution, tao->XL, tao->XU, tao->gradient);CHKERRQ(ierr); 109 ierr = TaoGradientNorm(tao, tao->gradient, NORM_2, &gnorm);CHKERRQ(ierr); 110 if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PetscObjectComm((PetscObject)tao),PETSC_ERR_USER, "User provided compute function generated Not-a-Number"); 111 tao->niter++; 112 ierr = TaoLogConvergenceHistory(tao,f,gnorm,0.0,tao->ksp_its);CHKERRQ(ierr); 113 ierr = TaoMonitor(tao,tao->niter,f,gnorm,0.0,stepsize);CHKERRQ(ierr); 114 ierr = (*tao->ops->convergencetest)(tao,tao->cnvP);CHKERRQ(ierr); 115 } 116 PetscFunctionReturn(0); 117 } 118 119 static PetscErrorCode TaoSetup_BLMVM(Tao tao) 120 { 121 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 122 PetscErrorCode ierr; 123 124 PetscFunctionBegin; 125 /* Existence of tao->solution checked in TaoSetup() */ 126 ierr = VecDuplicate(tao->solution,&blmP->Xold);CHKERRQ(ierr); 127 ierr = VecDuplicate(tao->solution,&blmP->Gold);CHKERRQ(ierr); 128 ierr = VecDuplicate(tao->solution, &blmP->unprojected_gradient);CHKERRQ(ierr); 129 130 if (!tao->stepdirection) { 131 ierr = VecDuplicate(tao->solution, &tao->stepdirection);CHKERRQ(ierr); 132 } 133 if (!tao->gradient) { 134 ierr = VecDuplicate(tao->solution,&tao->gradient);CHKERRQ(ierr); 135 } 136 if (!tao->XL) { 137 ierr = VecDuplicate(tao->solution,&tao->XL);CHKERRQ(ierr); 138 ierr = VecSet(tao->XL,PETSC_NINFINITY);CHKERRQ(ierr); 139 } 140 if (!tao->XU) { 141 ierr = VecDuplicate(tao->solution,&tao->XU);CHKERRQ(ierr); 142 ierr = VecSet(tao->XU,PETSC_INFINITY);CHKERRQ(ierr); 143 } 144 /* Allocate matrix for the limited memory approximation */ 145 ierr = MatLMVMAllocate(blmP->M,tao->solution,blmP->unprojected_gradient);CHKERRQ(ierr); 146 147 /* If the user has set a matrix to solve as the initial H0, set the options prefix here, and set up the KSP */ 148 if (blmP->H0) { 149 ierr = MatLMVMSetJ0(blmP->M, blmP->H0);CHKERRQ(ierr); 150 } 151 PetscFunctionReturn(0); 152 } 153 154 /* ---------------------------------------------------------- */ 155 static PetscErrorCode TaoDestroy_BLMVM(Tao tao) 156 { 157 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 158 PetscErrorCode ierr; 159 160 PetscFunctionBegin; 161 if (tao->setupcalled) { 162 ierr = VecDestroy(&blmP->unprojected_gradient);CHKERRQ(ierr); 163 ierr = VecDestroy(&blmP->Xold);CHKERRQ(ierr); 164 ierr = VecDestroy(&blmP->Gold);CHKERRQ(ierr); 165 } 166 ierr = MatDestroy(&blmP->M);CHKERRQ(ierr); 167 if (blmP->H0) { 168 PetscObjectDereference((PetscObject)blmP->H0); 169 } 170 ierr = PetscFree(tao->data);CHKERRQ(ierr); 171 PetscFunctionReturn(0); 172 } 173 174 /*------------------------------------------------------------*/ 175 static PetscErrorCode TaoSetFromOptions_BLMVM(PetscOptionItems* PetscOptionsObject,Tao tao) 176 { 177 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 178 PetscErrorCode ierr; 179 PetscBool is_spd; 180 181 PetscFunctionBegin; 182 ierr = PetscOptionsHead(PetscOptionsObject,"Limited-memory variable-metric method for bound constrained optimization");CHKERRQ(ierr); 183 ierr = PetscOptionsBool("-tao_blmvm_recycle","enable recycling of the BFGS matrix between subsequent TaoSolve() calls","",blmP->recycle,&blmP->recycle,NULL);CHKERRQ(ierr); 184 ierr = PetscOptionsTail();CHKERRQ(ierr); 185 ierr = TaoLineSearchSetFromOptions(tao->linesearch);CHKERRQ(ierr); 186 ierr = MatSetFromOptions(blmP->M);CHKERRQ(ierr); 187 ierr = MatGetOption(blmP->M, MAT_SPD, &is_spd);CHKERRQ(ierr); 188 if (!is_spd) SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_INCOMP, "LMVM matrix must be symmetric positive-definite"); 189 PetscFunctionReturn(0); 190 } 191 192 193 /*------------------------------------------------------------*/ 194 static PetscErrorCode TaoView_BLMVM(Tao tao, PetscViewer viewer) 195 { 196 TAO_BLMVM *lmP = (TAO_BLMVM *)tao->data; 197 PetscBool isascii; 198 PetscErrorCode ierr; 199 200 PetscFunctionBegin; 201 ierr = PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii);CHKERRQ(ierr); 202 if (isascii) { 203 ierr = PetscViewerASCIIPrintf(viewer, "Gradient steps: %D\n", lmP->grad);CHKERRQ(ierr); 204 ierr = PetscViewerPushFormat(viewer, PETSC_VIEWER_ASCII_INFO);CHKERRQ(ierr); 205 ierr = MatView(lmP->M, viewer);CHKERRQ(ierr); 206 ierr = PetscViewerPopFormat(viewer);CHKERRQ(ierr); 207 } 208 PetscFunctionReturn(0); 209 } 210 211 static PetscErrorCode TaoComputeDual_BLMVM(Tao tao, Vec DXL, Vec DXU) 212 { 213 TAO_BLMVM *blm = (TAO_BLMVM *) tao->data; 214 PetscErrorCode ierr; 215 216 PetscFunctionBegin; 217 PetscValidHeaderSpecific(tao,TAO_CLASSID,1); 218 PetscValidHeaderSpecific(DXL,VEC_CLASSID,2); 219 PetscValidHeaderSpecific(DXU,VEC_CLASSID,3); 220 if (!tao->gradient || !blm->unprojected_gradient) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Dual variables don't exist yet or no longer exist.\n"); 221 222 ierr = VecCopy(tao->gradient,DXL);CHKERRQ(ierr); 223 ierr = VecAXPY(DXL,-1.0,blm->unprojected_gradient);CHKERRQ(ierr); 224 ierr = VecSet(DXU,0.0);CHKERRQ(ierr); 225 ierr = VecPointwiseMax(DXL,DXL,DXU);CHKERRQ(ierr); 226 227 ierr = VecCopy(blm->unprojected_gradient,DXU);CHKERRQ(ierr); 228 ierr = VecAXPY(DXU,-1.0,tao->gradient);CHKERRQ(ierr); 229 ierr = VecAXPY(DXU,1.0,DXL);CHKERRQ(ierr); 230 PetscFunctionReturn(0); 231 } 232 233 /* ---------------------------------------------------------- */ 234 /*MC 235 TAOBLMVM - Bounded limited memory variable metric is a quasi-Newton method 236 for nonlinear minimization with bound constraints. It is an extension 237 of TAOLMVM 238 239 Options Database Keys: 240 . -tao_lmm_recycle - enable recycling of LMVM information between subsequent TaoSolve calls 241 242 Level: beginner 243 M*/ 244 PETSC_EXTERN PetscErrorCode TaoCreate_BLMVM(Tao tao) 245 { 246 TAO_BLMVM *blmP; 247 const char *morethuente_type = TAOLINESEARCHMT; 248 PetscErrorCode ierr; 249 250 PetscFunctionBegin; 251 tao->ops->setup = TaoSetup_BLMVM; 252 tao->ops->solve = TaoSolve_BLMVM; 253 tao->ops->view = TaoView_BLMVM; 254 tao->ops->setfromoptions = TaoSetFromOptions_BLMVM; 255 tao->ops->destroy = TaoDestroy_BLMVM; 256 tao->ops->computedual = TaoComputeDual_BLMVM; 257 258 ierr = PetscNewLog(tao,&blmP);CHKERRQ(ierr); 259 blmP->H0 = NULL; 260 blmP->recycle = PETSC_FALSE; 261 tao->data = (void*)blmP; 262 263 /* Override default settings (unless already changed) */ 264 if (!tao->max_it_changed) tao->max_it = 2000; 265 if (!tao->max_funcs_changed) tao->max_funcs = 4000; 266 267 ierr = TaoLineSearchCreate(((PetscObject)tao)->comm, &tao->linesearch);CHKERRQ(ierr); 268 ierr = PetscObjectIncrementTabLevel((PetscObject)tao->linesearch, (PetscObject)tao, 1);CHKERRQ(ierr); 269 ierr = TaoLineSearchSetType(tao->linesearch, morethuente_type);CHKERRQ(ierr); 270 ierr = TaoLineSearchUseTaoRoutines(tao->linesearch,tao);CHKERRQ(ierr); 271 ierr = TaoLineSearchSetOptionsPrefix(tao->linesearch,tao->hdr.prefix);CHKERRQ(ierr); 272 273 ierr = KSPInitializePackage();CHKERRQ(ierr); 274 ierr = MatCreate(((PetscObject)tao)->comm, &blmP->M);CHKERRQ(ierr); 275 ierr = MatSetType(blmP->M, MATLMVMBFGS);CHKERRQ(ierr); 276 ierr = PetscObjectIncrementTabLevel((PetscObject)blmP->M, (PetscObject)tao, 1);CHKERRQ(ierr); 277 ierr = MatSetOptionsPrefix(blmP->M, "tao_blmvm_");CHKERRQ(ierr); 278 PetscFunctionReturn(0); 279 } 280 281 /*@ 282 TaoLMVMRecycle - Enable/disable recycling of the QN history between subsequent TaoSolve calls. 283 284 Input Parameters: 285 + tao - the Tao solver context 286 - flg - Boolean flag for recycling (PETSC_TRUE or PETSC_FALSE) 287 288 Level: intermediate 289 @*/ 290 PetscErrorCode TaoLMVMRecycle(Tao tao, PetscBool flg) 291 { 292 TAO_LMVM *lmP; 293 TAO_BLMVM *blmP; 294 TaoType type; 295 PetscBool is_lmvm, is_blmvm; 296 PetscErrorCode ierr; 297 298 PetscFunctionBegin; 299 ierr = TaoGetType(tao, &type);CHKERRQ(ierr); 300 ierr = PetscStrcmp(type, TAOLMVM, &is_lmvm);CHKERRQ(ierr); 301 ierr = PetscStrcmp(type, TAOBLMVM, &is_blmvm);CHKERRQ(ierr); 302 303 if (is_lmvm) { 304 lmP = (TAO_LMVM *)tao->data; 305 lmP->recycle = flg; 306 } else if (is_blmvm) { 307 blmP = (TAO_BLMVM *)tao->data; 308 blmP->recycle = flg; 309 } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONGSTATE, "This routine applies to TAO_LMVM and TAO_BLMVM."); 310 PetscFunctionReturn(0); 311 } 312 313 /*@ 314 TaoLMVMSetH0 - Set the initial Hessian for the QN approximation 315 316 Input Parameters: 317 + tao - the Tao solver context 318 - H0 - Mat object for the initial Hessian 319 320 Level: advanced 321 322 .seealso: TaoLMVMGetH0(), TaoLMVMGetH0KSP() 323 @*/ 324 PetscErrorCode TaoLMVMSetH0(Tao tao, Mat H0) 325 { 326 TAO_LMVM *lmP; 327 TAO_BLMVM *blmP; 328 TaoType type; 329 PetscBool is_lmvm, is_blmvm; 330 PetscErrorCode ierr; 331 332 PetscFunctionBegin; 333 ierr = TaoGetType(tao, &type);CHKERRQ(ierr); 334 ierr = PetscStrcmp(type, TAOLMVM, &is_lmvm);CHKERRQ(ierr); 335 ierr = PetscStrcmp(type, TAOBLMVM, &is_blmvm);CHKERRQ(ierr); 336 337 if (is_lmvm) { 338 lmP = (TAO_LMVM *)tao->data; 339 ierr = PetscObjectReference((PetscObject)H0);CHKERRQ(ierr); 340 lmP->H0 = H0; 341 } else if (is_blmvm) { 342 blmP = (TAO_BLMVM *)tao->data; 343 ierr = PetscObjectReference((PetscObject)H0);CHKERRQ(ierr); 344 blmP->H0 = H0; 345 } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONGSTATE, "This routine applies to TAO_LMVM and TAO_BLMVM."); 346 PetscFunctionReturn(0); 347 } 348 349 /*@ 350 TaoLMVMGetH0 - Get the matrix object for the QN initial Hessian 351 352 Input Parameters: 353 . tao - the Tao solver context 354 355 Output Parameters: 356 . H0 - Mat object for the initial Hessian 357 358 Level: advanced 359 360 .seealso: TaoLMVMSetH0(), TaoLMVMGetH0KSP() 361 @*/ 362 PetscErrorCode TaoLMVMGetH0(Tao tao, Mat *H0) 363 { 364 TAO_LMVM *lmP; 365 TAO_BLMVM *blmP; 366 TaoType type; 367 PetscBool is_lmvm, is_blmvm; 368 Mat M; 369 370 PetscErrorCode ierr; 371 372 PetscFunctionBegin; 373 ierr = TaoGetType(tao, &type);CHKERRQ(ierr); 374 ierr = PetscStrcmp(type, TAOLMVM, &is_lmvm);CHKERRQ(ierr); 375 ierr = PetscStrcmp(type, TAOBLMVM, &is_blmvm);CHKERRQ(ierr); 376 377 if (is_lmvm) { 378 lmP = (TAO_LMVM *)tao->data; 379 M = lmP->M; 380 } else if (is_blmvm) { 381 blmP = (TAO_BLMVM *)tao->data; 382 M = blmP->M; 383 } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONGSTATE, "This routine applies to TAO_LMVM and TAO_BLMVM."); 384 ierr = MatLMVMGetJ0(M, H0);CHKERRQ(ierr); 385 PetscFunctionReturn(0); 386 } 387 388 /*@ 389 TaoLMVMGetH0KSP - Get the iterative solver for applying the inverse of the QN initial Hessian 390 391 Input Parameters: 392 . tao - the Tao solver context 393 394 Output Parameters: 395 . ksp - KSP solver context for the initial Hessian 396 397 Level: advanced 398 399 .seealso: TaoLMVMGetH0(), TaoLMVMGetH0KSP() 400 @*/ 401 PetscErrorCode TaoLMVMGetH0KSP(Tao tao, KSP *ksp) 402 { 403 TAO_LMVM *lmP; 404 TAO_BLMVM *blmP; 405 TaoType type; 406 PetscBool is_lmvm, is_blmvm; 407 Mat M; 408 PetscErrorCode ierr; 409 410 ierr = TaoGetType(tao, &type);CHKERRQ(ierr); 411 ierr = PetscStrcmp(type, TAOLMVM, &is_lmvm);CHKERRQ(ierr); 412 ierr = PetscStrcmp(type, TAOBLMVM, &is_blmvm);CHKERRQ(ierr); 413 414 if (is_lmvm) { 415 lmP = (TAO_LMVM *)tao->data; 416 M = lmP->M; 417 } else if (is_blmvm) { 418 blmP = (TAO_BLMVM *)tao->data; 419 M = blmP->M; 420 } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONGSTATE, "This routine applies to TAO_LMVM and TAO_BLMVM."); 421 ierr = MatLMVMGetJ0KSP(M, ksp);CHKERRQ(ierr); 422 PetscFunctionReturn(0); 423 } 424