1 #include <petsctaolinesearch.h> 2 #include <../src/tao/matrix/lmvmmat.h> 3 #include <../src/tao/unconstrained/impls/lmvm/lmvm.h> 4 #include <../src/tao/bound/impls/blmvm/blmvm.h> 5 6 /*------------------------------------------------------------*/ 7 #undef __FUNCT__ 8 #define __FUNCT__ "TaoSolve_BLMVM" 9 static PetscErrorCode TaoSolve_BLMVM(Tao tao) 10 { 11 PetscErrorCode ierr; 12 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 13 TaoConvergedReason reason = TAO_CONTINUE_ITERATING; 14 TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING; 15 PetscReal f, fold, gdx, gnorm; 16 PetscReal stepsize = 1.0,delta; 17 18 PetscFunctionBegin; 19 /* Project initial point onto bounds */ 20 ierr = TaoComputeVariableBounds(tao);CHKERRQ(ierr); 21 ierr = VecMedian(tao->XL,tao->solution,tao->XU,tao->solution);CHKERRQ(ierr); 22 ierr = TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);CHKERRQ(ierr); 23 24 25 /* Check convergence criteria */ 26 ierr = TaoComputeObjectiveAndGradient(tao, tao->solution,&f,blmP->unprojected_gradient);CHKERRQ(ierr); 27 ierr = VecBoundGradientProjection(blmP->unprojected_gradient,tao->solution, tao->XL,tao->XU,tao->gradient);CHKERRQ(ierr); 28 29 ierr = TaoGradientNorm(tao, tao->gradient,NORM_2,&gnorm);CHKERRQ(ierr); 30 if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf or NaN"); 31 32 ierr = TaoMonitor(tao, tao->niter, f, gnorm, 0.0, stepsize, &reason);CHKERRQ(ierr); 33 if (reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0); 34 35 /* Set initial scaling for the function */ 36 if (f != 0.0) { 37 delta = 2.0*PetscAbsScalar(f) / (gnorm*gnorm); 38 } else { 39 delta = 2.0 / (gnorm*gnorm); 40 } 41 ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr); 42 ierr = MatLMVMReset(blmP->M);CHKERRQ(ierr); 43 44 /* Set counter for gradient/reset steps */ 45 blmP->grad = 0; 46 blmP->reset = 0; 47 48 /* Have not converged; continue with Newton method */ 49 while (reason == TAO_CONTINUE_ITERATING) { 50 /* Compute direction */ 51 ierr = MatLMVMUpdate(blmP->M, tao->solution, tao->gradient);CHKERRQ(ierr); 52 ierr = MatLMVMSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 53 ierr = VecBoundGradientProjection(tao->stepdirection,tao->solution,tao->XL,tao->XU,tao->gradient);CHKERRQ(ierr); 54 55 /* Check for success (descent direction) */ 56 ierr = VecDot(blmP->unprojected_gradient, tao->gradient, &gdx);CHKERRQ(ierr); 57 if (gdx <= 0) { 58 /* Step is not descent or solve was not successful 59 Use steepest descent direction (scaled) */ 60 ++blmP->grad; 61 62 if (f != 0.0) { 63 delta = 2.0*PetscAbsScalar(f) / (gnorm*gnorm); 64 } else { 65 delta = 2.0 / (gnorm*gnorm); 66 } 67 ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr); 68 ierr = MatLMVMReset(blmP->M);CHKERRQ(ierr); 69 ierr = MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);CHKERRQ(ierr); 70 ierr = MatLMVMSolve(blmP->M,blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 71 } 72 ierr = VecScale(tao->stepdirection,-1.0);CHKERRQ(ierr); 73 74 /* Perform the linesearch */ 75 fold = f; 76 ierr = VecCopy(tao->solution, blmP->Xold);CHKERRQ(ierr); 77 ierr = VecCopy(blmP->unprojected_gradient, blmP->Gold);CHKERRQ(ierr); 78 ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr); 79 ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);CHKERRQ(ierr); 80 ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr); 81 82 if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) { 83 /* Linesearch failed 84 Reset factors and use scaled (projected) gradient step */ 85 ++blmP->reset; 86 87 f = fold; 88 ierr = VecCopy(blmP->Xold, tao->solution);CHKERRQ(ierr); 89 ierr = VecCopy(blmP->Gold, blmP->unprojected_gradient);CHKERRQ(ierr); 90 91 if (f != 0.0) { 92 delta = 2.0* PetscAbsScalar(f) / (gnorm*gnorm); 93 } else { 94 delta = 2.0/ (gnorm*gnorm); 95 } 96 ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr); 97 ierr = MatLMVMReset(blmP->M);CHKERRQ(ierr); 98 ierr = MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);CHKERRQ(ierr); 99 ierr = MatLMVMSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 100 ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr); 101 102 /* This may be incorrect; linesearch has values for stepmax and stepmin 103 that should be reset. */ 104 ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr); 105 ierr = TaoLineSearchApply(tao->linesearch,tao->solution,&f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);CHKERRQ(ierr); 106 ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr); 107 108 if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) { 109 tao->reason = TAO_DIVERGED_LS_FAILURE; 110 break; 111 } 112 } 113 114 /* Check for converged */ 115 ierr = VecBoundGradientProjection(blmP->unprojected_gradient, tao->solution, tao->XL, tao->XU, tao->gradient);CHKERRQ(ierr); 116 ierr = TaoGradientNorm(tao, tao->gradient, NORM_2, &gnorm);CHKERRQ(ierr); 117 118 119 if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Not-a-Number"); 120 tao->niter++; 121 ierr = TaoMonitor(tao, tao->niter, f, gnorm, 0.0, stepsize, &reason);CHKERRQ(ierr); 122 } 123 PetscFunctionReturn(0); 124 } 125 126 #undef __FUNCT__ 127 #define __FUNCT__ "TaoSetup_BLMVM" 128 static PetscErrorCode TaoSetup_BLMVM(Tao tao) 129 { 130 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 131 PetscInt n,N; 132 PetscErrorCode ierr; 133 KSP H0ksp; 134 135 PetscFunctionBegin; 136 /* Existence of tao->solution checked in TaoSetup() */ 137 ierr = VecDuplicate(tao->solution,&blmP->Xold);CHKERRQ(ierr); 138 ierr = VecDuplicate(tao->solution,&blmP->Gold);CHKERRQ(ierr); 139 ierr = VecDuplicate(tao->solution, &blmP->unprojected_gradient);CHKERRQ(ierr); 140 141 if (!tao->stepdirection) { 142 ierr = VecDuplicate(tao->solution, &tao->stepdirection);CHKERRQ(ierr); 143 } 144 if (!tao->gradient) { 145 ierr = VecDuplicate(tao->solution,&tao->gradient);CHKERRQ(ierr); 146 } 147 if (!tao->XL) { 148 ierr = VecDuplicate(tao->solution,&tao->XL);CHKERRQ(ierr); 149 ierr = VecSet(tao->XL,PETSC_NINFINITY);CHKERRQ(ierr); 150 } 151 if (!tao->XU) { 152 ierr = VecDuplicate(tao->solution,&tao->XU);CHKERRQ(ierr); 153 ierr = VecSet(tao->XU,PETSC_INFINITY);CHKERRQ(ierr); 154 } 155 /* Create matrix for the limited memory approximation */ 156 ierr = VecGetLocalSize(tao->solution,&n);CHKERRQ(ierr); 157 ierr = VecGetSize(tao->solution,&N);CHKERRQ(ierr); 158 ierr = MatCreateLMVM(((PetscObject)tao)->comm,n,N,&blmP->M);CHKERRQ(ierr); 159 ierr = MatLMVMAllocateVectors(blmP->M,tao->solution);CHKERRQ(ierr); 160 161 /* If the user has set a matrix to solve as the initial H0, set the options prefix here, and set up the KSP */ 162 if (blmP->H0) { 163 const char *prefix; 164 PC H0pc; 165 166 ierr = MatLMVMSetH0(blmP->M, blmP->H0);CHKERRQ(ierr); 167 ierr = MatLMVMGetH0KSP(blmP->M, &H0ksp);CHKERRQ(ierr); 168 169 ierr = TaoGetOptionsPrefix(tao, &prefix);CHKERRQ(ierr); 170 ierr = KSPSetOptionsPrefix(H0ksp, prefix);CHKERRQ(ierr); 171 ierr = PetscObjectAppendOptionsPrefix((PetscObject)H0ksp, "tao_h0_");CHKERRQ(ierr); 172 ierr = KSPGetPC(H0ksp, &H0pc);CHKERRQ(ierr); 173 ierr = PetscObjectAppendOptionsPrefix((PetscObject)H0pc, "tao_h0_");CHKERRQ(ierr); 174 175 ierr = KSPSetFromOptions(H0ksp);CHKERRQ(ierr); 176 ierr = KSPSetUp(H0ksp);CHKERRQ(ierr); 177 } 178 179 PetscFunctionReturn(0); 180 } 181 182 /* ---------------------------------------------------------- */ 183 #undef __FUNCT__ 184 #define __FUNCT__ "TaoDestroy_BLMVM" 185 static PetscErrorCode TaoDestroy_BLMVM(Tao tao) 186 { 187 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 188 PetscErrorCode ierr; 189 190 PetscFunctionBegin; 191 if (tao->setupcalled) { 192 ierr = MatDestroy(&blmP->M);CHKERRQ(ierr); 193 ierr = VecDestroy(&blmP->unprojected_gradient);CHKERRQ(ierr); 194 ierr = VecDestroy(&blmP->Xold);CHKERRQ(ierr); 195 ierr = VecDestroy(&blmP->Gold);CHKERRQ(ierr); 196 } 197 198 if (blmP->H0) { 199 PetscObjectDereference((PetscObject)blmP->H0); 200 } 201 202 ierr = PetscFree(tao->data);CHKERRQ(ierr); 203 PetscFunctionReturn(0); 204 } 205 206 /*------------------------------------------------------------*/ 207 #undef __FUNCT__ 208 #define __FUNCT__ "TaoSetFromOptions_BLMVM" 209 static PetscErrorCode TaoSetFromOptions_BLMVM(PetscOptionItems* PetscOptionsObject,Tao tao) 210 { 211 PetscErrorCode ierr; 212 213 PetscFunctionBegin; 214 ierr = PetscOptionsHead(PetscOptionsObject,"Limited-memory variable-metric method for bound constrained optimization");CHKERRQ(ierr); 215 ierr = TaoLineSearchSetFromOptions(tao->linesearch);CHKERRQ(ierr); 216 ierr = PetscOptionsTail();CHKERRQ(ierr); 217 PetscFunctionReturn(0); 218 } 219 220 221 /*------------------------------------------------------------*/ 222 #undef __FUNCT__ 223 #define __FUNCT__ "TaoView_BLMVM" 224 static int TaoView_BLMVM(Tao tao, PetscViewer viewer) 225 { 226 TAO_BLMVM *lmP = (TAO_BLMVM *)tao->data; 227 PetscBool isascii; 228 PetscErrorCode ierr; 229 230 PetscFunctionBegin; 231 ierr = PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii);CHKERRQ(ierr); 232 if (isascii) { 233 ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr); 234 ierr = PetscViewerASCIIPrintf(viewer, "Gradient steps: %D\n", lmP->grad);CHKERRQ(ierr); 235 ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr); 236 } 237 PetscFunctionReturn(0); 238 } 239 240 #undef __FUNCT__ 241 #define __FUNCT__ "TaoComputeDual_BLMVM" 242 static PetscErrorCode TaoComputeDual_BLMVM(Tao tao, Vec DXL, Vec DXU) 243 { 244 TAO_BLMVM *blm = (TAO_BLMVM *) tao->data; 245 PetscErrorCode ierr; 246 247 PetscFunctionBegin; 248 PetscValidHeaderSpecific(tao,TAO_CLASSID,1); 249 PetscValidHeaderSpecific(DXL,VEC_CLASSID,2); 250 PetscValidHeaderSpecific(DXU,VEC_CLASSID,3); 251 if (!tao->gradient || !blm->unprojected_gradient) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Dual variables don't exist yet or no longer exist.\n"); 252 253 ierr = VecCopy(tao->gradient,DXL);CHKERRQ(ierr); 254 ierr = VecAXPY(DXL,-1.0,blm->unprojected_gradient);CHKERRQ(ierr); 255 ierr = VecSet(DXU,0.0);CHKERRQ(ierr); 256 ierr = VecPointwiseMax(DXL,DXL,DXU);CHKERRQ(ierr); 257 258 ierr = VecCopy(blm->unprojected_gradient,DXU);CHKERRQ(ierr); 259 ierr = VecAXPY(DXU,-1.0,tao->gradient);CHKERRQ(ierr); 260 ierr = VecAXPY(DXU,1.0,DXL);CHKERRQ(ierr); 261 PetscFunctionReturn(0); 262 } 263 264 /* ---------------------------------------------------------- */ 265 /*MC 266 TAOBLMVM - Bounded limited memory variable metric is a quasi-Newton method 267 for nonlinear minimization with bound constraints. It is an extension 268 of TAOLMVM 269 270 Options Database Keys: 271 + -tao_lmm_vectors - number of vectors to use for approximation 272 . -tao_lmm_scale_type - "none","scalar","broyden" 273 . -tao_lmm_limit_type - "none","average","relative","absolute" 274 . -tao_lmm_rescale_type - "none","scalar","gl" 275 . -tao_lmm_limit_mu - mu limiting factor 276 . -tao_lmm_limit_nu - nu limiting factor 277 . -tao_lmm_delta_min - minimum delta value 278 . -tao_lmm_delta_max - maximum delta value 279 . -tao_lmm_broyden_phi - phi factor for Broyden scaling 280 . -tao_lmm_scalar_alpha - alpha factor for scalar scaling 281 . -tao_lmm_rescale_alpha - alpha factor for rescaling diagonal 282 . -tao_lmm_rescale_beta - beta factor for rescaling diagonal 283 . -tao_lmm_scalar_history - amount of history for scalar scaling 284 . -tao_lmm_rescale_history - amount of history for rescaling diagonal 285 - -tao_lmm_eps - rejection tolerance 286 287 Level: beginner 288 M*/ 289 #undef __FUNCT__ 290 #define __FUNCT__ "TaoCreate_BLMVM" 291 PETSC_EXTERN PetscErrorCode TaoCreate_BLMVM(Tao tao) 292 { 293 TAO_BLMVM *blmP; 294 const char *morethuente_type = TAOLINESEARCHMT; 295 PetscErrorCode ierr; 296 297 PetscFunctionBegin; 298 tao->ops->setup = TaoSetup_BLMVM; 299 tao->ops->solve = TaoSolve_BLMVM; 300 tao->ops->view = TaoView_BLMVM; 301 tao->ops->setfromoptions = TaoSetFromOptions_BLMVM; 302 tao->ops->destroy = TaoDestroy_BLMVM; 303 tao->ops->computedual = TaoComputeDual_BLMVM; 304 305 ierr = PetscNewLog(tao,&blmP);CHKERRQ(ierr); 306 blmP->H0 = NULL; 307 tao->data = (void*)blmP; 308 309 /* Override default settings (unless already changed) */ 310 if (!tao->max_it_changed) tao->max_it = 2000; 311 if (!tao->max_funcs_changed) tao->max_funcs = 4000; 312 313 ierr = TaoLineSearchCreate(((PetscObject)tao)->comm, &tao->linesearch);CHKERRQ(ierr); 314 ierr = TaoLineSearchSetType(tao->linesearch, morethuente_type);CHKERRQ(ierr); 315 ierr = TaoLineSearchUseTaoRoutines(tao->linesearch,tao);CHKERRQ(ierr); 316 ierr = TaoLineSearchSetOptionsPrefix(tao->linesearch,tao->hdr.prefix);CHKERRQ(ierr); 317 PetscFunctionReturn(0); 318 } 319 320 #undef __FUNCT__ 321 #define __FUNCT__ "TaoLMVMSetH0" 322 PETSC_EXTERN PetscErrorCode TaoLMVMSetH0(Tao tao, Mat H0) 323 { 324 TAO_LMVM *lmP; 325 TAO_BLMVM *blmP; 326 const TaoType type; 327 PetscBool is_lmvm, is_blmvm; 328 329 PetscErrorCode ierr; 330 331 ierr = TaoGetType(tao, &type);CHKERRQ(ierr); 332 ierr = PetscStrcmp(type, TAOLMVM, &is_lmvm);CHKERRQ(ierr); 333 ierr = PetscStrcmp(type, TAOBLMVM, &is_blmvm);CHKERRQ(ierr); 334 335 if (is_lmvm) { 336 lmP = (TAO_LMVM *)tao->data; 337 ierr = PetscObjectReference((PetscObject)H0);CHKERRQ(ierr); 338 lmP->H0 = H0; 339 } else if (is_blmvm) { 340 blmP = (TAO_BLMVM *)tao->data; 341 ierr = PetscObjectReference((PetscObject)H0);CHKERRQ(ierr); 342 blmP->H0 = H0; 343 } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONGSTATE, "This routine applies to TAO_LMVM and TAO_BLMVM."); 344 345 PetscFunctionReturn(0); 346 } 347 348 #undef __FUNCT__ 349 #define __FUNCT__ "TaoLMVMGetH0" 350 PETSC_EXTERN PetscErrorCode TaoLMVMGetH0(Tao tao, Mat *H0) 351 { 352 TAO_LMVM *lmP; 353 TAO_BLMVM *blmP; 354 const TaoType type; 355 PetscBool is_lmvm, is_blmvm; 356 Mat M; 357 358 PetscErrorCode ierr; 359 360 ierr = TaoGetType(tao, &type);CHKERRQ(ierr); 361 ierr = PetscStrcmp(type, TAOLMVM, &is_lmvm);CHKERRQ(ierr); 362 ierr = PetscStrcmp(type, TAOBLMVM, &is_blmvm);CHKERRQ(ierr); 363 364 if (is_lmvm) { 365 lmP = (TAO_LMVM *)tao->data; 366 M = lmP->M; 367 } else if (is_blmvm) { 368 blmP = (TAO_BLMVM *)tao->data; 369 M = blmP->M; 370 } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONGSTATE, "This routine applies to TAO_LMVM and TAO_BLMVM."); 371 372 ierr = MatLMVMGetH0(M, H0);CHKERRQ(ierr); 373 PetscFunctionReturn(0); 374 } 375 376 #undef __FUNCT__ 377 #define __FUNCT__ "TaoLMVMGetH0KSP" 378 PETSC_EXTERN PetscErrorCode TaoLMVMGetH0KSP(Tao tao, KSP *ksp) 379 { 380 TAO_LMVM *lmP; 381 TAO_BLMVM *blmP; 382 const TaoType type; 383 PetscBool is_lmvm, is_blmvm; 384 Mat M; 385 PetscErrorCode ierr; 386 387 ierr = TaoGetType(tao, &type);CHKERRQ(ierr); 388 ierr = PetscStrcmp(type, TAOLMVM, &is_lmvm);CHKERRQ(ierr); 389 ierr = PetscStrcmp(type, TAOBLMVM, &is_blmvm);CHKERRQ(ierr); 390 391 if (is_lmvm) { 392 lmP = (TAO_LMVM *)tao->data; 393 M = lmP->M; 394 } else if (is_blmvm) { 395 blmP = (TAO_BLMVM *)tao->data; 396 M = blmP->M; 397 } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONGSTATE, "This routine applies to TAO_LMVM and TAO_BLMVM."); 398 399 ierr = MatLMVMGetH0KSP(M, ksp);CHKERRQ(ierr); 400 PetscFunctionReturn(0); 401 } 402