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