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