1 #include <../src/tao/bound/impls/bqnk/bqnk.h> 2 3 static PetscErrorCode TaoBQNLSComputeHessian(Tao tao) 4 { 5 TAO_BNK *bnk = (TAO_BNK *)tao->data; 6 TAO_BQNK *bqnk = (TAO_BQNK*)bnk->ctx; 7 PetscErrorCode ierr; 8 PetscReal gnorm2, delta; 9 10 PetscFunctionBegin; 11 /* Compute the initial scaling and update the approximation */ 12 gnorm2 = bnk->gnorm*bnk->gnorm; 13 delta = 2.0 * PetscMax(1.0, PetscAbsScalar(bnk->f)) / PetscMax(gnorm2, PetscPowReal(PETSC_MACHINE_EPSILON, 2.0/3.0)); 14 ierr = MatSymBrdnSetDelta(bqnk->B, delta);CHKERRQ(ierr); 15 ierr = MatLMVMUpdate(bqnk->B, tao->solution, bnk->unprojected_gradient);CHKERRQ(ierr); 16 PetscFunctionReturn(0); 17 } 18 19 static PetscErrorCode TaoBQNLSComputeStep(Tao tao, PetscBool shift, KSPConvergedReason *ksp_reason, PetscInt *step_type) 20 { 21 TAO_BNK *bnk = (TAO_BNK *)tao->data; 22 TAO_BQNK *bqnk = (TAO_BQNK*)bnk->ctx; 23 PetscErrorCode ierr; 24 PetscInt nupdates; 25 26 PetscFunctionBegin; 27 ierr = MatSolve(bqnk->B, tao->gradient, tao->stepdirection);CHKERRQ(ierr); 28 ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr); 29 ierr = TaoBNKBoundStep(tao, bnk->as_type, tao->stepdirection);CHKERRQ(ierr); 30 *ksp_reason = KSP_CONVERGED_ATOL; 31 ierr = MatLMVMGetUpdateCount(bqnk->B, &nupdates);CHKERRQ(ierr); 32 if (nupdates == 0) { 33 *step_type = BNK_SCALED_GRADIENT; 34 } else { 35 *step_type = BNK_BFGS; 36 } 37 PetscFunctionReturn(0); 38 } 39 40 static PetscErrorCode TaoSetFromOptions_BQNLS(PetscOptionItems *PetscOptionsObject,Tao tao) 41 { 42 TAO_BNK *bnk = (TAO_BNK *)tao->data; 43 TAO_BQNK *bqnk = (TAO_BQNK*)bnk->ctx; 44 PetscErrorCode ierr; 45 KSPType ksp_type; 46 PetscBool is_spd; 47 48 PetscFunctionBegin; 49 ierr = PetscOptionsHead(PetscOptionsObject,"Quasi-Newton-Krylov method for bound constrained optimization");CHKERRQ(ierr); 50 ierr = PetscOptionsEList("-tao_bqnls_as_type", "active set estimation method", "", BNK_AS, BNK_AS_TYPES, BNK_AS[bnk->as_type], &bnk->as_type, 0);CHKERRQ(ierr); 51 ierr = PetscOptionsReal("-tao_bqnls_epsilon", "(developer) tolerance used when computing actual and predicted reduction", "", bnk->epsilon, &bnk->epsilon,NULL);CHKERRQ(ierr); 52 ierr = PetscOptionsReal("-tao_bqnls_as_tol", "(developer) initial tolerance used when estimating actively bounded variables", "", bnk->as_tol, &bnk->as_tol,NULL);CHKERRQ(ierr); 53 ierr = PetscOptionsReal("-tao_bqnls_as_step", "(developer) step length used when estimating actively bounded variables", "", bnk->as_step, &bnk->as_step,NULL);CHKERRQ(ierr); 54 ierr = PetscOptionsInt("-tao_bqnls_max_cg_its", "number of BNCG iterations to take for each Newton step", "", bnk->max_cg_its, &bnk->max_cg_its,NULL);CHKERRQ(ierr); 55 ierr = PetscOptionsTail();CHKERRQ(ierr); 56 ierr = TaoSetFromOptions(bnk->bncg);CHKERRQ(ierr); 57 ierr = TaoLineSearchSetFromOptions(tao->linesearch);CHKERRQ(ierr); 58 ierr = KSPSetFromOptions(tao->ksp);CHKERRQ(ierr); 59 ierr = KSPGetType(tao->ksp,&ksp_type);CHKERRQ(ierr); 60 bnk->is_nash = bnk->is_gltr = bnk->is_stcg = PETSC_FALSE; 61 ierr = MatSetFromOptions(bqnk->B);CHKERRQ(ierr); 62 ierr = MatGetOption(bqnk->B, MAT_SPD, &is_spd);CHKERRQ(ierr); 63 if (!is_spd) SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_INCOMP, "LMVM matrix must be symmetric positive-definite"); 64 PetscFunctionReturn(0); 65 } 66 67 PETSC_EXTERN PetscErrorCode TaoCreate_BQNLS(Tao tao) 68 { 69 TAO_BNK *bnk; 70 TAO_BQNK *bqnk; 71 PetscErrorCode ierr; 72 73 PetscFunctionBegin; 74 ierr = TaoCreate_BQNK(tao);CHKERRQ(ierr); 75 ierr = KSPSetOptionsPrefix(tao->ksp, "unused");CHKERRQ(ierr); 76 tao->ops->solve = TaoSolve_BNLS; 77 tao->ops->setfromoptions = TaoSetFromOptions_BQNLS; 78 79 bnk = (TAO_BNK*)tao->data; 80 bnk->update_type = BNK_UPDATE_STEP; 81 bnk->computehessian = TaoBQNLSComputeHessian; 82 bnk->computestep = TaoBQNLSComputeStep; 83 84 bqnk = (TAO_BQNK*)bnk->ctx; 85 ierr = MatSetOptionsPrefix(bqnk->B, "tao_bqnls_");CHKERRQ(ierr); 86 ierr = MatSetType(bqnk->B, MATLMVMBFGS);CHKERRQ(ierr); 87 PetscFunctionReturn(0); 88 }