/*@
TaoLineSearchComputeObjectiveAndGradient - Computes the objective function value at a given point
Collective on Tao
Input Parameters:
+ ls - the TaoLineSearch context
- x - input vector
Output Parameter:
+ f - Objective value at X
- g - Gradient vector at X
Notes: TaoLineSearchComputeObjectiveAndGradient() is typically used within line searches
so most users would not generally call this routine themselves.
Level: developer
.seealso: TaoLineSearchComputeGradient(), TaoLineSearchComputeObjectiveAndGradient(), TaoLineSearchSetObjectiveRoutine()
@*/
PetscErrorCode TaoLineSearchComputeObjectiveAndGradient(TaoLineSearch ls, Vec x, PetscReal *f, Vec g)
{
PetscErrorCode ierr;
PetscFunctionBegin;
PetscValidHeaderSpecific(ls,TAOLINESEARCH_CLASSID,1);
PetscValidHeaderSpecific(x,VEC_CLASSID,2);
PetscValidPointer(f,3);
PetscValidHeaderSpecific(g,VEC_CLASSID,4);
PetscCheckSameComm(ls,1,x,2);
PetscCheckSameComm(ls,1,g,4);
if (ls->usetaoroutines) {
ierr = TaoComputeObjectiveAndGradient(ls->tao,x,f,g);CHKERRQ(ierr);
} else {
ierr = PetscLogEventBegin(TaoLineSearch_EvalEvent,ls,0,0,0);CHKERRQ(ierr);
if (!ls->ops->computeobjective && !ls->ops->computeobjectiveandgradient) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Line Search does not have objective function set");
if (!ls->ops->computegradient && !ls->ops->computeobjectiveandgradient) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"Line Search does not have gradient function set");
PetscStackPush("TaoLineSearch user objective/gradient routine");
if (ls->ops->computeobjectiveandgradient) {
ierr = (*ls->ops->computeobjectiveandgradient)(ls,x,f,g,ls->userctx_funcgrad);CHKERRQ(ierr);
} else {
ierr = (*ls->ops->computeobjective)(ls,x,f,ls->userctx_func);CHKERRQ(ierr);
ierr = (*ls->ops->computegradient)(ls,x,g,ls->userctx_grad);CHKERRQ(ierr);
}
PetscStackPop;
ierr = PetscLogEventEnd(TaoLineSearch_EvalEvent,ls,0,0,0);CHKERRQ(ierr);
ierr = PetscInfo1(ls,"TaoLineSearch Function evaluation: %14.12e\n",(double)(*f));CHKERRQ(ierr);
ls->nfgeval++;
}
PetscFunctionReturn(0);
}
/* evaluate user info at current point */
PetscErrorCode IPMEvaluate(Tao tao)
{
TAO_IPM *ipmP = (TAO_IPM *)tao->data;
PetscErrorCode ierr;
PetscFunctionBegin;
ierr = TaoComputeObjectiveAndGradient(tao,tao->solution,&ipmP->kkt_f,tao->gradient);CHKERRQ(ierr);
ierr = TaoComputeHessian(tao,tao->solution,tao->hessian,tao->hessian_pre);CHKERRQ(ierr);
if (ipmP->me > 0) {
ierr = TaoComputeEqualityConstraints(tao,tao->solution,tao->constraints_equality);CHKERRQ(ierr);
ierr = TaoComputeJacobianEquality(tao,tao->solution,tao->jacobian_equality,tao->jacobian_equality_pre);CHKERRQ(ierr);
}
if (ipmP->mi > 0) {
ierr = TaoComputeInequalityConstraints(tao,tao->solution,tao->constraints_inequality);CHKERRQ(ierr);
ierr = TaoComputeJacobianInequality(tao,tao->solution,tao->jacobian_inequality,tao->jacobian_inequality_pre);CHKERRQ(ierr);
}
if (ipmP->nb > 0) {
/* Ai' = jac_ineq | I (w/lb) | -I (w/ub) */
ierr = IPMUpdateAi(tao);CHKERRQ(ierr);
}
PetscFunctionReturn(0);
}
static PetscErrorCode TaoSolve_OWLQN(Tao tao)
{
TAO_OWLQN *lmP = (TAO_OWLQN *)tao->data;
PetscReal f, fold, gdx, gnorm;
PetscReal step = 1.0;
PetscReal delta;
PetscErrorCode ierr;
PetscInt stepType;
PetscInt iter = 0;
TaoConvergedReason reason = TAO_CONTINUE_ITERATING;
TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING;
PetscFunctionBegin;
if (tao->XL || tao->XU || tao->ops->computebounds) {
ierr = PetscPrintf(((PetscObject)tao)->comm,"WARNING: Variable bounds have been set but will be ignored by owlqn algorithm\n");CHKERRQ(ierr);
}
/* Check convergence criteria */
ierr = TaoComputeObjectiveAndGradient(tao, tao->solution, &f, tao->gradient);CHKERRQ(ierr);
ierr = VecCopy(tao->gradient, lmP->GV);CHKERRQ(ierr);
ierr = ComputePseudoGrad_OWLQN(tao->solution,lmP->GV,lmP->lambda);CHKERRQ(ierr);
ierr = VecNorm(lmP->GV,NORM_2,&gnorm);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf or NaN");
ierr = TaoMonitor(tao, iter, f, gnorm, 0.0, step, &reason);CHKERRQ(ierr);
if (reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0);
/* Set initial scaling for the function */
if (f != 0.0) {
delta = 2.0 * PetscAbsScalar(f) / (gnorm*gnorm);
} else {
delta = 2.0 / (gnorm*gnorm);
}
ierr = MatLMVMSetDelta(lmP->M,delta);CHKERRQ(ierr);
/* Set counter for gradient/reset steps */
lmP->bfgs = 0;
lmP->sgrad = 0;
lmP->grad = 0;
/* Have not converged; continue with Newton method */
while (reason == TAO_CONTINUE_ITERATING) {
/* Compute direction */
ierr = MatLMVMUpdate(lmP->M,tao->solution,tao->gradient);CHKERRQ(ierr);
ierr = MatLMVMSolve(lmP->M, lmP->GV, lmP->D);CHKERRQ(ierr);
ierr = ProjDirect_OWLQN(lmP->D,lmP->GV);CHKERRQ(ierr);
++lmP->bfgs;
/* Check for success (descent direction) */
ierr = VecDot(lmP->D, lmP->GV , &gdx);CHKERRQ(ierr);
if ((gdx <= 0.0) || PetscIsInfOrNanReal(gdx)) {
/* Step is not descent or direction produced not a number
We can assert bfgsUpdates > 1 in this case because
the first solve produces the scaled gradient direction,
which is guaranteed to be descent
Use steepest descent direction (scaled) */
++lmP->grad;
if (f != 0.0) {
delta = 2.0 * PetscAbsScalar(f) / (gnorm*gnorm);
} else {
delta = 2.0 / (gnorm*gnorm);
}
ierr = MatLMVMSetDelta(lmP->M, delta);CHKERRQ(ierr);
ierr = MatLMVMReset(lmP->M);CHKERRQ(ierr);
ierr = MatLMVMUpdate(lmP->M, tao->solution, tao->gradient);CHKERRQ(ierr);
ierr = MatLMVMSolve(lmP->M,lmP->GV, lmP->D);CHKERRQ(ierr);
ierr = ProjDirect_OWLQN(lmP->D,lmP->GV);CHKERRQ(ierr);
lmP->bfgs = 1;
++lmP->sgrad;
stepType = OWLQN_SCALED_GRADIENT;
} else {
if (1 == lmP->bfgs) {
/* The first BFGS direction is always the scaled gradient */
++lmP->sgrad;
stepType = OWLQN_SCALED_GRADIENT;
} else {
++lmP->bfgs;
stepType = OWLQN_BFGS;
}
}
ierr = VecScale(lmP->D, -1.0);CHKERRQ(ierr);
/* Perform the linesearch */
fold = f;
ierr = VecCopy(tao->solution, lmP->Xold);CHKERRQ(ierr);
ierr = VecCopy(tao->gradient, lmP->Gold);CHKERRQ(ierr);
ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &f, lmP->GV, lmP->D, &step,&ls_status);CHKERRQ(ierr);
ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr);
while (((int)ls_status < 0) && (stepType != OWLQN_GRADIENT)) {
/* Reset factors and use scaled gradient step */
f = fold;
ierr = VecCopy(lmP->Xold, tao->solution);CHKERRQ(ierr);
ierr = VecCopy(lmP->Gold, tao->gradient);CHKERRQ(ierr);
ierr = VecCopy(tao->gradient, lmP->GV);CHKERRQ(ierr);
ierr = ComputePseudoGrad_OWLQN(tao->solution,lmP->GV,lmP->lambda);CHKERRQ(ierr);
switch(stepType) {
case OWLQN_BFGS:
/* Failed to obtain acceptable iterate with BFGS step
Attempt to use the scaled gradient direction */
if (f != 0.0) {
delta = 2.0 * PetscAbsScalar(f) / (gnorm*gnorm);
} else {
delta = 2.0 / (gnorm*gnorm);
}
ierr = MatLMVMSetDelta(lmP->M, delta);CHKERRQ(ierr);
ierr = MatLMVMReset(lmP->M);CHKERRQ(ierr);
ierr = MatLMVMUpdate(lmP->M, tao->solution, tao->gradient);CHKERRQ(ierr);
ierr = MatLMVMSolve(lmP->M, lmP->GV, lmP->D);CHKERRQ(ierr);
ierr = ProjDirect_OWLQN(lmP->D,lmP->GV);CHKERRQ(ierr);
lmP->bfgs = 1;
++lmP->sgrad;
stepType = OWLQN_SCALED_GRADIENT;
break;
case OWLQN_SCALED_GRADIENT:
/* The scaled gradient step did not produce a new iterate;
attempt to use the gradient direction.
Need to make sure we are not using a different diagonal scaling */
ierr = MatLMVMSetDelta(lmP->M, 1.0);CHKERRQ(ierr);
ierr = MatLMVMReset(lmP->M);CHKERRQ(ierr);
ierr = MatLMVMUpdate(lmP->M, tao->solution, tao->gradient);CHKERRQ(ierr);
ierr = MatLMVMSolve(lmP->M, lmP->GV, lmP->D);CHKERRQ(ierr);
ierr = ProjDirect_OWLQN(lmP->D,lmP->GV);CHKERRQ(ierr);
lmP->bfgs = 1;
++lmP->grad;
stepType = OWLQN_GRADIENT;
break;
}
ierr = VecScale(lmP->D, -1.0);CHKERRQ(ierr);
/* Perform the linesearch */
ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &f, lmP->GV, lmP->D, &step, &ls_status);CHKERRQ(ierr);
ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr);
}
if ((int)ls_status < 0) {
/* Failed to find an improving point*/
f = fold;
ierr = VecCopy(lmP->Xold, tao->solution);CHKERRQ(ierr);
ierr = VecCopy(lmP->Gold, tao->gradient);CHKERRQ(ierr);
ierr = VecCopy(tao->gradient, lmP->GV);CHKERRQ(ierr);
step = 0.0;
} else {
/* a little hack here, because that gv is used to store g */
ierr = VecCopy(lmP->GV, tao->gradient);CHKERRQ(ierr);
}
ierr = ComputePseudoGrad_OWLQN(tao->solution,lmP->GV,lmP->lambda);CHKERRQ(ierr);
/* Check for termination */
ierr = VecNorm(lmP->GV,NORM_2,&gnorm);CHKERRQ(ierr);
iter++;
ierr = TaoMonitor(tao,iter,f,gnorm,0.0,step,&reason);CHKERRQ(ierr);
if ((int)ls_status < 0) break;
}
PetscFunctionReturn(0);
}
static PetscErrorCode TaoSolve_TRON(Tao tao)
{
TAO_TRON *tron = (TAO_TRON *)tao->data;
PetscErrorCode ierr;
PetscInt its;
TaoConvergedReason reason = TAO_CONTINUE_ITERATING;
TaoLineSearchConvergedReason ls_reason = TAOLINESEARCH_CONTINUE_ITERATING;
PetscReal prered,actred,delta,f,f_new,rhok,gdx,xdiff,stepsize;
PetscFunctionBegin;
tron->pgstepsize=1.0;
tao->trust = tao->trust0;
/* Project the current point onto the feasible set */
ierr = TaoComputeVariableBounds(tao);CHKERRQ(ierr);
ierr = VecMedian(tao->XL,tao->solution,tao->XU,tao->solution);CHKERRQ(ierr);
ierr = TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);CHKERRQ(ierr);
ierr = TaoComputeObjectiveAndGradient(tao,tao->solution,&tron->f,tao->gradient);CHKERRQ(ierr);
ierr = ISDestroy(&tron->Free_Local);CHKERRQ(ierr);
ierr = VecWhichBetween(tao->XL,tao->solution,tao->XU,&tron->Free_Local);CHKERRQ(ierr);
/* Project the gradient and calculate the norm */
ierr = VecBoundGradientProjection(tao->gradient,tao->solution, tao->XL, tao->XU, tao->gradient);CHKERRQ(ierr);
ierr = VecNorm(tao->gradient,NORM_2,&tron->gnorm);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(tron->f) || PetscIsInfOrNanReal(tron->gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf pr NaN");
if (tao->trust <= 0) {
tao->trust=PetscMax(tron->gnorm*tron->gnorm,1.0);
}
tron->stepsize=tao->trust;
ierr = TaoMonitor(tao, tao->niter, tron->f, tron->gnorm, 0.0, tron->stepsize, &reason);CHKERRQ(ierr);
while (reason==TAO_CONTINUE_ITERATING){
tao->ksp_its=0;
ierr = TronGradientProjections(tao,tron);CHKERRQ(ierr);
f=tron->f; delta=tao->trust;
tron->n_free_last = tron->n_free;
ierr = TaoComputeHessian(tao,tao->solution,tao->hessian,tao->hessian_pre);CHKERRQ(ierr);
ierr = ISGetSize(tron->Free_Local, &tron->n_free);CHKERRQ(ierr);
/* If no free variables */
if (tron->n_free == 0) {
actred=0;
ierr = PetscInfo(tao,"No free variables in tron iteration.\n");CHKERRQ(ierr);
ierr = VecNorm(tao->gradient,NORM_2,&tron->gnorm);CHKERRQ(ierr);
ierr = TaoMonitor(tao, tao->niter, tron->f, tron->gnorm, 0.0, delta, &reason);CHKERRQ(ierr);
if (!reason) {
reason = TAO_CONVERGED_STEPTOL;
ierr = TaoSetConvergedReason(tao,reason);CHKERRQ(ierr);
}
break;
}
/* use free_local to mask/submat gradient, hessian, stepdirection */
ierr = TaoVecGetSubVec(tao->gradient,tron->Free_Local,tao->subset_type,0.0,&tron->R);CHKERRQ(ierr);
ierr = TaoVecGetSubVec(tao->gradient,tron->Free_Local,tao->subset_type,0.0,&tron->DXFree);CHKERRQ(ierr);
ierr = VecSet(tron->DXFree,0.0);CHKERRQ(ierr);
ierr = VecScale(tron->R, -1.0);CHKERRQ(ierr);
ierr = TaoMatGetSubMat(tao->hessian, tron->Free_Local, tron->diag, tao->subset_type, &tron->H_sub);CHKERRQ(ierr);
if (tao->hessian == tao->hessian_pre) {
ierr = MatDestroy(&tron->Hpre_sub);CHKERRQ(ierr);
ierr = PetscObjectReference((PetscObject)(tron->H_sub));CHKERRQ(ierr);
tron->Hpre_sub = tron->H_sub;
} else {
ierr = TaoMatGetSubMat(tao->hessian_pre, tron->Free_Local, tron->diag, tao->subset_type,&tron->Hpre_sub);CHKERRQ(ierr);
}
ierr = KSPReset(tao->ksp);CHKERRQ(ierr);
ierr = KSPSetOperators(tao->ksp, tron->H_sub, tron->Hpre_sub);CHKERRQ(ierr);
while (1) {
/* Approximately solve the reduced linear system */
ierr = KSPSTCGSetRadius(tao->ksp,delta);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, tron->R, tron->DXFree);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
tao->ksp_tot_its+=its;
ierr = VecSet(tao->stepdirection,0.0);CHKERRQ(ierr);
/* Add dxfree matrix to compute step direction vector */
ierr = VecISAXPY(tao->stepdirection,tron->Free_Local,1.0,tron->DXFree);CHKERRQ(ierr);
if (0) {
PetscReal rhs,stepnorm;
ierr = VecNorm(tron->R,NORM_2,&rhs);CHKERRQ(ierr);
ierr = VecNorm(tron->DXFree,NORM_2,&stepnorm);CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,"|rhs|=%g\t|s|=%g\n",(double)rhs,(double)stepnorm);CHKERRQ(ierr);
}
ierr = VecDot(tao->gradient, tao->stepdirection, &gdx);CHKERRQ(ierr);
ierr = PetscInfo1(tao,"Expected decrease in function value: %14.12e\n",(double)gdx);CHKERRQ(ierr);
ierr = VecCopy(tao->solution, tron->X_New);CHKERRQ(ierr);
ierr = VecCopy(tao->gradient, tron->G_New);CHKERRQ(ierr);
stepsize=1.0;f_new=f;
ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr);
ierr = TaoLineSearchApply(tao->linesearch, tron->X_New, &f_new, tron->G_New, tao->stepdirection,&stepsize,&ls_reason);CHKERRQ(ierr);CHKERRQ(ierr);
ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr);
ierr = MatMult(tao->hessian, tao->stepdirection, tron->Work);CHKERRQ(ierr);
ierr = VecAYPX(tron->Work, 0.5, tao->gradient);CHKERRQ(ierr);
ierr = VecDot(tao->stepdirection, tron->Work, &prered);CHKERRQ(ierr);
actred = f_new - f;
if (actred<0) {
rhok=PetscAbs(-actred/prered);
} else {
rhok=0.0;
}
/* Compare actual improvement to the quadratic model */
if (rhok > tron->eta1) { /* Accept the point */
/* d = x_new - x */
ierr = VecCopy(tron->X_New, tao->stepdirection);CHKERRQ(ierr);
ierr = VecAXPY(tao->stepdirection, -1.0, tao->solution);CHKERRQ(ierr);
ierr = VecNorm(tao->stepdirection, NORM_2, &xdiff);CHKERRQ(ierr);
xdiff *= stepsize;
/* Adjust trust region size */
if (rhok < tron->eta2 ){
delta = PetscMin(xdiff,delta)*tron->sigma1;
} else if (rhok > tron->eta4 ){
delta= PetscMin(xdiff,delta)*tron->sigma3;
} else if (rhok > tron->eta3 ){
delta=PetscMin(xdiff,delta)*tron->sigma2;
}
ierr = VecBoundGradientProjection(tron->G_New,tron->X_New, tao->XL, tao->XU, tao->gradient);CHKERRQ(ierr);
if (tron->Free_Local) {
ierr = ISDestroy(&tron->Free_Local);CHKERRQ(ierr);
}
ierr = VecWhichBetween(tao->XL, tron->X_New, tao->XU, &tron->Free_Local);CHKERRQ(ierr);
f=f_new;
ierr = VecNorm(tao->gradient,NORM_2,&tron->gnorm);CHKERRQ(ierr);
ierr = VecCopy(tron->X_New, tao->solution);CHKERRQ(ierr);
ierr = VecCopy(tron->G_New, tao->gradient);CHKERRQ(ierr);
break;
}
else if (delta <= 1e-30) {
break;
}
else {
delta /= 4.0;
}
} /* end linear solve loop */
tron->f=f; tron->actred=actred; tao->trust=delta;
tao->niter++;
ierr = TaoMonitor(tao, tao->niter, tron->f, tron->gnorm, 0.0, delta, &reason);CHKERRQ(ierr);
} /* END MAIN LOOP */
PetscFunctionReturn(0);
}
static PetscErrorCode TaoSolve_NTR(Tao tao)
{
TAO_NTR *tr = (TAO_NTR *)tao->data;
PC pc;
KSPConvergedReason ksp_reason;
TaoConvergedReason reason;
PetscReal fmin, ftrial, prered, actred, kappa, sigma, beta;
PetscReal tau, tau_1, tau_2, tau_max, tau_min, max_radius;
PetscReal f, gnorm;
PetscReal delta;
PetscReal norm_d;
PetscErrorCode ierr;
PetscInt iter = 0;
PetscInt bfgsUpdates = 0;
PetscInt needH;
PetscInt i_max = 5;
PetscInt j_max = 1;
PetscInt i, j, N, n, its;
PetscFunctionBegin;
if (tao->XL || tao->XU || tao->ops->computebounds) {
ierr = PetscPrintf(((PetscObject)tao)->comm,"WARNING: Variable bounds have been set but will be ignored by ntr algorithm\n");CHKERRQ(ierr);
}
tao->trust = tao->trust0;
/* Modify the radius if it is too large or small */
tao->trust = PetscMax(tao->trust, tr->min_radius);
tao->trust = PetscMin(tao->trust, tr->max_radius);
if (NTR_PC_BFGS == tr->pc_type && !tr->M) {
ierr = VecGetLocalSize(tao->solution,&n);CHKERRQ(ierr);
ierr = VecGetSize(tao->solution,&N);CHKERRQ(ierr);
ierr = MatCreateLMVM(((PetscObject)tao)->comm,n,N,&tr->M);CHKERRQ(ierr);
ierr = MatLMVMAllocateVectors(tr->M,tao->solution);CHKERRQ(ierr);
}
/* Check convergence criteria */
ierr = TaoComputeObjectiveAndGradient(tao, tao->solution, &f, tao->gradient);CHKERRQ(ierr);
ierr = VecNorm(tao->gradient,NORM_2,&gnorm);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf or NaN");
needH = 1;
ierr = TaoMonitor(tao, iter, f, gnorm, 0.0, 1.0, &reason);CHKERRQ(ierr);
if (reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0);
/* Create vectors for the limited memory preconditioner */
if ((NTR_PC_BFGS == tr->pc_type) &&
(BFGS_SCALE_BFGS != tr->bfgs_scale_type)) {
if (!tr->Diag) {
ierr = VecDuplicate(tao->solution, &tr->Diag);CHKERRQ(ierr);
}
}
switch(tr->ksp_type) {
case NTR_KSP_NASH:
ierr = KSPSetType(tao->ksp, KSPNASH);CHKERRQ(ierr);
if (tao->ksp->ops->setfromoptions) {
(*tao->ksp->ops->setfromoptions)(tao->ksp);
}
break;
case NTR_KSP_STCG:
ierr = KSPSetType(tao->ksp, KSPSTCG);CHKERRQ(ierr);
if (tao->ksp->ops->setfromoptions) {
(*tao->ksp->ops->setfromoptions)(tao->ksp);
}
break;
default:
ierr = KSPSetType(tao->ksp, KSPGLTR);CHKERRQ(ierr);
if (tao->ksp->ops->setfromoptions) {
(*tao->ksp->ops->setfromoptions)(tao->ksp);
}
break;
}
/* Modify the preconditioner to use the bfgs approximation */
ierr = KSPGetPC(tao->ksp, &pc);CHKERRQ(ierr);
switch(tr->pc_type) {
case NTR_PC_NONE:
ierr = PCSetType(pc, PCNONE);CHKERRQ(ierr);
if (pc->ops->setfromoptions) {
(*pc->ops->setfromoptions)(pc);
}
break;
case NTR_PC_AHESS:
ierr = PCSetType(pc, PCJACOBI);CHKERRQ(ierr);
if (pc->ops->setfromoptions) {
(*pc->ops->setfromoptions)(pc);
}
ierr = PCJacobiSetUseAbs(pc);CHKERRQ(ierr);
break;
case NTR_PC_BFGS:
ierr = PCSetType(pc, PCSHELL);CHKERRQ(ierr);
if (pc->ops->setfromoptions) {
(*pc->ops->setfromoptions)(pc);
}
ierr = PCShellSetName(pc, "bfgs");CHKERRQ(ierr);
ierr = PCShellSetContext(pc, tr->M);CHKERRQ(ierr);
ierr = PCShellSetApply(pc, MatLMVMSolveShell);CHKERRQ(ierr);
break;
default:
/* Use the pc method set by pc_type */
break;
}
/* Initialize trust-region radius */
switch(tr->init_type) {
case NTR_INIT_CONSTANT:
/* Use the initial radius specified */
break;
case NTR_INIT_INTERPOLATION:
/* Use the initial radius specified */
max_radius = 0.0;
for (j = 0; j < j_max; ++j) {
fmin = f;
sigma = 0.0;
if (needH) {
ierr = TaoComputeHessian(tao,tao->solution,tao->hessian,tao->hessian_pre);CHKERRQ(ierr);
needH = 0;
}
for (i = 0; i < i_max; ++i) {
ierr = VecCopy(tao->solution, tr->W);CHKERRQ(ierr);
ierr = VecAXPY(tr->W, -tao->trust/gnorm, tao->gradient);CHKERRQ(ierr);
ierr = TaoComputeObjective(tao, tr->W, &ftrial);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(ftrial)) {
tau = tr->gamma1_i;
}
else {
if (ftrial < fmin) {
fmin = ftrial;
sigma = -tao->trust / gnorm;
}
ierr = MatMult(tao->hessian, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = VecDot(tao->gradient, tao->stepdirection, &prered);CHKERRQ(ierr);
prered = tao->trust * (gnorm - 0.5 * tao->trust * prered / (gnorm * gnorm));
actred = f - ftrial;
if ((PetscAbsScalar(actred) <= tr->epsilon) &&
(PetscAbsScalar(prered) <= tr->epsilon)) {
kappa = 1.0;
}
else {
kappa = actred / prered;
}
tau_1 = tr->theta_i * gnorm * tao->trust / (tr->theta_i * gnorm * tao->trust + (1.0 - tr->theta_i) * prered - actred);
tau_2 = tr->theta_i * gnorm * tao->trust / (tr->theta_i * gnorm * tao->trust - (1.0 + tr->theta_i) * prered + actred);
tau_min = PetscMin(tau_1, tau_2);
tau_max = PetscMax(tau_1, tau_2);
if (PetscAbsScalar(kappa - 1.0) <= tr->mu1_i) {
/* Great agreement */
max_radius = PetscMax(max_radius, tao->trust);
if (tau_max < 1.0) {
tau = tr->gamma3_i;
}
else if (tau_max > tr->gamma4_i) {
tau = tr->gamma4_i;
}
else {
tau = tau_max;
}
}
else if (PetscAbsScalar(kappa - 1.0) <= tr->mu2_i) {
/* Good agreement */
max_radius = PetscMax(max_radius, tao->trust);
if (tau_max < tr->gamma2_i) {
tau = tr->gamma2_i;
}
else if (tau_max > tr->gamma3_i) {
tau = tr->gamma3_i;
}
else {
tau = tau_max;
}
}
else {
/* Not good agreement */
if (tau_min > 1.0) {
tau = tr->gamma2_i;
}
else if (tau_max < tr->gamma1_i) {
tau = tr->gamma1_i;
}
else if ((tau_min < tr->gamma1_i) && (tau_max >= 1.0)) {
tau = tr->gamma1_i;
}
else if ((tau_1 >= tr->gamma1_i) && (tau_1 < 1.0) &&
((tau_2 < tr->gamma1_i) || (tau_2 >= 1.0))) {
tau = tau_1;
}
else if ((tau_2 >= tr->gamma1_i) && (tau_2 < 1.0) &&
((tau_1 < tr->gamma1_i) || (tau_2 >= 1.0))) {
tau = tau_2;
}
else {
tau = tau_max;
}
}
}
tao->trust = tau * tao->trust;
}
if (fmin < f) {
f = fmin;
ierr = VecAXPY(tao->solution, sigma, tao->gradient);CHKERRQ(ierr);
ierr = TaoComputeGradient(tao,tao->solution, tao->gradient);CHKERRQ(ierr);
ierr = VecNorm(tao->gradient, NORM_2, &gnorm);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf or NaN");
needH = 1;
ierr = TaoMonitor(tao, iter, f, gnorm, 0.0, 1.0, &reason);CHKERRQ(ierr);
if (reason != TAO_CONTINUE_ITERATING) {
PetscFunctionReturn(0);
}
}
}
tao->trust = PetscMax(tao->trust, max_radius);
/* Modify the radius if it is too large or small */
tao->trust = PetscMax(tao->trust, tr->min_radius);
tao->trust = PetscMin(tao->trust, tr->max_radius);
break;
default:
/* Norm of the first direction will initialize radius */
tao->trust = 0.0;
break;
}
/* Set initial scaling for the BFGS preconditioner
This step is done after computing the initial trust-region radius
since the function value may have decreased */
if (NTR_PC_BFGS == tr->pc_type) {
if (f != 0.0) {
delta = 2.0 * PetscAbsScalar(f) / (gnorm*gnorm);
}
else {
delta = 2.0 / (gnorm*gnorm);
}
ierr = MatLMVMSetDelta(tr->M,delta);CHKERRQ(ierr);
}
/* Have not converged; continue with Newton method */
while (reason == TAO_CONTINUE_ITERATING) {
++iter;
tao->ksp_its=0;
/* Compute the Hessian */
if (needH) {
ierr = TaoComputeHessian(tao,tao->solution,tao->hessian,tao->hessian_pre);CHKERRQ(ierr);
needH = 0;
}
if (NTR_PC_BFGS == tr->pc_type) {
if (BFGS_SCALE_AHESS == tr->bfgs_scale_type) {
/* Obtain diagonal for the bfgs preconditioner */
ierr = MatGetDiagonal(tao->hessian, tr->Diag);CHKERRQ(ierr);
ierr = VecAbs(tr->Diag);CHKERRQ(ierr);
ierr = VecReciprocal(tr->Diag);CHKERRQ(ierr);
ierr = MatLMVMSetScale(tr->M,tr->Diag);CHKERRQ(ierr);
}
/* Update the limited memory preconditioner */
ierr = MatLMVMUpdate(tr->M, tao->solution, tao->gradient);CHKERRQ(ierr);
++bfgsUpdates;
}
while (reason == TAO_CONTINUE_ITERATING) {
ierr = KSPSetOperators(tao->ksp, tao->hessian, tao->hessian_pre);CHKERRQ(ierr);
/* Solve the trust region subproblem */
if (NTR_KSP_NASH == tr->ksp_type) {
ierr = KSPNASHSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
tao->ksp_tot_its+=its;
ierr = KSPNASHGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
} else if (NTR_KSP_STCG == tr->ksp_type) {
ierr = KSPSTCGSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
tao->ksp_tot_its+=its;
ierr = KSPSTCGGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
} else { /* NTR_KSP_GLTR */
ierr = KSPGLTRSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
tao->ksp_tot_its+=its;
ierr = KSPGLTRGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
}
if (0.0 == tao->trust) {
/* Radius was uninitialized; use the norm of the direction */
if (norm_d > 0.0) {
tao->trust = norm_d;
/* Modify the radius if it is too large or small */
tao->trust = PetscMax(tao->trust, tr->min_radius);
tao->trust = PetscMin(tao->trust, tr->max_radius);
}
else {
/* The direction was bad; set radius to default value and re-solve
the trust-region subproblem to get a direction */
tao->trust = tao->trust0;
/* Modify the radius if it is too large or small */
tao->trust = PetscMax(tao->trust, tr->min_radius);
tao->trust = PetscMin(tao->trust, tr->max_radius);
if (NTR_KSP_NASH == tr->ksp_type) {
ierr = KSPNASHSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
tao->ksp_tot_its+=its;
ierr = KSPNASHGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
} else if (NTR_KSP_STCG == tr->ksp_type) {
ierr = KSPSTCGSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
tao->ksp_tot_its+=its;
ierr = KSPSTCGGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
} else { /* NTR_KSP_GLTR */
ierr = KSPGLTRSetRadius(tao->ksp,tao->trust);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, tao->gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
tao->ksp_tot_its+=its;
ierr = KSPGLTRGetNormD(tao->ksp, &norm_d);CHKERRQ(ierr);
}
if (norm_d == 0.0) SETERRQ(PETSC_COMM_SELF,1, "Initial direction zero");
}
}
ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr);
ierr = KSPGetConvergedReason(tao->ksp, &ksp_reason);CHKERRQ(ierr);
if ((KSP_DIVERGED_INDEFINITE_PC == ksp_reason) &&
(NTR_PC_BFGS == tr->pc_type) && (bfgsUpdates > 1)) {
/* Preconditioner is numerically indefinite; reset the
approximate if using BFGS preconditioning. */
if (f != 0.0) {
delta = 2.0 * PetscAbsScalar(f) / (gnorm*gnorm);
}
else {
delta = 2.0 / (gnorm*gnorm);
}
ierr = MatLMVMSetDelta(tr->M, delta);CHKERRQ(ierr);
ierr = MatLMVMReset(tr->M);CHKERRQ(ierr);
ierr = MatLMVMUpdate(tr->M, tao->solution, tao->gradient);CHKERRQ(ierr);
bfgsUpdates = 1;
}
if (NTR_UPDATE_REDUCTION == tr->update_type) {
/* Get predicted reduction */
if (NTR_KSP_NASH == tr->ksp_type) {
ierr = KSPNASHGetObjFcn(tao->ksp,&prered);CHKERRQ(ierr);
} else if (NTR_KSP_STCG == tr->ksp_type) {
ierr = KSPSTCGGetObjFcn(tao->ksp,&prered);CHKERRQ(ierr);
} else { /* gltr */
ierr = KSPGLTRGetObjFcn(tao->ksp,&prered);CHKERRQ(ierr);
}
if (prered >= 0.0) {
/* The predicted reduction has the wrong sign. This cannot
happen in infinite precision arithmetic. Step should
be rejected! */
tao->trust = tr->alpha1 * PetscMin(tao->trust, norm_d);
}
else {
/* Compute trial step and function value */
ierr = VecCopy(tao->solution,tr->W);CHKERRQ(ierr);
ierr = VecAXPY(tr->W, 1.0, tao->stepdirection);CHKERRQ(ierr);
ierr = TaoComputeObjective(tao, tr->W, &ftrial);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(ftrial)) {
tao->trust = tr->alpha1 * PetscMin(tao->trust, norm_d);
} else {
/* Compute and actual reduction */
actred = f - ftrial;
prered = -prered;
if ((PetscAbsScalar(actred) <= tr->epsilon) &&
(PetscAbsScalar(prered) <= tr->epsilon)) {
kappa = 1.0;
}
else {
kappa = actred / prered;
}
/* Accept or reject the step and update radius */
if (kappa < tr->eta1) {
/* Reject the step */
tao->trust = tr->alpha1 * PetscMin(tao->trust, norm_d);
}
else {
/* Accept the step */
if (kappa < tr->eta2) {
/* Marginal bad step */
tao->trust = tr->alpha2 * PetscMin(tao->trust, norm_d);
}
else if (kappa < tr->eta3) {
/* Reasonable step */
tao->trust = tr->alpha3 * tao->trust;
}
else if (kappa < tr->eta4) {
/* Good step */
tao->trust = PetscMax(tr->alpha4 * norm_d, tao->trust);
}
else {
/* Very good step */
tao->trust = PetscMax(tr->alpha5 * norm_d, tao->trust);
}
break;
}
}
}
}
else {
/* Get predicted reduction */
if (NTR_KSP_NASH == tr->ksp_type) {
ierr = KSPNASHGetObjFcn(tao->ksp,&prered);CHKERRQ(ierr);
} else if (NTR_KSP_STCG == tr->ksp_type) {
ierr = KSPSTCGGetObjFcn(tao->ksp,&prered);CHKERRQ(ierr);
} else { /* gltr */
ierr = KSPGLTRGetObjFcn(tao->ksp,&prered);CHKERRQ(ierr);
}
if (prered >= 0.0) {
/* The predicted reduction has the wrong sign. This cannot
happen in infinite precision arithmetic. Step should
be rejected! */
tao->trust = tr->gamma1 * PetscMin(tao->trust, norm_d);
}
else {
ierr = VecCopy(tao->solution, tr->W);CHKERRQ(ierr);
ierr = VecAXPY(tr->W, 1.0, tao->stepdirection);CHKERRQ(ierr);
ierr = TaoComputeObjective(tao, tr->W, &ftrial);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(ftrial)) {
tao->trust = tr->gamma1 * PetscMin(tao->trust, norm_d);
}
else {
ierr = VecDot(tao->gradient, tao->stepdirection, &beta);CHKERRQ(ierr);
actred = f - ftrial;
prered = -prered;
if ((PetscAbsScalar(actred) <= tr->epsilon) &&
(PetscAbsScalar(prered) <= tr->epsilon)) {
kappa = 1.0;
}
else {
kappa = actred / prered;
}
tau_1 = tr->theta * beta / (tr->theta * beta - (1.0 - tr->theta) * prered + actred);
tau_2 = tr->theta * beta / (tr->theta * beta + (1.0 + tr->theta) * prered - actred);
tau_min = PetscMin(tau_1, tau_2);
tau_max = PetscMax(tau_1, tau_2);
if (kappa >= 1.0 - tr->mu1) {
/* Great agreement; accept step and update radius */
if (tau_max < 1.0) {
tao->trust = PetscMax(tao->trust, tr->gamma3 * norm_d);
}
else if (tau_max > tr->gamma4) {
tao->trust = PetscMax(tao->trust, tr->gamma4 * norm_d);
}
else {
tao->trust = PetscMax(tao->trust, tau_max * norm_d);
}
break;
}
else if (kappa >= 1.0 - tr->mu2) {
/* Good agreement */
if (tau_max < tr->gamma2) {
tao->trust = tr->gamma2 * PetscMin(tao->trust, norm_d);
}
else if (tau_max > tr->gamma3) {
tao->trust = PetscMax(tao->trust, tr->gamma3 * norm_d);
}
else if (tau_max < 1.0) {
tao->trust = tau_max * PetscMin(tao->trust, norm_d);
}
else {
tao->trust = PetscMax(tao->trust, tau_max * norm_d);
}
break;
}
else {
/* Not good agreement */
if (tau_min > 1.0) {
tao->trust = tr->gamma2 * PetscMin(tao->trust, norm_d);
}
else if (tau_max < tr->gamma1) {
tao->trust = tr->gamma1 * PetscMin(tao->trust, norm_d);
}
else if ((tau_min < tr->gamma1) && (tau_max >= 1.0)) {
tao->trust = tr->gamma1 * PetscMin(tao->trust, norm_d);
}
else if ((tau_1 >= tr->gamma1) && (tau_1 < 1.0) &&
((tau_2 < tr->gamma1) || (tau_2 >= 1.0))) {
tao->trust = tau_1 * PetscMin(tao->trust, norm_d);
}
else if ((tau_2 >= tr->gamma1) && (tau_2 < 1.0) &&
((tau_1 < tr->gamma1) || (tau_2 >= 1.0))) {
tao->trust = tau_2 * PetscMin(tao->trust, norm_d);
}
else {
tao->trust = tau_max * PetscMin(tao->trust, norm_d);
}
}
}
}
}
/* The step computed was not good and the radius was decreased.
Monitor the radius to terminate. */
ierr = TaoMonitor(tao, iter, f, gnorm, 0.0, tao->trust, &reason);CHKERRQ(ierr);
}
/* The radius may have been increased; modify if it is too large */
tao->trust = PetscMin(tao->trust, tr->max_radius);
if (reason == TAO_CONTINUE_ITERATING) {
ierr = VecCopy(tr->W, tao->solution);CHKERRQ(ierr);
f = ftrial;
ierr = TaoComputeGradient(tao, tao->solution, tao->gradient);
ierr = VecNorm(tao->gradient, NORM_2, &gnorm);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf or NaN");
needH = 1;
ierr = TaoMonitor(tao, iter, f, gnorm, 0.0, tao->trust, &reason);CHKERRQ(ierr);
}
}
PetscFunctionReturn(0);
}
static PetscErrorCode TaoSolve_SQPCON(Tao tao)
{
TAO_SQPCON *sqpconP = (TAO_SQPCON*)tao->data;
PetscInt iter=0;
TaoConvergedReason reason = TAO_CONTINUE_ITERATING;
TaoLineSearchConvergedReason ls_reason = TAOLINESEARCH_CONTINUE_ITERATING;
PetscReal step=1.0,f,fm, fold;
PetscReal cnorm, mnorm;
PetscBool use_update=PETSC_TRUE; /* don't update Q if line search failed */
PetscErrorCode ierr;
PetscFunctionBegin;
/* Scatter to U,V */
ierr = VecScatterBegin(sqpconP->state_scatter, tao->solution, sqpconP->U, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->state_scatter, tao->solution, sqpconP->U, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterBegin(sqpconP->design_scatter, tao->solution, sqpconP->V, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->design_scatter, tao->solution, sqpconP->V, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
/* Evaluate Function, Gradient, Constraints, and Jacobian */
ierr = TaoComputeObjectiveAndGradient(tao,tao->solution,&f,tao->gradient);CHKERRQ(ierr);
ierr = TaoComputeConstraints(tao,tao->solution, tao->constraints);CHKERRQ(ierr);
ierr = TaoComputeJacobianState(tao,tao->solution, &tao->jacobian_state, &tao->jacobian_state_pre, &tao->jacobian_state_inv, &sqpconP->statematflag);CHKERRQ(ierr);
ierr = TaoComputeJacobianDesign(tao,tao->solution, &tao->jacobian_design, &tao->jacobian_design_pre, &sqpconP->statematflag);CHKERRQ(ierr);
/* Scatter gradient to GU,GV */
ierr = VecScatterBegin(sqpconP->state_scatter, tao->gradient, sqpconP->GU, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->state_scatter, tao->gradient, sqpconP->GU, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterBegin(sqpconP->design_scatter, tao->gradient, sqpconP->GV, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->design_scatter, tao->gradient, sqpconP->GV, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecNorm(tao->gradient, NORM_2, &mnorm);CHKERRQ(ierr);
/* Evaluate constraint norm */
ierr = VecNorm(tao->constraints, NORM_2, &cnorm);CHKERRQ(ierr);
/* Monitor convergence */
ierr = TaoMonitor(tao, iter,f,mnorm,cnorm,step,&reason);CHKERRQ(ierr);
while (reason == TAO_CONTINUE_ITERATING) {
/* Solve tbar = -A\t (t is constraints vector) */
ierr = MatMult(tao->jacobian_state_inv, tao->constraints, sqpconP->Tbar);CHKERRQ(ierr);
ierr = VecScale(sqpconP->Tbar, -1.0);CHKERRQ(ierr);
/* aqwac = A'\(Q*Tbar + c) */
if (iter > 0) {
ierr = MatMult(sqpconP->Q,sqpconP->Tbar,sqpconP->WV);CHKERRQ(ierr);
} else {
ierr = VecCopy(sqpconP->Tbar, sqpconP->WV);CHKERRQ(ierr);
}
ierr = VecAXPY(sqpconP->WV,1.0,sqpconP->GU);CHKERRQ(ierr);
ierr = MatMultTranspose(tao->jacobian_state_inv, sqpconP->WV, sqpconP->aqwac);CHKERRQ(ierr);
/* Reduced Gradient dbar = d - B^t * aqwac */
ierr = MatMultTranspose(tao->jacobian_design,sqpconP->aqwac, sqpconP->dbar);CHKERRQ(ierr);
ierr = VecScale(sqpconP->dbar, -1.0);CHKERRQ(ierr);
ierr = VecAXPY(sqpconP->dbar,1.0,sqpconP->GV);CHKERRQ(ierr);
/* update reduced hessian */
ierr = MatLMVMUpdate(sqpconP->R, sqpconP->V, sqpconP->dbar);CHKERRQ(ierr);
/* Solve R*dv = -dbar using approx. hessian */
ierr = MatLMVMSolve(sqpconP->R, sqpconP->dbar, sqpconP->DV);CHKERRQ(ierr);
ierr = VecScale(sqpconP->DV, -1.0);CHKERRQ(ierr);
/* Backsolve for u = A\(g - B*dv) = tbar - A\(B*dv)*/
ierr = MatMult(tao->jacobian_design, sqpconP->DV, sqpconP->WL);CHKERRQ(ierr);
ierr = MatMult(tao->jacobian_state_inv, sqpconP->WL, sqpconP->DU);CHKERRQ(ierr);
ierr = VecScale(sqpconP->DU, -1.0);CHKERRQ(ierr);
ierr = VecAXPY(sqpconP->DU, 1.0, sqpconP->Tbar);CHKERRQ(ierr);
/* Assemble Big D */
ierr = VecScatterBegin(sqpconP->state_scatter, sqpconP->DU, tao->stepdirection, INSERT_VALUES, SCATTER_REVERSE);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->state_scatter, sqpconP->DU, tao->stepdirection, INSERT_VALUES, SCATTER_REVERSE);CHKERRQ(ierr);
ierr = VecScatterBegin(sqpconP->design_scatter, sqpconP->DV, tao->stepdirection, INSERT_VALUES, SCATTER_REVERSE);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->design_scatter, sqpconP->DV, tao->stepdirection, INSERT_VALUES, SCATTER_REVERSE);CHKERRQ(ierr);
/* Perform Line Search */
ierr = VecCopy(tao->solution, sqpconP->Xold);CHKERRQ(ierr);
ierr = VecCopy(tao->gradient, sqpconP->Gold);CHKERRQ(ierr);
fold = f;
ierr = TaoLineSearchComputeObjectiveAndGradient(tao->linesearch,tao->solution,&fm,sqpconP->GL);CHKERRQ(ierr);
ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);
ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &fm, sqpconP->GL, tao->stepdirection,&step, &ls_reason);CHKERRQ(ierr);
ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr);
if (ls_reason < 0) {
ierr = VecCopy(sqpconP->Xold, tao->solution);
ierr = VecCopy(sqpconP->Gold, tao->gradient);
f = fold;
ierr = VecAXPY(tao->solution, 1.0, tao->stepdirection);CHKERRQ(ierr);
ierr = PetscInfo(tao,"Line Search Failed, using full step.");CHKERRQ(ierr);
use_update=PETSC_FALSE;
} else {
use_update = PETSC_TRUE;
}
/* Scatter X to U,V */
ierr = VecScatterBegin(sqpconP->state_scatter, tao->solution, sqpconP->U, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->state_scatter, tao->solution, sqpconP->U, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterBegin(sqpconP->design_scatter, tao->solution, sqpconP->V, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->design_scatter, tao->solution, sqpconP->V, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
/* Evaluate Function, Gradient, Constraints, and Jacobian */
ierr = TaoComputeObjectiveAndGradient(tao,tao->solution,&f,tao->gradient);CHKERRQ(ierr);
ierr = TaoComputeConstraints(tao,tao->solution, tao->constraints);CHKERRQ(ierr);
ierr = TaoComputeJacobianState(tao,tao->solution, &tao->jacobian_state, &tao->jacobian_state_pre, &tao->jacobian_state_inv, &sqpconP->statematflag);CHKERRQ(ierr);
ierr = TaoComputeJacobianDesign(tao,tao->solution, &tao->jacobian_design, &tao->jacobian_design_pre, &sqpconP->designmatflag);CHKERRQ(ierr);
/* Scatter gradient to GU,GV */
ierr = VecScatterBegin(sqpconP->state_scatter, tao->gradient, sqpconP->GU, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->state_scatter, tao->gradient, sqpconP->GU, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterBegin(sqpconP->design_scatter, tao->gradient, sqpconP->GV, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(sqpconP->design_scatter, tao->gradient, sqpconP->GV, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
/* Update approx to hessian of the Lagrangian wrt state (Q)
with u_k+1, gu_k+1 */
if (use_update) {
ierr = MatApproxUpdate(sqpconP->Q,sqpconP->U,sqpconP->GU);CHKERRQ(ierr);
}
ierr = VecNorm(sqpconP->GL, NORM_2, &mnorm);CHKERRQ(ierr);
/* Evaluate constraint norm */
ierr = VecNorm(tao->constraints, NORM_2, &cnorm);CHKERRQ(ierr);
/* Monitor convergence */
iter++;
ierr = TaoMonitor(tao, iter,f,mnorm,cnorm,step,&reason);CHKERRQ(ierr);
}
PetscFunctionReturn(0);
}
static PetscErrorCode TaoSolve_GPCG(Tao tao)
{
TAO_GPCG *gpcg = (TAO_GPCG *)tao->data;
PetscErrorCode ierr;
PetscInt its;
PetscReal actred,f,f_new,gnorm,gdx,stepsize,xtb;
PetscReal xtHx;
TaoConvergedReason reason = TAO_CONTINUE_ITERATING;
TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING;
PetscFunctionBegin;
ierr = TaoComputeVariableBounds(tao);CHKERRQ(ierr);
ierr = VecMedian(tao->XL,tao->solution,tao->XU,tao->solution);CHKERRQ(ierr);
ierr = TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);CHKERRQ(ierr);
/* Using f = .5*x'Hx + x'b + c and g=Hx + b, compute b,c */
ierr = TaoComputeHessian(tao,tao->solution,tao->hessian,tao->hessian_pre);CHKERRQ(ierr);
ierr = TaoComputeObjectiveAndGradient(tao,tao->solution,&f,tao->gradient);CHKERRQ(ierr);
ierr = VecCopy(tao->gradient, gpcg->B);CHKERRQ(ierr);
ierr = MatMult(tao->hessian,tao->solution,gpcg->Work);CHKERRQ(ierr);
ierr = VecDot(gpcg->Work, tao->solution, &xtHx);CHKERRQ(ierr);
ierr = VecAXPY(gpcg->B,-1.0,gpcg->Work);CHKERRQ(ierr);
ierr = VecDot(gpcg->B,tao->solution,&xtb);CHKERRQ(ierr);
gpcg->c=f-xtHx/2.0-xtb;
if (gpcg->Free_Local) {
ierr = ISDestroy(&gpcg->Free_Local);CHKERRQ(ierr);
}
ierr = VecWhichBetween(tao->XL,tao->solution,tao->XU,&gpcg->Free_Local);CHKERRQ(ierr);
/* Project the gradient and calculate the norm */
ierr = VecCopy(tao->gradient,gpcg->G_New);CHKERRQ(ierr);
ierr = VecBoundGradientProjection(tao->gradient,tao->solution,tao->XL,tao->XU,gpcg->PG);CHKERRQ(ierr);
ierr = VecNorm(gpcg->PG,NORM_2,&gpcg->gnorm);CHKERRQ(ierr);
tao->step=1.0;
gpcg->f = f;
/* Check Stopping Condition */
ierr=TaoMonitor(tao,tao->niter,f,gpcg->gnorm,0.0,tao->step,&reason);CHKERRQ(ierr);
while (reason == TAO_CONTINUE_ITERATING){
tao->ksp_its=0;
ierr = GPCGGradProjections(tao);CHKERRQ(ierr);
ierr = ISGetSize(gpcg->Free_Local,&gpcg->n_free);CHKERRQ(ierr);
f=gpcg->f; gnorm=gpcg->gnorm;
ierr = KSPReset(tao->ksp);CHKERRQ(ierr);
if (gpcg->n_free > 0){
/* Create a reduced linear system */
ierr = VecDestroy(&gpcg->R);CHKERRQ(ierr);
ierr = VecDestroy(&gpcg->DXFree);CHKERRQ(ierr);
ierr = TaoVecGetSubVec(tao->gradient,gpcg->Free_Local, tao->subset_type, 0.0, &gpcg->R);CHKERRQ(ierr);
ierr = VecScale(gpcg->R, -1.0);CHKERRQ(ierr);
ierr = TaoVecGetSubVec(tao->stepdirection,gpcg->Free_Local,tao->subset_type, 0.0, &gpcg->DXFree);CHKERRQ(ierr);
ierr = VecSet(gpcg->DXFree,0.0);CHKERRQ(ierr);
ierr = TaoMatGetSubMat(tao->hessian, gpcg->Free_Local, gpcg->Work, tao->subset_type, &gpcg->Hsub);CHKERRQ(ierr);
if (tao->hessian_pre == tao->hessian) {
ierr = MatDestroy(&gpcg->Hsub_pre);CHKERRQ(ierr);
ierr = PetscObjectReference((PetscObject)gpcg->Hsub);CHKERRQ(ierr);
gpcg->Hsub_pre = gpcg->Hsub;
} else {
ierr = TaoMatGetSubMat(tao->hessian, gpcg->Free_Local, gpcg->Work, tao->subset_type, &gpcg->Hsub_pre);CHKERRQ(ierr);
}
ierr = KSPReset(tao->ksp);CHKERRQ(ierr);
ierr = KSPSetOperators(tao->ksp,gpcg->Hsub,gpcg->Hsub_pre);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp,gpcg->R,gpcg->DXFree);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
tao->ksp_tot_its+=its;
ierr = VecSet(tao->stepdirection,0.0);CHKERRQ(ierr);
ierr = VecISAXPY(tao->stepdirection,gpcg->Free_Local,1.0,gpcg->DXFree);CHKERRQ(ierr);
ierr = VecDot(tao->stepdirection,tao->gradient,&gdx);CHKERRQ(ierr);
ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr);
f_new=f;
ierr = TaoLineSearchApply(tao->linesearch,tao->solution,&f_new,tao->gradient,tao->stepdirection,&stepsize,&ls_status);CHKERRQ(ierr);
actred = f_new - f;
/* Evaluate the function and gradient at the new point */
ierr = VecBoundGradientProjection(tao->gradient,tao->solution,tao->XL,tao->XU, gpcg->PG);CHKERRQ(ierr);
ierr = VecNorm(gpcg->PG, NORM_2, &gnorm);CHKERRQ(ierr);
f=f_new;
ierr = ISDestroy(&gpcg->Free_Local);CHKERRQ(ierr);
ierr = VecWhichBetween(tao->XL,tao->solution,tao->XU,&gpcg->Free_Local);CHKERRQ(ierr);
} else {
actred = 0; gpcg->step=1.0;
/* if there were no free variables, no cg method */
}
tao->niter++;
ierr = TaoMonitor(tao,tao->niter,f,gnorm,0.0,gpcg->step,&reason);CHKERRQ(ierr);
gpcg->f=f;gpcg->gnorm=gnorm; gpcg->actred=actred;
if (reason!=TAO_CONTINUE_ITERATING) break;
} /* END MAIN LOOP */
PetscFunctionReturn(0);
}
static PetscErrorCode TaoSolve_BQPIP(Tao tao)
{
TAO_BQPIP *qp = (TAO_BQPIP*)tao->data;
PetscErrorCode ierr;
PetscInt iter=0,its;
PetscReal d1,d2,ksptol,sigma;
PetscReal sigmamu;
PetscReal dstep,pstep,step=0;
PetscReal gap[4];
TaoConvergedReason reason;
PetscFunctionBegin;
qp->dobj = 0.0;
qp->pobj = 1.0;
qp->gap = 10.0;
qp->rgap = 1.0;
qp->mu = 1.0;
qp->sigma = 1.0;
qp->dinfeas = 1.0;
qp->psteplength = 0.0;
qp->dsteplength = 0.0;
/* Tighten infinite bounds, things break when we don't do this
-- see test_bqpip.c
*/
ierr = VecSet(qp->XU,1.0e20);CHKERRQ(ierr);
ierr = VecSet(qp->XL,-1.0e20);CHKERRQ(ierr);
ierr = VecPointwiseMax(qp->XL,qp->XL,tao->XL);CHKERRQ(ierr);
ierr = VecPointwiseMin(qp->XU,qp->XU,tao->XU);CHKERRQ(ierr);
ierr = TaoComputeObjectiveAndGradient(tao,tao->solution,&qp->c,qp->C0);CHKERRQ(ierr);
ierr = TaoComputeHessian(tao,tao->solution,tao->hessian,tao->hessian_pre);CHKERRQ(ierr);
ierr = MatMult(tao->hessian, tao->solution, qp->Work);CHKERRQ(ierr);
ierr = VecDot(tao->solution, qp->Work, &d1);CHKERRQ(ierr);
ierr = VecAXPY(qp->C0, -1.0, qp->Work);CHKERRQ(ierr);
ierr = VecDot(qp->C0, tao->solution, &d2);CHKERRQ(ierr);
qp->c -= (d1/2.0+d2);
ierr = MatGetDiagonal(tao->hessian, qp->HDiag);CHKERRQ(ierr);
ierr = QPIPSetInitialPoint(qp,tao);CHKERRQ(ierr);
ierr = QPIPComputeResidual(qp,tao);CHKERRQ(ierr);
/* Enter main loop */
while (1){
/* Check Stopping Condition */
ierr = TaoMonitor(tao,iter++,qp->pobj,PetscSqrtScalar(qp->gap + qp->dinfeas),
qp->pinfeas, step, &reason);CHKERRQ(ierr);
if (reason != TAO_CONTINUE_ITERATING) break;
/*
Dual Infeasibility Direction should already be in the right
hand side from computing the residuals
*/
ierr = QPIPComputeNormFromCentralPath(qp,&d1);CHKERRQ(ierr);
if (iter > 0 && (qp->rnorm>5*qp->mu || d1*d1>qp->m*qp->mu*qp->mu) ) {
sigma=1.0;sigmamu=qp->mu;
sigma=0.0;sigmamu=0;
} else {
sigma=0.0;sigmamu=0;
}
ierr = VecSet(qp->DZ, sigmamu);CHKERRQ(ierr);
ierr = VecSet(qp->DS, sigmamu);CHKERRQ(ierr);
if (sigmamu !=0){
ierr = VecPointwiseDivide(qp->DZ, qp->DZ, qp->G);CHKERRQ(ierr);
ierr = VecPointwiseDivide(qp->DS, qp->DS, qp->T);CHKERRQ(ierr);
ierr = VecCopy(qp->DZ,qp->RHS2);CHKERRQ(ierr);
ierr = VecAXPY(qp->RHS2, 1.0, qp->DS);CHKERRQ(ierr);
} else {
ierr = VecZeroEntries(qp->RHS2);CHKERRQ(ierr);
}
/*
Compute the Primal Infeasiblitiy RHS and the
Diagonal Matrix to be added to H and store in Work
*/
ierr = VecPointwiseDivide(qp->DiagAxpy, qp->Z, qp->G);CHKERRQ(ierr);
ierr = VecPointwiseMult(qp->GZwork, qp->DiagAxpy, qp->R3);CHKERRQ(ierr);
ierr = VecAXPY(qp->RHS, -1.0, qp->GZwork);CHKERRQ(ierr);
ierr = VecPointwiseDivide(qp->TSwork, qp->S, qp->T);CHKERRQ(ierr);
ierr = VecAXPY(qp->DiagAxpy, 1.0, qp->TSwork);CHKERRQ(ierr);
ierr = VecPointwiseMult(qp->TSwork, qp->TSwork, qp->R5);CHKERRQ(ierr);
ierr = VecAXPY(qp->RHS, -1.0, qp->TSwork);CHKERRQ(ierr);
ierr = VecAXPY(qp->RHS2, 1.0, qp->RHS);CHKERRQ(ierr);
/* Determine the solving tolerance */
ksptol = qp->mu/10.0;
ksptol = PetscMin(ksptol,0.001);
ierr = MatDiagonalSet(tao->hessian, qp->DiagAxpy, ADD_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(tao->hessian,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(tao->hessian,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = KSPSetOperators(tao->ksp, tao->hessian, tao->hessian_pre);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, qp->RHS, tao->stepdirection);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
ierr = VecScale(qp->DiagAxpy, -1.0);CHKERRQ(ierr);
ierr = MatDiagonalSet(tao->hessian, qp->DiagAxpy, ADD_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(tao->hessian,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(tao->hessian,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = VecScale(qp->DiagAxpy, -1.0);CHKERRQ(ierr);
ierr = QPComputeStepDirection(qp,tao);CHKERRQ(ierr);
ierr = QPStepLength(qp); CHKERRQ(ierr);
/* Calculate New Residual R1 in Work vector */
ierr = MatMult(tao->hessian, tao->stepdirection, qp->RHS2);CHKERRQ(ierr);
ierr = VecAXPY(qp->RHS2, 1.0, qp->DS);CHKERRQ(ierr);
ierr = VecAXPY(qp->RHS2, -1.0, qp->DZ);CHKERRQ(ierr);
ierr = VecAYPX(qp->RHS2, qp->dsteplength, tao->gradient);CHKERRQ(ierr);
ierr = VecNorm(qp->RHS2, NORM_2, &qp->dinfeas);CHKERRQ(ierr);
ierr = VecDot(qp->DZ, qp->DG, gap);CHKERRQ(ierr);
ierr = VecDot(qp->DS, qp->DT, gap+1);CHKERRQ(ierr);
qp->rnorm=(qp->dinfeas+qp->psteplength*qp->pinfeas)/(qp->m+qp->n);
pstep = qp->psteplength; dstep = qp->dsteplength;
step = PetscMin(qp->psteplength,qp->dsteplength);
sigmamu= ( pstep*pstep*(gap[0]+gap[1]) +
(1 - pstep + pstep*sigma)*qp->gap )/qp->m;
if (qp->predcorr && step < 0.9){
if (sigmamu < qp->mu){
sigmamu=sigmamu/qp->mu;
sigmamu=sigmamu*sigmamu*sigmamu;
} else {sigmamu = 1.0;}
sigmamu = sigmamu*qp->mu;
/* Compute Corrector Step */
ierr = VecPointwiseMult(qp->DZ, qp->DG, qp->DZ);CHKERRQ(ierr);
ierr = VecScale(qp->DZ, -1.0);CHKERRQ(ierr);
ierr = VecShift(qp->DZ, sigmamu);CHKERRQ(ierr);
ierr = VecPointwiseDivide(qp->DZ, qp->DZ, qp->G);CHKERRQ(ierr);
ierr = VecPointwiseMult(qp->DS, qp->DS, qp->DT);CHKERRQ(ierr);
ierr = VecScale(qp->DS, -1.0);CHKERRQ(ierr);
ierr = VecShift(qp->DS, sigmamu);CHKERRQ(ierr);
ierr = VecPointwiseDivide(qp->DS, qp->DS, qp->T);CHKERRQ(ierr);
ierr = VecCopy(qp->DZ, qp->RHS2);CHKERRQ(ierr);
ierr = VecAXPY(qp->RHS2, -1.0, qp->DS);CHKERRQ(ierr);
ierr = VecAXPY(qp->RHS2, 1.0, qp->RHS);CHKERRQ(ierr);
/* Approximately solve the linear system */
ierr = MatDiagonalSet(tao->hessian, qp->DiagAxpy, ADD_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(tao->hessian,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(tao->hessian,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = KSPSolve(tao->ksp, qp->RHS2, tao->stepdirection);CHKERRQ(ierr);
ierr = KSPGetIterationNumber(tao->ksp,&its);CHKERRQ(ierr);
tao->ksp_its+=its;
ierr = MatDiagonalSet(tao->hessian, qp->HDiag, INSERT_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(tao->hessian,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(tao->hessian,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = QPComputeStepDirection(qp,tao);CHKERRQ(ierr);
ierr = QPStepLength(qp);CHKERRQ(ierr);
} /* End Corrector step */
/* Take the step */
pstep = qp->psteplength; dstep = qp->dsteplength;
ierr = VecAXPY(qp->Z, dstep, qp->DZ);CHKERRQ(ierr);
ierr = VecAXPY(qp->S, dstep, qp->DS);CHKERRQ(ierr);
ierr = VecAXPY(tao->solution, dstep, tao->stepdirection);CHKERRQ(ierr);
ierr = VecAXPY(qp->G, dstep, qp->DG);CHKERRQ(ierr);
ierr = VecAXPY(qp->T, dstep, qp->DT);CHKERRQ(ierr);
/* Compute Residuals */
ierr = QPIPComputeResidual(qp,tao);CHKERRQ(ierr);
/* Evaluate quadratic function */
ierr = MatMult(tao->hessian, tao->solution, qp->Work);CHKERRQ(ierr);
ierr = VecDot(tao->solution, qp->Work, &d1);CHKERRQ(ierr);
ierr = VecDot(tao->solution, qp->C0, &d2);CHKERRQ(ierr);
ierr = VecDot(qp->G, qp->Z, gap);CHKERRQ(ierr);
ierr = VecDot(qp->T, qp->S, gap+1);CHKERRQ(ierr);
qp->pobj=d1/2.0 + d2+qp->c;
/* Compute the duality gap */
qp->gap = (gap[0]+gap[1]);
qp->dobj = qp->pobj - qp->gap;
if (qp->m>0) qp->mu=qp->gap/(qp->m);
qp->rgap=qp->gap/( PetscAbsReal(qp->dobj) + PetscAbsReal(qp->pobj) + 1.0 );
} /* END MAIN LOOP */
PetscFunctionReturn(0);
}
static PetscErrorCode TaoSolve_BLMVM(Tao tao)
{
PetscErrorCode ierr;
TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data;
TaoConvergedReason reason = TAO_CONTINUE_ITERATING;
TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING;
PetscReal f, fold, gdx, gnorm;
PetscReal stepsize = 1.0,delta;
PetscFunctionBegin;
/* Project initial point onto bounds */
ierr = TaoComputeVariableBounds(tao);CHKERRQ(ierr);
ierr = VecMedian(tao->XL,tao->solution,tao->XU,tao->solution);CHKERRQ(ierr);
ierr = TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);CHKERRQ(ierr);
/* Check convergence criteria */
ierr = TaoComputeObjectiveAndGradient(tao, tao->solution,&f,blmP->unprojected_gradient);CHKERRQ(ierr);
ierr = VecBoundGradientProjection(blmP->unprojected_gradient,tao->solution, tao->XL,tao->XU,tao->gradient);CHKERRQ(ierr);
ierr = TaoGradientNorm(tao, tao->gradient,NORM_2,&gnorm);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf pr NaN");
ierr = TaoMonitor(tao, tao->niter, f, gnorm, 0.0, stepsize, &reason);CHKERRQ(ierr);
if (reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0);
/* Set initial scaling for the function */
if (f != 0.0) {
delta = 2.0*PetscAbsScalar(f) / (gnorm*gnorm);
} else {
delta = 2.0 / (gnorm*gnorm);
}
ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr);
/* Set counter for gradient/reset steps */
blmP->grad = 0;
blmP->reset = 0;
/* Have not converged; continue with Newton method */
while (reason == TAO_CONTINUE_ITERATING) {
/* Compute direction */
ierr = MatLMVMUpdate(blmP->M, tao->solution, tao->gradient);CHKERRQ(ierr);
ierr = MatLMVMSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = VecBoundGradientProjection(tao->stepdirection,tao->solution,tao->XL,tao->XU,tao->gradient);CHKERRQ(ierr);
/* Check for success (descent direction) */
ierr = VecDot(blmP->unprojected_gradient, tao->gradient, &gdx);CHKERRQ(ierr);
if (gdx <= 0) {
/* Step is not descent or solve was not successful
Use steepest descent direction (scaled) */
++blmP->grad;
if (f != 0.0) {
delta = 2.0*PetscAbsScalar(f) / (gnorm*gnorm);
} else {
delta = 2.0 / (gnorm*gnorm);
}
ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr);
ierr = MatLMVMReset(blmP->M);CHKERRQ(ierr);
ierr = MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);CHKERRQ(ierr);
ierr = MatLMVMSolve(blmP->M,blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr);
}
ierr = VecScale(tao->stepdirection,-1.0);CHKERRQ(ierr);
/* Perform the linesearch */
fold = f;
ierr = VecCopy(tao->solution, blmP->Xold);CHKERRQ(ierr);
ierr = VecCopy(blmP->unprojected_gradient, blmP->Gold);CHKERRQ(ierr);
ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr);
ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);CHKERRQ(ierr);
ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr);
if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) {
/* Linesearch failed
Reset factors and use scaled (projected) gradient step */
++blmP->reset;
f = fold;
ierr = VecCopy(blmP->Xold, tao->solution);CHKERRQ(ierr);
ierr = VecCopy(blmP->Gold, blmP->unprojected_gradient);CHKERRQ(ierr);
if (f != 0.0) {
delta = 2.0* PetscAbsScalar(f) / (gnorm*gnorm);
} else {
delta = 2.0/ (gnorm*gnorm);
}
ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr);
ierr = MatLMVMReset(blmP->M);CHKERRQ(ierr);
ierr = MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);CHKERRQ(ierr);
ierr = MatLMVMSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr);
ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr);
/* This may be incorrect; linesearch has values fo stepmax and stepmin
that should be reset. */
ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr);
ierr = TaoLineSearchApply(tao->linesearch,tao->solution,&f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);CHKERRQ(ierr);
ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr);
if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) {
tao->reason = TAO_DIVERGED_LS_FAILURE;
break;
}
}
/* Check for converged */
ierr = VecBoundGradientProjection(blmP->unprojected_gradient, tao->solution, tao->XL, tao->XU, tao->gradient);CHKERRQ(ierr);
ierr = TaoGradientNorm(tao, tao->gradient, NORM_2, &gnorm);CHKERRQ(ierr);
if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Not-a-Number");
tao->niter++;
ierr = TaoMonitor(tao, tao->niter, f, gnorm, 0.0, stepsize, &reason);CHKERRQ(ierr);
}
PetscFunctionReturn(0);
}
static PetscErrorCode TaoSolve_BMRM(Tao tao)
{
PetscErrorCode ierr;
TAO_DF df;
TAO_BMRM *bmrm = (TAO_BMRM*)tao->data;
/* Values and pointers to parts of the optimization problem */
PetscReal f = 0.0;
Vec W = tao->solution;
Vec G = tao->gradient;
PetscReal lambda;
PetscReal bt;
Vec_Chain grad_list, *tail_glist, *pgrad;
PetscInt i;
PetscMPIInt rank;
/* Used in converged criteria check */
PetscReal reg;
PetscReal jtwt = 0.0, max_jtwt, pre_epsilon, epsilon, jw, min_jw;
PetscReal innerSolverTol;
MPI_Comm comm;
PetscFunctionBegin;
ierr = PetscObjectGetComm((PetscObject)tao,&comm);CHKERRQ(ierr);
ierr = MPI_Comm_rank(comm, &rank);CHKERRQ(ierr);
lambda = bmrm->lambda;
/* Check Stopping Condition */
tao->step = 1.0;
max_jtwt = -BMRM_INFTY;
min_jw = BMRM_INFTY;
innerSolverTol = 1.0;
epsilon = 0.0;
if (!rank) {
ierr = init_df_solver(&df);CHKERRQ(ierr);
grad_list.next = NULL;
tail_glist = &grad_list;
}
df.tol = 1e-6;
tao->reason = TAO_CONTINUE_ITERATING;
/*-----------------Algorithm Begins------------------------*/
/* make the scatter */
ierr = VecScatterCreateToZero(W, &bmrm->scatter, &bmrm->local_w);CHKERRQ(ierr);
ierr = VecAssemblyBegin(bmrm->local_w);CHKERRQ(ierr);
ierr = VecAssemblyEnd(bmrm->local_w);CHKERRQ(ierr);
/* NOTE: In application pass the sub-gradient of Remp(W) */
ierr = TaoComputeObjectiveAndGradient(tao, W, &f, G);CHKERRQ(ierr);
ierr = TaoLogConvergenceHistory(tao,f,1.0,0.0,tao->ksp_its);CHKERRQ(ierr);
ierr = TaoMonitor(tao,tao->niter,f,1.0,0.0,tao->step);CHKERRQ(ierr);
ierr = (*tao->ops->convergencetest)(tao,tao->cnvP);CHKERRQ(ierr);
while (tao->reason == TAO_CONTINUE_ITERATING) {
/* Call general purpose update function */
if (tao->ops->update) {
ierr = (*tao->ops->update)(tao, tao->niter);CHKERRQ(ierr);
}
/* compute bt = Remp(Wt-1) - <Wt-1, At> */
ierr = VecDot(W, G, &bt);CHKERRQ(ierr);
bt = f - bt;
/* First gather the gradient to the master node */
ierr = VecScatterBegin(bmrm->scatter, G, bmrm->local_w, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(bmrm->scatter, G, bmrm->local_w, INSERT_VALUES, SCATTER_FORWARD);CHKERRQ(ierr);
/* Bring up the inner solver */
if (!rank) {
ierr = ensure_df_space(tao->niter+1, &df);CHKERRQ(ierr);
ierr = make_grad_node(bmrm->local_w, &pgrad);CHKERRQ(ierr);
tail_glist->next = pgrad;
tail_glist = pgrad;
df.a[tao->niter] = 1.0;
df.f[tao->niter] = -bt;
df.u[tao->niter] = 1.0;
df.l[tao->niter] = 0.0;
/* set up the Q */
pgrad = grad_list.next;
for (i=0; i<=tao->niter; i++) {
if (!pgrad) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Assert that there are at least tao->niter+1 pgrad available");
ierr = VecDot(pgrad->V, bmrm->local_w, ®);CHKERRQ(ierr);
df.Q[i][tao->niter] = df.Q[tao->niter][i] = reg / lambda;
pgrad = pgrad->next;
}
if (tao->niter > 0) {
df.x[tao->niter] = 0.0;
ierr = solve(&df);CHKERRQ(ierr);
} else
df.x[0] = 1.0;
/* now computing Jt*(alpha_t) which should be = Jt(wt) to check convergence */
jtwt = 0.0;
ierr = VecSet(bmrm->local_w, 0.0);CHKERRQ(ierr);
pgrad = grad_list.next;
for (i=0; i<=tao->niter; i++) {
jtwt -= df.x[i] * df.f[i];
ierr = VecAXPY(bmrm->local_w, -df.x[i] / lambda, pgrad->V);CHKERRQ(ierr);
pgrad = pgrad->next;
}
ierr = VecNorm(bmrm->local_w, NORM_2, ®);CHKERRQ(ierr);
reg = 0.5*lambda*reg*reg;
jtwt -= reg;
} /* end if rank == 0 */
/* scatter the new W to all nodes */
ierr = VecScatterBegin(bmrm->scatter,bmrm->local_w,W,INSERT_VALUES,SCATTER_REVERSE);CHKERRQ(ierr);
ierr = VecScatterEnd(bmrm->scatter,bmrm->local_w,W,INSERT_VALUES,SCATTER_REVERSE);CHKERRQ(ierr);
ierr = TaoComputeObjectiveAndGradient(tao, W, &f, G);CHKERRQ(ierr);
ierr = MPI_Bcast(&jtwt,1,MPIU_REAL,0,comm);CHKERRQ(ierr);
ierr = MPI_Bcast(®,1,MPIU_REAL,0,comm);CHKERRQ(ierr);
jw = reg + f; /* J(w) = regularizer + Remp(w) */
if (jw < min_jw) min_jw = jw;
if (jtwt > max_jtwt) max_jtwt = jtwt;
pre_epsilon = epsilon;
epsilon = min_jw - jtwt;
if (!rank) {
if (innerSolverTol > epsilon) innerSolverTol = epsilon;
else if (innerSolverTol < 1e-7) innerSolverTol = 1e-7;
/* if the annealing doesn't work well, lower the inner solver tolerance */
if(pre_epsilon < epsilon) innerSolverTol *= 0.2;
df.tol = innerSolverTol*0.5;
}
tao->niter++;
ierr = TaoLogConvergenceHistory(tao,min_jw,epsilon,0.0,tao->ksp_its);CHKERRQ(ierr);
ierr = TaoMonitor(tao,tao->niter,min_jw,epsilon,0.0,tao->step);CHKERRQ(ierr);
ierr = (*tao->ops->convergencetest)(tao,tao->cnvP);CHKERRQ(ierr);
}
/* free all the memory */
if (!rank) {
ierr = destroy_grad_list(&grad_list);CHKERRQ(ierr);
ierr = destroy_df_solver(&df);CHKERRQ(ierr);
}
ierr = VecDestroy(&bmrm->local_w);CHKERRQ(ierr);
ierr = VecScatterDestroy(&bmrm->scatter);CHKERRQ(ierr);
PetscFunctionReturn(0);
}
