ipopt-src 0.2.3+3.14.16

// Copyright (C) 2004, 2010 International Business Machines and others.
// All Rights Reserved.
// This code is published under the Eclipse Public License.
//
// Authors:  Carl Laird, Andreas Waechter     IBM    2004-08-13

#include "IpIpoptData.hpp"
#include "IpIpoptNLP.hpp"

namespace Ipopt
{

IpoptData::IpoptData(
   SmartPtr<IpoptAdditionalData> add_data /*= NULL*/
)
   : add_data_(add_data)
{ }

IpoptData::~IpoptData()
{ }

void IpoptData::RegisterOptions(
   const SmartPtr<RegisteredOptions>& roptions
)
{
   roptions->SetRegisteringCategory("Termination");
   roptions->AddLowerBoundedNumberOption(
      "tol",
      "Desired convergence tolerance (relative).",
      0., true,
#ifdef IPOPT_SINGLE
      1e-5,
#else
      1e-8,
#endif
      "Determines the convergence tolerance for the algorithm. "
      "The algorithm terminates successfully, if the (scaled) NLP error becomes smaller than this value, and "
      "if the (absolute) criteria according to \"dual_inf_tol\", \"constr_viol_tol\", and \"compl_inf_tol\" are met. "
      "This is epsilon_tol in Eqn. (6) in implementation paper. "
      "See also \"acceptable_tol\" as a second termination criterion. "
      "Note, some other algorithmic features also use this quantity to determine thresholds etc.");
}

bool IpoptData::Initialize(
   const Journalist&  jnlst,
   const OptionsList& options,
   const std::string& prefix
)
{
#if 0
   // I (AW) took the following heuristic out again, since it seemed
   // that the restoration phase tolerance became too tight by
   // default.  I originally probably put it in to avoid that a claim
   // of infeasibility is made prematurely...  let's see if someone
   // starts screaming...
   if (prefix == "resto.")
   {
      // The default for the restoration phase is 1e-2 time the value
      // for the regular algorithm
      if (!options.GetNumericValue("resto.tol", tol_, ""))
      {
         options.GetNumericValue("tol", tol_, prefix);
         tol_ *= 1e-2;
      }
   }
   else
   {
      options.GetNumericValue("tol", tol_, prefix);
   }
#else
   options.GetNumericValue("tol", tol_, prefix);
#endif

   iter_count_ = 0;
   curr_mu_ = -1.;
   mu_initialized_ = false;
   curr_tau_ = -1.;
   tau_initialized_ = false;
   have_prototypes_ = false;
   have_deltas_ = false;
   have_affine_deltas_ = false;

   free_mu_mode_ = false;
   tiny_step_flag_ = false;

   info_ls_count_ = 0;
   ResetInfo();
   info_last_output_ = -1.;
   info_iters_since_header_ = 1000; // need to be larger 10

   initialize_called_ = true;

   bool retval = true;

   if( IsValid(add_data_) )
   {
      retval = add_data_->Initialize(jnlst, options, prefix);
   }

   return retval;
}

bool IpoptData::InitializeDataStructures(
   IpoptNLP& ip_nlp,
   bool      want_x,
   bool      want_y_c,
   bool      want_y_d,
   bool      want_z_L,
   bool      want_z_U
)
{
   DBG_ASSERT(initialize_called_);
   /*
    * Allocate space for all the required linear algebra
    * structures
    */

   SmartPtr<Vector> new_x;
   SmartPtr<Vector> new_s;
   SmartPtr<Vector> new_y_c;
   SmartPtr<Vector> new_y_d;
   SmartPtr<Vector> new_z_L;
   SmartPtr<Vector> new_z_U;
   SmartPtr<Vector> new_v_L;
   SmartPtr<Vector> new_v_U;

   // clear curr_ from previous optimize, if any, in case ip_nlp.InitializeStructures() fails,
   // e.g., because OrigIpoptNLP throws a too-few-degrees-of-freedom exception
   curr_ = NULL;
   iterates_space_ = NULL;

   // Get the required linear algebra structures from the model
   bool retValue = ip_nlp.InitializeStructures(new_x, want_x, new_y_c, want_y_c, new_y_d, want_y_d, new_z_L, want_z_L,
                   new_z_U, want_z_U, new_v_L, new_v_U);
   if( !retValue )
   {
      return false;
   }

   new_s = new_y_d->MakeNew(); // same dimension as d

   iterates_space_ = new IteratesVectorSpace(*(new_x->OwnerSpace()), *(new_s->OwnerSpace()), *(new_y_c->OwnerSpace()),
      *(new_y_d->OwnerSpace()), *(new_z_L->OwnerSpace()), *(new_z_U->OwnerSpace()), *(new_v_L->OwnerSpace()),
      *(new_v_U->OwnerSpace()));

   curr_ = iterates_space_->MakeNewIteratesVector(*new_x, *new_s, *new_y_c, *new_y_d, *new_z_L, *new_z_U, *new_v_L,
           *new_v_U);
#if IPOPT_CHECKLEVEL > 0

   debug_curr_tag_ = curr_->GetTag();
   debug_curr_tag_sum_ = curr_->GetTagSum();
   debug_trial_tag_ = 0;
   debug_trial_tag_sum_ = 0;
   debug_delta_tag_ = 0;
   debug_delta_tag_sum_ = 0;
   debug_delta_aff_tag_ = 0;
   debug_delta_aff_tag_sum_ = 0;
#endif

   trial_ = NULL;

   // Set the pointers for storing steps to NULL
   delta_ = NULL;
   delta_aff_ = NULL;

   have_prototypes_ = true;
   have_deltas_ = false;
   have_affine_deltas_ = false;

   bool retval = true;
   if( IsValid(add_data_) )
   {
      retval = add_data_->InitializeDataStructures();
   }

   return retval;
}

void IpoptData::SetTrialPrimalVariablesFromStep(
   Number        alpha,
   const Vector& delta_x,
   const Vector& delta_s
)
{
   DBG_ASSERT(have_prototypes_);

   if( IsNull(trial_) )
   {
      trial_ = iterates_space_->MakeNewIteratesVector(false);
   }

   SmartPtr<IteratesVector> newvec = trial_->MakeNewContainer();
   newvec->create_new_x();
   newvec->x_NonConst()->AddTwoVectors(1., *curr_->x(), alpha, delta_x, 0.);

   newvec->create_new_s();
   newvec->s_NonConst()->AddTwoVectors(1., *curr_->s(), alpha, delta_s, 0.);

   set_trial(newvec);
}

void IpoptData::SetTrialEqMultipliersFromStep(
   Number        alpha,
   const Vector& delta_y_c,
   const Vector& delta_y_d
)
{
   DBG_ASSERT(have_prototypes_);

   SmartPtr<IteratesVector> newvec = trial()->MakeNewContainer();
   newvec->create_new_y_c();
   newvec->y_c_NonConst()->AddTwoVectors(1., *curr()->y_c(), alpha, delta_y_c, 0.);

   newvec->create_new_y_d();
   newvec->y_d_NonConst()->AddTwoVectors(1., *curr()->y_d(), alpha, delta_y_d, 0.);

   set_trial(newvec);
}

void IpoptData::SetTrialBoundMultipliersFromStep(
   Number        alpha,
   const Vector& delta_z_L,
   const Vector& delta_z_U,
   const Vector& delta_v_L,
   const Vector& delta_v_U
)
{
   DBG_ASSERT(have_prototypes_);

   SmartPtr<IteratesVector> newvec = trial()->MakeNewContainer();
   newvec->create_new_z_L();
   newvec->z_L_NonConst()->AddTwoVectors(1., *curr()->z_L(), alpha, delta_z_L, 0.);

   newvec->create_new_z_U();
   newvec->z_U_NonConst()->AddTwoVectors(1., *curr()->z_U(), alpha, delta_z_U, 0.);

   newvec->create_new_v_L();
   newvec->v_L_NonConst()->AddTwoVectors(1., *curr()->v_L(), alpha, delta_v_L, 0.);

   newvec->create_new_v_U();
   newvec->v_U_NonConst()->AddTwoVectors(1., *curr()->v_U(), alpha, delta_v_U, 0.);

   set_trial(newvec);
}

void IpoptData::AcceptTrialPoint()
{
   DBG_ASSERT(IsValid(trial_));
   DBG_ASSERT(IsValid(trial_->x()));
   DBG_ASSERT(IsValid(trial_->s()));
   DBG_ASSERT(IsValid(trial_->y_c()));
   DBG_ASSERT(IsValid(trial_->y_d()));
   DBG_ASSERT(IsValid(trial_->z_L()));
   DBG_ASSERT(IsValid(trial_->z_U()));
   DBG_ASSERT(IsValid(trial_->v_L()));
   DBG_ASSERT(IsValid(trial_->v_U()));

   CopyTrialToCurrent();

   // Set trial pointers to Null (frees memory unless someone else is
   // still referring to it, and it makes sure that indeed all trial
   // values are set before a new trial point is accepted)
   trial_ = NULL;

   // Free the memory for the affine-scaling step
   delta_aff_ = NULL;

   have_deltas_ = false;
   have_affine_deltas_ = false;

   if( IsValid(add_data_) )
   {
      add_data_->AcceptTrialPoint();
   }
}

} // namespace Ipopt