MoochoPack_NLPAlgoConfigIP.cpp

// @HEADER
// ***********************************************************************
// 
// Moocho: Multi-functional Object-Oriented arCHitecture for Optimization
//                  Copyright (2003) Sandia Corporation
// 
// Under terms of Contract DE-AC04-94AL85000, there is a non-exclusive
// license for use of this work by or on behalf of the U.S. Government.
// 
// This library is free software; you can redistribute it and/or modify
// it under the terms of the GNU Lesser General Public License as
// published by the Free Software Foundation; either version 2.1 of the
// License, or (at your option) any later version.
//  
// This library is distributed in the hope that it will be useful, but
// WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
// Lesser General Public License for more details.
//  
// You should have received a copy of the GNU Lesser General Public
// License along with this library; if not, write to the Free Software
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
// USA
// Questions? Contact Roscoe A. Bartlett (rabartl@sandia.gov) 
// 
// ***********************************************************************
// @HEADER

#include <assert.h>

#include <sstream>
#include <typeinfo>
#include <iostream>

#include "MoochoPack_NLPAlgoConfigIP.hpp"
#include "NLPInterfacePack_NLPBarrier.hpp"
#include "MoochoPack_NLPAlgo.hpp"
#include "MoochoPack_IpState.hpp"
#include "MoochoPack_NLPAlgoContainer.hpp"
#include "AbstractLinAlgPack_MatrixSymPosDefCholFactor.hpp"                     // rHL 
//#include "ConstrainedOptPack_MatrixSymPosDefInvCholFactor.hpp"    // .
#include "ConstrainedOptPack_MatrixSymPosDefLBFGS.hpp"        // .
//#include "ConstrainedOptPack_MatrixHessianSuperBasicInitDiagonal.hpp/ | rHL (super basics)
#include "AbstractLinAlgPack_MatrixSymDiagStd.hpp"                          // |

#include "NLPInterfacePack_NLPDirect.hpp"
#include "NLPInterfacePack_NLPVarReductPerm.hpp"
#include "NLPInterfacePack_CalcFiniteDiffProd.hpp"

// line search
#include "ConstrainedOptPack_DirectLineSearchArmQuad_Strategy.hpp"
#include "ConstrainedOptPack_DirectLineSearchArmQuad_StrategySetOptions.hpp"
#include "ConstrainedOptPack_MeritFuncNLPL1.hpp"
#include "ConstrainedOptPack_MeritFuncNLPModL1.hpp"

// Basis permutations and direct sparse solvers
#ifndef MOOCHO_NO_BASIS_PERM_DIRECT_SOLVERS
#include "ConstrainedOptPack_DecompositionSystemVarReductPerm.hpp"
#endif

#include "MoochoPack_MoochoAlgorithmStepNames.hpp"

#include "MoochoPack_UpdateBarrierParameter_Step.hpp"

#include "MoochoPack_PreEvalNewPointBarrier_Step.hpp"
#include "MoochoPack_PostEvalNewPointBarrier_Step.hpp"
#include "MoochoPack_ReducedGradientStd_Step.hpp"
//#include "MoochoPack_InitFinDiffReducedHessian_Step.hpp"
//#include "MoochoPack_InitFinDiffReducedHessian_StepSetOptions.hpp"
#include "MoochoPack_ReducedHessianSecantUpdateStd_Step.hpp"
#include "MoochoPack_ReducedHessianSecantUpdateBFGSFull_Strategy.hpp"
//#include "MoochoPack_ReducedHessianSecantUpdateBFGSProjected_Strategy.hpp"
//#include "MoochoPack_ReducedHessianSecantUpdateBFGSProjected_StrategySetOptions.hpp"
//#include "MoochoPack_ReducedHessianSecantUpdateLPBFGS_Strategy.hpp"
//#include "MoochoPack_ReducedHessianSecantUpdateLPBFGS_StrategySetOptions.hpp"
#include "MoochoPack_BFGSUpdate_Strategy.hpp"
#include "MoochoPack_BFGSUpdate_StrategySetOptions.hpp"
#include "MoochoPack_QuasiNormalStepStd_Step.hpp"
#include "MoochoPack_CheckDescentQuasiNormalStep_Step.hpp"
#include "MoochoPack_CheckDecompositionFromPy_Step.hpp"
#include "MoochoPack_CheckDecompositionFromRPy_Step.hpp"
#include "MoochoPack_TangentialStepIP_Step.hpp"
//#include "MoochoPack_TangentialStepWithoutBounds_Step.hpp"
//#include "MoochoPack_TangentialStepWithInequStd_Step.hpp"
//#include "MoochoPack_TangentialStepWithInequStd_StepSetOptions.hpp"
//#include "MoochoPack_SetDBoundsStd_AddedStep.hpp"
#include "MoochoPack_QPFailureReinitReducedHessian_Step.hpp"
#include "MoochoPack_CalcDFromYPYZPZ_Step.hpp"
#include "MoochoPack_CalcD_vStep_Step.hpp"

#include "MoochoPack_PreProcessBarrierLineSearch_Step.hpp"
#include "MoochoPack_PostProcessBarrierLineSearch_Step.hpp"
#include "MoochoPack_LineSearchFailureNewDecompositionSelection_Step.hpp"
#include "MoochoPack_LineSearchFilter_Step.hpp"
#include "MoochoPack_LineSearchFilter_StepSetOptions.hpp"
#include "MoochoPack_LineSearchFullStep_Step.hpp"
#include "MoochoPack_LineSearchDirect_Step.hpp"
//#include "MoochoPack_LineSearch2ndOrderCorrect_Step.hpp"
//#include "MoochoPack_LineSearch2ndOrderCorrect_StepSetOptions.hpp"
//#include "MoochoPack_FeasibilityStepReducedStd_Strategy.hpp"
//#include "MoochoPack_FeasibilityStepReducedStd_StrategySetOptions.hpp"
//#include "MoochoPack_QuasiRangeSpaceStepStd_Strategy.hpp"
//#include "MoochoPack_QuasiRangeSpaceStepTailoredApproach_Strategy.hpp"
//#include "MoochoPack_LineSearchWatchDog_Step.hpp"
//#include "MoochoPack_LineSearchWatchDog_StepSetOptions.hpp"
//#include "MoochoPack_LineSearchFullStepAfterKIter_Step.hpp"
//#include "MoochoPack_CalcLambdaIndepStd_AddedStep.hpp"
#include "MoochoPack_CalcReducedGradLagrangianStd_AddedStep.hpp"
#include "MoochoPack_CheckConvergenceStd_AddedStep.hpp"
#include "MoochoPack_CheckConvergenceIP_Strategy.hpp"
#include "MoochoPack_CheckSkipBFGSUpdateStd_StepSetOptions.hpp"
#include "MoochoPack_MeritFunc_PenaltyParamUpdate_AddedStepSetOptions.hpp"
#include "MoochoPack_MeritFunc_PenaltyParamUpdateMultFree_AddedStep.hpp"
//#include "MoochoPack_MeritFunc_PenaltyParamUpdateWithMult_AddedStep.hpp"
//#include "MoochoPack_MeritFunc_PenaltyParamsUpdateWithMult_AddedStep.hpp"
//#include "MoochoPack_MeritFunc_ModifiedL1LargerSteps_AddedStep.hpp"
//#include "MoochoPack_MeritFunc_ModifiedL1LargerSteps_AddedStepSetOptions.hpp"
//#include "MoochoPack_ActSetStats_AddedStep.hpp"
//#include "MoochoPack_NumFixedDepIndep_AddedStep.hpp"
#include "MoochoPack_UpdateReducedSigma_Step.hpp"

#include "MoochoPack_quasi_newton_stats.hpp"

// Misc utilities
#include "Teuchos_AbstractFactoryStd.hpp"
#include "Teuchos_dyn_cast.hpp"
#include "ReleaseResource_ref_count_ptr.hpp"
#include "Teuchos_TestForException.hpp"

// Stuff to read in options
#include "OptionsFromStreamPack_StringToIntMap.hpp"
#include "OptionsFromStreamPack_StringToBool.hpp"

// Stuff for exact reduced hessian
//#include "MoochoPack_ReducedHessianExactStd_Step.hpp"
//#include "MoochoPack_CrossTermExactStd_Step.hpp"
//#include "MoochoPack_DampenCrossTermStd_Step.hpp"

namespace {
  const double INF_BASIS_COND_CHANGE_FRAC      = 1e+20;
}

namespace MoochoPack {

//
// Here is where we define the default values for the algorithm.  These
// should agree with what are in the Moocho.opt.NLPAlgoConfigIP file.
//
NLPAlgoConfigIP::SOptionValues::SOptionValues()
  :max_basis_cond_change_frac_(-1.0)
  ,exact_reduced_hessian_(false)
  ,quasi_newton_(QN_AUTO)
  ,num_lbfgs_updates_stored_(-1)
  ,lbfgs_auto_scaling_(true)
  ,hessian_initialization_(INIT_HESS_AUTO)
  ,qp_solver_type_(QP_AUTO)
  ,reinit_hessian_on_qp_fail_(true)
  ,line_search_method_(LINE_SEARCH_AUTO)
  ,merit_function_type_(MERIT_FUNC_AUTO)
  ,l1_penalty_param_update_(L1_PENALTY_PARAM_AUTO)
  ,full_steps_after_k_(-1)
{}

NLPAlgoConfigIP::NLPAlgoConfigIP()
{}

NLPAlgoConfigIP::~NLPAlgoConfigIP()
{}

// overridden from NLPAlgoConfig

void NLPAlgoConfigIP::set_options( const options_ptr_t& options )
{
  options_ = options;
  decomp_sys_step_builder_.set_options(options);
}

const NLPAlgoConfig::options_ptr_t&
NLPAlgoConfigIP::get_options() const
{
  return options_;
}

void NLPAlgoConfigIP::config_algo_cntr(
  NLPAlgoContainer   *algo_cntr
  ,std::ostream       *trase_out
  )
{

  using Teuchos::RCP;
  using Teuchos::dyn_cast;
  using Teuchos::dyn_cast;

  if(trase_out) {
    *trase_out
      << std::endl
      << "*****************************************************************\n"
      << "*** NLPAlgoConfigIP configuration                               ***\n"
      << "***                                                           ***\n"
      << "*** Here, summary information about how the algorithm is      ***\n"
      << "*** configured is printed so that the user can see how the    ***\n"
      << "*** properties of the NLP and the set options influence       ***\n"
      << "*** how an algorithm is configured.                           ***\n"
      << "*****************************************************************\n";
  }

  // ////////////////////////////////////////////////////////////
  // A. ???

  // /////////////////////////////////////////////////////////////////////////
  // B. Create an algo object, give to algo_cntr, then give algo_cntr to algo

  if(trase_out)
    *trase_out << "\n*** Creating the NLPAlgo algo object ...\n";

  typedef Teuchos::RCP<NLPAlgo> algo_ptr_t;
  algo_ptr_t algo = Teuchos::rcp(new NLPAlgo);
  TEST_FOR_EXCEPT( !( algo.get() ) );
  algo_cntr->set_algo(algo);
  algo->set_algo_cntr(algo_cntr);

  // /////////////////////////////////////////////
  // C. Configure algo

  // /////////////////////////////////////////////////////
  // C.0 Set the nlp and track objects

  if(trase_out)
    *trase_out << "\n*** Setting the NLP and track objects to the algo object ...\n";

  algo->set_nlp( algo_cntr->get_nlp().get() );
  algo->set_track( algo_cntr->get_track() );

  // ////////////////////////////////////////////////
  // Determine what the options are:

  // Readin the options
  if(options_.get()) {
    readin_options( *options_, &uov_, trase_out );
  }
  else {
    if(trase_out) {
      *trase_out
        << "\n*** Warning, no OptionsFromStream object was set so a default set"
          " of options will be used!\n";
    }
  } 

  NLP &nlp = algo->nlp();
  nlp.initialize(algo->algo_cntr().check_results());
  // Get the dimensions of the NLP
  const size_type
    n   = nlp.n(),
    m   = nlp.m(),
    r   = m, // ToDo: Compute this for real!
    //dof = n - r,
    nb  = nlp.num_bounded_x();

  // Process the NLP
  NLPFirstOrder    *nlp_foi = NULL;
  NLPSecondOrder   *nlp_soi = NULL;
  NLPDirect        *nlp_fod = NULL;
  bool             tailored_approach = false;
  decomp_sys_step_builder_.process_nlp_and_options(
    trase_out, nlp
    ,&nlp_foi, &nlp_soi, &nlp_fod, &tailored_approach
    );

  const int max_dof_quasi_newton_dense
    = decomp_sys_step_builder_.current_option_values().max_dof_quasi_newton_dense_;

  // Make sure that we can handle this type of NLP currently
  TEST_FOR_EXCEPTION(
    m == 0, std::logic_error
    ,"NLPAlgoConfigIP::config_algo_cntr(...) : Error, "
    "can not currently solve an unconstrained NLP!" );
  TEST_FOR_EXCEPTION(
    n == m, std::logic_error
    ,"NLPAlgoConfigIP::config_algo_cntr(...) : Error, "
    "can not currently solve a square system of equations!" );

  // //////////////////////////////////////////////////////
  // C.1.  Sort out the options

  if(trase_out)
    *trase_out
      << "\n*** Sorting out some of the options given input options ...\n";

  if( tailored_approach ) {
    // Change the options for the tailored approach. 
    if(trase_out) {
      *trase_out
        << "\nThis is a tailored approach NLP (NLPDirect) which forces the following options:\n"
        << "merit_function_type         = L1;\n"
        << "l1_penalty_parameter_update = MULT_FREE;\n"
        << "null_space_matrix           = EXPLICIT;\n"
        ;
    }
    cov_.merit_function_type_
      = MERIT_FUNC_L1;
    cov_.l1_penalty_param_update_
      = L1_PENALTY_PARAM_MULT_FREE;
    decomp_sys_step_builder_.current_option_values().null_space_matrix_type_
      = DecompositionSystemStateStepBuilderStd::NULL_SPACE_MATRIX_EXPLICIT;
  }

  if( !tailored_approach && uov_.merit_function_type_ != MERIT_FUNC_L1  ) {
    if(trase_out) {
      *trase_out
        << "\nThe only merit function currently supported is:\n"
        << "merit_function_type         = L1;\n"
        ;
    }
    cov_.merit_function_type_   = MERIT_FUNC_L1;
  }

  // Decide what type of quasi newton update to use
  switch( uov_.quasi_newton_ ) {
    case QN_AUTO: {
      if(trase_out)
        *trase_out
          << "\nquasi_newton == AUTO:"
          << "\nnlp.num_bounded_x() == " << nlp.num_bounded_x() << ":\n";
      //if( n - r > cov_.max_dof_quasi_newton_dense_ ) {
      //  if(trase_out)
      //    *trase_out
      //      << "n-r = " << n-r << " > max_dof_quasi_newton_dense = "
      //      << cov_.max_dof_quasi_newton_dense_ <<  ":\n"
      //      << "setting quasi_newton == LBFGS\n";
      //  cov_.quasi_newton_ = QN_LBFGS;
      //}
      //else {
        if(trase_out)
          *trase_out
            << "n-r = " << n-r << " <= max_dof_quasi_newton_dense = "
            << max_dof_quasi_newton_dense << ":\n"
            << "setting quasi_newton == BFGS\n";
        cov_.quasi_newton_ = QN_BFGS;
        //}
      break;
    }
    case QN_BFGS:
    case QN_PBFGS:
    case QN_LBFGS:
    case QN_LPBFGS:
      cov_.quasi_newton_ = uov_.quasi_newton_;
      break;
      default:
      TEST_FOR_EXCEPT(true); // Invalid option!
  }

  // ToDo: Sort out the rest of the options!

  // Set the default options that where not already set yet
  set_default_options(uov_,&cov_,trase_out);

  // ToDo: Implement the 2nd order correction linesearch
  if( cov_.line_search_method_ == LINE_SEARCH_2ND_ORDER_CORRECT ) {
    if(trase_out)
      *trase_out <<
        "\nline_search_method == 2ND_ORDER_CORRECT:\n"
        "Sorry, the second order corrrection linesearch is not updated yet!\n"
        "setting line_search_method = DIRECT ...\n";
    cov_.line_search_method_ = LINE_SEARCH_DIRECT;
  }
  if( cov_.line_search_method_ == LINE_SEARCH_WATCHDOG ) {
    if(trase_out)
      *trase_out <<
        "\nline_search_method ==WATCHDOG:\n"
        "Sorry, the watchdog linesearch is not updated yet!\n"
        "setting line_search_method = DIRECT ...\n";
    cov_.line_search_method_ = LINE_SEARCH_DIRECT;
  }
  
  // /////////////////////////////////////////////////////
  // C.1. Create the decomposition system object

  typedef RCP<DecompositionSystem> decomp_sys_ptr_t;
  decomp_sys_ptr_t decomp_sys;
  decomp_sys_step_builder_.create_decomp_sys(
    trase_out, nlp, nlp_foi, nlp_soi, nlp_fod, tailored_approach
    ,&decomp_sys
    );

#ifndef MOOCHO_NO_BASIS_PERM_DIRECT_SOLVERS
  RCP<DecompositionSystemVarReductPerm>
    decomp_sys_perm = Teuchos::rcp_dynamic_cast<DecompositionSystemVarReductPerm>(decomp_sys);
#endif

  // /////////////////////////////////////////////////////
  // C.2. Create and set the state object

  if(trase_out)
    *trase_out
      << "\n*** Creating the state object and setting up iteration quantity objects ...\n";

  {
    //
    // Create the state object with the vector spaces
    //

    typedef RCP<IpState>   state_ptr_t;
    state_ptr_t
      state = Teuchos::rcp(
        new IpState(
          decomp_sys
          ,nlp.space_x()
          ,nlp.space_c()
          ,( tailored_approach
             ? ( nlp_fod->var_dep().size() 
               ? nlp.space_x()->sub_space(nlp_fod->var_dep())->clone()
               : Teuchos::null )
             : decomp_sys->space_range() // could be NULL for BasisSystemPerm
            )
          ,( tailored_approach
             ?( nlp_fod->var_indep().size()
              ? nlp.space_x()->sub_space(nlp_fod->var_indep())->clone()
              : Teuchos::null )
             : decomp_sys->space_null() // could be NULL for BasisSystemPerm
            )
          )
        );

    //
    // Set the iteration quantities for the NLP matrix objects
    //

    decomp_sys_step_builder_.add_iter_quantities(
      trase_out, nlp, nlp_foi, nlp_soi, nlp_fod, tailored_approach, decomp_sys
      ,state
      );

    // Set the iteration quantities for the barrier terms
    state->set_iter_quant(
      Vu_name
      ,Teuchos::rcp(
      new IterQuantityAccessContiguous<MatrixSymDiagStd>(
        1,
        Vu_name,
        Teuchos::rcp( new Teuchos::AbstractFactoryStd<MatrixSymDiagStd,MatrixSymDiagStd,
            MatrixSymDiagStd::PostMod>( nlp.space_x() ) 
        )
        )
      )
      );

    state->set_iter_quant(
      Vl_name
      ,Teuchos::rcp(
      new IterQuantityAccessContiguous<MatrixSymDiagStd>(
        1,
        Vl_name,
        Teuchos::rcp( new Teuchos::AbstractFactoryStd<MatrixSymDiagStd,MatrixSymDiagStd,
            MatrixSymDiagStd::PostMod>( nlp.space_x() ) 
        )
        )
      )
      );

    state->set_iter_quant(
      invXu_name
      ,Teuchos::rcp(
      new IterQuantityAccessContiguous<MatrixSymDiagStd>(
        1,
        invXu_name,
        Teuchos::rcp( new Teuchos::AbstractFactoryStd<MatrixSymDiagStd,MatrixSymDiagStd,
            MatrixSymDiagStd::PostMod>( nlp.space_x() ) 
        )
        )
      )
      );

    state->set_iter_quant(
      invXl_name
      ,Teuchos::rcp(
      new IterQuantityAccessContiguous<MatrixSymDiagStd>(
        1,
        invXl_name,
        Teuchos::rcp( new Teuchos::AbstractFactoryStd<MatrixSymDiagStd,MatrixSymDiagStd,
            MatrixSymDiagStd::PostMod>( nlp.space_x() ) 
        )
        )
      )
      );

    state->set_iter_quant(
      rHB_name
      ,Teuchos::rcp(
      new IterQuantityAccessContiguous<MatrixSymOp>(
        1,
        rHB_name,
        Teuchos::rcp(
        new Teuchos::AbstractFactoryStd<MatrixSymOp,MatrixSymPosDefCholFactor,MatrixSymPosDefCholFactor::PostMod>(
          MatrixSymPosDefCholFactor::PostMod(
          true      // maintain_original
          ,false    // maintain_factor
          ,true     // allow_factor      (always!)
          )
          )
        )
        )
      )
      );
        
    state->set_iter_quant(
      B_name
      ,Teuchos::rcp(
      new IterQuantityAccessContiguous<MatrixSymOp>(
        1,
        B_name,
        Teuchos::rcp(
        new Teuchos::AbstractFactoryStd<MatrixSymOp,MatrixSymPosDefCholFactor,MatrixSymPosDefCholFactor::PostMod>(
          MatrixSymPosDefCholFactor::PostMod(
          true      // maintain_original
          ,false    // maintain_factor
          ,true     // allow_factor      (always!)
          )
          )
        )
        )
      )
      );

    state->set_iter_quant(
      Sigma_name
      ,Teuchos::rcp(
      new IterQuantityAccessContiguous<MatrixSymDiagStd>(
        1,
        Sigma_name,
        Teuchos::rcp( new Teuchos::AbstractFactoryStd<MatrixSymDiagStd,MatrixSymDiagStd,
            MatrixSymDiagStd::PostMod>( nlp.space_x() ) 
        )
        )
      )
      );

    // These iteration quantities are defined in IpState,     
    // force their creation and resize them
    dyn_cast< IterQuantityAccessContiguous<value_type> >(state->barrier_obj()).resize(2);
    dyn_cast< IterQuantityAccessContiguous<VectorMutable> >(state->grad_barrier_obj()).resize(2);

    // Add reduced Hessian of the Lagrangian

    if( !cov_.exact_reduced_hessian_ ) {
      RCP<Teuchos::AbstractFactory<MatrixSymOp> >
        abstract_factory_rHL = Teuchos::rcp(
          new Teuchos::AbstractFactoryStd<MatrixSymOp,MatrixSymPosDefCholFactor,MatrixSymPosDefCholFactor::PostMod>(
            MatrixSymPosDefCholFactor::PostMod(
              true    // maintain_original
              ,false  // maintain_factor
              ,true   // allow_factor      (always!)
              )
            )
          );
      state->set_iter_quant(
        rHL_name
        ,Teuchos::rcp(
          new IterQuantityAccessContiguous<MatrixSymOp>(
            1
            ,rHL_name
            ,abstract_factory_rHL
            )
          )
        );
    }
    else {
      TEST_FOR_EXCEPT(true); // ToDo: Add rHL for an exact reduced Hessian!
    }
    
    //
    // Set the NLP merit function 
    //

    if( cov_.line_search_method_ != LINE_SEARCH_NONE 
       && cov_.line_search_method_ != LINE_SEARCH_FILTER) {
      RCP<Teuchos::AbstractFactory<MeritFuncNLP> >
        merit_func_factory = Teuchos::null;
      switch( cov_.merit_function_type_ ) {
        case MERIT_FUNC_L1:
          merit_func_factory = Teuchos::rcp(
            new Teuchos::AbstractFactoryStd<MeritFuncNLP,MeritFuncNLPL1>());
          break;
        case MERIT_FUNC_MOD_L1:
        case MERIT_FUNC_MOD_L1_INCR:
          merit_func_factory = Teuchos::rcp(
            new Teuchos::AbstractFactoryStd<MeritFuncNLP,MeritFuncNLPModL1>());
          break;
        default:
          TEST_FOR_EXCEPT(true);  // local programming error
      }
      state->set_iter_quant(
        merit_func_nlp_name
        ,Teuchos::rcp(
          new IterQuantityAccessContiguous<MeritFuncNLP>(
            1
            ,merit_func_nlp_name
            ,merit_func_factory
            )
          )
        );
    }

    if (cov_.line_search_method_ == LINE_SEARCH_FILTER)
      {
        // Add the filter iteration quantity
        state->set_iter_quant(
        FILTER_IQ_STRING
        ,Teuchos::rcp(
             new IterQuantityAccessContiguous<Filter_T>(1,FILTER_IQ_STRING)
        )
        );
      }

    //
    // Resize the number of storage locations (these can be changed later).
    //
    // Also, touch all of the value_type, index_type and vector iteration quantities
    // that we know about so that when state.dump_iter_quant() is called, all of the
    // iteration quantities will be included.
    //
    
    typedef IterQuantityAccessContiguous<value_type>            IQ_scalar_cngs;
    typedef IterQuantityAccessContiguous<VectorMutable>   IQ_vector_cngs;

    dyn_cast<IQ_vector_cngs>(state->x()).resize(2);
    dyn_cast<IQ_scalar_cngs>(state->f()).resize(2);
    if(m) dyn_cast<IQ_vector_cngs>(state->c()).resize(2);
    dyn_cast<IQ_vector_cngs>(state->Gf()).resize(2);
    if(m && nlp_foi) state->Gc();

    if( m
#ifndef MOOCHO_NO_BASIS_PERM_DIRECT_SOLVERS
       && decomp_sys_perm.get() == NULL
#endif
      ) state->py();
    if(m) dyn_cast<IQ_vector_cngs>(state->Ypy()).resize(2);
    if( m
#ifndef MOOCHO_NO_BASIS_PERM_DIRECT_SOLVERS
      && decomp_sys_perm.get() == NULL
#endif
      ) state->pz();
    if(m) dyn_cast<IQ_vector_cngs>(state->Zpz()).resize(2);
    dyn_cast<IQ_vector_cngs>(state->d()).resize(2);

    if( n > m ) {
      dyn_cast<IQ_vector_cngs>(state->rGf()).resize(2);
      state->w();
      state->zeta();
      state->qp_grad();
    }
    state->eta();
    
    dyn_cast<IQ_scalar_cngs>(state->barrier_parameter()).resize(2);

    dyn_cast<IQ_scalar_cngs>(state->alpha()).resize(2);
    dyn_cast<IQ_scalar_cngs>(state->mu()).resize(2);
    dyn_cast<IQ_scalar_cngs>(state->phi()).resize(2);

    dyn_cast<IQ_scalar_cngs>(state->opt_kkt_err()).resize(2);
    dyn_cast<IQ_scalar_cngs>(state->feas_kkt_err()).resize(2);
    if( n > m ) {
      dyn_cast<IQ_vector_cngs>(state->rGL()).resize(2);
    }
    if(m) dyn_cast<IQ_vector_cngs>(state->lambda()).resize(2);
    dyn_cast<IQ_vector_cngs>(state->nu()).resize(2);

    // Set the state object
    algo->set_state( state );
  }

  // /////////////////////////////////////////////////////
  // C.3  Create and set the step objects

  if(trase_out)
    *trase_out << "\n*** Creating and setting the step objects ...\n";

  {

    //
    // Create some standard step objects that will be used by many different
    // specific algorithms
    //
    
    typedef RCP<AlgorithmStep>   algo_step_ptr_t;

    // Create the EvalNewPoint step and associated objects
    algo_step_ptr_t                                    eval_new_point_step           = Teuchos::null;
    RCP<CalcFiniteDiffProd>                  calc_fd_prod                  = Teuchos::null;
    RCP<VariableBoundsTester>                bounds_tester                 = Teuchos::null;
    RCP<NewDecompositionSelection_Strategy>  new_decomp_selection_strategy = Teuchos::null;
    decomp_sys_step_builder_.create_eval_new_point(
      trase_out, nlp, nlp_foi, nlp_soi, nlp_fod, tailored_approach, decomp_sys
      ,&eval_new_point_step, &calc_fd_prod, &bounds_tester, &new_decomp_selection_strategy
      );

    // UpdateBarrierParameter_Step
    Teuchos::RCP<UpdateBarrierParameter_Step> updateBarrierParameter_step  = Teuchos::null;
    updateBarrierParameter_step = Teuchos::rcp(new UpdateBarrierParameter_Step());
    if(options_.get()) 
      {
      UpdateBarrierParameter_StepSetOptions options_setter(updateBarrierParameter_step.get());
      options_setter.set_options(*options_);
      }
    
    // PreEvalNewPointBarrier_Step
    Teuchos::RCP<PreEvalNewPointBarrier_Step>  preEvalNewPointBarrier_step  = Teuchos::null;
    preEvalNewPointBarrier_step = Teuchos::rcp(new PreEvalNewPointBarrier_Step());
    if(options_.get()) 
      {
      PreEvalNewPointBarrier_StepSetOptions
        options_setter(preEvalNewPointBarrier_step.get());
      options_setter.set_options(*options_);
      }

    // PostEvalNewPointBarrier_Step
    algo_step_ptr_t postEvalNewPointBarrier_step = Teuchos::rcp(new PostEvalNewPointBarrier_Step());

    // ReducedGradient_Step
    algo_step_ptr_t    reduced_gradient_step = Teuchos::null;
    if( !tailored_approach ) {
      reduced_gradient_step = Teuchos::rcp(new ReducedGradientStd_Step());
    }

    // RangeSpace_Step
    algo_step_ptr_t    quasi_normal_step_step = Teuchos::null;
    if( !tailored_approach ) {
      quasi_normal_step_step = Teuchos::rcp(new QuasiNormalStepStd_Step());
    }

    // Check and change decomposition 
    algo_step_ptr_t    check_decomp_from_py_step  = Teuchos::null;
    algo_step_ptr_t    check_decomp_from_Rpy_step = Teuchos::null;
    if( new_decomp_selection_strategy.get() && cov_.max_basis_cond_change_frac_ < INF_BASIS_COND_CHANGE_FRAC ) {
      check_decomp_from_py_step = Teuchos::rcp(
        new CheckDecompositionFromPy_Step(
          new_decomp_selection_strategy
          ,cov_.max_basis_cond_change_frac_
          ) );
      check_decomp_from_Rpy_step = Teuchos::rcp(
        new CheckDecompositionFromRPy_Step(
          new_decomp_selection_strategy
          ,cov_.max_basis_cond_change_frac_
          ) );
    }

    // CheckDescentQuasiNormalStep
    algo_step_ptr_t    check_descent_quasi_normal_step_step = Teuchos::null;
    if( algo->algo_cntr().check_results() ) {
      check_descent_quasi_normal_step_step = Teuchos::rcp(new CheckDescentQuasiNormalStep_Step(calc_fd_prod));
    }

    // ReducedGradient_Step
    //algo_step_ptr_t    reduced_gradient_step = Teuchos::null;
    //if( !tailored_approach ) {
    //  reduced_gradient_step = Teuchos::rcp(new ReducedGradientStd_Step());
    //}

    // CheckSkipBFGSUpdate
    algo_step_ptr_t    check_skip_bfgs_update_step = Teuchos::null;
    if(!cov_.exact_reduced_hessian_) {
      RCP<CheckSkipBFGSUpdateStd_Step>
        step = Teuchos::rcp(new CheckSkipBFGSUpdateStd_Step());
      if(options_.get()) {
        CheckSkipBFGSUpdateStd_StepSetOptions
          opt_setter( step.get() );
        opt_setter.set_options( *options_ );
      }
      check_skip_bfgs_update_step = step;
    }

    // ReducedHessian_Step
    algo_step_ptr_t    reduced_hessian_step = Teuchos::null;
    {
      // Get the strategy object that will perform the actual secant update.
      RCP<ReducedHessianSecantUpdate_Strategy>
        secant_update_strategy = Teuchos::null;
      switch( cov_.quasi_newton_ )
      {
        case QN_BFGS:
        case QN_PBFGS:
        case QN_LBFGS:
        case QN_LPBFGS:
        {
          // create and setup the actual BFGS strategy object
          typedef RCP<BFGSUpdate_Strategy> bfgs_strategy_ptr_t;
          bfgs_strategy_ptr_t
            bfgs_strategy = Teuchos::rcp(new BFGSUpdate_Strategy);
          if(options_.get()) { 
            BFGSUpdate_StrategySetOptions
              opt_setter( bfgs_strategy.get() );
            opt_setter.set_options( *options_ );
          }
          switch( cov_.quasi_newton_ ) {
            case QN_BFGS:
            case QN_LBFGS:
            {
              secant_update_strategy = Teuchos::rcp(new ReducedHessianSecantUpdateBFGSFull_Strategy(bfgs_strategy));
              break;
            }
            case QN_PBFGS:
            case QN_LPBFGS:
            {
              TEST_FOR_EXCEPTION(
                true, std::logic_error
                ,"NLPAlgoConfigIP::config_algo_cntr(...) : Error, "
                "The quansi_newton options of PBFGS and LPBFGS have not been updated yet!" );
              break;
            }
          }
          break;
        }
        default:
          TEST_FOR_EXCEPT(true);
      }
      
      // Finally build the step object
      reduced_hessian_step = Teuchos::rcp(
        new ReducedHessianSecantUpdateStd_Step( secant_update_strategy ) );
      // Add the QuasiNewtonStats iteration quantity
      algo->state().set_iter_quant(
        quasi_newton_stats_name
        ,Teuchos::rcp(new IterQuantityAccessContiguous<QuasiNewtonStats>(
          1
          ,quasi_newton_stats_name
#ifdef _MIPS_CXX
          ,Teuchos::RCP<Teuchos::AbstractFactoryStd<QuasiNewtonStats,QuasiNewtonStats> >(
            new Teuchos::AbstractFactoryStd<QuasiNewtonStats,QuasiNewtonStats>())
#endif
          )
        ));
    }

    // UpdateReducedSigma_Step
    Teuchos::RCP<UpdateReducedSigma_Step> updateReducedSigma_step = Teuchos::null;
    updateReducedSigma_step = Teuchos::rcp(new UpdateReducedSigma_Step());
    if(options_.get()) 
      {
      UpdateReducedSigma_StepSetOptions
        options_setter(updateReducedSigma_step.get());
      options_setter.set_options(*options_);
      }
    
    // NullSpace_Step
    algo_step_ptr_t    tangential_step = Teuchos::rcp(new TangentialStepIP_Step());
    /*    algo_step_ptr_t    set_d_bounds_step    = Teuchos::null;
    algo_step_ptr_t    tangential_step_step = Teuchos::null;
    if( nb == 0 ) {
      tangential_step_step = Teuchos::rcp(new TangentialStepWithoutBounds_Step());
    }
    else {
      // Step object that sets bounds for QP subproblem
      set_d_bounds_step = Teuchos::rcp(new SetDBoundsStd_AddedStep());
      // QP Solver object
      Teuchos::RCP<QPSolverRelaxed>  qp_solver = Teuchos::null;
        // ToDo: Create the QP solver!
      // QP solver tester
      Teuchos::RCP<QPSolverRelaxedTester> 
        qp_solver_tester = Teuchos::rcp(new QPSolverRelaxedTester());
      if(options_.get()) {
        QPSolverRelaxedTesterSetOptions
          opt_setter( qp_solver_tester.get() );
        opt_setter.set_options( *options_ );
      }
      // The null-space step
      Teuchos::RCP<TangentialStepWithInequStd_Step>
        tangential_step_with_inequ_step = Teuchos::rcp(
          new TangentialStepWithInequStd_Step(
            qp_solver, qp_solver_tester ) );
      if(options_.get()) {
        TangentialStepWithInequStd_StepSetOptions
          opt_setter( tangential_step_with_inequ_step.get() );
        opt_setter.set_options( *options_ );
      }
      tangential_step_step = tangential_step_with_inequ_step;
      // Step for reinitialization reduced Hessian on QP failure
      tangential_step_step = Teuchos::rcp(
        new QPFailureReinitReducedHessian_Step(tangential_step_step)
        );
        }*/

    // CalcDFromYPYZPZ_Step
    algo_step_ptr_t calc_d_from_Ypy_Zpy_step = Teuchos::null;
      {
      calc_d_from_Ypy_Zpy_step = Teuchos::rcp(new CalcDFromYPYZPZ_Step());
      }

    // CalcD_vStep_Step
    algo_step_ptr_t calc_d_v_step_step = Teuchos::rcp(new CalcD_vStep_Step());

    // build the barrier nlp decorator to be used by the line search
    Teuchos::RCP<NLPInterfacePack::NLPBarrier> barrier_nlp = Teuchos::rcp(new NLPInterfacePack::NLPBarrier());
    barrier_nlp->InitializeFromNLP( algo_cntr->get_nlp() );

    // PreProcessBarrierLineSearch_Step
    algo_step_ptr_t preprocess_barrier_linesearch_step = Teuchos::rcp(new PreProcessBarrierLineSearch_Step(barrier_nlp));

    // PostProcessBarrierLineSearch_Step
    algo_step_ptr_t postprocess_barrier_linesearch_step = Teuchos::rcp(new PostProcessBarrierLineSearch_Step(barrier_nlp));

    // CalcReducedGradLagrangianStd_AddedStep
    algo_step_ptr_t    calc_reduced_grad_lagr_step = Teuchos::null;
    {
      calc_reduced_grad_lagr_step = Teuchos::rcp(
        new CalcReducedGradLagrangianStd_AddedStep() );
    }

    // CheckConvergence_Step
    algo_step_ptr_t    check_convergence_step = Teuchos::null;
      {
      // Create the strategy object
      RCP<CheckConvergenceIP_Strategy>
        check_convergence_strategy = Teuchos::rcp(new CheckConvergenceIP_Strategy());

      if(options_.get()) 
        {
        CheckConvergence_StrategySetOptions
          opt_setter( check_convergence_strategy.get() );
        opt_setter.set_options( *options_ );
        }
      
      RCP<CheckConvergenceStd_AddedStep>
        _check_convergence_step = Teuchos::rcp(new CheckConvergenceStd_AddedStep(check_convergence_strategy));
      
      check_convergence_step = _check_convergence_step;
      }

    // MeritFuncPenaltyParamUpdate_Step
    algo_step_ptr_t    merit_func_penalty_param_update_step = Teuchos::null;
    if( cov_.line_search_method_ == LINE_SEARCH_FILTER ) {
      // We don't need to update a penalty parameter for the filter method :-)
    }
    else if( cov_.line_search_method_ != LINE_SEARCH_NONE ) {
      RCP<MeritFunc_PenaltyParamUpdate_AddedStep>
        param_update_step = Teuchos::null;
      switch( cov_.merit_function_type_ ) {
        case MERIT_FUNC_L1: {
          switch(cov_.l1_penalty_param_update_) {
            case L1_PENALTY_PARAM_WITH_MULT:
//              param_update_step
//                = Teuchos::rcp(new  MeritFunc_PenaltyParamUpdateWithMult_AddedStep());
              TEST_FOR_EXCEPTION(
                true, std::logic_error
                ,"NLPAlgoConfigIP::config_algo_cntr(...) : Error, "
                "The l1_penalty_parameter_update option of MULT_FREE has not been updated yet!" );
              break;
            case L1_PENALTY_PARAM_MULT_FREE:
              param_update_step
                = Teuchos::rcp(new  MeritFunc_PenaltyParamUpdateMultFree_AddedStep());
              break;
            default:
              TEST_FOR_EXCEPT(true);
          }
          break;
        }
        case MERIT_FUNC_MOD_L1:
        case MERIT_FUNC_MOD_L1_INCR:
//          param_update_step = new  MeritFunc_PenaltyParamsUpdateWithMult_AddedStep(
//                      Teuchos::rcp_implicit_cast<MeritFuncNLP>(merit_func) );
          TEST_FOR_EXCEPTION(
            true, std::logic_error
            ,"NLPAlgoConfigIP::config_algo_cntr(...) : Error, "
            "The merit_function_type options of MODIFIED_L1 and MODIFIED_L1_INCR have not been updated yet!" );
          break;
        default:
          TEST_FOR_EXCEPT(true);  // local programming error
      }
      if(options_.get()) {
        MeritFunc_PenaltyParamUpdate_AddedStepSetOptions
          ppu_options_setter( param_update_step.get() );
        ppu_options_setter.set_options( *options_ );
      }
      merit_func_penalty_param_update_step = param_update_step;
    }

    // LineSearch_Step
    algo_step_ptr_t    line_search_full_step_step = Teuchos::null;
    {
      line_search_full_step_step = Teuchos::rcp(new LineSearchFullStep_Step(bounds_tester));
    }

    // LineSearch_Step
    algo_step_ptr_t    line_search_step = Teuchos::null;
    if( cov_.line_search_method_ != LINE_SEARCH_NONE ) {
      RCP<DirectLineSearchArmQuad_Strategy>
        direct_line_search = Teuchos::rcp(new  DirectLineSearchArmQuad_Strategy());
      if(options_.get()) {
        ConstrainedOptPack::DirectLineSearchArmQuad_StrategySetOptions
          ls_options_setter( direct_line_search.get(), "DirectLineSearchArmQuadSQPStep" );
        ls_options_setter.set_options( *options_ );
      }
      switch( cov_.line_search_method_ ) {
        case LINE_SEARCH_DIRECT: {
          line_search_step = Teuchos::rcp(new LineSearchDirect_Step(direct_line_search));
          break;
        }
        case LINE_SEARCH_2ND_ORDER_CORRECT: {
          TEST_FOR_EXCEPTION(
            true, std::logic_error
            ,"NLPAlgoConfigIP::config_algo_cntr(...) : Error, "
            "The line_search_method option of 2ND_ORDER_CORRECT has not been updated yet!" );
          break;
        }
        case LINE_SEARCH_WATCHDOG: {
          TEST_FOR_EXCEPTION(
            true, std::logic_error
            ,"NLPAlgoConfigIP::config_algo_cntr(...) : Error, "
            "The line_search_method option of WATCHDOG has not been updated yet!" );
          break;
        }
        case LINE_SEARCH_FILTER: 
          {
          Teuchos::RCP<LineSearchFilter_Step> 
            line_search_filter_step = Teuchos::rcp(new LineSearchFilter_Step(barrier_nlp, barrier_obj_name, grad_barrier_obj_name));

          if(options_.get()) 
            {
            LineSearchFilter_StepSetOptions options_setter(line_search_filter_step.get());
            options_setter.set_options(*options_);
            }

          line_search_step = line_search_filter_step;         
          break; 
          }
      }
    }

    // LineSearchFailure
    if( new_decomp_selection_strategy.get() ) {
      line_search_step = Teuchos::rcp(
        new LineSearchFailureNewDecompositionSelection_Step(
          line_search_step
          ,new_decomp_selection_strategy
          )
        );
    }
  
    //
    // Create the algorithm depending on the type of NLP we are trying to solve.
    //

    if( m == 0 ) {
      if( nb == 0 ) {
        //
        // Unconstrained NLP (m == 0, num_bounded_x == 0)
        //
        if(trase_out)
          *trase_out 
            << "\nConfiguring an algorithm for an unconstrained "
            << "NLP (m == 0, num_bounded_x == 0) ...\n";
        TEST_FOR_EXCEPTION(
          m == 0 && nb == 0, std::logic_error
          ,"NLPAlgoConfigIP::config_alg_cntr(...) : Error, "
          "Unconstrained NLPs are not supported yet!" );
      }
      else {
        //
        // Simple bound constrained NLP (m == 0, num_bounded_x > 0)
        //
        if(trase_out)
          *trase_out 
            << "\nConfiguring an algorithm for a simple bound constrained "
            << "NLP (m == 0, num_bounded_x > 0) ...\n";
        TEST_FOR_EXCEPTION(
          m == 0 && nb == 0, std::logic_error
          ,"NLPAlgoConfigIP::config_alg_cntr(...) : Error, "
          "Bound constrained NLPs are not supported yet!" );
      }
    }
    else if( n == m ) {
      //
      // System of Nonlinear equations (n == m)
      //
      if(trase_out)
        *trase_out 
          << "\nConfiguring an algorithm for a system of nonlinear equations "
          << "NLP (n == m) ...\n";
      TEST_FOR_EXCEPTION(
        n == m, std::logic_error
        ,"NLPAlgoConfigIP::config_alg_cntr(...) : Error, "
        "Nonlinear equation (NLE) problems are not supported yet!" );
      TEST_FOR_EXCEPT(true); // ToDo: add the step objects for this algorithm
    }
    else if ( m > 0 || nb > 0 ) {
      //
      // General nonlinear NLP ( m > 0 )
      //
      if( nb == 0 ) {
        //
        // Nonlinear equality constrained NLP ( m > 0 && num_bounded_x == 0 )
        //
        if(trase_out)
          *trase_out 
            << "\nConfiguring an algorithm for a nonlinear equality constrained "
            << "NLP ( m > 0 && num_bounded_x == 0) ...\n";
      }
      else {
        //
        // Nonlinear inequality constrained NLP ( num_bounded_x > 0 )
        //
        if(trase_out)
          *trase_out 
            << "\nConfiguring an algorithm for a nonlinear generally constrained "
            << "NLP ( num_bounded_x > 0 ) ...\n";
      }



      // Add all the steps to the algorithm
      
      int step_num       = 0;
      int assoc_step_num = 0;
  
       // UpdateBarrierParameter
      //algo->insert_step( ++step_num, "UpdateBarrierParameter", updateBarrierParameter_step );

      // EvalNewPoint
      algo->insert_step( ++step_num, EvalNewPoint_name, eval_new_point_step );

      //* EvalNewPoint pre steps
      // PreEvalNewPointBarrier
      algo->insert_assoc_step( step_num, IterationPack::PRE_STEP, 1, "PreEvalNewPointBarrier", preEvalNewPointBarrier_step);

      //* EvalNewPoint post steps
      if( check_descent_quasi_normal_step_step.get() && tailored_approach && algo->algo_cntr().check_results() )
        {
        algo->insert_assoc_step(
          step_num
          ,IterationPack::POST_STEP
          ,++assoc_step_num
          ,"CheckDescentQuasiNormalStep"
          ,check_descent_quasi_normal_step_step
          );
        }

      // PostEvalNewPointBarrier
      algo->insert_assoc_step( step_num, IterationPack::POST_STEP, ++assoc_step_num, "PostEvalNewPointBarrier", postEvalNewPointBarrier_step);
      assoc_step_num = 0;
      
      // ReducedGradient
      if( !tailored_approach ) {
        algo->insert_step( ++step_num, ReducedGradient_name, reduced_gradient_step );
        } 

         // CalcReducedGradLagrangian
        algo->insert_step( ++step_num, CalcReducedGradLagrangian_name, calc_reduced_grad_lagr_step );

        // CalcLagrangeMultDecomposed
        // Compute these here so that in case we converge we can report them
        if( !tailored_approach ) {
           // ToDo: Insert this step
        }

        // CheckConvergence
        algo->insert_step( ++step_num, CheckConvergence_name, check_convergence_step );

        //} 

       // UpdateBarrierParameter
      algo->insert_step( ++step_num, "UpdateBarrierParameter", updateBarrierParameter_step );

      // QuasiNormalStep
      if( !tailored_approach ) {
        algo->insert_step( ++step_num, QuasiNormalStep_name, quasi_normal_step_step );
        assoc_step_num = 0;
        if( check_decomp_from_py_step.get() )
          algo->insert_assoc_step(
            step_num
            ,IterationPack::POST_STEP
            ,++assoc_step_num
            ,"CheckDecompositionFromPy"
            ,check_decomp_from_py_step
            );
        if( check_decomp_from_Rpy_step.get() )
          algo->insert_assoc_step(
            step_num
            ,IterationPack::POST_STEP
            ,++assoc_step_num
            ,"CheckDecompositionFromRPy"
            ,check_decomp_from_Rpy_step
            );
        if( check_descent_quasi_normal_step_step.get() )
          algo->insert_assoc_step(
            step_num
            ,IterationPack::POST_STEP
            ,++assoc_step_num
            ,"CheckDescentQuasiNormalStep"
            ,check_descent_quasi_normal_step_step
            );
      }

      // ReducedHessian
      algo->insert_step( ++step_num, ReducedHessian_name, reduced_hessian_step );

      // UpdateReducedSigma_Step
      algo->insert_step( ++step_num, "UpdateReducedSigma", updateReducedSigma_step);

      // TangentialStep
      algo->insert_step( ++step_num, "TangentialStepIP", tangential_step);
      // CalcDFromYPYZPZ
      algo->insert_step( ++step_num, CalcDFromYPYZPZ_name, calc_d_from_Ypy_Zpy_step );
      
      // CalcD_vStep_Step
      algo->insert_step( ++step_num, "CalcD_vStep_Step", calc_d_v_step_step );

      // PreProcessBarrierLineSearch_Step
      algo->insert_step( ++step_num, "PreProcessBarrierLineSearch_Step", preprocess_barrier_linesearch_step );

      // LineSearch
      if( cov_.line_search_method_ == LINE_SEARCH_NONE ) {
        algo->insert_step( ++step_num, LineSearch_name, line_search_full_step_step );
      }
      else {
        // Main line search step
        algo->insert_step( ++step_num, LineSearch_name, line_search_step );
        // Insert presteps
        Algorithm::poss_type
          pre_step_i = 0;
        // (.-?) LineSearchFullStep
        //algo->insert_assoc_step(
        //  step_num
        //  ,IterationPack::PRE_STEP
        //  ,++pre_step_i
        //  ,"LineSearchFullStep"
        //  ,line_search_full_step_step
        //  );
        // (.-?) MeritFunc_PenaltyPramUpdate
        if(merit_func_penalty_param_update_step.get()) {
          algo->insert_assoc_step(
            step_num
            ,IterationPack::PRE_STEP
            ,++pre_step_i
            ,"MeritFunc_PenaltyParamUpdate"
            ,merit_func_penalty_param_update_step
            );
        }
      }

      // PostProcessBarrierLineSearch_Step
      algo->insert_step( ++step_num, "PostProcessBarrierLineSearch_Step", postprocess_barrier_linesearch_step );

      // CheckConvergence
      //algo->insert_step( ++step_num, CheckConvergence_name, check_convergence_step );

    }
    else {
      TEST_FOR_EXCEPT(true); // Error, this should not ever be called!
    }
  }
  
}

void NLPAlgoConfigIP::init_algo(NLPAlgoInterface* _algo)
{
  using Teuchos::dyn_cast;

  TEST_FOR_EXCEPTION(
    _algo == NULL, std::invalid_argument
    ,"NLPAlgoConfigIP::init_algo(_algo) : Error, "
    "_algo can not be NULL" );

  NLPAlgo           &algo    = dyn_cast<NLPAlgo>(*_algo);
  NLPAlgoState      &state   = algo.rsqp_state();
  NLP               &nlp     = algo.nlp();

  algo.max_iter( algo.algo_cntr().max_iter() );
  algo.max_run_time( algo.algo_cntr().max_run_time() );

  // Reset the iteration count to zero
  state.k(0);

  // Get organized output of vectors and matrices even if setw is not used by Step objects.
  algo.track().journal_out()
    << std::setprecision(algo.algo_cntr().journal_print_digits())
    << std::scientific;

  // set the first step
  algo.do_step_first(1);

  // The rest of the algorithm should initialize itself
}

// private

void NLPAlgoConfigIP::readin_options(
    const OptionsFromStreamPack::OptionsFromStream     &options
  , SOptionValues                                      *ov
  , std::ostream                                       *trase_out
  )
{
  namespace ofsp = OptionsFromStreamPack;
  using   ofsp::OptionsFromStream;
  typedef   OptionsFromStream::options_group_t    options_group_t;
  using   ofsp::StringToIntMap;
  using   ofsp::StringToBool;

  TEST_FOR_EXCEPT( !( ov ) ); // only a local class error

  // Get the options group for "NLPAlgoConfigIP"
  const std::string opt_grp_name = "NLPAlgoConfigIP";
  const OptionsFromStream::options_group_t optgrp = options.options_group( opt_grp_name );
  if( OptionsFromStream::options_group_exists( optgrp ) ) {

    // Define map for options group "IpConfig".
    const int num_opts = 11;
    enum EIpConfig {
      MAX_BASIS_COND_CHANGE_FRAC
      ,EXACT_REDUCED_HESSIAN
      ,QUASI_NEWTON
      ,NUM_LBFGS_UPDATES_STORED
      ,LBFGS_AUTO_SCALING
      ,HESSIAN_INITIALIZATION
      ,QP_SOLVER
      ,REINIT_HESSIAN_ON_QP_FAIL
      ,LINE_SEARCH_METHOD
      ,MERIT_FUNCTION_TYPE
      ,L1_PENALTY_PARAM_UPDATE
    };
    const char* SIpConfig[num_opts] = {
      "max_basis_cond_change_frac"
      ,"exact_reduced_hessian"
      ,"quasi_newton"
      ,"num_lbfgs_updates_stored"
      ,"lbfgs_auto_scaling"
      ,"hessian_initialization"
      ,"qp_solver"
      ,"reinit_hessian_on_qp_fail"
      ,"line_search_method"
      ,"merit_function_type"
      ,"l1_penalty_parameter_update"
    };
    StringToIntMap  map(  opt_grp_name, num_opts, SIpConfig );

    options_group_t::const_iterator itr = optgrp.begin();
    for( ; itr != optgrp.end(); ++itr ) {
      switch( (EIpConfig)map( ofsp::option_name(itr) ) ) {
        case MAX_BASIS_COND_CHANGE_FRAC:
          ov->max_basis_cond_change_frac_ = std::atof( ofsp::option_value(itr).c_str() );
          break;
        case EXACT_REDUCED_HESSIAN:
          ov->exact_reduced_hessian_ = StringToBool( "exact_reduced_hessian", ofsp::option_value(itr).c_str() );
          break;
        case QUASI_NEWTON:
        {
          const std::string &opt_val = ofsp::option_value(itr);
          if( opt_val == "AUTO" )
            ov->quasi_newton_ = QN_AUTO;
          else if( opt_val == "BFGS" )
            ov->quasi_newton_ = QN_BFGS;
          else if( opt_val == "PBFGS" )
            ov->quasi_newton_ = QN_PBFGS;
          else if( opt_val == "LBFGS" )
            ov->quasi_newton_ = QN_LBFGS;
          else if( opt_val == "LPBFGS" )
            ov->quasi_newton_ = QN_LPBFGS;
          else
            TEST_FOR_EXCEPTION(
              true, std::invalid_argument
              ,"NLPAlgoConfigIP::readin_options(...) : "
              "Error, incorrect value for \"quasi_newton\" "
              ", Only options of BFGS, PBFGS"
              ", LBFGS, LPBFGS and AUTO are avalible."
              );
          break;
        }
        case NUM_LBFGS_UPDATES_STORED:
          ov->num_lbfgs_updates_stored_ = std::atoi( ofsp::option_value(itr).c_str() );
          break;
        case LBFGS_AUTO_SCALING:
          ov->lbfgs_auto_scaling_
            = StringToBool( "lbfgs_auto_scaling", ofsp::option_value(itr).c_str() );
          break;
        case HESSIAN_INITIALIZATION:
        {
          const std::string &opt_val = ofsp::option_value(itr);
          if( opt_val == "IDENTITY" )
            ov->hessian_initialization_ = INIT_HESS_IDENTITY;
          else if( opt_val == "FINITE_DIFF_SCALE_IDENTITY" )
            ov->hessian_initialization_ = INIT_HESS_FIN_DIFF_SCALE_IDENTITY;
          else if( opt_val == "FINITE_DIFF_DIAGONAL" )
            ov->hessian_initialization_ = INIT_HESS_FIN_DIFF_SCALE_DIAGONAL;
          else if( opt_val == "FINITE_DIFF_DIAGONAL_ABS" )
            ov->hessian_initialization_ = INIT_HESS_FIN_DIFF_SCALE_DIAGONAL_ABS;
          else if( opt_val == "AUTO" )
            ov->hessian_initialization_ = INIT_HESS_AUTO;
          else
            TEST_FOR_EXCEPTION(
              true, std::invalid_argument
              ,"NLPAlgoConfigIP::readin_options(...) : "
              "Error, incorrect value for \"hessian_initialization\" "
              ", Only options of IDENTITY, FINITE_DIFF_SCALE_IDENTITY,"
              " FINITE_DIFF_DIAGONAL, FINITE_DIFF_DIAGONAL_ABS and AUTO"
              " are available"  );
          break;
        }
        case QP_SOLVER:
        {
          const std::string &qp_solver = ofsp::option_value(itr);
          if( qp_solver == "AUTO" ) {
            ov->qp_solver_type_ = QP_AUTO;
          } else if( qp_solver == "QPSOL" ) {
            ov->qp_solver_type_ = QP_QPSOL;
          } else if( qp_solver == "QPOPT" ) {
#ifdef CONSTRAINED_OPTIMIZATION_PACK_USE_QPOPT
            ov->qp_solver_type_ = QP_QPOPT;
#else
            TEST_FOR_EXCEPTION(
              true, std::invalid_argument
              ,"NLPAlgoConfigIP::readin_options(...) : QPOPT is not supported,"
              " must define CONSTRAINED_OPTIMIZATION_PACK_USE_QPOPT!" );
#endif
          } else if( qp_solver == "QPKWIK" ) {
            ov->qp_solver_type_ = QP_QPKWIK;
          } else if( qp_solver == "QPSCHUR" ) {
            ov->qp_solver_type_ = QP_QPSCHUR;
          } else {
            TEST_FOR_EXCEPTION(
              true, std::invalid_argument
              ,"NLPAlgoConfigIP::readin_options(...) : "
              "Error, incorrect value for \"qp_solver\" "
              "Only qp solvers QPOPT, QPSOL, QPKWIK, QPSCHUR and AUTO are avalible."  );
          }
          break;
        }
        case REINIT_HESSIAN_ON_QP_FAIL:
          ov->reinit_hessian_on_qp_fail_ = StringToBool( "reinit_hessian_on_qp_fail", ofsp::option_value(itr).c_str() );
          break;
        case LINE_SEARCH_METHOD:
        {
          const std::string &option = ofsp::option_value(itr);
          if( option == "NONE" ) {
            ov->line_search_method_ = LINE_SEARCH_NONE;
          } else if( option == "DIRECT" ) {
            ov->line_search_method_ = LINE_SEARCH_DIRECT;
          } else if( option == "2ND_ORDER_CORRECT" ) {
            ov->line_search_method_ = LINE_SEARCH_2ND_ORDER_CORRECT;
          } else if( option == "WATCHDOG" ) {
            ov->line_search_method_ = LINE_SEARCH_WATCHDOG;
          } else if( option == "AUTO" ) {
            ov->line_search_method_ = LINE_SEARCH_AUTO;
          } else if( option == "FILTER" ) {
            ov->line_search_method_ = LINE_SEARCH_FILTER;
          } else {
            TEST_FOR_EXCEPTION(
              true, std::invalid_argument
              ,"NLPAlgoConfigIP::readin_options(...) : "
              "Error, incorrect value for \"line_search_method\".\n"
              "Only the options NONE, DIRECT, 2ND_ORDER_CORRECT, FILTER, WATCHDOG "
              "and AUTO are avalible." );
          }
          break;
        }
        case MERIT_FUNCTION_TYPE:
        {
          const std::string &option = ofsp::option_value(itr);
          if( option == "L1" )
            ov->merit_function_type_ = MERIT_FUNC_L1;
          else if( option == "MODIFIED_L1" )
            ov->merit_function_type_ = MERIT_FUNC_MOD_L1;
          else if( option == "MODIFIED_L1_INCR" )
            ov->merit_function_type_ = MERIT_FUNC_MOD_L1_INCR;
          else if( option == "AUTO" )
            ov->merit_function_type_ = MERIT_FUNC_AUTO;
          else
            TEST_FOR_EXCEPTION(
              true, std::invalid_argument
              ,"NLPAlgoConfigIP::readin_options(...) : "
              "Error, incorrect value for \"merit_function_type\".\n"
              "Only the options L1, MODIFIED_L1, MODIFIED_L1_INCR "
              "and AUTO are avalible." );
          break;
        }
        case L1_PENALTY_PARAM_UPDATE:
        {
          const std::string &option = ofsp::option_value(itr);
          if( option == "WITH_MULT" )
            ov->l1_penalty_param_update_
              = L1_PENALTY_PARAM_WITH_MULT;
          else if( option == "MULT_FREE" )
            ov->l1_penalty_param_update_
              = L1_PENALTY_PARAM_MULT_FREE;
          else if( option == "AUTO" )
            ov->l1_penalty_param_update_
              = L1_PENALTY_PARAM_AUTO;
          else
            TEST_FOR_EXCEPTION(
              true, std::invalid_argument
              ,"NLPAlgoConfigIP::readin_options(...) : "
              "Error, incorrect value for \"l1_penalty_param_update\".\n"
              "Only the options WITH_MULT, MULT_FREE and AUTO"
              "are avalible."  );
          break;
        }
        default:
          TEST_FOR_EXCEPT(true);  // this would be a local programming error only.
      }
    }
  }
  else {
    if(trase_out)
      *trase_out
        << "\n\n*** Warning!  The options group \"NLPAlgoConfigIP\" was not found.\n"
        << "Using a default set of options instead ... \n";
  }
}

//
// This is where some of the default options are set and the user is alerted to what their
// value is.
//
void NLPAlgoConfigIP::set_default_options( 
  const SOptionValues     &uov
  ,SOptionValues          *cov
  ,std::ostream           *trase_out
  )
{
  if(trase_out)
    *trase_out
      << "\n*** Setting option defaults for options not set by the user or determined some other way ...\n";

  if( cov->max_basis_cond_change_frac_ < 0.0 &&  uov.max_basis_cond_change_frac_ < 0.0 ) {
    if(trase_out)
      *trase_out
        << "\nmax_basis_cond_change_frac < 0 : setting max_basis_cond_change_frac = 1e+4 \n";
    cov->max_basis_cond_change_frac_ = 1e+4;
  }
  else {
    cov->max_basis_cond_change_frac_ = uov.max_basis_cond_change_frac_;
  }
  cov->exact_reduced_hessian_ = uov.exact_reduced_hessian_;
  if( cov->quasi_newton_ == QN_AUTO && uov.quasi_newton_ == QN_AUTO ) {
    if(trase_out)
      *trase_out
        << "\nquasi_newton == AUTO: setting quasi_newton = BFGS\n";
    cov->quasi_newton_ = QN_BFGS;
  }
  else if(cov->quasi_newton_ == QN_AUTO) {
    cov->quasi_newton_ = uov.quasi_newton_;
  }
  if( cov->num_lbfgs_updates_stored_ < 0 && uov.num_lbfgs_updates_stored_ < 0 ) {
    if(trase_out)
      *trase_out
        << "\nnum_lbfgs_updates_stored < 0 : setting num_lbfgs_updates_stored = 10\n";
    cov->num_lbfgs_updates_stored_ = 10;
  }
  else if(cov->num_lbfgs_updates_stored_ < 0) {
    cov->num_lbfgs_updates_stored_ = uov.num_lbfgs_updates_stored_;
  }
  cov->lbfgs_auto_scaling_ = uov.lbfgs_auto_scaling_;
  if( cov->hessian_initialization_ == INIT_HESS_AUTO && uov.hessian_initialization_ == INIT_HESS_AUTO ) {
    if(trase_out)
      *trase_out
        << "\nhessian_initialization == AUTO: setting hessian_initialization = FINITE_DIFF_DIAGONAL_ABS\n";
    cov->hessian_initialization_ = INIT_HESS_FIN_DIFF_SCALE_DIAGONAL_ABS;
  }
  else if(cov->hessian_initialization_ == INIT_HESS_AUTO) {
    cov->hessian_initialization_ = uov.hessian_initialization_;
  }
  if( cov->qp_solver_type_ == QP_AUTO && uov.qp_solver_type_ == QP_AUTO ) {
    if(trase_out)
      *trase_out
        << "\nqp_solver_type == AUTO: setting qp_solver_type = QPSCHUR\n";
    cov->qp_solver_type_ = QP_QPSCHUR;
  }
  else if(cov->qp_solver_type_ == QP_AUTO) {
    cov->qp_solver_type_ = uov.qp_solver_type_;
  }
  cov->reinit_hessian_on_qp_fail_ = uov.reinit_hessian_on_qp_fail_;
  if( cov->line_search_method_ == LINE_SEARCH_AUTO && uov.line_search_method_ == LINE_SEARCH_AUTO ) {
    if(trase_out)
      *trase_out
        << "\nline_search_method == AUTO: setting line_search_method = FILTER\n";
    cov->line_search_method_ = LINE_SEARCH_FILTER;
  }
  else if(cov->line_search_method_ == LINE_SEARCH_AUTO) {
    cov->line_search_method_ = uov.line_search_method_;
  }
  if( cov->merit_function_type_ == MERIT_FUNC_AUTO && uov.merit_function_type_ == MERIT_FUNC_AUTO ) {
    if(trase_out)
      *trase_out
        << "\nmerit_function_type == AUTO: setting merit_function_type = MODIFIED_L1_INCR\n";
    cov->merit_function_type_ = MERIT_FUNC_MOD_L1_INCR;
  }
  else if(cov->merit_function_type_ == MERIT_FUNC_AUTO) {
    cov->merit_function_type_ = uov.merit_function_type_;
  }
  if( cov->l1_penalty_param_update_ == L1_PENALTY_PARAM_AUTO && uov.l1_penalty_param_update_ == L1_PENALTY_PARAM_AUTO ) {
    if(trase_out)
      *trase_out
        << "\nl1_penalty_param_update == AUTO: setting l1_penalty_param_update = MULT_FREE\n";
    cov->l1_penalty_param_update_ = L1_PENALTY_PARAM_MULT_FREE;
  }
  else if(cov->l1_penalty_param_update_ == L1_PENALTY_PARAM_AUTO) {
    cov->l1_penalty_param_update_ = uov.l1_penalty_param_update_;
  }
  if( cov->full_steps_after_k_ < 0 && uov.full_steps_after_k_ < 0 ) {
    if(trase_out)
      *trase_out
        << "\nfull_steps_after_k < 0 : the line search will never be turned off after so many iterations\n";
  }
  else {
    cov->full_steps_after_k_ = uov.full_steps_after_k_;
  }
  if(trase_out)
    *trase_out
      << "\n*** End setting default options\n";
}

} // end namespace MoochoPack