grid_function_util.h

Go to the documentation of this file.

/*
 * Copyright (c) 2010-2015:  G-CSC, Goethe University Frankfurt
 * Author: Andreas Vogel
 * 
 * This file is part of UG4.
 * 
 * UG4 is free software: you can redistribute it and/or modify it under the
 * terms of the GNU Lesser General Public License version 3 (as published by the
 * Free Software Foundation) with the following additional attribution
 * requirements (according to LGPL/GPL v3 §7):
 * 
 * (1) The following notice must be displayed in the Appropriate Legal Notices
 * of covered and combined works: "Based on UG4 (www.ug4.org/license)".
 * 
 * (2) The following notice must be displayed at a prominent place in the
 * terminal output of covered works: "Based on UG4 (www.ug4.org/license)".
 * 
 * (3) The following bibliography is recommended for citation and must be
 * preserved in all covered files:
 * "Reiter, S., Vogel, A., Heppner, I., Rupp, M., and Wittum, G. A massively
 *   parallel geometric multigrid solver on hierarchically distributed grids.
 *   Computing and visualization in science 16, 4 (2013), 151-164"
 * "Vogel, A., Reiter, S., Rupp, M., Nägel, A., and Wittum, G. UG4 -- a novel
 *   flexible software system for simulating pde based models on high performance
 *   computers. Computing and visualization in science 16, 4 (2013), 165-179"
 * 
 * This program 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.
 */
 
#ifndef __H__UG__LIB_DISC__FUNCTION_SPACE__GRID_FUNCTION_UTIL__
#define __H__UG__LIB_DISC__FUNCTION_SPACE__GRID_FUNCTION_UTIL__
 
#include <vector>
#include <string>
#include <cmath>  // for isinf, isnan
#include <boost/function.hpp>
 
 
#include "common/util/file_util.h"
 
#include "lib_algebra/cpu_algebra/sparsematrix_print.h"
#include "lib_algebra/operator/interface/matrix_operator.h"
#include "lib_algebra/operator/debug_writer.h"
#include "lib_algebra/operator/vector_writer.h"
#include "lib_algebra/common/matrixio/matrix_io_mtx.h"
#include "lib_algebra/common/connection_viewer_output.h"
#include "lib_algebra/common/csv_gnuplot_output.h"
#include "lib_disc/io/vtkoutput.h"
#include "lib_disc/spatial_disc/constraints/constraint_interface.h"
#include "lib_disc/dof_manager/dof_distribution.h"
#include "lib_disc/spatial_disc/user_data/user_data.h"
#include "lib_disc/common/groups_util.h"
#include "lib_disc/common/geometry_util.h"
#include "lib_disc/function_spaces/integrate.h"
#include "lib_grid/algorithms/debug_util.h"  // for ElementDebugInfo
#include "lib_grid/tools/periodic_boundary_manager.h"
 
#include "grid_function.h"
#include "dof_position_util.h"
 
#ifdef UG_PARALLEL
#include "pcl/pcl.h"
#endif
 
namespace ug {
 
#ifndef isnan
using boost::math::isnan;
#endif
 
#ifndef isinf
using boost::math::isinf;
#endif
 
 
template <typename TBaseElem, typename TGridFunction>
static void ScaleGFOnElems
(
  ConstSmartPtr<DoFDistribution> dd,
  SmartPtr<TGridFunction> vecOut,
  ConstSmartPtr<TGridFunction> vecIn,
  const std::vector<number>& vScale
)
{
  typename DoFDistribution::traits<TBaseElem>::const_iterator iter, iterEnd;
  std::vector<DoFIndex> vInd;
 
  try
  {
    // iterate all elements (including SHADOW_RIM_COPY!)
    iter = dd->template begin<TBaseElem>(SurfaceView::ALL);
    iterEnd = dd->template end<TBaseElem>(SurfaceView::ALL);
    for (; iter != iterEnd; ++iter)
    {
      for (size_t fi = 0; fi < dd->num_fct(); ++fi)
      {
        size_t nInd = dd->inner_dof_indices(*iter, fi, vInd);
 
        // remember multi indices
        for (size_t dof = 0; dof < nInd; ++dof)
          DoFRef(*vecOut, vInd[dof]) = vScale[fi] * DoFRef(*vecIn, vInd[dof]);
      }
    }
  }
  UG_CATCH_THROW("Error while scaling vector.")
}
 
 
template <typename TGridFunction>
void ScaleGF
(
  SmartPtr<TGridFunction> scaledVecOut,
  ConstSmartPtr<TGridFunction> vecIn,
  const std::vector<number>& scalingFactors
)
{
  // check that the correct numbers of scaling factors are given
  size_t n = scalingFactors.size();
  UG_COND_THROW(n != vecIn->num_fct(), "Number of scaling factors (" << n << ") "
      "does not match number of functions given in dimless vector (" << vecIn->num_fct() << ").");
 
  // check that input and output vectors have the same number of components and dofs
  UG_COND_THROW(n != scaledVecOut->num_fct(), "Input and output vectors do not have "
      "the same number of functions (" << n << " vs. " << scaledVecOut->num_fct() << ").");
  for (size_t fct = 0; fct < n; ++fct)
  {
    UG_COND_THROW(vecIn->num_dofs(fct) != scaledVecOut->num_dofs(fct),
        "Input and output vectors do not have the same number of DoFs for function " << fct
        << " (" << vecIn->num_dofs(fct) << " vs. " << scaledVecOut->num_dofs(fct) << ").");
  }
 
  ConstSmartPtr<DoFDistribution> dd = vecIn->dof_distribution();
 
  if (dd->max_dofs(VERTEX))
    ScaleGFOnElems<Vertex, TGridFunction>(dd, scaledVecOut, vecIn, scalingFactors);
  if (dd->max_dofs(EDGE))
    ScaleGFOnElems<Edge, TGridFunction>(dd, scaledVecOut, vecIn, scalingFactors);
  if (dd->max_dofs(FACE))
    ScaleGFOnElems<Face, TGridFunction>(dd, scaledVecOut, vecIn, scalingFactors);
  if (dd->max_dofs(VOLUME))
    ScaleGFOnElems<Volume, TGridFunction>(dd, scaledVecOut, vecIn, scalingFactors);
}
 
 
 
 
//  AverageComponent
 
template<typename TGridFunction, typename TBaseElem>
void SubtractValueFromComponent(SmartPtr<TGridFunction> spGF, size_t fct, number sub)
{
  typedef TGridFunction GF;
  typedef typename GF::template traits<TBaseElem>::const_iterator iter_type;
 
  iter_type iter = spGF->template begin<TBaseElem>();
  iter_type iterEnd = spGF->template end<TBaseElem>();
 
  PeriodicBoundaryManager* pbm = spGF->domain()->grid()->periodic_boundary_manager();
 
//  loop elems
  std::vector<DoFIndex> vMultInd;
  for(; iter != iterEnd; ++iter)
  {
  //  get element
    TBaseElem* elem = *iter;
 
  //  skip periodic ghosts
    if (pbm && pbm->is_slave(elem)) continue;
 
  //  get global indices
    spGF->inner_dof_indices(elem, fct, vMultInd);
 
  //  sum up value
    for(size_t i = 0; i < vMultInd.size(); ++i)
    {
      DoFRef(*spGF, vMultInd[i]) -= sub;
    }
  }
}
 
template<typename TGridFunction>
void AdjustMeanValue
(
  SmartPtr<TGridFunction> spGF, 
  const std::vector<std::string>& vCmp, 
  number mean 
)
{
  typedef TGridFunction GF;
  PROFILE_FUNC_GROUP("gmg");
 
  if(vCmp.empty())
    return;
 
  if(spGF.invalid())
    UG_THROW("AverageComponent: expects a valid GridFunction.");
 
  ConstSmartPtr<DoFDistributionInfo> ddinfo =
                spGF->approx_space()->dof_distribution_info();
 
//  compute integral of components
  const number area = Integral(1.0, spGF);
  std::vector<number> vIntegral(vCmp.size(), 0.0);
  for(size_t f = 0; f < vCmp.size(); f++){
    const size_t fct = spGF->fct_id_by_name(vCmp[f].c_str());
    vIntegral[f] = Integral(spGF, vCmp[f].c_str(), NULL, ddinfo->lfeid(fct).order());
  }
 
//  subtract value
  for(size_t f = 0; f < vCmp.size(); f++)
  {
    const number sub = (vIntegral[f] - mean) / area;
    const size_t fct = spGF->fct_id_by_name(vCmp[f].c_str());
 
    if(ddinfo->max_fct_dofs(fct, VERTEX)) SubtractValueFromComponent<GF, Vertex>(spGF, fct, sub);
    if(ddinfo->max_fct_dofs(fct, EDGE)) SubtractValueFromComponent<GF, Edge>(spGF, fct, sub);
    if(ddinfo->max_fct_dofs(fct, FACE)) SubtractValueFromComponent<GF, Face>(spGF, fct, sub);
    if(ddinfo->max_fct_dofs(fct, VOLUME)) SubtractValueFromComponent<GF, Volume>(spGF, fct, sub);
  }
}
 
template<typename TGridFunction>
void AdjustMeanValue(SmartPtr<TGridFunction> spGF, const std::vector<std::string>& vCmp)
{
  AdjustMeanValue(spGF, vCmp, 0.0);
}
 
template<typename TGridFunction>
void AdjustMeanValue(SmartPtr<TGridFunction> spGF, const std::string& fcts)
{
  AdjustMeanValue(spGF, TokenizeTrimString(fcts), 0.0);
}
 
template<typename TGridFunction>
void AdjustMeanValue(SmartPtr<TGridFunction> spGF, const std::string& fcts, number mean)
{
  AdjustMeanValue(spGF, TokenizeTrimString(fcts), mean);
}
 
//  Summing up components
 
template <typename TGridFunction, typename TBaseElem>
number SumGFValuesAt
(
  TGridFunction * u, 
  size_t fct 
)
{
  typedef typename TGridFunction::template traits<TBaseElem>::const_iterator t_elem_iter;
  
  std::vector<DoFIndex> ind;
  number sum (1);
  
  sum = 0;
//  Loop the elements in the subset
  for (t_elem_iter vi = u->template begin<TBaseElem> ();
        vi != u->template end<TBaseElem> (); ++vi)
  {
    TBaseElem * vert = *vi;
    
  //  indices at this element
    u->inner_dof_indices (vert, fct, ind);
    
    if (ind.size () != 1)
      UG_THROW ("SumGFValuesAt: The function must be scalar!");
  
  //  add the contribution
    sum += DoFRef (*u, ind [0]);
  }
  
#ifdef UG_PARALLEL
  pcl::ProcessCommunicator procComm;
  sum = procComm.allreduce (sum, PCL_RO_SUM);
#endif
  
  return sum;
}
 
template <typename TGridFunction, typename TBaseElem>
number SumGFValuesAt
(
  TGridFunction * u, 
  const char * fct_names 
)
{
//  Get the function index
  std::vector<std::string> vfctNames;
  TokenizeString (fct_names, vfctNames);
  if (vfctNames.size () != 1)
    UG_THROW ("SumGFValuesAt: Exactly one function name must be specified.");
  FunctionGroup fctGroup (u->function_pattern ());
  fctGroup.add (vfctNames [0]);
 
//  Compute the sum
  return SumGFValuesAt<TGridFunction, TBaseElem> (u, fctGroup[0]);
}
 
template <typename TGridFunction, typename TBaseElem>
number SumGFValuesAt
(
  TGridFunction * u, 
  size_t fct, 
  SubsetGroup & ssGroup 
)
{
  typedef typename TGridFunction::template traits<TBaseElem>::const_iterator t_elem_iter;
  
  std::vector<DoFIndex> ind;
  number sum (1);
  
//  Loop the selected subsets
  sum = 0;
  for (size_t i = 0; i < ssGroup.size (); i++)
  {
    int ssi = ssGroup [i];
    
  //  Loop the elements in the subset
    for (t_elem_iter vi = u->template begin<TBaseElem> (ssi);
          vi != u->template end<TBaseElem> (ssi); ++vi)
    {
        TBaseElem * vert = *vi;
        
    //  indices at this element
      u->inner_dof_indices (vert, fct, ind);
      
      if (ind.size () != 1)
        UG_THROW ("SumGFValuesAt: The function must be scalar!");
    
    //  add the contribution
      sum += DoFRef (*u, ind [0]);
    }
  }
  
#ifdef UG_PARALLEL
  pcl::ProcessCommunicator procComm;
  sum = procComm.allreduce (sum, PCL_RO_SUM);
#endif
  
  return sum;
}
 
template <typename TGridFunction, typename TBaseElem>
number SumGFValuesAt
(
  TGridFunction * u, 
  const char * fct_names, 
  const char * subset_names 
)
{
//  Get the function index
  std::vector<std::string> vfctNames;
  TokenizeString (fct_names, vfctNames);
  if (vfctNames.size () != 1)
    UG_THROW ("SumGFValuesAt: Exactly one function name must be specified.");
  FunctionGroup fctGroup (u->function_pattern ());
  fctGroup.add (vfctNames [0]);
 
//  Get the subset indices
  std::vector<std::string> vssNames;
  TokenizeString (subset_names, vssNames);
  SubsetGroup ssGroup (u->domain()->subset_handler ());
  ssGroup.add (vssNames);
  
//  Compute the sum
  return SumGFValuesAt<TGridFunction, TBaseElem> (u, fctGroup[0], ssGroup);
}
 
//  Checking components for nan and inf
 
template <typename TGridFunction, typename TBaseElem>
bool CheckGFforNaN
(
  const TGridFunction * u, 
  size_t fct 
)
{
  typedef typename TGridFunction::template traits<TBaseElem>::const_iterator t_elem_iter;
  
  std::vector<DoFIndex> ind;
  
//  Loop the elements in the subset
  for (t_elem_iter vi = u->template begin<TBaseElem> ();
        vi != u->template end<TBaseElem> (); ++vi)
  {
    TBaseElem * vert = *vi;
    
  //  indices at this element
    u->inner_dof_indices (vert, fct, ind);
    
    if (ind.size () != 1)
      UG_THROW ("CheckGFforNaN: The function must be scalar!");
  
  //  check the value
    number value = DoFRef (*u, ind [0]);
    if (isnan (value))
    {
      int si = u->domain()->subset_handler()->get_subset_index (vert);
      UG_LOG ("nan at index " << ind [0] << ", grid data idx " << vert->grid_data_index ());
      if (si >= 0)
        UG_LOG (", subset " << u->domain()->subset_handler()->get_subset_name (si))
      UG_LOG ('\n');
      return true;
    }
    if (isinf (value))
    {
      int si = u->domain()->subset_handler()->get_subset_index (vert);
      UG_LOG ("inf at index " << ind [0] << ", grid data idx " << vert->grid_data_index ());
      if (si >= 0)
        UG_LOG (", subset " << u->domain()->subset_handler()->get_subset_name (si))
      UG_LOG ('\n');
      return true;
    }
  }
  
  return false;
}
 
template <typename TGridFunction, typename TBaseElem>
bool CheckGFforNaN
(
  const TGridFunction * u, 
  const char * fct_names 
)
{
//  Get the function index
  std::vector<std::string> vfctNames;
  TokenizeString (fct_names, vfctNames);
  FunctionGroup fctGroup (u->function_pattern ());
  for (size_t i = 0; i < vfctNames.size (); i++)
    fctGroup.add (vfctNames [i]);
 
//  Check the functions
  bool result = false;
  for (size_t i = 0; i < vfctNames.size (); i++)
  {
    UG_LOG ("Checking " << vfctNames[i] << " ... ");
    if (CheckGFforNaN<TGridFunction, TBaseElem> (u, fctGroup[i]))
      result = true;
    else
      UG_LOG ("OK\n");
  }
  return result;
}
 
template <typename TGridFunction, typename TBaseElem>
bool CheckGFforNaN
(
  const TGridFunction * u, 
  const FunctionGroup & fctGroup 
)
{
//  Check the functions
  bool result = false;
  for (size_t i = 0; i < fctGroup.size (); i++)
  {
    UG_LOG ("Checking fct #" << i << " ... ");
    if (CheckGFforNaN<TGridFunction, TBaseElem> (u, fctGroup[i]))
      result = true;
    else
      UG_LOG ("OK\n");
  }
  return result;
}
 
// //////////////////////////////////////////////////////////////////////////////
//  Check values are within bounds
// //////////////////////////////////////////////////////////////////////////////
template <typename TGridFunction, typename TBaseElem>
bool CheckGFValuesWithinBounds
(
  ConstSmartPtr<TGridFunction> u, 
  size_t cmp,  
  number lowerBnd,
  number upperBnd
)
{
  typedef typename TGridFunction::template traits<TBaseElem>::const_iterator elem_it;
 
  bool ret = true;
 
  // loop the elements in the subset
  std::vector<DoFIndex> vDI;
  elem_it it = u->template begin<TBaseElem>();
  elem_it itEnd = u->template end<TBaseElem>();
  for (; it != itEnd; ++it)
  {
    TBaseElem* elem = *it;
 
    // loop indices at this element
    const size_t nInd = u->inner_dof_indices(elem, cmp, vDI, true);
    for (size_t i = 0; i < nInd; ++i)
    {
      const number& val = DoFRef(*u, vDI[i]);
      if (val < lowerBnd || val > upperBnd)
      {
        UG_LOG_ALL_PROCS("Function value for component " << cmp << " (" << val << ") "
          "is outside the specified range [" << lowerBnd << ", " << upperBnd << "] "
          "at " << ElementDebugInfo(*u->domain()->grid(), elem) << std::endl);
        ret = false;
      }
    }
  }
 
#ifdef UG_PARALLEL
  if (pcl::NumProcs() > 1)
  {
    pcl::ProcessCommunicator pc;
    int retInt = ret;
    ret = pc.allreduce(retInt, PCL_RO_BAND);
  }
#endif
 
  return ret;
}
 
template <typename TGridFunction>
bool CheckGFValuesWithinBounds
(
  ConstSmartPtr<TGridFunction> u, 
  size_t cmp,  
  number lowerBnd,
  number upperBnd
)
{
  bool ret = true;
  if (u->max_fct_dofs(cmp, 0))
    ret = ret && CheckGFValuesWithinBounds<TGridFunction, Vertex>(u, cmp, lowerBnd, upperBnd);
  if (u->max_fct_dofs(cmp, 1))
    ret = ret && CheckGFValuesWithinBounds<TGridFunction, Edge>(u, cmp, lowerBnd, upperBnd);
  if (u->max_fct_dofs(cmp, 2))
    ret = ret && CheckGFValuesWithinBounds<TGridFunction, Face>(u, cmp, lowerBnd, upperBnd);
  if (u->max_fct_dofs(cmp, 3))
    ret = ret && CheckGFValuesWithinBounds<TGridFunction, Volume>(u, cmp, lowerBnd, upperBnd);
 
  return ret;
}
 
template <typename TGridFunction>
bool CheckGFValuesWithinBounds
(
  ConstSmartPtr<TGridFunction> u, 
  const char* fctNames,  
  number lowerBnd,
  number upperBnd
)
{
  std::vector<std::string> vFctNames;
  TokenizeTrimString (fctNames, vFctNames);
  FunctionGroup fctGroup (u->function_pattern());
  const size_t nFct = vFctNames.size();
  for (size_t f = 0; f < nFct; ++f)
  {
    try {fctGroup.add(vFctNames[f]);}
    UG_CATCH_THROW("Could not add function " << vFctNames[f] << " to function group.");
  }
 
  bool ret = true;
  const size_t fctGrpSz = fctGroup.size();
  for (size_t f = 0; f < fctGrpSz; ++f)
    ret = ret && CheckGFValuesWithinBounds(u, fctGroup[f], lowerBnd, upperBnd);
 
  return ret;
}
 
 
 
 
//  Writing algebra
 
template<typename TDomain>
void WriteAlgebraIndices(std::string name, ConstSmartPtr<TDomain> domain,  ConstSmartPtr<DoFDistribution> dd)
{
  PROFILE_FUNC_GROUP("debug");
 
#ifdef UG_PARALLEL
  name = ConnectionViewer::GetParallelName(name, dd->layouts()->proc_comm());
#endif
 
  std::vector<size_t> fctIndex;
  std::vector<std::string> fctNames;
 
  ExtractAlgebraIndices<TDomain>(domain, dd, fctIndex);
 
  size_t numFct = dd->num_fct();
  if(numFct <= 1) return;
 
  fctNames.resize(numFct);
  for(size_t i=0; i<numFct; i++)
    fctNames[i] = dd->name(i);
 
  name.append(".indices");
  std::fstream file(name.c_str(), std::ios::out);
 
  //std::cout << "name is " << name << "\n";
  file << "NUMDOF " << fctNames.size() << "\n";
  for(size_t i=0; i<numFct; i++)
    file << fctNames[i] << "\n";
 
  for(size_t i=0; i<fctIndex.size(); i++)
    file << fctIndex[i] << "\n";
}
 
template<class TFunction>
void WriteMatrixToConnectionViewer(const char *filename,
    const typename TFunction::algebra_type::matrix_type &A,
    const TFunction &u) {
 
  PROFILE_FUNC_GROUP("debug");
//  check name
  if ( !FileTypeIs( filename, ".mat") ) {
    UG_THROW( "Only '.mat' format supported for matrices, but"
              " filename is '" << filename << "'." );
  }
 
//  position array
  const static int dim = TFunction::domain_type::dim;
  std::vector<MathVector<dim> > vPos;
  ExtractPositions(u, vPos);
 
//  write matrix
  if(vPos.empty())
    ConnectionViewer::WriteMatrixPar( filename, A, (MathVector<dim>*)NULL, dim );
  else
    ConnectionViewer::WriteMatrixPar( filename, A, &vPos[0], dim );
 
  WriteAlgebraIndices(filename, u.domain(),u.dof_distribution());
 
}
 
template<typename TGridFunction>
void SaveMatrixForConnectionViewer(
    TGridFunction& u,
    MatrixOperator<typename TGridFunction::algebra_type::matrix_type,
        typename TGridFunction::vector_type>& A,
    const char* filename) {
  PROFILE_FUNC_GROUP("debug");
//  forward
  WriteMatrixToConnectionViewer(filename, A.get_matrix(), u);
}
 
// TODO extend MatrixIO to support other than CPUAlgebra
inline void SaveMatrixToMTX( const char *filename,
    MatrixOperator< CPUAlgebra::matrix_type, CPUAlgebra::vector_type >& A,
    std::string comment="%Generated with ug4." ) {
  PROFILE_FUNC_GROUP("debug");
 
  // check name
  if ( !FileTypeIs( filename, ".mtx" ) ) {
    UG_THROW( "Please use '.mtx' as file extension for MatrixMarket exchange files."
              " (Filename is '" << filename << "')" );
  }
  
  // automatically detect the file mode to use
  MatrixIO::OpenMode openMode = (FileExists( filename )) ? MatrixIO::EXISTING : MatrixIO::NEW;
  
  // create MatrixIO object for handling the export
  MatrixIOMtx mtx( filename, openMode );
  mtx.write_from( A.get_matrix(), comment );
}
 
template<class TFunction>
void WriteVectorToConnectionViewer(const char *filename,
    const typename TFunction::algebra_type::vector_type &b,
    const TFunction &u,
    const typename TFunction::algebra_type::vector_type *pCompareVec = NULL) {
  PROFILE_FUNC_GROUP("debug");
//  check name
  if ( !FileTypeIs( filename, ".vec") ) {
    UG_THROW( "Only '.vec' format supported for vectors, but"
              " filename is '" << filename << "'." );
  }
 
//  get positions of vertices
  const static int dim = TFunction::domain_type::dim;
  std::vector<MathVector<dim> > vPos;
  ExtractPositions(u, vPos);
 
//  write vector
  ConnectionViewer::WriteVectorPar( filename, b, &vPos[0], dim, pCompareVec);
  WriteAlgebraIndices(filename, u.domain(),u.dof_distribution());
}
 
template<class TFunction>
void WriteVectorToConnectionViewer(
    const char *filename,
    const typename TFunction::algebra_type::matrix_type &A,
    const typename TFunction::algebra_type::vector_type &b,
    const TFunction &u,
    const typename TFunction::algebra_type::vector_type *pCompareVec = NULL) {
  PROFILE_FUNC_GROUP("debug");
//  get dimension
  const static int dim = TFunction::domain_type::dim;
 
//  check name
  if ( !FileTypeIs( filename, ".vec") ) {
    UG_THROW( "Only '.vec' format supported for vectors." );
  }
 
//  get positions of vertices
  std::vector<MathVector<dim> > positions;
  ExtractPositions(u, positions);
 
//  write vector
  ConnectionViewer::WriteVectorPar( filename, A, b, &positions[0], dim, pCompareVec );
}
 
template<typename TGridFunction>
void SaveVectorForConnectionViewer(TGridFunction& b, const char* filename) {
  WriteVectorToConnectionViewer(filename, b, b);
}
 
template<typename TGridFunction>
void SaveVectorDiffForConnectionViewer(TGridFunction& b, TGridFunction& bCompare, const char* filename) {
  WriteVectorToConnectionViewer(filename, b, b, &bCompare);
}
 
template<typename TGridFunction>
void SaveVectorForConnectionViewer(
    TGridFunction& u,
    MatrixOperator<typename TGridFunction::algebra_type::matrix_type,
        typename TGridFunction::vector_type>& A,
    const char* filename) {
  WriteVectorToConnectionViewer(filename, A.get_matrix(), u, u);
}
 
template<typename TGridFunction>
void SaveVectorDiffForConnectionViewer(
    TGridFunction& u,
    TGridFunction& compareVec,
    MatrixOperator<typename TGridFunction::algebra_type::matrix_type,
        typename TGridFunction::vector_type>& A,
    const char* filename) {
//  forward
  WriteVectorToConnectionViewer(filename, A.get_matrix(), u, u, &compareVec);
}
 
// from connection_viewer_input.h
// with additional checks
template<typename vector_type>
bool ReadVector(std::string filename, vector_type &vec,int dim)
{
    Progress p;
  std::cout << " Reading std::vector from " <<  filename << "... ";
  std::fstream matfile(filename.c_str(), std::ios::in);
  if(matfile.is_open() == false) { std::cout << "failed.\n"; return false; }
 
  int version=-1, dimension=-1, gridsize;
 
  matfile >> version;
  matfile >> dimension;
  matfile >> gridsize;
 
  assert(version == 1);
  assert(dimension == dim);
  // todo check positions and not just size
  assert(gridsize == (int)vec.size());
  
 
  PROGRESS_START(prog, gridsize*2, "ReadVector " << dimension << "d from " << filename << " , " << gridsize << " x " << gridsize);
  for(int i=0; i<gridsize; i++)
  {
    if(i%100) { PROGRESS_UPDATE(prog, i); }
    if(matfile.eof())
    {
      std::cout << " failed.\n";
      assert(0);
      return false;
    }
    double x, y, z;
    matfile >> x >> y;
    if(dimension==3) matfile >> z;
  }
 
  int printStringsInWindow;
  matfile >> printStringsInWindow;
 
  // vec.resize(gridsize);
  bool bEOF = matfile.eof();
  while(!bEOF)
  {
    int from, to; double value;
    char c = matfile.peek();
    if(c == -1 || c == 'c' || c == 'v' || matfile.eof())
      break;
 
    matfile >> from >> to >> value;
    assert(from == to);
    vec[from] = value;
    if(from%100) { PROGRESS_UPDATE(prog, from); }
    bEOF = matfile.eof();
  }
  return true;
}
 
// load vector that has been saved in connection viewer format and write it
// into grid function
template<typename TGridFunction>
void LoadVector(TGridFunction& u,const char* filename){
  PROFILE_FUNC();
  typename TGridFunction::algebra_type::vector_type b;
  b.resize(u.num_indices());
  ReadVector(filename,b,TGridFunction::dim);
  u.assign(b);
}
 
// Same as before, but for comma separated value (CSV)
template<class TFunction>
void WriteVectorCSV(const char *filename,
    const typename TFunction::algebra_type::vector_type &b,
    const TFunction &u) {
  PROFILE_FUNC_GROUP("debug");
//  get dimension
  const static int dim = TFunction::domain_type::dim;
 
//  check name
  if ( !FileTypeIs( filename, ".csv") ) {
    UG_THROW( "Only '.csv' format supported for vectors, but"
              " filename is '" << filename << "'." );
  }
 
//  extended filename
//  add p000X extension in parallel
#ifdef UG_PARALLEL
  std::string name(filename);
  size_t iExtPos = name.find_last_of(".");
  name.resize(iExtPos);
  int rank = pcl::ProcRank();
  char ext[20]; snprintf(ext, 20, "_p%05d.csv", rank);
  name.append(ext);
#endif
 
//  get positions of vertices
  std::vector<MathVector<dim> > positions;
  ExtractPositions(u, positions);
 
//  write vector
  WriteVectorCSV( filename, b, &positions[0], dim );
}
 
template<typename TGridFunction>
void SaveVectorCSV(TGridFunction& b, const char* filename) {
  PROFILE_FUNC();
  WriteVectorCSV(filename, b, b);
}
 
template<typename TDomain, typename TAlgebra>
number AverageFunctionDifference(
    SmartPtr< GridFunction<TDomain, TAlgebra> > spGridFct,
    std::string subset, std::string fct1, std::string fct2 )
{
  PROFILE_FUNC();
  // get subset index
  size_t subSetID = spGridFct->subset_id_by_name( subset.c_str() );
 
  // get function indices
  size_t fct1ID = spGridFct->fct_id_by_name( fct1.c_str() );
  size_t fct2ID = spGridFct->fct_id_by_name( fct2.c_str() );
 
  // create space for sum of difference
  number sum = 0.0;
  size_t numElements = 0;
 
  // loop over all vertices in given subset and compare values of fct1 and fct2
  typedef typename GridFunction<TDomain, TAlgebra>::template traits<Vertex>::const_iterator gridFctIterator;
  for( gridFctIterator iter = spGridFct->template begin<Vertex>((int)subSetID); 
         iter != spGridFct->template end<Vertex>((int)subSetID); ++iter ) {
    // get dof_indices for the two functions on given subset
    std::vector< DoFIndex > indFct1, indFct2;
    spGridFct->template dof_indices<Vertex>( *iter, fct1ID, indFct1 );
    spGridFct->template dof_indices<Vertex>( *iter, fct2ID, indFct2 );
 
    // calculate the difference between the two functions at this grid point
    sum += DoFRef( *spGridFct, indFct1[0] ) - DoFRef( *spGridFct, indFct2[0] );
    numElements++;
  }
  
  // return overal arithmetic average of differences
  return sum / numElements;
}
 
 
//  Grid Function Debug Writer
 
template<typename TDomain, typename TAlgebra>
class GridFunctionDebugWriter: public IDebugWriter<TAlgebra>
{
  static const int dim = TDomain::dim;
 
public:
  typedef TAlgebra algebra_type;
 
  typedef typename algebra_type::vector_type vector_type;
 
  typedef typename algebra_type::matrix_type matrix_type;
 
  typedef ApproximationSpace<TDomain> approximation_space_type;
 
  typedef IDebugWriter<TAlgebra> super;
  using super::get_base_dir;
 
public:
  GridFunctionDebugWriter(
      SmartPtr<ApproximationSpace<TDomain> > spApproxSpace) :
      m_spApproxSpace(spApproxSpace), bConnViewerOut(
          true), bConnViewerIndices(false), bVTKOut(true), m_printConsistent(true)
  {
    IVectorDebugWriter<vector_type>::m_currentDim = dim;
    reset();
  }
 
  virtual ~GridFunctionDebugWriter() {};
 
  void set_grid_level(const GridLevel& gridLevel) {
    m_glFrom = gridLevel; m_glTo = gridLevel;
  }
 
  void set_grid_levels(const GridLevel& glFrom, const GridLevel& glTo) {
    m_glFrom = glFrom; m_glTo = glTo;
  }
 
  GridLevel grid_level() const {return m_glFrom;}
 
  void reset() {
    set_grid_level(GridLevel(GridLevel::TOP, GridLevel::SURFACE));
  }
 
 
  void set_vtk_output(bool b) {bVTKOut = b;}
 
  void set_conn_viewer_output(bool b) {bConnViewerOut = b;}
 
  void set_conn_viewer_indices(bool b) { bConnViewerIndices = b; }
 
  void set_print_consistent(bool b) {m_printConsistent = b;}
 
  virtual void write_vector(const vector_type& vec, const char* filename) {
    //  write to conn viewer
    if (bConnViewerOut)
      write_vector_to_conn_viewer(vec, filename);
 
    //  write to vtk
    if (bVTKOut)
      write_vector_to_vtk(vec, filename);
  }
 
  virtual void update_positions()
  {
    extract_positions(m_glFrom);
  }
 
  virtual void write_matrix(const matrix_type& mat, const char* filename) {
    PROFILE_FUNC_GROUP("debug");
    //  something to do ?
    if (!bConnViewerOut)
      return;
 
    update_positions();
 
    std::string name;
    this->compose_file_path (name);
    name += filename;
 
    if ( !FileTypeIs( filename, ".mat" ) ) {
      UG_THROW( "Only '.mat' format supported for matrices, but"
                " filename is '" << filename << "'." );
    }
 
    //  write to file
    write_algebra_indices_CV(name);
    if(m_glFrom == m_glTo){
      if(mat.num_rows() != mat.num_cols())
        UG_THROW("DebugWriter: grid level the same, but non-square matrix.");
 
      const std::vector<MathVector<dim> >& vPos = this->template get_positions<dim>();
      if(vPos.empty())
        ConnectionViewer::WriteMatrixPar(name.c_str(), mat,(MathVector<dim>*)NULL, dim);
      else
        ConnectionViewer::WriteMatrixPar(name.c_str(), mat, &vPos[0], dim);
    }
    else{
      const std::vector<MathVector<dim> >& vFromPos = this->template get_positions<dim>();
 
      std::vector<MathVector<dim> > vToPos;
      ExtractPositions<TDomain>(m_spApproxSpace->domain(),
                                m_spApproxSpace->dof_distribution(m_glTo),
                                vToPos);
 
      if(mat.num_cols() == vFromPos.size() && mat.num_rows() == vToPos.size())
      {
        ConnectionViewer::WriteMatrixPar(name, mat, vFromPos, vToPos, dim);
      }
      else{
        UG_THROW("GridFunctionDebugWriter: Wrong size of matrix for writing"
            "matrix ("<<m_glTo<<" x "<<m_glFrom<<"), that would be a ("
            <<vToPos.size()<<" x "<<vFromPos.size()<<") matrix. But "
            "passed matrix of size ("<<mat.num_rows()<<" x "
            <<mat.num_cols()<<").");
      }
    }
  }
 
protected:
 
  void write_algebra_indices_CV(std::string name)
  {
    if(bConnViewerIndices)
      WriteAlgebraIndices<TDomain>(name, m_spApproxSpace->domain(), m_spApproxSpace->dof_distribution(m_glFrom));
  }
 
  void extract_positions(const GridLevel& gridLevel) {
    //  extract positions for this grid function
    std::vector<MathVector<dim> >& vPos =
        this->template positions<dim>();
 
    vPos.clear();
    ExtractPositions<TDomain>(
        m_spApproxSpace->domain(),
        m_spApproxSpace->dof_distribution(gridLevel),
        vPos);
  }
 
  virtual void write_vector_to_conn_viewer(const vector_type& vec,
      const char* filename)
  {
    PROFILE_FUNC_GROUP("debug");
    update_positions();
 
    std::string name;
    this->compose_file_path (name);
    name += filename;
 
    if ( !FileTypeIs( filename, ".vec" ) ) {
      UG_THROW( "Only '.vec' format supported for vectors, but"
                " filename is '" << name << "'.");
    }
 
    //  write
    write_algebra_indices_CV(filename);
    const std::vector<MathVector<dim> >& vPos =
        this->template get_positions<dim>();
    if(vPos.empty())
      ConnectionViewer::WriteVectorPar(name.c_str(), vec, (MathVector<dim>*)NULL, dim);
    else
      ConnectionViewer::WriteVectorPar(name.c_str(), vec, &vPos[0], dim);
  }
 
  void write_vector_to_vtk(const vector_type& vec, const char* filename) {
    PROFILE_FUNC_GROUP("debug");
    //  check name
 
    std::string name;
    this->compose_file_path (name);
    name += filename;
 
    typedef GridFunction<TDomain, TAlgebra> TGridFunction;
    TGridFunction vtkFunc(
        m_spApproxSpace,
        m_spApproxSpace->dof_distribution(m_glFrom));
    vtkFunc.resize_values(vec.size());
    vtkFunc.assign(vec);
    VTKOutput<dim> out;
    out.print(name.c_str(), vtkFunc, m_printConsistent);
  }
 
protected:
  //  underlying approximation space
  SmartPtr<approximation_space_type> m_spApproxSpace;
 
  //  flag if write to conn viewer
  bool bConnViewerOut;
  bool bConnViewerIndices;
 
  //  flag if write to vtk
  bool bVTKOut;
 
  //  flag if data shall be made consistent before printing
  bool m_printConsistent;
 
  //  current grid level
  GridLevel m_glFrom, m_glTo;
};
 
//  Grid Function Position Provider
 
template<typename TGridFunction>
class GridFunctionPositionProvider: public IPositionProvider<
    TGridFunction::domain_type::dim> {
public:
  GridFunctionPositionProvider() :
      m_pGridFunc(NULL) {
  }
  GridFunctionPositionProvider(const TGridFunction& u) :
      m_pGridFunc(&u) {
  }
  void set_reference_grid_function(const TGridFunction& u) {
    m_pGridFunc = &u;
  }
 
  virtual bool get_positions(
      std::vector<MathVector<TGridFunction::domain_type::dim> >&vec) {
    UG_ASSERT(m_pGridFunc != NULL,
        "provide a grid function with set_reference_grid_function");
    ExtractPositions(*m_pGridFunc, vec);
    return true;
  }
 
protected:
  const TGridFunction *m_pGridFunc;
};
 
//  Grid Function Vector Writer
 
template<typename TGridFunction, typename TVector>
class GridFunctionVectorWriter: public IVectorWriter<TVector> {
public:
  typedef typename TVector::value_type value_type;
  typedef typename TGridFunction::domain_type domain_type;
  typedef TVector vector_type;
 
public:
  GridFunctionVectorWriter() :
      m_pGridFunc(NULL) {
  }
 
  void set_user_data(SmartPtr<UserData<number, domain_type::dim> > userData) {
    m_userData = userData;
  }
 
  void set_reference_grid_function(const TGridFunction& u) {
    m_pGridFunc = &u;
  }
 
  /*virtual double calculate(MathVector<3> &v, double time)
   {
   }*/
 
  virtual bool update(vector_type &vec) {
    PROFILE_FUNC_GROUP("debug");
    UG_ASSERT(m_pGridFunc != NULL,
        "provide a grid function with set_reference_grid_function");
    UG_ASSERT(m_userData.valid(), "provide user data with set_user_data");
 
    const TGridFunction &u = *m_pGridFunc;
 
    //  get position accessor
 
    const typename domain_type::position_accessor_type& aaPos =
        u.domain()->position_accessor();
 
    //  number of total dofs
    int nr = u.num_indices();
 
    //  resize positions
    vec.resize(nr);
 
    typedef typename TGridFunction::template traits<Vertex>::const_iterator const_iterator;
 
    //  loop all subsets
    for (int si = 0; si < u.num_subsets(); ++si) {
      //  loop all vertices
      const_iterator iter = u.template begin<Vertex>(si);
      const_iterator iterEnd = u.template end<Vertex>(si);
 
      for (; iter != iterEnd; ++iter) {
        //  get vertex
        Vertex* v = *iter;
 
        //  algebra indices vector
        std::vector<size_t> ind;
 
        //  load indices associated with vertex
        u.inner_algebra_indices(v, ind);
 
        number t = 0.0;
 
        number d;
        (*m_userData)(d, aaPos[v], t, si);
 
        //  write
        for (size_t i = 0; i < ind.size(); ++i) {
          const size_t index = ind[i];
          for (size_t alpha = 0; alpha < GetSize(vec[index]); ++alpha)
            BlockRef(vec[index], alpha) = d;
        }
      }
    }
    return true;
  }
 
protected:
  const TGridFunction *m_pGridFunc;
  SmartPtr<UserData<number, domain_type::dim> > m_userData;
};
 
//  Grid Function Vector Writer Dirichlet 0
 
template<typename TGridFunction>
class GridFunctionVectorWriterDirichlet0: public IVectorWriter<
    typename TGridFunction::algebra_type::vector_type> {
public:
  typedef typename TGridFunction::domain_type domain_type;
 
  typedef typename TGridFunction::approximation_space_type approximation_space_type;
 
  typedef typename TGridFunction::algebra_type algebra_type;
  typedef typename algebra_type::vector_type vector_type;
  typedef typename vector_type::value_type value_type;
 
public:
  GridFunctionVectorWriterDirichlet0() :
      m_pApproxSpace(NULL), m_spPostProcess(NULL), m_level(-1) {
  }
 
  void set_level(size_t level) {
    m_level = level;
  }
 
  void init(SmartPtr<IConstraint<algebra_type> > pp,
      approximation_space_type& approxSpace) {
    m_spPostProcess = pp;
    m_pApproxSpace = &approxSpace;
  }
 
  /*virtual double calculate(MathVector<3> &v, double time)
   {
   }*/
 
  virtual bool update(vector_type &vec) {
    PROFILE_FUNC_GROUP("debug");
    UG_ASSERT(m_spPostProcess.valid(), "provide a post process with init");
    UG_ASSERT(m_pApproxSpace != NULL, "provide approximation space init");
 
    size_t numDoFs = 0;
    GridLevel gl;
    if (m_level == (size_t) -1) {
      numDoFs = m_pApproxSpace->dof_distribution(GridLevel(GridLevel::TOP, GridLevel::SURFACE))->num_indices();
      gl = GridLevel();
    } else {
      numDoFs =
          m_pApproxSpace->dof_distribution(GridLevel(m_level, GridLevel::LEVEL, true))->num_indices();
      gl = GridLevel(m_level, GridLevel::LEVEL);
    }
    vec.resize(numDoFs);
    vec.set(1.0);
 
    m_spPostProcess->adjust_defect(vec, vec, m_pApproxSpace->dof_distribution(gl), CT_DIRICHLET);
 
    return true;
  }
 
protected:
  approximation_space_type * m_pApproxSpace;
  SmartPtr<IConstraint<algebra_type> > m_spPostProcess;
  size_t m_level;
};
 
} // end namespace ug
 
#endif /* __H__UG__LIB_DISC__FUNCTION_SPACE__GRID_FUNCTION_UTIL__ */