transport.cc

Go to the documentation of this file.

/*!
 *
 * Copyright (C) 2015 Technical University of Liberec.  All rights reserved.
 * 
 * This program is free software; you can redistribute it and/or modify it under
 * the terms of the GNU General Public License version 3 as published by the
 * Free Software Foundation. (http://www.gnu.org/licenses/gpl-3.0.en.html)
 * 
 * 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 General Public License for more details.
 *
 * 
 * @file    transport.cc
 * @ingroup transport
 * @brief   Transport
 */

#include <memory>

#include "system/system.hh"
#include "system/sys_profiler.hh"

#include "mesh/side_impl.hh"
#include "mesh/long_idx.hh"
#include "mesh/mesh.h"
#include "mesh/partitioning.hh"
#include "mesh/accessors.hh"
#include "mesh/range_wrapper.hh"
#include "mesh/neighbours.h"
#include "transport/transport.h"

#include "la/distribution.hh"

#include "la/sparse_graph.hh"
#include <iostream>
#include <iomanip>
#include <string>

#include "io/output_time.hh"
#include "tools/time_governor.hh"
#include "coupling/balance.hh"
#include "input/accessors.hh"
#include "input/input_type.hh"

#include "fields/field_algo_base.hh"
#include "fields/field_values.hh"
#include "fields/field_fe.hh"
#include "fields/fe_value_handler.hh"
#include "fields/generic_field.hh"

#include "reaction/isotherm.hh" // SorptionType enum
#include "flow/mh_dofhandler.hh"


FLOW123D_FORCE_LINK_IN_CHILD(convectionTransport);


namespace IT = Input::Type;

const string _equation_name = "Solute_Advection_FV";

const int ConvectionTransport::registrar =
        Input::register_class< ConvectionTransport, Mesh &, const Input::Record >(_equation_name) +
        ConvectionTransport::get_input_type().size();

const IT::Record &ConvectionTransport::get_input_type()
{
    return IT::Record(_equation_name, "Finite volume method, explicit in time, for advection only solute transport.")
            .derive_from(ConcentrationTransportBase::get_input_type())
            .declare_key("input_fields", IT::Array(
                    EqData().make_field_descriptor_type(_equation_name)),
                    IT::Default::obligatory(),
                    "")
            .declare_key("output",
                    EqData().output_fields.make_output_type(_equation_name, ""),
                    IT::Default("{ \"fields\": [ \"conc\" ] }"),
                    "Specification of output fields and output times.")
            .close();
}


ConvectionTransport::EqData::EqData() : TransportEqData()
{
    *this += bc_conc.name("bc_conc")
            .description("Boundary condition for concentration of substances.")
            .input_default("0.0")
            .units( UnitSI().kg().m(-3) );

    *this += init_conc.name("init_conc")
            .description("Initial values for concentration of substances.")
            .input_default("0.0")
            .units( UnitSI().kg().m(-3) );

    output_fields += *this;
    output_fields += conc_mobile.name("conc")
            .units( UnitSI().kg().m(-3) )
            .flags(FieldFlag::equation_result)
            .description("Concentration solution.");
    output_fields += region_id.name("region_id")
            .units( UnitSI::dimensionless())
            .flags(FieldFlag::equation_external_output)
            .description("Region ids.");
    output_fields += subdomain.name("subdomain")
            .units( UnitSI::dimensionless() )
            .flags(FieldFlag::equation_external_output)
            .description("Subdomain ids of the domain decomposition.");
}


ConvectionTransport::ConvectionTransport(Mesh &init_mesh, const Input::Record in_rec)
: ConcentrationTransportBase(init_mesh, in_rec),
  is_mass_diag_changed(false),
  input_rec(in_rec),
  mh_dh(nullptr),
  sources_corr(nullptr)
{
    START_TIMER("ConvectionTransport");
    this->eq_data_ = &data_;

    transport_matrix_time = -1.0; // or -infty
    transport_bc_time = -1.0;
    is_convection_matrix_scaled = false;
    is_src_term_scaled = false;
    is_bc_term_scaled = false;
}

void ConvectionTransport::initialize()
{
    target_mark_type = time_->equation_fixed_mark_type();

    cfl_max_step = time_->end_time();

    data_.set_components(substances_.names());
    data_.set_input_list( input_rec.val<Input::Array>("input_fields"), *time_ );
    data_.set_mesh(*mesh_);

    make_transport_partitioning();
    alloc_transport_vectors();
    alloc_transport_structs_mpi();

    // register output vectors
    data_.output_fields.set_components(substances_.names());
    data_.output_fields.set_mesh(*mesh_);
    data_.output_fields.output_type(OutputTime::ELEM_DATA);
    data_.conc_mobile.setup_components();
    data_.region_id = GenericField<3>::region_id(*mesh_);
    data_.subdomain = GenericField<3>::subdomain(*mesh_);
    for (unsigned int sbi=0; sbi<n_substances(); sbi++)
    {
        // create shared pointer to a FieldFE and push this Field to output_field on all regions
        output_field_ptr[sbi] = create_field<3, FieldValue<3>::Scalar>(out_conc[sbi], *mesh_, 1);
        data_.conc_mobile[sbi].set_field(mesh_->region_db().get_region_set("ALL"), output_field_ptr[sbi], 0);
    }
    //output_stream_->add_admissible_field_names(input_rec.val<Input::Array>("output_fields"));
    //output_stream_->mark_output_times(*time_);

    data_.output_fields.initialize(output_stream_, mesh_, input_rec.val<Input::Record>("output"), time() );
    //cout << "Transport." << endl;
    //cout << time().marks();

    balance_->allocate(el_ds->lsize(), 1);

}


//=============================================================================
// MAKE TRANSPORT
//=============================================================================
void ConvectionTransport::make_transport_partitioning() {

//    int * id_4_old = new int[mesh_->n_elements()];
//    int i = 0;
//    for (auto ele : mesh_->elements_range()) id_4_old[i++] = ele.index();
//    mesh_->get_part()->id_maps(mesh_->n_elements(), id_4_old, el_ds, el_4_loc, row_4_el);
//    delete[] id_4_old;
    el_ds = mesh_->get_el_ds();
    el_4_loc = mesh_->get_el_4_loc();
    row_4_el = mesh_->get_row_4_el();

    // TODO: make output of partitioning is usefull but makes outputs different
    // on different number of processors, which breaks tests.
    //
    // Possible solution:
    // - have flag in ini file to turn this output ON
    // - possibility to have different ref_output for different num of proc.
    // - or do not test such kind of output
    //
    //for (auto ele : mesh_->elements_range()) {
    //    ele->pid()=el_ds->get_proc(row_4_el[ele.index()]);
    //}

}



ConvectionTransport::~ConvectionTransport()
{
    unsigned int sbi;

    if (sources_corr) {
        //Destroy mpi vectors at first
        chkerr(MatDestroy(&tm));
        chkerr(VecDestroy(&mass_diag));
        chkerr(VecDestroy(&vpmass_diag));
        chkerr(VecDestroy(&vcfl_flow_));
        chkerr(VecDestroy(&vcfl_source_));
        delete cfl_flow_;
        delete cfl_source_;

        for (sbi = 0; sbi < n_substances(); sbi++) {
            // mpi vectors
            chkerr(VecDestroy(&vconc[sbi]));
            chkerr(VecDestroy(&vpconc[sbi]));
            chkerr(VecDestroy(&bcvcorr[sbi]));
            chkerr(VecDestroy(&vcumulative_corr[sbi]));
            chkerr(VecDestroy(&v_tm_diag[sbi]));
            chkerr(VecDestroy(&v_sources_corr[sbi]));

            // arrays of arrays
            delete conc[sbi];
            delete cumulative_corr[sbi];
            delete tm_diag[sbi];
            delete sources_corr[sbi];
        }

        // arrays of mpi vectors
        delete vconc;
        delete vpconc;
        delete bcvcorr;
        delete vcumulative_corr;
        delete v_tm_diag;
        delete v_sources_corr;
        
        // arrays of arrays
        delete conc;
        delete cumulative_corr;
        delete tm_diag;
        delete sources_corr;
    }
}





void ConvectionTransport::set_initial_condition()
{
    for (auto elem : mesh_->elements_range()) {
        if (!el_ds->is_local(row_4_el[ elem.idx() ])) continue;

        LongIdx index = row_4_el[ elem.idx() ] - el_ds->begin();
        ElementAccessor<3> ele_acc = mesh_->element_accessor( elem.idx() );

        for (unsigned int sbi=0; sbi<n_substances(); sbi++) // Optimize: SWAP LOOPS
            conc[sbi][index] = data_.init_conc[sbi].value(elem.centre(), ele_acc);
    }

}

//=============================================================================
//  ALLOCATE OF TRANSPORT VARIABLES (ELEMENT & NODES)
//=============================================================================
void ConvectionTransport::alloc_transport_vectors() {

    unsigned int i, sbi, n_subst;
    n_subst = n_substances();
    
    sources_corr = new double*[n_subst];
    tm_diag = new double*[n_subst];
    cumulative_corr = new double*[n_subst];
    for (sbi = 0; sbi < n_subst; sbi++) {
      cumulative_corr[sbi] = new double[el_ds->lsize()];
      sources_corr[sbi] = new double[el_ds->lsize()];
      tm_diag[sbi] = new double[el_ds->lsize()];
    }

    conc = new double*[n_subst];
    out_conc.clear();
    out_conc.resize(n_subst);
    output_field_ptr.clear();
    output_field_ptr.resize(n_subst);
    for (sbi = 0; sbi < n_subst; sbi++) {
        conc[sbi] = new double[el_ds->lsize()];
        out_conc[sbi].resize( el_ds->size() );
        for (i = 0; i < el_ds->lsize(); i++) {
            conc[sbi][i] = 0.0;
        }
    }
    
    cfl_flow_ = new double[el_ds->lsize()];
    cfl_source_ = new double[el_ds->lsize()];
}

//=============================================================================
//  ALLOCATION OF TRANSPORT VECTORS (MPI)
//=============================================================================
void ConvectionTransport::alloc_transport_structs_mpi() {

    int sbi, n_subst, rank, np;
    n_subst = n_substances();

    MPI_Barrier(PETSC_COMM_WORLD);
    MPI_Comm_rank(PETSC_COMM_WORLD, &rank);
    MPI_Comm_size(PETSC_COMM_WORLD, &np);

    vconc = new Vec[n_subst];
    vpconc = new Vec[n_subst];
    bcvcorr = new Vec[n_subst];
    vcumulative_corr = new Vec[n_subst];
    v_tm_diag = new Vec[n_subst];
    v_sources_corr = new Vec[n_subst];
    

    for (sbi = 0; sbi < n_subst; sbi++) {
        VecCreateMPI(PETSC_COMM_WORLD, el_ds->lsize(), mesh_->n_elements(), &bcvcorr[sbi]);
        VecZeroEntries(bcvcorr[sbi]);
        VecCreateMPIWithArray(PETSC_COMM_WORLD,1, el_ds->lsize(), mesh_->n_elements(), conc[sbi],
                &vconc[sbi]);

        VecCreateMPI(PETSC_COMM_WORLD, el_ds->lsize(), mesh_->n_elements(), &vpconc[sbi]);
        VecZeroEntries(vconc[sbi]);
        VecZeroEntries(vpconc[sbi]);

        // SOURCES
        VecCreateMPIWithArray(PETSC_COMM_WORLD,1, el_ds->lsize(), mesh_->n_elements(),
                cumulative_corr[sbi],&vcumulative_corr[sbi]);
        
        VecCreateMPIWithArray(PETSC_COMM_WORLD,1, el_ds->lsize(), mesh_->n_elements(),
                sources_corr[sbi],&v_sources_corr[sbi]);
        
        VecCreateMPIWithArray(PETSC_COMM_WORLD,1, el_ds->lsize(), mesh_->n_elements(),
                tm_diag[sbi],&v_tm_diag[sbi]);

        VecZeroEntries(vcumulative_corr[sbi]);
        VecZeroEntries(out_conc[sbi].petsc_vec());
    }


    MatCreateAIJ(PETSC_COMM_WORLD, el_ds->lsize(), el_ds->lsize(), mesh_->n_elements(),
            mesh_->n_elements(), 16, PETSC_NULL, 4, PETSC_NULL, &tm);
    
    VecCreateMPI(PETSC_COMM_WORLD, el_ds->lsize(), mesh_->n_elements(), &mass_diag);
    VecCreateMPI(PETSC_COMM_WORLD, el_ds->lsize(), mesh_->n_elements(), &vpmass_diag);

    VecCreateMPIWithArray(PETSC_COMM_WORLD,1, el_ds->lsize(), mesh_->n_elements(),
            cfl_flow_,&vcfl_flow_);
    VecCreateMPIWithArray(PETSC_COMM_WORLD,1, el_ds->lsize(), mesh_->n_elements(),
            cfl_source_,&vcfl_source_);
}


void ConvectionTransport::set_boundary_conditions()
{
    START_TIMER ("set_boundary_conditions");

    ElementAccessor<3> elm;

    unsigned int sbi, loc_el;
    
    // Assembly bcvcorr vector
    for(sbi=0; sbi < n_substances(); sbi++) VecZeroEntries(bcvcorr[sbi]);

    balance_->start_flux_assembly(subst_idx);

    for (loc_el = 0; loc_el < el_ds->lsize(); loc_el++) {
        elm = mesh_->element_accessor( el_4_loc[loc_el] );
        if (elm->boundary_idx_ != NULL) {
            LongIdx new_i = row_4_el[ elm.idx() ];

            for (unsigned int si=0; si<elm->n_sides(); si++) {
                Boundary *b = elm.side(si)->cond();
                if (b != NULL) {
                    double flux = mh_dh->side_flux( *(elm.side(si)) );
                    if (flux < 0.0) {
                        double aij = -(flux / elm.measure() );

                        for (sbi=0; sbi<n_substances(); sbi++)
                        {
                            double value = data_.bc_conc[sbi].value( b->element_accessor().centre(), b->element_accessor() );
                            
                            VecSetValue(bcvcorr[sbi], new_i, value * aij, ADD_VALUES);

                            // CAUTION: It seems that PETSc possibly optimize allocated space during assembly.
                            // So we have to add also values that may be non-zero in future due to changing velocity field.
                            balance_->add_flux_matrix_values(subst_idx[sbi], elm.side(si), {row_4_el[el_4_loc[loc_el]]}, {0.});
                            balance_->add_flux_vec_value(subst_idx[sbi], elm.side(si), flux*value);
                        }
                    } else {
                        for (sbi=0; sbi<n_substances(); sbi++)
                            VecSetValue(bcvcorr[sbi], new_i, 0, ADD_VALUES);
                        
                        for (unsigned int sbi=0; sbi<n_substances(); sbi++)
                        {
                            balance_->add_flux_matrix_values(subst_idx[sbi], elm.side(si), {row_4_el[el_4_loc[loc_el]]}, {flux});
                            balance_->add_flux_vec_value(subst_idx[sbi], elm.side(si), 0);
                        }
                    }
                }
            }

        }
    }

    balance_->finish_flux_assembly(subst_idx);

    for (sbi=0; sbi<n_substances(); sbi++)      VecAssemblyBegin(bcvcorr[sbi]);
    for (sbi=0; sbi<n_substances(); sbi++)      VecAssemblyEnd(bcvcorr[sbi]);

    // we are calling set_boundary_conditions() after next_time() and
    // we are using data from t() before, so we need to set corresponding bc time
    transport_bc_time = time_->last_t();
}


//=============================================================================
// COMPUTE SOURCES
//=============================================================================
void ConvectionTransport::compute_concentration_sources() {

  //temporary variables
  unsigned int loc_el, sbi;
  double csection, source, diag;

  ElementAccessor<3> ele_acc;
  arma::vec3 p;
    
  //TODO: would it be possible to check the change in data for chosen substance? (may be in multifields?)
  
  //checking if the data were changed
    if( (data_.sources_density.changed() )
          || (data_.sources_conc.changed() )
          || (data_.sources_sigma.changed() )
          || (data_.cross_section.changed()))
    {
        START_TIMER("sources_reinit"); 
        balance_->start_source_assembly(subst_idx);
        
        for (loc_el = 0; loc_el < el_ds->lsize(); loc_el++) 
        {
            ele_acc = mesh_->element_accessor( el_4_loc[loc_el] );
            p = ele_acc.centre();
            csection = data_.cross_section.value(p, ele_acc);
            
            // read for all substances
            double max_cfl=0;
            for (sbi = 0; sbi < n_substances(); sbi++)
            {      
                double src_sigma = data_.sources_sigma[sbi].value(p, ele_acc);
                
                source = csection * (data_.sources_density[sbi].value(p, ele_acc) + src_sigma * data_.sources_conc[sbi].value(p, ele_acc));
                // addition to RHS
                sources_corr[sbi][loc_el] = source;
                // addition to diagonal of the transport matrix
                diag = src_sigma * csection;
                tm_diag[sbi][loc_el] = - diag;
                
                // compute maximal cfl condition over all substances
                max_cfl = std::max(max_cfl, fabs(diag));
                
                balance_->add_source_matrix_values(sbi, ele_acc.region().bulk_idx(), {row_4_el[el_4_loc[loc_el]]}, 
                                                    {- src_sigma * ele_acc.measure() * csection});
                balance_->add_source_vec_values(sbi, ele_acc.region().bulk_idx(), {row_4_el[el_4_loc[loc_el]]}, 
                                                {source * ele_acc.measure()});
            }
            
            cfl_source_[loc_el] = max_cfl;
        }
        
        balance_->finish_source_assembly(subst_idx);
        
        END_TIMER("sources_reinit");
    }
}



void ConvectionTransport::zero_time_step()
{
    OLD_ASSERT_EQUAL(time_->tlevel(), 0);

    data_.mark_input_times(*time_);
    data_.set_time(time_->step(), LimitSide::right);
    std::stringstream ss; // print warning message with table of uninitialized fields
    if ( FieldCommon::print_message_table(ss, "convection transport") ) {
        WarningOut() << ss.str();
    }

    set_initial_condition();
    create_mass_matrix();
    
    START_TIMER("Convection balance zero time step");

    create_transport_matrix_mpi();
    compute_concentration_sources();
    set_boundary_conditions();

    // write initial condition
    output_data();
}


bool ConvectionTransport::evaluate_time_constraint(double& time_constraint)
{
    OLD_ASSERT(mh_dh, "Null MH object.\n" );
    data_.set_time(time_->step(), LimitSide::right); // set to the last computed time
    
    START_TIMER("data reinit");
    
    bool cfl_changed = false;
    
    // if FLOW or DATA changed ---------------------> recompute transport matrix
    if (mh_dh->time_changed() > transport_matrix_time)
    {
        create_transport_matrix_mpi();
        is_convection_matrix_scaled=false;
        cfl_changed = true;
        DebugOut() << "CFL changed - flow.\n";
    }
    
    if (is_mass_diag_changed)
    {
        create_mass_matrix();
        cfl_changed = true;
        DebugOut() << "CFL changed - mass matrix.\n";
    }
    
    // if DATA changed ---------------------> recompute concentration sources (rhs and matrix diagonal)
    if( data_.sources_density.changed() || data_.sources_conc.changed() || data_.sources_sigma.changed()
       || data_.cross_section.changed())
    {
        compute_concentration_sources();
        is_src_term_scaled = false;
        cfl_changed = true;
        DebugOut() << "CFL changed - source.\n";
    }
    
    // now resolve the CFL condition
    if(cfl_changed)
    {
        // find maximum of sum of contribution from flow and sources: MAX(vcfl_flow_ + vcfl_source_)
        Vec cfl;
        VecCreateMPI(PETSC_COMM_WORLD, el_ds->lsize(),PETSC_DETERMINE, &cfl);
        VecWAXPY(cfl, 1.0, vcfl_flow_, vcfl_source_);
        VecMaxPointwiseDivide(cfl,mass_diag, &cfl_max_step);
        // get a reciprocal value as a time constraint
        cfl_max_step = 1 / cfl_max_step;
        DebugOut().fmt("CFL constraint (transport): {}\n", cfl_max_step);
    }
    
    // although it does not influence CFL, compute BC so the full system is assembled
    if ( (mh_dh->time_changed() > transport_bc_time)
        || data_.porosity.changed()
        || data_.water_content.changed()
        || data_.bc_conc.changed() )
    {
        set_boundary_conditions();
        is_bc_term_scaled = false;
    }

    END_TIMER("data reinit");
    
    // return time constraint
    time_constraint = cfl_max_step;
    return cfl_changed;
}

void ConvectionTransport::update_solution() {

    START_TIMER("convection-one step");
    
    // proceed to next time - which we are about to compute
    // explicit scheme looks one step back and uses data from previous time
    // (data time set previously in assess_time_constraint())
    time_->next_time();
    
    double dt_new = time_->dt(),                    // current time step we are about to compute
           dt_scaled = dt_new / time_->last_dt();   // scaling ratio to previous time step
    
    START_TIMER("time step rescaling");
    
    // if FLOW or DATA or BC or DT changed ---------------------> rescale boundary condition
    if( ! is_bc_term_scaled || time_->is_changed_dt() )
    {
        DebugOut() << "BC - rescale dt.\n";
        //choose between fresh scaling with new dt or rescaling to a new dt
        double dt = (!is_bc_term_scaled) ? dt_new : dt_scaled;
        for (unsigned int  sbi=0; sbi<n_substances(); sbi++)
            VecScale(bcvcorr[sbi], dt);
        is_bc_term_scaled = true;
    }
    

    // if DATA or TIME STEP changed -----------------------> rescale source term
    if( !is_src_term_scaled || time_->is_changed_dt()) {
        DebugOut() << "SRC - rescale dt.\n";
        //choose between fresh scaling with new dt or rescaling to a new dt
        double dt = (!is_src_term_scaled) ? dt_new : dt_scaled;
        for (unsigned int sbi=0; sbi<n_substances(); sbi++)
        {
            VecScale(v_sources_corr[sbi], dt);
            VecScale(v_tm_diag[sbi], dt);
        }
        is_src_term_scaled = true;
    }
    
    // if DATA or TIME STEP changed -----------------------> rescale transport matrix
    if ( !is_convection_matrix_scaled || time_->is_changed_dt()) {
        DebugOut() << "TM - rescale dt.\n";
        //choose between fresh scaling with new dt or rescaling to a new dt
        double dt = (!is_convection_matrix_scaled) ? dt_new : dt_scaled;
        
        MatScale(tm, dt);
        is_convection_matrix_scaled = true;
    }
    
    END_TIMER("time step rescaling");
    
    
    data_.set_time(time_->step(), LimitSide::right); // set to the last computed time
    if (data_.cross_section.changed() || data_.water_content.changed() || data_.porosity.changed())
    {
        VecCopy(mass_diag, vpmass_diag);
        create_mass_matrix();
    } else is_mass_diag_changed = false;
    

    // Compute new concentrations for every substance.
    
    for (unsigned int sbi = 0; sbi < n_substances(); sbi++) {
      // one step in MOBILE phase
      START_TIMER("mat mult");
      
      // tm_diag is a diagonal part of transport matrix, which depends on substance data (sources_sigma)
      // Wwe need keep transport matrix independent of substance, therefore we keep this diagonal part
      // separately in a vector tm_diag.
      // Computation: first, we compute this diagonal addition D*pconc and save it temporaly into RHS
        
      // RHS = D*pconc, where D is diagonal matrix represented by a vector
      VecPointwiseMult(vcumulative_corr[sbi], v_tm_diag[sbi], vconc[sbi]); //w = x.*y
      
      // Then we add boundary terms ans other source terms into RHS.
      // RHS = 1.0 * bcvcorr + 1.0 * v_sources_corr + 1.0 * rhs
      VecAXPBYPCZ(vcumulative_corr[sbi], 1.0, 1.0, 1.0, bcvcorr[sbi], v_sources_corr[sbi]);   //z = ax + by + cz
      
      // Then we set the new previous concentration.
      VecCopy(vconc[sbi], vpconc[sbi]); // pconc = conc
      // And finally proceed with transport matrix multiplication.
      if (is_mass_diag_changed) {
        VecPointwiseMult(vconc[sbi], vconc[sbi], vpmass_diag); // vconc*=vpmass_diag
        MatMultAdd(tm, vpconc[sbi], vconc[sbi], vconc[sbi]);   // vconc+=tm*vpconc
        VecAXPY(vconc[sbi], 1, vcumulative_corr[sbi]);         // vconc+=vcumulative_corr
        VecPointwiseDivide(vconc[sbi], vconc[sbi], mass_diag); // vconc/=mass_diag
      } else {
        MatMultAdd(tm, vpconc[sbi], vcumulative_corr[sbi], vconc[sbi]); // vconc =tm*vpconc+vcumulative_corr
        VecPointwiseDivide(vconc[sbi], vconc[sbi], mass_diag);          // vconc/=mass_diag
        VecAXPY(vconc[sbi], 1, vpconc[sbi]);                            // vconc+=vpconc
      }

      END_TIMER("mat mult");
    }
    
    for (unsigned int sbi=0; sbi<n_substances(); ++sbi)
      balance_->calculate_cumulative(sbi, vpconc[sbi]);
    
    END_TIMER("convection-one step");
}


void ConvectionTransport::set_target_time(double target_time)
{

    //sets target_mark_type (it is fixed) to be met in next_time()
    time_->marks().add(TimeMark(target_time, target_mark_type));

    // This is done every time TOS calls update_solution.
    // If CFL condition is changed, time fixation will change later from TOS.
    
    // Set the same constraint as was set last time.
    
    // TODO: fix this hack, remove this method completely, leaving just the first line at the calling point
    // in TransportOperatorSplitting::update_solution()
    // doing this directly leads to choose of large time step violationg CFL condition
    if (cfl_max_step > time_->dt()*1e-10)
    time_->set_upper_constraint(cfl_max_step, "CFL condition used from previous step.");

    // fixing convection time governor till next target_mark_type (got from TOS or other)
    // may have marks for data changes
    time_->fix_dt_until_mark();
}


void ConvectionTransport::create_mass_matrix()
{
    ElementAccessor<3> elm;
    
    VecZeroEntries(mass_diag);
    
    balance_->start_mass_assembly(subst_idx);

    for (unsigned int loc_el = 0; loc_el < el_ds->lsize(); loc_el++) {
        elm = mesh_->element_accessor( el_4_loc[loc_el] );

        double csection = data_.cross_section.value(elm.centre(), elm);
        //double por_m = data_.porosity.value(elm.centre(), elm->element_accessor());
        double por_m = data_.water_content.value(elm.centre(), elm);

        for (unsigned int sbi=0; sbi<n_substances(); ++sbi)
            balance_->add_mass_matrix_values(subst_idx[sbi], elm.region().bulk_idx(), {row_4_el[el_4_loc[loc_el]]}, {csection*por_m*elm.measure()} );
        
        VecSetValue(mass_diag, row_4_el[ elm.idx() ], csection*por_m, INSERT_VALUES);
    }
    
    balance_->finish_mass_assembly(subst_idx);
    
    VecAssemblyBegin(mass_diag);
    VecAssemblyEnd(mass_diag);
    
    is_mass_diag_changed = true;
}


//=============================================================================
// CREATE TRANSPORT MATRIX
//=============================================================================
void ConvectionTransport::create_transport_matrix_mpi() {

    START_TIMER("convection_matrix_assembly");

    ElementAccessor<3> el2;
    ElementAccessor<3> elm;
    const Edge *edg;
    int j, np, rank;
    LongIdx new_j, new_i;
    double aij, aii;
        
    MatZeroEntries(tm);

    MPI_Comm_rank(PETSC_COMM_WORLD, &rank);
    MPI_Comm_size(PETSC_COMM_WORLD, &np);

    double flux, flux2, edg_flux;

    aii = 0.0;

    for (unsigned int loc_el = 0; loc_el < el_ds->lsize(); loc_el++) {
        elm = mesh_->element_accessor( el_4_loc[loc_el] );
        new_i = row_4_el[ elm.idx() ];

        for (unsigned int si=0; si<elm->n_sides(); si++) {
            // same dim
            flux = mh_dh->side_flux( *(elm.side(si)) );
            if (elm.side(si)->cond() == NULL) {
                 edg = elm.side(si)->edge();
                 edg_flux = 0;
                 for( int s=0; s < edg->n_sides; s++) {
                   flux2 = mh_dh->side_flux( *(edg->side(s)) );
                   if ( flux2 > 0)  edg_flux+= flux2;
                 }
                 for(unsigned int s=0; s<edg->n_sides; s++)
                    // this test should also eliminate sides facing to lower dim. elements in comp. neighboring
                    // These edges on these sides should have just one side
                    if (edg->side(s) != elm.side(si)) {
                        j = edg->side(s)->element().idx();
                        new_j = row_4_el[j];

                        flux2 = mh_dh->side_flux( *(edg->side(s)));
                        if ( flux2 > 0.0 && flux <0.0)
                            aij = -(flux * flux2 / ( edg_flux * elm.measure() ) );
                        else aij =0;
                        MatSetValue(tm, new_i, new_j, aij, INSERT_VALUES);
                    }
            }
            if (flux > 0.0)
              aii -= (flux / elm.measure() );
        }  // end same dim     //ELEMENT_SIDES

        for (unsigned int n=0; n<elm->n_neighs_vb(); n++) // comp model
            {
                el2 = mesh_->element_accessor( elm->neigh_vb[n]->side()->element().idx() ); // higher dim. el.
                ASSERT( el2.idx() != elm.idx() ).error("Elm. same\n");
                new_j = row_4_el[ el2.idx() ];
                flux = mh_dh->side_flux( *(elm->neigh_vb[n]->side()) );

                // volume source - out-flow from higher dimension
                if (flux > 0.0)  aij = flux / elm.measure();
                else aij=0;
                MatSetValue(tm, new_i, new_j, aij, INSERT_VALUES);
                // out flow from higher dim. already accounted

                // volume drain - in-flow to higher dimension
                if (flux < 0.0) {
                  aii -= (-flux) / elm.measure();                           // diagonal drain
                  aij = (-flux) / el2.measure();
                } else aij=0;
                MatSetValue(tm, new_j, new_i, aij, INSERT_VALUES);
            }

        MatSetValue(tm, new_i, new_i, aii, INSERT_VALUES);
        
        cfl_flow_[loc_el] = fabs(aii);
        aii = 0.0;
    } // END ELEMENTS

    MatAssemblyBegin(tm, MAT_FINAL_ASSEMBLY);
    MatAssemblyEnd(tm, MAT_FINAL_ASSEMBLY);

    is_convection_matrix_scaled = false;
    END_TIMER("convection_matrix_assembly");

    transport_matrix_time = time_->t();
}





//=============================================================================
//      OUTPUT VECTOR GATHER
//=============================================================================
void ConvectionTransport::output_vector_gather() {

    unsigned int sbi;
    IS is;

    ISCreateGeneral(PETSC_COMM_SELF, mesh_->n_elements(), row_4_el, PETSC_COPY_VALUES, &is); //WithArray
    VecScatterCreate(vconc[0], is, out_conc[0].petsc_vec(), PETSC_NULL, &vconc_out_scatter);
    for (sbi = 0; sbi < n_substances(); sbi++) {
        VecScatterBegin(vconc_out_scatter, vconc[sbi], out_conc[sbi].petsc_vec(), INSERT_VALUES, SCATTER_FORWARD);
        VecScatterEnd(vconc_out_scatter, vconc[sbi], out_conc[sbi].petsc_vec(), INSERT_VALUES, SCATTER_FORWARD);
    }
    chkerr(VecScatterDestroy(&(vconc_out_scatter)));
    chkerr(ISDestroy(&(is)));
}


double **ConvectionTransport::get_concentration_matrix() {
    return conc;
}

void ConvectionTransport::get_par_info(LongIdx * &el_4_loc_out, Distribution * &el_distribution_out){
    el_4_loc_out = this->el_4_loc;
    el_distribution_out = this->el_ds;
    return;
}

//int *ConvectionTransport::get_el_4_loc(){
//  return el_4_loc;
//}

LongIdx *ConvectionTransport::get_row_4_el(){
    return row_4_el;
}



void ConvectionTransport::output_data() {

    data_.output_fields.set_time(time().step(), LimitSide::right);
    //if ( data_.output_fields.is_field_output_time(data_.conc_mobile, time().step()) ) {
    output_vector_gather();
    //}

    for (unsigned int sbi = 0; sbi < n_substances(); sbi++) {
        fill_output_data(out_conc[sbi], output_field_ptr[sbi]);
    }

    data_.output_fields.output(time().step());
    
    START_TIMER("TOS-balance");
    for (unsigned int sbi=0; sbi<n_substances(); ++sbi)
      balance_->calculate_instant(sbi, vconc[sbi]);
    balance_->output();
    END_TIMER("TOS-balance");
}

void ConvectionTransport::set_balance_object(std::shared_ptr<Balance> balance)
{
    balance_ = balance;
    subst_idx = balance_->add_quantities(substances_.names());
}