richards_lmh.hh

Go to the documentation of this file.

/*
 * richards_lmh.hh
 *
 *  Created on: Sep 16, 2015
 *      Author: jb
 */
 
#ifndef SRC_FLOW_RICHARDS_LMH_HH_
#define SRC_FLOW_RICHARDS_LMH_HH_
 
 
#include <memory>                    // for shared_ptr
#include "fields/field.hh"           // for Field
#include "fields/field_values.hh"    // for FieldValue<>::Scalar, FieldValue
#include "la/vector_mpi.hh"          // for VectorMPI
#include "flow/darcy_flow_mh.hh"     // for DarcyMH, DarcyMH::EqData
#include "input/type_base.hh"        // for Array
#include "input/type_generic.hh"     // for Instance
#include "petscvec.h"                // for VecScatter, _p_VecScatter
 
class Mesh;
class SoilModelBase;
namespace Input {
    class Record;
    namespace Type {
        class Record;
    }
}
 
/**
 * @brief Edge lumped mixed-hybrid solution of unsteady Darcy flow.
 *
 * The time term and sources are evenly distributed form an element to its edges.
 * This applies directly to the second Schur complement. After this system for pressure traces is solved we reconstruct pressures and side flows as follows:
 *
 * -# Element pressure is  average of edge pressure. This is in fact same as the MH for steady case so we let SchurComplement class do its job.
 *
 * -# We let SchurComplement to reconstruct fluxes and then account time term and sources which are evenly distributed from an element to its sides.
 *    It can be proved, that this keeps continuity of the fluxes over the edges.
 *
 * This lumping technique preserves discrete maximum principle for any time step provided one use acute mesh. But in practice even worse meshes are tractable.
 *
 * Ideas how to unify steady and unsteady flow:
 * zero_time_step:
 *
 * -# Set initial time.
 * -# Read initial condition. Reconstruct pressures.
 * -# Assembly system (possibly in matrix free way).
 * -# Reconstruct velocities (schur complement resolve).
 * -# Solve iteratively as regular time step if an input flag "steady_initial_time" is set.
 * -# (Detect that there is no time term. I such case use arbitrary long time step up to next change of data.
 *     Some kind of time step estimator would be nice.
 *
 * update solution:
 *
 * -# Move to the next time.
 * -# Update fields
 * -# Nonlinear solve.
 * -# In case of slow convergence, use shorter time-step, within estimated limits. Otherwise there is a different problem.
 */
class RichardsLMH : public DarcyMH
{
public:
    /// Class with all fields used in the equation DarcyFlow.
    /// This is common to all implementations since this provides interface
    /// to this equation for possible coupling.
    class EqData : public DarcyMH::EqData {
    public:
        EqData();
        // input fields
        Field<3, FieldValue<3>::Scalar > water_content_saturated;
        Field<3, FieldValue<3>::Scalar > water_content_residual;
        Field<3, FieldValue<3>::Scalar > genuchten_p_head_scale;
        Field<3, FieldValue<3>::Scalar > genuchten_n_exponent;
 
        //output fields
 
        // Auxiliary assembly fields.
        //std::unordered_map<unsigned int, unsigned int> *edge_new_local_4_mesh_idx_;
        VectorMPI phead_edge_;
        VectorMPI water_content_previous_it;
        VectorMPI water_content_previous_time;
        VectorMPI capacity;
        // source terms to be added to the side fluxes, in order to get proper (continuous) velocity field
        VectorMPI postprocess_side_sources;
 
 
        // This is necessary in the assembly
        // TODO: store time information in the field set and in fields, is it ok also for more complex discretization methods?
        double time_step_;
        std::shared_ptr<SoilModelBase> soil_model_;
    };
 
    RichardsLMH(Mesh &mesh, const Input::Record in_rec);
 
    static const Input::Type::Record & get_input_type();
protected:
    /// Registrar of class to factory
    static const int registrar;
 
    bool zero_time_term(bool time_global=false) override;
 
    void initialize_specific() override;
    //void local_assembly_specific(LocalAssemblyData &local_data) override;
    void assembly_source_term() override;
 
    void read_initial_condition() override;
    void assembly_linear_system() override;
    void setup_time_term() override;
    void prepare_new_time_step() override;
    void postprocess() override;
private:
 
    std::shared_ptr<EqData> data_;
    /// PETSC scatter from the solution vector to the parallel edge vector with ghost values.
    VecScatter solution_2_edge_scatter_;
 
    /*
    Vec steady_diagonal;
    Vec steady_rhs;
    Vec new_diagonal;
    Vec previous_solution;
*/
 
    //Vec time_term;
};
 
 
 
 
#endif /* SRC_FLOW_RICHARDS_LMH_HH_ */