]> https://gitweb.dealii.org/ - dealii.git/commitdiff
Add tests for ADHelperResidualLinearization 7338/head
authorJean-Paul Pelteret <jppelteret@gmail.com>
Fri, 12 Oct 2018 06:32:56 +0000 (08:32 +0200)
committerJean-Paul Pelteret <jppelteret@gmail.com>
Fri, 12 Oct 2018 18:02:43 +0000 (20:02 +0200)
13 files changed:
tests/ad_common_tests/step-44-helper_res_lin_01.h [new file with mode: 0644]
tests/adolc/step-44-helper_res_lin_01_1.cc [new file with mode: 0644]
tests/adolc/step-44-helper_res_lin_01_1.output [new file with mode: 0644]
tests/adolc/step-44-helper_res_lin_01_2.cc [new file with mode: 0644]
tests/adolc/step-44-helper_res_lin_01_2.output [new file with mode: 0644]
tests/sacado/step-44-helper_res_lin_01_1.cc [new file with mode: 0644]
tests/sacado/step-44-helper_res_lin_01_1.output [new file with mode: 0644]
tests/sacado/step-44-helper_res_lin_01_2.cc [new file with mode: 0644]
tests/sacado/step-44-helper_res_lin_01_2.output [new file with mode: 0644]
tests/sacado/step-44-helper_res_lin_01_3.cc [new file with mode: 0644]
tests/sacado/step-44-helper_res_lin_01_3.output [new file with mode: 0644]
tests/sacado/step-44-helper_res_lin_01_4.cc [new file with mode: 0644]
tests/sacado/step-44-helper_res_lin_01_4.output [new file with mode: 0644]

diff --git a/tests/ad_common_tests/step-44-helper_res_lin_01.h b/tests/ad_common_tests/step-44-helper_res_lin_01.h
new file mode 100644 (file)
index 0000000..5b53d78
--- /dev/null
@@ -0,0 +1,1750 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2016 - 2018 by the deal.II authors
+//
+// This file is part of the deal.II library.
+//
+// The deal.II library is free software; you can use it, 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.
+// The full text of the license can be found in the file LICENSE at
+// the top level of the deal.II distribution.
+//
+// ---------------------------------------------------------------------
+
+// Header file:
+// This is a modified version of step-44, which tests the implementation of
+// cell-level auto-differentiation (linearisation of a residual vector).
+
+#include <deal.II/base/function.h>
+#include <deal.II/base/parameter_handler.h>
+#include <deal.II/base/point.h>
+#include <deal.II/base/quadrature_lib.h>
+#include <deal.II/base/quadrature_point_data.h>
+#include <deal.II/base/symmetric_tensor.h>
+#include <deal.II/base/tensor.h>
+#include <deal.II/base/timer.h>
+#include <deal.II/base/work_stream.h>
+
+#include <deal.II/differentiation/ad.h>
+
+#include <deal.II/dofs/dof_renumbering.h>
+#include <deal.II/dofs/dof_tools.h>
+
+#include <deal.II/fe/fe_dgp_monomial.h>
+#include <deal.II/fe/fe_q.h>
+#include <deal.II/fe/fe_system.h>
+#include <deal.II/fe/fe_tools.h>
+#include <deal.II/fe/fe_values.h>
+#include <deal.II/fe/mapping_q_eulerian.h>
+
+#include <deal.II/grid/grid_generator.h>
+#include <deal.II/grid/grid_in.h>
+#include <deal.II/grid/grid_tools.h>
+#include <deal.II/grid/tria.h>
+#include <deal.II/grid/tria_boundary_lib.h>
+
+#include <deal.II/lac/block_sparse_matrix.h>
+#include <deal.II/lac/block_vector.h>
+#include <deal.II/lac/constraint_matrix.h>
+#include <deal.II/lac/dynamic_sparsity_pattern.h>
+#include <deal.II/lac/full_matrix.h>
+#include <deal.II/lac/linear_operator.h>
+#include <deal.II/lac/packaged_operation.h>
+#include <deal.II/lac/precondition_selector.h>
+#include <deal.II/lac/solver_cg.h>
+#include <deal.II/lac/solver_selector.h>
+#include <deal.II/lac/sparse_direct.h>
+
+#include <deal.II/numerics/data_out.h>
+#include <deal.II/numerics/vector_tools.h>
+
+#include <fstream>
+#include <functional>
+#include <iostream>
+
+#include "../tests.h"
+namespace Step44
+{
+  using namespace dealii;
+  namespace AD = dealii::Differentiation::AD;
+  namespace Parameters
+  {
+    struct FESystem
+    {
+      unsigned int poly_degree;
+      unsigned int quad_order;
+      static void
+      declare_parameters(ParameterHandler &prm);
+      void
+      parse_parameters(ParameterHandler &prm);
+    };
+    void
+    FESystem::declare_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Finite element system");
+      {
+        prm.declare_entry("Polynomial degree",
+                          "2",
+                          Patterns::Integer(0),
+                          "Displacement system polynomial order");
+        prm.declare_entry("Quadrature order",
+                          "3",
+                          Patterns::Integer(0),
+                          "Gauss quadrature order");
+      }
+      prm.leave_subsection();
+    }
+    void
+    FESystem::parse_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Finite element system");
+      {
+        poly_degree = prm.get_integer("Polynomial degree");
+        quad_order  = prm.get_integer("Quadrature order");
+      }
+      prm.leave_subsection();
+    }
+    struct Geometry
+    {
+      unsigned int global_refinement;
+      double       scale;
+      double       p_p0;
+      static void
+      declare_parameters(ParameterHandler &prm);
+      void
+      parse_parameters(ParameterHandler &prm);
+    };
+    void
+    Geometry::declare_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Geometry");
+      {
+        prm.declare_entry("Global refinement",
+                          "2",
+                          Patterns::Integer(0),
+                          "Global refinement level");
+        prm.declare_entry("Grid scale",
+                          "1e-3",
+                          Patterns::Double(0.0),
+                          "Global grid scaling factor");
+        prm.declare_entry("Pressure ratio p/p0",
+                          "100",
+                          Patterns::Selection("20|40|60|80|100"),
+                          "Ratio of applied pressure to reference pressure");
+      }
+      prm.leave_subsection();
+    }
+    void
+    Geometry::parse_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Geometry");
+      {
+        global_refinement = prm.get_integer("Global refinement");
+        scale             = prm.get_double("Grid scale");
+        p_p0              = prm.get_double("Pressure ratio p/p0");
+      }
+      prm.leave_subsection();
+    }
+    struct Materials
+    {
+      double nu;
+      double mu;
+      static void
+      declare_parameters(ParameterHandler &prm);
+      void
+      parse_parameters(ParameterHandler &prm);
+    };
+    void
+    Materials::declare_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Material properties");
+      {
+        prm.declare_entry("Poisson's ratio",
+                          "0.4999",
+                          Patterns::Double(-1.0, 0.5),
+                          "Poisson's ratio");
+        prm.declare_entry("Shear modulus",
+                          "80.194e6",
+                          Patterns::Double(),
+                          "Shear modulus");
+      }
+      prm.leave_subsection();
+    }
+    void
+    Materials::parse_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Material properties");
+      {
+        nu = prm.get_double("Poisson's ratio");
+        mu = prm.get_double("Shear modulus");
+      }
+      prm.leave_subsection();
+    }
+    struct LinearSolver
+    {
+      std::string type_lin;
+      double      tol_lin;
+      double      max_iterations_lin;
+      bool        use_static_condensation;
+      std::string preconditioner_type;
+      double      preconditioner_relaxation;
+      static void
+      declare_parameters(ParameterHandler &prm);
+      void
+      parse_parameters(ParameterHandler &prm);
+    };
+    void
+    LinearSolver::declare_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Linear solver");
+      {
+        prm.declare_entry("Solver type",
+                          "CG",
+                          Patterns::Selection("CG|Direct"),
+                          "Type of solver used to solve the linear system");
+        prm.declare_entry("Residual",
+                          "1e-6",
+                          Patterns::Double(0.0),
+                          "Linear solver residual (scaled by residual norm)");
+        prm.declare_entry(
+          "Max iteration multiplier",
+          "1",
+          Patterns::Double(0.0),
+          "Linear solver iterations (multiples of the system matrix size)");
+        prm.declare_entry("Use static condensation",
+                          "true",
+                          Patterns::Bool(),
+                          "Solve the full block system or a reduced problem");
+        prm.declare_entry("Preconditioner type",
+                          "ssor",
+                          Patterns::Selection("jacobi|ssor"),
+                          "Type of preconditioner");
+        prm.declare_entry("Preconditioner relaxation",
+                          "0.65",
+                          Patterns::Double(0.0),
+                          "Preconditioner relaxation value");
+      }
+      prm.leave_subsection();
+    }
+    void
+    LinearSolver::parse_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Linear solver");
+      {
+        type_lin                  = prm.get("Solver type");
+        tol_lin                   = prm.get_double("Residual");
+        max_iterations_lin        = prm.get_double("Max iteration multiplier");
+        use_static_condensation   = prm.get_bool("Use static condensation");
+        preconditioner_type       = prm.get("Preconditioner type");
+        preconditioner_relaxation = prm.get_double("Preconditioner relaxation");
+      }
+      prm.leave_subsection();
+    }
+    struct NonlinearSolver
+    {
+      unsigned int max_iterations_NR;
+      double       tol_f;
+      double       tol_u;
+      static void
+      declare_parameters(ParameterHandler &prm);
+      void
+      parse_parameters(ParameterHandler &prm);
+    };
+    void
+    NonlinearSolver::declare_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Nonlinear solver");
+      {
+        prm.declare_entry("Max iterations Newton-Raphson",
+                          "10",
+                          Patterns::Integer(0),
+                          "Number of Newton-Raphson iterations allowed");
+        prm.declare_entry("Tolerance force",
+                          "1.0e-9",
+                          Patterns::Double(0.0),
+                          "Force residual tolerance");
+        prm.declare_entry("Tolerance displacement",
+                          "1.0e-6",
+                          Patterns::Double(0.0),
+                          "Displacement error tolerance");
+      }
+      prm.leave_subsection();
+    }
+    void
+    NonlinearSolver::parse_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Nonlinear solver");
+      {
+        max_iterations_NR = prm.get_integer("Max iterations Newton-Raphson");
+        tol_f             = prm.get_double("Tolerance force");
+        tol_u             = prm.get_double("Tolerance displacement");
+      }
+      prm.leave_subsection();
+    }
+    struct Time
+    {
+      double delta_t;
+      double end_time;
+      static void
+      declare_parameters(ParameterHandler &prm);
+      void
+      parse_parameters(ParameterHandler &prm);
+    };
+    void
+    Time::declare_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Time");
+      {
+        prm.declare_entry("End time", "1", Patterns::Double(), "End time");
+        prm.declare_entry("Time step size",
+                          "0.1",
+                          Patterns::Double(),
+                          "Time step size");
+      }
+      prm.leave_subsection();
+    }
+    void
+    Time::parse_parameters(ParameterHandler &prm)
+    {
+      prm.enter_subsection("Time");
+      {
+        end_time = prm.get_double("End time");
+        delta_t  = prm.get_double("Time step size");
+      }
+      prm.leave_subsection();
+    }
+    struct AllParameters : public FESystem,
+                           public Geometry,
+                           public Materials,
+                           public LinearSolver,
+                           public NonlinearSolver,
+                           public Time
+    {
+      AllParameters(const std::string &input_file);
+      static void
+      declare_parameters(ParameterHandler &prm);
+      void
+      parse_parameters(ParameterHandler &prm);
+    };
+    AllParameters::AllParameters(const std::string &input_file)
+    {
+      ParameterHandler prm;
+      declare_parameters(prm);
+      prm.parse_input(input_file);
+      parse_parameters(prm);
+    }
+    void
+    AllParameters::declare_parameters(ParameterHandler &prm)
+    {
+      FESystem::declare_parameters(prm);
+      Geometry::declare_parameters(prm);
+      Materials::declare_parameters(prm);
+      LinearSolver::declare_parameters(prm);
+      NonlinearSolver::declare_parameters(prm);
+      Time::declare_parameters(prm);
+    }
+    void
+    AllParameters::parse_parameters(ParameterHandler &prm)
+    {
+      FESystem::parse_parameters(prm);
+      Geometry::parse_parameters(prm);
+      Materials::parse_parameters(prm);
+      LinearSolver::parse_parameters(prm);
+      NonlinearSolver::parse_parameters(prm);
+      Time::parse_parameters(prm);
+    }
+  } // namespace Parameters
+  template <int dim>
+  class StandardTensors
+  {
+  public:
+    static const SymmetricTensor<2, dim> I;
+    static const SymmetricTensor<4, dim> IxI;
+    static const SymmetricTensor<4, dim> II;
+    static const SymmetricTensor<4, dim> dev_P;
+  };
+  template <int dim>
+  const SymmetricTensor<2, dim>
+    StandardTensors<dim>::I = unit_symmetric_tensor<dim>();
+  template <int dim>
+  const SymmetricTensor<4, dim> StandardTensors<dim>::IxI = outer_product(I, I);
+  template <int dim>
+  const SymmetricTensor<4, dim>
+    StandardTensors<dim>::II = identity_tensor<dim>();
+  template <int dim>
+  const SymmetricTensor<4, dim>
+    StandardTensors<dim>::dev_P = deviator_tensor<dim>();
+  class Time
+  {
+  public:
+    Time(const double time_end, const double delta_t)
+      : timestep(0)
+      , time_current(0.0)
+      , time_end(time_end)
+      , delta_t(delta_t)
+    {}
+    virtual ~Time()
+    {}
+    double
+    current() const
+    {
+      return time_current;
+    }
+    double
+    end() const
+    {
+      return time_end;
+    }
+    double
+    get_delta_t() const
+    {
+      return delta_t;
+    }
+    unsigned int
+    get_timestep() const
+    {
+      return timestep;
+    }
+    void
+    increment()
+    {
+      time_current += delta_t;
+      ++timestep;
+    }
+
+  private:
+    unsigned int timestep;
+    double       time_current;
+    const double time_end;
+    const double delta_t;
+  };
+  template <int dim>
+  class Material_Compressible_Neo_Hook_Three_Field
+  {
+  public:
+    Material_Compressible_Neo_Hook_Three_Field(const double mu, const double nu)
+      : kappa((2.0 * mu * (1.0 + nu)) / (3.0 * (1.0 - 2.0 * nu)))
+      , c_1(mu / 2.0)
+    {
+      Assert(kappa > 0, ExcInternalError());
+    }
+    ~Material_Compressible_Neo_Hook_Three_Field()
+    {}
+
+    template <typename NumberType>
+    SymmetricTensor<2, dim, NumberType>
+    get_tau(const Tensor<2, dim, NumberType> &F,
+            const NumberType &                p_tilde) const
+    {
+      return get_tau_iso(F) + get_tau_vol(F, p_tilde);
+    }
+    template <typename NumberType>
+    NumberType
+    get_dPsi_vol_dJ(const NumberType &J_tilde) const
+    {
+      return (kappa / 2.0) * (J_tilde - 1.0 / J_tilde);
+    }
+
+  protected:
+    const double kappa;
+    const double c_1;
+
+    template <typename NumberType>
+    SymmetricTensor<2, dim, NumberType>
+    get_tau_vol(const Tensor<2, dim, NumberType> &F,
+                const NumberType &                p_tilde) const
+    {
+      const NumberType det_F = determinant(F);
+      // return p_tilde * det_F * StandardTensors<dim>::I;
+      SymmetricTensor<2, dim, NumberType> tau_vol(StandardTensors<dim>::I);
+      tau_vol *= p_tilde * det_F;
+      return tau_vol;
+    }
+    template <typename NumberType>
+    SymmetricTensor<2, dim, NumberType>
+    get_tau_iso(const Tensor<2, dim, NumberType> &F) const
+    {
+      // return StandardTensors<dim>::dev_P * get_tau_bar(F);
+
+      const SymmetricTensor<2, dim, NumberType> tau_bar = get_tau_bar(F);
+      SymmetricTensor<2, dim, NumberType>       tau_iso;
+      for (unsigned int i = 0; i < dim; ++i)
+        for (unsigned int j = i; j < dim; ++j)
+          for (unsigned int k = 0; k < dim; ++k)
+            for (unsigned int l = 0; l < dim; ++l)
+              tau_iso[i][j] +=
+                StandardTensors<dim>::dev_P[i][j][k][l] * tau_bar[k][l];
+
+      return tau_iso;
+    }
+    template <typename NumberType>
+    SymmetricTensor<2, dim, NumberType>
+    get_tau_bar(const Tensor<2, dim, NumberType> &F) const
+    {
+      using adtl::pow;
+      using std::pow;
+
+      const NumberType                    det_F = determinant(F);
+      SymmetricTensor<2, dim, NumberType> b_bar = symmetrize(F * transpose(F));
+      b_bar *= std::pow(det_F, -2.0 / dim);
+      return 2.0 * c_1 * b_bar;
+    }
+  };
+  template <int dim>
+  class PointHistory
+  {
+  public:
+    PointHistory()
+    {}
+    virtual ~PointHistory()
+    {}
+    void
+    setup_lqp(const Parameters::AllParameters &parameters)
+    {
+      material.reset(
+        new Material_Compressible_Neo_Hook_Three_Field<dim>(parameters.mu,
+                                                            parameters.nu));
+    }
+
+    template <typename NumberType>
+    SymmetricTensor<2, dim, NumberType>
+    get_tau(const Tensor<2, dim, NumberType> &F,
+            const NumberType &                p_tilde) const
+    {
+      return material->get_tau(F, p_tilde);
+    }
+    template <typename NumberType>
+    NumberType
+    get_dPsi_vol_dJ(const NumberType &J_tilde) const
+    {
+      return material->get_dPsi_vol_dJ(J_tilde);
+    }
+
+  private:
+    std::shared_ptr<Material_Compressible_Neo_Hook_Three_Field<dim>> material;
+  };
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  class Solid
+  {
+  public:
+    Solid(const std::string &input_file);
+    virtual ~Solid();
+    void
+    run();
+
+  private:
+    struct PerTaskData_ASM;
+    struct ScratchData_ASM;
+    struct PerTaskData_SC;
+    struct ScratchData_SC;
+    void
+    make_grid();
+    void
+    system_setup();
+    void
+    determine_component_extractors();
+    void
+    assemble_system(const BlockVector<double> &solution_delta);
+    void
+    assemble_system_one_cell(
+      const typename DoFHandler<dim>::active_cell_iterator &cell,
+      ScratchData_ASM &                                     scratch,
+      PerTaskData_ASM &                                     data) const;
+    void
+    copy_local_to_global_system(const PerTaskData_ASM &data);
+    void
+    assemble_sc();
+    void
+    assemble_sc_one_cell(
+      const typename DoFHandler<dim>::active_cell_iterator &cell,
+      ScratchData_SC &                                      scratch,
+      PerTaskData_SC &                                      data);
+    void
+    copy_local_to_global_sc(const PerTaskData_SC &data);
+    void
+    make_constraints(const int &it_nr);
+    void
+    setup_qph();
+    void
+    solve_nonlinear_timestep(BlockVector<double> &solution_delta);
+    std::pair<unsigned int, double>
+    solve_linear_system(BlockVector<double> &newton_update);
+    BlockVector<double>
+    get_total_solution(const BlockVector<double> &solution_delta) const;
+    void
+                              output_results() const;
+    Parameters::AllParameters parameters;
+    double                    vol_reference;
+    Triangulation<dim>        triangulation;
+    Time                      time;
+    mutable TimerOutput       timer;
+    CellDataStorage<typename Triangulation<dim>::cell_iterator,
+                    PointHistory<dim>>
+                                     quadrature_point_history;
+    const unsigned int               degree;
+    const FESystem<dim>              fe;
+    DoFHandler<dim>                  dof_handler_ref;
+    const unsigned int               dofs_per_cell;
+    const FEValuesExtractors::Vector u_fe;
+    const FEValuesExtractors::Scalar p_fe;
+    const FEValuesExtractors::Scalar J_fe;
+    static const unsigned int        n_blocks          = 3;
+    static const unsigned int        n_components      = dim + 2;
+    static const unsigned int        first_u_component = 0;
+    static const unsigned int        p_component       = dim;
+    static const unsigned int        J_component       = dim + 1;
+    enum
+    {
+      u_dof = 0,
+      p_dof = 1,
+      J_dof = 2
+    };
+    std::vector<types::global_dof_index> dofs_per_block;
+    std::vector<types::global_dof_index> element_indices_u;
+    std::vector<types::global_dof_index> element_indices_p;
+    std::vector<types::global_dof_index> element_indices_J;
+    const QGauss<dim>                    qf_cell;
+    const QGauss<dim - 1>                qf_face;
+    const unsigned int                   n_q_points;
+    const unsigned int                   n_q_points_f;
+    ConstraintMatrix                     constraints;
+    BlockSparsityPattern                 sparsity_pattern;
+    BlockSparseMatrix<double>            tangent_matrix;
+    BlockVector<double>                  system_rhs;
+    BlockVector<double>                  solution_n;
+    struct Errors
+    {
+      Errors()
+        : norm(1.0)
+        , u(1.0)
+        , p(1.0)
+        , J(1.0)
+      {}
+      void
+      reset()
+      {
+        norm = 1.0;
+        u    = 1.0;
+        p    = 1.0;
+        J    = 1.0;
+      }
+      void
+      normalise(const Errors &rhs)
+      {
+        if (rhs.norm != 0.0)
+          norm /= rhs.norm;
+        if (rhs.u != 0.0)
+          u /= rhs.u;
+        if (rhs.p != 0.0)
+          p /= rhs.p;
+        if (rhs.J != 0.0)
+          J /= rhs.J;
+      }
+      double norm, u, p, J;
+    };
+    Errors error_residual, error_residual_0, error_residual_norm, error_update,
+      error_update_0, error_update_norm;
+    void
+    get_error_residual(Errors &error_residual);
+    void
+    get_error_update(const BlockVector<double> &newton_update,
+                     Errors &                   error_update);
+    std::pair<double, double>
+    get_error_dilation(const BlockVector<double> &solution_total) const;
+    void
+    print_conv_header();
+    void
+    print_conv_footer(const BlockVector<double> &solution_delta);
+  };
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  Solid<dim, number_t, ad_type_code>::Solid(const std::string &input_file)
+    : parameters(input_file)
+    , triangulation(Triangulation<dim>::maximum_smoothing)
+    , time(parameters.end_time, parameters.delta_t)
+    , timer(std::cout, TimerOutput::never, TimerOutput::wall_times)
+    , degree(parameters.poly_degree)
+    , fe(FE_Q<dim>(parameters.poly_degree),
+         dim, // displacement
+         FE_DGPMonomial<dim>(parameters.poly_degree - 1),
+         1, // pressure
+         FE_DGPMonomial<dim>(parameters.poly_degree - 1),
+         1)
+    , // dilatation
+    dof_handler_ref(triangulation)
+    , dofs_per_cell(fe.dofs_per_cell)
+    , u_fe(first_u_component)
+    , p_fe(p_component)
+    , J_fe(J_component)
+    , dofs_per_block(n_blocks)
+    , qf_cell(parameters.quad_order)
+    , qf_face(parameters.quad_order)
+    , n_q_points(qf_cell.size())
+    , n_q_points_f(qf_face.size())
+  {
+    Assert(dim == 2 || dim == 3,
+           ExcMessage("This problem only works in 2 or 3 space dimensions."));
+    determine_component_extractors();
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  Solid<dim, number_t, ad_type_code>::~Solid()
+  {
+    dof_handler_ref.clear();
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::run()
+  {
+    make_grid();
+    system_setup();
+    {
+      ConstraintMatrix constraints;
+      constraints.close();
+      const ComponentSelectFunction<dim> J_mask(J_component, n_components);
+      VectorTools::project(dof_handler_ref,
+                           constraints,
+                           QGauss<dim>(degree + 2),
+                           J_mask,
+                           solution_n);
+    }
+    output_results();
+    time.increment();
+    BlockVector<double> solution_delta(dofs_per_block);
+    while (time.current() < time.end())
+      {
+        solution_delta = 0.0;
+        solve_nonlinear_timestep(solution_delta);
+        solution_n += solution_delta;
+        output_results();
+        // Output displacement at centre of traction surface
+        {
+          const Point<dim> soln_pt(
+            dim == 3 ? Point<dim>(0.0, 1.0, 0.0) * parameters.scale :
+                       Point<dim>(0.0, 1.0) * parameters.scale);
+          typename DoFHandler<dim>::active_cell_iterator
+            cell = dof_handler_ref.begin_active(),
+            endc = dof_handler_ref.end();
+          for (; cell != endc; ++cell)
+            for (unsigned int v = 0; v < GeometryInfo<dim>::vertices_per_cell;
+                 ++v)
+              if (cell->vertex(v).distance(soln_pt) < 1e-6 * parameters.scale)
+                {
+                  Tensor<1, dim> soln;
+                  for (unsigned int d = 0; d < dim; ++d)
+                    soln[d] = solution_n(cell->vertex_dof_index(v, u_dof + d));
+                  deallog << "Timestep " << time.get_timestep() << ": " << soln
+                          << std::endl;
+                }
+        }
+        time.increment();
+      }
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  struct Solid<dim, number_t, ad_type_code>::PerTaskData_ASM
+  {
+    FullMatrix<double>                   cell_matrix;
+    Vector<double>                       cell_rhs;
+    std::vector<types::global_dof_index> local_dof_indices;
+    PerTaskData_ASM(const unsigned int dofs_per_cell)
+      : cell_matrix(dofs_per_cell, dofs_per_cell)
+      , cell_rhs(dofs_per_cell)
+      , local_dof_indices(dofs_per_cell)
+    {}
+    void
+    reset()
+    {
+      cell_matrix = 0.0;
+      cell_rhs    = 0.0;
+    }
+  };
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  struct Solid<dim, number_t, ad_type_code>::ScratchData_ASM
+  {
+    const BlockVector<double> &solution_total;
+    FEValues<dim>              fe_values_ref;
+    FEFaceValues<dim>          fe_face_values_ref;
+    ScratchData_ASM(const FiniteElement<dim> & fe_cell,
+                    const QGauss<dim> &        qf_cell,
+                    const UpdateFlags          uf_cell,
+                    const QGauss<dim - 1> &    qf_face,
+                    const UpdateFlags          uf_face,
+                    const BlockVector<double> &solution_total)
+      : solution_total(solution_total)
+      , fe_values_ref(fe_cell, qf_cell, uf_cell)
+      , fe_face_values_ref(fe_cell, qf_face, uf_face)
+    {}
+    ScratchData_ASM(const ScratchData_ASM &rhs)
+      : solution_total(rhs.solution_total)
+      , fe_values_ref(rhs.fe_values_ref.get_fe(),
+                      rhs.fe_values_ref.get_quadrature(),
+                      rhs.fe_values_ref.get_update_flags())
+      , fe_face_values_ref(rhs.fe_face_values_ref.get_fe(),
+                           rhs.fe_face_values_ref.get_quadrature(),
+                           rhs.fe_face_values_ref.get_update_flags())
+    {}
+    void
+    reset()
+    {}
+  };
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  struct Solid<dim, number_t, ad_type_code>::PerTaskData_SC
+  {
+    FullMatrix<double>                   cell_matrix;
+    std::vector<types::global_dof_index> local_dof_indices;
+    FullMatrix<double>                   k_orig;
+    FullMatrix<double>                   k_pu;
+    FullMatrix<double>                   k_pJ;
+    FullMatrix<double>                   k_JJ;
+    FullMatrix<double>                   k_pJ_inv;
+    FullMatrix<double>                   k_bbar;
+    FullMatrix<double>                   A;
+    FullMatrix<double>                   B;
+    FullMatrix<double>                   C;
+    PerTaskData_SC(const unsigned int dofs_per_cell,
+                   const unsigned int n_u,
+                   const unsigned int n_p,
+                   const unsigned int n_J)
+      : cell_matrix(dofs_per_cell, dofs_per_cell)
+      , local_dof_indices(dofs_per_cell)
+      , k_orig(dofs_per_cell, dofs_per_cell)
+      , k_pu(n_p, n_u)
+      , k_pJ(n_p, n_J)
+      , k_JJ(n_J, n_J)
+      , k_pJ_inv(n_p, n_J)
+      , k_bbar(n_u, n_u)
+      , A(n_J, n_u)
+      , B(n_J, n_u)
+      , C(n_p, n_u)
+    {}
+    void
+    reset()
+    {}
+  };
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  struct Solid<dim, number_t, ad_type_code>::ScratchData_SC
+  {
+    void
+    reset()
+    {}
+  };
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::make_grid()
+  {
+    GridGenerator::hyper_rectangle(
+      triangulation,
+      (dim == 3 ? Point<dim>(0.0, 0.0, 0.0) : Point<dim>(0.0, 0.0)),
+      (dim == 3 ? Point<dim>(1.0, 1.0, 1.0) : Point<dim>(1.0, 1.0)),
+      true);
+    GridTools::scale(parameters.scale, triangulation);
+    triangulation.refine_global(std::max(1U, parameters.global_refinement));
+    vol_reference = GridTools::volume(triangulation);
+    std::cout << "Grid:\n\t Reference volume: " << vol_reference << std::endl;
+    typename Triangulation<dim>::active_cell_iterator cell = triangulation
+                                                               .begin_active(),
+                                                      endc =
+                                                        triangulation.end();
+    for (; cell != endc; ++cell)
+      for (unsigned int face = 0; face < GeometryInfo<dim>::faces_per_cell;
+           ++face)
+        {
+          if (cell->face(face)->at_boundary() == true &&
+              cell->face(face)->center()[1] == 1.0 * parameters.scale)
+            {
+              if (dim == 3)
+                {
+                  if (cell->face(face)->center()[0] < 0.5 * parameters.scale &&
+                      cell->face(face)->center()[2] < 0.5 * parameters.scale)
+                    cell->face(face)->set_boundary_id(6);
+                }
+              else
+                {
+                  if (cell->face(face)->center()[0] < 0.5 * parameters.scale)
+                    cell->face(face)->set_boundary_id(6);
+                }
+            }
+        }
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::system_setup()
+  {
+    timer.enter_subsection("Setup system");
+    std::vector<unsigned int> block_component(n_components,
+                                              u_dof); // Displacement
+    block_component[p_component] = p_dof;             // Pressure
+    block_component[J_component] = J_dof;             // Dilatation
+    dof_handler_ref.distribute_dofs(fe);
+    DoFRenumbering::Cuthill_McKee(dof_handler_ref);
+    DoFRenumbering::component_wise(dof_handler_ref, block_component);
+    DoFTools::count_dofs_per_block(dof_handler_ref,
+                                   dofs_per_block,
+                                   block_component);
+    std::cout << "Triangulation:"
+              << "\n\t Number of active cells: "
+              << triangulation.n_active_cells()
+              << "\n\t Number of degrees of freedom: "
+              << dof_handler_ref.n_dofs() << std::endl;
+    tangent_matrix.clear();
+    {
+      const types::global_dof_index n_dofs_u = dofs_per_block[u_dof];
+      const types::global_dof_index n_dofs_p = dofs_per_block[p_dof];
+      const types::global_dof_index n_dofs_J = dofs_per_block[J_dof];
+      BlockDynamicSparsityPattern   dsp(n_blocks, n_blocks);
+      dsp.block(u_dof, u_dof).reinit(n_dofs_u, n_dofs_u);
+      dsp.block(u_dof, p_dof).reinit(n_dofs_u, n_dofs_p);
+      dsp.block(u_dof, J_dof).reinit(n_dofs_u, n_dofs_J);
+      dsp.block(p_dof, u_dof).reinit(n_dofs_p, n_dofs_u);
+      dsp.block(p_dof, p_dof).reinit(n_dofs_p, n_dofs_p);
+      dsp.block(p_dof, J_dof).reinit(n_dofs_p, n_dofs_J);
+      dsp.block(J_dof, u_dof).reinit(n_dofs_J, n_dofs_u);
+      dsp.block(J_dof, p_dof).reinit(n_dofs_J, n_dofs_p);
+      dsp.block(J_dof, J_dof).reinit(n_dofs_J, n_dofs_J);
+      dsp.collect_sizes();
+      Table<2, DoFTools::Coupling> coupling(n_components, n_components);
+      for (unsigned int ii = 0; ii < n_components; ++ii)
+        for (unsigned int jj = 0; jj < n_components; ++jj)
+          if (((ii < p_component) && (jj == J_component)) ||
+              ((ii == J_component) && (jj < p_component)) ||
+              ((ii == p_component) && (jj == p_component)))
+            coupling[ii][jj] = DoFTools::none;
+          else
+            coupling[ii][jj] = DoFTools::always;
+      DoFTools::make_sparsity_pattern(
+        dof_handler_ref, coupling, dsp, constraints, false);
+      sparsity_pattern.copy_from(dsp);
+    }
+    tangent_matrix.reinit(sparsity_pattern);
+    system_rhs.reinit(dofs_per_block);
+    system_rhs.collect_sizes();
+    solution_n.reinit(dofs_per_block);
+    solution_n.collect_sizes();
+    setup_qph();
+    timer.leave_subsection();
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::determine_component_extractors()
+  {
+    element_indices_u.clear();
+    element_indices_p.clear();
+    element_indices_J.clear();
+    for (unsigned int k = 0; k < fe.dofs_per_cell; ++k)
+      {
+        const unsigned int k_group = fe.system_to_base_index(k).first.first;
+        if (k_group == u_dof)
+          element_indices_u.push_back(k);
+        else if (k_group == p_dof)
+          element_indices_p.push_back(k);
+        else if (k_group == J_dof)
+          element_indices_J.push_back(k);
+        else
+          {
+            Assert(k_group <= J_dof, ExcInternalError());
+          }
+      }
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::setup_qph()
+  {
+    std::cout << "    Setting up quadrature point data..." << std::endl;
+    quadrature_point_history.initialize(triangulation.begin_active(),
+                                        triangulation.end(),
+                                        n_q_points);
+    for (typename Triangulation<dim>::active_cell_iterator cell =
+           triangulation.begin_active();
+         cell != triangulation.end();
+         ++cell)
+      {
+        const std::vector<std::shared_ptr<PointHistory<dim>>> lqph =
+          quadrature_point_history.get_data(cell);
+        Assert(lqph.size() == n_q_points, ExcInternalError());
+        for (unsigned int q_point = 0; q_point < n_q_points; ++q_point)
+          lqph[q_point]->setup_lqp(parameters);
+      }
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::solve_nonlinear_timestep(
+    BlockVector<double> &solution_delta)
+  {
+    std::cout << std::endl
+              << "Timestep " << time.get_timestep() << " @ " << time.current()
+              << "s" << std::endl;
+    BlockVector<double> newton_update(dofs_per_block);
+    error_residual.reset();
+    error_residual_0.reset();
+    error_residual_norm.reset();
+    error_update.reset();
+    error_update_0.reset();
+    error_update_norm.reset();
+    print_conv_header();
+    unsigned int newton_iteration = 0;
+    for (; newton_iteration < parameters.max_iterations_NR; ++newton_iteration)
+      {
+        std::cout << " " << std::setw(2) << newton_iteration << " "
+                  << std::flush;
+        make_constraints(newton_iteration);
+        assemble_system(solution_delta);
+        get_error_residual(error_residual);
+        if (newton_iteration == 0)
+          error_residual_0 = error_residual;
+        error_residual_norm = error_residual;
+        error_residual_norm.normalise(error_residual_0);
+        if (newton_iteration > 0 && error_update_norm.u <= parameters.tol_u &&
+            error_residual_norm.u <= parameters.tol_f)
+          {
+            std::cout << " CONVERGED! " << std::endl;
+            print_conv_footer(solution_delta);
+            break;
+          }
+        const std::pair<unsigned int, double> lin_solver_output =
+          solve_linear_system(newton_update);
+        get_error_update(newton_update, error_update);
+        if (newton_iteration == 0)
+          error_update_0 = error_update;
+        error_update_norm = error_update;
+        error_update_norm.normalise(error_update_0);
+        solution_delta += newton_update;
+        std::cout << " | " << std::fixed << std::setprecision(3) << std::setw(7)
+                  << std::scientific << lin_solver_output.first << "  "
+                  << lin_solver_output.second << "  "
+                  << error_residual_norm.norm << "  " << error_residual_norm.u
+                  << "  " << error_residual_norm.p << "  "
+                  << error_residual_norm.J << "  " << error_update_norm.norm
+                  << "  " << error_update_norm.u << "  " << error_update_norm.p
+                  << "  " << error_update_norm.J << "  " << std::endl;
+      }
+    AssertThrow(newton_iteration <= parameters.max_iterations_NR,
+                ExcMessage("No convergence in nonlinear solver!"));
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::print_conv_header()
+  {
+    static const unsigned int l_width = 144;
+    for (unsigned int i = 0; i < l_width; ++i)
+      std::cout << "_";
+    std::cout << std::endl;
+    std::cout << "           SOLVER STEP             "
+              << " |  LIN_IT   LIN_RES    RES_NORM    "
+              << " RES_U     RES_P      RES_J     NU_NORM     "
+              << " NU_U       NU_P       NU_J " << std::endl;
+    for (unsigned int i = 0; i < l_width; ++i)
+      std::cout << "_";
+    std::cout << std::endl;
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::print_conv_footer(
+    const BlockVector<double> &solution_delta)
+  {
+    static const unsigned int l_width = 144;
+    for (unsigned int i = 0; i < l_width; ++i)
+      std::cout << "_";
+    std::cout << std::endl;
+    const std::pair<double, double> error_dil =
+      get_error_dilation(get_total_solution(solution_delta));
+    std::cout << "Relative errors:" << std::endl
+              << "Displacement:\t" << error_update.u / error_update_0.u
+              << std::endl
+              << "Force: \t\t" << error_residual.u / error_residual_0.u
+              << std::endl
+              << "Dilatation:\t" << error_dil.first << std::endl
+              << "v / V_0:\t" << error_dil.second * vol_reference << " / "
+              << vol_reference << " = " << error_dil.second << std::endl;
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  std::pair<double, double>
+  Solid<dim, number_t, ad_type_code>::get_error_dilation(
+    const BlockVector<double> &solution_total) const
+  {
+    double        vol_current  = 0.0;
+    double        dil_L2_error = 0.0;
+    FEValues<dim> fe_values_ref(
+      fe, qf_cell, update_values | update_gradients | update_JxW_values);
+    std::vector<Tensor<2, dim>> solution_grads_u_total(qf_cell.size());
+    std::vector<double>         solution_values_J_total(qf_cell.size());
+    for (typename DoFHandler<dim>::active_cell_iterator cell =
+           dof_handler_ref.begin_active();
+         cell != dof_handler_ref.end();
+         ++cell)
+      {
+        fe_values_ref.reinit(cell);
+        fe_values_ref[u_fe].get_function_gradients(solution_total,
+                                                   solution_grads_u_total);
+        fe_values_ref[J_fe].get_function_values(solution_total,
+                                                solution_values_J_total);
+        const std::vector<std::shared_ptr<const PointHistory<dim>>> lqph =
+          quadrature_point_history.get_data(cell);
+        Assert(lqph.size() == n_q_points, ExcInternalError());
+        for (unsigned int q_point = 0; q_point < n_q_points; ++q_point)
+          {
+            const double det_F_qp = determinant(
+              StandardTensors<dim>::I + solution_grads_u_total[q_point]);
+            const double J_tilde_qp = solution_values_J_total[q_point];
+            const double the_error_qp_squared =
+              std::pow((det_F_qp - J_tilde_qp), 2);
+            const double JxW = fe_values_ref.JxW(q_point);
+            dil_L2_error += the_error_qp_squared * JxW;
+            vol_current += det_F_qp * JxW;
+          }
+      }
+    Assert(vol_current > 0.0, ExcInternalError());
+
+    return std::make_pair(std::sqrt(dil_L2_error), vol_current / vol_reference);
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::get_error_residual(Errors &error_residual)
+  {
+    BlockVector<double> error_res(dofs_per_block);
+    for (unsigned int i = 0; i < dof_handler_ref.n_dofs(); ++i)
+      if (!constraints.is_constrained(i))
+        error_res(i) = system_rhs(i);
+    error_residual.norm = error_res.l2_norm();
+    error_residual.u    = error_res.block(u_dof).l2_norm();
+    error_residual.p    = error_res.block(p_dof).l2_norm();
+    error_residual.J    = error_res.block(J_dof).l2_norm();
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::get_error_update(
+    const BlockVector<double> &newton_update,
+    Errors &                   error_update)
+  {
+    BlockVector<double> error_ud(dofs_per_block);
+    for (unsigned int i = 0; i < dof_handler_ref.n_dofs(); ++i)
+      if (!constraints.is_constrained(i))
+        error_ud(i) = newton_update(i);
+    error_update.norm = error_ud.l2_norm();
+    error_update.u    = error_ud.block(u_dof).l2_norm();
+    error_update.p    = error_ud.block(p_dof).l2_norm();
+    error_update.J    = error_ud.block(J_dof).l2_norm();
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  BlockVector<double>
+  Solid<dim, number_t, ad_type_code>::get_total_solution(
+    const BlockVector<double> &solution_delta) const
+  {
+    BlockVector<double> solution_total(solution_n);
+    solution_total += solution_delta;
+    return solution_total;
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::assemble_system(
+    const BlockVector<double> &solution_delta)
+  {
+    timer.enter_subsection("Assemble system");
+    std::cout << " ASM_SYS " << std::flush;
+    tangent_matrix = 0.0;
+    system_rhs     = 0.0;
+    const BlockVector<double> solution_total(
+      get_total_solution(solution_delta));
+    const UpdateFlags uf_cell(update_values | update_gradients |
+                              update_JxW_values);
+    const UpdateFlags uf_face(update_values | update_normal_vectors |
+                              update_JxW_values);
+    PerTaskData_ASM   per_task_data(dofs_per_cell);
+    ScratchData_ASM   scratch_data(
+      fe, qf_cell, uf_cell, qf_face, uf_face, solution_total);
+    // ADOL-C is incompatible with TBB
+    // WorkStream::run(dof_handler_ref.begin_active(),
+    //                 dof_handler_ref.end(),
+    //                 std::bind(&Solid<dim,number_t,ad_type_code>::assemble_system_one_cell,
+    //                                 this,
+    //                                 std::placeholders::_1,
+    //                                 std::placeholders::_2,
+    //                                 std::placeholders::_3),
+    //                 std::bind(&Solid<dim,number_t,ad_type_code>::copy_local_to_global_system,
+    //                                 this,
+    //                                 std::placeholders::_1),
+    //                 scratch_data,
+    //                 per_task_data);
+    for (typename DoFHandler<dim>::active_cell_iterator cell =
+           dof_handler_ref.begin_active();
+         cell != dof_handler_ref.end();
+         ++cell)
+      {
+        assemble_system_one_cell(cell, scratch_data, per_task_data);
+        copy_local_to_global_system(per_task_data);
+      }
+    timer.leave_subsection();
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::copy_local_to_global_system(
+    const PerTaskData_ASM &data)
+  {
+    if (data.cell_matrix.frobenius_norm() > 1e-12)
+      constraints.distribute_local_to_global(data.cell_matrix,
+                                             data.cell_rhs,
+                                             data.local_dof_indices,
+                                             tangent_matrix,
+                                             system_rhs);
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::assemble_system_one_cell(
+    const typename DoFHandler<dim>::active_cell_iterator &cell,
+    ScratchData_ASM &                                     scratch,
+    PerTaskData_ASM &                                     data) const
+  {
+    data.reset();
+    scratch.reset();
+    scratch.fe_values_ref.reinit(cell);
+    cell->get_dof_indices(data.local_dof_indices);
+
+    const std::vector<std::shared_ptr<const PointHistory<dim>>> lqph =
+      quadrature_point_history.get_data(cell);
+    Assert(lqph.size() == n_q_points, ExcInternalError());
+
+    const unsigned int n_independent_variables = data.local_dof_indices.size();
+    const unsigned int n_dependent_variables   = dofs_per_cell;
+    Assert(n_dependent_variables == n_independent_variables,
+           ExcMessage("Expect square system."));
+
+    typedef AD::ADHelperResidualLinearization<ad_type_code, number_t> ADHelper;
+    typedef typename ADHelper::ad_type ADNumberType;
+    ADHelper ad_helper(n_independent_variables, n_dependent_variables);
+    ad_helper.set_tape_buffer_sizes(); // Increase the buffer size from the
+                                       // default values
+
+    const int  tape_no = 1;
+    const bool is_recording =
+      ad_helper.start_recording_operations(tape_no, //  material_id
+                                           true,    // overwrite_tape
+                                           true);   // keep
+
+    if (is_recording == true)
+      {
+        // Set the values for all DoFs
+        ad_helper.register_dof_values(scratch.solution_total,
+                                      data.local_dof_indices);
+        const std::vector<ADNumberType> dof_values_ad =
+          ad_helper.get_sensitive_dof_values();
+
+        // Compute all values, gradients etc. based on sensitive AD DoF values
+        std::vector<Tensor<2, dim, ADNumberType>> Grad_u(
+          n_q_points, Tensor<2, dim, ADNumberType>());
+        std::vector<ADNumberType> p_tilde(n_q_points, ADNumberType(0.0));
+        std::vector<ADNumberType> J_tilde(n_q_points, ADNumberType(0.0));
+        scratch.fe_values_ref[u_fe]
+          .get_function_gradients_from_local_dof_values(dof_values_ad, Grad_u);
+        scratch.fe_values_ref[p_fe].get_function_values_from_local_dof_values(
+          dof_values_ad, p_tilde);
+        scratch.fe_values_ref[J_fe].get_function_values_from_local_dof_values(
+          dof_values_ad, J_tilde);
+
+        // Compute the residual
+        // Note: It is critical that this vector be initialised with zero'd
+        // values otherwise the results may be garbage!
+        std::vector<ADNumberType> residual_ad(n_dependent_variables,
+                                              ADNumberType(0.0));
+        for (unsigned int q_point = 0; q_point < n_q_points; ++q_point)
+          {
+            const Tensor<2, dim, ADNumberType> F =
+              unit_symmetric_tensor<dim>() + Grad_u[q_point];
+            const Tensor<2, dim, ADNumberType> F_inv = invert(F);
+            const ADNumberType                 det_F = determinant(F);
+            Assert(numbers::value_is_greater_than(det_F, 0.0),
+                   ExcMessage("Negative jacobian detected!"));
+
+            const SymmetricTensor<2, dim, ADNumberType> tau =
+              lqph[q_point]->get_tau(F, p_tilde[q_point]);
+            const ADNumberType dPsi_vol_dJ =
+              lqph[q_point]->get_dPsi_vol_dJ(J_tilde[q_point]);
+
+            const double JxW = scratch.fe_values_ref.JxW(q_point);
+            for (unsigned int i = 0; i < dofs_per_cell; ++i)
+              {
+                const unsigned int i_group =
+                  fe.system_to_base_index(i).first.first;
+                if (i_group == u_dof)
+                  {
+                    const SymmetricTensor<2, dim, ADNumberType> symm_grad_Nx_i =
+                      symmetrize(
+                        scratch.fe_values_ref[u_fe].gradient(i, q_point) *
+                        F_inv);
+                    residual_ad[i] += (symm_grad_Nx_i * tau) * JxW;
+                  }
+                else if (i_group == p_dof)
+                  {
+                    const double N_i =
+                      scratch.fe_values_ref[p_fe].value(i, q_point);
+                    residual_ad[i] += N_i * (det_F - J_tilde[q_point]) * JxW;
+                  }
+                else if (i_group == J_dof)
+                  {
+                    const double N_i =
+                      scratch.fe_values_ref[J_fe].value(i, q_point);
+                    residual_ad[i] +=
+                      N_i * (dPsi_vol_dJ - p_tilde[q_point]) * JxW;
+                  }
+                else
+                  Assert(i_group <= J_dof, ExcInternalError());
+              }
+          }
+        for (unsigned int face = 0; face < GeometryInfo<dim>::faces_per_cell;
+             ++face)
+          if (cell->face(face)->at_boundary() == true &&
+              cell->face(face)->boundary_id() == 6)
+            {
+              scratch.fe_face_values_ref.reinit(cell, face);
+              for (unsigned int f_q_point = 0; f_q_point < n_q_points_f;
+                   ++f_q_point)
+                {
+                  const Tensor<1, dim> &N =
+                    scratch.fe_face_values_ref.normal_vector(f_q_point);
+                  static const double p0 =
+                    -4.0 / (parameters.scale * parameters.scale);
+                  const double time_ramp = (time.current() / time.end());
+                  const double pressure  = p0 * parameters.p_p0 * time_ramp;
+                  const Tensor<1, dim> traction = pressure * N;
+                  for (unsigned int i = 0; i < dofs_per_cell; ++i)
+                    {
+                      const unsigned int i_group =
+                        fe.system_to_base_index(i).first.first;
+                      if (i_group == u_dof)
+                        {
+                          const unsigned int component_i =
+                            fe.system_to_component_index(i).first;
+                          const double Ni =
+                            scratch.fe_face_values_ref.shape_value(i,
+                                                                   f_q_point);
+                          const double JxW =
+                            scratch.fe_face_values_ref.JxW(f_q_point);
+                          residual_ad[i] -= (Ni * traction[component_i]) * JxW;
+                        }
+                    }
+                }
+            }
+
+        ad_helper.register_residual_vector(residual_ad);
+        ad_helper.stop_recording_operations(false /*write_tapes_to_file*/);
+      }
+    else
+      {
+        Assert(is_recording == true, ExcInternalError());
+      }
+
+    // Unnecessary when keep == true
+    // ad_helper.activate_recorded_tape(tape_no);
+    // ad_helper.set_dof_values(scratch.solution_total,
+    //                          data.local_dof_indices);
+
+    // Compute the residual values and their jacobian for the new evaluation
+    // point
+    ad_helper.compute_residual(data.cell_rhs);
+    data.cell_rhs *= -1.0; // RHS = - residual
+    ad_helper.compute_linearization(data.cell_matrix);
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::make_constraints(const int &it_nr)
+  {
+    std::cout << " CST " << std::flush;
+    if (it_nr > 1)
+      return;
+    constraints.clear();
+    const bool                       apply_dirichlet_bc = (it_nr == 0);
+    const FEValuesExtractors::Scalar x_displacement(0);
+    const FEValuesExtractors::Scalar y_displacement(1);
+    {
+      const int boundary_id = 0;
+      if (apply_dirichlet_bc == true)
+        VectorTools::interpolate_boundary_values(
+          dof_handler_ref,
+          boundary_id,
+          ZeroFunction<dim>(n_components),
+          constraints,
+          fe.component_mask(x_displacement));
+      else
+        VectorTools::interpolate_boundary_values(
+          dof_handler_ref,
+          boundary_id,
+          ZeroFunction<dim>(n_components),
+          constraints,
+          fe.component_mask(x_displacement));
+    }
+    {
+      const int boundary_id = 2;
+      if (apply_dirichlet_bc == true)
+        VectorTools::interpolate_boundary_values(
+          dof_handler_ref,
+          boundary_id,
+          ZeroFunction<dim>(n_components),
+          constraints,
+          fe.component_mask(y_displacement));
+      else
+        VectorTools::interpolate_boundary_values(
+          dof_handler_ref,
+          boundary_id,
+          ZeroFunction<dim>(n_components),
+          constraints,
+          fe.component_mask(y_displacement));
+    }
+    if (dim == 3)
+      {
+        const FEValuesExtractors::Scalar z_displacement(2);
+        {
+          const int boundary_id = 3;
+          if (apply_dirichlet_bc == true)
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              (fe.component_mask(x_displacement) |
+               fe.component_mask(z_displacement)));
+          else
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              (fe.component_mask(x_displacement) |
+               fe.component_mask(z_displacement)));
+        }
+        {
+          const int boundary_id = 4;
+          if (apply_dirichlet_bc == true)
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              fe.component_mask(z_displacement));
+          else
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              fe.component_mask(z_displacement));
+        }
+        {
+          const int boundary_id = 6;
+          if (apply_dirichlet_bc == true)
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              (fe.component_mask(x_displacement) |
+               fe.component_mask(z_displacement)));
+          else
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              (fe.component_mask(x_displacement) |
+               fe.component_mask(z_displacement)));
+        }
+      }
+    else
+      {
+        {
+          const int boundary_id = 3;
+          if (apply_dirichlet_bc == true)
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              (fe.component_mask(x_displacement)));
+          else
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              (fe.component_mask(x_displacement)));
+        }
+        {
+          const int boundary_id = 6;
+          if (apply_dirichlet_bc == true)
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              (fe.component_mask(x_displacement)));
+          else
+            VectorTools::interpolate_boundary_values(
+              dof_handler_ref,
+              boundary_id,
+              ZeroFunction<dim>(n_components),
+              constraints,
+              (fe.component_mask(x_displacement)));
+        }
+      }
+    constraints.close();
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::assemble_sc()
+  {
+    timer.enter_subsection("Perform static condensation");
+    std::cout << " ASM_SC " << std::flush;
+    PerTaskData_SC per_task_data(dofs_per_cell,
+                                 element_indices_u.size(),
+                                 element_indices_p.size(),
+                                 element_indices_J.size());
+    ScratchData_SC scratch_data;
+    WorkStream::run(dof_handler_ref.begin_active(),
+                    dof_handler_ref.end(),
+                    *this,
+                    &Solid::assemble_sc_one_cell,
+                    &Solid::copy_local_to_global_sc,
+                    scratch_data,
+                    per_task_data);
+    timer.leave_subsection();
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::copy_local_to_global_sc(
+    const PerTaskData_SC &data)
+  {
+    for (unsigned int i = 0; i < dofs_per_cell; ++i)
+      for (unsigned int j = 0; j < dofs_per_cell; ++j)
+        tangent_matrix.add(data.local_dof_indices[i],
+                           data.local_dof_indices[j],
+                           data.cell_matrix(i, j));
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::assemble_sc_one_cell(
+    const typename DoFHandler<dim>::active_cell_iterator &cell,
+    ScratchData_SC &                                      scratch,
+    PerTaskData_SC &                                      data)
+  {
+    data.reset();
+    scratch.reset();
+    cell->get_dof_indices(data.local_dof_indices);
+    data.k_orig.extract_submatrix_from(tangent_matrix,
+                                       data.local_dof_indices,
+                                       data.local_dof_indices);
+    data.k_pu.extract_submatrix_from(data.k_orig,
+                                     element_indices_p,
+                                     element_indices_u);
+    data.k_pJ.extract_submatrix_from(data.k_orig,
+                                     element_indices_p,
+                                     element_indices_J);
+    data.k_JJ.extract_submatrix_from(data.k_orig,
+                                     element_indices_J,
+                                     element_indices_J);
+    data.k_pJ_inv.invert(data.k_pJ);
+    data.k_pJ_inv.mmult(data.A, data.k_pu);
+    data.k_JJ.mmult(data.B, data.A);
+    data.k_pJ_inv.Tmmult(data.C, data.B);
+    data.k_pu.Tmmult(data.k_bbar, data.C);
+    data.k_bbar.scatter_matrix_to(element_indices_u,
+                                  element_indices_u,
+                                  data.cell_matrix);
+    data.k_pJ_inv.add(-1.0, data.k_pJ);
+    data.k_pJ_inv.scatter_matrix_to(element_indices_p,
+                                    element_indices_J,
+                                    data.cell_matrix);
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  std::pair<unsigned int, double>
+  Solid<dim, number_t, ad_type_code>::solve_linear_system(
+    BlockVector<double> &newton_update)
+  {
+    unsigned int lin_it  = 0;
+    double       lin_res = 0.0;
+    if (parameters.use_static_condensation == true)
+      {
+        BlockVector<double> A(dofs_per_block);
+        BlockVector<double> B(dofs_per_block);
+        {
+          assemble_sc();
+          tangent_matrix.block(p_dof, J_dof)
+            .vmult(A.block(J_dof), system_rhs.block(p_dof));
+          tangent_matrix.block(J_dof, J_dof)
+            .vmult(B.block(J_dof), A.block(J_dof));
+          A.block(J_dof) = system_rhs.block(J_dof);
+          A.block(J_dof) -= B.block(J_dof);
+          tangent_matrix.block(p_dof, J_dof)
+            .Tvmult(A.block(p_dof), A.block(J_dof));
+          tangent_matrix.block(u_dof, p_dof)
+            .vmult(A.block(u_dof), A.block(p_dof));
+          system_rhs.block(u_dof) -= A.block(u_dof);
+          timer.enter_subsection("Linear solver");
+          std::cout << " SLV " << std::flush;
+          if (parameters.type_lin == "CG")
+            {
+              const int solver_its = tangent_matrix.block(u_dof, u_dof).m() *
+                                     parameters.max_iterations_lin;
+              const double tol_sol =
+                parameters.tol_lin * system_rhs.block(u_dof).l2_norm();
+              SolverControl solver_control(solver_its, tol_sol, false, false);
+              GrowingVectorMemory<Vector<double>> GVM;
+              SolverCG<Vector<double>> solver_CG(solver_control, GVM);
+              PreconditionSelector<SparseMatrix<double>, Vector<double>>
+                preconditioner(parameters.preconditioner_type,
+                               parameters.preconditioner_relaxation);
+              preconditioner.use_matrix(tangent_matrix.block(u_dof, u_dof));
+              solver_CG.solve(tangent_matrix.block(u_dof, u_dof),
+                              newton_update.block(u_dof),
+                              system_rhs.block(u_dof),
+                              preconditioner);
+              lin_it  = solver_control.last_step();
+              lin_res = solver_control.last_value();
+            }
+          else if (parameters.type_lin == "Direct")
+            {
+              SparseDirectUMFPACK A_direct;
+              A_direct.initialize(tangent_matrix.block(u_dof, u_dof));
+              A_direct.vmult(newton_update.block(u_dof),
+                             system_rhs.block(u_dof));
+              lin_it  = 1;
+              lin_res = 0.0;
+            }
+          else
+            Assert(false, ExcMessage("Linear solver type not implemented"));
+          timer.leave_subsection();
+        }
+        constraints.distribute(newton_update);
+        timer.enter_subsection("Linear solver postprocessing");
+        std::cout << " PP " << std::flush;
+        {
+          tangent_matrix.block(p_dof, u_dof)
+            .vmult(A.block(p_dof), newton_update.block(u_dof));
+          A.block(p_dof) *= -1.0;
+          A.block(p_dof) += system_rhs.block(p_dof);
+          tangent_matrix.block(p_dof, J_dof)
+            .vmult(newton_update.block(J_dof), A.block(p_dof));
+        }
+        constraints.distribute(newton_update);
+        {
+          tangent_matrix.block(J_dof, J_dof)
+            .vmult(A.block(J_dof), newton_update.block(J_dof));
+          A.block(J_dof) *= -1.0;
+          A.block(J_dof) += system_rhs.block(J_dof);
+          tangent_matrix.block(p_dof, J_dof)
+            .Tvmult(newton_update.block(p_dof), A.block(J_dof));
+        }
+        constraints.distribute(newton_update);
+        timer.leave_subsection();
+      }
+    else
+      {
+        std::cout << " ------ " << std::flush;
+        timer.enter_subsection("Linear solver");
+        std::cout << " SLV " << std::flush;
+        if (parameters.type_lin == "CG")
+          {
+            const Vector<double> &f_u = system_rhs.block(u_dof);
+            const Vector<double> &f_p = system_rhs.block(p_dof);
+            const Vector<double> &f_J = system_rhs.block(J_dof);
+            Vector<double> &      d_u = newton_update.block(u_dof);
+            Vector<double> &      d_p = newton_update.block(p_dof);
+            Vector<double> &      d_J = newton_update.block(J_dof);
+            const auto            K_uu =
+              linear_operator(tangent_matrix.block(u_dof, u_dof));
+            const auto K_up =
+              linear_operator(tangent_matrix.block(u_dof, p_dof));
+            const auto K_pu =
+              linear_operator(tangent_matrix.block(p_dof, u_dof));
+            const auto K_Jp =
+              linear_operator(tangent_matrix.block(J_dof, p_dof));
+            const auto K_JJ =
+              linear_operator(tangent_matrix.block(J_dof, J_dof));
+            PreconditionSelector<SparseMatrix<double>, Vector<double>>
+              preconditioner_K_Jp_inv("jacobi");
+            preconditioner_K_Jp_inv.use_matrix(
+              tangent_matrix.block(J_dof, p_dof));
+            ReductionControl solver_control_K_Jp_inv(
+              tangent_matrix.block(J_dof, p_dof).m() *
+                parameters.max_iterations_lin,
+              1.0e-30,
+              parameters.tol_lin);
+            SolverSelector<Vector<double>> solver_K_Jp_inv;
+            solver_K_Jp_inv.select("cg");
+            solver_K_Jp_inv.set_control(solver_control_K_Jp_inv);
+            const auto K_Jp_inv =
+              inverse_operator(K_Jp, solver_K_Jp_inv, preconditioner_K_Jp_inv);
+            const auto K_pJ_inv     = transpose_operator(K_Jp_inv);
+            const auto K_pp_bar     = K_Jp_inv * K_JJ * K_pJ_inv;
+            const auto K_uu_bar_bar = K_up * K_pp_bar * K_pu;
+            const auto K_uu_con     = K_uu + K_uu_bar_bar;
+            PreconditionSelector<SparseMatrix<double>, Vector<double>>
+              preconditioner_K_con_inv(parameters.preconditioner_type,
+                                       parameters.preconditioner_relaxation);
+            preconditioner_K_con_inv.use_matrix(
+              tangent_matrix.block(u_dof, u_dof));
+            ReductionControl solver_control_K_con_inv(
+              tangent_matrix.block(u_dof, u_dof).m() *
+                parameters.max_iterations_lin,
+              1.0e-30,
+              parameters.tol_lin);
+            SolverSelector<Vector<double>> solver_K_con_inv;
+            solver_K_con_inv.select("cg");
+            solver_K_con_inv.set_control(solver_control_K_con_inv);
+            const auto K_uu_con_inv =
+              inverse_operator(K_uu_con,
+                               solver_K_con_inv,
+                               preconditioner_K_con_inv);
+            d_u =
+              K_uu_con_inv * (f_u - K_up * (K_Jp_inv * f_J - K_pp_bar * f_p));
+            timer.leave_subsection();
+            timer.enter_subsection("Linear solver postprocessing");
+            std::cout << " PP " << std::flush;
+            d_J     = K_pJ_inv * (f_p - K_pu * d_u);
+            d_p     = K_Jp_inv * (f_J - K_JJ * d_J);
+            lin_it  = solver_control_K_con_inv.last_step();
+            lin_res = solver_control_K_con_inv.last_value();
+          }
+        else if (parameters.type_lin == "Direct")
+          {
+            SparseDirectUMFPACK A_direct;
+            A_direct.initialize(tangent_matrix);
+            A_direct.vmult(newton_update, system_rhs);
+            lin_it  = 1;
+            lin_res = 0.0;
+            std::cout << " -- " << std::flush;
+          }
+        else
+          Assert(false, ExcMessage("Linear solver type not implemented"));
+        timer.leave_subsection();
+        constraints.distribute(newton_update);
+      }
+    return std::make_pair(lin_it, lin_res);
+  }
+  template <int dim, typename number_t, enum AD::NumberTypes ad_type_code>
+  void
+  Solid<dim, number_t, ad_type_code>::output_results() const
+  {
+    DataOut<dim> data_out;
+    std::vector<DataComponentInterpretation::DataComponentInterpretation>
+      data_component_interpretation(
+        dim, DataComponentInterpretation::component_is_part_of_vector);
+    data_component_interpretation.push_back(
+      DataComponentInterpretation::component_is_scalar);
+    data_component_interpretation.push_back(
+      DataComponentInterpretation::component_is_scalar);
+    std::vector<std::string> solution_name(dim, "displacement");
+    solution_name.push_back("pressure");
+    solution_name.push_back("dilatation");
+    data_out.attach_dof_handler(dof_handler_ref);
+    data_out.add_data_vector(solution_n,
+                             solution_name,
+                             DataOut<dim>::type_dof_data,
+                             data_component_interpretation);
+    Vector<double> soln(solution_n.size());
+    for (unsigned int i = 0; i < soln.size(); ++i)
+      soln(i) = solution_n(i);
+    MappingQEulerian<dim> q_mapping(degree, dof_handler_ref, soln);
+    data_out.build_patches(q_mapping, degree);
+    std::ostringstream filename;
+    filename << "solution-" << dim << "d-" << time.get_timestep() << ".vtk";
+    std::ofstream output(filename.str().c_str());
+    data_out.write_vtk(output);
+  }
+} // namespace Step44
diff --git a/tests/adolc/step-44-helper_res_lin_01_1.cc b/tests/adolc/step-44-helper_res_lin_01_1.cc
new file mode 100644 (file)
index 0000000..86555ce
--- /dev/null
@@ -0,0 +1,68 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2016 - 2018 by the deal.II authors
+//
+// This file is part of the deal.II library.
+//
+// The deal.II library is free software; you can use it, 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.
+// The full text of the license can be found in the file LICENSE at
+// the top level of the deal.II distribution.
+//
+// ---------------------------------------------------------------------
+
+// This is a modified version of step-44, which tests the implementation of
+// cell-level auto-differentiation (linearisation of a residual vector).
+//
+// AD number type: ADOL-C taped
+
+#include "../ad_common_tests/step-44-helper_res_lin_01.h"
+#include "../tests.h"
+
+int
+main(int argc, char **argv)
+{
+  initlog();
+
+  Utilities::MPI::MPI_InitFinalize mpi_initialization(
+    argc, argv, testing_max_num_threads());
+
+  using namespace dealii;
+  namespace AD = dealii::Differentiation::AD;
+  using namespace Step44;
+  try
+    {
+      const unsigned int                               dim = 2;
+      Solid<dim, double, AD::NumberTypes::adolc_taped> solid(
+        SOURCE_DIR "/prm/parameters-step-44.prm");
+      solid.run();
+    }
+  catch (std::exception &exc)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Exception on processing: " << std::endl
+                << exc.what() << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  catch (...)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Unknown exception!" << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  return 0;
+}
diff --git a/tests/adolc/step-44-helper_res_lin_01_1.output b/tests/adolc/step-44-helper_res_lin_01_1.output
new file mode 100644 (file)
index 0000000..8ba79ae
--- /dev/null
@@ -0,0 +1,11 @@
+
+DEAL::Timestep 1: 0.00 -0.000119
+DEAL::Timestep 2: 0.00 -0.000228
+DEAL::Timestep 3: 0.00 -0.000322
+DEAL::Timestep 4: 0.00 -0.000402
+DEAL::Timestep 5: 0.00 -0.000470
+DEAL::Timestep 6: 0.00 -0.000527
+DEAL::Timestep 7: 0.00 -0.000576
+DEAL::Timestep 8: 0.00 -0.000617
+DEAL::Timestep 9: 0.00 -0.000653
+DEAL::Timestep 10: 0.00 -0.000683
diff --git a/tests/adolc/step-44-helper_res_lin_01_2.cc b/tests/adolc/step-44-helper_res_lin_01_2.cc
new file mode 100644 (file)
index 0000000..a86d4e2
--- /dev/null
@@ -0,0 +1,68 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2016 - 2018 by the deal.II authors
+//
+// This file is part of the deal.II library.
+//
+// The deal.II library is free software; you can use it, 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.
+// The full text of the license can be found in the file LICENSE at
+// the top level of the deal.II distribution.
+//
+// ---------------------------------------------------------------------
+
+// This is a modified version of step-44, which tests the implementation of
+// cell-level auto-differentiation (linearisation of a residual vector).
+//
+// AD number type: ADOL-C tapeless
+
+#include "../ad_common_tests/step-44-helper_res_lin_01.h"
+#include "../tests.h"
+
+int
+main(int argc, char **argv)
+{
+  initlog();
+
+  Utilities::MPI::MPI_InitFinalize mpi_initialization(
+    argc, argv, testing_max_num_threads());
+
+  using namespace dealii;
+  namespace AD = dealii::Differentiation::AD;
+  using namespace Step44;
+  try
+    {
+      const unsigned int                                  dim = 2;
+      Solid<dim, double, AD::NumberTypes::adolc_tapeless> solid(
+        SOURCE_DIR "/prm/parameters-step-44.prm");
+      solid.run();
+    }
+  catch (std::exception &exc)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Exception on processing: " << std::endl
+                << exc.what() << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  catch (...)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Unknown exception!" << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  return 0;
+}
diff --git a/tests/adolc/step-44-helper_res_lin_01_2.output b/tests/adolc/step-44-helper_res_lin_01_2.output
new file mode 100644 (file)
index 0000000..8ba79ae
--- /dev/null
@@ -0,0 +1,11 @@
+
+DEAL::Timestep 1: 0.00 -0.000119
+DEAL::Timestep 2: 0.00 -0.000228
+DEAL::Timestep 3: 0.00 -0.000322
+DEAL::Timestep 4: 0.00 -0.000402
+DEAL::Timestep 5: 0.00 -0.000470
+DEAL::Timestep 6: 0.00 -0.000527
+DEAL::Timestep 7: 0.00 -0.000576
+DEAL::Timestep 8: 0.00 -0.000617
+DEAL::Timestep 9: 0.00 -0.000653
+DEAL::Timestep 10: 0.00 -0.000683
diff --git a/tests/sacado/step-44-helper_res_lin_01_1.cc b/tests/sacado/step-44-helper_res_lin_01_1.cc
new file mode 100644 (file)
index 0000000..7bd0c69
--- /dev/null
@@ -0,0 +1,68 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2016 - 2018 by the deal.II authors
+//
+// This file is part of the deal.II library.
+//
+// The deal.II library is free software; you can use it, 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.
+// The full text of the license can be found in the file LICENSE at
+// the top level of the deal.II distribution.
+//
+// ---------------------------------------------------------------------
+
+// This is a modified version of step-44, which tests the implementation of
+// cell-level auto-differentiation (linearisation of a residual vector).
+//
+// AD number type: Sacado DFad
+
+#include "../ad_common_tests/step-44-helper_res_lin_01.h"
+#include "../tests.h"
+
+int
+main(int argc, char **argv)
+{
+  initlog();
+
+  Utilities::MPI::MPI_InitFinalize mpi_initialization(
+    argc, argv, testing_max_num_threads());
+
+  using namespace dealii;
+  namespace AD = dealii::Differentiation::AD;
+  using namespace Step44;
+  try
+    {
+      const unsigned int                               dim = 2;
+      Solid<dim, double, AD::NumberTypes::sacado_dfad> solid(
+        SOURCE_DIR "/prm/parameters-step-44.prm");
+      solid.run();
+    }
+  catch (std::exception &exc)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Exception on processing: " << std::endl
+                << exc.what() << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  catch (...)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Unknown exception!" << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  return 0;
+}
diff --git a/tests/sacado/step-44-helper_res_lin_01_1.output b/tests/sacado/step-44-helper_res_lin_01_1.output
new file mode 100644 (file)
index 0000000..8ba79ae
--- /dev/null
@@ -0,0 +1,11 @@
+
+DEAL::Timestep 1: 0.00 -0.000119
+DEAL::Timestep 2: 0.00 -0.000228
+DEAL::Timestep 3: 0.00 -0.000322
+DEAL::Timestep 4: 0.00 -0.000402
+DEAL::Timestep 5: 0.00 -0.000470
+DEAL::Timestep 6: 0.00 -0.000527
+DEAL::Timestep 7: 0.00 -0.000576
+DEAL::Timestep 8: 0.00 -0.000617
+DEAL::Timestep 9: 0.00 -0.000653
+DEAL::Timestep 10: 0.00 -0.000683
diff --git a/tests/sacado/step-44-helper_res_lin_01_2.cc b/tests/sacado/step-44-helper_res_lin_01_2.cc
new file mode 100644 (file)
index 0000000..b0ac5f5
--- /dev/null
@@ -0,0 +1,68 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2016 - 2018 by the deal.II authors
+//
+// This file is part of the deal.II library.
+//
+// The deal.II library is free software; you can use it, 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.
+// The full text of the license can be found in the file LICENSE at
+// the top level of the deal.II distribution.
+//
+// ---------------------------------------------------------------------
+
+// This is a modified version of step-44, which tests the implementation of
+// cell-level auto-differentiation (linearisation of a residual vector).
+//
+// AD number type: Sacado DFad-DFad
+
+#include "../ad_common_tests/step-44-helper_res_lin_01.h"
+#include "../tests.h"
+
+int
+main(int argc, char **argv)
+{
+  initlog();
+
+  Utilities::MPI::MPI_InitFinalize mpi_initialization(
+    argc, argv, testing_max_num_threads());
+
+  using namespace dealii;
+  namespace AD = dealii::Differentiation::AD;
+  using namespace Step44;
+  try
+    {
+      const unsigned int                                    dim = 2;
+      Solid<dim, double, AD::NumberTypes::sacado_dfad_dfad> solid(
+        SOURCE_DIR "/prm/parameters-step-44.prm");
+      solid.run();
+    }
+  catch (std::exception &exc)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Exception on processing: " << std::endl
+                << exc.what() << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  catch (...)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Unknown exception!" << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  return 0;
+}
diff --git a/tests/sacado/step-44-helper_res_lin_01_2.output b/tests/sacado/step-44-helper_res_lin_01_2.output
new file mode 100644 (file)
index 0000000..8ba79ae
--- /dev/null
@@ -0,0 +1,11 @@
+
+DEAL::Timestep 1: 0.00 -0.000119
+DEAL::Timestep 2: 0.00 -0.000228
+DEAL::Timestep 3: 0.00 -0.000322
+DEAL::Timestep 4: 0.00 -0.000402
+DEAL::Timestep 5: 0.00 -0.000470
+DEAL::Timestep 6: 0.00 -0.000527
+DEAL::Timestep 7: 0.00 -0.000576
+DEAL::Timestep 8: 0.00 -0.000617
+DEAL::Timestep 9: 0.00 -0.000653
+DEAL::Timestep 10: 0.00 -0.000683
diff --git a/tests/sacado/step-44-helper_res_lin_01_3.cc b/tests/sacado/step-44-helper_res_lin_01_3.cc
new file mode 100644 (file)
index 0000000..8b9be16
--- /dev/null
@@ -0,0 +1,68 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2016 - 2018 by the deal.II authors
+//
+// This file is part of the deal.II library.
+//
+// The deal.II library is free software; you can use it, 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.
+// The full text of the license can be found in the file LICENSE at
+// the top level of the deal.II distribution.
+//
+// ---------------------------------------------------------------------
+
+// This is a modified version of step-44, which tests the implementation of
+// cell-level auto-differentiation (linearisation of a residual vector).
+//
+// AD number type: Sacado Rad
+
+#include "../ad_common_tests/step-44-helper_res_lin_01.h"
+#include "../tests.h"
+
+int
+main(int argc, char **argv)
+{
+  initlog();
+
+  Utilities::MPI::MPI_InitFinalize mpi_initialization(
+    argc, argv, testing_max_num_threads());
+
+  using namespace dealii;
+  namespace AD = dealii::Differentiation::AD;
+  using namespace Step44;
+  try
+    {
+      const unsigned int                              dim = 2;
+      Solid<dim, double, AD::NumberTypes::sacado_rad> solid(
+        SOURCE_DIR "/prm/parameters-step-44.prm");
+      solid.run();
+    }
+  catch (std::exception &exc)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Exception on processing: " << std::endl
+                << exc.what() << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  catch (...)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Unknown exception!" << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  return 0;
+}
diff --git a/tests/sacado/step-44-helper_res_lin_01_3.output b/tests/sacado/step-44-helper_res_lin_01_3.output
new file mode 100644 (file)
index 0000000..8ba79ae
--- /dev/null
@@ -0,0 +1,11 @@
+
+DEAL::Timestep 1: 0.00 -0.000119
+DEAL::Timestep 2: 0.00 -0.000228
+DEAL::Timestep 3: 0.00 -0.000322
+DEAL::Timestep 4: 0.00 -0.000402
+DEAL::Timestep 5: 0.00 -0.000470
+DEAL::Timestep 6: 0.00 -0.000527
+DEAL::Timestep 7: 0.00 -0.000576
+DEAL::Timestep 8: 0.00 -0.000617
+DEAL::Timestep 9: 0.00 -0.000653
+DEAL::Timestep 10: 0.00 -0.000683
diff --git a/tests/sacado/step-44-helper_res_lin_01_4.cc b/tests/sacado/step-44-helper_res_lin_01_4.cc
new file mode 100644 (file)
index 0000000..a77459a
--- /dev/null
@@ -0,0 +1,68 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2016 - 2018 by the deal.II authors
+//
+// This file is part of the deal.II library.
+//
+// The deal.II library is free software; you can use it, 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.
+// The full text of the license can be found in the file LICENSE at
+// the top level of the deal.II distribution.
+//
+// ---------------------------------------------------------------------
+
+// This is a modified version of step-44, which tests the implementation of
+// cell-level auto-differentiation (linearisation of a residual vector).
+//
+// AD number type: Sacado Rad-DFad
+
+#include "../ad_common_tests/step-44-helper_res_lin_01.h"
+#include "../tests.h"
+
+int
+main(int argc, char **argv)
+{
+  initlog();
+
+  Utilities::MPI::MPI_InitFinalize mpi_initialization(
+    argc, argv, testing_max_num_threads());
+
+  using namespace dealii;
+  namespace AD = dealii::Differentiation::AD;
+  using namespace Step44;
+  try
+    {
+      const unsigned int                                   dim = 2;
+      Solid<dim, double, AD::NumberTypes::sacado_rad_dfad> solid(
+        SOURCE_DIR "/prm/parameters-step-44.prm");
+      solid.run();
+    }
+  catch (std::exception &exc)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Exception on processing: " << std::endl
+                << exc.what() << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  catch (...)
+    {
+      std::cerr << std::endl
+                << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      std::cerr << "Unknown exception!" << std::endl
+                << "Aborting!" << std::endl
+                << "----------------------------------------------------"
+                << std::endl;
+      return 1;
+    }
+  return 0;
+}
diff --git a/tests/sacado/step-44-helper_res_lin_01_4.output b/tests/sacado/step-44-helper_res_lin_01_4.output
new file mode 100644 (file)
index 0000000..8ba79ae
--- /dev/null
@@ -0,0 +1,11 @@
+
+DEAL::Timestep 1: 0.00 -0.000119
+DEAL::Timestep 2: 0.00 -0.000228
+DEAL::Timestep 3: 0.00 -0.000322
+DEAL::Timestep 4: 0.00 -0.000402
+DEAL::Timestep 5: 0.00 -0.000470
+DEAL::Timestep 6: 0.00 -0.000527
+DEAL::Timestep 7: 0.00 -0.000576
+DEAL::Timestep 8: 0.00 -0.000617
+DEAL::Timestep 9: 0.00 -0.000653
+DEAL::Timestep 10: 0.00 -0.000683

In the beginning the Universe was created. This has made a lot of people very angry and has been widely regarded as a bad move.

Douglas Adams


Typeset in Trocchi and Trocchi Bold Sans Serif.