step-47.h

Go to the documentation of this file.

 
 = 0) const override
 *         {
 *           return std::sin(PI * p[0]) * std::sin(PI * p[1]);
 *         }
 *   
 *         virtual Tensor<1, dim>
 *         gradient(const Point<dim> &p,
 *                  const unsigned int /*component*/ = 0) const override
 *         {
 *           Tensor<1, dim> r;
 *           r[0] = PI * std::cos(PI * p[0]) * std::sin(PI * p[1]);
 *           r[1] = PI * std::cos(PI * p[1]) * std::sin(PI * p[0]);
 *           return r;
 *         }
 *   
 *         virtual void
 *         hessian_list(const std::vector<Point<dim>>        &points,
 *                      std::vector<SymmetricTensor<2, dim>> &hessians,
 *                      const unsigned int /*component*/ = 0) const override
 *         {
 *           for (unsigned i = 0; i < points.size(); ++i)
 *             {
 *               const double x = points[i][0];
 *               const double y = points[i][1];
 *   
 *               hessians[i][0][0] = -PI * PI * std::sin(PI * x) * std::sin(PI * y);
 *               hessians[i][0][1] = PI * PI * std::cos(PI * x) * std::cos(PI * y);
 *               hessians[i][1][1] = -PI * PI * std::sin(PI * x) * std::sin(PI * y);
 *             }
 *         }
 *       };
 *   
 *   
 *       template <int dim>
 *       class RightHandSide : public Function<dim>
 *       {
 *       public:
 *         static_assert(dim == 2, "Only dim==2 is implemented");
 *   
 *         virtual double value(const Point<dim> &p,
 *                              const unsigned int /*component*/ = 0) const override
 *   
 *         {
 *           return 4 * Utilities::fixed_power<4>(PI) * std::sin(PI * p[0]) *
 *                  std::sin(PI * p[1]);
 *         }
 *       };
 *     } // namespace ExactSolution
 *   
 *   
 *   
 * @endcode
 * 
 * 
 * <a name="step_47-Themainclass"></a> 
 * <h3>The main class</h3>
 *   
 
 * 
 * The following is the principal class of this tutorial program. It has
 * the structure of many of the other tutorial programs and there should
 * really be nothing particularly surprising about its contents or
 * the constructor that follows it.
 * 
 * @code
 *     template <int dim>
 *     class BiharmonicProblem
 *     {
 *     public:
 *       BiharmonicProblem(const unsigned int fe_degree);
 *   
 *       void run();
 *   
 *     private:
 *       void make_grid();
 *       void setup_system();
 *       void assemble_system();
 *       void solve();
 *       void compute_errors();
 *       void output_results(const unsigned int iteration) const;
 *   
 *       Triangulation<dim> triangulation;
 *   
 *       MappingQ<dim> mapping;
 *   
 *       const FE_Q<dim>           fe;
 *       DoFHandler<dim>           dof_handler;
 *       AffineConstraints<double> constraints;
 *   
 *       SparsityPattern      sparsity_pattern;
 *       SparseMatrix<double> system_matrix;
 *   
 *       Vector<double> solution;
 *       Vector<double> system_rhs;
 *     };
 *   
 *   
 *   
 *     template <int dim>
 *     BiharmonicProblem<dim>::BiharmonicProblem(const unsigned int fe_degree)
 *       : mapping(1)
 *       , fe(fe_degree)
 *       , dof_handler(triangulation)
 *     {}
 *   
 *   
 *   
 * @endcode
 * 
 * Next up are the functions that create the initial mesh (a once refined
 * unit square) and set up the constraints, vectors, and matrices on
 * each mesh. Again, both of these are essentially unchanged from many
 * previous tutorial programs.
 * 
 * @code
 *     template <int dim>
 *     void BiharmonicProblem<dim>::make_grid()
 *     {
 *       GridGenerator::hyper_cube(triangulation, 0., 1.);
 *       triangulation.refine_global(1);
 *   
 *       std::cout << "Number of active cells: " << triangulation.n_active_cells()
 *                 << std::endl
 *                 << "Total number of cells: " << triangulation.n_cells()
 *                 << std::endl;
 *     }
 *   
 *   
 *   
 *     template <int dim>
 *     void BiharmonicProblem<dim>::setup_system()
 *     {
 *       dof_handler.distribute_dofs(fe);
 *   
 *       std::cout << "   Number of degrees of freedom: " << dof_handler.n_dofs()
 *                 << std::endl;
 *   
 *       constraints.clear();
 *       DoFTools::make_hanging_node_constraints(dof_handler, constraints);
 *   
 *       VectorTools::interpolate_boundary_values(dof_handler,
 *                                                0,
 *                                                ExactSolution::Solution<dim>(),
 *                                                constraints);
 *       constraints.close();
 *   
 *   
 *       DynamicSparsityPattern dsp(dof_handler.n_dofs());
 *       DoFTools::make_flux_sparsity_pattern(dof_handler, dsp, constraints, true);
 *       sparsity_pattern.copy_from(dsp);
 *       system_matrix.reinit(sparsity_pattern);
 *   
 *       solution.reinit(dof_handler.n_dofs());
 *       system_rhs.reinit(dof_handler.n_dofs());
 *     }
 *   
 *   
 *   
 * @endcode
 * 
 * 
 * <a name="step_47-Assemblingthelinearsystem"></a> 
 * <h4>Assembling the linear system</h4>
 *   
 
 * 
 * The following pieces of code are more interesting. They all relate to the
 * assembly of the linear system. While assembling the cell-interior terms
 * is not of great difficulty -- that works in essence like the assembly
 * of the corresponding terms of the Laplace equation, and you have seen
 * how this works in @ref step_4 "step-4" or @ref step_6 "step-6", for example -- the difficulty
 * is with the penalty terms in the formulation. These require the evaluation
 * of gradients of shape functions at interfaces of cells. At the least,
 * one would therefore need to use two FEFaceValues objects, but if one of the
 * two sides is adaptively refined, then one actually needs an FEFaceValues
 * and one FESubfaceValues objects; one also needs to keep track which
 * shape functions live where, and finally we need to ensure that every
 * face is visited only once. All of this is a substantial overhead to the
 * logic we really want to implement (namely the penalty terms in the
 * bilinear form). As a consequence, we will make use of the
 * FEInterfaceValues class -- a helper class in deal.II that allows us
 * to abstract away the two FEFaceValues or FESubfaceValues objects and
 * directly access what we really care about: jumps, averages, etc.
 *   
 
 * 
 * But this doesn't yet solve our problem of having to keep track of
 * which faces we have already visited when we loop over all cells and
 * all of their faces. To make this process simpler, we use the
 * MeshWorker::mesh_loop() function that provides a simple interface
 * for this task: Based on the ideas outlined in the WorkStream
 * namespace documentation, MeshWorker::mesh_loop() requires three
 * functions that do work on cells, interior faces, and boundary
 * faces. These functions work on scratch objects for intermediate
 * results, and then copy the result of their computations into
 * copy data objects from where a copier function copies them into
 * the global matrix and right hand side objects.
 *   
 
 * 
 * The following structures then provide the scratch and copy objects
 * that are necessary for this approach. You may look up the WorkStream
 * namespace as well as the
 * @ref threads "Parallel computing with multiple processors"
 * topic for more information on how they typically work.
 * 
 * @code
 *     template <int dim>
 *     struct ScratchData
 *     {
 *       ScratchData(const Mapping<dim>       &mapping,
 *                   const FiniteElement<dim> &fe,
 *                   const unsigned int        quadrature_degree,
 *                   const UpdateFlags         update_flags,
 *                   const UpdateFlags         interface_update_flags)
 *         : fe_values(mapping, fe, QGauss<dim>(quadrature_degree), update_flags)
 *         , fe_interface_values(mapping,
 *                               fe,
 *                               QGauss<dim - 1>(quadrature_degree),
 *                               interface_update_flags)
 *       {}
 *   
 *   
 *       ScratchData(const ScratchData<dim> &scratch_data)
 *         : fe_values(scratch_data.fe_values.get_mapping(),
 *                     scratch_data.fe_values.get_fe(),
 *                     scratch_data.fe_values.get_quadrature(),
 *                     scratch_data.fe_values.get_update_flags())
 *         , fe_interface_values(scratch_data.fe_values.get_mapping(),
 *                               scratch_data.fe_values.get_fe(),
 *                               scratch_data.fe_interface_values.get_quadrature(),
 *                               scratch_data.fe_interface_values.get_update_flags())
 *       {}
 *   
 *       FEValues<dim>          fe_values;
 *       FEInterfaceValues<dim> fe_interface_values;
 *     };
 *   
 *   
 *   
 *     struct CopyData
 *     {
 *       CopyData(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)
 *       {}
 *   
 *   
 *       CopyData(const CopyData &) = default;
 *   
 *   
 *       CopyData(CopyData &&) = default;
 *   
 *   
 *       ~CopyData() = default;
 *   
 *   
 *       CopyData &operator=(const CopyData &) = default;
 *   
 *   
 *       CopyData &operator=(CopyData &&) = default;
 *   
 *   
 *       struct FaceData
 *       {
 *         FullMatrix<double>                   cell_matrix;
 *         std::vector<types::global_dof_index> joint_dof_indices;
 *       };
 *   
 *       FullMatrix<double>                   cell_matrix;
 *       Vector<double>                       cell_rhs;
 *       std::vector<types::global_dof_index> local_dof_indices;
 *       std::vector<FaceData>                face_data;
 *     };
 *   
 *   
 *   
 * @endcode
 * 
 * The more interesting part is where we actually assemble the linear system.
 * Fundamentally, this function has five parts:
 * - The definition of the `cell_worker` lambda function, a small
 * function that is defined within the `assemble_system()`
 * function and that will be responsible for computing the local
 * integrals on an individual cell. It will work on a copy of the
 * `ScratchData` class and put its results into the corresponding
 * `CopyData` object.
 * - The definition of the `face_worker` lambda function that does
 * the integration of all terms that live on the interfaces between
 * cells.
 * - The definition of the `boundary_worker` function that does the
 * same but for cell faces located on the boundary of the domain.
 * - The definition of the `copier` function that is responsible
 * for copying all of the data the previous three functions have
 * put into copy objects for a single cell, into the global matrix
 * and right hand side.
 *   
 
 * 
 * The fifth part is the one where we bring all of this together.
 *   
 
 * 
 * Let us go through each of these pieces necessary for the assembly
 * in turns.
 * 
 * @code
 *     template <int dim>
 *     void BiharmonicProblem<dim>::assemble_system()
 *     {
 *       using Iterator = typename DoFHandler<dim>::active_cell_iterator;
 *   
 * @endcode
 * 
 * The first piece is the `cell_worker` that does the assembly
 * on the cell interiors. It is a (lambda) function that takes
 * a cell (input), a scratch object, and a copy object (output)
 * as arguments. It looks like the assembly functions of many
 * other of the tutorial programs, or at least the body of the
 * loop over all cells.
 *     
 
 * 
 * The terms we integrate here are the cell contribution
 * @f{align*}{
 * A^K_{ij} = \int_K \nabla^2\varphi_i(x) : \nabla^2\varphi_j(x) dx
 * @f}
 * to the global matrix, and
 * @f{align*}{
 * f^K_i = \int_K \varphi_i(x) f(x) dx
 * @f}
 * to the right hand side vector.
 *     
 
 * 
 * We use the same technique as used in the assembly of @ref step_22 "step-22"
 * to accelerate the function: Instead of calling
 * `fe_values.shape_hessian(i, qpoint)` in the innermost loop,
 * we create a variable `hessian_i` that evaluates this
 * value once in the loop over `i` and re-use the so-evaluated
 * value in the loop over `j`. For symmetry, we do the same with a
 * variable `hessian_j`, although it is indeed only used once and
 * we could have left the call to `fe_values.shape_hessian(j,qpoint)`
 * in the instruction that computes the scalar product between
 * the two terms.
 * 
 * @code
 *       auto cell_worker = [&](const Iterator   &cell,
 *                              ScratchData<dim> &scratch_data,
 *                              CopyData         &copy_data) {
 *         copy_data.cell_matrix = 0;
 *         copy_data.cell_rhs    = 0;
 *   
 *         FEValues<dim> &fe_values = scratch_data.fe_values;
 *         fe_values.reinit(cell);
 *   
 *         cell->get_dof_indices(copy_data.local_dof_indices);
 *   
 *         ExactSolution::RightHandSide<dim> right_hand_side;
 *   
 *         const unsigned int dofs_per_cell =
 *           scratch_data.fe_values.get_fe().n_dofs_per_cell();
 *   
 *         for (unsigned int qpoint = 0; qpoint < fe_values.n_quadrature_points;
 *              ++qpoint)
 *           {
 *             for (unsigned int i = 0; i < dofs_per_cell; ++i)
 *               {
 *                 const Tensor<2, dim> &hessian_i =
 *                   fe_values.shape_hessian(i, qpoint);
 *   
 *                 for (unsigned int j = 0; j < dofs_per_cell; ++j)
 *                   {
 *                     const Tensor<2, dim> &hessian_j =
 *                       fe_values.shape_hessian(j, qpoint);
 *   
 *                     copy_data.cell_matrix(i, j) +=
 *                       scalar_product(hessian_i,   // nabla^2 phi_i(x)
 *                                      hessian_j) * // nabla^2 phi_j(x)
 *                       fe_values.JxW(qpoint);      // dx
 *                   }
 *   
 *                 copy_data.cell_rhs(i) +=
 *                   fe_values.shape_value(i, qpoint) * // phi_i(x)
 *                   right_hand_side.value(
 *                     fe_values.quadrature_point(qpoint)) * // f(x)
 *                   fe_values.JxW(qpoint);                  // dx
 *               }
 *           }
 *       };
 *   
 *   
 * @endcode
 * 
 * The next building block is the one that assembles penalty terms on each
 * of the interior faces of the mesh. As described in the documentation of
 * MeshWorker::mesh_loop(), this function receives arguments that denote
 * a cell and its neighboring cell, as well as (for each of the two
 * cells) the face (and potentially sub-face) we have to integrate
 * over. Again, we also get a scratch object, and a copy object
 * for putting the results in.
 *     
 
 * 
 * The function has three parts itself. At the top, we initialize
 * the FEInterfaceValues object and create a new `CopyData::FaceData`
 * object to store our input in. This gets pushed to the end of the
 * `copy_data.face_data` variable. We need to do this because
 * the number of faces (or subfaces) over which we integrate for a
 * given cell differs from cell to cell, and the sizes of these
 * matrices also differ, depending on what degrees of freedom
 * are adjacent to the face or subface. As discussed in the documentation
 * of MeshWorker::mesh_loop(), the copy object is reset every time a new
 * cell is visited, so that what we push to the end of
 * `copy_data.face_data()` is really all that the later `copier` function
 * gets to see when it copies the contributions of each cell to the global
 * matrix and right hand side objects.
 * 
 * @code
 *       auto face_worker = [&](const Iterator     &cell,
 *                              const unsigned int &f,
 *                              const unsigned int &sf,
 *                              const Iterator     &ncell,
 *                              const unsigned int &nf,
 *                              const unsigned int &nsf,
 *                              ScratchData<dim>   &scratch_data,
 *                              CopyData           &copy_data) {
 *         FEInterfaceValues<dim> &fe_interface_values =
 *           scratch_data.fe_interface_values;
 *         fe_interface_values.reinit(cell, f, sf, ncell, nf, nsf);
 *   
 *         copy_data.face_data.emplace_back();
 *         CopyData::FaceData &copy_data_face = copy_data.face_data.back();
 *   
 *         copy_data_face.joint_dof_indices =
 *           fe_interface_values.get_interface_dof_indices();
 *   
 *         const unsigned int n_interface_dofs =
 *           fe_interface_values.n_current_interface_dofs();
 *         copy_data_face.cell_matrix.reinit(n_interface_dofs, n_interface_dofs);
 *   
 * @endcode
 * 
 * The second part deals with determining what the penalty
 * parameter should be. By looking at the units of the various
 * terms in the bilinear form, it is clear that the penalty has
 * to have the form @f$\frac{\gamma}{h_K}@f$ (i.e., one over length
 * scale), but it is not a priori obvious how one should choose
 * the dimension-less number @f$\gamma@f$. From the discontinuous
 * Galerkin theory for the Laplace equation, one might
 * conjecture that the right choice is @f$\gamma=p(p+1)@f$ is the
 * right choice, where @f$p@f$ is the polynomial degree of the
 * finite element used. We will discuss this choice in a bit
 * more detail in the results section of this program.
 *       
 
 * 
 * In the formula above, @f$h_K@f$ is the size of cell @f$K@f$. But this
 * is not quite so straightforward either: If one uses highly
 * stretched cells, then a more involved theory says that @f$h@f$
 * should be replaced by the diameter of cell @f$K@f$ normal to the
 * direction of the edge in question.  It turns out that there
 * is a function in deal.II for that. Secondly, @f$h_K@f$ may be
 * different when viewed from the two different sides of a face.
 *       
 
 * 
 * To stay on the safe side, we take the maximum of the two values.
 * We will note that it is possible that this computation has to be
 * further adjusted if one were to use hanging nodes resulting from
 * adaptive mesh refinement.
 * 
 * @code
 *         const unsigned int p = fe.degree;
 *         const double       gamma_over_h =
 *           std::max((1.0 * p * (p + 1) /
 *                     cell->extent_in_direction(
 *                       GeometryInfo<dim>::unit_normal_direction[f])),
 *                    (1.0 * p * (p + 1) /
 *                     ncell->extent_in_direction(
 *                       GeometryInfo<dim>::unit_normal_direction[nf])));
 *   
 * @endcode
 * 
 * Finally, and as usual, we loop over the quadrature points and
 * indices `i` and `j` to add up the contributions of this face
 * or sub-face. These are then stored in the
 * `copy_data.face_data` object created above. As for the cell
 * worker, we pull the evaluation of averages and jumps out of
 * the loops if possible, introducing local variables that store
 * these results. The assembly then only needs to use these
 * local variables in the innermost loop. Regarding the concrete
 * formula this code implements, recall that the interface terms
 * of the bilinear form were as follows:
 * @f{align*}{
 * -\sum_{e \in \mathbb{F}} \int_{e}
 * \jump{ \frac{\partial v_h}{\partial \mathbf n}}
 * \average{\frac{\partial^2 u_h}{\partial \mathbf n^2}} \ ds
 * -\sum_{e \in \mathbb{F}} \int_{e}
 * \average{\frac{\partial^2 v_h}{\partial \mathbf n^2}}
 * \jump{\frac{\partial u_h}{\partial \mathbf n}} \ ds
 * + \sum_{e \in \mathbb{F}}
 * \frac{\gamma}{h_e}
 * \int_e
 * \jump{\frac{\partial v_h}{\partial \mathbf n}}
 * \jump{\frac{\partial u_h}{\partial \mathbf n}} \ ds.
 * @f}
 * 
 * @code
 *         for (unsigned int qpoint = 0;
 *              qpoint < fe_interface_values.n_quadrature_points;
 *              ++qpoint)
 *           {
 *             const auto &n = fe_interface_values.normal(qpoint);
 *   
 *             for (unsigned int i = 0; i < n_interface_dofs; ++i)
 *               {
 *                 const double av_hessian_i_dot_n_dot_n =
 *                   (fe_interface_values.average_of_shape_hessians(i, qpoint) * n *
 *                    n);
 *                 const double jump_grad_i_dot_n =
 *                   (fe_interface_values.jump_in_shape_gradients(i, qpoint) * n);
 *   
 *                 for (unsigned int j = 0; j < n_interface_dofs; ++j)
 *                   {
 *                     const double av_hessian_j_dot_n_dot_n =
 *                       (fe_interface_values.average_of_shape_hessians(j, qpoint) *
 *                        n * n);
 *                     const double jump_grad_j_dot_n =
 *                       (fe_interface_values.jump_in_shape_gradients(j, qpoint) *
 *                        n);
 *   
 *                     copy_data_face.cell_matrix(i, j) +=
 *                       (-av_hessian_i_dot_n_dot_n       // - {grad^2 v n n }
 *                          * jump_grad_j_dot_n           // [grad u n]
 *                        - av_hessian_j_dot_n_dot_n      // - {grad^2 u n n }
 *                            * jump_grad_i_dot_n         // [grad v n]
 *                        +                               // +
 *                        gamma_over_h *                  // gamma/h
 *                          jump_grad_i_dot_n *           // [grad v n]
 *                          jump_grad_j_dot_n) *          // [grad u n]
 *                       fe_interface_values.JxW(qpoint); // dx
 *                   }
 *               }
 *           }
 *       };
 *   
 *   
 * @endcode
 * 
 * The third piece is to do the same kind of assembly for faces that
 * are at the boundary. The idea is the same as above, of course,
 * with only the difference that there are now penalty terms that
 * also go into the right hand side.
 *     
 
 * 
 * As before, the first part of the function simply sets up some
 * helper objects:
 * 
 * @code
 *       auto boundary_worker = [&](const Iterator     &cell,
 *                                  const unsigned int &face_no,
 *                                  ScratchData<dim>   &scratch_data,
 *                                  CopyData           &copy_data) {
 *         FEInterfaceValues<dim> &fe_interface_values =
 *           scratch_data.fe_interface_values;
 *         fe_interface_values.reinit(cell, face_no);
 *         const auto &q_points = fe_interface_values.get_quadrature_points();
 *   
 *         copy_data.face_data.emplace_back();
 *         CopyData::FaceData &copy_data_face = copy_data.face_data.back();
 *   
 *         const unsigned int n_dofs =
 *           fe_interface_values.n_current_interface_dofs();
 *         copy_data_face.joint_dof_indices =
 *           fe_interface_values.get_interface_dof_indices();
 *   
 *         copy_data_face.cell_matrix.reinit(n_dofs, n_dofs);
 *   
 *         const std::vector<double> &JxW = fe_interface_values.get_JxW_values();
 *         const std::vector<Tensor<1, dim>> &normals =
 *           fe_interface_values.get_normal_vectors();
 *   
 *   
 *         ExactSolution::Solution<dim> exact_solution;
 *         std::vector<Tensor<1, dim>>  exact_gradients(q_points.size());
 *         exact_solution.gradient_list(q_points, exact_gradients);
 *   
 *   
 * @endcode
 * 
 * Positively, because we now only deal with one cell adjacent to the
 * face (as we are on the boundary), the computation of the penalty
 * factor @f$\gamma@f$ is substantially simpler:
 * 
 * @code
 *         const unsigned int p = fe.degree;
 *         const double       gamma_over_h =
 *           (1.0 * p * (p + 1) /
 *            cell->extent_in_direction(
 *              GeometryInfo<dim>::unit_normal_direction[face_no]));
 *   
 * @endcode
 * 
 * The third piece is the assembly of terms. This is now
 * slightly more involved since these contains both terms for
 * the matrix and for the right hand side. The former is exactly
 * the same as for the interior faces stated above if one just
 * defines the jump and average appropriately (which is what the
 * FEInterfaceValues class does). The latter requires us to
 * evaluate the boundary conditions @f$j(\mathbf x)@f$, which in the
 * current case (where we know the exact solution) we compute
 * from @f$j(\mathbf x) = \frac{\partial u(\mathbf x)}{\partial
 * {\mathbf n}}@f$. The term to be added to the right hand side
 * vector is then
 * @f$\frac{\gamma}{h_e}\int_e
 * \jump{\frac{\partial v_h}{\partial \mathbf n}} j \ ds@f$.
 * 
 * @code
 *         for (unsigned int qpoint = 0; qpoint < q_points.size(); ++qpoint)
 *           {
 *             const auto &n = normals[qpoint];
 *   
 *             for (unsigned int i = 0; i < n_dofs; ++i)
 *               {
 *                 const double av_hessian_i_dot_n_dot_n =
 *                   (fe_interface_values.average_of_shape_hessians(i, qpoint) * n *
 *                    n);
 *                 const double jump_grad_i_dot_n =
 *                   (fe_interface_values.jump_in_shape_gradients(i, qpoint) * n);
 *   
 *                 for (unsigned int j = 0; j < n_dofs; ++j)
 *                   {
 *                     const double av_hessian_j_dot_n_dot_n =
 *                       (fe_interface_values.average_of_shape_hessians(j, qpoint) *
 *                        n * n);
 *                     const double jump_grad_j_dot_n =
 *                       (fe_interface_values.jump_in_shape_gradients(j, qpoint) *
 *                        n);
 *   
 *                     copy_data_face.cell_matrix(i, j) +=
 *                       (-av_hessian_i_dot_n_dot_n  // - {grad^2 v n n}
 *                          * jump_grad_j_dot_n      //   [grad u n]
 *                                                   
 *                        - av_hessian_j_dot_n_dot_n // - {grad^2 u n n}
 *                            * jump_grad_i_dot_n    //   [grad v n]
 *                                                   
 *                        + gamma_over_h             //  gamma/h
 *                            * jump_grad_i_dot_n    // [grad v n]
 *                            * jump_grad_j_dot_n    // [grad u n]
 *                        ) *
 *                       JxW[qpoint]; // dx
 *                   }
 *   
 *                 copy_data.cell_rhs(i) +=
 *                   (-av_hessian_i_dot_n_dot_n *       // - {grad^2 v n n }
 *                      (exact_gradients[qpoint] * n)   //   (grad u_exact . n)
 *                    +                                 // +
 *                    gamma_over_h                      //  gamma/h
 *                      * jump_grad_i_dot_n             // [grad v n]
 *                      * (exact_gradients[qpoint] * n) // (grad u_exact . n)
 *                    ) *
 *                   JxW[qpoint]; // dx
 *               }
 *           }
 *       };
 *   
 * @endcode
 * 
 * Part 4 is a small function that copies the data produced by the
 * cell, interior, and boundary face assemblers above into the
 * global matrix and right hand side vector. There really is not
 * very much to do here: We distribute the cell matrix and right
 * hand side contributions as we have done in almost all of the
 * other tutorial programs using the constraints objects. We then
 * also have to do the same for the face matrix contributions
 * that have gained content for the faces (interior and boundary)
 * and that the `face_worker` and `boundary_worker` have added
 * to the `copy_data.face_data` array.
 * 
 * @code
 *       auto copier = [&](const CopyData &copy_data) {
 *         constraints.distribute_local_to_global(copy_data.cell_matrix,
 *                                                copy_data.cell_rhs,
 *                                                copy_data.local_dof_indices,
 *                                                system_matrix,
 *                                                system_rhs);
 *   
 *         for (const auto &cdf : copy_data.face_data)
 *           {
 *             constraints.distribute_local_to_global(cdf.cell_matrix,
 *                                                    cdf.joint_dof_indices,
 *                                                    system_matrix);
 *           }
 *       };
 *   
 *   
 * @endcode
 * 
 * Having set all of this up, what remains is to just create a scratch
 * and copy data object and call the MeshWorker::mesh_loop() function
 * that then goes over all cells and faces, calls the respective workers
 * on them, and then the copier function that puts things into the
 * global matrix and right hand side. As an additional benefit,
 * MeshWorker::mesh_loop() does all of this in parallel, using
 * as many processor cores as your machine happens to have.
 * 
 * @code
 *       const unsigned int n_gauss_points = dof_handler.get_fe().degree + 1;
 *       ScratchData<dim>   scratch_data(mapping,
 *                                     fe,
 *                                     n_gauss_points,
 *                                     update_values | update_gradients |
 *                                       update_hessians | update_quadrature_points |
 *                                       update_JxW_values,
 *                                     update_values | update_gradients |
 *                                       update_hessians | update_quadrature_points |
 *                                       update_JxW_values | update_normal_vectors);
 *       CopyData           copy_data(dof_handler.get_fe().n_dofs_per_cell());
 *       MeshWorker::mesh_loop(dof_handler.begin_active(),
 *                             dof_handler.end(),
 *                             cell_worker,
 *                             copier,
 *                             scratch_data,
 *                             copy_data,
 *                             MeshWorker::assemble_own_cells |
 *                               MeshWorker::assemble_boundary_faces |
 *                               MeshWorker::assemble_own_interior_faces_once,
 *                             boundary_worker,
 *                             face_worker);
 *     }
 *   
 *   
 *   
 * @endcode
 * 
 * 
 * <a name="step_47-Solvingthelinearsystemandpostprocessing"></a> 
 * <h4>Solving the linear system and postprocessing</h4>
 *   
 
 * 
 * The show is essentially over at this point: The remaining functions are
 * not overly interesting or novel. The first one simply uses a direct
 * solver to solve the linear system (see also @ref step_29 "step-29"):
 * 
 * @code
 *     template <int dim>
 *     void BiharmonicProblem<dim>::solve()
 *     {
 *       std::cout << "   Solving system..." << std::endl;
 *   
 *       SparseDirectUMFPACK A_direct;
 *       A_direct.initialize(system_matrix);
 *       A_direct.vmult(solution, system_rhs);
 *   
 *       constraints.distribute(solution);
 *     }
 *   
 *   
 *   
 * @endcode
 * 
 * The next function evaluates the error between the computed solution
 * and the exact solution (which is known here because we have chosen
 * the right hand side and boundary values in a way so that we know
 * the corresponding solution). In the first two code blocks below,
 * we compute the error in the @f$L_2@f$ norm and the @f$H^1@f$ semi-norm.
 * 
 * @code
 *     template <int dim>
 *     void BiharmonicProblem<dim>::compute_errors()
 *     {
 *       {
 *         Vector<float> norm_per_cell(triangulation.n_active_cells());
 *         VectorTools::integrate_difference(mapping,
 *                                           dof_handler,
 *                                           solution,
 *                                           ExactSolution::Solution<dim>(),
 *                                           norm_per_cell,
 *                                           QGauss<dim>(fe.degree + 2),
 *                                           VectorTools::L2_norm);
 *         const double error_norm =
 *           VectorTools::compute_global_error(triangulation,
 *                                             norm_per_cell,
 *                                             VectorTools::L2_norm);
 *         std::cout << "   Error in the L2 norm           :     " << error_norm
 *                   << std::endl;
 *       }
 *   
 *       {
 *         Vector<float> norm_per_cell(triangulation.n_active_cells());
 *         VectorTools::integrate_difference(mapping,
 *                                           dof_handler,
 *                                           solution,
 *                                           ExactSolution::Solution<dim>(),
 *                                           norm_per_cell,
 *                                           QGauss<dim>(fe.degree + 2),
 *                                           VectorTools::H1_seminorm);
 *         const double error_norm =
 *           VectorTools::compute_global_error(triangulation,
 *                                             norm_per_cell,
 *                                             VectorTools::H1_seminorm);
 *         std::cout << "   Error in the H1 seminorm       : " << error_norm
 *                   << std::endl;
 *       }
 *   
 * @endcode
 * 
 * Now also compute an approximation to the @f$H^2@f$ seminorm error. The actual
 * @f$H^2@f$ seminorm would require us to integrate second derivatives of the
 * solution @f$u_h@f$, but given the Lagrange shape functions we use, @f$u_h@f$ of
 * course has kinks at the interfaces between cells, and consequently second
 * derivatives are singular at interfaces. As a consequence, we really only
 * integrate over the interior of cells and ignore the interface
 * contributions. This is *not* an equivalent norm to the energy norm for
 * the problem, but still gives us an idea of how fast the error converges.
 *     
 
 * 
 * We note that one could address this issue by defining a norm that
 * is equivalent to the energy norm. This would involve adding up not
 * only the integrals over cell interiors as we do below, but also adding
 * penalty terms for the jump of the derivative of @f$u_h@f$ across interfaces,
 * with an appropriate scaling of the two kinds of terms. We will leave
 * this for later work.
 * 
 * @code
 *       {
 *         const QGauss<dim>            quadrature_formula(fe.degree + 2);
 *         ExactSolution::Solution<dim> exact_solution;
 *         Vector<double> error_per_cell(triangulation.n_active_cells());
 *   
 *         FEValues<dim> fe_values(mapping,
 *                                 fe,
 *                                 quadrature_formula,
 *                                 update_values | update_hessians |
 *                                   update_quadrature_points | update_JxW_values);
 *   
 *         const FEValuesExtractors::Scalar scalar(0);
 *         const unsigned int               n_q_points = quadrature_formula.size();
 *   
 *         std::vector<SymmetricTensor<2, dim>> exact_hessians(n_q_points);
 *         std::vector<Tensor<2, dim>>          hessians(n_q_points);
 *         for (auto &cell : dof_handler.active_cell_iterators())
 *           {
 *             fe_values.reinit(cell);
 *             fe_values[scalar].get_function_hessians(solution, hessians);
 *             exact_solution.hessian_list(fe_values.get_quadrature_points(),
 *                                         exact_hessians);
 *   
 *             double local_error = 0;
 *             for (unsigned int q_point = 0; q_point < n_q_points; ++q_point)
 *               {
 *                 local_error +=
 *                   ((exact_hessians[q_point] - hessians[q_point]).norm_square() *
 *                    fe_values.JxW(q_point));
 *               }
 *             error_per_cell[cell->active_cell_index()] = std::sqrt(local_error);
 *           }
 *   
 *         const double error_norm =
 *           VectorTools::compute_global_error(triangulation,
 *                                             error_per_cell,
 *                                             VectorTools::L2_norm);
 *         std::cout << "   Error in the broken H2 seminorm: " << error_norm
 *                   << std::endl;
 *       }
 *     }
 *   
 *   
 *   
 * @endcode
 * 
 * Equally uninteresting is the function that generates graphical output.
 * It looks exactly like the one in @ref step_6 "step-6", for example.
 * 
 * @code
 *     template <int dim>
 *     void
 *     BiharmonicProblem<dim>::output_results(const unsigned int iteration) const
 *     {
 *       std::cout << "   Writing graphical output..." << std::endl;
 *   
 *       DataOut<dim> data_out;
 *   
 *       data_out.attach_dof_handler(dof_handler);
 *       data_out.add_data_vector(solution, "solution");
 *       data_out.build_patches();
 *   
 *       const std::string filename =
 *         ("output_" + Utilities::int_to_string(iteration, 6) + ".vtu");
 *       std::ofstream output_vtu(filename);
 *       data_out.write_vtu(output_vtu);
 *     }
 *   
 *   
 *   
 * @endcode
 * 
 * The same is true for the `run()` function: Just like in previous
 * programs.
 * 
 * @code
 *     template <int dim>
 *     void BiharmonicProblem<dim>::run()
 *     {
 *       make_grid();
 *   
 *       const unsigned int n_cycles = 4;
 *       for (unsigned int cycle = 0; cycle < n_cycles; ++cycle)
 *         {
 *           std::cout << "Cycle " << cycle << " of " << n_cycles << std::endl;
 *   
 *           triangulation.refine_global(1);
 *           setup_system();
 *   
 *           assemble_system();
 *           solve();
 *   
 *           output_results(cycle);
 *   
 *           compute_errors();
 *           std::cout << std::endl;
 *         }
 *     }
 *   } // namespace Step47
 *   
 *   
 *   
 * @endcode
 * 
 * 
 * <a name="step_47-Themainfunction"></a> 
 * <h3>The main() function</h3>
 * 
 
 * 
 * Finally for the `main()` function. There is, again, not very much to see
 * here: It looks like the ones in previous tutorial programs. There
 * is a variable that allows selecting the polynomial degree of the element
 * we want to use for solving the equation. Because the C0IP formulation
 * we use requires the element degree to be at least two, we check with
 * an assertion that whatever one sets for the polynomial degree actually
 * makes sense.
 * 
 * @code
 *   int main()
 *   {
 *     try
 *       {
 *         using namespace dealii;
 *         using namespace Step47;
 *   
 *         const unsigned int fe_degree = 2;
 *         Assert(fe_degree >= 2,
 *                ExcMessage("The C0IP formulation for the biharmonic problem "
 *                           "only works if one uses elements of polynomial "
 *                           "degree at least 2."));
 *   
 *         BiharmonicProblem<2> biharmonic_problem(fe_degree);
 *         biharmonic_problem.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;
 *   }
 * @endcode
<a name="step_47-Results"></a><h1>Results</h1>
 
 
We run the program with right hand side and boundary values as
discussed in the introduction. These will produce the
solution @f$u = \sin(\pi x) \sin(\pi y)@f$ on the domain @f$\Omega = (0,1)^2@f$.
We test this setup using @f$Q_2@f$, @f$Q_3@f$, and @f$Q_4@f$ elements, which one can
change via the `fe_degree` variable in the `main()` function. With mesh
refinement, the @f$L_2@f$ convergence rates, @f$H^1@f$-seminorm rate,
and @f$H^2@f$-seminorm convergence of @f$u@f$
should then be around 2, 2, 1 for @f$Q_2@f$ (with the @f$L_2@f$ norm
sub-optimal as discussed in the introduction); 4, 3, 2 for
@f$Q_3@f$; and 5, 4, 3 for @f$Q_4@f$, respectively.
 
From the literature, it is not immediately clear what
the penalty parameter @f$\gamma@f$ should be. For example,
@cite Brenner2010 state that it needs to be larger than one, and
choose @f$\gamma=5@f$. The FEniCS/Dolphin tutorial chooses it as
@f$\gamma=8@f$, see
https://fenicsproject.org/docs/dolfin/1.6.0/python/demo/documented/biharmonic/python/documentation.html
. @cite Wells2007 uses a value for @f$\gamma@f$ larger than the
number of edges belonging to an element for Kirchhoff plates (see
their Section 4.2). This suggests that maybe
@f$\gamma = 1@f$, @f$2@f$, are too small; on the other hand, a value
@f$p(p+1)@f$ would be reasonable,
where @f$p@f$ is the degree of polynomials. The last of these choices is
the one one would expect to work by comparing
to the discontinuous Galerkin formulations for the Laplace equation
(see, for example, the discussions in @ref step_39 "step-39" and @ref step_74 "step-74"),
and it will turn out to also work here.
But we should check what value of @f$\gamma@f$ is right, and we will do so
below; changing @f$\gamma@f$ is easy in the two `face_worker` and
`boundary_worker` functions defined in `assemble_system()`.
 
 
<a name="step_47-TestresultsoniQsub2subiwithigammapp1i"></a><h3>Test results on <i>Q<sub>2</sub></i> with <i>&gamma; = p(p+1)</i> </h3>
 
 
We run the code with differently refined meshes
and get the following convergence rates.
 
<table align="center" class="doxtable">
  <tr>
   <th>Number of refinements </th><th>  @f$\|u-u_h\|_{L_2}@f$ </th><th>  Conv. rates  </th><th>  @f$|u-u_h|_{H^1}@f$ </th><th> Conv. rates </th><th> @f$|u-u_h|^\circ_{H^2}@f$ </th><th> Conv. rates </th>
  </tr>
  <tr>
   <td>   2                  </td><td>   8.780e-03 </td><td>       </td><td>  7.095e-02   </td><td>           </td><td>  1.645 </td><td>   </td>
  </tr>
  <tr>
   <td>   3                  </td><td>   3.515e-03   </td><td>  1.32 </td><td> 2.174e-02  </td><td>     1.70     </td><td> 8.121e-01  </td><td>  1.018  </td>
  </tr>
  <tr>
   <td>   4                  </td><td>   1.103e-03   </td><td>  1.67   </td><td> 6.106e-03    </td><td>  1.83        </td><td>   4.015e-01 </td><td> 1.016  </td>
  </tr>
  <tr>
   <td>   5                  </td><td>  3.084e-04  </td><td>  1.83   </td><td>  1.622e-03   </td><td>    1.91        </td><td> 1.993e-01 </td><td>  1.010   </td>
  </tr>
</table>
We can see that the @f$L_2@f$ convergence rates are around 2,
@f$H^1@f$-seminorm convergence rates are around 2,
and @f$H^2@f$-seminorm convergence rates are around 1. The latter two
match the theoretically expected rates; for the former, we have no
theorem but are not surprised that it is sub-optimal given the remark
in the introduction.
 
 
<a name="step_47-TestresultsoniQsub3subiwithigammapp1i"></a><h3>Test results on <i>Q<sub>3</sub></i> with <i>&gamma; = p(p+1)</i> </h3>
 
 
 
<table align="center" class="doxtable">
  <tr>
   <th>Number of refinements </th><th>  @f$\|u-u_h\|_{L_2}@f$ </th><th>  Conv. rates  </th><th>  @f$|u-u_h|_{H^1}@f$ </th><th> Conv. rates </th><th> @f$|u-u_h|^\circ_{H^2}@f$ </th><th> Conv. rates </th>
  </tr>
  <tr>
   <td>   2                  </td><td>    2.045e-04 </td><td>       </td><td>   4.402e-03   </td><td>           </td><td> 1.641e-01 </td><td>   </td>
  </tr>
  <tr>
   <td>   3                  </td><td>   1.312e-05   </td><td> 3.96  </td><td>  5.537e-04  </td><td>   2.99     </td><td> 4.096e-02 </td><td>  2.00  </td>
  </tr>
  <tr>
   <td>   4                  </td><td>   8.239e-07 </td><td>  3.99  </td><td> 6.904e-05   </td><td> 3.00     </td><td> 1.023e-02 </td><td> 2.00 </td>
  </tr>
  <tr>
   <td>   5                  </td><td>   5.158e-08  </td><td>  3.99 </td><td> 8.621e-06 </td><td>  3.00      </td><td> 2.558e-03  </td><td>  2.00  </td>
  </tr>
</table>
We can see that the @f$L_2@f$ convergence rates are around 4,
@f$H^1@f$-seminorm convergence rates are around 3,
and @f$H^2@f$-seminorm convergence rates are around 2.
This, of course, matches our theoretical expectations.
 
 
<a name="step_47-TestresultsoniQsub4subiwithigammapp1i"></a><h3>Test results on <i>Q<sub>4</sub></i> with <i>&gamma; = p(p+1)</i> </h3>
 
 
<table align="center" class="doxtable">
  <tr>
   <th>Number of refinements </th><th>  @f$\|u-u_h\|_{L_2}@f$ </th><th>  Conv. rates  </th><th>  @f$|u-u_h|_{H^1}@f$ </th><th> Conv. rates </th><th> @f$|u-u_h|^\circ_{H^2}@f$ </th><th> Conv. rates </th>
  </tr>
  <tr>
   <td>   2                  </td><td>    6.510e-06 </td><td>       </td><td> 2.215e-04   </td><td>           </td><td>  1.275e-02 </td><td>   </td>
  </tr>
  <tr>
   <td>   3                  </td><td>   2.679e-07  </td><td>  4.60  </td><td> 1.569e-05  </td><td>   3.81    </td><td> 1.496e-03 </td><td>  3.09  </td>
  </tr>
  <tr>
   <td>   4                  </td><td>   9.404e-09  </td><td> 4.83   </td><td> 1.040e-06    </td><td> 3.91       </td><td> 1.774e-04 </td><td> 3.07 </td>
  </tr>
  <tr>
   <td>   5                  </td><td>   7.943e-10 </td><td>  3.56  </td><td>   6.693e-08 </td><td> 3.95     </td><td> 2.150e-05  </td><td> 3.04    </td>
  </tr>
</table>
We can see that the @f$L_2@f$ norm convergence rates are around 5,
@f$H^1@f$-seminorm convergence rates are around 4,
and @f$H^2@f$-seminorm convergence rates are around 3.
On the finest mesh, the @f$L_2@f$ norm convergence rate
is much smaller than our theoretical expectations
because the linear solver becomes the limiting factor due
to round-off. Of course the @f$L_2@f$ error is also very small already in
that case.
 
 
<a name="step_47-TestresultsoniQsub2subiwithigamma1i"></a><h3>Test results on <i>Q<sub>2</sub></i> with <i>&gamma; = 1</i> </h3>
 
 
For comparison with the results above, let us now also consider the
case where we simply choose @f$\gamma=1@f$:
 
<table align="center" class="doxtable">
  <tr>
   <th>Number of refinements </th><th>  @f$\|u-u_h\|_{L_2}@f$ </th><th>  Conv. rates  </th><th>  @f$|u-u_h|_{H^1}@f$ </th><th> Conv. rates </th><th> @f$|u-u_h|^\circ_{H^2}@f$ </th><th> Conv. rates </th>
  </tr>
  <tr>
   <td>   2                  </td><td>   7.350e-02 </td><td>       </td><td>   7.323e-01   </td><td>           </td><td> 10.343 </td><td>   </td>
  </tr>
  <tr>
   <td>   3                  </td><td>   6.798e-03   </td><td> 3.43  </td><td> 1.716e-01   </td><td>   2.09    </td><td>4.836 </td><td>  1.09 </td>
  </tr>
  <tr>
   <td>   4                  </td><td>  9.669e-04   </td><td> 2.81   </td><td> 6.436e-02    </td><td> 1.41      </td><td>  3.590 </td><td> 0.430 </td>
  </tr>
  <tr>
   <td>   5                  </td><td>   1.755e-04 </td><td> 2.46 </td><td>  2.831e-02  </td><td>    1.18      </td><td>3.144  </td><td>  0.19  </td>
  </tr>
</table>
Although @f$L_2@f$ norm convergence rates of @f$u@f$ more or less
follows the theoretical expectations,
the @f$H^1@f$-seminorm and @f$H^2@f$-seminorm do not seem to converge as expected.
Comparing results from @f$\gamma = 1@f$ and @f$\gamma = p(p+1)@f$, it is clear that
@f$\gamma = p(p+1)@f$ is a better penalty.
Given that @f$\gamma=1@f$ is already too small for @f$Q_2@f$ elements, it may not be surprising that if one repeated the
experiment with a @f$Q_3@f$ element, the results are even more disappointing: One again only obtains convergence
rates of 2, 1, zero -- i.e., no better than for the @f$Q_2@f$ element (although the errors are smaller in magnitude).
Maybe surprisingly, however, one obtains more or less the expected convergence orders when using @f$Q_4@f$
elements. Regardless, this uncertainty suggests that @f$\gamma=1@f$ is at best a risky choice, and at worst an
unreliable one and that we should choose @f$\gamma@f$ larger.
 
 
<a name="step_47-TestresultsoniQsub2subiwithigamma2i"></a><h3>Test results on <i>Q<sub>2</sub></i> with <i>&gamma; = 2</i> </h3>
 
 
Since @f$\gamma=1@f$ is clearly too small, one might conjecture that
@f$\gamma=2@f$ might actually work better. Here is what one obtains in
that case:
 
<table align="center" class="doxtable">
  <tr>
   <th>Number of refinements </th><th>  @f$\|u-u_h\|_{L_2}@f$ </th><th>  Conv. rates  </th><th>  @f$|u-u_h|_{H^1}@f$ </th><th> Conv. rates </th><th> @f$|u-u_h|^\circ_{H^2}@f$ </th><th> Conv. rates </th>
  </tr>
  <tr>
   <td>   2                  </td><td>   4.133e-02 </td><td>       </td><td>  2.517e-01   </td><td>           </td><td> 3.056 </td><td>   </td>
  </tr>
  <tr>
   <td>   3                  </td><td>  6.500e-03   </td><td>2.66  </td><td> 5.916e-02  </td><td>  2.08    </td><td>1.444 </td><td>  1.08 </td>
  </tr>
  <tr>
   <td>   4                  </td><td> 6.780e-04   </td><td> 3.26  </td><td> 1.203e-02    </td><td> 2.296      </td><td> 6.151e-01 </td><td> 1.231 </td>
  </tr>
  <tr>
   <td>   5                  </td><td> 1.622e-04 </td><td> 2.06 </td><td>  2.448e-03  </td><td>   2.297     </td><td> 2.618e-01  </td><td> 1.232  </td>
  </tr>
</table>
In this case, the convergence rates more or less follow the
theoretical expectations, but, compared to the results from @f$\gamma =
p(p+1)@f$, are more variable.
Again, we could repeat this kind of experiment for @f$Q_3@f$ and @f$Q_4@f$ elements. In both cases, we will find that we
obtain roughly the expected convergence rates. Of more interest may then be to compare the absolute
size of the errors. While in the table above, for the @f$Q_2@f$ case, the errors on the finest grid are comparable between
the @f$\gamma=p(p+1)@f$ and @f$\gamma=2@f$ case, for @f$Q_3@f$ the errors are substantially larger for @f$\gamma=2@f$ than for
@f$\gamma=p(p+1)@f$. The same is true for the @f$Q_4@f$ case.
 
 
<a name="step_47-Conclusionsforthechoiceofthepenaltyparameter"></a><h3> Conclusions for the choice of the penalty parameter </h3>
 
 
The conclusions for which of the "reasonable" choices one should use for the penalty parameter
is that @f$\gamma=p(p+1)@f$ yields the expected results. It is, consequently, what the code
uses as currently written.
 
 
<a name="step_47-Possibilitiesforextensions"></a><h3> Possibilities for extensions </h3>
 
 
There are a number of obvious extensions to this program that would
make sense:
 
- The program uses a square domain and a uniform mesh. Real problems
  don't come this way, and one should verify convergence also on
  domains with other shapes and, in particular, curved boundaries. One
  may also be interested in resolving areas of less regularity by
  using adaptive mesh refinement.
 
- From a more theoretical perspective, the convergence results above
  only used the "broken" @f$H^2@f$ seminorm @f$|\cdot|^\circ_{H^2}@f$ instead
  of the "equivalent" norm @f$|\cdot|_h@f$. This is good enough to
  convince ourselves that the program isn't fundamentally
  broken. However, it might be interesting to measure the error in the
  actual norm for which we have theoretical results. Implementing this
  addition should not be overly difficult using, for example, the
  FEInterfaceValues class combined with MeshWorker::mesh_loop() in the
  same spirit as we used for the assembly of the linear system.
 
 
<a name="step_47-Derivationforthesimplysupportedplates"></a>  <h4> Derivation for the simply supported plates </h4>
 
 
  Similar to the "clamped" boundary condition addressed in the implementation,
  we will derive the @f$C^0@f$ IP finite element scheme for simply supported plates:
  @f{align*}{
    \Delta^2 u(\mathbf x) &= f(\mathbf x)
    \qquad \qquad &&\forall \mathbf x \in \Omega,
    u(\mathbf x) &= g(\mathbf x) \qquad \qquad
    &&\forall \mathbf x \in \partial\Omega, \\
    \Delta u(\mathbf x) &= h(\mathbf x) \qquad \qquad
    &&\forall \mathbf x \in \partial\Omega.
  @f}
  We multiply the biharmonic equation by the test function @f$v_h@f$ and integrate over @f$ K @f$ and get:
  @f{align*}{
    \int_K v_h (\Delta^2 u_h)
     &= \int_K (D^2 v_h) : (D^2 u_h)
       + \int_{\partial K} v_h \frac{\partial (\Delta u_h)}{\partial \mathbf{n}}
       -\int_{\partial K} (\nabla v_h) \cdot (\frac{\partial \nabla u_h}{\partial \mathbf{n}}).
  @f}
 
  Summing up over all cells @f$K \in  \mathbb{T}@f$,since normal directions of @f$\Delta u_h@f$ are pointing at
  opposite directions on each interior edge shared by two cells and @f$v_h = 0@f$ on @f$\partial \Omega@f$,
  @f{align*}{
  \sum_{K \in \mathbb{T}} \int_{\partial K} v_h \frac{\partial (\Delta u_h)}{\partial \mathbf{n}} = 0,
  @f}
  and by the definition of jump over cell interfaces,
  @f{align*}{
  -\sum_{K \in \mathbb{T}} \int_{\partial K} (\nabla v_h) \cdot (\frac{\partial \nabla u_h}{\partial \mathbf{n}}) = -\sum_{e \in \mathbb{F}} \int_{e} \jump{\frac{\partial v_h}{\partial \mathbf{n}}} (\frac{\partial^2 u_h}{\partial \mathbf{n^2}}).
  @f}
  We separate interior faces and boundary faces of the domain,
  @f{align*}{
  -\sum_{K \in \mathbb{T}} \int_{\partial K} (\nabla v_h) \cdot (\frac{\partial \nabla u_h}{\partial \mathbf{n}}) = -\sum_{e \in \mathbb{F}^i} \int_{e} \jump{\frac{\partial v_h}{\partial \mathbf{n}}} (\frac{\partial^2 u_h}{\partial \mathbf{n^2}})
  - \sum_{e \in \partial \Omega} \int_{e} \jump{\frac{\partial v_h}{\partial \mathbf{n}}} h,
  @f}
  where @f$\mathbb{F}^i@f$ is the set of interior faces.
  This leads us to
  @f{align*}{
  \sum_{K \in \mathbb{T}} \int_K (D^2 v_h) : (D^2 u_h) \ dx - \sum_{e \in \mathbb{F}^i} \int_{e} \jump{\frac{\partial v_h}{\partial \mathbf{n}}} (\frac{\partial^2 u_h}{\partial \mathbf{n^2}}) \ ds
  = \sum_{K \in \mathbb{T}}\int_{K} v_h f  \ dx + \sum_{e\subset\partial\Omega} \int_{e} \jump{\frac{\partial v_h}{\partial \mathbf{n}}} h \ ds.
  @f}
 
  In order to symmetrize and stabilize the discrete problem,
  we add symmetrization and stabilization term.
  We finally get the @f$C^0@f$ IP finite element scheme for the biharmonic equation:
  find @f$u_h@f$ such that @f$u_h =g@f$ on @f$\partial \Omega@f$ and
  @f{align*}{
  \mathcal{A}(v_h,u_h)&=\mathcal{F}(v_h) \quad \text{holds for all test functions } v_h,
  @f}
  where
  @f{align*}{
  \mathcal{A}(v_h,u_h):=&\sum_{K \in \mathbb{T}}\int_K D^2v_h:D^2u_h \ dx
  \\
  &
   -\sum_{e \in \mathbb{F}^i} \int_{e}
    \jump{\frac{\partial v_h}{\partial \mathbf n}}
    \average{\frac{\partial^2 u_h}{\partial \mathbf n^2}} \ ds
   -\sum_{e \in \mathbb{F}^i} \int_{e}
   \average{\frac{\partial^2 v_h}{\partial \mathbf n^2}}
   \jump{\frac{\partial u_h}{\partial \mathbf n}} \ ds
  \\
  &+ \sum_{e \in \mathbb{F}^i}
   \frac{\gamma}{h_e}
   \int_e
   \jump{\frac{\partial v_h}{\partial \mathbf n}}
   \jump{\frac{\partial u_h}{\partial \mathbf n}} \ ds,
  @f}
  and
  @f{align*}{
  \mathcal{F}(v_h)&:=\sum_{K \in \mathbb{T}}\int_{K} v_h f \ dx
  +
  \sum_{e\subset\partial\Omega}
  \int_e \jump{\frac{\partial v_h}{\partial \mathbf n}} h \ ds.
  @f}
  The implementation of this boundary case is similar to the "clamped" version
  except that `boundary_worker` is no longer needed for system assembling
  and the right hand side is changed according to the formulation.
 *
 *
<a name="step_47-PlainProg"></a>
<h1> The plain program</h1>
@include "step-47.cc"
*/