]> https://gitweb.dealii.org/ - dealii-svn.git/commitdiff
compute correct energy norm and L2 norm
authorkanschat <kanschat@0785d39b-7218-0410-832d-ea1e28bc413d>
Thu, 28 Jul 2011 16:43:36 +0000 (16:43 +0000)
committerkanschat <kanschat@0785d39b-7218-0410-832d-ea1e28bc413d>
Thu, 28 Jul 2011 16:43:36 +0000 (16:43 +0000)
git-svn-id: https://svn.dealii.org/trunk@23980 0785d39b-7218-0410-832d-ea1e28bc413d

deal.II/examples/step-39/postprocess.pl
deal.II/examples/step-39/step-39.cc

index 5d60b44f6032de78c2d6564171513a8cfda35593..567b14e68511cc06a9656bae7e2ba9c18cab99ff 100644 (file)
@@ -8,19 +8,21 @@ use strict;
 
 my $step;     # The iteration step in the adaptive loop
 my @dofs;     # The number of degrees of freedom in each step
-my @error;    # The error of the solution
+my @error;    # The energy error of the solution
+my @l2error;  # The L2-error of the solution
 my @estimate; # The a posteriori error estimate
 my @steps;    # The number of multigrid iteration steps
 while(<>)
 {
     $step = $1 if m/DEAL::Step\s*(\d+)/;
     $dofs[$step] = $1 if m/DEAL::DoFHandler\s*(\d+)/;
-    $error[$step] = $1 if m/DEAL::Error\s*(\S+)/;
+    $error[$step] = $1 if m/DEAL::energy-error:\s*(\S+)/;
+    $l2error[$step] = $1 if m/DEAL::L2-error:\s*(\S+)/;
     $estimate[$step] = $1 if m/DEAL::Estimate\s*(\S+)/;
     $steps[$step] = $1 if m/DEAL:\w+::Convergence step\s*(\S+)/;
 }
 
 for (my $i=0;$i<=$step;++$i)
 {
-    printf "%-3d\t%-7d\t%g\t%g\t%d\n", $i, $dofs[$i], $error[$i], $estimate[$i], $steps[$i];
+    printf "%-3d\t%-7d\t%g\t%g\t%g\t%d\n", $i, $dofs[$i], $error[$i], $estimate[$i], $l2error[$i], $steps[$i];
 }
index 67a51756c0622885d77844bbd58d273afd59081b..5d3e16a537971b1a798dbf797e97b40d101dee0e 100644 (file)
@@ -321,6 +321,153 @@ void Estimator<dim>::face(MeshWorker::DoFInfo<dim>& dinfo1,
   dinfo2.value(0) = dinfo1.value(0);
 }
 
+                                // Finally we have an integrator for
+                                // the error. Since the energy norm
+                                // for discontinuous Galerkin
+                                // problems not only involves the
+                                // difference of the gradient inside
+                                // the cells, but also the jump terms
+                                // across faces and at the boundary,
+                                // we cannot just use
+                                // VectorTools::integrate_difference().
+                                // Instead, we use the MeshWorker
+                                // interface to compute the error
+                                // ourselves.
+
+                                // There are several different ways
+                                // to define this energy norm, but
+                                // all of them are equivalent to each
+                                // other uniformly with mesh size
+                                // (some not uniformly with
+                                // polynomial degree). Here, we
+                                // choose
+                                // @f[
+                                // \|u\|_{1,h} = \sum_{K\in \mathbb
+                                // T_h} \|\nabla u\|_K^2
+                                // + \sum_{F \in F_h^i}
+                                // 4\sigma_F\|\{\!\{ u \mathbf
+                                // n\}\!\}\|^2_F
+                                // + \sum_{F \in F_h^b} 2\sigma_F\|u\|^2_F
+                                // @f]
+
+template <int dim>
+class ErrorIntegrator : public Subscriptor
+{
+  public:
+    static void cell(MeshWorker::DoFInfo<dim>& dinfo, typename MeshWorker::IntegrationInfo<dim>& info);
+    static void boundary(MeshWorker::DoFInfo<dim>& dinfo, typename MeshWorker::IntegrationInfo<dim>& info);
+    static void face(MeshWorker::DoFInfo<dim>& dinfo1,
+                    MeshWorker::DoFInfo<dim>& dinfo2,
+                    typename MeshWorker::IntegrationInfo<dim>& info1,
+                    typename MeshWorker::IntegrationInfo<dim>& info2);
+};
+
+                                // Here we have the integration on
+                                // cells. There is currently no good
+                                // interfce in MeshWorker that would
+                                // allow us to access values of
+                                // regular functions in the
+                                // quadrature points. Thus, we have
+                                // to create the vectors for the
+                                // exact function's values and
+                                // gradients inside the cell
+                                // integrator. After that, everything
+                                // is as before and we just add up
+                                // the squares of the differences.
+
+                                // Additionally to computing the error
+                                // in the energy norm, we use the
+                                // capability of the mesh worker to
+                                // compute two functionals at the
+                                // same time and compute the
+                                // <i>L<sup>2</sup></i>-error in the
+                                // same loop. Obviously, this one
+                                // does not have any jump terms and
+                                // only appears in the integration on
+                                // cells.
+template <int dim>
+void ErrorIntegrator<dim>::cell(
+  MeshWorker::DoFInfo<dim>& dinfo,
+  typename MeshWorker::IntegrationInfo<dim>& info)
+{
+  const FEValuesBase<dim>& fe = info.fe_values();
+  std::vector<Tensor<1,dim> > exact_gradients(fe.n_quadrature_points);
+  std::vector<double> exact_values(fe.n_quadrature_points);
+  
+  exact_solution.gradient_list(fe.get_quadrature_points(), exact_gradients);
+  exact_solution.value_list(fe.get_quadrature_points(), exact_values);
+  
+  const std::vector<Tensor<1,dim> >& Duh = info.gradients[0][0];
+  const std::vector<double>& uh = info.values[0][0];
+
+  for (unsigned k=0;k<fe.n_quadrature_points;++k)
+    {
+      double sum = 0;
+      for (unsigned int d=0;d<dim;++d)
+       {
+         const double diff = exact_gradients[k][d] - Duh[k][d];
+         sum += diff*diff;
+       }
+      const double diff = exact_values[k] - uh[k];
+      dinfo.value(0) +=  sum * fe.JxW(k);
+      dinfo.value(1) +=  diff*diff * fe.JxW(k);
+    }
+  dinfo.value(0) = std::sqrt(dinfo.value(0));
+  dinfo.value(1) = std::sqrt(dinfo.value(1));
+}
+
+
+template <int dim>
+void ErrorIntegrator<dim>::boundary(
+  MeshWorker::DoFInfo<dim>& dinfo,
+  typename MeshWorker::IntegrationInfo<dim>& info)
+{
+  const FEValuesBase<dim>& fe = info.fe_values();
+  
+  std::vector<double> exact_values(fe.n_quadrature_points);
+  exact_solution.value_list(fe.get_quadrature_points(), exact_values);
+  
+  const std::vector<double>& uh = info.values[0][0];
+  
+  const unsigned int deg = fe.get_fe().tensor_degree();
+  const double penalty = 2. * deg * (deg+1) * dinfo.face->measure() / dinfo.cell->measure();
+  
+  for (unsigned k=0;k<fe.n_quadrature_points;++k)
+    {
+      const double diff = exact_values[k] - uh[k];
+      dinfo.value(0) += penalty * diff * diff * fe.JxW(k);
+    }
+  dinfo.value(0) = std::sqrt(dinfo.value(0));  
+}
+
+
+template <int dim>
+void ErrorIntegrator<dim>::face(
+  MeshWorker::DoFInfo<dim>& dinfo1,
+  MeshWorker::DoFInfo<dim>& dinfo2,
+  typename MeshWorker::IntegrationInfo<dim>& info1,
+  typename MeshWorker::IntegrationInfo<dim>& info2)
+{
+  const FEValuesBase<dim>& fe = info1.fe_values();
+  const std::vector<double>& uh1 = info1.values[0][0];
+  const std::vector<double>& uh2 = info2.values[0][0];
+  
+  const unsigned int deg = fe.get_fe().tensor_degree();
+  const double penalty1 = deg * (deg+1) * dinfo1.face->measure() / dinfo1.cell->measure();
+  const double penalty2 = deg * (deg+1) * dinfo2.face->measure() / dinfo2.cell->measure();
+  const double penalty = penalty1 + penalty2;
+  
+  for (unsigned k=0;k<fe.n_quadrature_points;++k)
+    {
+      double diff = uh1[k] - uh2[k];
+      dinfo1.value(0) += (penalty * diff*diff)
+                        * fe.JxW(k);
+    }
+  dinfo1.value(0) = std::sqrt(dinfo1.value(0));
+  dinfo2.value(0) = dinfo1.value(0);
+}
+
+
 
                                 // @sect3{The main class}
 
@@ -766,23 +913,6 @@ Step39<dim>::solve()
   solver.solve(matrix, solution, right_hand_side, preconditioner);
 }
 
-                                // Here we compare our finite element
-                                // solution with the (known) exact
-                                // solution and compute the mean
-                                // quadratic error of the gradient.
-template <int dim>
-void
-Step39<dim>::error()
-{
-  const unsigned int n_gauss_points = dof_handler.get_fe().tensor_degree()+2;
-  Vector<double> cell_errors(triangulation.n_active_cells());
-  
-  QGauss<dim> quadrature(n_gauss_points);
-  VectorTools::integrate_difference(mapping, dof_handler, solution, exact_solution,
-                                   cell_errors, quadrature, VectorTools::H1_seminorm);
-  deallog << "Error    " << cell_errors.l2_norm() << std::endl;
-}
-
 
                                 // Another clone of the assemble
                                 // function. The big difference to
@@ -867,6 +997,66 @@ Step39<dim>::estimate()
   return estimates.block(0).l2_norm();
 }
 
+                                // Here we compare our finite element
+                                // solution with the (known) exact
+                                // solution and compute the mean
+                                // quadratic error of the gradient
+                                // and the function itself. This
+                                // function is a clone of the
+                                // estimation function right above.
+
+                                // Since we compute the error in the
+                                // energy and the
+                                // <i>L<sup>2</sup></i>-norm,
+                                // respectively, our block vector
+                                // needs two blocks here.
+template <int dim>
+void
+Step39<dim>::error()
+{
+  BlockVector<double> errors(2);
+  errors.block(0).reinit(triangulation.n_active_cells());
+  errors.block(1).reinit(triangulation.n_active_cells());
+  unsigned int i=0;
+  for (typename Triangulation<dim>::active_cell_iterator cell = triangulation.begin_active();
+       cell != triangulation.end();++cell,++i)
+    cell->set_user_index(i);
+
+  MeshWorker::IntegrationInfoBox<dim> info_box;
+  const unsigned int n_gauss_points = dof_handler.get_fe().tensor_degree()+1;
+  info_box.initialize_gauss_quadrature(n_gauss_points, n_gauss_points+1, n_gauss_points);
+
+  NamedData<Vector<double>* > solution_data;
+  solution_data.add(&solution, "solution");
+  
+  info_box.cell_selector.add("solution", true, true, false);
+  info_box.boundary_selector.add("solution", true, false, false);
+  info_box.face_selector.add("solution", true, false, false);
+  
+  info_box.add_update_flags_cell(update_quadrature_points);
+  info_box.add_update_flags_boundary(update_quadrature_points);
+  info_box.initialize(fe, mapping, solution_data);
+  
+  MeshWorker::DoFInfo<dim> dof_info(dof_handler);
+  
+  MeshWorker::Assembler::CellsAndFaces<double> assembler;  
+  NamedData<BlockVector<double>* > out_data;
+  BlockVector<double>* est = &errors;
+  out_data.add(est, "cells");
+  assembler.initialize(out_data, false);
+  
+  MeshWorker::integration_loop<dim, dim> (
+    dof_handler.begin_active(), dof_handler.end(),
+    dof_info, info_box,
+    &ErrorIntegrator<dim>::cell,
+    &ErrorIntegrator<dim>::boundary,
+    &ErrorIntegrator<dim>::face,
+    assembler);
+
+  deallog << "energy-error: " << errors.block(0).l2_norm() << std::endl;
+  deallog << "L2-error:     " << errors.block(1).l2_norm() << std::endl;
+}
+
 
                                 // Some graphical output
 template <int dim>

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.