]> https://gitweb.dealii.org/ - dealii.git/commitdiff
Add TensorProductMatrixCreator namespace 14316/head
authorPeter Munch <peterrmuench@gmail.com>
Mon, 19 Sep 2022 17:45:53 +0000 (19:45 +0200)
committerPeter Munch <peterrmuench@gmail.com>
Wed, 28 Sep 2022 06:06:07 +0000 (08:06 +0200)
include/deal.II/lac/tensor_product_matrix.h
include/deal.II/numerics/tensor_product_matrix_creator.h [new file with mode: 0644]

index 0c167ea0c90368638f128504204d6769b8ab94da..86a32dcb26016df82d2a8447b5ab6d1c9faa64b8 100644 (file)
@@ -472,7 +472,6 @@ private:
 };
 
 
-
 /*----------------------- Inline functions ----------------------------------*/
 
 #ifndef DOXYGEN
diff --git a/include/deal.II/numerics/tensor_product_matrix_creator.h b/include/deal.II/numerics/tensor_product_matrix_creator.h
new file mode 100644 (file)
index 0000000..1242e64
--- /dev/null
@@ -0,0 +1,375 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2022 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.md at
+// the top level directory of deal.II.
+//
+// ---------------------------------------------------------------------
+
+#ifndef dealii_tensor_product_matrix_creator_h
+#define dealii_tensor_product_matrix_creator_h
+
+
+#include <deal.II/base/config.h>
+
+#include <deal.II/base/quadrature.h>
+
+#include <deal.II/dofs/dof_handler.h>
+
+#include <deal.II/fe/fe.h>
+#include <deal.II/fe/fe_tools.h>
+#include <deal.II/fe/fe_values.h>
+#include <deal.II/fe/mapping_q1.h>
+
+#include <deal.II/grid/grid_generator.h>
+#include <deal.II/grid/tria.h>
+
+#include <set>
+
+DEAL_II_NAMESPACE_OPEN
+
+
+/**
+ * A namespace with functions that create input for certain standard matrices
+ * for the classes TensorProductMatrixSymmetricSum and
+ * TensorProductMatrixSymmetricSumCache.
+ */
+namespace TensorProductMatrixCreator
+{
+  /**
+   * Boundary type that can be used in create_laplace_tensor_product_matrix();
+   */
+  enum LaplaceBoundaryType
+  {
+    dirichlet,
+    neumann,
+    internal_boundary,
+  };
+
+  /**
+   * Create 1D mass matrix and 1D derivative matrix for a scalar
+   * constant-coefficient
+   * Laplacian for a @p dim dimensional Cartesian cell. Its boundary types
+   * can be specified with @p boundary_ids. The cell extent (including the cell extent
+   * of each neighbor) can be specified via @p cell_extent. With @p n_overlap, an
+   * overlap with neighboring cells can be specified. The default value is one,
+   * which correspond to all matrix entries restricted to the cell-local DoFs.
+   */
+  template <int dim, typename Number>
+  std::pair<std::array<FullMatrix<Number>, dim>,
+            std::array<FullMatrix<Number>, dim>>
+  create_laplace_tensor_product_matrix(
+    const FiniteElement<1> &                            fe,
+    const Quadrature<1> &                               quadrature,
+    const dealii::ndarray<LaplaceBoundaryType, dim, 2> &boundary_ids,
+    const dealii::ndarray<double, dim, 3> &             cell_extent,
+    const unsigned int                                  n_overlap = 1);
+
+  /**
+   * Same as above but the boundary IDs are extracted from the given @p cell
+   * and are mapped to the boundary type via the sets @p dirichlet_boundaries and @p neumann_boundaries.
+   */
+  template <int dim, typename Number>
+  std::pair<std::array<FullMatrix<Number>, dim>,
+            std::array<FullMatrix<Number>, dim>>
+  create_laplace_tensor_product_matrix(
+    const typename Triangulation<dim>::cell_iterator &cell,
+    const std::set<types::boundary_id> &              dirichlet_boundaries,
+    const std::set<types::boundary_id> &              neumann_boundaries,
+    const FiniteElement<1> &                          fe,
+    const Quadrature<1> &                             quadrature,
+    const dealii::ndarray<double, dim, 3> &           cell_extent,
+    const unsigned int                                n_overlap = 1);
+
+} // namespace TensorProductMatrixCreator
+
+
+
+/*----------------------- Inline functions ----------------------------------*/
+
+
+namespace TensorProductMatrixCreator
+{
+  namespace internal
+  {
+    template <typename Number>
+    std::tuple<FullMatrix<Number>, FullMatrix<Number>, bool>
+    create_reference_mass_and_stiffness_matrices(
+      const FiniteElement<1> &fe,
+      const Quadrature<1> &   quadrature)
+    {
+      Triangulation<1> tria;
+      GridGenerator::hyper_cube(tria);
+
+      DoFHandler<1> dof_handler(tria);
+      dof_handler.distribute_dofs(fe);
+
+      MappingQ1<1> mapping;
+
+      const unsigned int n_dofs_1D = fe.n_dofs_per_cell();
+
+      FullMatrix<Number> mass_matrix_reference(n_dofs_1D, n_dofs_1D);
+      FullMatrix<Number> derivative_matrix_reference(n_dofs_1D, n_dofs_1D);
+
+      FEValues<1> fe_values(mapping,
+                            fe,
+                            quadrature,
+                            update_values | update_gradients |
+                              update_JxW_values);
+
+      fe_values.reinit(tria.begin());
+
+      const auto lexicographic_to_hierarchic_numbering =
+        Utilities::invert_permutation(
+          FETools::hierarchic_to_lexicographic_numbering<1>(
+            fe.tensor_degree()));
+
+      for (const unsigned int q_index : fe_values.quadrature_point_indices())
+        for (const unsigned int i : fe_values.dof_indices())
+          for (const unsigned int j : fe_values.dof_indices())
+            {
+              mass_matrix_reference(i, j) +=
+                (fe_values.shape_value(lexicographic_to_hierarchic_numbering[i],
+                                       q_index) *
+                 fe_values.shape_value(lexicographic_to_hierarchic_numbering[j],
+                                       q_index) *
+                 fe_values.JxW(q_index));
+
+              derivative_matrix_reference(i, j) +=
+                (fe_values.shape_grad(lexicographic_to_hierarchic_numbering[i],
+                                      q_index) *
+                 fe_values.shape_grad(lexicographic_to_hierarchic_numbering[j],
+                                      q_index) *
+                 fe_values.JxW(q_index));
+            }
+
+      return {mass_matrix_reference, derivative_matrix_reference, false};
+    }
+  } // namespace internal
+
+
+
+  template <int dim, typename Number>
+  std::pair<std::array<FullMatrix<Number>, dim>,
+            std::array<FullMatrix<Number>, dim>>
+  create_laplace_tensor_product_matrix(
+    const FiniteElement<1> &                            fe,
+    const Quadrature<1> &                               quadrature,
+    const dealii::ndarray<LaplaceBoundaryType, dim, 2> &boundary_ids,
+    const dealii::ndarray<double, dim, 3> &             cell_extent,
+    const unsigned int                                  n_overlap)
+  {
+    // 1) create element mass and siffness matrix (without overlap)
+    const auto create_reference_mass_and_stiffness_matrices =
+      internal::create_reference_mass_and_stiffness_matrices<Number>(
+        fe, quadrature);
+
+    const auto &M_ref =
+      std::get<0>(create_reference_mass_and_stiffness_matrices);
+    const auto &K_ref =
+      std::get<1>(create_reference_mass_and_stiffness_matrices);
+    const auto &is_dg =
+      std::get<2>(create_reference_mass_and_stiffness_matrices);
+
+    AssertIndexRange(n_overlap, M_ref.n());
+    AssertIndexRange(0, n_overlap);
+    AssertThrow(is_dg == false, ExcNotImplemented());
+
+    // 2) loop over all dimensions and create 1D mass and stiffness
+    // matrices so that boundary conditions and overlap are considered
+
+    const unsigned int n_dofs_1D              = M_ref.n();
+    const unsigned int n_dofs_1D_with_overlap = M_ref.n() - 2 + 2 * n_overlap;
+
+    std::array<FullMatrix<Number>, dim> Ms;
+    std::array<FullMatrix<Number>, dim> Ks;
+
+    const auto clear_row_and_column = [&](const unsigned int n, auto &matrix) {
+      for (unsigned int i = 0; i < n_dofs_1D_with_overlap; ++i)
+        {
+          matrix[i][n] = 0.0;
+          matrix[n][i] = 0.0;
+        }
+    };
+
+    for (unsigned int d = 0; d < dim; ++d)
+      {
+        Ms[d].reinit(n_dofs_1D_with_overlap, n_dofs_1D_with_overlap);
+        Ks[d].reinit(n_dofs_1D_with_overlap, n_dofs_1D_with_overlap);
+
+        // inner cell
+        for (unsigned int i = 0; i < n_dofs_1D; ++i)
+          for (unsigned int j = 0; j < n_dofs_1D; ++j)
+            {
+              const unsigned int i0 = i + n_overlap - 1;
+              const unsigned int j0 = j + n_overlap - 1;
+              Ms[d][i0][j0]         = M_ref[i][j] * cell_extent[d][1];
+              Ks[d][i0][j0]         = K_ref[i][j] / cell_extent[d][1];
+            }
+
+        // left neighbor or left boundary
+        if (boundary_ids[d][0] == LaplaceBoundaryType::internal_boundary)
+          {
+            // left neighbor
+            Assert(cell_extent[d][0] > 0.0, ExcInternalError());
+
+            for (unsigned int i = 0; i < n_overlap; ++i)
+              for (unsigned int j = 0; j < n_overlap; ++j)
+                {
+                  const unsigned int i0 = n_dofs_1D - n_overlap + i;
+                  const unsigned int j0 = n_dofs_1D - n_overlap + j;
+                  Ms[d][i][j] += M_ref[i0][j0] * cell_extent[d][0];
+                  Ks[d][i][j] += K_ref[i0][j0] / cell_extent[d][0];
+                }
+          }
+        else
+          {
+            if (boundary_ids[d][0] == LaplaceBoundaryType::dirichlet)
+              {
+                // left DBC
+                const unsigned i0 = n_overlap - 1;
+                clear_row_and_column(i0, Ms[d]);
+                clear_row_and_column(i0, Ks[d]);
+              }
+            else if (boundary_ids[d][0] == LaplaceBoundaryType::neumann)
+              {
+                // left NBC -> nothing to do
+              }
+            else
+              {
+                AssertThrow(false, ExcNotImplemented());
+              }
+          }
+
+        // right neighbor or right boundary
+        if (boundary_ids[d][1] == LaplaceBoundaryType::internal_boundary)
+          {
+            Assert(cell_extent[d][2] > 0.0, ExcInternalError());
+
+            for (unsigned int i = 0; i < n_overlap; ++i)
+              for (unsigned int j = 0; j < n_overlap; ++j)
+                {
+                  const unsigned int i0 = n_overlap + n_dofs_1D + i - 2;
+                  const unsigned int j0 = n_overlap + n_dofs_1D + j - 2;
+                  Ms[d][i0][j0] += M_ref[i][j] * cell_extent[d][2];
+                  Ks[d][i0][j0] += K_ref[i][j] / cell_extent[d][2];
+                }
+          }
+        else
+          {
+            if (boundary_ids[d][1] == LaplaceBoundaryType::dirichlet)
+              {
+                // right DBC
+                const unsigned i0 = n_overlap + n_dofs_1D - 2;
+                clear_row_and_column(i0, Ms[d]);
+                clear_row_and_column(i0, Ks[d]);
+              }
+            else if (boundary_ids[d][1] == LaplaceBoundaryType::neumann)
+              {
+                // right NBC -> nothing to do
+              }
+            else
+              {
+                AssertThrow(false, ExcNotImplemented());
+              }
+          }
+      }
+
+    return {Ms, Ks};
+  }
+
+
+  template <int dim, typename Number>
+  std::pair<std::array<FullMatrix<Number>, dim>,
+            std::array<FullMatrix<Number>, dim>>
+  create_laplace_tensor_product_matrix(
+    const typename Triangulation<dim>::cell_iterator &cell,
+    const std::set<types::boundary_id> &              dirichlet_boundaries,
+    const std::set<types::boundary_id> &              neumann_boundaries,
+    const FiniteElement<1> &                          fe,
+    const Quadrature<1> &                             quadrature,
+    const dealii::ndarray<double, dim, 3> &           cell_extent,
+    const unsigned int                                n_overlap)
+  {
+    dealii::ndarray<LaplaceBoundaryType, dim, 2> boundary_ids;
+
+    for (unsigned int d = 0; d < dim; ++d)
+      {
+        // left neighbor or left boundary
+        if ((cell->at_boundary(2 * d) == false) ||
+            cell->has_periodic_neighbor(2 * d))
+          {
+            // left neighbor
+            Assert(cell_extent[d][0] > 0.0, ExcInternalError());
+
+            boundary_ids[d][0] = LaplaceBoundaryType::internal_boundary;
+          }
+        else
+          {
+            const auto bid = cell->face(2 * d)->boundary_id();
+            if (dirichlet_boundaries.find(bid) !=
+                dirichlet_boundaries.end() /*DBC*/)
+              {
+                // left DBC
+                boundary_ids[d][0] = LaplaceBoundaryType::dirichlet;
+              }
+            else if (neumann_boundaries.find(bid) !=
+                     neumann_boundaries.end() /*NBC*/)
+              {
+                // left NBC
+                boundary_ids[d][0] = LaplaceBoundaryType::neumann;
+              }
+            else
+              {
+                AssertThrow(false, ExcNotImplemented());
+              }
+          }
+
+        // right neighbor or right boundary
+        if ((cell->at_boundary(2 * d + 1) == false) ||
+            cell->has_periodic_neighbor(2 * d + 1))
+          {
+            Assert(cell_extent[d][2] > 0.0, ExcInternalError());
+
+            boundary_ids[d][1] = LaplaceBoundaryType::internal_boundary;
+          }
+        else
+          {
+            const auto bid = cell->face(2 * d + 1)->boundary_id();
+            if (dirichlet_boundaries.find(bid) !=
+                dirichlet_boundaries.end() /*DBC*/)
+              {
+                // right DBC
+                boundary_ids[d][1] = LaplaceBoundaryType::dirichlet;
+              }
+            else if (neumann_boundaries.find(bid) !=
+                     neumann_boundaries.end() /*NBC*/)
+              {
+                // right NBC
+                boundary_ids[d][1] = LaplaceBoundaryType::neumann;
+              }
+            else
+              {
+                AssertThrow(false, ExcNotImplemented());
+              }
+          }
+      }
+
+    return create_laplace_tensor_product_matrix<dim, Number>(
+      fe, quadrature, boundary_ids, cell_extent, n_overlap);
+  }
+
+} // namespace TensorProductMatrixCreator
+
+
+
+DEAL_II_NAMESPACE_CLOSE
+
+#endif

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.