Reference documentation for deal.II version GIT cd34b0849a 2022-05-26 19:50:01+00:00
\(\newcommand{\dealvcentcolon}{\mathrel{\mathop{:}}}\) \(\newcommand{\dealcoloneq}{\dealvcentcolon\mathrel{\mkern-1.2mu}=}\) \(\newcommand{\jump}[1]{\left[\!\left[ #1 \right]\!\right]}\) \(\newcommand{\average}[1]{\left\{\!\left\{ #1 \right\}\!\right\}}\)
Changes since the last release

This is the list of changes made since the last release of deal.II. All entries are signed with the names of the authors.


Following are a few modifications to the library that unfortunately are incompatible with previous versions of the library, but which we deem necessary for the future maintainability of the library. Unfortunately, some of these changes will require modifications to application programs. We apologize for the inconvenience this causes.

  1. Changed: For ParameterHandler::OutputStyle::JSON or ParameterHandler::OutputStyle::ShortJSON, the function ParameterHandler::print_parameters() demangles now the parameters before printing the output. Users do not need to demangle the output themselves anymore.
    (Magdalena Schreter, Peter Munch, 2022/05/25)

  2. Improved: GridTools::get_coarse_mesh_description() no longer returns subcell data corresponding to default values (i.e., when the manifold_id is numbers::flat_manifold_id and when the boundary_id is numbers::internal_face_boundary_id). By skipping this data, which is not required by Triangulation::create_triangulation(), the total amount of data returned by this function is significantly lowered (and, therefore, so is the cost of creating a new Triangulation from the output).
    (David Wells, 2022/05/09)

  3. Changed: All functions n_nonzero_elements(), which are members of the SparseMatrix and SparsityPattern family of classes from PETSc and Trilinos, now always return std::uint64_t instead of types::global_dof_index.
    (Marc Fehling, 2022/04/04)

  4. Changed: To be able to guarantee correctness of the application of constraints in MatrixFree, we require that users call AffineConstraints::close() before calling MatrixFree::reinit() if AffineConstraints objects are passed that are not empty. One can check if an AffineConstraints object is closed with the new function AffineConstraints::is_closed().
    (Peter Munch, 2022/03/08)

  5. Fixed: Callback functions attached to any weighted load balancing signal of parallel::distributed::Triangulation objects now need to handle cells with CELL_INVALID status explicitely.
    If a cell gets refined, only the first child has the CELL_REFINE status, while all other children are CELL_INVALID.
    (Marc Fehling, 2022/02/25)

  6. Changed: For weighted load balancing with parallel::distributed::Triangulation objects, an intial weight of 1000 will no longer be assigned to each cell.
    The old signal Triangulation::Signals::cell_weight has been deprecated. Please use the new signal Triangulation::Signals::weight instead.
    You can invoke the old behavior by connecting a function to the new signal that returns the base weight like this:

    -> unsigned int { return 1000; });
    typename ::Triangulation< dim, spacedim >::CellStatus CellStatus
    Definition: tria.h:294
    typename ::Triangulation< dim, spacedim >::cell_iterator cell_iterator
    Definition: tria.h:270
    const ::parallel::distributed::Triangulation< dim, spacedim > * triangulation

    (Marc Fehling, 2022/02/25)

  7. Updated: Spurious inclusions of the header files deal.II/grid/grid_tools.h and deal.II/grid/grid_tools_cache.h from other headers of deal.II have been removed. User codes relying on this implicit inclusion will need to add the respective include files.
    (Martin Kronbichler, 2022/02/22)

  8. Deprecated: We have introduced new versions of DoFTools::extract_locally_active_dofs(), DoFTools::extract_locally_active_level_dofs(), DoFTools::extract_locally_relevant_dofs(), and DoFTools::extract_locally_relevant_level_dofs(). These versions return the index sets directly. The old versions have been early deprecated.
    (Peter Munch, 2022/02/23)

  9. Changed: The function MatrixFreeTools::compute_diagonal() does not initialize the output vector anymore. Users need to call MatrixFree:initialize_dof_vector() themselves. This change is required to be able to also support block vectors.
    (Magdalena Schreter, Peter Munch, 2022/02/01)

  10. Changed: The two algorithms in the ConsensusAlgorithms namespace (along with their base class) required a callback function that correctly sized the buffer into which the answer to a request was to be written. These algorithms have been rewritten in ways that make this no longer necessary, and they now correctly size these buffers themselves. As a consequence, the ConsensusAlgorithms::Process class no longer has the prepare_for_answer() function, and the ConsensusAlgorithms::AnonymousProcess class no longer has a corresponding member variable; in the latter case, the constructor of the class also no longer takes an argument to this effect.
    (Wolfgang Bangerth, Peter Munch, 2022/01/18)

  11. Updated: The header inclusions of several files in the deal.II/matrix_free subfolder have been cleaned up to make the library more maintainable. As a result, some headers are no longer implicitly included and a user code might need to add the respective include file.
    (Martin Kronbichler, 2021/11/30)

  12. Renamed: The namespace Functions::LevelSet has been renamed to Functions::SignedDistance.
    (Peter Munch, 2021/10/19)

  13. Changed: For parallel::distributed::Triangulation, vertices will no longer be communicated automatically to p4est in debug mode. Please set the corresponding flag parallel::distributed::Triangulation::Settings::communicate_vertices_to_p4est while constructing your triangulation object.
    (Marc Fehling, 2021/10/13)

  14. Changed: The 3D implementation of GridTools::rotate() with an integer for a Cartesian coordinate direction has been superseded by the version that accepts unit vectors as rotation axes. Further, Physics::Transformations::Rotations::rotation_matrix_3d() now requires a Tensor<1,3> object instead of a Point<3> as an axis.
    (Marc Fehling, 2021/10/11)

  15. Updated: deal.II now requires ginkgo version 1.4.0 or newer (Fabian Castelli, 2021/09/01)

  16. Deprecated: The Tensor and SymmetricTensor classes had functions begin_raw() and end_raw() that suggested that the elements of these objects are stored in arrays and consequently could be accessed consecutively through iterators. While they are indeed stored contiguously in memory, they are not part of an array and, as a consequence, C++ does not actually allow this kind of access. The functions have therefore been deprecated and will be removed in due time.
    As a consequence, the make_array_view() functions that allow turning a Tensor or SymmetricTensor into an ArrayView object have now also been deprecated.
    (Wolfgang Bangerth, 2021/08/19)

  17. Changed: The DiagonalMatrix objects set up by MatrixFreeOperators::Base and its inheriting classes no set ghost entries.
    (David Wells, 2021/08/17)

  18. Deprecated: Many of the member functions of the FEInterfaceValues class, as well as the FEInterfaceViews::Scalar and FEInterfaceViews::Vector classes, that are used to compute jumps and averages of shape function-related quantities have been renamed in such a way that they are more expressive about what they compute.
    (Jean-Paul Pelteret, 2021/07/31)

  19. Changed: MatrixFreeOperators::MassOperator now computes its true diagonal (and not a lumped mass matrix) as its diagonal.
    (David Wells, 2021/07/29)

  20. Deprecated: The autopartition parameter has been removed from parallel::DistributedTriangulationBase::load() and all inheriting classes.
    (Marc Fehling, 2021/07/24)

  21. Changed: Custom repartitioning via Triangulation::Signals::cell_weight is no longer part of parallel::distributed::Triangulation::load(). If desired, call parallel::distributed::Triangulation::repartition() manually after load().
    (Marc Fehling, 2021/07/23)

  22. Changed: The class MappingQ now applies a high-order mapping to a cells in the mesh, not just the ones touching the boundary. This makes the class completely equivalent to the previous class MappingQGeneric. All functionality in the latter class has been integrated into the MappingQ class. This avoids errors with curved manifolds in the interior of the domain, where MappingQ could lead to gaps or overlaps in the computational domain. In case the slightly better performance of using MappingQ1 in interior cells is desired, hp::MappingCollection can be used to replicate the old behavior.
    (Martin Kronbichler, 2021/07/10)

  23. Updated: FE_RaviartThomasNodal now uses a different polynomial space to allow for a simpler use for faces in non-standard orientation. The new polynomials are anisotropic tensor products of Lagrange polynomials on the points of a Gauss–Lobatto quadrature formula. This change leads to different entries in the matrices and constraints, for example, but as the resulting polynomial space spans the same polynomials, no change in accuracy should be expected.
    (Martin Kronbichler, Katharina Kormann, Konrad Simon, 2021/06/12)

  24. Removed: The PETSc Eisenstat preconditioner wrapper, which has never worked correctly in parallel, has been removed.
    (David Wells, 2021/06/11)

  25. Removed: The deprecated exception ParameterHandler::ExcInvalidEntryForPatternXML and the deprecated overload of SparsityTools::gather_sparsity_pattern() have been removed.
    (Daniel Arndt, 2021/06/07)

  26. Removed: The deprecated member function TriangulationBase::compute_vertices_with_ghost_neighbors() has been removed.
    (Daniel Arndt, 2021/06/07)

  27. Removed: The deprecated constructors for FESeries::Fourier and FESeries::Legendre have been removed.
    (Daniel Arndt, 2021/06/07)

  28. Removed: The deprecated specializations of VectorTools::integrate_difference() and VectorTools::project_boundary_values_curl_conforming() have been removed.
    (Daniel Arndt, 2021/06/04)

  29. Removed: The deprecated overloads of FETools::lexicographic_to_hierarchic_numbering() and FETools::hierarchic_to_lexicographic_numbering() have been removed.
    (Daniel Arndt, 2021/06/03)

  30. Removed: The deprecated member variable VectorizedArray::n_array_elements has been removed.
    (Daniel Arndt, 2021/06/03)

  31. Removed: The deprecated member function CellAccessor::active() has been removed.
    (Daniel Arndt, 2021/06/03)

  32. Removed: The deprecated member functions DataOut::first_cell() and DataOut::next_cell() have been removed.
    (Daniel Arndt, 2021/06/02)

  33. Changed: Trilinos support in deal.II requires both deal.II and Trilinos to be configured with MPI support.
    (Daniel Arndt, 2021/06/02)

  34. Removed: The deprecated build_matrices() member functions in various multigrid transfer classes have been removed.
    (Daniel Arndt, 2021/06/01)

  35. Removed: The deprecated versions of DoFTools::count_dofs_per_component() and DoFTools::count_dofs_per_block() have been removed.
    (Daniel Arndt, 2021/06/01)

  36. Removed: The deprecated overloads for DoFTools::extract_hanging_node_dofs(), and DoFTools::extract_dofs() have been removed.
    (Daniel Arndt, 2021/05/28)

  37. Removed: Deprecated parallel::CellWeights member functions have been removed.
    (Daniel Arndt, 2021/05/27)

  38. Removed: The overloads in CUDAWrappers::FEEvaluation that take a local dof index or a quadrature point as arguement have been removed. Use the ones that don't use these arguments in their interface instead.
    (Daniel Arndt, 2021/05/26)

  39. Removed: The deprecated parallel::Triangulation class has been removed. Use parallel::TriangulationBase instead.
    (Daniel Arndt, 2021/05/26)

  40. Removed: The deprecated member field MatrixFree::AdditionalData::level_mg_handler.
    (Peter Munch, 2021/05/25)

  41. Removed: The deprecated parallel::distributed::CellDataTransfer::CoarseningStrategies struct has been removed. Use AdaptationStrategies::Coarsening instead.
    (Daniel Arndt, 2021/05/25)

  42. Removed: The deprecated member functions DoFHandler::locally_owned_dofs_per_processor(), DoFHandler::n_locally_owned_dofs_per_processor(), and DoFHandler::n_locally_owned_mg_dofs_per_processor() have been removed.
    (Daniel Arndt, 2021/05/24)

  43. Deprecated: The template arguments of the following classes have been changed to avoid the legacy argument DoFHandlerType:

    Please change your code accordingly.
    If for some reason, you need a code that is compatible with deal.II 9.3 and the subsequent release, a Legacy namespace has been introduced with aliases of these classes with their original interface. You can make the following substitutions to your code for each of the affected classes:

    • X → Legacy::X

    To perform this substitution automatically, you may use a search and replace script like the following made for the bash shell:

    classes=(SolutionTransfer parallel::distributed::SolutionTransfer Functions::FEFieldFunction DataOut DataOut_DoFData DataOutFaces DataOutRotation)
    for c in \${classes[@]}; do
    find /path/to/your/code -type f -exec sed -i -E "/(\w\${c}|\${c}[^<]|Particles::\${c}|distributed::\${c}|^\s*(\/\/|\*))/! s/\${c}/Legacy::\${c}/g" {} \;

    (Marc Fehling, 2020/11/21)


  1. New: Added support for the CGAL library ( The following features are supported:

    • Conversion from deal.II to CGAL point types and viceversa
    • Conversion from deal.II cell types to CGAL::Surface_mesh and viceversa
    • Conversion from deal.II Triangulation to CGAL::Surface_mesh and viceversa
    • Insertion of deal.II points in CGAL triangulation types
    • Conversion from CGAL::Triangulation_2 to Triangulation<dim, 2>
    • Conversion from CGAL::Triangulation_3 to Triangulation<dim, 3>
    • Conversion from CGAL::Surface_mesh or CGAL::Polyhedron_3 to Triangulation<2, 3>

    (Marco Feder, Luca Heltai, 2022/04/26)

  2. New: Added a new function DoFRenumbering::support_point_wise() which renumbers DoFs so that DoFs with identical support points are now numbered adjacently.
    (David Wells, 2021/03/11)

  3. New: A tutorial, step-85, which shows how to use functionality in the NonMatching namespace to solve a PDE with the cut finite element method, when the domain is described by a level set function.
    (Simon Sticko, 2022/01/07)

  4. New: Geometry GridGenerator::pipe_junction() creates an intersection of arbitrary truncated cones in 3D. It can be used to generate T-pipes, Y-pipes, corner pieces and other fittings.
    (Marc Fehling, Maximilian Bergbauer, Peter Munch, Martin Kronbichler, 2021/12/01)

  5. New: The step-49 tutorial now has an extended section about generating meshes from Gmsh with a more interesting example.
    (Timo Heister, Sean Ingimarson, 2021/10/08)

  6. New: Added an interface to p4est_search.h (>=v2.2) to find the MPI rank of cells owning remote points without communication. This works in the special (but not uncommon) case that no nontrivial manifold is attached to the triangulation, i.e., the triangulation has only FlatManifold attached to it.
    (Konrad Simon, 2021/09/24)

  7. New: The tutorial program step-82 (local discontinuous Galerkin method for the bi-Laplacian problem) has been added.
    (Diane Guignard, 2021/09/17)

  8. Added: Evaluation and integration of hessians both on cells and on faces is now available in the matrix-free framework.
    (Maximilian Bergbauer, Martin Kronbichler, Peter Munch, 2021/09/06)

  9. New: VTU files have traditionally been used for visualization purposes, but more recently they have also been used for more general postprocessing tasks since they can easily be imported into Python, for example. By setting the DataOutBase::VtkFlags::physical_units variable and attaching such a flags object to a DataOut, Particles::DataOut, or MatrixOut object, one can now specify the physical units associated with variables that are written into VTU files, so that they can be parsed by scripts that do further postprocessing.
    (Wolfgang Bangerth, 2021/07/09)

  10. Improved: DataOut_DoFData and the derived classes create internally a copy of vectors attached via add_data_vector() and are now responsible for updating ghost values so that users do not need to call update_ghost_values() and potentially zero_out_ghost_values().
    (Peter Munch, Magdalena Schreter, 2021/05/27)

  11. New: The new postprocessing class DataOutResample is similar to DataOut but allows to interpolate the result onto a second triangulation.
    (Peter Munch, Martin Kronbichler, Magdalena Schreter, 2021/06/08)

  12. Improved: The internal particle handler storage structure has been reworked to allow for significantly faster particle sorting and iteration as well as for future improvements and extensions.
    (Rene Gassmoeller, Bruno Blais, Martin Kronbichler, Peter Munch, 2021/06/06)

Specific improvements

  1. New: PreconditionChebyshev can now also estimate eigenvalues with the power method, rather than the Lanczos algorithm underlying the SolverCG estimation. This is useful for matrices that are slightly non-symmetric such that the eigenvalue estimate of SolverCG breaks down, but not too much so that the Chebyshev iteration is still a useful iteration.
    (Martin Kronbichler, 2022/05/23)

  2. New: The Utilities::pack() and Utilities::unpack() functions are now optimized for the rather common case where the argument is of type std::vector<T> where T is a type that can be copied bit by bit (using memcpy()).
    (Wolfgang Bangerth, 2022/05/05)

  3. Improved: The thread-local storage in FunctionParser has been reworked to avoid extra locking and unlocking calls. This makes evaluations of simple expressions about three times faster.
    (David Wells, 2022/05/16)

  4. New: The hdf5 DataOut now correctly supports very large output files with more than 2 billion nodes/vertices.
    (Sean Ingimarson, Timo Heister, Jiaqi Zhang, 2022/05/10)

  5. New: The class SolverCG now supports the interleaving of vector operations with the matrix-vector product. The prerequisite is an associated MatrixType class to provide a vmult class with two std::function objects to specify the operation before and after the matrix-vector product, and a PreconditionerType class that provides either a function apply to apply the preconditioner action on a single element or and apply_to_subrange(const unsigned int, const unsigned int) const that can selectively apply the precondition on a part of a vector. For optimal performance, the matrix and preconditioner types need to agree on suitable sizes for the sub-ranges.
    (Dmytro Sashko, Martin Kronbichler, Peter Munch, 2022/05/09)

  6. New: The class SolverFlexibleCG implements a flexible variant of the conjugate gradient method that runs the so-called Polak-Ribiere update formula instead of the default Fletcher-Reeves formula of SolverCG.
    (Martin Kronbichler, 2022/05/06)

  7. New: There is now a function ReferenceCell::get_midpoint_quadrature() that generalizes QMidpoint.
    (Wolfgang Bangerth, 2022/05/05)

  8. New: There are now functions ReferenceCell::volume() and ReferenceCells::barycenter() that compute what their names suggest for the various kinds of reference cells deal.II supports.
    (Wolfgang Bangerth, 2022/05/04)

  9. New: MGConstrainedDoFs now has a function add_boundary_indices() to add boundary indices on levels.
    (Jiaqi Zhang, 2021/05/02)

  10. Changed: Improved: The internal indexing in FESystem::fill_fe_values() has been improved and is now about twice as fast.
    (David Wells, 2022/04/30)

  11. New: The class DataPostprocessors::BoundaryIds can be used to visualize the boundary indicators used in a mesh.
    (Wolfgang Bangerth, 2022/04/27)

  12. Improved: The step-44 tutorial now uses FE_DGP elements to discretize the pressure and dilatation fields, rather than FE_DGPMonomial. This improves the condition number of the linear system, necessitating fewer iterations to compute its iterative inverse.
    (Jean-Paul Pelteret, 2022/04/26)

  13. Fixed: The build process now works well with MPI option on Windows systems. If configured properly, this allows exploring the capability of distributed computing on Windows system with deal.II.
    (Daniel Sun, 2022/04/26)

  14. New: There is a new function DoFRenumbering::matrix_free_data_locality() that re-orders the indices of the unknowns in a DoFHandler to give optimal data locality of pre- and post-operations in matrix-free loops.
    (Martin Kronbichler, Peter Munch, Alexander Roschlaub, Dmytro Sashko, 2022/04/21)

  15. Added: New function enable_abort_on_exception.
    (Pasquale Claudio Africa, 2022/04/20)

  16. New: FEInterfaceValues now has functions get_*_function_values() to evaluate solution values across cell faces, such as the jump in (or average of) values, gradients, and Hessians.
    (Jiaqi Zhang, 2022/04/11)

  17. Fixed: Renumbering of the support points of FE_RaviartThomasNodal, which previously had x and z coordinates switched for the second component with dim=3.
    (Niklas Wik, 2022/04/04)

  18. Changed: The bundled version of muParser has been updated to version 2.3.3.
    (David Wells, 2022/03/28)

  19. Fixed: deal.II, compiled with the bundled copy of muParser, can now be safely linked alongside other projects which use a separate copy of muParser.
    (David Wells, 2022/03/25)

  20. Fixed: The functions in DoFRenumbering now work correctly (they do not deadlock) when a processor has zero locally owned degrees of freedom.
    (David Wells, 2022/03/23)

  21. Improved: MeshWorker::CopyData gains a new template argument for the scalar type, which allows the collected local matrix or vector contributions to be stored as types other than double.
    (Jean-Paul Pelteret, 2022/03/14)

  22. New: Add new wrappers ArborXWrappers::SphereIntersectPredicate and ArborXWrappers::SphereNearestPredicate to perform geometrical search on spheres using ArborX.
    (Bruno Turcksin, 2022/03/11)

  23. New: There is now a function ReferenceCell::contains_point() that provides the generalized equivalent of a function in the GeometryInfo class that will eventually be deprecated. type.
    (Wolfgang Bangerth, 2022/03/05)

  24. Improved: The Utilities::MPI::broadcast() has been optimized for types that are natively supported by MPI.
    (Francesco Andreuzzi, Wolfgang Bangerth, 2022/03/04)

  25. New: There is now a template variable Utilities::MPI::is_mpi_type that can be used to query whether a data type is a natively supported MPI type.
    (Wolfgang Bangerth, 2022/03/01)

  26. Changed: We now require the MPI library to support at least the MPI 3.0 standard.
    (Pengfei Jia, Matthias Maier, Timo Heister, 2022/02/28)

  27. New: The template variable Utilities::MPI::mpi_type_id_for_type provides information about the MPI_Datatype that corresponds to a given type. This information is necessary in all MPI calls where one needs to provide an MPI_Datatype to identify the kinds of objects stored in a send or receive buffer.
    (Wolfgang Bangerth, 2022/02/27)

  28. Fixed: We can now load certain msh files generated by Gmsh that contain some cells with inverted volume.
    (Sean Ingimarson, Timo Heister, 2022/02/26)

  29. Improved: The asserts in parallel::distributed::SolutionTransfer have been improved. Before this change, an assert has been only called in the parallel case for distributed vectors for cells neighboring interprocess boundaries. Now, we check for all DoFs that the contributions from all cells have the same values. This condition might be not given if the solution has high gradients at hanging nodes.
    (Peter Munch, Magdalena Schreter, 2022/02/25)

  30. Fixed: ParsedConvergenceTable::difference() was not working. This is now fixed.
    (Luca Heltai, 2022/02/22)

  31. Fixed: The function parallel::distributed::copy_triangulation() used to copy the reference to the attached SolutionTransfer instance, which resulted in calling the same SolutionTransfer instance multiple times if the triangulation and the new triangulation are refined/coarsened independently. This has been fixed.
    (Ivo Dravins, Peter Munch, 2022/02/22)

  32. Fixed: GridOut::write_msh() with gmsh api now works also with non standard combinations of boundary and manifold ids.
    (Luca Heltai, 2022/02/20)

  33. Fixed: NonMatching::create_coupling_sparsity_pattern() and NonMatching::create_coupling_mass_matrix() were not working properly with hanging node constraints.
    (Luca Heltai, 2022/02/18)

  34. New: AffineConstraints::add_entries_local_to_global() now supports also the case where columns don't use the same constraints as rows.
    (Luca Heltai, 2022/02/18)

  35. Added: The DEAL_II_QUERY_GIT_INFORMATION() CMake macro now also records the date and time of the commit in the CMake variable GIT_TIMESTAMP.
    (Matthias Maier, 2022/02/06)

  36. New: GridGenerator::generate_from_name_and_argument() now supports also GridGenerator::hyper_ball_balanced()
    (Luca Heltai, 2022/02/17)

  37. Fixed: Patterns::Tools::Convert now supports also std::array types.
    (Luca Heltai, 2022/02/16)

  38. New: Add new DistributedTree class based on ArborX. This class performs multiple kinds of geometric search on distributed objects: intersection of bounding boxes and intersection of bounding boxes with points. This class is the distributed equivalent of ArborXWrappers::BVH.
    (Bruno Turcksin, 2022/02/15)

  39. Fixed: BlockDynamicSparsityPattern in step-55 will be initialized with the correct IndexSet for locally relevant dofs.
    (Marc Fehling, 2022/02/10)

  40. Fixed: GridGenerator::hyper_ball_balanced() now works correctly with nonzero centers.
    (David Wells, 2022/02/09)

  41. New: The FE_Nothing class can now return its (empty) constant modes. DoFTools::extract_constant_modes can now work with DoFHandlers that contain FE_Nothing.
    (Sebastian Proell, 2022/02/07)

  42. Improved: To be consistent with the FEValues classes, one can now loop over all DoF and quadrature-point indices in a range-based loop with the help of the new functions dof_indices() and quadrature_point_indices().
    (Peter Munch, 2022/02/06)

  43. Changed: The minimum CMake version required to configure and build deal.II has been raised to 3.3.0.
    (Matthias Maier, 2022/02/06)

  44. Improved: In MatrixFreeTools::compute_diagonal a template argument VectorType has been introduced to be applicable for arbitrary vector types.
    (Magdalena Schreter, Peter Munch, 2022/02/02)

  45. New: Added support for hessians to FE_NedelecSZ
    (Sebastian Kinnewig, 2022/01/28)

  46. Fixed: MappingQ::transform_points_real_to_unit_cell has been made more robust, correcting an unsuitable initial guess for the Newton iteration to compute the inverse transformation in the case of bilinear/trilinear cells.
    (Bruno Blais, Martin Kronbichler, 2022/01/22)

  47. New: MeshWorker::ScratchData and MeshWorker::CopyData now have hp-capabilities. Creation of a ScratchData object with an hp::FECollection and hp::QCollection enables those capabilities. A few specialized MeshWorker::ScratchData::reinit() were added to support face integration where the integration rule and mapping differs on either side of an interface. Additionally, some MeshWorker::CopyData::reinit() methods have been implemented to cater for the change in number of degrees-of-freedom between cells iterated over during assembly.
    (Jean-Paul Pelteret, 2022/01/09)

  48. Fixed: GridGenerator::convert_hypercube_to_simplex_mesh() now correctly copies line manifold ids in 3D.
    (David Wells, 2022/01/07)

  49. Changed: The step-43 tutorial program used Amg_preconditioner and Mp_preconditioner to denote the preconditioners for the top left and bottom right block of a mixed-Laplace matrix. These variable names were taken from step-31/step-32 where they made sense, but in the current context, the names do not reflect what is in the corresponding matrix blocks. They have been renamed to top_left_preconditioner and bottom_right_preconditioner, respectively.
    (Wolfgang Bangerth, 2022/01/07)

  50. Fixed: Serializing empty sparsity patterns ran into a segmentation fault. This is now fixed.
    (Lucas Myers, Wolfgang Bangerth, 2022/01/05)

  51. Fixed: deal.II has been updated to support the new OneAPI api interface introduced by the Intel Threading Building Blocks Library. Note that the threading support of the matrix-free backend will be disabled in this case: For now, the MatrixFree::AdditionalData::tasks_parallel_scheme control has no effect and simply defaults to the serial loop.
    (Wolfgang Bangerth, Matthias Maier, 2022/01/06)

  52. Fixed: The FEInterfaceValues class is now made compatible with filtered iterators. Specifically, it is now possible to call FEInterfaceValues::reinit() with the active cell being a filtered iterator and the type returned by cell->neighbor() being some other non-filtered iterator type.
    (Simon Sticko, Jean-Paul Pelteret, 2022/01/05)

  53. Improved: The class PreconditionRelaxation has been refactored. Similarly to PreconditionChebyshev, it can be now used stand-alone and users can attach their own preconditioners. Furthermore, users can specify multiple iteration steps, which is particularly useful if PreconditionRelaxation is used as smoother in a multigrid algorithm.
    (Peter Munch, 2022/01/01)

  54. New: More "getter" functions have been added to MeshWorker::ScratchData, facilitating the construction of other FEValues-type objects with some of the same objects as those used in a ScratchData instance.
    (Jean-Paul Pelteret, 2021/12/31)

  55. Fixed: The SparseDirectUMFPACK class had a bug where it did not correctly set up the data passed to the UMFPACK package if (i) the matrix is a block matrix, and (ii) the elements are complex-valued. This is now fixed.
    (Kuljit S. Virk, 2021/12/30)

  56. New: The IteratorFilters::BoundaryIdEqualTo and IteratorFilters::ManifoldIdEqualTo filtered iterator predicates have been added to help select cell faces that have, respectively, one or more specified boundary ID(s) and manifold ID(s).
    (Jean-Paul Pelteret, 2021/12/30)

  57. Fixed: GridGenerator::flatten_triangulation() did not copy information on the manifold ids of interior faces and edges. This is now fixed.
    (Wolfgang Bangerth, 2021/12/27)

  58. Fixed: GridGenerator::flatten_triangulation() did not copy information of edges of three-dimensional triangulations, even though it copied information from faces. This is now fixed.
    (Wolfgang Bangerth, 2021/12/26)

  59. Changed: The step-60 tutorial now works also with non-zero Dirichlet BCs and non-zero right-hand side
    (Marco Feder, 2021/12/24)

  60. New: Build also a docker image with root user, to be used in github actions directly.
    (Luca Heltai, 2021/12/24)

  61. Deprecated: The Threads::new_thread() functions have been deprecated. These functions return a Threads::Thread object, which is a class that had previously been deprecated, and should no longer be used: use the std::thread or std::jthread classes and related functionality instead.
    (Wolfgang Bangerth, 2021/12/23)

  62. Fixed: The matrix-free FEFaceEvaluation would produce wrong results in case hanging nodes were present on faces with non-standard orientation. This is now fixed.
    (Martin Kronbichler, 2021/12/20)

  63. Fixed: IndexSet::subtract_set would previously run in quadratic complexity in the number of ranges. This is now fixed.
    (Martin Kronbichler, 2021/12/13)

  64. New: GridOut::write_msh() now also works for triangular and tetrahedral meshes.
    (Wolfgang Bangerth, 2021/12/06)

  65. New: The DataPostprocessorInputs::CommonInputs class now also stores the face number that is currently being worked on when using a DataOutFaces object. This allows accessing face-related data like the boundary id from a DataPostprocessor.
    (Wolfgang Bangerth, 2021/12/03)

  66. Changed: The function TrilinosWrappers::Vector::is_non_negative() now also works in parallel.
    (Justin O'Connor, 2021/12/02)

  67. Fixed: MatrixFreeOperators::CellwiseInverseMassMatrix with template argument -1 for the polynomial degree now chooses the pre-compiled code with fast tensor-product algorithms if available, rather than a slower general-purpose code.
    (Martin Kronbichler, 2021/11/29)

  68. New: Namespace Physics::VectorRelations features functions to compute angles between (spatial) vectors.
    (Marc Fehling, 2021/11/29)

  69. New: The SD::BatchOptimizer::copy_from() function can be used to duplicate the data from one batch optimizer instance to another.
    (Jean-Paul Pelteret, 2021/11/27)

  70. Fixed: The step-22 tutorial now uses FEValuesExtractors also for the right-hand side
    (Marco Feder, 2021/11/23)

  71. New: Utilities::MPI::create_mpi_data_type_n_bytes() helps with large MPI communication and IO.
    (Timo Heister, 2021/11/17)

  72. New: In the same spirit as provided by the C++20 range adaptors feature, it is now possible to "filter" ranges over the active cells of Triangulation and DoFHandler objects.
    This allows to replace:

    DoFHandler<dim> dof_handler;
    for (const auto &cell : dof_handler.active_cell_iterators())
    if (cell->is_locally_owned())
    fe_values.reinit (cell); the local integration on 'cell'...;
    void reinit(const Triangulation< dim, spacedim > &tria)
    IteratorRange< active_cell_iterator > active_cell_iterators() const


    DoFHandler<dim> dof_handler;
    for (const auto &cell :
    fe_values.reinit (cell); the local integration on 'cell'...;

    Here, the operator| is to be interpreted in the same way as is done in the range adaptors feature that is part of C++20. It has the same meaning as the | symbol on the command line: It takes what is on its left as its inputs, and filters and transforms to produce some output. In the example above, it "filters" all of the active cell iterators and removes those that do not satisfy the predicate – that is, it produces a range of iterators that only contains those cells that are both active and locally owned.
    (Wolfgang Bangerth, 2021/11/03)

  73. New: The GridOut::write_gnuplot() function can now deal with tetrahedral meshes, as well as with wedges and pyramids.
    (Wolfgang Bangerth, 2021/11/15)

  74. New: The GridIn::read_comsol_mphtxt() function adds the ability to read .mphtxt files generated by COMSOL.
    (Wolfgang Bangerth, 2021/11/15)

  75. New: The new class Functions::SignedDistance::Ellipsoid computes a signed distance function for ellipsoids. It is implemented for 1D and 2D ellipsoids (ellipses).
    (Nils Much, 2021/11/05)

  76. New: One can now output simplex, pyramid, and wedge cells in GNUPLOT format via DataOut and other classes that are based on DataOutBase and call DataOutBase::write_gnuplot().
    (Wolfgang Bangerth, 2021/11/03)

  77. Changed: The DataOutBase::write_deal_II_intermediate() function had a bug when outputting data that corresponds to cells other than hypercubes. This is now fixed. In the process, the format version for this function's output has been incremented from 3 to 4, which also includes changing the order in which vertices are outputted.
    The DataOutReader class was also updated to match the new output format.
    (Wolfgang Bangerth, 2021/10/25)

  78. New: The ReferenceCell class now has input and output operators (>> and <<).
    (Wolfgang Bangerth, 2021/10/25)

  79. Fixed: GridIn::read_unv() can now ignore any sections before 2411. In other words, the unv file can start with any number sections (before 2411) as long as they have a proper format (must begin and end with a "-1").
    (Vachan Potluri, 2021/10/22)

  80. New: There is now a function ReferenceCell::vertex() that returns the coordinates of a vertex of a reference cell.
    (Wolfgang Bangerth, 2021/10/13)

  81. Fixed: The DataOutFaces class can now also be used with simplex meshes. It previously aborted with exceptions.
    (Wolfgang Bangerth, 2021/10/13)

  82. Improved: PreconditionChebyshev can now run the vector updates on a part of the vector for better data locality. The prerequisite is an associated MatrixType class to provide a vmult class with two std::function objects to specify the operation before and after the matrix-vector product.
    (Martin Kronbichler, 2021/10/12)

  83. Fixed: The KellyErrorEstimator for 1D can now handle the hp case.
    (Fabian Castelli, 2021/10/12)

  84. Fixed: Deleted duplicate occurrence of VectorTools::interpolate_boundary_values() in setup_dofs() of step-45.
    (Raghunandan Pratoori, 2021/10/11)

  85. Changed: The 3D implementation of GridTools::rotate() now accepts unit vectors as rotation axes.
    (Marc Fehling, 2021/10/11)

  86. New: VectorizedArray can now be constructed from an initializer list.
    (Peter Munch, 2021/10/08)

  87. Fixed: The FEPointEvaluation class always evaluated a solution starting at the component 0 of the selected vector-valued base element, even if a different component was selected by the caller. This has been fixed to correctly start evaluating at the selected component.
    (Rene Gassmoeller, 2021/10/01)

  88. Fixed: The payload for Trilinos-based linear operators is now able to use its own type as an examplar operator. This means that one can now use linear operators based on Trilinos matrices as exampler operators when initializing a LinearOperator for a Trilinos preconditioner.
    (Wenyu Lei, Jean-Paul Pelteret, 2021/09/12)

  89. Fixed: ParticleHandler::insert_particles() runs into a deadlock in parallel when a processor is not inserting any particles. This is now fixed.
    (Sebastian Fuchs, 2021/08/31)

  90. Fixed: There was a bug in the default constructor of TimeStepping::EmbeddedExplicitRungeKutta(). Some parameters were left uninitialized which triggered a segmentation fault when calling evolve_one_time_step().
    (Praveen Chandrashekar, Bruno Turcksin, 2021/08/30)

  91. Changed: For 1d cells, the following code did not compile:

    for (const auto &face : cell->face_iterators())
    if (face->at_boundary())

    That was because, unlike in the 2d and 3d case, the class that describes the faces of 1d cells had a set_boundary_id() function that was not marked as const. This has now been fixed by adding the const specification for this function, along with the one on the set_all_boundary_ids() function.
    (Wolfgang Bangerth, Tyler Anderson, 2021/08/21)

  92. New: The hp::FECollection::hp_vertex_dof_identities() function (and related functions for lines and quads) allows for the computation of multiway computations of DoF indentities in the hp-context.
    (Wolfgang Bangerth, 2021/08/20)

  93. New: There is now an overload for Particles::ParticleAccessor::set_properties() that takes a Tensor as argument. It is used in step-19.
    (Wolfgang Bangerth, 2021/08/19)

  94. Fixed: Exceptions thrown via AssertThrow did not have a stacktrace attached to them, unlike exceptions thrown with Assert. This is now fixed.
    (Wolfgang Bangerth, Paras Kumar, 2021/08/20)

  95. Fixed: It was not possible to default-construct objects of type DoFHandler::face_iterator for dim==1. As a consequence, one could also not call cell->face_iterators() for the cells of 1d DoFHandler objects. This is now fixed.
    (Wolfgang Bangerth, Tyler Anderson, 2021/08/19)

  96. New: MeshWorker::ScratchData is now able to evaluate the jumps in, and averages of, finite element functions and their derivatives across an interface.
    (Jean-Paul Pelteret, 2021/08/14)

  97. New: FEInterfaceViews now has functions get_*_function_values() and get_*_function_values_from_local_dof_values() to evaluate solution values across cell faces, such as the jump in (or average of) values, gradients, and Hessians. With these functions, FEInterfaceValues can now be used with automatic differentiation.
    (Jiaqi Zhang, 2021/08/11)

  98. New: The functions FEFaceValuesBase::face_number(), FEInterfaceValues::get_cell() and FEInterfaceValues::face_number() have been implemented to provide introspection as to which entities these objects have been initialized for.
    (Jean-Paul Pelteret, 2021/08/06)

  99. New: The function FEInterfaceValues::dof_indices() has been added to make writing range-based for loops over all local interface degrees of freedom much simpler.
    (Jean-Paul Pelteret, 2021/08/05)

  100. New: Diagnostic functions AffineConstraints::n_identities() and AffineConstraints::n_inhomogeneities() to analyse your constraints.
    (Marc Fehling, 2021/08/04)

  101. Fixed: Use tolerances consistently in RemotePointEvaluation, VectorTools::point_values(), and GridTools::distributed_compute_point_locations().
    (Peter Munch, Magdalena Schreter, 2021/07/31)

  102. New: QWitherdenVincentSimplex now implements even-order rules in addition to the standard Gauss-like odd-order rules.
    (David Wells, 2021/07/29)

  103. Fixed: AlignedVector objects for which AlignedVector::replicate_across_communicator() had previously been called, could not be move-constructed or move-assigned: The resulting object was usable, but destruction of the target object led to memory errors that aborted the program. This is now fixed. As a side effect of the patch, memory consumption of AlignedVector objects has also been shrunk for the most common case where AlignedVector::replicate_across_communicator() has not actually been called.
    (Wolfgang Bangerth, 2021/07/28)

  104. New: The new function TableBase::clear() allows to empty a Table object.
    (Jiaqi Zhang, 2021/07/21)

  105. Fixed: MappingFEField now works with simplex faces correctly.
    (David Wells, 2021/07/19)

  106. New: Added a class QIteratedSimplex for building composite simplex quadrature rules.
    (David Wells, 2021/07/15)

  107. New: ParticleAccessor::get_local_index() is a new function returning particle indices local to each MPI process, which is useful for direct access into arrays with temporary results associated to particles.
    (Martin Kronbichler, 2021/07/06)

  108. Fixed: step-34 had a minor memory leak whereby the function get_singular_quadrature() allocated some memory the first time it was called, and then never released it. This is now fixed. Separately, this function was not thread-safe, and now it is.
    (Wolfgang Bangerth, 2021/07/03)

  109. New: When calling cell->vertex_dof_index(...) on a DoFHandler that uses the hp capability, then one needed to provide the last argument indicating what the active fe index is. But for cells, there can only be one active fe index, so leaving the last argument at its default value is now interpreted as using the active fe index of the cell. The argument still needs to be provided if the vertex_dof_index() function is called on faces, edges, or vertices.
    (Wolfgang Bangerth, 2021/07/02)

  110. New: The new function Triangulation::n_global_coarse_cells() allows to query the global number of coarse cells. This function is particularly useful for parallel::fullydistributed::Triangulation, since in this case the coarse cells might differ on each process.
    (Peter Munch, 2021/06/26)

  111. New: The new unified function MGTwoLevelTransfer::reinit() selects automatically if MGTwoLevelTransfer::reinit_geometric_transfer() or MGTwoLevelTransfer::reinit_polynomial_transfer() is needed.
    (Peter Munch, 2021/06/23)

  112. New: The new function CellAccessor::is_ghost_on_level() allows to check if a cell is a ghost cell on a multigrid level.
    (Peter Munch, 2021/06/23)

  113. Improved: i) Standardized the format of the references in the file doc/doxygen/references.bib using in particular the convention "M. Mo" or "R.B. Kellogg" for names and the full name for journals. ii) Fixed the display of some titles (e.g., upper case for proper noun). iii) Added an URL entry for the references with a DOI.
    (Diane Guignard, 2021/06/22)

  114. Fixed: The Gauss-like rules implemented by QGaussSimplex now integrate 2 * n_points_1D - 1 degree polynomials exactly.
    (David Wells, 2021/06/22)

  115. Fixed: The step-21 program called a function that was disabled on Microsoft Windows to work around a bug in the Microsoft Visual Studio compiler. This is now fixed and the program should run successfully again on Windows.
    (Wolfgang Bangerth, 2021/06/08)

  116. Fixed: MappingQCache::get_vertices() would previously always return the actual vertices, ignoring a possible displacement that is possible with this class. This is now fixed.
    (Martin Kronbichler, 2021/06/07)

  117. Improved: An assert has been added in the constructor of FEPointEvaluation to prevent accessing non-existing components.
    (Judith Pauen, 2021/06/05)

  118. New: The new function Utilities::MPI::reduce() allows to reduce arbitrary types with a user-specified binary operation.
    (Peter Munch, 2021/05/27)

  119. New: Fixed a bug in clear user data for standard Triangulation.
    (Nicola Giuliani, 2021/05/25)

  120. Addes conformity tests for several vector elements in 2D/3D as well as minor fixes for 2d Nedelec and Raviart-Thomas to make them work on non-standard meshes.
    (Konrad Simon, 2021/05/08)

  121. New: The class MGTransferMatrixFree::build() now also accepts an optional function for initializing the internal level vectors. This is useful if one uses the transfer operators in the context of smoothers that are built around MatrixFree objects.
    (Peter Munch, 2021/03/09)

  122. New: You can now perform range-based iterations on hp::FECollecion, hp::QCollection, and hp::MappingCollection objects.
    (Peter Munch, 2021/10/27)