Reference documentation for deal.II version GIT relicensing864g63b55056d6 20240620 03:10:02+00:00

This is the list of changes made between the release of deal.II version 8.5.0 and that of 9.0.0. All entries are signed with the names of the author.
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.
New: Every function in the GridGenerator namespace now attaches a default manifold to the curved parts of the boundary of the domain, and sets reasonable defaults for manifold indicators in the domain, independently on the boundary indicators.
(Luca Heltai, Daniel Arndt, JeanPaul Pelteret, David Wells, 2018/04/17)
New: The implementation of the divergence in FEValuesExtractors::Tensor was changed so that the gradient operator is contracted from the right. This is done in order to make it consistent with gradient of the second order tensor, namely \(Grad(T) : I = Div(T)\).
(Denis Davydov, 2018/04/12)
New: The Manifold class now requires the implementation of a Manifold::clone() method in all derived classes.
(Luca Heltai, 2018/04/08)
New: The class FE_DGQHermite has been switched to the more stable polynomial basis Polynomials::HermiteLikeInterpolation. This changes the representation of the numbers by the FE_DGQHermite class. The change is yet highly beneficial because it significantly improves the accuracy (in terms of roundoff) for this basis and also reduces iteration counts for some iterative solvers with simple preconditioner ingredients such as pointJacobi or pointJacobi smoothing.
(Martin Kronbichler, 2018/02/09)
Changed: FETools::get_fe_by_name returns a std::unique_ptr instead of a owning raw pointer to prevent creating memory leaks.
(Daniel Arndt, 2018/02/09)
Changed: The function Mapping::clone (and all inheriting classes' implementations, such as MappingQ1Eulerian::clone) now return a unique_ptr<Mapping<dim,spacedim>>
.
(David Wells, 2018/02/08)
Changed: The GridRefinement functions only allow Vector objects to be passed in as criteria.
(Daniel Arndt, 2018/02/04)
Changed: TableIndices uses std::size_t for index access, LAPACKFullMatrix::size_type is types::blas_int and FullMatrix::size_type is std::size_t.
(Daniel Arndt, 2018/02/02)
Changed: The classes OpenCASCADE::NormalProjectionBoundary, OpenCASCADE::DirectionalProjectionBoundary and OpenCASCADE::NormalToMeshProjectionBoundary are rebased from Boundary to FlatManifold.
(Daniel Arndt, 2018/01/05)
Changed: It is now necessary to explicitly enable MPI support when configuring deal.II via DDEAL_II_WITH_MPI=ON, or DWITH_MPI=ON on the command line. The old behavior of just specifying an MPI compiler wrapper via environment, or CMAKE_(CCXX)_COMPILER does not work any more.
(Matthias Maier, 2018/01/01)
Changed: The type LAPACKFullMatrix::size_type
is now unsigned int
, which matches FullMatrix::size_type
.
(David Wells, 2017/12/20)
Changed: Change the logic of DoFTools::get_subdomain_association to have the degrees of freedom along a refinement edge be now all given to the processor with the smallest subdomain_id.
(Alexander Grayver, 2017/12/11)
Changed: The parameters contained in struct SolverQMRS::AdditionalData have been changed in order to meet the rewritten algorithm. See the class documentation for more details.
(Ingo Kligge, 2017/12/06)
Changed: The default partitioner for parallel::shared::Triangulation is Zoltan.
(Daniel Arndt, 2017/11/28)
Changed: The minimal supported Trilinos version is 12.4.
(Daniel Arndt, 2017/11/18)
Removed: The default constructor for ArrayView was removed since an object constructed in this way can't be used in any sensible way.
(Daniel Arndt, 2017/11/17)
Changed: The many KellyErrorEstimator::estimate() overloads all have an argument that represents the Neumann boundary values (if the solution of the PDE satisfies Neumann boundary conditions on parts of the boundary). They used to just be of type Function<dim>, but this has now been changed to Function<dim,number> (or, more precisely, FunctionMap<dim,number>, which stores a map to Function<dim,number> objects) where number
is the underlying data type of the input vector. In other words, if you are using a regular Vector<double> (or PETSc or Trilinos equivalents), then nothing has changed since number
is double
, which is the default for the Function class. On the other hand, if you are using a Vector<std::complex<double>>, for example, then you need to pass a Function<dim,std::complex<double>> – which of course makes sense because the Neumann boundary values of a complexvalued solution should also be complexvalued.
(Wolfgang Bangerth, 2017/11/17)
Changed: The distribution of degrees of freedom along a refinement edge for a parallel::shared::Triangulation has been changed to mimic that of a parallel::distributed::Triangulation, that is, they are now all given to the processor with the smallest subdomain_id.
(Conrad Clevenger, 2017/11/13)
Changed: ParameterHandler::print_parameters() in LaTeX format now correctly escapes special characters. As a consequence, the names of labels changed because they are now mangled. See the documentation for more details.
(Timo Heister, 2017/11/03)
Removed: The BZIP2 dependency has been removed as it was not used in the library anymore.
(Daniel Arndt, 2017/10/25)
Changed: The direction argument of the DoFRenumbering::downstream() and related functions have been changed from a Point<dim> to a Tensor<1,dim> type, since that is the correct data type for a direction vector.
(Wolfgang Bangerth, 2017/10/23)
Changed: LAPACKSupport::Properties is renamed to LAPACKSupport::Property to be consistent with LAPACKSupport::State.
(Denis Davydov, 2017/08/24)
Changed: The virtual function SolverRichardson::criterion() now receives the residual and preconditioned residual vectors as arguments, rather than accessing it through the member variables of the class. It has also been made const
.
(Wolfgang Bangerth, 2017/09/12)
Changed: The methods
have had their declarations changed in an incompatible manner: these three methods now take ArrayView arguments instead of std::vector
s. This change was done for performance reasons: before this change about a third of the time spent generating a curved grid (in this particular benchmark, a circle described with polar coordinates and a transfinite interpolation) was used in allocating and freeing memory used by the std::vector
arguments to Manifold::get_new_point() even though the sizes of the arrays are usually known at compilation time. This change completely eliminates this allocation cost. The method Manifold::add_new_points()
has been removed in favor of Manifold::get_new_points(), which also uses ArrayView arguments instead of std::vector
s.
(David Wells, 2017/09/10)
Changed: deal.II now requires BOOST version 1.59 or newer.
(David Wells, 2017/09/04)
Changed: LAPACKSupport::upper_triangle and LAPACKSupport::lower_triangle are renamed to LAPACKSupport::upper_triangular and LAPACKSupport::lower_triangular, respectively.
(Denis Davydov, 2017/08/24)
Changed: SolverGMRES used to notify via deallog when reorthogonalization of the Arnoldi vectors kicks in. Now, there is a slot one can connect to retrieve this information. By default nothing is printed.
(Daniel Arndt, 2017/09/02)
Fixed: FiniteElement::get_generalized_support_points() now always returns a list of unique points. This is in contrast to the old behavior where get_generalized_support_points() returned a repeated list of (nodal) support points for an FESystem consisting of Lagrangian elements.
(Matthias Maier, 2017/08/31)
Changed: Specialization of the ProductType class are now implemented through specialization of the class internal::ProductTypeImpl . This was done in order to ensure that product operations performed with qualified number types do not result in the intended specializations being overlooked by the compiler.
(JeanPaul Pelteret, Wolfgang Bangerth, 2017/08/24)
Changed: The VectorMemory::Pointer class used to have an automatic conversion operator to the underlying pointertovector object. As part of a rewrite of this class, this conversion (rarely used in practice) has been removed.
(Wolfgang Bangerth, 2017/08/22)
Changed: The constructors SymmetricTensor (const Tensor &) and Tensor (const array_type &) have been marked as 'explicit'.
(Daniel Arndt, 2017/08/20)
Changed: The configuration file deal.II/base/config.h
has been thoroughly cleaned up. As a result, the following preprocessor symbols (which were either never used or were workarounds for older, now unsupported compilers) are no longer defined:
DEAL_II_EXPLICIT_CONSTRUCTOR_BUG
DEAL_II_MEMBER_VAR_SPECIALIZATION_BUG
DEAL_II_BOOST_BIND_COMPILER_BUG
DEAL_II_HAVE_ISNAN
DEAL_II_STD_ISNAN
DEAL_II_HAVE_UNDERSCORE_ISNAN
DEAL_II_HAVE_ISFINITE
DEAL_II_HAVE_SYS_TIMES_H
DEAL_II_HAVE_TIMES
DEAL_II_HAVE_SYS_TYPES_H
DEAL_II_HAVE_SYS_TIMES_H
(Matthias Maier and David Wells, 2017/08/08  2017/08/24)
Changed: The member function Boundary::project_to_surface is now ignored in the function GridTools::fix_up_distorted_child_cells. Instead, GridTools::fix_up_distorted_child_cells uses an internal function that duplicates the algorithm used by StraightBoundary::project_to_surface, which was the only place project_to_surface was implemented in the library.
(Luca Heltai, David Wells, 2017/08/12)
Changed: Triangulation::get_manifold() will now return a FlatManifold instead of a StraightBoundary in the case that no manifold description has been attached to a provided manifold ID.
(Luca Heltai, David Wells, 2017/08/12)
Removed: Some old functionality has been removed from LogStream:
test_mode()
, threshold_double()
, and threshold_float()
. Use numdiff, or other tools that can do output comparison with small numerical differences instead. log_cerr()
function has been removed. timestamp()
, log_execution_time()
and log_time_differences()
. Use the Timer, TimerOutput and TimerOutput::Scope classes instead.
(Matthias Maier, 2017/08/10)
Changed: DataOutBase (and derived classes) no longer sets floating point output precision of ostreams.
(Matthias Maier, 2017/08/09)
Changed: The types::boundary_id and types::material_id types have been changed from unsigned char to unsigned int.
(JeanPaul Pelteret, 2017/08/07)
Deprecated: A number of classes have been deprecated in favor of the LinearOperator concept:
Use the LinearOperator class instead, see the module on linear operators for more details.
(Matthias Maier, 2017/07/14)
Changed: In 1D, MappingQGeneric now considers a cell to have multiple manifolds if the manifold attached to the cell does not match one of the manifolds attached to a face (i.e., vertex).
(David Wells, 2017/07/10)
Changed: The SynchronousIteartors::iterators member variable has been made private, as mentioned in the changelog of the previous release. It can be accessed via SynchronousIterators::operator*(), however.
(Wolfgang Bangerth, 2017/07/11)
Changed: The TableIndices class had a single constructor that just took as many indices as one wanted, and padded any unassigned indices with numbers::invalid_unsigned_int. This invited mistakes and indeed led to difficult to track down errors.
This constructor is therefore now deprecated, and has been replaced by a series of constructors with between one and five arguments that can be used for objects of type TableIndices<1> to TableIndices<5>.
(Wolfgang Bangerth, 2017/07/10)
Changed: In 1D, the face values calculated by MappingManifold
now use the manifold attached to the relevant vertex instead of the manifold on the current face.
(David Wells, 2017/07/08)
Changed: Timer::print_data doesn't restrict its output to the first MPI process anymore but leaves such a choice to the stream given.
(Daniel Arndt, 2017/06/27)
Changed: The restart_parameter member variable of TrilinosWrappers::SolverGMRES::AdditionalData has been removed. This value is now stored as gmres_restart_parameter in the base class TrilinosWrappers::SolverBase::AdditionalData.
(JeanPaul Pelteret, 2017/06/26)
Changed: The deprecated data in SolverCG::AdditionalData and SolverGMRES::AdditionalData::compute_eigenvalues have been removed. Use the respective connect_* member functions instead.
(Daniel Arndt, 2017/06/12)
Changed: The deprecated member functions add(), normalize(), conjugate(), abs(), sadd(), equ() and mult() in the vector classes have been removed.
(Daniel Arndt, 2017/06/12)
Deprecated: The ParameterHandler::print_parameters_section() function has been deprecated.
(Wolfgang Bangerth, 2017/06/07)
Changed: Versions of PETSc prior to 3.3.0 are no longer supported.
(David Wells, 2017/06/03)
Changed: The FiniteElement::clone() function and all of its implementations in derived classes now return a std::unique_ptr rather than a plain pointer to a finite element object. Userimplemented finite element classes need to be adjusted accordingly.
(Wolfgang Bangerth, 2017/06/01)
Changed: The deprecated function SparsityTools::reorder_Cuthill_McKee() acting on a SparsityPattern object has been removed. Use the one acting on a DynamicSparsityPattern instead.
(Daniel Arndt, 2017/05/15)
Changed: The deprecated functions BlockSparseMatrixEZ::n_rows() and BlockSparseMatrixEZ::n_cols() have been removed. Use BlockSparseMatrixEZ::m() and BlockSparseMatrixEZ::n() instead.
(Daniel Arndt, 2017/05/14)
Changed: The deprecated function Utilities::System::job_supports_mpi() has been removed. Use Utilities::MPI::job_supports_mpi() instead.
(Daniel Arndt, 2017/05/14)
Removed: The deprecated serial Trilinos vector classes have been removed.
(Daniel Arndt, 2017/05/08)
Removed: The deprecated serial PETSc vector classes have been removed.
(David Wells, 2017/05/06)
Changed: The deprecated member functions in the classes TrilinosWrappers::BlockVector, TrilinosWrappers::MPI::BlockVector, TrilinosWrappers::Vector and TrilinosWrappers::MPI::Vector have been removed.
(Daniel Arndt, 2017/05/05)
Changed: The deprecated version of Manifold::get_new_point() that took an argument of type Quadrature has been removed. Use the other variant of that function instead. The same has been done to all implementations of that interface in derived classes.
(Wolfgang Bangerth, 2017/05/05)
Changed: The deprecated typedef FEFieldFunction::ExcPointNotAvailableHere has been removed. Use VectorTools::ExcPointNotAvailableHere instead.
(Wolfgang Bangerth, 2017/05/04)
Changed: The deprecated variant of GridTools::get_face_connectivity_of_cells() has been removed. Use the other variant instead.
(Wolfgang Bangerth, 2017/05/04)
Changed: The deprecated constructors of MappingQEulerian and MappingQ1Eulerian have been removed. Use the other constructor of each class instead.
(Wolfgang Bangerth, 2017/05/04)
Changed: The deprecated function FEValuesBase::transform() has been removed. It only forwarded the call to the Mapping object used by the FEValuesBase object. Use the corresponding function of the mapping instead.
(Wolfgang Bangerth, 2017/05/04)
Changed: The deprecated member variables supports_distributed_data
that was present in all vector classes has been removed. If you needed this functionality, use the type trait is_serial_vector instead.
(Wolfgang Bangerth, 2017/04/30)
Changed: The deprecated functions DoFHandler::get_tria() and hp::DoFHandler::get_tria() have been removed. Use DoFHandler::get_triangulation() and hp::DoFHandler::get_triangulation() instead.
(Wolfgang Bangerth, 2017/04/25)
Changed: The deprecated functions DataOutInterface::write_pvd_record() and DataOutBase::write_visit_record() have been removed. Use the corresponding functions in namespace DataOutBase.
(Wolfgang Bangerth, 2017/04/25)
Changed: The deprecated function FEValuesBase::get_normal_vectors() that returned a vector of Point objects has been removed. Its replacement, FEValuesBase::get_all_normal_vectors() has now itself been deprecated, and we have created a new function FEValuesBase::get_normal_vectors() that returns a vector of Tensor<1,dim> objects. The net effect is that the function with the old name has simply gotten a new return type.
(Wolfgang Bangerth, 2017/04/25)
Removed: DataPostprocessor had member functions compute_derived_quantities_*()
that had previously already been deprecated. These have now been removed.
(Wolfgang Bangerth, 2017/04/24)
Removed: The ParameterHandler::read_input() function and friends have been removed. They were already deprecated in the previous release.
(Wolfgang Bangerth, 2017/04/21)
Changed: It was previously allowed to copy one FESystem object to another via the copy constructor. There are probably few reasons to do so, but they complicated the design of data structures. Consequently, it is now no longer allowed to copy such objects.
(Wolfgang Bangerth, 2017/04/17)
Removed: The FETools::compute_node_matrix() variant that takes two arguments has been removed. It was previously already deprecated. Use the variant with just one argument instead.
(Wolfgang Bangerth, 2017/04/16)
Changed: The AssertGlobalIndexRange
macro has been removed: the expansion of this macro involved an undeclared template and always generates compiler errors if used.
(David Wells, 2017/04/13)
Removed: A number of deprecated classes have been removed in favor of the new LinearOperator concept:
Use the LinearOperator class instead, see the module on linear operators for more details.
(Matthias Maier, 2017/04/06  2018/05/03)
Removed: The FiniteElement::interpolate() function and all of its implementations in derived classes has been removed. It was previously already deprecated. Use FiniteElement::convert_generalized_support_point_values_to_nodal_values() instead.
(Wolfgang Bangerth, 2017/04/05)
Removed: The deprecated MPI communicator and the constructor which used it were removed from the MatrixFree::AdditionalData class.
(Denis Davydov, 2017/03/26)
Removed: The deprecated CMake flag DEAL_II_CXX11_FLAG
has been removed.
(David Wells, 2017/03/23)
Changed: deal.II now requires a compiler supporting (very nearly) the entire C++11 standard. The minimal version of GCC supported is now 4.8.
(David Wells, 2017/03/23)
New: The tutorial step59 shows a matrixfree solver for the Poisson equation discretized with the symmetric interior penalty discontinuous Galerkin method.
(Katharina Kormann, Martin Kronbichler, 2018/05/04)
New: The MatrixFree framework has been reworked to also support face integrals in DG. The new function MatrixFree::loop() takes three function pointers for workers on cells, interior faces and boundary faces. The loop can perform the data exchange with MPI and is also parallelized with threads taking into account the face dependencies. The new class FEFaceEvaluation implements the respective evaluation and access routines for face terms.
(Katharina Kormann, Martin Kronbichler, 2018/04/27)
New: Triangulation now uses std::unique_ptr to store Manifolds, and duplicates the Manifolds when Triangulation::set_manifold() is called. This allows the construction of the Manifold objects to be independent of the Triangulation. In particular, it will be possible to associate to the Triangulation reasonable default Manifold descriptors, that will be copied around together with the Triangulation itself.
(Luca Heltai, 2018/04/06)
Changed: The bundled version of Intel Threading Building Blocks has been updated to 2018 U2.
(David Wells, 2018/03/02)
Improved: The manifold smoothing algorithms applied in the Triangulation class and MappingQGeneric have been changed from the old Laplacestyle smoothing to a transfinite interpolation that linearly blends between the descriptions on the faces around a cell. The old transformation introduced boundary layers inside cells that prevented convergence rates from exceeding 3.5 in the global L2 errors on typical settings. This change also considerably improves mesh quality on settings where curved descriptions are only applied to the boundary rather than the whole volume.
(Martin Kronbichler, 2017/12/04)
New: Add ScaLAPACKMatrix and ProcessGrid wrappers for highperformance dense linear algebra routines for parallel distributed memory machines.
(Denis Davydov, Benjamin Brands, 2017/11/20)
New: The new ParticleHandler class can store and organize a collection of particles, and provide information about their properties. It is currently limited to parallel distributed computations and specific applications, but will be extended over time.
(Rene Gassmoeller, 2017/11/10)
New: The majority of deal.II classes that deal with solution vectors are now also instantiated for complexvalued vectors. In other words, complexvalued vectors should now be supported at the same level as realvalued vectors.
(Wolfgang Bangerth, 2017/11/08)
New: The new namespace Rol contains an adaptor class that provides the implementation of the ROL::Vector interface. The Trilinos package, Rapid Optimization Library (ROL), can solve unconstrained and constrained optimization problems, and optimization problems under uncertainty.
(Vishal Boddu 2017/11/02)
New: Using the Physics::Notation::Kelvin class, it is possible to store and retrieve tensors and symmetric tensors in or from a compressed format.
(JeanPaul Pelteret, 2017/10/16)
New: The function distribute_mg_dofs has been written for a parallel::shared::Triangulation. This allows for geometric multigrid computations on an adaptively refined mesh using a shared triangulation with the possibility of a user defined partitioning of the active and level cells.
(Conrad Clevenger, 2017/10/10)
New: Added support for the KINSOL solver of the SUNDIALS library. KINSOL is a solver for nonlinear algebraic systems. It includes a NewtonKrylov solver as well as Picard and fixed point solvers, both of which can be accelerated with Anderson acceleration.
(Luca Heltai, 2017/09/28)
New: Added support for the ARKode solver of the SUNDIALS library. ARKode is a solver library that provides adaptivestep time integration of the initial value problem for systems of stiff, nonstiff, and multirate systems of ordinary differential equations (ODEs) given in linearly implicit form.
(Luca Heltai, 2017/09/27)
New: There is now support for the storage of Particles and their properties in the new namespace Particles.
(Rene Gassmoeller, 2017/09/22)
New: A new ParameterAcceptor class has been added to the library. The class is intended to be used as a base class for any class that wants to handle parameters using the ParameterHandler class. If you derive all your classes from ParameterAcceptor, and declare your parameters either with parse/declare_parameters methods or via the ParameterAcceptor::add_parameter() method, then the declaring and parsing of your parameter files will be automatically managed by the ParameterAcceptor::initialize() function.
(Luca Heltai, 2017/09/20)
New: Added support for the Open Asset Import Library (Assimp) (https://assimp.sourceforge.net/). This library can be used to read about 40 different 3D graphics formats, used in 3D modelers (such as Blender, Maya, etc.). Some of these formats contain mesh information, that in turn can be read into deal.II Triangulation<2,3> objects.
(Luca Heltai, 2017/09/16)
New: A new GridTools::Cache class allows caching of some expensive data of the Triangulation class, computed generally using functions in GridTools.
(Luca Heltai, 2017/09/14)
New: A new SUNDIALS::IDA class has been added that interfaces the SUNDIALS IDA library (https://computation.llnl.gov/projects/sundials) This class can be used to solve differential algebraic equations using IDA (an implicit differential algebraic equation solver, that supports variable step and variable order BDF schemes).
(Luca Heltai, 2017/09/12)
Improved: The support for nonLagrangian elements with generalized support points has been vastly improved. FESystem::get_generalized_support_point() now returns a list of unique generalized support points for the finite element system. In order to do interpolation a function FiniteElement::convert_generalized_support_points_to_dof_values() can be used. This interface is implement for a wide variety of base classes and FESystem. TODO
(Matthias Maier, Luca Heltai, 2017/08/31)
New: A new KDTreeDistance class has been added that interfaces the nanoflann library (https://github.com/jlblancoc/nanoflann). This class can be used to extract nearest neighbour information on collection of points, query for the closest points to a target point or all points contained within a given distance.
(Luca Heltai, 2017/08/13)
New: A new MeshWorker::mesh_loop() function has been added that performs the same tasks of the MeshWorker::loop() function without forcing the users to adhere to a specific interface.
(Luca Heltai, 2017/08/12)
New: The eigenvectors of a rank2 symmetric tensor can now be computed using one of three approaches through the eigenvectors() function. The three algorithms that have been implemented are:
New: The eigenvalues of a rank2 symmetric tensor can now be computed using an analytical approach via the eigenvalues() function.
(JeanPaul Pelteret, Ester Comellas, 2017/07/27)
New: The new namespace Patterns::Tools contains utilities that can be used to convert from complicated types to strings and vice versa. These tools have been put to use in the method ParameterHandler::add_parameter() that allows users to perform in one single call the operations ParameterHandler::declare_parameter(), ParameterHandler::get() and to convert the string to a valid value stored in the variable that is given as a parameter to ParameterHandler::add_parameter().
(Luca Heltai, 2017/07/20)
New: The cmake configuration now supports unity builds with the option DDEAL_II_UNITY_BUILD=ON
. This option speeds up the build by about 10 to 25%.
(David Wells, 2017/07/20)
New: The hp::DoFHandler class can now work with shared triangulations of type parallel::shared::Triangulation.
(Wolfgang Bangerth, 2017/07/17)
New: SolverFIRE implements FIRE (Fast Inertial Relaxation Engine) for solving the problem of minimization of a given objective function.
(Vishal Boddu, Denis Davydov 2017/07/11)
New: A new manifold class TransfiniteInterpolationManifold implementing an interpolation from a curved boundary description to the interior has been added. This class enables highorder convergence rates of more than three in the power of the mesh size for situations where a curved manifold can only be prescribed to the boundary but not in a whole volume.
(Martin Kronbichler, Luca Heltai, 2017/06/01)
New: A new tutorial program, step60, shows how to deal with nonmatching discretizations using Distributed Lagrange Multipliers.
(Luca Heltai, Giovanni Alzetta, 2018/05/07)
New: Gmsh::create_triangulation_from_boundary_curve is a new method that constructs a grid from an OpenCASCADE TopoDS_Edge object. It offers the reverse functionality of the method OpenCASCADE::create_curves_from_triangulation_boundary().
(Luca Heltai, 2018/05/03)
New: NonMatching::create_coupling_sparsity_pattern and NonMatching::create_coupling_mass_matrix now have an overloaded version that takes an additional GridTools::Cache, instead of building it inside the functions. In the process, NonMatching::create_coupling_sparsity_pattern gained also an additional ConstraintMatrix argument to reflect the same argument list of its companion function NonMatching::create_coupling_mass_matrix.
(Luca Heltai, 2018/05/02)
New: Utilities::MPI::create_group allows using the functionality provided by MPI_Comm_create_group to be used also for a nonMPI3.0conforming MPI dependency.
(Daniel Arndt, 2018/04/28)
New: There are new functions FEEvaluation::gather_evaluate() and FEEvaluation::integrate_scatter() that combine the vector access through FEEvaluation::read_dof_values() and FEEvaluation::evaluate() as well as FEEvaluation::integrate() and FEEvaluation::distribute_local_to_global(), respectively. This allows to write somewhat more compact code and is also more efficient in the case of FEFaceEvaluation where equivalent functions exists.
(Katharina Kormann, Martin Kronbichler, 2018/04/27)
New: The function MatrixFree::cell_loop takes an optional boolean argument that enables the class to zero vectors within the loop for better cache locality, clearing only that part of the vector that will be accessed soon.
(Katharina Kormann, Martin Kronbichler, 2018/04/27)
New: The new field MatrixFree::AdditionalData::cell_vectorization_categories allows to control the way batches of cells are formed in MatrixFree by assigning different numbers to different parts in the domain.
(Katharina Kormann, Martin Kronbichler, 2018/04/27)
New: The class FEEvaluation can now be constructed for only a subset of the components inside an FESystem, picking e.g. the pressure part outside of a TaylorHood element.
(Katharina Kormann, Martin Kronbichler, 2018/04/27)
New: FECollection gained an equality comparison operator. Both FiniteElement and FECollcetion have a nonequality comparison operator now.
(Daniel Arndt, 2018/04/25)
New: Added missing implementation of MappingFEField::get_vertices.
(Luca Heltai, 2018/04/24)
Updated: deal.II is now compatible with PETSc version 3.9.0, SLEPC version 3.9.0
(Matthias Maier, 2018/04/23)
New: GridTools::minimal_cell_diameter and GridTools::maximal_cell_diameter now take an optional Mapping argument, that allows one to compute minimal and maximal cell diameters of deformed grids.
(Luca Heltai, 2018/04/23)
Fixed: GridTools::distributed_compute_point_locations now returns the correct maps values (third component of the output tuple). Added a test to check it.
(Giovanni Alzetta, 2018/04/23)
New: Add direct solvers (Cholesky and LU factorization) to solve problems on the GPU
(Bruno Turcksin, 2018/04/23)
New: Utilities::System::get_current_vectorization_level() returns the currently used vectorization support in string format.
(Timo Heister, 2018/04/21)
Changed: In accordance with DoFHandler, hp::DoFHandler stores a copy of the FECollection instead of a pointer.
(Daniel Arndt, 2018/04/21)
Add default constructor and member function initialize
to hp::DoFHandler, making the interface consistent with DoFHandler.
(Ce Qin, 2018/04/20)
New: Cell weighting can now be taken into account by the Metis partitioner. This can be done directly using the new GridTools::partition_triangulation() function that accepts a vector of cell weights, or by adding the appropriate signal to the triangulation itself (Triangulation::Signals::cell_weight).
(JeanPaul Pelteret, 2018/04/17)
Changed: GridTools::distributed_compute_point_locations now takes as input the global description of the manifold using bounding boxes.
(Giovanni Alzetta, 2018/04/16)
New: Introduced Triangulation::reset_manifold(), as a substitute for Triangulation::set_manifold() with only one argument. The set_manifold() method with one argument only was used to reset the manifold object. This method is now deprecated, in favor of the new, more explicative, method. An additional Triangulation::reset_all_manifolds() method has been added to remove all manifold objects from the triangulation.
(Luca Heltai, 2018/04/14)
New: GridTools::map_boundary_to_manifold_ids() allows you to set manifold ids on the boundary according to a given map of boundary to manifold ids.
(Luca Heltai, 2018/04/13)
Fixed: Ensure that numbers::NumberTraits and its corresponding operations are well defined for all supported autodifferentiable numbers.
(JeanPaul Pelteret, 2018/04/12)
Fixed: The SymmetricTensor class was previously not instantiated for autodifferentiable numbers.
(JeanPaul Pelteret, 2018/04/12)
Fixed: When using diagonal SymmetricTensors with automaticdifferentiable numbers, computations using the eigenvalue() and eigenvector() functions would return correct values of the eigenvalues/vectors. However, the derivatives of these values/vectors were incorrect as the sensitivities of the eigenvalues/vectors with respect to one another was not encoded in the returned result. Therefore, under these specific conditions the returned result is now a more coarse approximation for the eigenvalues/vectors but with the tradeoff that a meaningful approximation of the derivative of the result can now be computed.
(JeanPaul Pelteret, 2018/04/12)
Fixed: The Tensor class was previously not instantiated for autodifferentiable numbers.
(JeanPaul Pelteret, 2018/04/12)
Fixed: The Tensor::invert() function would previously not work with some Sacado number types. This has now been fixed.
(JeanPaul Pelteret, 2018/04/12)
Fixed: Some compiletimes errors would previously appear for some arithmetic functions when using Sacado::Fad::DFad types (e.g. Physics::Elasticity::Kinematics::F_iso() ). By defining the ProductType Sacado expression templates, this issue is now avoided.
(JeanPaul Pelteret, 2018/04/12)
New: FEValuesExtractors::Tensor now supports calculation of gradients.
(Denis Davydov, 2018/04/12)
New: The two new functions Polynomials::jacobi_polynomial_value() and Polynomials::jacobi_polynomial_roots() provide a uservisible implementation of Jacobi polynomials. This functionality has previously only been available internally in the library.
(Martin Kronbichler, 2018/04/06)
New: Patterns::Tools::to_string() and Patterns::Tools::to_value() simplify the conversion to and from strings of arbitrarily complex types.
(Luca Heltai, 2018/04/06)
New: Extend GridGenerator::extrude_triangulation to using exact slicing zaxis values and reimplement the existing GridGenerator::extrude_triangulation using the newly developed overload.
(Weixiong Zheng, 2018/04/05)
New: There is now a function Utilities::dynamic_unique_cast() that does for std::unique_ptr
objects what dynamic_cast
does for regular pointers.
(Wolfgang Bangerth, 2018/04/03)
Improved: VectorTools::get_position_vector() now supports parallel triangulations.
(Denis Davydov, 2018/04/01)
New: Multigrid classes obtained the new signals mg::Signals::transfer_to_mg, mg::Signals::transfer_to_global, mg::Signals::coarse_solve, mg::Signals::restriction, mg::Signals::prolongation, mg::Signals::pre_smoother_step, mg::Signals::post_smoother_step that functions can be connected to.
(Daniel Arndt, Timo Heister, 2018/03/31)
Fixed: The VectorTools::integrate_difference() function allows users to provide a weight function that can also serve as a component mask to select individual components of the solution vector for error computation. For components not selected, such a mask would then simply be zero.
In some cases, the solution vector contains NaN numbers, for example when one uses the FE_FaceQ element for certain components of the solution vector and uses a quadrature formula for error evaluation that has quadrature points in the interior of the cell. For any "regular" solution component for which the component mask has a zero weight, the value of that component will be multiplied by zero and consequently does not add anything to the error computation. However, if the NaNs of a FE_FaceQ are multiplied with zero weights, the result is still a NaN, and adding it to the values times weights of the other components results in NaNs – in effect rendering it impossible to get any information out of the VectorTools::integrate_difference() function if one of the finite elements involved is FE_FaceQ.
This is now fixed by simply skipping vector components for which the weight vector is zero. This has the same result as before for all "normal" situations, but also properly skips the NaN case outlined above.
(Wolfgang Bangerth, 2018/03/29)
Fixed: Threads::Thread<T> and Threads::Task<T> objects can now also be used when T
is a type that is only moveconstructable or movable, but not necessarily copyconstructible or copyable. Consequently, one can now also call functions that return such objects on tasks and threads. In particular, this is relevant for functions that return std::unique_ptr objects that have this property.
(Wolfgang Bangerth, 2018/03/28)
Fixed: LinearAlgebra::distributed::Vector::compress(VectorOperation::insert) now flashes ghost part of the vector in Release mode.
(Denis Davydov, 2018/03/17)
Fixed: GridTools::minimal_cell_diameter() and GridTools::maximal_cell_diameter() return the maximal respectively minimal cell diameter over the whole mesh for parallel::distributed::Triangulation object, too.
(Daniel Arndt, 2018/03/25)
New: Added a third test variant to the testsuite driver: For a test consisting of a test.prm.in (and a test.output) file the test.prm.in file will be configured/preprocessed to a test.prm file. This is done with the CMake macro CONFIGURE_FILE that replaces all strings @VARIABLE@ with the contents of the corresponding CMake variable. This is useful in particular to conveniently substitute @SOURCE_DIR@ with the full source directory path of the test.
(Matthias Maier, 2018/03/21)
Fixed: Previously, it was not possible to iterate over the local range of rows of PETSc matrix objects because one would have to call matrix.end(row)
where row
is the last locally owned row, and that triggered an assertion because this end iterator is also the begin iterator of the next row – which is not locally owned any more.
This is now fixed.
(Feimi Yu, Wolfgang Bangerth, 2018/03/20)
Fixed: The SparsityPattern::copy_from() variant that takes a FullMatrix argument was previously of quadratic complexity in the number of nonzero entries per row. This is now fixed, and the function is now linear.
(Ben Shields, Wolfgang Bangerth, 2018/03/20)
Fixed: FETools::get_interpolation_difference_matrix() used to not clear content of the matrix passed as argument, but instead just add to it. This resulted in wrong output, but is now fixed.
(Wolfgang Bangerth, 2018/03/18)
New: Add LAPACKFullMatrix<number>::set(const size_type, const size_type, const number) to set an element of the matrix.
(Denis Davydov, 2018/03/17)
Fixed: FullMatrix::residual(), FullMatrix::add_col(), and FullMatrix::add_row() now work correctly with rectangular matrices.
(David Wells, 2018/03/17)
New: JSON files which are written out by the parameter handler can be read in again with the function ParameterHandler::parse_input_from_json().
(Menno Fraters, 2018/03/12)
New: The testsuite can be run using valgrind via 'ctest S ../tests/run_memorycheck.cmake'.
(Daniel Arndt, 2018/03/11)
New: The Utilities::dealii_version_string() function returns a string representation of the deal.II version being used.
(Wolfgang Bangerth, 2018/03/09)
Fixed: Copying a Trilinos::MPI::Vector to a local deal.II Vector using operator=() resulted in only a partial copy of data to the local vector. In addition, the copied elements were offset incorrectly on each process. Similar held for copying a Trilinos::MPI::BlockVector to a local deal.II BlockVector. This has been fixed and now works as expected.
(JeanPaul Pelteret, 2018/03/07)
New: LAPACKFullMatrix::compute_inverse_svd_with_kernel allows to set an expected kernel size for an inverse singular value decomposition.
(Daniel Arndt, 2018/03/07)
New: The ParameterAcceptorProxy class allows you to wrap a class that provides a static member declare_parameters and a member parse_parameters into a ParameterAcceptor subclass.
(Luca Heltai, 2018/03/06)
New: NonMatching::create_coupling_sparsity_pattern and NonMatching::create_coupling_mass_matrix functions to handle L2 projections between arbitrary nonaligned grids.
(Luca Heltai, 2018/03/05)
Fixed: GridTools::distort_random works with parallel::shared::Triangulation objects.
(Daniel Arndt, 2018/03/05)
New: A template class LinearIndexIterator has been added with the intent of using it to generalize iterators over contiguously stored data.
(David Wells, 2018/03/01)
Changed: The GridRefinement::hyper_sphere() function used to have two template arguments (dim
and spacedim
), but it really only existed if dim == spacedim1
. Consequently, it now has lost its dim
template argument and has only retained spacedim
.
(Wolfgang Bangerth, 2018/01/25)
New: TransposeTable (the base class of LAPACKFullMatrix and ScaLAPACKMatrix) now has a random access iterator implementation similar to the one provided by SparseMatrix.
(David Wells, 2018/02/25)
New: A top level target 'expand_all_instantiations' generates all .inst files.
(Timo Heister, Daniel Arndt, 2018/02/20)
New: LAPACKFullMatrix: make operators *= and /= use Lapack function, LAPACKFullMatrix::add() uses BLAS 1 routine instead of handwritten loops
(Benjamin Brands, 2018/02/20)
Changed: The GridOut::write_mesh_per_processor_as_vtu function now only includes processor info in the .vtu filename and writes a .pvtu file when using a parallel::Triangulation
.
(Conrad Clevenger, 2018/02/19)
New: Add a new class CUDAWrappers::SparseMatrix, i.e., wrappers for cuSPARSE csr sparse matrix. The matrix is copied from deal.II own's SparseMatrix and copied to the device.
(Bruno Turcksin, 2018/02/18)
New: Add LAPACKFullMatrix<number>::Tmmult(LAPACKFullMatrix<number> &, const LAPACKFullMatrix<number> &, const Vector<number> &, const bool) const to do a triple matrix product with a diagonal matrix in the middle. Add LAPACKFullMatrix::scale_rows() to scale rows via a given vector. Make LAPACKFullMatrix::Tmmult() and LAPACKFullMatrix::mTmult() use Xsyrk if A==B.
(Denis Davydov, 2018/02/14)
New: add parallel::distributed::BlockVector::mmult(BlockVector &, const FullMatrixType &, const NumberType, const NumberType) const and parallel::distributed::BlockVector::multivector_inner_product_with_metric(const FullMatrixType &, const BlockVector &V, const bool) const to operate on multivectors with a metric tensor.
(Denis Davydov, 2018/02/10)
New: The new polynomial class Polynomials::HermiteLikeInterpolation is a modification of the Hermite polynomials with good conditioning of interpolation for all degrees, as opposed to the Polynomials::HermiteInterpolation class.
(Anian Fuchs, Martin Kronbichler, 2018/02/09)
Extend VectorTools::project() function with MatrixFree quadrature data to optionally take the finite element component index.
(Denis Davydov, 2018/02/07)
New: Added routines to perform addition, multiplication and scaling for ScaLAPACKMatrix
(Benjamin Brands, 2018/02/07)
Fixed: MGTransferMatrixFree::restrict() and MGTransferMatrixFree::prolongate() would produce a segmentation fault when used with multicomponent systems for polynomial degrees larger than 10. This is now fixed.
(Martin Kronbichler, 2018/01/30)
New: Add AssertCusparse macro to assert the error code of cuSPARSE function
(Bruno Turcksin, 2018/01/29)
New: There are new overloads of make_array_view for Tensor, SymmetricTensor, LAPACKFullMatrix, Cstyle array and Vector.
(Daniel Arndt, 2018/01/25)
New: Tensor::begin_raw, Tensor::end_raw, SymmetricTensor::begin_raw and SymmetricTensor::end_raw provide access to the underlying storage for Tensor and SymmetricTensor.
(Daniel Arndt, 2018/01/25)
New: Add save/load functions for ScaLAPACKMatrix to save/load distributed matrix to/from disc using HDF5. If HDF is configured with MPI, parallel I/O is used to save/load the matrix.
(Benjamin Brands, 2018/01/25)
New: Added Utilities::MPI::gather wrapper and tests to gather objects from all to one MPI process.
(Benjamin Brands, 2018/01/25)
Improved: The implementation of the SolverQMRS iteration has been renewed to perform much faster and to be adaptable to left and right side preconditioning of the system matrix.
(Ingo Kligge, 2018/01/23)
New: New function GridTools::distributed_compute_point_locations ; similarly to GridTools::compute_point_locations , given a vector of points, it returns vectors containing them, their reference position and the process owning them as it works with shared and distributed meshes.
(Giovanni Alzetta, 2018/01/19)
Fixed: In some situations, DoFTools::locally_owned_dofs_per_subdomain() returned a vector of the wrong size. Specifically, this happened if the processor with the highest subdomain id owned no degrees of freedom. This is now fixed.
(JeanPaul Pelteret, Wolfgang Bangerth, 2018/01/18)
Fixed: In FE_Enriched element, avoid evaluating quadrature points if no dofs are assigned. This happens when FE_Nothing is used together with other FE (i.e. FE_Q) as enrichments in FE_Enriched constructor.
(Nivesh Dommaraju, 2018/01/15)
New: Enhanced RaviartThomas finite element FE_RT_Bubbles, allows for local elimination of a vector variable in multipoint flux mixed finite element methods and similar.
(Eldar Khattatov, 2018/01/11)
New: Enhanced RaviartThomas polynomial space PolynomialsRT_Bubbles consisting of RT_k + curls bubbles.
(Eldar Khattatov, 2018/01/09)
Fixed: The DataOut::write_vtu output pretended to write double precision data, while in reality the data was first converted to float and then written as double. This was fixed by writing all data (including vertex positions) as single precision float and adjusting the output types accordingly.
(Rene Gassmoeller, 2018/01/04)
New: Implementing CylindricalManifold::push_forward_gradient allows to compute normal vectors for boundaries described by CylindricalManifold objects.
(Daniel Arndt, 2018/01/04)
Fixed: In parallel computations, the DoFRenumbering::hierarchical() function created DoF indices that were dependent on the previous DoF indices owned by each processor. This was not intended: the new DoF indices were supposed to only depend on the order of cells, not any previous numbering. This is now fixed.
(Wolfgang Bangerth, 2017/12/31)
New: There is now a function VectorizedArray::streaming_store() that stores the content of a vectorized array into another array which bypasses the CPU's caches on supported architectures using _mm_stream_pd() intrinsics. This operation is useful because it can avoid the readforownership memory access on large arrays and directly write to the destination for streaming stores. However, it can also be detrimental to performance in case the store would fit into caches. This new function is not used within the library because only the final user code knows the properties of hardware and whether a particular store pattern is so large that it will not fit into caches and thus benefit from this optimization.
(Martin Kronbichler, 2017/12/27)
Fixed: implement missing instantiation of TrilinosWrappers::SparsityPattern::copy_from().
(Timo Heister, 2017/12/22)
Improved: SparseMatrix::mmult(..) is now using the DynamicSparsityPattern::compute_mmult_pattern(..) function to create the sparsity pattern of the final matrix C.
(Christoph Goering, 2017/12/22)
New: Added QSimplex, QTrianglePolar, QDuffy, and QSplit classes to perform quadratures on reference simplices, on their affine transformations, and on hyper cubes split by a given point.
(Luca Heltai, 2017/12/21)
New: add LAPACK_WITH_64BIT_BLAS_INDICES configure option and introduce types::blas_int to support 64bit BLAS indices.
(Denis Davydov, 2017/12/21)
New: add LinearAlgebra::distributed::BlockVector::norm_sqr() and LinearAlgebra::distributed::Vector::norm_sqr() that return square of the l2 norm.
(Denis Davydov, 2017/12/21)
New: LAPACKFullMatrix::remove_row_and_column() to remove certain row and column of the matrix.
(Denis Davydov, 2017/12/20)
Fixed: parallel::distributed::Triangulation::add_periodicity missed to update the ghost_owners in the NumberCache member variable.
(Daniel Arndt, Sambit Das, 2017/12/11)
New: DynamicSparsityPattern::compute_mmult_pattern(left, right) with two arguents of either a DynamicSparsityPattern or a SparsityPattern; or any combination of those. The result is the pattern which is obtained by multiplying the two sparse matrices on the given sparsity patterns.
(Christoph Goering, 2017/12/19)
New: LAPACKFullMatrix::rank1_update(const number, const Vector<number> &) performs a rank1 update of a matrix.
(Denis Davydov, 2017/12/17)
New: Add move constructor for TrilinosWrappers::SparseMatrix and TrilinosWrappers::SparsityPattern
(Bruno Turcksin, 2017/12/10)
New: Add Vector::grow_or_shrink() and LAPACKFullMatrix::grow_or_shrink() to (partly) keep the previous values upon resizing.
(Denis Davydov, 2017/12/16)
Fixed: parallel::distributed::Triangulation::communicate_locally_moved_vertices treats periodic faces correctly now.
(Daniel Arndt, Sambit Das, 2017/12/11)
New: Add Jupyter Notebook explaining how to use the python wrappers.
(Bruno Turcksin, 2017/12/10)
Improved: DoFTools::extract_hanging_node_dofs works for parallel::shared::Triangualtion and parallel::distributed::Triangulation and reports locally relevant DoFs.
(Daniel Arndt, 2017/12/21)
New: Now opencascade also works for spacedim == 2.
(Luca Heltai, 2017/12/08)
New: Improve support of triangular matrices in LAPACKFullMatrix wrappers. Add LAPACKFullMatrix::solve() to solve a system of equations either when the matrix is factorized (Cholesky/LU) or triangular.
(Denis Davydov, 2017/12/07)
New: Created a new method OpenCASCADE::Utilities::create_curves_from_triangulation_boundary that smoothly interpolates the boundary of two dimensional triangulations into a vector of OpenCASCADE TopoDS_Edge closed curves, representing the connected components of the boundary.
(Dirk Peschka, Luca Heltai, 2017/12/05)
Fixed: SparsityPattern::print_svg(...) has the right size for the white rectangle so there is a small margin between the first and last row/column of the printed matrix.
(Christoph Goering, 2017/12/05)
New: Add color_sparsity_pattern function to color a graph represented by SparsityPattern Object. The function uses coloring function from ZOLTAN library.
(Nivesh Dommaraju, 2017/12/04)
New: GridGenerator::extract_boundary_mesh now correctly extracts also Manifold information.
(Luca Heltai, Dirk Peschka, 2017/12/04)
Fixed: Manifold::normal_vector() used to pick two directions that are almost linearly dependent quite often, which resulted in very bad accuracy of the normal vectors. This has been fixed. The downstream use of this function on curved geometries in VectorTools::compute_no_normal_flux_constraints() has now become much more accurate and robust.
(Martin Kronbichler, 2017/12/02)
Improved: SphericalManifold::get_new_points now computes a lot of information outside of the loop over all points in get_new_point, which saves a significant amount of time for spherical geometries.
(Rene Gassmoeller, 2017/12/01)
Fixed: MappingQGeneric::transform_real_to_unit_cell() with degree 1 would previously fail for Cartesian meshes with certain combinations of roundoff errors. This is now fixed.
(Martin Kronbichler, 2017/12/01)
New: Added CellId::serialize function and a test for it using pack/unpack
(Giovanni Alzetta, 2017/11/28)
New: Added BoundingBox::serialize function and a test for it using pack/unpack functions
(Giovanni Alzetta, 2017/11/24)
Extend: partition_triangulation with options for choosing between METIS or ZOLTAN partitioners. METIS (the original partitioner) is the default.
(Nivesh Dommaraju, 2017/11/23)
New: Added random_value<T>() and random_point<dim>() in tests.h, to simplify and unify our way to generate random numbers and points.
(Luca Heltai, 2017/11/21)
Fixed: Added a tolerance inside GridTools::find_active_cell_around_point to errors with boundary points and a test with such points.
(Giovanni Alzetta, 2017/11/21)
Fixed: Added convert_generalized_support_point_values_to_dof_values into FE_FaceQ and FE_TraceQ. This is needed by VectorTools::interpolate
(Praveen Chandrashekar, 2017/11/19)
Fixed: ConvergenceTable::evaluate_convergence_rates now also supports values and keys of type 'unsigned long long int' as used for types::global_dof_index with 64 bit integers enabled.
(Martin Kronbichler, 2017/11/19)
New: Add Utilities::pack and Utilities::unpack, to serialize and unserialize arbitrary objects that support boost::serialization/deserialization.
(Luca Heltai, 2017/11/18)
New: Utilities::MPI::all_gather and Utilities::MPI::some_to_some functions have been added to perform general collective communications between processors.
(Giovanni Alzetta, Luca Heltai, 2017/11/18)
Fixed: The function GridRefinement::refine_and_coarsen_fixed_number no longer produces an EXC_BAD_ACCESS exception when either of the arguments top_fraction or bottom_fraction are set to 1.0.
(Oliver Sutton, 2017/11/17)
Changed: The specializations of the scalar_product function for Sacado::Fad::DFad numbers were redundant and have been removed.
(JeanPaul Pelteret, 2017/11/17)
New: ArrayView has a data() member function making it more similar to the standard containers.
(Daniel Arndt, 2017/11/17)
New: Math operations for some automatically differentiable numbers have been extended. We also import some math functions for these number types into the standard namespace. This gives full compatibility with the Tensor and SymmetricTensor classes, and facilitates the integration of these number types into existing code.
(JeanPaul Pelteret, 2017/11/16)
Changed: Core functions in the FEValues and FEValuesViews classes have been updated to allow the use of certain automatically differentiable numbers to represent degreeoffreedom values.
(JeanPaul Pelteret, 2017/11/16)
New: In parallel, the starting indices of DoFRenumbering::Cuthill_McKee() can now also be locally active indices. They no longer need to be locally owned DoF indices.
(Wolfgang Bangerth, 2017/11/16)
Fixed and improved: The GridTools::compute_point_locations() now always return the correct number of cells. The new algorithm is also significantly faster.
(Giovanni Alzetta, 2017/11/15)
New: Given a vector of points, the GridTools::guess_point_owner() function uses bounding boxes describing the mesh to guess the processes that own these points. A test for the function has been added.
(Giovanni Alzetta, 2017/11/13)
Fixed: The VectorTools::interpolate_boundary_values() function that takes a hp::MappingCollection argument was declared, but not instantiated, and could consequently not be called. This is now fixed.
(Wolfgang Bangerth, 2017/11/12)
New: TrilinosWrappers::DirectSolver::solve() with two LinearAlgebra::distributed::Vector arguments now reuses the factorization of a matrix set in the TrilinosWrappers::DirectSolver::initialize() call.
(Martin Kronbichler, 2017/11/10)
Modified: Function Functions::FEFieldFunction::compute_point_locations now calls GridTools::compute_point_locations Modified: Function GridTools::compute_point_locations now accepts a cell hint (a test for this hint has been added)
(Giovanni Alzetta, 2017/11/06)
Fixed: The algorithm in Manifold::normal_vector could cause divisionbyzero errors for special cases. This was fixed.
(Rene Gassmoeller, 2017/11/03)
New: A uniform interface to automatically differentiable numbers that will be supported by the deal.II library has been implemented. A corresponding set of type traits classes that help query information about these numbers has also been implemented.
(JeanPaul Pelteret, 2017/11/03)
Improved: Both Sundials2.7.0 and Sundials3.0.0 are supported.
(Daniel Arndt, 2017/11/03)
Fixed: ChartManifold now implements the method Manifold::get_intermediate_point().
(Martin Kronbichler, 2017/11/02)
Improved: SphericalManifold::get_new_point uses a Newtontype optimization algorithm instead of a simple averaging.
(Daniel Arndt, 2017/11/02)
Fixed: The spherical manifold used to do a division by zero inside of SphericalManifold::get_new_point() if the new point was exactly at the center of the manifold. This is fixed now.
(Rene Gassmoeller, 2017/10/31)
New: Added GridTools::exchange_local_bounding_boxes function and tests to exchange vectors of bounding boxes and gather them in every process.
(Giovanni Alzetta, 2017/10/30)
Fixed: The treatment of inhomogeneous boundary conditions in VectorTools::project has been fixed.
(Daniel Arndt, 2017/10/30)
Improved: The step37 tutorial program now describes how to deal with inhomogeneous Dirichlet boundary values with the matrixfree framework in the results section of the tutorial.
(Martin Kronbichler, 2017/10/29)
Fixed: The member variable FEEvaluation::dofs_per_cell and the static variable FEEvaluation::tensor_dofs_per_cell only returned the number of degrees of freedom per scalar component also for vectorvalued evaluators, rather than returning the number accumulated over all components in the evaluator. This is now fixed.
(Martin Kronbichler, 2017/10/28)
New: Created a new Patterns::Tuple class, that allows parsing arbitrary tuples of objects in parameter files.
(Luca Heltai, 2017/10/26)
New: Removed a typo in QTelles and added a new singular integration test.
(Nicola Giuliani, 25/10/2017
New: Add LAPACKFullMatrix<number>::add (const number a, const LAPACKFullMatrix<number> &A)
(Denis Davydov, 2017/10/25)
Fixed: The computation of the inverse jacobians in the MappingCartesian class was broken. It would trigger an assertion at runtime complaining that only an assignment with zero is allowed. This is fixed now.
(Rene Gassmoeller, 2017/10/24)
New: The function LinearAlgebra::distributed::Vector::copy_locally_owned_data_from() can copy the locally owned range in a parallel vector without considering the ghost entries which might be different between the two vectors.
(Martin Kronbichler, 2017/10/24)
Added: Function GridTools::compute_point_locations and a test for it. The function uses the recently added Cache to improve performance.
(Giovanni Alzetta, 2017/10/24)
Added: Test for Functions::FEFieldFunction::compute_point_locations
(Giovanni Alzetta, 2017/10/23)
New: DoFHandler::renumber_dofs() for multigrid levels is now implemented for parallel::distributed::Triangulation for the special case that the numbering does not change over processor boundaries.
(Martin Kronbichler, 2017/10/20)
New: Added BoundingBox::get_neighbor_type() which returns the neighboring relation between two bounding boxes. Added the function GridTools::compute_mesh_predicate_bounding_box() which returns a vector of BoundingBoxes covering the cells for which a given predicate is true.
(Giovanni Alzetta, 2017/10/20)
Changed: The enum
VectorOperation is now declared in its own header instead of the header for Vector.
(David Wells, 2017/10/20)
New: MatrixFreeOperators::Base now also stores the diagonal of the operator that shall be populated in derived class in compute_diagonal() method. The read access is provided by MatrixFreeOperators::Base::get_matrix_diagonal().
(Denis Davydov, 2017/10/19)
New: A new version of GridTools::find_active_cell_around_point() has been added that exploits a GridTools::Cache object.
(Luca Heltai, 2017/10/18)
New: The class Utilities::MPI::Partitioner can now handle the MPI import and export calls that were previously hidden in the vector class. This allows to use MPI communication of integers besides the previously supported floating point data.
(Martin Kronbichler, 2017/10/17)
New: The class Utilities::MPI::Partitioner now supports handling only a selection of ghost indices within a larger ghost set. Use this feature by adding a second, larger, IndexSet to Partitioner::set_ghost_indices().
(Martin Kronbichler, 2017/10/17)
Fixed: The transformation operation for rank2 and rank4 (nonsymmetric) tensors in the Physics::Transformations was incorrectly implemented. This has now been fixed.
(JeanPaul Pelteret, 2017/10/17)
New: Add LAPACKFullMatrix::trace() to calculate trace of a matrix.
(Denis Davydov, 2017/10/09)
New: FiniteElement::get_sub_fe() allows you to return a contained FiniteElement based on a ComponentMask.
(Timo Heister, 2017/10/07)
Fixed: It was previously not possible to create objects of type Table with elements that are not copyable or copy constructible. An example of such a type is std::unique_ptr
. This has now been fixed.
(Wolfgang Bangerth, 2017/10/02)
New: MGTransferBlockMatrixFree allows to use a separate DoFHandler for each block. The required interface is supported by PreconditionMG.
(Daniel Arndt, 2017/10/01)
New: Add Utilities::MPI::sum() for LAPACKFullMatrix objects.
(Denis Davydov, 2017/09/04)
New: Add LinearAlgebra::distributed::BlockVector<Number>::multivector_inner_product() to perform inner product between each block of the two block vectors.
(Denis Davydov, 2017/09/30)
Fixed: One of the functions GridTools::find_active_cell_around_point() supports an additional mapping
argument. This mapping was used to actually transform points from the real to the reference space, but somewhere in the algorithm we look for the closest_vertex
to the target point. Such search was done ignoring if the mapping
actually changes the location of the vertices. This is now fixed, and the internal function that looks for the closest vertex now takes into account the mapping
transformation.
(Luca Heltai, 2017/09/27)
Changed: Split TensorProductMatrixSymmetricSum into base and derived class. Added a template specialization for TensorProductMatrixSymmetricSum in case of having a VectorizedArray as arithmetic type. In addition, TensorProductMatrixSymmetricSum is now able to handle distinct 1D matrices in each tensor dimension.
(Julius Witte, 2017/09/27)
New: Add MGTransferMatrixFree::interpolate_to_mg() to transfer finelevel solution to a multigrid vector
(Denis Davydov, 2017/09/27)
New: Extend MappingQEulerian to work with multigrid vectors.
(Denis Davydov, 2017/09/27)
Fixed: The hp version of DoFTools::make_flux_sparsity_pattern() with masks for DoF couplings now works with nonprimitive FE spaces, similarly to its nonhp version. The hp version of VectorTools::project_boundary_values() is modified to work properly with Hdiv conforming finite elements.
(Eldar Khattatov, 2017/09/26)
New: Extend LAPACKFullMatrix to provide (i) Cholesky factorization (and use it for inversion); (ii) an estimate of the reciprocal condition number; (iii) l1_norm(), linfty_norm() and frobenius_norm().
(Denis Davydov, 2017/09/26)
New: There are new signals in the Triangulation class that signal the beginning and end of a parallel::distributed refinement and serialization. This allows to connect functions that should be called uniquely once before and after refinement and serialization for parallel distributed Triangulations (something not possible with the existing signals).
(Rene Gassmoeller, 2017/09/25)
Changed: The new methods ArrayView::cbegin() and ArrayView::cend() return const_iterators and are const themselves. ArrayView::begin() and ArrayView::end() return (nonconst) iterators and are const themselves.
(Daniel Arndt, Julius Witte, 2017/09/24)
New: Added two methods to the class BoundingBox. The first, merge_with(BoundingBox) merges the current BoundingBox with BBox. The second, volume(), returns the volume (dimdimensional measure) of the current BoundingBox
(Giovanni Alzetta, 2017/09/22)
New: A GridTools::find_closest_vertex() has been added that takes a map of indexvertex position and a target point, and returns the index of the closest point in the map.
(Luca Heltai, 2017/09/19)
New: A GridTools::extract_used_vertices() has been added that returns a map of indexvertex position, using an optional Mapping<dim,spacedim> object to compute the vertex locations.
(Luca Heltai, 2017/09/18)
New: A new GridIn::read_assimp() function allows to read in a file in one of the (40) formats supported by Assimp.
(Luca Heltai, 2017/09/16)
New: Added the base class BoundingBox. The BoundingBox is constructed from a std::pair of points in real space, ordered following the convention bottomleft, topright. The class has the method BoundingBox::point_inside(Point<spacedim> p) which checks for the presence of a point inside it.
(Giovanni Alzetta, 2017/09/15)
Fixed: The destructor of TimerOutput::Scope will now always exit the subsection it entered in its constructor, instead of just exiting the last subsection that was created in the provided TimerOutput object. This fixes timing output measurements for multithreaded applications and cases where subsections are nested.
(David Wells, 2017/09/14)
New: FiniteElement and FESystem have move constructors.
(Daniel Arndt, 2017/09/11)
New: There is now a conversion constructor that converts std::vector objects to ArrayView objects.
(Wolfgang Bangerth, 2017/09/11)
New: Add more Python wrappers to generate meshes and to flatten a Triangulation.
(Bruno Turcksin, 2017/09/10)
New: The newly implemented FiniteElement::operator^ allows in combination with the new variadic template constructor or the new std::initializer_list constructor for FESystem to construct FESystem objects using syntax like
(Daniel Arndt, 2017/09/05)
New: Add Utilities::MPI::sum() for FullMatrix objects.
(Denis Davydov, 2017/09/04)
Changed: The various overloads for Threads::new_thread() and Threads::new_task() taking from 0 to 9 parameters have been replaced by two variadic template versions.
(Daniel Arndt, 2017/09/01)
New: The new DataPostprocessorTensor can help visualize tensorvalued quantities.
(Wolfgang Bangerth, 2017/08/30)
New: The DataPostprocessorVector class now has an extensive example of its use.
(Wolfgang Bangerth, 2017/08/29)
Changed: DoFHandler::get_finite_element(unsigned int) has been removed again in favor of equipping DoFHandler::get_fe() with a (defaulted) unsigned int parameter. hp::DoFHandler::get_finite_element(unsigned int) has been renamed to hp::DoFHandler::get_fe(unsigned int).
(Daniel Arndt, 2017/08/25)
New: Extend Point::distance_square() to work with VectorizedArray numbers.
(Denis Davydov, 2017/08/24)
Fixed: On Vector::add(..), a pointer to the first element of a std::vector is gotten by std::vector::data() instead of using the "&v[0]" idiom, which results in an undefined behaviour.
(Tulio Ligneul, 2017/08/23)
Fixed: It turns out that it was not possible to copy an invalid operator, i.e., the following code would yield an exception:
This made no sense, and has now been fixed.
(Wolfgang Bangerth, 2017/08/23)
New: The struct is_base_of_all is a generalization of std::is_base_of to template parameter packs and tests if all classes in the parameter pack have a given class as base class or are this class.
(Daniel Arndt, 2017/08/21)
New: The constructors for hp::FECollection taking a fixed number of FiniteElement objects as argument were replaced by a variadic template constructor taking an arbitrary number of objects.
(Daniel Arndt, 2017/08/20)
New: DoFHandler::get_fe() and hp::DoFHandler::get_fe() have been deprecated in favor of DoFHandler::get_finite_element(), DoFHandler::get_fe_collection(), hp::DoFHandler::get_finite_element() and hp::DoFHandler::get_fe_collection() that share a common signature each.
(Daniel Arndt, 2017/08/19)
New: Add MGTransferBlockMatrixFree which allows using geometric multigrids with LinearAlgebra::distributed::BlockVector<Number> where each block uses the same index space of degrees of freedom.
(Denis Davydov, 2017/08/18)
Fixed: The nonmember invert() function operating on SymmetricTensors has been generalized for all number types.
(Wolfgang Bangerth, JeanPaul Pelteret, 2017/08/17)
New: An additional nonmember operator*() function, performing a single contraction between a symmetric and normal tensor of arbitrary rank, has been added to the SymmetricTensor class.
(Matthias Maier, JeanPaul Pelteret, 2017/08/17)
Fixed: The following nonmember functions operating on SymmetricTensors have been generalized for all number types:
Fixed: The Tensor::l1_norm() and Tensor::linfty_norm() functions have now been generalized for all number types.
(JeanPaul Pelteret, 2017/08/17)
Improved: The CMake CUDA detection and configure has been completely rewritten. CUDA support now requires CMake version 3.9 or newer.
(Matthias Maier, 2017/08/08)
Fixed: ConsecutiveControl now asserts that check() is being called for zeroth iteration.
(Denis Davydov, 2017/08/16)
Fixed: Constructing an IndexSet from an Epetra_Map generated the wrong size.
(Timo Heister, 2017/08/15)
Deprecated: types_are_equal has been deprecated in favor of std::is_same.
(David Wells, 2017/08/14)
Deprecated: constraint_and_return_value has been deprecated in favor of std::enable_if.
(Daniel Arndt, 2017/08/14)
New: A new version of GridTools::find_active_cell_around_point() has been added that exploits local maps constructed using standard GridTools utilities.
(Rene Gassmoeller, Luca Heltai, 2017/08/13)
New: GridTools::exchange_cell_data_to_ghosts() to exchange ghost data.
(Timo Heister, 2017/08/12)
New: All of the FEValuesViews classes now have a set of functions named get_function_*_from_local_dof_values that return the function values, gradients etc. from a local vector of degreeoffreedom values. They also now contain a data structure called OutputType that that provides the output type for the product of the value and derivatives of basis functions of the view and any number type.
(JeanPaul Pelteret, Luca Heltai, 2017/08/10)
Deprecated: internal::bool2type and int2type has been depreated in favor of std::intergral_constant.
(Daniel Arndt, 2017/08/10)
Fixed: Creation of coverage information via compiling with 'DEAL_II_SETUP_COVERAGE=ON' and 'ctest S ../tests/run_coverage.cmake' works again.
(Matthias Maier, Daniel Arndt, 2017/08/09)
Improved: ConstraintMatrix now has a member function ConstraintMatrix::get_lines() that returns a range object that allows iteration over all constraint entries.
(Matthias Maier, 2017/08/08)
Changed: ConstraintMatrix::ConstraintLine is now a public member class definition.
(Matthias Maier, 2017/08/08)
New: MappingQGeneric uses tensorized evaluation for maybe_update_Jacobians, maybe_compute_q_points and maybe_update_jacobian_grads in case the Quadrature object represents a tensor product of identical onedimensional quadrature formulas.
(Daniel Arndt, 2017/08/04)
New: Quadrature has a (defaulted) move assignment operator.
(Daniel Arndt, 2017/08/04)
New: Add Utilities::LinearAlgebra::chebyshev_filter() to apply Chebyshev filter of a given degree.
(Denis Davydov, 2017/08/02)
New: Add Utilities::LinearAlgebra::lanczos_largest_eigenvalue() to estimate the largest eigenvalue of a symmetric linear operator using ksteps of Lanczos algorithm.
(Denis Davydov, 2017/08/01)
Fixed: DiagonalMatrix<VectorType> can now be used as a linear operator.
(Denis Davydov, 2017/08/01)
New: Extend pArpack solver to cover mode 1 (standard eigenvalue problem) and mode 2 (generalized eigenvalue problem without spectral transformation).
(Denis Davydov, 2017/08/01)
New: Quadrature::is_tensor_product() returns if the corresponding quadrature formula can be represented as a tensor product and the quadrature points are sorted lexicographically. The corresponding 1D objects can be queried using Quadrature::get_tensor_basis().
(Daniel Arndt, 2017/07/26)
Fixed: DataOutBase::write_hdf5_parallel() can also be used in case deal.II and HDF5 are not compiled with MPI support.
(Daniel Arndt, 2017/07/26)
Fixed: The FunctionParser class demonstrated an incompatibility with (very) old versions of the Threading Building Blocks used in deal.II. This is now worked around.
(Alberto Sartori, Wolfgang Bangerth, 2017/07/26)
Initialization of Chebyshev smoother: Make initial guess robust with respect to number of processors by operating on the global index.
(Niklas Fehn, 2017/07/25)
Fixed: MappingQGeneric::transform_real_to_unit_cell had a bug where, for certain parallelograms, a problem with subtracting nearly equal floating point numbers ruined most of the digits of accuracy in the coordinate transformation.
(Giovanni Di Ilio, David Wells, JeanPaul Pelteret, 2017/07/25)
New: There is now a function ConstraintMatrix::copy_from() that allows copying objects of type ConstraintMatrix.
(Wolfgang Bangerth, 2017/07/27)
Fixed: The ghost layer computation for the case when periodic boundary conditions were set on all boundaries in 3D missed some cells leading to a deadlock when used on 43 MPI ranks or more. This is now fixed.
(Martin Kronbichler, 2017/07/20)
New: DataOutBase now supports writing zerodimensional data in a higher spatial dimension (e.g. vertices, quadrature points, particles) for a number of output formats (vtu,vtk,gnuplot,hdf5). A consequence of the change is that HDF5 now in general supports output for dim<spacedim.
(Rene Gassmoeller, 2017/07/18)
Improved: linear_operator gained an overload that directly takes a LinearOperator as exemplar object to directly copy the reinit_(domainrange)_vector functions from.
(Matthias Maier, 2017/07/14)
New: Adding the functions FE_FaceQ::hp_vertex_dof_identities(), FE_FaceQ::hp_line_dof_identities() and FE_FaceQ::hp_quad_dof_identities() allows FE_FaceQ to be used in combination with a hp::DoFHandler.
(Samuel Imfeld, Daniel Arndt, 2017/07/11)
New: A new class TensorProductMatrixSymmetricSum that represents a tensor product of a mass and derivative matrix has been added. It implements both the matrixvector operation (TensorProductMatrixSymmetricSum::vmult()) as well as the inverse operation (TensorProductMatrixSymmetricSum::apply_inverse()) operation in tensorial form with optimal complexity of size^{dim+1}
operations. The inverse operation uses the fast diagonalization method.
(Martin Kronbichler, 2017/07/11)
Fixed: The DoFHandler::n_boundary_dofs() and hp::DoFHandler::n_boundary_dofs() functions did not take into account that a onedimensional triangulation may consist of more than one segment (and then may have more than two vertices on the boundary). This is now fixed.
(Wolfgang Bangerth, 2017/07/10)
New: MGConstrainedDoFs object now stores information for periodicity constraints which is used during the multigrid transfer by MGTransferMatrixFree and MGTransferPrebuilt.
(Conrad Clevenger, Martin Kronbichler, 2017/07/08)
Changed: The DoFRenumbering::compute_component_wise() function used to take two template arguments that denote iterator types, one for the begin and one for the end iterator. This has been changed: It now only takes an iterator type argument for the begin iterator, and the end iterator is automatically casted to that same type.
(Wolfgang Bangerth, 2017/07/08)
New: The DoFRenumbering::component_wise() function was available for all dimensions and space dimensions for regular DoFHandler objects, but only for dimension == space_dimension for hp::DoFHandler. This is now fixed.
(Wolfgang Bangerth, 2017/07/07)
Fixed: TrilinosWrappers::PreconditionBlock* classes now work correctly if one of the processors does not own any rows.
(Timo Heister, 2017/07/04)
Deprecated: The ParpackSolver::Shift class has been deprecated.
(Matthias Maier, 2017/07/03)
New: The determinant of a LAPACKMatrix can now be computed via a call to LAPACKMatrix::determinant().
(JeanPaul Pelteret, 2017/07/02)
Fixed: FullMatrix::left_invert() and FullMatrix::right_invert() no longer fail when the matrix is square.
(Andreas Kergassner, JeanPaul Pelteret, 2017/07/02)
Fixed: FullMatrix::determinant() is now implemented for square matrices of size greater than 3. In this case, it is computed through the LU decomposition of the matrix using LAPACK.
(Andreas Kergassner, JeanPaul Pelteret, 2017/07/02)
Changed: The complete_index_set() function now returns a compressed version of the IndexSet that contains all indices.
(Wolfgang Bangerth, 2017/06/24)
New: The Timer class now has new functions Timer::get_total_data(), Timer::print_total_data, Timer::last_wall_time, Timer::cpu_time and Timer::last_cpu_time() to access information about the total run as well as the last lap.
(Daniel Arndt, 2017/06/23)
Deprecated: The static member variables DoFHandler::invalid_dof_index and hp::DoFHandler::invalid_dof_index are now deprecated. Use numbers::invalid_dof_index instead.
(Wolfgang Bangerth, 2017/06/20)
New: Patterns::Map now accepts a new optional parameter, that allows to specify also the separator between keyvalue pairs.
(Luca Heltai, 2017/06/14)
New: Utilities::split_string_list() now has a version that allows to pass a string instead of a character. With this it is possible to split on double chars, for example: "alpha ;; beta ;; gamma".
(Luca Heltai, 2017/06/14)
Fixed: The passing of additional settings to TrilinosWrappers::SolverBase from derived solvers was never completed. This has now been corrected.
(Julian Andrej, JeanPaul Pelteret, 2017/06/14)
Fixed: Patterns derived from PatternBase now return std::unique_ptr<PatternBase> in their clone() and create() functions.
(Luca Heltai, 2017/06/12)
Changed: ParameterHandler::print_parameters() was at times looking at individual bits instead of just the declared values of ParameterHandler::OutputStyle. This presumably allowed for calling that function with a combination of the OutputStyle flags, for reasons that no longer seem particularly relevant nor obvious.
This possibility has now been removed from the current implementation of the function, but is for the moment retained for the (deprecated) function ParameterHandler::print_parameters_section().
(Wolfgang Bangerth, 2017/06/09)
Fixed: The ParameterHandler::print_parameters() function used the previously set fill character of the output stream, but it should have filled with spaces instead. This is now fixed.
(Wolfgang Bangerth, 2017/06/07)
Fixed: FullMatrix::print_formatted() and LAPACKFullMatrix::print_formatted would previously print NaN values as zero. This is now fixed.
(Simon Sticko, Martin Kronbichler, 2017/06/08)
Initialization of Chebyshev smoother: project vector onto space of vectors with zero mean which is necessary in some cases where the matrix is singular. Rename variable is_initialized > eigenvalues_are_initialized in order to improve code readability.
(Niklas Fehn, 2017/06/07)
Improved: The multigrid smoothers now provide an additional apply() method that specializes the smoothing for the case the input vector is set to zero. This method is used in the Multigrid class and saves one matrixvector product, which can speed up the multigrid algorithm by up to 10 percent, depending on the cost of the smoother.
(Martin Kronbichler, 2017/06/07)
Changed: The ParameterHandler::print_parameters() function is now const
, as one would expect it to be.
(Wolfgang Bangerth, 2017/06/07)
New: Added a new polynomial class IntegratedLegendreSZ. This implements the integrated Legendre polynomials described in the 2006 PhD thesis of Sabine Zaglmayr.
(Ross Kynch, 2017/06/06)
New: A function TriaAccessor::real_to_unit_cell_affine_approximation that computes an approximation to a point in unit coordinates has been added.
(Martin Kronbichler, 2017/06/05)
Improved: CylindricalManifold is now really using cylindrical coordinates instead of taking an average in space coordinates.
(Daniel Arndt, 2017/06/05)
Deprecated: The ParameterHandler::print_parameters_section() function has been deprecated.
(Wolfgang Bangerth, 2017/06/02)
New: The ConeBoundary class now has a member function ConeBoundary::normal_vector().
(Anne Glerum, 2017/05/30)
Fixed: DataOutInterface::set_flags() was broken when passed an object of type DataOutBase::GnuplotFlags. This is now fixed.
(Wolfgang Bangerth, 2017/05/30)
New: deal.IIFeatureConfig.cmake will contain detailed information about feature configuration and is installed along with other CMake files. These variables can be loaded by calling INCLUDE(${DEAL_II_FEATURE_CONFIG}). This is useful when users want to link in additional parts of dependencies, not required by deal.II (i.e. filesystem library from Boost package).
(Denis Davydov and Matthias Maier, 2017/05/29)
Removed: The build_tests category of tests has been removed. Build tests now compile all configurable example steps as part of the regular build stage.
(Matthias Maier, 2017/04/06)
New: There is a new header, lac/trilinos_index_access.h
, that provides index and size functions for some common Trilinos objects that work correctly for both 32 and 64 bit code.
(David Wells, 2017/05/10)
New: Add method to access timing data from TimerOutput from code.
(Jonathan Robey, 2017/05/22)
New: Added a GridTools::regularize_corner_cells function that detects if the boundary cells of a mesh at corner positions (with dim adjacent faces on the boundary) need to be split into cells with smaller angles.
(Luca Heltai, Martin Kronbichler, 2017/05/18)
New: Add a new class ConsecutiveControl which returns SolverControl::State::success if and only if a certain positive number of consecutive iterations satisfy the specified tolerance.
(Denis Davydov, 2017/05/17)
New: The vertex (and face) iterator for dimension 1 triangulations now implements the get_manifold method.
(David Wells, 2017/05/10)
New: Add support for matrixfree operator application on GPU using CUDA. Constraints (Dirichlet boundary conditions, hanging nodes, etc.) cannot be applied.
(Bruno Turcksin and Karl Ljungkvist, 2017/05/10)
New: TrilinosWrappers::MPI::BlockMatrix can now return its MPI_Comm via get_mpi_communicator() and its range and domain indices via locally_owned_range_indices() and locally_owned_domain_indices() relying on the information of the TrilinosWrappers::MPI::Matrix object it is based on.
(Daniel Arndt, 2017/05/09)
New: ConstraintMatrix::distribute_local_to_global can now also assemble to rectangular matrices where rows and columns are described by different ConstraintMatrix objects.
(Martin Kronbichler, 2017/05/04)
Changed: The PETScWrappers::MPI::Vector class always corresponds to a PETSc vector with type mpi
, even when empty.
(David Wells, 2017/03/29)
New: TriaAccessor::enclosing_ball() computes and return a pair of Point and double corresponding to the center and the radius of a reasonably small enclosing ball of the TriaAccessor object.
(Vishal Boddu, Denis Davydov 2017/04/26)
Fixed: Copying Patterns::List or Patterns::Map objects previously led to memory corruption. This is now fixed.
(Wolfgang Bangerth, 2017/04/25)
New: Add Point<dim,Number>::distance_square() which calculates the squared Euclidean distance.
(Denis Davydov, 2017/04/24)
New: The ParameterHandler::parse_input() function and friends now make a guarantee that if a parameter value does not match its pattern, or an associated action throws an exception, that the enclosing ParameterHandler object will be reset to the same subsection level it was in before ParameterHandler::parse_input() was called.
(Wolfgang Bangerth, 2017/04/21)
Changed: We no longer support Visual Studio 2013 because it lacks important c++11 features like constexpr. It is now possible to use MSVC 2017 in addition to MSVC 2015.
(Timo Heister, 2017/04/20)
New: The new function ParameterHandler::add_action() allows to register actions that should be performed when a parameter is read from somewhere. This allows, in particular, initializing member variables that store parameter values without having to explicitly call ParameterHandler::get(), ParameterHandler::get_integer(), or similar functions.
(Wolfgang Bangerth, 2017/04/16)
Fixed: The MeshWorker framework can also be used in case dim != spacedim.
(Daniel Arndt, 2017/04/16)
Changed: The PointValueHistory class still used the old, deprecated interface for DataPostprocessor objects in that variant of the PointValueHistory::evaluate_field() function that takes such an object. This is now fixed: it uses DataPostprocessor::evaluate_scalar_field() and DataPostprocessor::evaluate_vector_field(), like all other users of the DataPostprocessor class.
(Wolfgang Bangerth, 2017/04/16)
Fixed: FETools::extrapolate also works for BlockVector types.
(Daniel Arndt, 2017/04/13)
Fixed: When initializing a LinearAlgebra::distributed::Vector and omitting zeroing the entries, ghost entries were left undefined, leading to use of invalid memory when calling compress() without a previous zero_out_ghosts() also when the local range was completely valid. This is now fixed.
(Martin Kronbichler, 2017/04/12)
New: Augment python interface of Point: add operators to modify point, a function to compute the distance between points, and functions to compute norms.
(Bruno Turcksin, 2017/04/11)
New: There is now a function TableHandler::declare_entry() that creates a column of a table without actually putting a value into it.
(Wolfgang Bangerth, 2017/04/11)
New: There is now a function TableHandler::start_new_row() that fills all entries of the current row (if any) that hadn't been filled, and thus starts a new row of the table. interface.
(Wolfgang Bangerth, 2017/04/09)
New: The class FESystem now implements the FiniteElement::convert_generalized_support_point_values_to_dof_values() interface.
(Wolfgang Bangerth, Matthias Maier, 2017/08/31)
New: The classes FE_Q, FE_DGQ, and FE_DGQArbitraryNodes now implement the FiniteElement::convert_generalized_support_point_values_to_dof_values() interface.
(Wolfgang Bangerth, 2017/04/05)
New: The FEEvaluation::read_dof_values and FEEvaluation::distribute_local_to_global now use dedicate gather (AVX2, AVX512) and scatter (AVX512) instructions for faster vector access if those are available.
(Martin Kronbichler, 2017/03/24)
New: The tensor product kernels used by FEEvaluation have gained a faster variant for evaluating gradients derived from spectral identities. This improves performance of the Laplace operator evaluation by 1020% in the computationbound case.
(Martin Kronbichler, 2017/03/24)
New: DoFTools::extract_dofs_with_support_contained_within() returns a set of degrees of freedom whose support is entirely contained within the cells for which the predicate returns true.
(Denis Davydov, 2017/03/24)
New: GridTools::compute_bounding_box() computes a bounding box of a subdomain whose active cells conform to a given predicate. GridTools::compute_active_cell_layer_within_distance() computes a collection of active cells that are within a given distance from predicate subdomain. GridTools::compute_ghost_cell_layer_within_distance() computes a collection of active cells that are within a given distance from locally owned active cells.
(Vishal Boddu, Denis Davydov 2017/03/23)
Extend: The two overloaded functions GridTools::find_active_cell_around_point() now take an optional custom mask for vertices to narrow down the search for surrounding cells.
(Vishal Boddu, 2017/03/22)