Changes between Version 6.2 and 6.3

This is the list of changes made between the deal.II releases listed above. made to the three sub-libraries base, lac, and deal.II, as well as changes to the general infrastructure, documentation, etc.

All entries are signed with the names of the author. Regular contributor's names are abbreviated by WB (Wolfgang Bangerth), GK (Guido Kanschat), RH (Ralf Hartmann).

Incompatibilities

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. Removed: The class TrilinosWrappers::SolverBlock and TrilinosWrappers::PreconditionBlock have been deleted for compatibility reasons with Trilinos 10.2. These solvers can be exchanged by deal.II's own iterative solvers with block matrices and vectors without loss in performance.
   (Martin Kronbichler 2010/06/24)

2. Changed: The SparseMatrix::operator() used to always check that an entry exists when accessing it. If it didn't, it would throw an exception that could be caught in a user program. Since these accesses are very frequent, this check now only happens in debug mode when the program aborts if a nonexistent element is accessed. If you access a nonexistent element in optimized mode, anything might happen (as in many other functions if input arguments do not satisfy their constraints).
   (WB 2010/06/04)

3. Removed: The interface to PETSc has been simplified to better handle incremental changes in PETSc versions and accommodate changes in functionality between versions. As a part of this process, the deal.II configure script no longer handles PETSc versions <2.3.0. Attempting to configure deal.II with PETSc versions that are not supported will result in the error message, "Unknown PETSc version". The usage of the PETScWrappers are otherwise not affected by this change.
   (Toby D. Young 2010/03/06)

4. Removed: The class TrilinosWrappers::PreconditionStokes has been deleted because it did not work properly, and it is too specific to be part of the general library. A preconditioner that has the same properties is explained in the The step-31 tutorial program step-31 tutorial program.
   (Martin Kronbichler 2009/11/23)

5. Changed: The class MGSmootherRelaxation now instead of a preconditioner takes a relaxation method with the functions step and Tstep. These perform a complete relaxation step, thus saving an auxiliary vector and computational effort for Gauss-Seidel type methods.

   While all relaxation preconditioners have been provided with the new two functions, you may have used MGSmootherRelaxation with your own preconditioner. In that case, you have two options:

   1. Configure deal.II with –enable-mgcompatibility, which restores the old behavior of MGSmootherRelaxation.
   2. Use MGSmootherPrecondition, which does what MGSmootherRelaxation did before.

   
   (GK 2009/08/04)

General

1. New: The new tutorial program step-45, contributed by Markus Bürg, shows how to implement periodic boundary conditions.
   (Markus Bürg, 2010/06/23)

2. Improved: Exception classes declared locally to another class using DeclException0 through DeclException5 were previously shown as local classes in the class overview of the library. This was annoying since they were not really of interest but made up most of the list. The documentation now shows them in one central place in the Exceptions and assertions module.
   (WB 2010/04/26)

3. Improved: We now compile all files with their full path name on the command line. This makes it simpler for tools like debuggers or profilers (e.g. valgrind) to find the source files that corresponds to an executable.
   (WB 2010/04/15)

4. Improved: Over the last few months, the multigrid implementation has seen significant rewrites, with much of the work done by Bärbel Janssen. The goal — now achieved — was to finally fully support multigrid also for continuous finite elements on adaptively refined meshes (uniformly refined meshes and discontinuous elements have worked for a long time). As part of this process, step-16 has been rewritten and now solves the same problem step-6 solves, just with a multigrid solver.
   (Bärbel Janssen, WB 2010/02/13)

5. New: The version of boost included in the contrib/ directory has been updated to 1.41.0.
   (WB 2009/12/10)

6. New: There is now a new tutorial program, step-35, contributed by Abner Salgado-Gonzalez, that implements a solver for the Navier-Stokes equations using a decoupled projection scheme.
   (Abner Salgado-Gonzalez 2009/10/07)

7. New: A report has been added on the codimension one capabilities of the library (by Antonio DeSimone, Luca Heltai and Cataldo Manigrasso, SISSA, Trieste, Italy). It explains in detail how to use the library for the solution of problems defined on codimension one manifolds, such as, for example, Boundary Element Methods.
   (Luca Heltai, 2009/09/23)

8. New: The configure switch –with-cpu=... now allows the value native, indicating that we would like the compiler to figure out which CPU we are running on and optimize for it. The resulting libraries may not work on other (previous generation) processors, however.
   (WB 2009/08/20)

9. Fixed: step-31 had a bug in the computation of the global scaling parameter in the function that evaluates the artificial viscosity: we computed \(c(\mathbf{u},T) = c_R\ \|\mathbf{u}\|_{L^\infty(\Omega)} \ \mathrm{var}(T) \frac{1}{|\mathrm{diam}(\Omega)|^{\alpha-2}}\) when it should have been \(c(\mathbf{u},T) = c_R\ \|\mathbf{u}\|_{L^\infty(\Omega)} \ \mathrm{var}(T) \ |\mathrm{diam}(\Omega)|^{\alpha-2}\). This didn't matter much in this program because \(\mathrm{diam}(\Omega)=2^{1/\textrm{dim}}\) and so is close to one. It would matter, however, if the domain had been different, as it is, for example, in the future step 32.
   (WB 2009/08/19)

10. Changed: When using Trilinos wrapper objects in parallel through MPI, each object now uses a separate and distinct MPI communicator object. This ensures that different objects (such as different matrices, or different vectors) communicate on separate channels, thereby simplifying debugging and possibly the parallelization of programs.
    (WB 2009/08/10)

11. New: There is now a new tutorial program, step-36, contributed by Toby D. Young and Wolfgang Bangerth, that demonstrates solving eigenvalue problems.
    (Toby D. Young, WB 2009/07/29)

12. Changed: When configuring to use METIS for partitioning meshes in parallel, the METIS header files had to be modified by hand. In addition, with some MPI implementations one would get into trouble if mpi.h included mpicxx.h. These two problems have now been worked around.
    (WB 2009/07/06)

13. New: As a primary means of parallelizing programs, deal.II now uses a task-based, rather than thread-based approach, in which one uses a high-level description of what needs to be done, rather than how these jobs have to be mapped onto threads. We then use the Threading Building Blocks (TBB) library to schedule tasks onto available hardware resources. This new scheme of describing parallelism and various abstractions to make programming in this framework easier are described in great detail in the Parallel computing with multiple processors module. In addition, most of the parallelism already used within deal.II has been converted to use tasks, rather than threads, and so have some of the tutorial programs.
    (WB 2009/01/09)

14. Changed: The support for threading has been completely re-written. In particular, the Threads::spawn functions have been deprecated, and new functions Threads::new_thread have been introduced. Threading is now discussed in a lot of detail in the Parallel computing with multiple processors module.
    (WB 2009/01/09)

15. Changed: Previously, one had to give the two flags –enable-multithreading –with-multithreading to ./configure to enable thread usage throughout the library. This has now been simplified: only the flag –enable-threads is now necessary. Furthermore, since most current machines have multiple cores these days, the default is now to use threads. This can be switched off using –disable-threads, however.
    (WB 2008/09/29)

base

1. New: The Timer class can now accumulate and average run times of pieces of code across multiple MPI processes.
   (Timo Heister 2010/06/07)

2. New: The Utilities::System::compute_point_to_point_communication_pattern function can be used to compute who wants to send messages to the current processor in unstructured point-to-point MPI communications.
   (WB 2010/06/07)

3. New: The DataOutBase class (and all derived classes such as DataOut, MatrixOut, etc) can now produce the XML-based version of the VTK file format (the so-called VTU format). Furthermore, the DataOutInterfaces::write_pvtu_record function can be used to describe a set of parallel VTU files as part of a single visualization set.
   (Scott Miller 2010/06/01)

4. Changed: The Function::vector_gradient_list function was previously implemented by calling Function::gradient on each point and each component. It has been changed to now call Function::vector_gradient on each point only, and derived classes should implement this function accordingly.
   (WB 2010/02/10)

5. Fixed: The file data_out_base.cc could not be compiled when Tecplot was available. This should now be fixed.
   (WB 2010/02/10)

6. Fixed: The PolynomialsBDM, PolynomialsABF and Functions::FlowFunction classes had a race condition in multithreaded programs. This now fixed.
   (WB 2010/01/27)

7. New: The new class IndexSet can represent sets and ranges of indices.
   (WB 2009/10/09)

8. New: There is now an operator << for the TableIndices class.
   (WB 2009/09/24)

9. Fixed: If anything had been put into a LogStream object without flushing it with std::endl before the destruction of the log stream, it was lost. This is now fixed.
   (WB 2009/09/23)

10. New: SymmetricTensor::component_to_unrolled_index() and SymmetricTensor::unrolled_to_component_indices() allow to convert between the indices of an element of a symmetric tensor and the index within an unrolled vector.
    (WB 2009/09/23)

11. New: classes NamedData and NamedSelection provide an interface to store and retrieve data objects with name identifiers.
    (GK 2009/09/13)

12. New: The Utilities::System::job_supports_mpi() can be used to query whether the current job runs under MPI or not.
    (WB 2009/08/14)

13. New: The Utilities::Trilinos::comm_self function return an MPI communicator that consists only of the current processor.
    (WB 2009/08/07)

14. New: The Utilities::Trilinos::duplicate_communicator function allows to duplicate an Epetra_Comm object to get a unique parallel MPI communicator out of an existing one. Utilities::Trilinos::duplicate_map creates a map that has the same members as the given template but uses a separate communicator.
    (WB 2009/08/06)

15. New: There is now a specialization Tensor<0,dim> of tensors of rank 0. Since rank-0 tensors are scalars, this class essentially acts like a scalar, but it allows for some neat template tricks that involve tensors of arbitrary rank.
    (WB 2009/07/15)

16. New: The GeometryInfo::alternating_form_at_vertices can be used to investigate the degree of distortion of cells.
    (WB 2009/06/28)

17. New: The GeometryInfo::d_linear_shape_function and GeometryInfo::d_linear_shape_function_gradient functions can be used to represent the \(d\)-linear shape functions that are frequently used to map the reference cell to real cells (though the Mapping class hierarchy also allows to use higher order mappings).
    (WB 2009/06/28)

18. New: The determinant() function is now implemented for rank-2 Tensor arguments of all sizes. The implementation is not efficient for very large matrix sizes, however.
    (WB 2009/06/28)

19. Improved: The QGaussLobatto::gamma function now returns a long double instead of an unsigned int, otherwise we will get an overflow and thus meaningless weights for higher QGaussLobatto quadrature rules.
    (Tobias Leicht, RH 2009/06/05)

20. New: The new function Utilities::duplicate_communicator can be used to duplicate an MPI communicator to produce a unique duplicate.
    (WB 2009/05/13)

lac

1. New: The ConstraintMatrix class can now handle storing only a subset of all constraints, for example only for degrees of freedom that are relevant for the subdomain that is owned by one process in an MPI universe.
   (Timo Heister, Martin Kronbichler 2010/06/07)

2. New: The PETScWrappers::MPI::Vector and TrilinosWrappers::MPI::Vector classes can now handle ghost elements, i.e. elements that are not owned by the current processor but are available for reading anyway. The simplest form of ghosting would be to simply import an entire vector to local memory, but the new function allow to select the elements we need to support the case of computations where importing all elements of even a single vector would exceed available memory.
   (Timo Heister 2010/06/07)

3. New: A class SparseDirectMumps that provides an interface to the MUltifrontal Massively Parallel sparse direct Solver (MUMPS).

   
   (Markus Buerg 2010/05/10)

4. Fixed: BlockSparsityPattern::copy_from accidentally only copied n_block_rows times n_block_rows blocks, instead of n_block_rows times n_block_cols. This is now fixed.

   
   (WB 2010/01/06)

5. Fixed: SparsityPattern::copy_from crashed whenever a compressed sparsity pattern was copied that had either zero rows or zero columns. This is now fixed.

   
   (WB 2010/01/06)

6. New: The function Householder::least_squares can handle BlockVectors as well now. Note that in one place we still have to copy to a Vector to use the function backward from FullMatrix.

   
   (Bärbel Janssen 2010/01/05)

7. New: There are now two ConstraintMatrix::add_lines functions that can add several constraints at once.

   
   (WB 2010/01/05)

8. Improved: The Vector class has been equipped with an improved way to calculate sums in inner products and norms. This reduces the accumulation of round-off errors. Especially the solution with float vectors should profit from the new implementation.

   
   (Martin Kronbichler 2009/11/05)

9. Improved: The ConstraintMatrix class now uses a cache for random access to the constraint lines. This considerably increases performance of the _local_to_global functions, where such an access pattern is usual. Moreover, the ConstraintMatrix class has now a function get_dof_values that can import data from a global vector to a cell vector with respecting the constraints.

   
   (Martin Kronbichler 2009/09/30)

10. Fixed: SparsityTools::reorder_Cuthill_McKee would produce an error if the input graph had disconnected components. This is now fixed.

    
    (WB 2009/09/25)

11. Fixed: When using the TrilinosWrappers::MPI::Vector::reinit() function with a parallel vector, and if the vector initialized and the vector given had a local range on one of the processors that exactly matched, the program would freeze if the local ranges on the other processors did not also match exactly. This is now fixed.

    
    (WB 2009/09/02)

12. Improved: BlockVector and several of the block sparsity patterns can now be initialized with BlockIndices objects. Therefore, if an application needs such an object, it does not have to store a vector of block sizes separately.
    (GK 2009/08/26)

13. New: The class PreconditionChebyshev implements a preconditioner based on Chebyshev polynomials. It is based on matrix-vector products together with some vector updates.
    (Martin Kronbichler 2009/08/25)

14. Fixed: Crash or strange behaviour (wrong matrix entries written) in PETScWrappers::MPI::BlockSparseMatrix when adding or setting elements through any of the set() and add() routines. This happened when different CPUs access different blocks at the start of assembly or when switching between adding and setting.
    (Timo Heister 2009/08/05)

15. New: The relaxation preconditioners PreconditionJacobi, PreconditionSOR and PreconditionSSOR, as well as their blocked versions PreconditionBlockJacobi, PreconditionBlockSOR and PreconditionBlockSSOR now have functions step and Tstep performing one complete step of these methods.
    (GK 2009/08/04)

16. New: There are new functions FullMatrix::cholesky and FullMatrix::outer_product. FullMatrix::cholesky finds the Cholesky decomposition of a matrix in lower triangular form. FullMatrix::outer_product calculates *this \(= VW^T\) where \(V\) and \(W\) are vectors.
    (Jean Marie Linhart 2009/07/27)

17. Fixed: The TrilinosWrappers::MPI::BlockVector class declares an assignment operator from the non-Trilinos BlockVector class but it could not be compiled due to an oversight. This is now fixed.
    (WB 2009/06/29)

18. New: Based on work by Francisco Alvaro, the existing SLEPcWrappers now have a handle on the generalized eigenvalue problem where B=I.
    (Toby D. Young 2009/06/25)

19. New: Based on work with Francisco Alvaro and Jose E. Roman, the new SLEPcWrappers give a handle on some of the features of SLEPc (Scalable Library for Eigenvalue Problem Computations): (1) The SLEPcWrappers::SolverBase class can be used for specifying an eigenvalue problem, either in standard or generalized form, on serial or parallel architectures with support for a few solver types; and (2) The SLEPcWrappers::TransformationBase class encapsulates a variety of spectral transformations providing some functionality required for acceleration techniques based on the transformation of the spectrum.
    (Toby D. Young 2009/06/25)

20. Fixed: The TrilinosWrappers::BlockVector class declares an assignment operator from the non-Trilinos BlockVector class but it wasn't implemented. This is now fixed.
    (WB 2009/06/24)

21. New: The SparseMatrix class has now a function SparseMatrix::mmult that can multiply two sparse matrices with each other.
    (Martin Kronbichler 2009/05/04)

deal.II

1. New: The namespace MeshWorker contains a generic MeshWorker::loop() over all cells and faces as well as auxiliary classes, which allow to program integrals over mesh cells and faces in a very generic and local way. In particular, an application programmer will not have to distinguish between regular faces and faces with hanging nodes anymore. Two tutorial programs (step-12 and step-39) highlight the functionality of this framework.
   (GK 2010/06/24)

2. New: The FE_Q_Hierarchical class now has functions FE_Q_Hierarchical::hp_constraints_are_implemented and FE_Q_Hierarchical::hp_vertex_dof_identities.
   (Markus Bürg 2010/06/08)

3. New: The FEValuesViews::Vector class now has functions FEValuesViews::Vector::curl and FEValuesViews::Vector::get_function_curls.
   (Markus Bürg 2010/05/13)

4. New: TriaAccessor::extent_in_direction() returns the length of an object in a given direction.
   (James Avery 2010/05/10)

5. New: There is a new function DoFTools::extract_dofs_with_support_on_boundary().
   (WB 2010/05/07)

6. Fixed: FE_DGQ::has_support_on_face() returned the wrong value in 1d if the polynomial degree of the finite element equals zero (i.e. for piecewise constants) where the lone shape function is nonzero on all faces. This is now fixed.
   (WB 2010/05/07)

7. Fixed: VectorTools::interpolate_boundary_values inadvertently produces an exception when used with hp::DoFHandler objects in 1d. This is now fixed.
   (WB 2010/05/04)

8. Fixed: The GridIn::read_msh function got into trouble if the mesh file had a section that listed physical names for variables. This is now fixed.
   (WB 2010/05/03)

9. Improved: DoFHandler iterators now can be assigned from a Triangulation iterator after the dof handler was set once.
   (GK 2010/03/25)

10. New: The function DoFRenumbering::downstream has now an additional bool argument. If enabled, the downstream comparison is performed on a DoF basis, as opposed to the cell-based comparison that is used with a false argument.
    (Martin Kronbichler 2010/03/19)

11. Improved: DoFHandler now has a BlockInfo object, automatically updated after DoFHandler::distribute_dofs() and accessible by DoFHandler::block_info(). This object can be used to initialize block vectors and obliterates the necessity to count dofs per block (or dofs per component in legacy code) in application programs.
    (GK 2010/03/18)

12. New: The class MGTransferSelect is prepared for use on adaptively refined meshes.
    (Bärbel Janssen 2010/02/05)

13. New: The function Cuthill_McKee in namespace DoFRenumbering is now also compiled for MGDoFHandler as well as the make_sparsity_pattern functions in DoFTools.
    (Bärbel Janssen 2010/01/08)

14. New: Constructor of FESystem now also exists for four base elements.
    (Thomas Wick 2010/01/08)

15. New: The functions in namespace DoFRenumbering::boost are now also compiled for hp::DoFHandler arguments.
    (WB 2009/12/14)

16. Fixed: The functions DoFTools::count_dofs_per_component and DoFTools::extract_dofs produced wrong results for elements that consist of two or more nested FESystems. Moreoever, a bug in DoFTools::extract_constant_modes has been corrected and DoFTools::distribute_cell_to_dof_vector now works according to the documentation, namely leaving unselected components in the result vector unchanged, instead of setting these to zero as was done before.
    (Martin Kronbichler 2009/12/14)

17. Fixed: The function Triangulation::n_levels() accidentally turned out to be quite expensive, in particular if the mesh has been coarsened significantly in the past. Since Triangulation::n_active_cells() calls Triangulation::n_levels() repeatedly, by consequence it also is surprisingly expensive, which is particularly annoying since this is a frequently called function. This has now been fixed by caching the result of both functions.
    (Wolfgang Bangerth 2009/11/22)

18. Fixed: The function DataOut class got confused when its DataOut::first_cell() and DataOut::next_cell() were overloaded and cell data was given. This is now fixed.
    (Wolfgang Bangerth 2009/11/05)

19. New: The function DoFRenumbering::component_wise is now also implemented for arguments of type hp::DoFHandler.
    (Markus Bürg 2009/10/01)

20. New: The class BlockInfo stores and computes all BlockIndices objects related to DoFHandler and MGDoFHandler based on FESystem.
    (GK 2009/09/13)

21. Improved: MGDoFHandler now also has a typedef for MGDoFHandler::Container
    (GK 2009/09/13)

22. FETools::compute_block_renumbering() can nor return block sizes instead of start indices.
    (GK 2009/08/26)

23. New: The function GridGenerator::truncated_cone() and the class ConeBoundary can now be used to describe conical objects.
    (Markus Bürg 2009/08/17)

24. New: Instead of asking face or edge iterators for their boundary indicator using TriaAccessor::boundary_indicator() and then the triangulation for the boundary object using Triangulation::get_boundary(), you can now directly ask the iterator for the boundary object using TriaAccessor::get_boundary().
    (WB 2009/07/31)

25. Fixed: The CellAccessor::recursively_set_material_id function did not set the material id for all children, but only for the first two, which is obviously a bug. This should now be fixed.
    (WB 2009/07/14)

26. Fixed: The GridIn class sometimes had problems with input files that had whitespace at the end of lines. This should now be fixed.
    (WB 2009/07/10)

27. New: Previously, the Triangulation::create_triangulation function silently accepted input meshes with inverted cells (i.e. cells with a zero or negative determinant of the Jacobian of the mapping from the reference cell to the real cell). This can been changed now: By passing the appropriate flag to the constructor of the Triangulation class, the Triangulation::create_triangulation function checks whether cells are distorted or inverted, and may throw an exception containing a list of cells for which this is the case. If you know that this is harmless, for example if you have cells with collapsed vertices in your mesh but you do not intend to integrate on them, then you can catch and ignore this message. In all other cases, the output of your computations are likely to be wrong anyway.
    The same is true for the Triangulation::execute_coarsening_and_refinement function: if it creates cells that are distorted, it throws a list of cells whose children are distorted.
    The whole issue is described in some detail in the entry on distorted cells in the glossary.
    (WB 2009/06/29)

28. New: The new hp::DoFHandler::set_active_fe_indices function allows to distribute all active FE indices at once based on a given vector. This might be useful if this information is stored somewhere and has to be reconstructed or else if two DoFHandler objects with the same FE index distribution should be created. There is now also a corresponding hp::DoFHandler::get_active_fe_indices function.
    (Tobias Leicht, RH 2009/06/12)

29. Fixed: The projection of quadrature points to subfaces in MappingQ in case of 3d anisotropic refinement did not respect non-standard face orientation/flip/rotation cases. This has now been fixed.
    (Tobias Leicht, RH 2009/06/12)

30. New: The new Triangulation::n_raw_faces() function forwards to Triangulation::n_raw_lines() in 2d and Triangulation::n_raw_quads() in 3d.
    (Tobias Leicht, RH 2009/06/12)

31. New: There is now a new DataOutFaces::build_patches function which takes a Mapping argument. For higher order mappings this allows to represent curved boundaries by using more subdivisions. This function is also useful in the context of MappingQ1Eulerian.
    (Tobias Leicht, RH 2009/06/05)

32. New: For empty triangulations the new Triangulation::set_mesh_smoothing function allows to override the MeshSmoothing given to the constructor.
    (RH 2009/06/05)

33. New: The new function TriaAccessor::is_translation_of computes whether a cell, face, or edge is a translation of another.
    (Martin Kronbichler, WB 2009/05/19)

34. New: The DoFTools::make_sparsity_pattern functions have acquired a new parameter subdomain_id. If a value other than the default value is passed for it, the function only assembles the sparsity pattern on those cells that have the given subdomain id. This is useful, for example, in conjunction with the TrilinosWrappers::SparsityPattern class that can hold a sparsity pattern distributed across several MPI processes; in that case, it is not necessary that each process builds the entire sparsity pattern.
    (WB 2009/04/29)

35. Fixed: The DoFRenumbering::component_wise function for MGDoFHandler objects did a few things in parallel that weren't thread-safe. This is now fixed.
    (WB, 2009/01/20)

36. Changed: The two DataOut::build_patches, DataOutFaces::build_patches, and DataOutRotation::build_patches functions have lost the argument that indicated the number of threads with which they should build the intermediate representation. This is something that now happens transparently in the background and doesn't need caller input any more.
    (WB 2008/12/16)

37. Changed: The KellyErrorEstimator::estimate functions had a parameter that indicates the number of threads to be used in the computation. This parameter continues to exist for compatibility, but is now ignored. Rather, the number of threads is determined automatically by scheduling the requested computations on available compute resources.
    (WB, 2008/12/29)

38. New: The new function internal::hp::FEValuesBase::get_fe_collection function allows to query the finite element collection currently in used in an hp::FEValues, hp::FEFaceValues, or hp::FESubfaceValues object.
    (WB 2008/09/30)

39. New: The new function FEValuesBase::get_update_flags allows to query the update flags that are currently set on an FEValues, FEFaceValues, or FESubfaceValues object.
    (WB 2008/09/29)