Reference documentation for deal.II version Git 1448398 2018-02-21 21:14:18 -0500
Changes between Version 6.1 and 6.2

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. Changed: The implementation of the reordering algorithms in the boost namespace created multiple edges between two nodes. The new implementation submits not more than one edge, but that affects the ordering of the degrees of freedom and deviates from previous behavior.
(Martin Kronbichler 2009/04/22)

2. Changed: The files implementing the class ConstraintMatrix are now saved in the lac/ subdirectory, named constraint_matrix.h and included by the line #include <lac/constraint_matrix.h>. After all, constraints are just algebraic objects and do not depend on anything in the deal.II directory (in particular, they are dimension-independent). The old #include <dofs/dof_constraints.h> include line is still enabled by a reference to the new file, but users should try to adapt their codes to the new file names as the reference might be removed in the future.
(Martin Kronbichler 2009/03/04)

3. Changed: The function DoFTools::get_subdomain_association function used to assign degrees of freedom to the subdomain of the last cell on which the degree of freedom is a part. This introduced a bias for degrees of freedom's subdomains located on boundaries of subdomains, and consequently to unequal numbers of DoFs per subdomain even if the number of cells is roughly equal across subdomains. This behavior has been changed by assigning degrees of freedom pseudo-randomly to any of the subdomains on which they are located. This is a deviation from previous behavior, however.
(Timo Heister, WB 2008/11/02)

4. Changed: The way we set up threads in the Threads::spawn functions and friends has been completely written, using the boost::bind and boost::function libraries. This has made things significantly simpler and allowed us to remove some 4,100 lines of code. The only user-visible side effect is that you can now no longer spawn functions with 10 arguments (this was the maximum before) whereas 9 arguments continues to work; the reason for this is a limitation in the boost::bind library. This limit will be lifted with the next C++ standard, however, using variadic templates.
(WB 2008/10/10)

5. Changed: The SolutionTransfer class used to take a type as second template argument that denoted the scalar upon which Vector objects were built, in order to allow interpolating Vector<float> objects, for example. This argument has now been changed to a vector type, and been given a default of Vector<double>; however, one can now also pass in BlockVector objects, or objects of type PETScWrappers::Vector, etc. On the downside, the old SolutionTransfer::refine_interpolate function with only a single argument has been deleted since there is no reliable way to resize a vector unless it is a plain Vector<double>.
(WB 2008/08/28)

6. Changed: The FiniteElement::get_prolongation_matrix and FiniteElement::get_restriction_matrix functions now have an additional argument of type RefinementCase to enable a meaningful use of these matrices in case of anisotropic refinement. These matrices are used for solution transfer and multigrid operations, but there should be no need to use them in user codes directly. Note, that the matrices are not fully implemented for all finite elements up to now.
(Tobias Leicht 2008/07/08 as of branch_anisotropic at 2006/07/04)

7. Changed: GeometryInfo::children_per_cell has been replaced by GeometryInfo::max_children_per_cell, which represents the maximum number of children a cell might have, i.e. the number of children in the case of isotropic refinement. But note, that this number will rarely be needed in user codes. In general, the number of children of a cell varies from cell to cell and can be obtained by cell->n_children(), which gives the number of children of a specific cell which is refined iso- or anisotropically.
(RH 2008/07/08 as of branch_anisotropic at 2005/02/26)

8. Changed: The analogous change applies to the number of children of a face. GeometryInfo::subfaces_per_face has been replaced by GeometryInfo::max_children_per_face, which represents the maximum number of children a face might have, i.e. the number of children in the case of an isotropic refined face. But note, that this number will rarely be needed in user codes. In general, the number of children of a face varies from face to face and can be obtained by face->n_children(), which gives the number of children of a specific face which is refined iso- or anisotropically.
(RH 2008/07/08 as of branch_anisotropic at 2005/02/26)

9. Changed: The GeometryInfo::child_cell_on_face function is generalized to anisotropic refinement and has now an additional RefinementCase argument. This function will rarely be used in user codes, as e.g. the neighbor's child at a specific face and subface should be accessed through cell->neighbor_child_on_subface rather than through the GeometryInfo::child_cell_on_face function.
(RH 2008/07/08 as of branch_anisotropic at 2005/02/27)

### General

1. Updated: The step-22 tutorial program now uses some advanced features of the ConstraintMatrix such as imhomogeneities for implementing Dirichlet boundary condition and condensation of constraints on the fly.
(Martin Kronbichler 2009/04/20)

2. New: There is now a program, contributed by Jean-Paul Pelteret, that can convert Cubit generated meshes into UCD format that the GridIn class can read. The program along with a description is in the contrib/mesh_conversion directory.
(Jean-Paul Pelteret 2009/04/16)

3. Updated: Configure mechanism and a few PETSc functions in the lac directory have been upgraded to support migration to PETSc version >=3.0.0 in anaology to the existing interface.
(Toby D. Young 2009/03/31)

4. New: A new tutorial program step-34 was added to the library that shows the usage of the new codimension one functionality recently added to the library. In this tutorial we show the use of boundary element methods on piecewise constant functions defined over a surface, and we solve the irrotational flow problem, or exterior Neumann Laplace problem.
(Luca Heltai 2009/03/10)

5. Fixed: When linking with both Trilinos and BLAS, the BLAS libraries appeared first on the command line; however, if the Trilinos libraries required BLAS then this was the wrong order. This is now fixed.
(WB 2009/02/11)

6. Changed: The subversion repository for deal.II development is now located on commercially hosted space at http://www.dealii.org/svn/dealii/.
(GK 2009/01/20)

7. Changed: Some parts of the library used to use classes and functions from the BOOST library. Since many of the components of BOOST have been voted into what will be the next C++ standard, we now use a namespace std_cxx1x (coined on the provisional name C++0x used for the next C++ standard) into which we import BOOST components as necessary. If a compiler supports C++0x, we can then later replace BOOST components by elements from namespace std in one central place, rather than all throughout the code.
(WB 2009/01/12)

8. Updated: The step-33 tutorial program now uses the Trilinos wrapper classes for accessing Trilinos linear algebra routines instead of implementing that by hand.
(Martin Kronbichler 2009/01/07)

9. Updated: The version of the BOOST library that is in the contrib/ directory and is used in various places of the library has been upgraded to 1.37.
(WB 2008/12/07)

10. New: Based on work by Cataldo Manigrasso and Luca Heltai developed over the last few months on a branch, deal.II now supports meshes that are embedded in higher space dimensions. This allows, for example, to solve equations on the surface of a sphere (e.g. to use boundary element methods for exterior problems).
In practice, this is implemented by giving a large number of classes a second template argument. Whereas the first one, typically named dim still denotes the dimensionality of the triangulation (e.g. dim=2 for triangulations of quadrilaterals, or dim=3 for those made up of hexahedra), the second template argument spacedim denotes the dimensionality of the space this all is embedded in. As a consequence, if one wants to have a triangulation of a two-dimensional manifold in three-dimensional space, one would use dim=2, spacedim=3 as template arguments.
In order to remain backward compatible, the new second template argument always has a default value that equals dim meaning it is still sufficient to write Triangulation<2> if you intend two-dimensional triangulations of two-dimensional domains.
The new functionality is described in the step-34 tutorial program.
(Cataldo Manigrasso, Luca Heltai 2008/12/05)

11. Changed: The various accessor classes have been rewritten entirely. Where previously we had these accessor classes as separate specializations for 1d, 2d, and 3d, each implementing the same set of functions, these classes have now all been merged into common templates TriaAccessor, DoFAccessor, and MGDoFAccessor, along with the existing CellAccessor, DoFCellAccessor and MGDoFCellAccessor. As a result, the documentation of individual functions should now be much easier to find.
(WB 2008/12/05)

12. New: In analogy to the existing interfaces to the PETSc library, there are now also interfaces to much of the linear algebra part of the Trilinos library.
(Martin Kronbichler 2008/12/01)

13. New: The deal.II configure script now autodetects SLEPc, an eigenvalue solver. Currently on SLEPc versions >=3.0.0 are supported.
(Toby D. Young 2008/11/27)

14. New: A new tutorial program, step-31, is contributed by Martin Kronbichler and Wolfgang Bangerth and shows the solution of the Boussinesq approximation to thermally driven convection in an incompressible fluid.
(Martin Kronbichler 2008/11/14)

15. Updated: In the step-22 tutorial program the generation of the sparsity pattern using the class BlockCompressedSetSparsityPattern has been replaced by a pattern of the class BlockCompressedSimpleSparsityPattern, which uses far less memory and is slightly faster.
(Martin Kronbichler, WB 2008/11/12)

16. New: The shared libraries we create are now versioned, i.e. they have the form libdeal_II_2d.g.so.6.2.0. The library without the version number is simply a link to the versioned library name.
(WB 2008/10/07)

17. Updated: The version of the BOOST library that is in the contrib/ directory and is used in various places of the library has been upgraded to 1.36.
(WB 2008/09/19)

18. Fixed: In the step-23 tutorial program the terms corresponding to the external force were computed incorrectly. The error wasn't visible right away since in the program the right hand side is assumed to be zero. This has now been fixed.
(Johan Lorentzon, WB 2008/09/12)

19. New: A new tutorial program, step-28, is contributed by Yaqi Wang. It illustrates the solution of a coupled system of diffusion equations relevant to nuclear reactor physics where we use different meshes for different components of a vector-valued solution.
(Yaqi Wang 2008/09/10)

20. New: Contributed by Tobias Leicht as well as Ralf Hartmann, deal.II now supports anisotropic refinement of meshes. This functionality is explained in the step-30 tutorial program.
(Tobias Leicht, RH 2008/07/08)

21. Fixed: In the new step-33 tutorial program there was a place where we incorrectly passed the diameter of a cell, rather than a face, to a function. This is now fixed.
(Chih-Che Chueh, WB 2008/06/28)

22. Fixed: The new step-33 tutorial program had a place where we didn't release some memory. This is now fixed.
(Chih-Che Chueh, WB 2008/06/02)

23. Fixed: A missing include file prevented the ./configure script from detecting the presence of the demangler with recent versions of the gcc compiler. The result is that backtraces after failed assertions only show the mangles function names, not their plain text equivalent. This is now fixed.
(WB 2008/05/27)

### base

1. New: There is now a new QGaussOneOverR class, that allows for integration on the two dimensional reference element of arbitrary polynomial functions with weight 1/R. This class is only instantiated for dim=2, and it is intended for use with collocation type boundary element methods of order 1, where the singularities are collocated on the vertices of the quadrilaterals.
(Luca Heltai 2009/03/11)

2. New: There is now a new QGaussLogR class, that generalizes the QGaussLog class to allow for arbitrary location of singularities, and singularity factors.
(Luca Heltai 2009/03/11)

3. New: The FunctionParser class now supports the fparser library's interface to use units (like cm, or km) in expressions. An example is given in the documentation of that class.
(Victor Prosolin 2009/03/01)

(WB 2009/01/14)

5. New: There is now a class TimerOutput that allows to neatly measure computing times in different sections of a program and write that information to screen.
(Martin Kronbichler 2009/01/12)

6. New: The Timer class can now also return the wall time using the method Timer::wall_time(), as opposed to the CPU time returned by Timer::operator ().
(Martin Kronbichler 2009/01/07)

7. New: The new class types_are_equal allows to write some templates more efficient by allowing to figure out whether certain template types are, for example, equal to double or float (in which case we can use BLAS functions, or could do something else special).
(WB 2008/10/31)

8. New: The Utilities::reverse_permutation and Utilities::invert_permutation compute the reverse and inverse of a given permutation of indices.
(WB 2008/10/31)

9. Fixed: The PolynomialsRaviartThomas class had a bug that led to random results when used from multiple threads. As a consequence the FE_RaviartThomas class was unusable in a multithreaded context. This has now been fixed.
(WB 2008/10/13)

10. New: There is a new function scalar_product(const Tensor<2,dim> &, const Tensor<2,dim> &) that computes the scalar product $$a:b=\sum_{i,j} a_{ij}b_{ij}$$ between two tensors of rank 2.
(WB 2008/08/15)

11. New: If the compiler allows to do #include <mpi.h>, then the preprocessor flag DEAL_II_COMPILER_SUPPORTS_MPI is now set in base/include/base/config.h. This also fixes a problem in base/include/base/utilities.h if a compiler capable of including mpi.h was used but not PETSc.
(WB 2008/08/15)

12. Fixed: A misplaced #include directive prevented the file base/source/data_out_base.cc from being compilable by the PGI C++ compiler. This is now fixed.
(WB 2008/08/05)

13. New: There are now a new GeometryInfo::min_cell_refinement_case_for_face_refinement (resp. GeometryInfo::min_cell_refinement_case_for_line_refinement) function which returns the RefinementCase representing the smallest refinement case of a cell for a given refinement of one of its faces (resp. lines). In 2D for example a cell has to be refined at least with RefinementCase::cut_y if the left line (line 0) shall be refined. Another refinement possibility for the cell would be RefinementCase::cut_xy, but that is not the minimal case.
(Tobias Leicht 2008/07/08 as of branch_anisotropic at 2006/06/28)

14. New: There is now a new GeometryInfo::line_refinement_case function which returns the RefinementCase representing the refinement case of a line for a given refinement case of the corresponding cell.
(Tobias Leicht 2008/07/08 as of branch_anisotropic at 2006/06/28)

15. New: The new GeometryInfo::n_children(refinement_case) function returns the number of children a cell/face has when refined with the RefinementCase refinement_case.
(RH 2008/07/08 as of branch_anisotropic at 2005/11/07)

16. New: Given a RefinementCase of a cell the new GeometryInfo::face_refinement_case function returns the RefinementCase of a face.
(RH 2008/07/08 as of branch_anisotropic at 2005/11/07)

17. New: There is now a new RefinementCase GeometryInfo::isotropic_refinement representing the isotropic refinement case in dim dimensions, i.e. GeometryInfo<1>::isotropic_refinement = RefinementCase::cut_x, GeometryInfo<2>::isotropic_refinement = RefinementCase::cut_xy and GeometryInfo<3>::isotropic_refinement = RefinementCase::cut_xyz.
(RH 2008/07/08 as of branch_anisotropic at 2005/03/03)

18. New: The class QGaussLog allows the quadrature of integrals with logarithmic kernels.
(Cataldo Manigrasso 2008/05/28)

### lac

1. Updated: The SparseMatrix::precondition_SSOR function was implemented inefficiently. The new implementation uses about half the instruction count and reduces execution times on typical applications by ten to thirty percent.
(Martin Kronbichler 2009/04/21)

2. Updated: The SparseILU::vmult kernel has been re-written to use similar data structures as SparseMatrix::vmult, which reduces the count of operations by one third and the execution times on typical applications by ten to twenty percent.
(Martin Kronbichler 2009/04/16)

3. New: There is now a new class VectorView<Number> that allows views of arbitrary areas of memory to be seen as a Vector<Number>, simplifying a lot of works for portability issues between different libraries, and allowing for subviews of Vector<Number> classes without the need to copy back and forth large chunk of memories.
To be used with EXTREME caution, and only when you know exactly what you are doing, and speed is a necessity.
(Luca Heltai 2009/04/05)

4. Updated: The local_to_global functions in ConstraintMatrix got smarter, which accelerates sparsity pattern generation and makes writing into sparse matrices using distribute_local_to_global faster.
(Martin Kronbichler 2009/03/16)

5. Fixed: The FullMatrix::swap_row and FullMatrix::swap_col functions had bugs that made them only work in case the matrix was square. This is now fixed.
(WB 2009/03/05)

6. New: Added a few simple helper functions (to VectorBase) that allow some manipulation of PETSc vectors. These functions do what they say in the documentation.
(Toby D. Young 2009/01/08)

7. New: There is now a class TrilinosWrappers::SparsityPattern that allows to construct distributed sparsity patterns that can be used for initializing Trilinos sparse matrices. This means that for large problems on several MPI processors the memory requirements are reduced and the initialization speed is increased.
(Martin Kronbichler 2008/12/29)

8. New: All SparseMatrix classes (SparseMatrix<number>, PETSc sparse matrices, Trilinos sparse matrices, block sparse matrices) can now directly add and set a FullMatrix and some other arrays into their value list. This is faster and more convenient than an element-by-element addition/set.
(Martin Kronbichler 2008/11/26)

9. New: The class LAPACKFullMatrix can now invert full matrices using the (optimized) LAPACK functions getrf and getri. The speedup over the FullMatrix::gauss_jordan() function is a factor of two for matrices with 100 rows and columns, and grows with matrix size.
(Martin Kronbichler 2008/11/11)

10. Fixed: The BlockMatrixBase::clear() function that is used by all other block matrix type classes had a memory leak in that the memory allocated by all sub-objects was not freed. This is now fixed.
(WB 2008/11/05)

11. New: The function SparsityTools::reorder_Cuthill_McKee reorders the nodes of a graph based on their connectivity to other nodes.
(WB 2008/10/31)

12. New: The function GridTools::get_face_connectivity_of_cells produces a sparsity pattern that describes the connectivity of cells of a triangulation based on whether they share common faces.
(WB 2008/10/31)

13. Changed: The function SparsityPattern::partition has been deprecated. It is now available in a new namespace SparsityTools that collects algorithms that work on sparsity patterns or connectivity graphs.
(WB 2008/10/31)

14. Fixed: Whereas the Vector class copy operator resized the left hand side operand whenever necessary, the corresponding operator of the BlockVector class did not. This is now fixed.
(Christian Cornelssen, WB 2008/10/28)

15. Changed: The SparseDirectUMFPACK class now calls the umfpack_dl_* routines instead of umfpack_di_*. On machines with 64-bit longs this allows the UMFPACK solver to allocate more than 2GB of memory for large problems.
(Moritz Allmaras 2008/10/16)

16. Improved: The SparseILU::initialize function, for some reason, required the second argument, SparseILU::AdditionalParameters, to be present even if it is a default constructed object. This argument now has a default value that equates to a default constructed object that can therefore be omitted when so desired.
(WB 2008/09/23)

17. New: Added the CompressedSimpleSparsityPattern as an alternative to CompressedSparsityPattern and CompressedSetSparsityPattern, which should be faster in most cases but its memory usage is somewhere inbetween.
(Timo Heister 2008/09/03)

18. Improved: The CompressedSparsityPattern can now elide some operations upon entering entries that may have been added before already.
(Timo Heister, WB 2008/08/28)

19. Fixed: There are now functions CompressedSparsityPattern::print and CompressedSetSparsityPattern::print, in analogy to SparsityPattern::print .
(WB 2008/07/31)

20. Fixed: The FullMatrix::copy_from(FullMatrix,bool) function had a bug that resulted in an exception being triggered whenever the given matrix had rows where the diagonal element is zero and the second parameter to the function was set to true. This is now fixed.
(WB 2008/07/30)

21. New: Added two new functions to calculate the left or the right inverse of a given rectangular matrix (FullMatrix::left_invert, FullMatrix::right_invert). They throw an exception if the matrices are not invertible.
(Cataldo Manigrasso 2008/6/12)

22. Fixed: FullMatrix::Tadd can now transpose correctly rectangular matrices, there was a mismatch about the indexes in the code.
(Cataldo Manigrasso 2008/6/12)

### deal.II

1. Fixed: VectorTools::integrate_difference now works also on codimension one grids.
(Luca Heltai 2009/04/10)

2. Changed: The ConstraintMatrix class can now also handle inhomogeneous constraints. This means that e.g. Dirichlet boundary conditions are now a special case of algebraic constraints. There are new condense functions that simultaneously act on matrices and vectors for use of this feature.
(Martin Kronbichler 2009/02/25)

3. Changed: The DoFTools::count_dofs_per_component and DoFTools::count_dofs_per_block erroneously resized the output argument to the number of components or blocks in the finite element, respectively, even if the target component/block list given as an additional argument needed a different number of output elements. This is now fixed.
(WB 2009/02/05)

4. New: The GridGenerator::half_hyper_shell function now also exists in 3d. However, to use it with the HalfHyperShellBoundary class, one has to pass inner and outer radii to the constructor.
(Martin Kronbichler, WB, 2009/01/13)

5. New: The restriction and embedding matrices for FE_Q are now implemented.
(Tobias Leicht, RH, 2008/12/23)

6. Improved: The GeometryInfo::cell_to_child_coordinates and GeometryInfo::child_to_cell_coordinates functions are extended to anisotropic refinement.
(Tobias Leicht, RH, 2008/12/23)

7. Fixed: There were cases with both anisotropic refinement and non-standard line orientation where we created wrong meshes in 3d. This is now fixed.
(Tobias Leicht, RH, 2008/12/23)

8. Fixed: The FEValuesExtractors classes were not copyable though there are legitimate reasons to do so. This is now fixed.
(WB 2008/12/17)

9. Fixed: The MappingQEulerian function was not thread-safe. This is now fixed.
(WB 2008/12/16)

10. New: The functions FETools::lexicographic_to_hierarchic_numbering and FETools::hierarchic_to_lexicographic_numbering now also exist in variants returning their result by value, rather than through their last argument.
(WB 2008/12/16)

11. New: FEValues objects can now directly calculate the Laplacians of solution functions at quadrature points. This is equivalent to the trace of the Hessians, but faster and more convenient in user code. step-31 program has also been updated to use this.
(Martin Kronbichler 2008/12/15)

12. Improved: The FEValuesViews objects that one gets when writing things like fe_values[velocities] have acquired the ability to extract individual components of solutions at quadrature points by writing code like fe_values[velocities].get_function_values (global_solution, local_velocity_values). See the new section at the end of the Handling vector valued problems module for an overview of this facility. The step-31 program has also been updated to use this.
(WB 2008/12/13)

13. Improved: The FEValuesViews objects that one gets when writing things like fe_values[velocities] (see Handling vector valued problems) have become a lot smarter. They now compute a significant amount of data at creation time, rather than on the fly. This means that creating such objects becomes more expensive but using them is cheaper. To offset this cost, FEValuesBase objects now create all possible FEValuesViews objects at creation time, rather than whenever you do things like fe_values[velocities], and simply return a reference to a pre-generated object. This turns an $$O(N)$$ effort into an $$O(1)$$ effort, where $$N$$ is the number of cells.
(WB 2008/12/10)

14. Upgraded: The FunctionParser classes now use version 2.83 of the fparser library.
(Luca Heltai 2008/12/08)

15. Fixed: The GridGenerator::laplace_transform would only do at most 1000 iterations in its solver, irrespective of the actual number of nodes to be moved around. This is now fixed: the maximum number now equals the number of nodes.
(Luca D'Auria 2008/12/05)

16. New: The function DoFTools::make_zero_boundary_constraints() computes the constraints that result from requiring the degrees of freedom at the boundary to be zero. Use cases are when the sparsity pattern is not known / not wanted / not inaccessible.
(Toby D. Young 2008/12/04)

17. Updated: The function ConstraintMatrix::distribute_local_to_global() for matrices does now use row-wise addition into sparse matrices, which accelerates the transfer from local to global data.
(Martin Kronbichler, WB 2008/11/27)

18. Fixed: The VectorTools::interpolate_boundary_values function was implemented a bit clumsily and was using much more time than necessary. This should be fixed now.
(WB 2008/11/25)

19. Fixed: The GridIn::read_msh function had a bug that made it reject MSH input files if they contained type-15 cells with more than one associated vertex.
(WB 2008/11/05)

20. Fixed: The DoFRenumbering::Cuthill_McKee algorithm had a bug when applied to MGDoFHandler objects and if the reverse flag was set. This should now be fixed.
(WB 2008/10/31)

21. New: MatrixTools::apply_boundary_values() also for PETScWrappers::MPI::BlockSparseMatrix.
(Timo Heister 2008/10/27)

22. New: When calling function DoFTools::make_sparsity_pattern with a ConstraintMatrix, it is now possible to set a bool argument keep_constrained_dofs. When this flag is set to false, constrained rows and columns will not be part of the sparsity pattern, which increases the performance of matrix operations and decrease memory consumption in case there are many constraints.
(Martin Kronbichler 2008/10/21)

23. New: There is now a second DoFTools::count_dofs_with_subdomain_association function that calculates the number of degrees of freedom associated with a certain subdomain and splits the result up according to the vector component of each degree of freedom. This function is needed when splitting block matrices in parallel computations.
(WB 2008/10/07)

24. Fixed: The GridOut::write_gnuplot function had a bug that made it output only the very first cell. This is now fixed.
(WB 2008/10/09)

25. New: The GridIn::read_msh function can now read version 2 of the MSH format described here.
(WB 2008/10/07)

26. Fixed: In rare cases, when Triangulation::limit_level_difference_at_vertices is passed to the constructor of the Triangulation class, meshes could be generated that do not honor this flag. This is now fixed.
(WB 2008/10/06)

27. New: The class FE_Q can now alternatively be constructed based on support points from a given one-dimensional quadrature rule.
(Katharina Kormann, Martin Kronbichler, 2008/09/07)

28. Fixed: Using the ConstraintMatrix class, when a degree of freedom was constrained against another DoF, and that other DoF was constrained to be equal to zero (i.e. the trivial linear combination of a third set of degrees of freedom), an exception resulted. This is now fixed.
(WB 2008/08/15)

29. New: It is now possible to get the inverse of the Jacobian matrix from the transformation form the real to the unit cell by using FEValues::inverse_jacobian.
(Martin Kronbichler 2008/08/13)

30. New: There is a second GridTools::partition_triangulation function that takes a cell connectivity pattern as argument, rather than computing it itself as the existing function. Use cases are discussed in the documentation of the new function.
(WB 2008/08/06)

31. Fixed: GridTools::find_cells_adjacent_to_vertex had a bug that prevented its correct functioning in three dimensions. Some cases were left out due to uncorrect assumptions on the various refinement possibilities.
(Luca Heltai 2008/07/17)

New: There is now a new Triangulation::prevent_distorted_boundary_cells function which is only useful in case of anisotropic refinement. At the boundary of the domain, the new point on the face may be far inside the current cell, if the boundary has a strong curvature. If we allow anistropic refinement here, the resulting cell may be strongly distorted, especially if it is refined again later on. To prevent this problem, this function flags such cells for isotropic refinement. It is called automatically from Triangulation::prepare_coarsening_and_refinement. Therefore this should have no effect on user codes.
(Tobias Leicht 2008/07/08 as of branch_anisotropic at 2006/08/02)

32. New: There is now a new Triangulation::create_children function which actually sets up the children of a cell and updates the neighbor information. This work has been done in Triangulation::execute_refinement so far. Memory allocation has to be done prior to the function call. However, this is really something internal to the library and there should never be the need to use this function in user_codes. Calling Triangulation::execute_coarsening_and_refinement will be all you need.
(Tobias Leicht 2008/07/08 as of branch_anisotropic at 2006/06/29)

33. New: A part of the functionality of Triangulation::execute_coarsening has been implemented in a new way and shifted to the new function Triangulation::coarsening_allowed. This function decides, depending on the refinement situation of all a cells neighbors, if the cell may be coarsened, in which case the bool value true is returned. This is a functionality which is partly dimension dependend.
However, there should never be any reason to use this function in user codes.
(Tobias Leicht 2008/07/08 as of branch_anisotropic at 2006/06/28)

34. New: There is now a new CellAccessor::neighbor_is_coarser function, which returns a bool value true, if the requested neighbor is a coarser one and false if the neighbor is as refined as the current cell. This new functionality is needed in the internals of the library to decide, if neighbor_of_neighbor or neighbor_of_coarser_neighbor has to be called. This decision is trivial in case of isotropic refinement, where the level of the cell and its neighbor is all the information needed. In case of aniostropic refinement, this new function can be used.
(Tobias Leicht 2008/07/08 as of branch_anisotropic at 2006/06/28)

35. New: There is now a new Triangulation::MeshSmoothing::allow_anisotropic_smoothing smoothing flag for triangulations. An important part of the smoothing process for refinement and coarsen flags is to ensure, that no double refinement takes places at any face. If a cell is refined twice, its neighbor has to be refined at least once. However, technically it is not always necessary to refine the neighbor isotropically, it may be sufficient to choose only one direction for the refinement. While this allows to reduce the number of newly created cells to a minimum, it is incompatible with older versions of the library, as anisotropic refinement takes place, even if all the explicitly set refinement flags ask for isotropic refinement.
Therefore this functionality is off by default. However, in order to use the new feature of anisotropic refinement to full extend, this flag should be set explicitly.
Note, that for reasons of backwards compatibility this flag is NOT included in the general MeshSmoothing::maximum_smoothing flag.
(Tobias Leicht 2008/07/08 as of branch_anisotropic at 2006/06/28)

36. New: There is now a new TriaObjectAccessor::clear_refinement_case function.
(RH 2008/07/08 as of branch_anisotropic at 2005/12/19)

37. Extended: The CellAccessor::neighbor_of_neighbor function is now extended to anisotropic refinement.
(RH 2008/07/08 as of branch_anisotropic at 2005/12/15)

38. Extended (internal): Lines on level>0 have always been stored pairwise. In order to allow the creation of single interior lines we now also allow storage of single lines. The TriangulationLevel<1>::reserve_space function now takes an additional n_consecutive_lines parameter which allows to create new single as well as pairs for lines n_consecutive_lines=1 or n_consecutive_lines=2, respectively.
(RH 2008/07/08 as of branch_anisotropic at 2005/12/15)

39. Changed: When allowing anisotropic refinement it cannot be guaranteed that all children of a cell are stored in consecutive components of the TriaLevel::cells vector, respectively. It is only known that children are stored at least in pairs. Therefore, it is not sufficient any more to store the index of the first child of a cell, only. Now the indices of every second child, i.e. of the even numbered children, must be stored. For this, the TriaObjectAccessor::set_children function for cells now has a new argument representing the number of the child for which the index shall be set. This function can only be called for even numbered children. The same applies to the respective function for faces.
Finally, we note that these functions are for internal use, only.
(RH 2008/07/08 as of branch_anisotropic at 2005/03/03)

40. Changed: The CellAccessor::refine_flag_set function now returns a RefinementCase argument instead of a boolean. Be aware, that you now still can ask if(!cell->refine_flag_set()) , if(cell->refine_flag_set()==false) and if(cell->refine_flag_set()) , but you cannot ask if(cell->refine_flag_set()==true) any more.
(RH 2008/07/08 as of branch_anisotropic at 2005/03/03)

41. Extended: The CellAccessor::set_refine_flag function now has a new RefinementCase argument which defaults to GeometryInfo<dim>::isotropic_refinement.
(RH 2008/07/08 as of branch_anisotropic at 2005/03/03)

42. New: There is now a new class RefinementCase defined in geometry_info.h, which allows to describe all possible (anisotropic and isotropic) refinement cases in (1,) 2 and 3 dimensions.
(RH 2008/07/08 as of branch_anisotropic at 2005/02/26)

43. New: There is now a new TriaObjectAccessor::n_children function which returns the number of children of the cell or face it was invoked for.
(RH 2008/07/08 as of branch_anisotropic at 2005/02/26)

44. New: There is now a new TriaObjectAccessor::refinement_case function which returns the RefinementCase a cell or face is refined with.
(RH 2008/07/08 as of branch_anisotropic at 2005/02/26)

45. New (for internal use only): There is now a new TriaObjectAccessor::set_refinement_case function. This function is mainly for internal use (required by Triangulation::execute_coarsening_and_refinement). It is not provided for dim=1 as there the refinement case defaults to isotropic refinement.
Note, that users should still use the CellAccessor::set_refine_flag function for setting the RefinementCase of cells to be refined within the next Triangulation::execute_coarsening_and_refinement function.
(RH 2008/07/08 as of branch_anisotropic at 2005/02/26)

46. New: New vector<RefinementCase> refinement_cases vectors have been introduced in TriaObjects. For memory efficiency (but with a penalty on run-time) they might be replaced by vector<vector<bool> > (dim, vector<bool> (n_quads/n_hexes)), later.
(RH 2008/07/08 as of branch_anisotropic at 2005/02/26)

47. Fixed: When using a higher order mapping of degree at least 3 (i.e. the MappingQ class) on meshes that have cells with non-standard face orientation and that touch the boundary of the domain, then some interpolation points were wrongly computed.
(Tobias Leicht, Timo Heister, WB 2008/06/10)

48. Fixed: The ParameterHandler::get_integer function now throws an exception when called for non-integer parameters. This exception has been commented out some time ago but is now reincluded.
(RH 2008/06/11)