This is the list of changes made between the deal.II releases listed above. All entries are signed with the names of the author.

General

New: example program step-11.
(WB 2001/09/17)
New: example program step-10.
(WB, RH 2001/09/12)
New: the ./configure script now also recognizes Compaq's cxx compiler. The library now also compiles cleanly with this compiler.
(WB 2001/07/06)
Changed: The libraries are no more linked using the C++ compilation flags, but rather using LDFLAGS. Some compilers object to compilation flags on the linker line.
(WB 2001/05/23)
Changed: If in multithreaded mode, the ACE library is now automatically added to the $(LIBS) Makefile variable. There is no need anymore for a special clause in your Makefile.
(WB 2001/05/23)
New: the ./configure script now also recognizes the Intel ICC compiler that was released for Linux lately. The library now also compiles cleanly with this compiler.
(WB 2001/05/22)
Fixed: the program that generated HTML from the example programs was broken on some platforms for reasons beyond our knowledge. This is now fixed.
(Roger Young, WB 2001/03/22)
Improved: libraries are now all in a subdirectory lib in the deal.II home directory. It is now easy to make this directory a link to put the libraries on a faster disk.
(GK 2001/03/14)
New Design of FiniteElements and Mappings

New: The finite element classes have been redesigned from scratch to allow also elements that depend on the actual cell shape, like Raviart-Thomas, BDM or edge elements. We have implemented continuous Lagrange elements up to degree four and discontinuous Lagrange elements up to degree 10. They can be easily extended to higher degree.

Furthermore we have totally reimplemented the mapping between unit and real cell. In future we won't be restricted to Q₁ mapping any more, but we will have Q_p mappings of arbitrary degree p. Presently implemented are mappings up to degree 10 in 2d, and degree three in 3d. This allows to approximate curved boundaries not only by straight lines (linear approximation) but also by quadratic, cubic, etc approximation. The finite elements will be able to be combined with arbitrary mappings. So in future we can use subparametric, iso- as well as superparametric elements.

The new implementation uses a totally new structure of communication between FEValues, the FiniteElements and the Mappings. Despite of this, the new structure will almost not be 'visible' to the user of deal.II as we tried to retain the interface to the user (mainly that of FEValues) as far as possible.

Together with this new design comes a reduction of 25000(!) lines of deal.II code. This elimination mainly affected code that previously was machine generated (e.g. by maple). In view of almost unchanged execution times of applications, the faster compilation and additional functionality constitutes a significant improvement of deal.II. Results produced by the new code match those of the old design up to machine precision.
(RH & GK 2001/03/13)
New: There is now some support to include and use routines from the Harwell Subroutine Library.
(WB 2001/01/30)
New: The ./configure script now checks for the existence of a Fortran 77 compiler, and sets its path, some compiler flags and the libraries to be linked in when mixing C++ and F77 in some variables in the file common/Make.global_options.
(WB 2000/12/30)

base

New: Color function DataOutBase::EpsFlags::reverse_grey_scale_color_function.
(WB 2001/08/24)
New: Function QProjector::project_to_child generates quadrature formulae which act on the area which a child would occupy.
(WB 2001/08/23)
Changed: The examples classes in the base directory are now moved into a namespace Functions of their own. This improves encapsulation, but also keeps the documentation of these functions together, as they were previously scrambled all over the screen in the documentation page of the base library.
(WB 2001/07/18)
New: classes FourierSineFunction and FourierCosineFunction, resembling one mode of a Fourier decomposition. Classes FourierSineSum and FourierCosineSum, resembling sums of such modes of a Fourier decomposition.
(WB 2001/07/18)
New: class vector2d was introduced in analogy to STL class vector. The bew class provides a two-dimensional array and replaces the use of FullMatrix in the base library.
(GK 2001/05/21)
Improved: JobIdentifier::operator() includes host name if supported by the operating system
(GK 2001/05/17)
New: There is now a new AutoDerivativeFunction class that automatically computes the gradient of a function by employing numerical difference quotients. This only, if the user function does not provide the gradient function himself. This class can replace the Function class as base class for user defined Function classes.
This new class can be used if the user only implements the value function but wants to call also the gradient functions.
(RH 2001/05/15)
New: The Quadrature class now has a constructor that only accepts quadrature points and sets the weights to invalid values.
(GK 2001/04/26)
New: The function Logstream::get_prefix allows access to the prefix string used for log-files.
(GK 2001/04/26)
New: There is now a global function trace that computes the trace of a tensor of rank 2.
(WB 2001/04/12)
New: The Threads now has a barrier class in order to synchronise multiple threads. In multithreaded mode, this is an alias to the ACE_Barrier class of the ACE library, if deal.II is not configured for multithreading, then the class is a dummy class that does nothing.
(WB 2001/03/19)
New: We now support the output format of the Visualization Toolkit (Vtk) from the DataOutBase class and all derived classes.
(WB 2001/03/19)
New: The class TensorProductPolynomials<dim> performs the tensor product of 1d polynomials, computes its values, its first and second derivatives. If n polynomials are given to the constructor of this class, it constructs n^dim tensor product polynomials.
(RH 2001/03/14)
New: LagrangeEquidistant is a new class derived from Polynomial providing the 1d Lagrange interpolation of degree n of n+1 equidistant support points in [0,1]. It is implemented up to degree 10. This class is used for the implementation of the continuous and discontinuous Lagrange elements.
(RH 2001/03/14)
New: The new Polynomial class can be used as base class for all 1d polynomials. It stores the coefficients of the polynomial and uses the Horner scheme to evaluate values and all derivates.
(RH 2001/03/14)
New: function contract for two arguments of Tensor<1,dim>
(GK 2001/03/05)
New: Logstream::log_time_differences (bool) makes Logstream print the time since the last log line instead of accumulated time since program start.
(GK 2001/03/05)
Fix: Logstream::pop() does not perform anything on empty stacks.
(GK 2001/03/05)
Changed: Sort the quadrature points of each Quadrature<1> in ascending order. This actually changed the order of the points of only QGauss4 and QGauss5.
(Ralf Hartmann 2001/01/22)
New: function contract for two arguments of Tensor<1,dim>
(GK 2001/01/15)

lac

New: Function FullMatrix::symmetrize().
(WB 2001/09/20)
Improved: the stopping criterion of SolverBicgstab without computing the exact residual is implemented
(GK 2001/09/11)
New: The FullMatrix class now has a function operator*=, which simply scales the matrix.
(WB 2001/09/02)
New: Function BlockSparseMatrix::el(), just like SparseMatrix::el().
(WB 2001/08/21)
New: There is now a class MatrixOut which can be used to convert a matrix into a graphical output, for example in order to see visually that a matrix is diagonally dominant or not.
(WB 2001/08/21)
Changed: Base class Solver and all Preconditioner classes are now derived from Subscriptor. Class PreconditionLACSolver now uses smartpointers to the given solver and preconditioner objects. You will, therefore, have to derive your own preconditioners and solvers from Subscriptor if you want to use it with PreconditionLACSolver.
(WB 2001/08/16)
New: Classes Vector and BlockVector now have member functions operator*= which scale the vectors by a constant factor. These functions are aliases for the scale member functions except that they return a reference to themselves.
(WB 2001/08/14)
New: There is now a function FullMatrix::precondition_Jacobi. The PreconditionJacobi class is therefore now also applicable with the full matrix class.
(WB 2001/08/11)
New: The Vector and BlockVector classes can now be initialized using a new constructor that takes two iterators that denote a range of elements which are to be copied.
(WB 2001/08/08)
Changed: The SolverCG class now saves the initial matrix vector product if the initial value of the solution is vector is zero, as it is common practice to use this is starting vector.
(WB 2001/06/25)
New: SparsityPattern::reinit no more throws an error if the given maximal row lengths are all zero.
(WB 2001/06/25)
New: Class CompressedBlockSparsityPattern may be used as an intermediate form of the BlockSparsityPattern.
(WB 2001/06/23)
New: Class CompressedSparsityPattern may be used as an intermediate form of the SparsityPattern class if memory requirements are tight during construction of the sparsity pattern.
(WB 2001/06/22)
New: There are now functions SparsityPattern::copy_from and SparseMatrix::copy_from that can be used to construct sparsity patterns and matrix at once, i.e. without setting up a sparsity pattern element by element, possibly after using a way too large value for the maximal row length, then calling SparsityPattern::compress etc.
(WB 2001/05/07)
New: BlockIndices::block_size returns the size of a specified block.
(WB 2001/05/07)
New: There is now a (private) function SparsityPattern::optimized_lower_bound that is used as an optimized replacement for std::lower_bound for searching in the column number arrays. It unrolls small loops and it also seems that the compiler is able to optimized it better due to eliminated template parameters, making it about twice as fast as the standard implementation. In effect, it also speeds up the SSOR preconditioner that spent about one third of its time in that function by approximately 15 per cent.
(WB 2001/04/25)
New: The FilteredMatrix class is a replacement for the MatrixTools::apply_boundary_values function for cases where you would like to solve several times with the same matrix, either for different right hand sides, or for different boundary values.
(WB 2001/04/27)
New: There is now a function Vector::scale(Vector) that scales each element of the vector by the corresponding element of the argument.
(WB 2001/04/23)
Changed: Solver functions solve return void now. If the solver has not converged within the maximum number of iterations or encounters a breakdown, it throws an exception of type SolverControl::NoConvergence instead of returning a special value.
(GK 2001/03/29)
New: The functions FullMatrix::mmult and Tmmult now have an additional adding argument. If this flag is true, the matrix-matrix product is added to the resulting matrix; if false, the resulting matrix is set to (overwritten by) the matrix-matrix product. The latter is the behaviour these functions had before. Hence the default value is set to be false to ensure backward compatibility.
(RH 2001/03/29)
New: class SchurMatrix implements a Schur complement for block matrices. It provides matrix-vector multiplication suitable for iterative methods as well as pre- and post-processing of right hand side and slution, respectively.
(GK 2001/03/22)
Removed: The explicit instantiations of SparseMatrix<long double> are removed as a prerelease of gcc3.0 fails to compile it. A user of SparseMatrix<long double> needs now to include lac/include/lac/sparse_matrix.templates.h into their source file and to use an appropriate compiler, e.g. gcc2.95.2 or a future version of gcc3.0 (that will then hopefully be fixed).
(RH 2001/03/14)
New: class BlockMatrixArray<MATRIX> implements a block matrix based on an array of matrix pointers. Since this array may be incomplete, no direct access to entries is allowed. Matrix blocks may be scaled and transposed.
(GK 2001/02/13)
New: There is now some support to include and use routines from the Harwell Subroutine Library, and support classes SparseDirectMA27 and SparseDirectMA47 for the sparse direct solvers MA27 and MA47.
(WB 2001/01/30)

deal.II

New: Class MappingQ1Eulerian implementing an Eulerian mapping.
(Michael Stadler 2001/09/24)
New: VectorTools::create_boundary_right_hand_side integrates boundary forces for inhomogeneous Neumann boundary values.
(WB 2001/09/13)
New: DoFTools::make_flux_sparsity_pattern now exists also for 1d.
(WB 2001/09/03)
New: There are now two functions FETools::hierarchic_to_lexicographic_numbering and FETools::lexicographic_to_hierarchic_numbering which map the hierarchical numbering used in continuous finite element classes to a lexicographical numbering and back.
(WB 2001/08/31)
New: ConstraintMatrix::close now simply returns instead of throwing an exception, if the matrix was already closed.
(WB 2001/08/30)
New: Member function ConstraintMatrix::is_identity_constrained.
(WB 2001/08/29)
Fixed: in a rather rare case, some work was done twice in the KellyErrorEstimator class when in multithread mode. This is now fixed.
(WB 2001/08/24)
New: There is now a class GridTools which provides algorithms working on triangulations. At present, it offers a function computing the diameter of a triangulation.
(WB 2001/08/16)
Changed: The MatrixCreator and MatrixTools class have lost their template arguments. Rather, the individual functions now are templated on their own, so the compiler can pick out the correct space dimension on its own.
(WB 2001/08/15)
Extended: ConstraintMatrix::merge can now handle arguments which further constrain the present object.
(WB 2001/07/30)
New: Implement DoFTools::make_sparsity_pattern, DoFTools::make_boundary_sparsity_pattern, and ConstraintMatrix::condense to work on the CompressedSparsityPattern and CompressedBlockSparsityPattern and classes.
(WB 2001/06/22)
New: FEValues::get_quadrature() and FEFaceValuesBase::get_quadrature() return a reference to the quadrature formula used by a FEValues object.
(WB 2001/06/21)
New: DoFTools::compute_intergrid_transfer_representation is a function that generates a matrix representation that can be used to transfer data from one grid to one component of the fields on another, finer, grid.
(WB 2001/05/24)
Changed: the GeometryInfo class has been reverted from a general template that calculates the values of its member variables by recursion/induction to a set of explicitly specialized classes. This seemed necessary since most compilers objected to the old implementation as the declaration of the values of the class required knowledge of the elements of the class with one lower dimension which was, however, only being declared at present except for 1d. Also, allocating space for variables was difficult as that would mean declaring specializations after their first use. The new (and very old) implementation is entirely compatible to the previous one.
(WB 2001/05/23)
Changed: the classes that denote flags to the TimeStepBase_Tria class have been move from local classes to a namespace of their own, names TimeStepBase_Tria_Flags.
(WB 2001/05/23)
Fixed: due to a bug in gcc, the compiler did not check that we did not use the set of given boundary indicators to the DoFTools::extract_boundary_dofs function in 1d. That set was therefore ignored. This is now fixed.
(WB 2001/05/18)
Changed: the flags which are given to the GridOut class to modify the appearance of the output have been moved from local classes of the GridOut class to a namespace names GridOutFlags and have lost the trailing Flags part in their name. This change was necessary as C++ does not allow explicit specialization of member classes; the previous use in the library was only accepted by GCC as an extension.
(WB 2001/05/17)
New: The functions DoFTools::map_dof_to_boundary_indices, DoFTools::make_boundary_sparsity_pattern, DoFHandler::n_boundary_dofs, DoFHandler::max_couplings_between_boundary_dofs, and VectorTools::project_boundary_values, are now also implemented in 1d.
(WB 2001/04/29)
New: There are now functions DoFTools::map_dofs_to_support_points and DoFTools::map_support_points_to_dofs that generate a map between DoF indices and the support points of these DoFs, and the other way round.
(WB 2001/04/24)
New: Implement the VectorTools::interpolate_boundary_values a second time, this time taking a FunctionMap object as do all the other functions of this type.
(WB 2001/04/23)
Fixed: The vertex numbers written by the GridOut::write_ucd_faces function are now also 1-based. This, to be consistent with the vertex numbers given by the GridOut::write_ucd function.
(RH 2001/04/20)
Extended: the DerivativeApproximation class now also works in 3d, as the spectral norm of a symmetric 3x3 matrix can now be computed.
(Roger Young 2001/04/12)
Fixed: the DoFRenumbering::Cuthill_McKee function did not work correctly when giving the reversed_numbering flag (off-by-one indexing). This is now fixed.
(Oliver Rheinbach 2001/04/12)
Changed: A typedef FunctionMap was declared in at least four places previously (in classes DoFHandler, MatrixTools, VectorTools, and KellyErrorEstimator). It is now unified in one place and is called FunctionMap<dim>::type (it is a local typedef in a class called FunctionMap). This type is defined in the file dofs/function_map.h.
(WB 2001/04/09)
Fixed: When using Neuman boundary functions in the KellyErrorEstimator class, it was assumed that the function object had Function::vector_value overloaded, even in the scalar case. We now use Function::value instead.
(WB 2001/04/09)
New: The various MatrixCreator::create_*_matrix functions are now using multiple threads to do their work, if the library is configured to use multithreading.
(WB 2001/04/08)
New: The following functions are extended to work on arbitrary mappings:
The static create_mass_matrix, create_boundary_mass_matrix and create_laplace_matrix member functions of the class MatrixCreator,
the static project and project_boundary_values member functions of the class VectorTools,
the two versions of the static estimate member functions of the class KellyErrorEstimator,
and the static approximate_gradient and approximate_second_derivative member functions of the class DerivativeApproximation.

All these functions now take a Mapping object as additional argument.
For backward compatibility there exist second versions of each of these functions that still don't have a Mapping argument. These functions implicitly use a mapping of the class MappingQ1.
(RH 2001/04/05)
Removed: Along with the redesign of the FiniteElement and the Mapping classes the FiniteElement::get_local_mass_matrix function is now removed. This was necessary as the local mass matrix depends not only on the considered FiniteElement but also on the mapping between the unit to the real cell. Hence this function cannot be a member function of a FiniteElement any more.

As a consequence also the MatrixCreator::create_mass_matrix function with two arguments is removed, as it relied on the get_local_mass_matrix function. If in future the user wants to create a mass matrix they need to use one of the remaining MatrixCreator::create_mass_matrix functions that require an appropriate Quadrature as argument.
(RH 2001/04/04)
New/Fixed: Now there exists a new Triangulation::ExcMultiplySetLineInfoOfLine exception, that is thrown if the SubCellData that is given to Triangulation::create_triangulation, multiply includes the line info of a specific line. Before the fix the wrong ExcInteriorLineCantBeBoundary exception was thrown.
(RH 2001/04/03)
Fixed: Missing ucd is now added to the list of supported output formats returned by GridOut::get_output_format_names.
(RH 2001/04/03)
New/fixed: In some obscure corner cases, the detection logic in DataOut_DoFData::add_data_vector would not have been able to detect whether something is a DoF data vector or a vector of cell data, and in some equally rare cases this would also have made a difference. This is now fixed by adding another argument to the function telling it either to automatically detect the vector type (default) or to assume that it has a certain type (for these corner cases).
(WB 2001/03/30)
Removed: the ProblemBase class, which has been deprecated since before the release of deal.II 3.0, has finally been removed. The same applies for the classes Assembler, Equation, MassMatrix, and LaplaceMatrix.
(WB 2001/03/27)
New: There is now a class MappingC1 that implements a continuously differentiable C¹ mapping of the boundary by using a cubic mapping with continuous derivatives at cell boundaries. The class presently only implements something for 2d and 1d (where it does nothing).
(WB 2001/03/27)
New: The static interpolate, create_right_hand_side, interpolate_boundary_values, integrate_difference and compute_mean_value member functions of the class VectorTools are extended to work on arbitrary mappings. All these functions now take a Mapping object as additional argument.
For backward compatibility there exist second versions of each of these functions that still don't have a Mapping argument. These functions implicitly use a mapping of the class MappingQ1.
(RH 2001/03/27)
New: Boundary and derived classes now have a function get_normals_at_vertices that returns a multiple of the normal vector to the boundary at the vertices of a given face. This is used in the construction of C¹ mappings of the boundary.
(WB 2001/03/23)
New: The new function PersistentTriangulation<dim>::restore(unsigned int) allows to restore the triangulation step by step.
New: Now there exists a function PersistentTriangulation<dim>::clear_flags to allow to re-read_flags without the need of clearing the triangulation in advance.
New: The new function PersistentTriangulation<dim>::n_refinement_steps returns the number of refinement steps stored in refine_flags.
(RH 2001/03/20)
New: GridGenerator::hyper_rectangle creates coordinate-parallel rectangles in arbitrary dimension.
(GK 2001/03/16)
Changed: The syntax of the FiniteElement<dim>::get_unit_support_points function is changed; it returns a reference to the vector of points in lieu of taking this vector as argument. The unit support points are now computed by the constructor of the FiniteElement and not on each FiniteElement<dim>::get_unit_support_points function call as before.
(WB 2001/03/14)
Removed: The function FiniteElement<dim>::get_support_points is removed as the FiniteElement cannot know the position of the support points by itself. This is because the support points depend on the unit support points (known to the FiniteElement class) but they also depend on the mapping.

In future the support points can be computed using a FEValues object initialized with the Mapping and a Quadrature that uses the unit support points as quadrature points.
(RH 2001/03/14)
New: The class Boundary has two new functions get_intermediate_points_on_line and get_intermediate_points_on_quad that needs to be implemented in derived classes only if GridOut is used with MappingQ of degree p>2.
(RH 2001/03/14)
New: The functions GridOut::write_gnuplot and GridOut::write_eps now take a pointer to a Mapping object as additional argument. This allows to write grids in 2d whereby cells with a curved boundary are transformed by the given Mapping class. The default mapping is MappingQ1. Note, that the grids do not show the ‘original’ domain with boundaries described by the Boundary classes but the discretized domain whose boundary cells are transformed using the given mapping.

There are also a new GnuplotFlags::n_boundary_face_points and EpsFlags::n_boundary_face_points variables to set the number of additional points written to represent the curved boundary faces.
(RH 2001/03/14)
New: The constructor of the FEValues and FE(Sub)FaceValues classes now take a reference of a Mapping object as additional argument. This is the new possibility to combine a FiniteElement with an arbitrary Mapping, see also New FE and Mapping Design.

For backward compatibility there still exists a constructor of FEValues without a Mapping argument that uses a MappingQ1 by default.
(RH 2001/03/14)
Changed: We now unconditionally include deal.II/include/grid/tria_accessor.templates.h (which contains some inline functions for triangulation accessor classes) into deal.II/include/grid/tria_accessor.h to work around a problem with gcc3.0 which does not place a copy of these functions into the library. Previously we only included the file in optimized mode.
(RH 2001/03/14)
New: The class GridReordering::Cell has now a copy constructor to work around a bug in a gcc3.0 snapshot.
(RH, WB 2001/03/14)
Changed: when refining a hexahedron in 3d, the rules by which the new vertices at the centers of the faces and at the center of the cell are placed, are changed. They are changed in a way as to minimize the distortion of the resulting cells from the optimal shape and to make them look as much alike as possible to generate a smoother grid.
(WB 2001/03/02)
Fix: the DoFTools::compute_intergrid_constraints function took memory quadratic in the number of degrees of freedom. This is now reduced to linear behaviour, with a constant that depends on the number of levels by which the two grids differ.
(WB 2001/02/26)
Fix: in the triangulation, the straight_boundary variable, which is a reference, was assigned the address of a temporary object. It is unclear how this could have worked for three years, but it apparently did...
(WB 2001/02/26)
New: The DoFTools class now has a function count_dofs_per_component that counts the number of degrees of freedom in each of the components of the finite element, i.e. how many degrees of freedom there are on the global mesh for each variable (field).
(WB 2001/02/22)
New: The CellAccessor class now has a function has_boundary_lines that mostly has the same semantics that at_boundary has, but also covers the case that a hexahedron in 3d may be at the boundary only by one line, rather than by a whole face. In that case, at_boundary reports false, while has_boundary_lines reports true.
(WB 2001/02/21)
New: There is now a function ConstraintMatrix::merge that merges the constraints represented by two constraint matrices into one object.
(WB 2001/01/31)