GitHub

Ferrite changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Removed

  • The deprecated third type parameter for interpolations have been removed. Old code which tries to use three parameters will now throw the somewhat cryptic error:
    julia> Lagrange{2, RefCube, 1}()
ERROR: too many parameters for type
    (#1083)

Other

v1.0.0 - 2024-09-30

Ferrite version 1.0 is a relatively large release, with a lot of new features, improvements, deprecations and some removals. These changes are made to make the code base more consistent and more suitable for future improvements. With this 1.0 release we are aiming for long time stability, and there is no breaking release 2.0 on the horizon.

Unfortunately this means that code written for Ferrite version 0.3 will have to be updated. All changes, with upgrade paths, are listed in the sections below. Since these sections include a lot of other information as well (new features, internal changes, ...) there is also a dedicated section about upgrading code from Ferrite 0.3 to 1.0 (see below) which include the most common changes that are required. In addition, in all cases where possible, you will be presented with a descriptive error message telling you what needs to change.

Deprecations for 1.0 will be removed during the 1.x release series. When upgrading old code it is therefore recommended to use Ferrite 1.0 as a first stepping stone since this release contain descriptive deprecation error messages that might not exist in e.g. Ferrite version 1.2.

Upgrading code from Ferrite 0.3 to 1.0

This section give a short overview of the most common required changes. More details and motivation are included in the following sections (with links to issues/pull request for more discussion).

  • Interpolations: remove the first parameter (the reference dimension) and use new reference shapes.

    Examples:

    # Linear Lagrange interpolation for a line
- Lagrange{1, RefCube, 1}()
+ Lagrange{RefLine, 1}()

# Linear Lagrange interpolation for a quadrilateral
- Lagrange{2, RefCube, 1}()
+ Lagrange{RefQuadrilateral, 1}()

# Quadratic Lagrange interpolation for a triangle
- Lagrange{2, RefTetrahedron, 2}()
+ Lagrange{RefTriangle, 2}()

    For vector valued problems it is now required to explicitly vectorize the interpolation using the new VectorizedInterpolation. This is required when passing the interpolation to CellValues and when adding fields to the DofHandler using add!. In both of these places the interpolation was implicitly vectorized in Ferrite 0.3.

    Examples:

    # Linear Lagrange interpolation for a vector problem on the triangle (vector dimension
# same as the reference dimension)
ip_scalar = Lagrange{RefTriangle, 1}()
ip_vector = ip_scalar ^ 2 # or VectorizedInterpolation{2}(ip_scalar)

  • Quadrature: remove the first parameter (the reference dimension) and use new reference shapes.

    Examples:

    # Quadrature for a line
- QuadratureRule{1, RefCube}(quadrature_order)
+ QuadratureRule{RefLine}(quadrature_order)

# Quadrature for a quadrilateral
- QuadratureRule{2, RefCube}(quadrature_order)
+ QuadratureRule{RefQuadrilateral}(quadrature_order)

# Quadrature for a tetrahedron
- QuadratureRule{3, RefTetrahedron}(quadrature_order)
+ QuadratureRule{RefTetrahedron}(quadrature_order)

  • Quadrature for face integration (FacetValues): replace QuadratureRule{dim-1, reference_shape}(quadrature_order) with FacetQuadratureRule{reference_shape}(quadrature_order).

    Examples:

    # Quadrature for the facets of a quadrilateral
- QuadratureRule{1, RefCube}(quadrature_order)
+ FacetQuadratureRule{RefQuadrilateral}(quadrature_order)

# Quadrature for the facets of a triangle
- QuadratureRule{1, RefTetrahedron}(quadrature_order)
+ FacetQuadratureRule{RefTriangle}(quadrature_order)

# Quadrature for the facets of a hexhedron
- QuadratureRule{2, RefCube}(quadrature_order)
+ FacetQuadratureRule{RefHexahedron}(quadrature_order)

  • CellValues: replace usage of CellScalarValues and CellVectorValues with CellValues. For vector valued problems the interpolation passed to CellValues should be vectorized to a VectorizedInterpolation (see above).

    Examples:

    # CellValues for a scalar problem with triangle elements
- qr = QuadratureRule{2, RefTetrahedron}(quadrature_order)
- ip = Lagrange{2, RefTetrahedron, 1}()
- cv = CellScalarValues(qr, ip)
+ qr = QuadratureRule{RefTriangle}(quadrature_order)
+ ip = Lagrange{RefTriangle, 1}()
+ cv = CellValues(qr, ip)

# CellValues for a vector problem with hexahedronal elements
- qr = QuadratureRule{3, RefCube}(quadrature_order)
- ip = Lagrange{3, RefCube, 1}()
- cv = CellVectorValues(qr, ip)
+ qr = QuadratureRule{RefHexahedron}(quadrature_order)
+ ip = Lagrange{RefHexahedron, 1}() ^ 3
+ cv = CellValues(qr, ip)

    If you use CellScalarValues or CellVectorValues in method signature you must replace them with CellValues. Note that the type parameters are different.

    Examples:

    - function do_something(cvs::CellScalarValues, cvv::CellVectorValues)
+ function do_something(cvs::CellValues, cvv::CellValues)

    The default geometric interpolation have changed from the function interpolation to always use linear Lagrange interpolation. If you use linear elements in the grid, and a higher order interpolation for the function you can now rely on the new default:

    qr = QuadratureRule(...)
- ip_function = Lagrange{2, RefTetrahedron, 2}()
- ip_geometry = Lagrange{2, RefTetrahedron, 1}()
- cv = CellScalarValues(qr, ip_function, ip_geometry)
+ ip_function = Lagrange{2, RefTetrahedron, 2}()
+ cv = CellValues(qr, ip_function)

    and if you have quadratic (or higher order) elements in the grid you must now pass the corresponding interpolation to the constructor:

    qr = QuadratureRule(...)
- ip_function = Lagrange{2, RefTetrahedron, 2}()
- cv = CellScalarValues(qr, ip_function)
+ ip_function = Lagrange{2, RefTetrahedron, 2}()
+ ip_geometry = Lagrange{2, RefTetrahedron, 1}()
+ cv = CellValues(qr, ip_function, ip_geometry)

  • FacetValues: replace usage of FaceScalarValues and FaceVectorValues with FacetValues. For vector valued problems the interpolation passed to FacetValues should be vectorized to a VectorizedInterpolation (see above). The input quadrature rule should be a FacetQuadratureRule instead of a QuadratureRule.

    Examples:

    # FacetValues for a scalar problem with triangle elements
- qr = QuadratureRule{1, RefTetrahedron}(quadrature_order)
- ip = Lagrange{2, RefTetrahedron, 1}()
- cv = FaceScalarValues(qr, ip)
+ qr = FacetQuadratureRule{RefTriangle}(quadrature_order)
+ ip = Lagrange{RefTriangle, 1}()
+ cv = FacetValues(qr, ip)

# FaceValues for a vector problem with hexahedronal elements
- qr = QuadratureRule{2, RefCube}(quadrature_order)
- ip = Lagrange{3, RefCube, 1}()
- cv = FaceVectorValues(qr, ip)
+ qr = FacetQuadratureRule{RefHexahedron}(quadrature_order)
+ ip = Lagrange{RefHexahedron, 1}() ^ 3
+ cv = FacetValues(qr, ip)

  • DofHandler construction: it is now required to pass the interpolation explicitly when adding new fields using add! (previously it was optional, defaulting to the default interpolation of the elements in the grid). For vector-valued fields the interpolation should be vectorized, instead of passing the number of components to add! as an integer.

    Examples:

    dh = DofHandler(grid) # grid with triangles

# Vector field :u
- add!(dh, :u, 2)
+ add!(dh, :u, Lagrange{RefTriangle, 1}()^2)

# Scalar field :p
- add!(dh, :u, 1)
+ add!(dh, :u, Lagrange{RefTriangle, 1}())

  • Boundary conditions: The entity enclosing a cell was previously called face, but is now denoted a facet. When applying boundary conditions, rename getfaceset to getfacetset and addfaceset! is now addfacetset!. These sets are now described by FacetIndex instead of FaceIndex. When looping over the facets of a cell, change nfaces to nfacets.

    Examples:

    # Dirichlet boundary conditions
- addfaceset!(grid, "dbc", x -> x[1] ≈ 1.0)
+ addfacetset!(grid, "dbc", x -> x[1] ≈ 1.0)

- dbc = Dirichlet(:u, getfaceset(grid, "dbc"), Returns(0.0))
+ dbc = Dirichlet(:u, getfacetset(grid, "dbc"), Returns(0.0))

# Neumann boundary conditions
- for facet in 1:nfaces(cell)
-     if (cellid(cell), facet) ∈ getfaceset(grid, "Neumann Boundary")
+ for facet in 1:nfacets(cell)
+     if (cellid(cell), facet) ∈ getfacetset(grid, "Neumann Boundary")
          # ...

  • VTK Export: The VTK export has been changed #692.

    - vtk_grid(name, dh) do vtk
-     vtk_point_data(vtk, dh, a)
-     vtk_point_data(vtk, nodal_data, "my node data")
-     vtk_point_data(vtk, proj, projected_data, "my projected data")
-     vtk_cell_data(vtk, proj, projected_data, "my projected data")
+ VTKGridFile(name, dh) do vtk
+     write_solution(vtk, dh, a)
+     write_node_data(vtk, nodal_data, "my node data")
+     write_projection(vtk, proj, projected_data, "my projected data")
+     write_cell_data(vtk, cell_data, "my projected data")
end

    When using a paraview_collection collection for e.g. multiple timesteps the VTKGridFile object can be used instead of the previous type returned from vtk_grid.

  • Sparsity pattern and global matrix construction: since there is now explicit support for working with the sparsity pattern before instantiating a matrix the function create_sparsity_pattern has been removed. To recover the old functionality that return a sparse matrix from the DofHandler directly use allocate_matrix instead.

    Examples:

    # Create sparse matrix from DofHandler
- K = create_sparsity_pattern(dh)
+ K = allocate_matrix(dh)

# Create condensed sparse matrix from DofHandler + ConstraintHandler
- K = create_sparsity_pattern(dh, ch)
+ K = allocate_matrix(dh, ch)

Added

  • InterfaceValues for computing jumps and averages over interfaces. (#743)

  • InterfaceIterator and InterfaceCache for iterating over interfaces. (#747)

  • FacetQuadratureRule implementation for RefPrism and RefPyramid. (#779)

  • The DofHandler now support selectively adding fields on sub-domains (rather than the full domain). This new functionality is included with the new SubDofHandler struct, which, as the name suggest, is a DofHandler for a subdomain. (#624, #667, #735)

  • New reference shape structs RefLine, RefTriangle, RefQuadrilateral, RefTetrahedron, RefHexahedron, and RefPrism have been added. These encode the reference dimension, and will thus replace the old reference shapes for which it was necessary to always pair with an explicit dimension (i.e. RefLine replaces (RefCube, 1), RefTriangle replaces (RefTetrahedron, 2), etc.). For writing "dimension independent code" it is possible to use Ferrite.RefHypercube{dim} and Ferrite.RefSimplex{dim}. (#679)

  • New methods for adding entitysets that are located on the boundary of the grid: addboundaryfacetset! and addboundaryvertexset!. These work similar to addfacetset! and addvertexset!, but filters out all instances not on the boundary (this can be used to avoid accidental inclusion of internal entities in sets used for boundary conditions, for example). (#606)

  • New interpolation VectorizedInterpolation which vectorizes scalar interpolations for vector-valued problems. A VectorizedInterpolation is created from a (scalar) interpolation ip using either ip ^ dim or VectorizedInterpolation{dim}(ip). For convenience, the method VectorizedInterpolation(ip) vectorizes the interpolation to the reference dimension of the interpolation. (#694, #736)

  • New (scalar) interpolation Lagrange{RefQuadrilateral, 3}(), i.e. third order Lagrange interpolation for 2D quadrilaterals. (#701, #731)

  • CellValues now support embedded elements. Specifically you can now embed elements with reference dimension 1 into spatial dimension 2 or 3, and elements with reference dimension 2 in to spatial dimension 3. (#651)

  • CellValues now support (vector) interpolations with dimension different from the spatial dimension. (#651)

  • FacetQuadratureRule have been added and should be used for FacetValues. A FacetQuadratureRule for integration of the facets of e.g. a triangle can be constructed by FacetQuadratureRule{RefTriangle}(order) (similar to how QuadratureRule is constructed). (#716)

  • New functions Ferrite.reference_shape_value(::Interpolation, ξ::Vec, i::Int) and Ferrite.reference_shape_gradient(::Interpolation, ξ::Vec, i::Int) for evaluating the value/gradient of the ith shape function of an interpolation in local reference coordinate ξ. These methods are public but not exported. (Note that these methods return the value/gradient wrt. the reference coordinate ξ, whereas the corresponding methods for CellValues etc return the value/gradient wrt the spatial coordinate x.) (#721)

  • FacetIterator and FacetCache have been added. These work similarly to CellIterator and CellCache but are used to iterate over (boundary) face sets instead. These simplify boundary integrals in general, and in particular Neumann boundary conditions are more convenient to implement now that you can loop directly over the face set instead of checking all faces of a cell inside the element routine. (#495)

  • The ConstraintHandler now support adding Dirichlet boundary conditions on discontinuous interpolations. (#729)

  • collect_periodic_faces now have a keyword argument tol that can be used to relax the default tolerance when necessary. (#749)

  • VTK export now work with QuadraticHexahedron elements. (#714)

  • The function bounding_box(::AbstractGrid) has been added. It computes the bounding box for a given grid (based on its node coordinates), and returns the minimum and maximum vertices of the bounding box. (#880)

  • Support for working with sparsity patterns has been added. This means that Ferrite exposes the intermediate "state" between the DofHandler and the instantiated matrix as the new struct SparsityPattern. This make it possible to insert custom equations or couplings in the pattern before instantiating the matrix. The function create_sparsity_pattern have been removed. The new function allocate_matrix is instead used to instantiate the matrix. Refer to the documentation for more details. (#888)

    To upgrade: if you want to recover the old functionality and don't need to work with the pattern, replace any usage of create_sparsity_pattern with allocate_matrix.

  • A new function, geometric_interpolation, is exported, which gives the geometric interpolation for each cell type. This is equivalent to the deprecated Ferrite.default_interpolation function. (#953)

  • CellValues and FacetValues can now store and map second order gradients (Hessians). The number of gradients computed in CellValues/FacetValues is specified using the keyword arguments update_gradients::Bool (default true) and update_hessians::Bool (default false) in the constructors, i.e. CellValues(...; update_hessians=true). (#953)

  • L2Projector supports projecting on grids with mixed celltypes. (#949)

Changed

  • It is now possible to create sparsity patterns with interface couplings, see the new function add_interface_entries! and the rework of sparsity pattern construction. (#710)

  • The AbstractCell interface has been reworked. This change should not affect user code, but may in some cases be relevant for code parsing external mesh files. In particular, the generic Cell struct have been removed in favor of concrete cell implementations (Line, Triangle, ...). (#679, #712)

    To upgrade replace any usage of Cell{...}(...) with calls to the concrete implementations.

  • The default geometric mapping in CellValues and FacetValues have changed. The new default is to always use Lagrange{refshape, 1}(), i.e. linear Lagrange polynomials, for the geometric interpolation. Previously, the function interpolation was (re) used also for the geometry interpolation. (#695)

    To upgrade, if you relied on the previous default, simply pass the function interpolation also as the third argument (the geometric interpolation).

  • All interpolations are now categorized as either scalar or vector interpolations. All (previously) existing interpolations are scalar. (Scalar) interpolations must now be explicitly vectorized, using the new VectorizedInterpolation, when used for vector problems. (Previously implicit vectorization happened in the CellValues constructor, and when adding fields to the DofHandler). (#694)

  • It is now required to explicitly pass the interpolation to the DofHandler when adding a new field using add!. For vector fields the interpolation should be vectorized, instead of passing number of components as an integer. (#694)

    To upgrade don't pass the dimension as an integer, and pass the interpolation explicitly. See more details in Upgrading code from Ferrite 0.3 to 1.0.

  • Interpolations should now be constructed using the new reference shapes. Since the new reference shapes encode the reference dimension the first type parameter of interpolations have been removed. (#711) To upgrade replace e.g. Lagrange{1, RefCube, 1}() with Lagrange{RefLine, 1}(), and Lagrange{2, RefTetrahedron, 1}() with Lagrange{RefTriangle, 1}(), etc.

  • QuadratureRules should now be constructed using the new reference shapes. Since the new reference shapes encode the reference dimension the first type parameter of QuadratureRule have been removed. (#711, #716) To upgrade replace e.g. QuadratureRule{1, RefCube}(order) with QuadratureRule{RefLine}(order), and QuadratureRule{2, RefTetrahedron}(1) with Lagrange{RefTriangle}(order), etc.

  • CellScalarValues and CellVectorValues have been merged into CellValues, FaceScalarValues and FaceVectorValues have been merged into FacetValues, and PointScalarValues and PointVectorValues have been merged into PointValues. The differentiation between scalar and vector have thus been moved to the interpolation (see above). Note that previously CellValues, FaceValues, and PointValues where abstract types, but they are now concrete implementations with different type parameters, except FaceValues which is now FacetValues (#708) To upgrade, for scalar problems, it is enough to replace CellScalarValues with CellValues, FaceScalarValues with FacetValues and PointScalarValues with PointValues, respectively. For vector problems, make sure to vectorize the interpolation (see above) and then replace CellVectorValues with CellValues, FaceVectorValues with FacetValues, and PointVectorValues with PointValues.

  • The quadrature rule passed to FacetValues should now be of type FacetQuadratureRule rather than of type QuadratureRule. (#716) To upgrade replace the quadrature rule passed to FacetValues with a FacetQuadratureRule.

  • Checking if a face (ele_id, local_face_id) ∈ faceset has been previously implemented by type piracy. In order to be invariant to the underlying Set datatype as well as omitting type piracy, (#835) implemented isequal and hash for BoundaryIndex datatypes.

  • VTK export: Ferrite no longer extends WriteVTK.vtk_grid and associated functions, instead the new type VTKGridFile should be used instead. New methods exists for writing to a VTKGridFile, e.g. write_solution, write_cell_data, write_node_data, and write_projection. See #692.

  • Definitions: Previously, face and edge referred to codimension 1 relative reference shape. In Ferrite v1, volume, face, edge, and vertex refer to 3, 2, 1, and 0 dimensional entities, and facet replaces the old definition of face. No direct replacement for edges exits. See #914 and #914. The main implications of this change are

    • FaceIndex -> FacetIndex (FaceIndex still exists, but has a different meaning)
    • FaceValues -> FacetValues
    • nfaces -> nfacets (nfaces is now an internal method with different meaning)
    • addfaceset! -> addfacetset
    • getfaceset -> getfacetset

    Furthermore, subtypes of Interpolation should now define vertexdof_indices, edgedof_indices, facedof_indices, volumedof_indices (and similar) according to these definitions.

  • Ferrite.getdim has been changed into Ferrite.getrefdim for getting the dimension of the reference shape and Ferrite.getspatialdim to get the spatial dimension (of the grid). (#943)

  • Ferrite.getfielddim(::AbstractDofHandler, args...) has been renamed to Ferrite.n_components. (#943)

  • The constructor for ExclusiveTopology only accept an AbstractGrid as input, removing the alternative of providing a Vector{<:AbstractCell}, as knowing the spatial dimension is required for correct code paths. Furthermore, it uses a new internal data structure, ArrayOfVectorViews, to store the neighborhood information more efficiently The datatype for the neighborhood has thus changed to a view of a vector, instead of the now removed EntityNeighborhood container. This also applies to vertex_star_stencils. (#974).

  • project(::L2Projector, data, qr_rhs) now expects data to be indexed by the cellid, as opposed to the index in the vector of cellids passed to the L2Projector. The data may be passed as an AbstractDict{Int, <:AbstractVector}, as an alternative to AbstractArray{<:AbstractVector}. (#949)

Deprecated

  • The rarely (if ever) used methods of function_value, function_gradient, function_divergence, and function_curl taking vectorized dof values as in put have been deprecated. (#698)

  • The function reshape_to_nodes have been deprecated in favor of evaluate_at_grid_nodes. (#703)

  • start_assemble([n::Int]) has been deprecated in favor of calling Ferrite.COOAssembler() directly (#916, #1058).

  • start_assemble(f, K) have been deprecated in favor of the "canonical" start_assemble(K, f). (#707)

  • assemble!(assembler, dofs, fe, Ke) have been deprecated in favor of the "canonical" assemble!(assembler, dofs, Ke, fe). (#1059)

  • end_assemble have been deprecated in favor of finish_assemble. (#754)

  • get_point_values have been deprecated in favor of evaluate_at_points. (#754)

  • transform! have been deprecated in favor of transform_coordinates!. (#754)

  • Ferrite.default_interpolation has been deprecated in favor of geometric_interpolation. (#953)

Removed

  • MixedDofHandler + FieldHandler have been removed in favor of DofHandler + SubDofHandler. Note that the syntax has changed, and note that SubDofHandler is much more capable compared to FieldHandler. Previously it was often required to pass both the MixedDofHandler and the FieldHandler to e.g. the assembly routine, but now it is enough to pass the SubDofHandler since it can be used for e.g. DoF queries etc. (#624, #667, #735)

  • Some old methods to construct the L2Projector have been removed after being deprecated for several releases. (#697)

  • The option project_to_nodes have been removed from project(::L2Projector, ...). The returned values are now always ordered according to the projectors internal DofHandler. (#699)

  • The function compute_vertex_values have been removed. (#700)

  • The names getweights, getpoints, getcellsets, getnodesets, getfacesets, getedgesets, and getvertexsets have been removed from the list of exported names. (For now you can still use them by prefixing Ferrite., e.g. Ferrite.getweights.) (#754)

  • The onboundary function (and the associated boundary_matrix property of the Grid datastructure) have been removed (#924). Instead of first checking onboundary and then check whether a facet belong to a specific facetset, check the facetset directly. For example:

    - if onboundary(cell, local_face_id) && (cell_id, local_face_id) in getfacesets(grid, "traction_boundary")
+ if (cell_id, local_face_id) in getfacesets(grid, "traction_boundary")
     # integrate the "traction_boundary" boundary
  end

Fixed

  • Benchmarks now work with master branch. (#751, #855)

  • Topology construction have been generalized to, in particular, fix construction for 1D and for wedge elements. (#641, #670, #684)

Other improvements

  • Documentation:

    • The documentation is now structured according to the Diataxis framework. There is now also clear separation between tutorials (for teaching) and code gallery (for showing off). (#737, #756)
    • New section in the developer documentation that describes the (new) reference shapes and their numbering scheme. (#688)

  • Performance:

    • Ferrite.transform!(grid, f) (for transforming the node coordinates in the grid according to a function f) is now faster and allocates less. (#675)
    • Slight performance improvement in construction of PointEvalHandler (faster reverse coordinate lookup). (#719)
    • Various performance improvements to topology construction. (#753, #759)

  • Internal improvements:

    • The dof distribution interface have been updated to support higher order elements (future work). (#627, #732, #733)
    • The AbstractGrid and AbstractDofHandler interfaces are now used more consistently internally. This will help with the implementation of distributed grids and DofHandlers. (#655)
    • VTK export now uses the (geometric) interpolation directly when evaluating the finite element field instead of trying to work backwards how DoFs map to nodes. (#703)
    • Improved bounds checking in assemble!. (#706)
    • Internal methods Ferrite.value and Ferrite.derivative for computing the value/gradient of all shape functions have been removed. (#720)
    • Ferrite.create_incidence_matrix now work with any AbstractGrid (not just Grid). (#726)

v0.3.14 - 2023-04-03

Added

  • Support reordering dofs of a MixedDofHandler by the built-in orderings FieldWise and ComponentWise. This includes support for reordering dofs of fields on subdomains. (#645)
  • Support specifying the coupling between fields in a MixedDofHandler when creating the sparsity pattern. (#650)
  • Support Metis dof reordering with coupling information for MixedDofHandler. (#650)
  • Pretty printing for MixedDofHandler and L2Projector. (#465)

Other improvements

  • The MixedDofHandler have gone through a performance review (see #629) and now performs the same as DofHandler. This was part of the push to merge the two DoF handlers. Since MixedDofHandler is strictly more flexible, and now equally performant, it will replace DofHandler in the next breaking release. (#637, #639, #642, #643, #656, #660)

Internal changes

Changes listed here should not affect regular usage, but listed here in case you have been poking into Ferrite internals:

  • Ferrite.ndim(dh, fieldname) has been removed, use Ferrite.getfielddim(dh, fieldname) instead. (#658)
  • Ferrite.nfields(dh) has been removed, use length(Ferrite.getfieldnames(dh)) instead. (#444, #653)
  • getfielddims(::FieldHandler) and getfieldinterpolations(::FieldHandler) have been removed (#647, #659)

v0.3.13 - 2023-03-23

Added

  • Support for classical trilinear and triquadratic wedge elements. (#581)
  • Symmetric quadrature rules up to order 10 for prismatic elements. (#581)
  • Finer granulation of dof distribution, allowing to distribute different amounts of dofs per entity. (#581)

Fixed

  • Dof distribution for embedded elements. (#581)
  • Improve numerical accuracy in shape function evaluation for the Lagrange{2,Tetrahedron,(3|4|5)} interpolations. (#582, #633)

Other improvements

  • Documentation:
    • New "Developer documentation" section in the manual for documenting Ferrite.jl internals and developer tools. (#611)
    • Fix a bug in constraint computation in Stoke's flow example. (#614)
  • Performance:
    • Benchmarking infrastructure to help tracking performance changes. (#388)
    • Performance improvements for various accessor functions for MixedDofHandler. (#621)

Internal changes

  • To clarify the dof management vertices(ip), edges(ip) and faces(ip) has been deprecated in favor of vertexdof_indices(ip), edgedof_indices(ip) and facedof_indices(ip). (#581)
  • Duplicate grid representation has been removed from the MixedDofHandler. (#577)

v0.3.12 - 2023-02-28

Added

  • Added a basic show method for assemblers. (#598)

Fixed

  • Fix an issue in constraint application of Symmetric-wrapped sparse matrices (i.e. obtained from create_symmatric_sparsity_pattern). In particular, apply!(K::Symmetric, f, ch) would incorrectly modify f if any of the constraints were inhomogeneous. (#592)
  • Properly disable the Metis extension on Julia 1.9 instead of causing precompilation errors. (#588)
  • Fix adding Dirichlet boundary conditions on nodes when using MixedDofHandler. (#593, #594)
  • Fix accidentally slow implementation of show for Grids. (#599)
  • Fixes to topology functionality. (#453, #518, #455)
  • Fix grid coloring for cell sets with 0 or 1 cells. (#600)

Other improvements

  • Documentation improvements:
    • Simplications and clarifications to hyperelasticity example. (#591)
    • Remove duplicate docstring entry for vtk_point_data. (#602)
    • Update documentation about initial conditions. (#601, #604)

v0.3.11 - 2023-01-17

Added

  • Metis.jl extension for fill-reducing DoF permutation. This uses Julias new package extension mechanism (requires Julia 1.10) to support a new DoF renumbering order DofOrder.Ext{Metis}() that can be passed to renumber! to renumber DoFs using the Metis.jl library. (#393, #549)
  • BlockArrays.jl extension for creating a globally blocked system matrix. create_sparsity_pattern(BlockMatrix, dh, ch; kwargs...) return a matrix that is blocked by field (requires DoFs to be (re)numbered by field, i.e. renumber!(dh, DofOrder.FieldWise())). For custom blocking it is possible to pass an uninitialized BlockMatrix with the correct block sizes (see BlockArrays.jl docs). This functionality is useful for e.g. special solvers where individual blocks need to be extracted. Requires Julia version 1.9 or above. (#567)
  • New function apply_analytical! for setting the values of the degrees of freedom for a specific field according to a spatial function f(x). (#532)
  • New cache struct CellCache to be used when iterating over the cells in a grid or DoF handler. CellCache caches nodes, coordinates, and DoFs, for the cell. The cache cc can be re-initialized for a new cell index ci by calling reinit!(cc, ci). This can be used as an alternative to CellIterator when more control over which element to loop over is needed. See documentation for CellCache for more information. (#546)
  • It is now possible to create the sparsity pattern without constrained entries (they will be zeroed out later anyway) by passing keep_constrained=false to create_sparsity_pattern. This naturally only works together with local condensation of constraints since there won't be space allocated in the global matrix for the full (i.e. "non-condensed") element matrix. Creating the matrix without constrained entries reduces the memory footprint, but unless a significant amount of DoFs are constrained (e.g. high mesh resolution at a boundary) the savings are negligible. (#539)

Changed

  • ConstraintHandler: update! is now called implicitly in close!. This was easy to miss, and somewhat of a strange requirement when solving problems without time stepping. (#459)
  • The function for computing the inhomogeneity in a Dirichlet constraint can now be specified as either f(x) or f(x, t), where x is the spatial coordinate and t the time. (#459)
  • The elements of a CellIterator are now CellCache instead of the iterator itself, which was confusing in some cases. This change does not affect typical user code. (#546)

Deprecated

  • Adding fields to a DoF handler with push!(dh, ...) has been deprecated in favor of add!(dh, ...). This is to make it consistent with how constraints are added to a constraint handler. (#578)

Fixed

  • Fix shape_value for the linear, discontinuous Lagrange interpolation. (#553)
  • Fix reference_coordinate dispatch for discontinuous Lagrange interpolations. (#559)
  • Fix show(::Grid) for custom cell types. (#570)
  • Fix apply_zero!(Δa, ch) when using inhomogeneous affine constraints (#575)

Other improvements

  • Internal changes defining a new global matrix/vector "interface". These changes make it easy to enable more array types (e.g. BlockMatrix support added in this release) and solvers in the future. (#562, #571)
  • Performance improvements:
    • Reduced time and memory allocations for global sparse matrix creation (Julia >= 1.10). (#563)
  • Documentation improvements:
    • Added an overview of the Examples section. (#531)
    • Added an example showing topology optimization. (#531)
    • Various typo fixes. (#574)
    • Fix broken links. (#583)

v0.3.10 - 2022-12-11

Added

  • New functions apply_local! and apply_assemble! for applying constraints locally on the element level before assembling to the global system. (#528)
  • New functionality to renumber DoFs by fields or by components. This is useful when you need the global matrix to be blocked. (#378, #545)
  • Functionality to renumber DoFs in DofHandler and ConstraintHandler simultaneously: renumber!(dh::DofHandler, ch::ConstraintHandler, order). Previously renumbering had to be done before creating the ConstraintHandler since otherwise DoF numbers would be inconsistent. However, this was inconvenient in cases where the constraints impact the new DoF order permutation. (#542)
  • The coupling between fields can now be specified when creating the global matrix with create_sparsity_pattern by passing a Matrix{Bool}. For example, in a problem with unknowns (u, p) and corresponding test functions (v, q), if there is no coupling between p and q it is unnecessary to allocate entries in the global matrix corresponding to these DoFs. This can now be communicated to create_sparsity_pattern by passing the coupling matrix [true true; true false] in the keyword argument coupling. (#544)

Changed

  • Runtime and allocations for application of boundary conditions in apply! and apply_zero! have been improved. As a result, the strategy keyword argument is obsolete and thus ignored. (#489)
  • The internal representation of Dirichlet boundary conditions and AffineConstraints in the ConstraintHandler have been unified. As a result, conflicting constraints on DoFs are handled more consistently: the constraint added last to the ConstraintHandler now always override any previous constraints. Conflicting constraints could previously cause problems when a DoF where prescribed by both Dirichlet and AffineConstraint. (#529)
  • Entries in local matrix/vector are now ignored in the assembly procedure. This allows, for example, using a dense local matrix [a b; c d] even if no entries exist in the global matrix for the d block, i.e. in [A B; C D] the D block is zero, and these global entries might not exist in the sparse matrix. (Such sparsity patterns can now be created by create_sparsity_pattern, see #544.) (#543)

Fixed

  • Fix affine constraints with prescribed DoFs in the right-hand-side. In particular, DoFs that are prescribed by just an inhomogeneity are now handled correctly, and nested affine constraints now give an error instead of silently giving the wrong result. (#530, #535)
  • Fixed internal inconsistency in edge ordering for 2nd order RefTetrahedron and RefCube. (#520, #523)

Other improvements

  • Performance improvements:
    • Reduced time and memory allocations in DoF distribution for MixedDofHandler. (#533)
    • Reduced time and memory allocations reductions in getcoordinates!. (#536)
    • Reduced time and memory allocations in affine constraint condensation. (#537, #541, #550)
  • Documentation improvements:
    • Use :static scheduling for threaded for-loop (#534)
    • Remove use of @inbounds (#547)
  • Unification of create_sparsity_pattern methods to remove code duplication between DofHandler and MixedDofHandler. (#538, #540)

v0.3.9 - 2022-10-19

Added

  • New higher order function interpolations for triangles (Lagrange{2,RefTetrahedron,3}, Lagrange{2,RefTetrahedron,4}, and Lagrange{2,RefTetrahedron,5}). (#482, #512)
  • New Gaussian quadrature formula for triangles up to order 15. (#514)
  • Add debug mode for working with Ferrite internals. (#524)

Changed

  • The default components to constrain in Dirichlet and PeriodicDirichlet have changed from component 1 to all components of the field. For scalar problems this has no effect. (#506, #509)

v0.3.8 - 2022-10-05

Added

  • Ferrite.jl now has a logo! (#464)
  • New keyword argument search_nneighbors::Int in PointEvalHandler for specifying how many neighboring elements to consider in the kNN search. The default is still 3 (usually sufficient). (#466)
  • The IJV-assembler now support assembling non-square matrices. (#471)
  • Periodic boundary conditions have been reworked and generalized. It now supports arbitrary relations between the mirror and image boundaries (e.g. not only translations in x/y/z direction). (#478, #481, #496, #501)

Fixed

  • Fix PointEvalHandler when the first point is missing. (#466)
  • Fix the ordering of nodes on the face for (Quadratic)Tetrahedron cells. (#475)

Other improvements

  • Many improvements to the documentation. (#467, #473, #487, #494, #500)
  • Improved error messages in reinit! when number of geometric base functions and number of element coordinates mismatch. (#469)
  • Remove some unnecessary function parametrizations. (#503)
  • Remove some unnecessary allocations in grid coloring. (#505)
  • More efficient way of creating the sparsity pattern when using AffineConstraints and/or PeriodicDirichlet. (#436)

v0.3.7 - 2022-07-05

Fixed

  • Fix tests for newer version of WriteVTK (no functional change). (#462)

Other improvements

  • Various improvements to the heat equation example and the hyperelasticity example in the documentation. (#460, #461)

v0.3.6 - 2022-06-30

Fixed

  • Fix a bug with L2Projection of mixed grid. (#456)

Other improvements

  • Expanded manual section of Dirichlet BCs. (#458)

v0.3.5 - 2022-05-30

Added

  • Functionality for querying information about the grid topology (e.g. neighboring cells, boundaries, ...). (#363)

Fixed

  • Fix application of boundary conditions when combining RHSData and affine constraints. (#431)

v0.3.4 - 2022-02-25

Added

  • Affine (linear) constraints between degrees-of-freedom. (#401)
  • Periodic Dirichlet boundary conditions. (#418)
  • Evaluation of arbitrary quantities in FE space. (#425)

Changed

  • Interpolation(s) and the quadrature rule are now stored as part of the CellValues structs (cv.func_interp, cv.geo_interp, and cv.qr). (#428)

v0.3.3 - 2022-02-04

Changed

  • Verify user input in various functions to eliminate possible out-of-bounds accesses. (#407, #411)

v0.3.2 - 2022-01-18

Added

  • Support for new interpolation types: DiscontinuousLagrange, BubbleEnrichedLagrange, and CrouzeixRaviart. (#352, #392)

Changed

  • Julia version 1.0 is no longer supported for Ferrite versions >= 0.3.2. Use Julia version >= 1.6. (#385)
  • Quadrature data for L2 projection can now be given as a matrix of size "number of elements" x "number of quadrature points per element". (#386)
  • Projected values from L2 projection can now be exported directly to VTK. (#390)
  • Grid coloring can now act on a subset of cells. (#402)
  • Various functions related to cell values now use traits to make it easier to extend and reuse functionality in external code. (#404)

Fixed

  • Exporting tensors to VTK now use correct names for the components. (#406)