GitLab

Resolve "How to pass a python grid function back to C++?"

Closes #70

See merge request !358

Fix typos

See merge request !356

Remove unused-parameter warnings

See merge request !353

Improve the documentation of grid-functions

See merge request !345

Make derivative of AnalyticGridFunction bound to an element if the GridFunction itself is bound

See merge request !355

Add assertion in localIndex for out of bound indices

See merge request !352

Use math.hh's power instead of StaticPower

See merge request !346

Constrain the constructor of DefaultGlobalBasis that gets a factory

Closes #67

See merge request !349

USE HAVE_DUNE_UGGRID instead of removed HAVE_UG

See merge request !344

Make DiscreteGlobalBasisFunction work non-istl range types like array and TupleVector

See merge request !343

Cleanup poisson-mfem

See merge request !341

This allows to also use types like `TupleVector` as range.

This way assignment also works for types like `TupleVector`

While the pressure is piecewise constant for a mixed RT0
discretization, writing it as vertex data is much more
helpful, because paraview cannot use 'warp by scalar'
for cell data.

* Fix quadrature order for operator assembly.
* Improve boundary DOF marking. Mark all flux DOFs
  associated to subentities of faces contain in a
  boundary patch. As before the patch is determined using
  an indicator function. This avoids having to interpolate
  special functions to filter out the boundary DOFs.
* Describe boundary normal flux as multiple of boundary vector.
  The previous version used a scalar function `f` which will implicitly
  be multiplied by the one vector `(1,...,1)`. Hence the interpolated
  normal flux is `f(x) <(1,...,1),n(x)> n(x)` insetad of the probably
  intended `f(x)n(x)`.

Add FaceNormalGridFunction

See merge request !340

This implements element normals as (discontinuous) grid function.
When evaluated at a point `x` of an element `e`, it computes the
unit outward normal of the 'closest face' of this element.

* The 'closes face' is determined in local coordinates within the
  reference elements.
* The implementation also works for curved faces, since the normal
  of the real element is computed using the (potentially non-constant)
  covariant Piola transformation of the (constante) normal in the
  reference element.
* Notice that the function is an L^2 function in general. It is
  defined piecewise with non-unique value on intersections.
  On elements, with affine geometries, it is piecewise constant.
* This also adds a test checking for consistency with
  `intersection.unitOuterNormal()` when restricting the function
  to `intersection`. A grid with distorted cubes serves as test case
  for non-constant normals.
* This satisfied the standard grid function interface. Extending this
  to an interface for skeleton functions to be bound on intersections
  may be interesting. But such an interface was not defined so far.

Avoid nesting of SubspaceBasis by resolving nested ones

See merge request !337

[bugfix] Fix jacobian part of Piola transformations

See merge request !338

Since `jacobian(Inverse)Transposed` may be a sparse matrix
(e.g. `DiagonalMatrix` for `YaspGrid`) we must not use
`jacobian(Inverse)Transposed[i][j]`. Instead iterate over
the sparse row entries using `sparseRange`.

Also test with `YaspGrid` and higher order RT elements.
Checking against `YaspGrid` reveales a bug: The jacobian
Piola transformations assume that `jacobian(Inverse)Transposed`
is dense, which may not be the case.