Remove an unused type

See merge request agnumpde/dune-elasticity!26

Replace local energies by local densities

See merge request agnumpde/dune-elasticity!24

 - wrap the densities by a generic Integral energy
 - this preserves all material properties
 - reduces redundant code

Fix/damping

See merge request agnumpde/dune-elasticity!25

Various minor improvements by Lisa

See merge request agnumpde/dune-elasticity!23

Dune does not support this anymore post version 2.7.

lisa_julia.nebel_at_tu-dresden.de authored Apr 16, 2020 and oliver.sander_at_tu-dresden.de committed Apr 16, 2020

Otherwise distributed computations lead to a lot of cluttering.

lisa_julia.nebel_at_tu-dresden.de authored Apr 16, 2020 and oliver.sander_at_tu-dresden.de committed Apr 16, 2020

Mainly issues related to dune-parmg.

lisa_julia.nebel_at_tu-dresden.de authored Apr 16, 2020 and oliver.sander_at_tu-dresden.de committed Apr 16, 2020

Previously the tape number was hard-wired to '1'.  Using the
process number instead avoid two processes writing to the
same tape file.

lisa_julia.nebel_at_tu-dresden.de authored Apr 16, 2020 and oliver.sander_at_tu-dresden.de committed Apr 16, 2020

In distributed situations, assembly has to happen on interior elements
only.  Select this partition type in a single location; previously
it was set several times in different places.

Fix adolcmaterialtest.cc

See merge request agnumpde/dune-elasticity!20

The test erroneously checked for a small *absolute* error, but the
numbers in this test are very large (around 1e8), so checking
for a small *relative* error is more appropriate.  This change
makes the test pass.

The clang 6 CI images does not exist anymore.

Update the dune-elasticity build system some more

See merge request agnumpde/dune-elasticity!16

The dune-elasticity module dates from an era where that file
was not needed.

In particular:
* Introduce enable_all_packages()
* Introduce (empty) subdirectory 'cmake/modules'

Nobody I know still uses AmiraMesh.

It simplifies the code a little.

It was empty anyway, and all implementations were always required
to implement the methods.  With the new LocalEnergy interface
class this can now be properly documented in the interface.

... rather than from LocalFEStiffness.  The latter is for
implementations that provide first and second derivatives,
which none of the materials do.

As the name says, it is really only an energy, and nothing more.

... rather than a LocalFEStiffness.  Because it really only needs
the energy evaluation, and implements the rest itself.

This is supposed to the base class for everything that can
integrate energy densities over one grid element.

LocalFEStiffness inherits from the new LocalEnergy.

That will always be false, even if the underlying bases
are the same.

This is a fixup to commit 8abd4e13

This patch uses the multigrid implementation from the dune-parmg
module for distributed solution of finite-strain elasticity
problems.  It is still a bit rough about the edges, and there
are still bugs concerning coarse grid obstacles.  These will
hopefully be fixed in the nearer future.

All this new code is optional: if dune-parmg is not present,
then the old sequential code will be used.

In that case, the correction is a fortiori smaller than the
tolerance, and the program will stop.  However, it does not
mean that we have reached the solution.  Therefore, be
verbose about the reason for termination.

* Sum energy across all processes
* Compute gradient norm across all processes
* Screen output only on rank 1