Ned Coltman authored 4 years ago and Timo Koch committed 4 years ago

* Loads in solution from rans2c for a rans2cni test
* Tests rans2cni faster, plus loading a partial solution for ICs

This commit also modifies the tracer_tracer test, as this failed for box
after removing the elembctypes. This is because it was using Neumann
no-flow boundary conditions for the fracture domain on the left boundary
while the bulk problem used Dirichlet BCs. The old elembctypes for box
were discarding Dirichlet BCs on those nodes on the boundary that also
couple to fractures in order for the coupling conditions to be
fulfilled. However, that discards the boundary condition. With this
commit, Dirichlet BCs are always fulfilled, but for the tracer test this
means that as a consequence, the pressure in the fracture on the left
side is equal to the bulk pressure - because it uses Dirichlet-type
coupling conditions. This commit changes the test such that it uses
Dirichlet BCs in the bulk domain only on the lower part of the boundary
such that the fracture touches the domain boundary at a no-flow
boundary.

* Reduced inital dt
* Fixed timesteps
* Adapted the reference solution

Johannes Hommel authored 5 years ago and Dennis Gläser committed 4 years ago

with contributions by Felix Weinhardt
and some proof reading by Timo Koch and Dennis Gläser

The distribution of an unstructured across several processes is
not stable in the sense that a process will always get the same
elements. Not over several Dune versions and possibly even not
between different machines. Threfore, don't compare the solution
for a single process.

Theresa Schollenberger authored 4 years ago and Timo Koch committed 4 years ago

[test][2pncminni] increase runtime to test precipitation, use fixed time-step size increment, adapt reference

[examples][swe][fix] Update reference solution after change of boundary conditions in 8732166a

Update 2ptracer reference solution after correcting derivative

Timo Koch authored 5 years ago and Ned Coltman committed 5 years ago

The idea of the tracer model is really tied to Fickian diffusion and components
with small concentration diffusing in a bulk phase. Hence also it's name.
The Maxwell-Stefan test actual models 3 gaseous components where none of
the components is really the main phase but most importantly there is no
bulk phase with a main component that is not considered in the equations.
Thus the 1pnc model is the right choice for such a test.
This changes the test to derive from the 1pnc model. As for the given
setting with only Neumann boundaries the 1pnc wouldn't convergence for
a compressible fluid, the Ideal gas law is used to model a
compressible gas.

The same test actually exists for the 1pnc model already in 1pnc/implicit/1p3c.
Thus this commit simply removes the tracer test.

Simon Emmert authored 5 years ago and Timo Koch committed 5 years ago

with changes over processor boundaries adapt other tests to match new injection rate, harmonize timesteps with lower DtInitial and MaxTimeStepSize

Melanie Lipp authored 5 years ago and Timo Koch committed 5 years ago

        stationary, outflow at outlet, block profile at inlet.

Dennis Gläser authored 5 years ago and Timo Koch committed 5 years ago

Actually use the surface grid. Update references.

the correct state initialization in the Newton method
results in a new solution