The default linear solver is set to AMG. The solver is parallel in contrast to the current default sequential solver and shows the best performance in a wide range of problems.
After changing this, some tests failed. If the differences in the solution were in the order of machine epsilon, the zero threshold value has been adjusted.
In the remaining cases the origin of the tests to fail after changing the linear solver could not be detected and the original linear solver has been set as a property for these tests. (it quite possibly has to be a bug in the models as the same jacobian should lead to the same solution, the newton convergence criteria were tried in the most precise configuration and there were still differences)
TODO
-
test corner point grid
This fixes #325 (closed).