The former implementation did not produce expected results for 2D and 3D case. Rewriting and implementing the Jacobian for the Newton scheme. Keeping K matrix for the Picard scheme as applying the BC to the r.h.s. only significantly increases number of non-linear iterations because the problem becomes non-linear.
A method of manufactured solutions is used to verify implementation of the Robin-type boundary condition. Grid convergence was assessed and the observed order of convergence is 1.99, which, as expected, is close enough to 2.
In the same setup a Neumann and Dirichlet-type boundary conditions can (and were) tested with same convergence order as the Robin-type BC.
The test is testing both non-linear solver schemes, Newton and Picard. With the new implementation of the Robin-type BC the convergence rates of the non-linear solver are satisfying, usually 2 iterations per time step for both schemes.
Four tests testing the axisymmetry were updated, three of them verified against a given manufactured solution, and the TES one is relying on the numerical results, i.e. the 3D solution was mapped to 2D axisymmetric case and is used as reference.
-
Feature description was added to the changelog -
Tests covering your feature were added?
Review commit-wise. Many details are provided in the commit messages.