This is a follow-up to PR #211. The adjustment to finish exactly at end_time is good for most cases, but adjusting the target timestep size does not work well for RRK time refinement studies. Therefore, I am proposing to add a parameter to make this optional, and have adjusted the tests likewise.
The issue arises in the following case:
Target timestep is decreased to reach final_time exactly.
Adjusted timestep (from RRK) is smaller than target timestep.
A second (very) small timestep is taken to reach final_time exactly.
This results in an extra timestep compared to the old behaviour, which invalidates the average relaxation parameter calculation.
This is a follow-up to PR #211. The adjustment to finish exactly at
end_timeis good for most cases, but adjusting the target timestep size does not work well for RRK time refinement studies. Therefore, I am proposing to add a parameter to make this optional, and have adjusted the tests likewise.The issue arises in the following case:
The following text file demonstrates the behaviour: convergence_behaviour.txt .
I've also included some improvements/corrections to the time refinement study tests from my branch in PR #195.