The underlying algorithm for detecting a steady state solution while using the Transient Executioner has changed.
David Andrs made the changes in this PR
The basic issue is that the old relative norm algorithm was wrong. It was doing:
abs( norm(u_old) - norm(u_new) ) / norm(u_new)
That is definitely wrong! The new algorithm is:
norm(u_old - u_new) / norm(u_new)
Which is a "proper" relative norm.
I don't expect this to actually change much for people using this capability. The old algorithm had the possibility of generating false-positives (i.e. it would "detect" a steady state that wasn't steady!) so the new algorithm should yield better results.
A good side effect of David's work is that you can now output this relative norm as a Postprocessor.
While we're on the subject I also want to draw some attention to an Issue I created yesterday about renaming the parameters for steady state detection. Definitely go comment on that issue if you use the current parameters and have an opinion!Share on Twitter Share on Facebook