Performance of BlockNonlinearForm compared to NonlinearForm

[thelfer]

I just wanted to know if anyone already tested if using a BlockNonlinearForm has a performance hit compared to a NonlinearForm for problems with a unique unknowns.

The context of this question is that I am writting an application on top of MFEM for (generalised) non linear solid mechanics. So I will solve problem with only one unknowns (the displacement), and other with several unknowns (mechanics strongly coupled with heat transfer for example). It would be nice to avoid code duplication, so I wondered if building all my code around BlockNonlinearForm from the start would be a good idea, whatever the problem treated.

Thanks for any advice !

[jandrej]

There should be no performance overhead using BlockNonlinearForm. We have no formal benchmarks to present though.

[thelfer]

@jandrej Thanks for the anwser.

As a corollary, is there any advantage of using NonlinearForm rather than BlockNonlinearForm? Sorry if the question is a bit rude

[jandrej]

BlockNonlinearForm evolved out of the necessity to have a more convenient interface for multi-variable problems. If you only have a problem with one variable you could use NonlinearForm to avoid the Array notation. I don't see any other major advantage.

[thelfer]

@jandrej Thanks again. I'll thus change my app to use BlockNonlinearForm for each case.