BOUT-dev
BOUT-dev copied to clipboard
boutproject
Issues with LaplacePDD
Does anyone use LaplacePDD? I think there's at least one bug in it:
https://github.com/boutproject/BOUT-dev/blob/123e774f865d4c98d378b909b89accd7efd56dc3/src/invert/laplace/impls/pdd/pdd.cxx#L172-L174
data.jy should probably come before kz. I only noticed because I'm getting a warning about converting kz * kwaveFactor to an int which definitely smells wrong.
Secondly, test-laplace fails if I use PDD (and run it only with nproc=4), with max differences between 0.1 and 10. This requires the above fix, or else it throws out-of-bounds.
The header to PDD says it neglects "small" corrections and has a big warning:
CHECK IF THIS ALGORITHM PRODUCES REASONABLE RESULTS FOR YOUR PROBLEM
This suggests it probably needs its own tests, and its assumptions stating up front. I guess test-laplace violates these assumptions somehow.
This was a test, the paper promised great things, but I don't think it ever worked. The bug probably didn't help, but I suspect it's just not a very good method. I vote for removing in 5
Do we still want to remove LaplacePDD? We never "officially" deprecated it
Yes I vote for removing it. Joseph's methods are certainly better and probably faster