precision difference OF-2.3.0 and OF-2.3.1
 

Hello,

for some special analysis I need a very high numerical precision of of the flow field (so, double precision everywhere). And accidentally, I came along a small difference between the versions OF-2.3.0 and OF-2.3.1.
Running the same code in the same case, I get max. velocities which differ after 6 digits.
Some example results:

code compiled with OF-2.3.0
Umax max(mag(U)) [0 1 -1 0 0 0 0] 1.90072158967443
Umax max(mag(U)) [0 1 -1 0 0 0 0] 1.90072234086763
Umax max(mag(U)) [0 1 -1 0 0 0 0] 1.9007235882843

code compiled with OF-2.3.1
Umax max(mag(U)) [0 1 -1 0 0 0 0] 1.90072222726156
Umax max(mag(U)) [0 1 -1 0 0 0 0] 1.90072298134827
Umax max(mag(U)) [0 1 -1 0 0 0 0] 1.9007242315462

I compiled the two different OF-versions on the same machine, using the same compiler, but I always get these small differences (in serial and parallel). I searched in the src folder (using meld), but didn't find the files which could explain this difference.

Does anyone has an idea?

Thanks,
Matze

PS: It is a simple 2D airfoil using the SIMPLE algorithm and inletOutlet BCs on a circular domain. writePrecision is set to 15.

After running several more simulations and checking the codes again, I found that the relevant difference is the delta() function in
src/finiteVolume/fvMesh/fvPatches/fvPatch/fvPatch.C

OF-2.3.0:
return Cf() - Cn();

OF-2.3.1:
// Use patch-normal delta for all non-coupled BCs
const vectorField nHat(nf());
return nHat*(nHat & (Cf() - Cn()));

So, the difference is that in OF-2.3.1 only the normal part of the cell-centre to face-centre vector is used.
As far as I know, this delta is used for the computation of the Laplacian term, but I'm still not sure which one is correct and why this change was done...
Any ideas?