View Issue Details
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0001468||OpenFOAM||Bug||public||2014-12-25 10:18||2015-05-01 00:49|
|Platform||Desktop and Cluster||OS||Linux Mint / SUSE||OS Version||17 / Enterprise|
|Summary||0001468: Something wrong in the CrankNicolson ddt scheme (2.3.x)|
|Description||When running LES calculation on 2.3.x with CrankNicolson ddt scheme, |
maximum courant number gradually increases and finally calculation blows.
I cannot see the problem with backward ddt scheme or CrankNicolson in "2.2.x".
So, I can guess any changes between 2.2.x and 2.3.x causes the problem.
In addition, I have been faced with the problem since 2.3.0 released.
So, It might not be caused by recent commitment.
|Steps To Reproduce||Running tutorial case (pisoFoam/pitzDaily) with:|
default CrankNicolson 1;
Initial residuals of each velocity component remains higher (near 0.9) and courant number starts weaving.
Finally, calculation will stop at Time=0.02656 (in my case).
|Tags||No tags attached.|
Thanks for the report.
In OpenFOAM-2.3.? I changed ddtCorr to handle mesh-motion in a more general way, accurately including the effects of face rotation as well as translation. However, it was not clear that this approach needed the Crank-Nicolson corrections and tests showed that the Euler ddtCorr handling was sufficient for the Crank-Nicolson scheme. But these tests were done using a small amount of off-centering:
default CrankNicolson 0.9;
which has been found necessary in order to stabilize this scheme for complex flows, providing similar accuracy and stability as the backward scheme. I have now run tests without off-centering as you are running and also find that for cases where this is stable in OpenFOAM-2.2.x it is not stable in OpenFOAM-2.3.x, unless the Crank-Nicolson corrections to ddtCorr are reinstated. I have now pushed this change to OpenFOAM-2.3.x.
Resolved by commit 9a79633f0723bd714aa25f3eec21e4d90623ee90