Cyclic Non-Convergence in Parallel PimpleDyMFoam

Collin · February 27, 2017, 21:43

Hello,

I have run across an issue where a single-core process of PimpleDyMFoam converges at a particular time step, yet the multi-core process has a cyclic non-convergence issue. (In my case the pressure converges until it trips a new iteration of omega, that new omega causes the pressure convergence to start over; this process repeats ad infinitum).

I have attempted changing the relaxation factors, the number of correctors, and the pressure relTol. None yielded an improvement.

Everything is identical between these two cases except that one is being run in parallel while the other is not. I am running on foam-extend-3.2.

Any suggestions for other things to try?

Collin · April 13, 2017, 21:42

As it turns out, when you use "patch-constrained" decomposition, not all point which contact the constrained patch move uniformly. Periodic points would move at half of the intended rate (those points happened to be those which corresponded to a joining of 3 cells at the surface rather than 2 cells (for an extruded semi-2D case)).

Thus, if anyone comes across this issue in the future, the solution is to not use patch-constrained decomposition in conjunction with mesh motion.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Problem with cyclic boundary conditions in parallel	meng9019	OpenFOAM Running, Solving & CFD	3	February 5, 2019 10:11
Cyclic Boundary Condition Errors?	nyflyer	OpenFOAM Running, Solving & CFD	2	April 26, 2016 15:14
Force can not converge	colopolo	CFX	13	October 4, 2011 23:03
Problem with cyclic bc when run on parallel	Tek	OpenFOAM	2	January 20, 2011 15:40
Parallel convergence worse than serial!	Michael B	FLUENT	3	December 9, 2006 22:46

April 13, 2017, 21:42		#2
Collin New Member Collin Strassburger Join Date: Feb 2017 Location: Oak Ridge, TN, USA Posts: 10 Rep Power: 9	As it turns out, when you use "patch-constrained" decomposition, not all point which contact the constrained patch move uniformly. Periodic points would move at half of the intended rate (those points happened to be those which corresponded to a joining of 3 cells at the surface rather than 2 cells (for an extruded semi-2D case)). Thus, if anyone comes across this issue in the future, the solution is to not use patch-constrained decomposition in conjunction with mesh motion.