[makaveli_lcf]

Hi all!
As I already wrote at thread http://www.cfd-online.com/Forums/ope...tml#post227242, I noticed a difference between serial and parallel solutions in OpenFOAM. Following pictures illustrate it for 1 CPU and 2 CPUs being used:

1cpu.jpg2cpus.jpg

Velocity magnitude field is shown on both pictures at the end of simulation when we get steady state. Top of the domain is velocity inlet, bottom is pressure outlet, left wall is nonslip with cooled zone where porous media grows due to cooling processes, right is symmetry plain.
For the numerical solution two systems (mass and heat transfer) are solved together until they reach given convergence criteria for each time-step, thus solution is transient.
We can see that for single processor calculations global convergence criteria is fulfilled. When multiprocessor calculations are involved, parallel libraries do not care about global convergence, so from my point of view data is transmitted between subdomains only after each time-step finalization, rather than after each iteration.
So I'm looking for your advises, how to control global convergence of numerical solution for parallel run?