Periodic Boundary Conditions and parallelization

In some LBM-tutorials periodic boundary conditions are used. For example,
f’(1,1,j) =f (1,501,j) for a grid 501x501. But it is “a bad thing” for parallel calculations since the cells i=1 and i=501 are on different CPUs. Whether I can use instead of ‘periodic conditions’ usual ‘bounce-back’ or something similar?

if you need periodic boundary conditions you’ll have to do the communication. There is no way to avoid it. But it should not cost too much since you anyway have to do some communication between CPUs for normal bulk cells no?


  1. I have met periodic BCs for the 2-D lid-driven cavity benchmark. Whether they are obligatory for this test? Why I cannot replace them by no-slip conditions?
  2. Communications between CPUs for normal bulk cells (inside grid) are certainly done. But transfer between i=1 and i=501 is not normal and is too cost. So it is better to remove it.

you are right that periodic BC is more expensive than normal streaming, but if your physics tells you to have periodicity of your domain then you have to pay the price for it.

For the lid-driven cavity i don’t understand why one should want periodic BC. There is certainly no periodicity in it. If the right and left walls are periodic then it is not a cavity anymore but rather a couette flow.

Thanks, I agree with you.