Discrepancy between RZ/3D (FBPIC/HiPACE++) for beam-driven wakefield about fbpic HOT 7 CLOSED

@AlexanderSinn thanks! Such a simple fix, sorry! I also re-ran with the wider box and got much better agreement between the codes. Thanks all for your help! @RemiLehe @AngelFP

from fbpic.

I used HiPACE++ to run the provided input script unchanged (blue) and with

amr.n_cell = 2047 2047 1024
geometry.prob_lo     = -100.e-6 -100.e-6 -155.e-6
geometry.prob_hi     =  100.e-6  100.e-6  0.e-6

to extend the domain transversely by 2x (orange). This now looks closer to the RZ codes.

from fbpic.

Thanks a lot for reporting this!
This discrepancy is indeed not expected. The fact that Wake-T and HiPACE++ do not have agreement is maybe the most surprising, since they both use the quasistatic approximation (unlike FBPIC).
@MaxThevenet @AngelFP @AlexanderSinn Have you seen a similar effect in the past? Are you able to reproduce the issue? And should we post the same issue on the Wake-T/HiPACE++ repository?

from fbpic.

Thanks for bringing up this issue. It looks very interesting. One suspicion I have is that the x,y boundary in HiPACE++ is too close to the wake, causing it to be shorter. HiPACE++ uses Dirichlet boundary conditions, while the RZ codes allow a 1/r field to go to infinity (is this correct, Remi?). Thanks to the inclusion of the input scripts in the issue, I will be able to do further HiPACE++ simulations to verify/falsify my suspicion about the boundary shortly. 

from fbpic.

Hi all,
It's good to see that FBPIC and Wake-T agree so well, but in principle there is no reason why Hipace should be giving different results here.

I've done similar convergence tests with Wake-T and hipace (beam driven case) and they both agree extremely well.

How many particles do you have in the driver? That's something where hipace might need more due to the 3d grid.

from fbpic.

Thanks all for the suggestions!

How many particles do you have in the driver? That's something where hipace might need more due to the 3d grid.

@AngelFP I only used 1e5 particles in the driver for these tests, so can try increasing this.

from fbpic.

Looks great!

from fbpic.