I am studying an LWR core (with symmetry present of course) but the radial power profile is severely asymmetrical. I know this is not due to histories (as I have run 2,500 batches with 100,000 particles per batch). the profile I am getting is shown here. the boundary conditions are vacuum on the outermost boundaries. I have checked to see if the core universe is correctly input as symmetric.
It can take a very large number of cycles to average out the cycle-to-cycle correlations you encounter in a very optically thick problem like this one. The convergence rate of the tallies will not actually be what you’d expect from the standard central limit theorem. You can increase the number of generations per batch to get a more realistic estimate of the tally variance. You should then see that your asymmetry levels fall within the noise level of the reported uncertainties.
the results I posted are with 50 inactive cycles, but I have done the same calculation with 200 inactive cycles, too, with no significant difference. the other thing is if you look at the bottom and top rows you can see huge differences (greater than what I would like to associate with the statistics). Also, the value at the centre is supposed to be ~0.7 and I am getting ~0.5.
On top of what has already been said, I would consider cutting your problem in quarters or half during this testing phase. (Maybe half if you are comparing bottom vs top values?)
As was mentioned this is a very tough to converge case as the core is very large and decoupled. Cutting it in half might help to eliminate oscillations as the fission source converges and quicken your convergence, just for testing purposes.
Also 200 iterations might not be enough. Have you plotted basic Shannon entropy values and checked for convergence there? Sometimes for these large decoupled problems, we need to spend many more iterations than we’d like to get the true fission source. I would keep the number of particles the same, but just skip possibly many many more generations.
Finally, sometimes the best best practices recommend running many calculations and taking the average among the results as the true answer (so long as the answers from each individual calc are also converged properly). I am not sure how often this is done in practice however. Have you tried running other calculations with just a different random number? I imagine that if the fission source is truly uncoverged, the discrepancy between tallies in symmetric regions wouldn’t be constant. Right now I see 1.6 at the top and 1.4 at the bottom. If you rerun a couple times though and the fission source is not converged, I’d expect the bottom to sometimes be 1.6 or so and the top to be 1.4 or so. But if the top or bottom remains consistent after many runs, then it’s likely an actual input error somewhere.
I have plotted the Shannon entropy for the quarter core and indeed by increasing the particles I can get a symmetric power profile. now I am using 1E6 particles instead of 1E5 particles for 1000 total batches and 200 inactive batches.
