Dear respected professors and developers,
I have encountered an issue during my OpenMC‑based simulations. My calculation is performed on an OECD‑proposed ADS core model, as shown in Figure 1. The original vacuum boundary region was replaced with a reflector layer, and the thickness of model was extended to 172 cm. I adopted the MAGIC‑generated global weight‑window scheme to generate a weight_windows.h5 file, which was subsequently applied to guide the particle‑transport calculation. A benchmark comparison was conducted against the analog‑mode simulation.
It turns out that the analog simulation yields far‑superior results compared to the weight‑window‑assisted calculation. The error distribution from the analog simulation is plotted in Figure 2, while that of the MAGIC‑weight‑window case is presented in Figure 3. The runtime of the weight‑window transport simulation is 1534 seconds and the analog‑only simulation is 1253 seconds.
The simulation was executed in eigenvalue mode with 100 total batches, 60 inactive batches and 300 000 particles per batch. The weight‑window file was updated every 10 batches. My weight‑window generator settings are as follows，mesh_zcy_real is the global mesh:
weight_window_gen2 = openmc.WeightWindowGenerator(
mesh=mesh_zcy_real,
#energy_bounds***=energy_bounds,***
particle_type=‘neutron’,
method=‘magic’,
max_realizations=settings.batches,
update_interval=1,
on_the_fly=True
)
I am confused by this phenomenon. Although the weight‑window‑aided simulation consumes more computation time, its outcomes are inferior. The reflector region in the outer domain is theoretically supposed to be well‑optimized by the weight‑window method. On the contrary, it presents larger relative errors for neutron‑flux results. I would appreciate any relevant assistance.
