I’ve been playing around with the SpatialLegendreFilter for computing Legendre moments of the angular flux, and I’ve found that, for the same number of histories, I can often get much better statistics from a mesh tally of the flux, which I then use to compute Legendre moments by a quadrature approximation like the Trapezoidal rule than I can get directly from the Monte Carlo computed Legendre moments themselves. Perhaps there’s a more stable (i.e. robust to MC uncertainty) way to reconstruct the flux profile from the Legendre moments that I haven’t heard of.
Intuitively, I would expect throwing out moments that are smaller than MC noise would be a good idea, but I’m still unable to reconstruct the spatial flux distribution better than with a plain old mesh tally. If mesh tallies are just more robust in general, is there any benefit to using the SpatialLegendreMoment filter? Maybe it’s more memory efficient than a mesh tally?