Higher order cumulants of point processes, such as skew and kurtosis, require significant computational effort to calculate. The traditional counts-in-cells method implicitly requires a large amount of computation since, for each sampling sphere, a count of particles is necessary. Although alternative methods based on tree algorithms can reduce execution time considerably, such methods still suffer from shot noise when measuring moments on low amplitude signals. We present a novel method for calculating higher order moments that is based upon first top-hat filtering the point process data on to a grid. After correcting for the smoothing process, we are able to sample this grid using an interpolation technique to calculate the statistics of interest. The filtering technique also suppresses noise and allows us to calculate skew and kurtosis when the point process is highly homogeneous. The algorithm can be implemented efficiently in a shared memory parallel environment provided a data-local random sampling technique is used. The local sampling technique allows us to obtain close to optimal speed-up for the sampling process on the Alphaserver GS320 NUMA architecture.

Non-specialist paper, 9 pages, 5 figures, accepted for publication in Int. J. of High Perf. Comp. & Networking