Errata/note added Oct. 22, 2005, regarding slides from my Sherwood 2005 talk, http://w3.pppl.gov/~hammett/talks/2005/Sherwood05-noise.pdf
The following shows the sensitivity of some of the predictions of chi_noise to some assumptions. There are no significant changes in the conclusions.
On my slide 37, the prediction for chi_noise(t) assumed that 20 particles/cell were used and assumed that the effects of filtering and interpolation could be estimated from previous simulations in the way described on my slide 38, and assumed that the filtering and interpolation methods were equivalent in all three directions. With these assumptions chi_noise at late times was ~2.1 chi_GB, while I think the PIC simulation gave chi ~ 3 chi_GB.
The figure at http://w3.pppl.gov/~hammett/talks/2005/etg20-obs-var.pdf investigates the sensitivity of chi_noise to these assumptions. For 20 particles per cell, changing from the original filter assumption (on slide 37) to assuming linear interpolation with a "binomial compensated filter" causes chi_noise to drop from about 2.1 GB to 1.6 chi_GB. Lowering the assumed particle number from 20 to 5 particles/cell causes chi_noise to rise from 1.6 to 4.2 chi_GB. Changing the assumed filter from a "binomial compensated" filter to a binomial filter cause chi_noise to drop from 4.2 to 3.1 chi_GB. [This is following the terminology in Appendix C of Birdsall and Langdon, "Plasma Physics Via Computer Simulation", 1991, where they describe two kinds of 5 point filters. Their second kind of 5-point "binomial compensated" filter is broader (does less filtering) than the 3-point binomial filter, which in turn does less filtering than the first kind of 5-point filter they describe, which is equivalent to a binomial squared filter.]
Here are links to the official proceedings of the 20th IAEA Fusion Energy Conference 2004, and the paper and presentation by Z. Lin et al., "Electron Thermal Transport in Tokamak: ETG or TEM Turbulences?".