Sept 11, 1997
I ran the Chi vs. epsilon (r/R) scaling with higher resolution, larger perpendicular boxes, and for longer times. Although these new runs are slightly different in detail, the conclusions based on the preliminary results I put up last week hold.
The higher resolution run time histories to compare with Andris's '94 IAEA Fig. 2 are shown below:
These runs used our "4+2" equations and a parallel box length of -pi to pi, the same as in the GK particle runs.
These three runs seem reasonably converged and saturated. They represent 32 hours of C90 CPU time and 95 hours of C90 wall clock time.
The trend is still pretty clear: the difference between GF and GK does not scale strongly with epsilon. If the Hinton-Rosenbluth effect were the dominant cause of the difference between the GF and GK results, chi_GF/chi_GK would scale something like 1+const*sqrt(eps), as shown below in red.
Back to index of cyclone pages
Back to Mike Beer's home page