
Re: 64bit KISS RNGs
Posted:
Mar 1, 2009 4:49 AM


In article <godlg7$a4j$1@soup.linux.pwf.cam.ac.uk>, nmm1@cam.ac.uk writes:
> Galathaea's concern was about the period, and she is very right to > be so concerned. While a long period does not guarantee pseudo > randomness, it is a prerequisite for it  in particular, the pseudo > random properties in N dimensions are often limited by the Nth root > of the period. And, despite common belief, that is NOT solely true > for multiplicative congruential generators.
The period of RANLUX is huge10**164 or something. (In other words, many orders of magnitude larger than the number of distinct bit combinations. Many lesser generators have a period much LESS than the number of distinct bit combinations and some algorithms can have a period at most as long as the number of distinct bit combinations. In these cases, of course, a given number x is always followed by a given number y, which with RANLUX is not the case.)

