Gaussian RNG?



>>>>> "Liam" == Liam M Healy <Liam.Healy@nrl.navy.mil> writes:

    Liam> Probably won't win on speed but I've always thought the Box-Muller
    Liam> algorithm was quite cool.

I've implemented 5 different Gaussian RNGs (you can find them as a
part of CLOCC).  Of these, the NA algorithm was the slowest, the
Ziggurat method was the fasted, and the rest were pretty close:

                                time in sec
1. Box-Muller                   0.45
2. Ahrens NA algorithm          0.88
3. Trigonometric                0.5
4. Ratio                        0.58
5. Ziggurat                     0.28

(Time for 1 million numbers on a 866 MHz Pentium II, with CMUCL.)

I don't expect maxima being used in such a way that number crunching
is the crucial element, but if we can easily make it fast, why not?

Ray