Re: [Help-gsl] better generator for gaussian random numbers

[John D Lamb]

Brian Gough wrote:
> John D Lamb writes:
>  > Can you tell me whether anything's happened with the Gamma generator,
>  > also due to Marsaglia-Tsang? It would be good to implement that too.
>  > 
>  > I'm happy to test and or write any of this.
> 
> If you could put up a page like Jochen's with some analysis and
> testing that would be great.
> 
Here it is. Sorry the web page is a little hurriedly constructed.

http://www.uyea.btinternet.co.uk/gamma/gamma.html


The gamma generator of Marsaglia and Tsang also substantially
outperforms GSL and has the virtue of running about the same speed for
all a >= 1. (a < 1 is very unusual).

It might also be worthwhile to add one other distribution to GSL.

gsl_ran_erlang( int a )

This is just a gamma distribution with integer coefficients. I assume
that gsl_ran_chisq uses gsl_ran_gamma and so could also be made to run
much faster. In any case, I suspect that an erlang generator would be
used much more than a chi-squared generator because of its applications
to queueing models.


-- 
JDL