Gaussian RNG?



 >Maxima has a Gaussian random number generator (look in bessel.lisp!),
 >but it's rather odd.  It sums up 12 numbers generated from a uniform
 >RNG on [0, 1), so the result can't be negative.

That's odd.

 >I'd like to replace this with a real generator, or at least one that
 >can produce negative numbers and one that's better than summing 12
 >uniform numbers.

Isn't the standard proceedure to generate the largest datatype random 
integer, divide by RAND_MAX and then approximate as floating point, 
assume uniform [0,1], then do a transformation to standard normal and 
then transform to the other normal.

I don't know much about LISP, I use C mostly, but the only reason I can 
think of to sum multiple random numbers would be that RAND_MAX is too 
small, so there is not enough continuity in the distribution

 >Anyone depending on that behavior?  Should I just give the new one a
 >different name?

No, if anything, give the old one a new name. If the numbers are only 
positive, then it is not a Gaussian distribution and is wrong.