Hi,

 I have written a piece of code to implement the multinomial CDF. GSL currently provides random variates from a multinomial and the multinomial PDF. The CDF is implemented as the approximation from,

"A Representation for Multinomial Cumulative Distribution Functions",  Bruce Levin, The Annals of Statistics, v.9, n.5, pp.1123-1126, 1981

The approximation is meant to be pretty good.

Because the multinomial is a vector random variate the "P" version of the CDF is implemented.

The P version corresponds to P(X1<=n1,X2<=n2,...,XK<=nK)

In keeping with the pdf function already in GSL no checking of the consistency/validity of the inputs is done.

Code attached along with a small test function.


Glenn Stone
CSIRO Mathematical and Information Sciences
North Ryde
+61 2 9325 3259