[Bug-gsl] bug in gsl_ran_multinomial_lnpdf function

[mingxi]

To whom it may concern,

I encountered a bug in gsl_ran_multinomial_lnpdf function.

The case this function does not consider is the degenerated cases.

Suppose I have input size_t K=3, p[]={0.5,0.5,0}; n[]={2,3,0}. This 
function will not return correct results.

In this case, we can just ignore the last class probability, which is 
p[]={0.5,0.5}, n[]={2,3}.

BTW: I use this library for 2 years. It is good software!

Sincerely,

-Mingxi Wu
University of Florida