I found the following in https://github.com/poliastro/cephes/blob/master/src/pdtr.c *  *  * SYNOPSIS:  *  * int k;  * double m, y, pdtr();  *  * y = pdtr( k, m );  *  *  *  * DESCRIPTION:  *  * Returns the sum of the first k terms of the Poisson  * distribution:  *  *   k         j  *   --   -m  m  *   >   e    --  *   --       j!  *  j=0  *  * The terms are not summed directly; instead the incomplete  * Gamma integral is employed, according to the relation  *  * y = pdtr( k, m ) = igamc( k+1, m ).  *  * The arguments must both be positive. Alecos Papadopoulos PhD Athens University of Economics and Business web: alecospapadopoulos.wordpress.com/ On 28/9/2019 01:00, gretl-users-request(a)gretlml.univpm.it wrote: > Hmm, it seems there's something odd about the cephes function we're=20 > using to give the answer here, namely pdtri(). For the most part the=20 > cephes suite, written by Stephen Moshier, is fast and accurate, but=20 > I agree that something looks to be amiss here. I'll try to find out=20 > what. > > (In case anyone is able to help: cephes' pdtri() takes two=20 > arguments, an integer k and a double 0 < y < 1 (probability). And=20 > the doc says "Finds the Poisson variable x such that the integral=20 > from 0 to x of the Poisson density is equal to the given probability=20 > y." So what's "k"? If it's the mean (=3D variance) it shouldn't have=20 > to be an integer, should it?)