[Bug-gsl] Bug in normalized incomplete Gamma function/upper tail Gamma dist'n probs.

[Brian Williams]

Version number of GSL:  1.6
Hardware & OS:  Dell PowerEdge 6850, Intel Xeon processor at 3.0GHz/8MB
Cache, Linux 2.4.21-27.EL
Compiler:  gcc 3.2.3 20030502 (Red Hat Linux 3.2.3-52)
   gcc -o test.out -Wall -lgsl -lgslcblas -lm test.c
Description:  For some inputs, answers are returned that are incorrect. 
Here is a test program (test.c):

#include <stdio.h>
#include <gsl/gsl_sf_gamma.h>
#include <gsl/gsl_cdf.h>
 
int main (void)
{
   gsl_sf_result r;
 
   gsl_sf_gamma_inc_Q_e (1008.02,5.0039*168.156,&r);
   printf("r = %g %g\n", r.val, r.err);
 
   printf("sf = %g\n", gsl_cdf_gamma_Q(168.156,1008.02,1.0/5.0039));
 
   return(1);
}

Here is the output:

r = -1.37412e-08 2.0289e-22
sf = -1.37412e-08

The output calculated from R version 2.1.1:

sf =  1

Brian