Hello,
The current implementation of the chi-squared distribution pdf in GSL
returns 0.0 for inputs with x<=0.0.
The domain of this function is [0.0, inf), (see e.g. Abramovitz-Stegun
here: http://people.math.sfu.ca/~cbm/aands/page_940.htm ),
so wrong values at x=0.0 are returned for nu=1.0
(gsl_ran_chisq_pdf(0.0, 1.0) should be inf) and nu=2.0 (should be
0.5).
Attached is a patch which fixes this.
Best,
Teemu
=== modified file 'randist/chisq.c'
--- randist/chisq.c 2007-09-10 18:09:35 +0000
+++ randist/chisq.c 2010-11-10 13:19:34 +0000
@@ -39,16 +39,23 @@
double
gsl_ran_chisq_pdf (const double x, const double nu)
{
- if (x <= 0)
+ if (x < 0)
{
return 0 ;
}
else
{
- double p;
- double lngamma = gsl_sf_lngamma (nu / 2);
-
- p = exp ((nu / 2 - 1) * log (x/2) - x/2 - lngamma) / 2;
- return p;
+ if(nu == 2.0)
+ {
+ return exp(-x/2.0 - log(2.0));
+ }
+ else
+ {
+ double p;
+ double lngamma = gsl_sf_lngamma (nu / 2);
+
+ p = exp ((nu / 2 - 1) * log (x/2) - x/2 - lngamma) / 2;
+ return p;
+ }
}
}
_______________________________________________
Bug-gsl mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/bug-gsl
