Re: incorrect test on gamma function ?

[Michael D Godfrey]

 On 10/27/2010 10:57 AM, Marco Atzeri wrote:
> gamma(0) = Inf
> gamma(-1) = NaN
>
> > > > > > >>>>>  processing /pub/hg/octave/src/mappers.cc
>    ***** test
>   x = [-1, 0, 1, Inf];
>   v = [NaN, Inf, 1, Inf];
>   assert (gamma(x), v);
>   assert (gamma(single (x)), single (v));
> !!!!! test failed
> assert (gamma (x),v) expected
>     NaN   Inf     1   Inf
> but got
>     Inf   Inf     1   Inf
> NaNs don't match
>
> considering the graphics
> http://en.wikipedia.org/wiki/Gamma_function
>
> I presume the two values should be both Inf (or both NaN)
>
> Marco
The 2 values should both be Inf  (also, that is the Matlab result).
In looking at this I noticed that both matlab and Octave restrict
the argument to reals.  It would be useful to provide gamma
for complex argument.  Does an m file for this exist somewhere?
Numerical method for this is in Abramowitz and Stegun, but newer
methods may be around. Gamma is defined for complex argument,
so just modifying the current gamma to accept complex would be
appropriate.

Michael