Re: [Bug-gsl] gsl_sf_coupling_3j bug report

[Alexey A. Illarionov]

Hi Brian,

I just want to emphasize that the Grigory's patch does not solve the
problem it only extends a little bit range of possible j and m where
algorithm gives reasonable answer. For example,
    (200 200 200)
    (-10 60 -50) = 2.55804e+18 +- 1.35857e+17
while the correct answer would be approximately
    7.4939e-4



On 10/10/11 09:17 AM, Brian Gough wrote:
> At Sun, 2 Oct 2011 13:25:50 +0400,
> Grigory I. Rubtsov wrote:
>>
>> Hi Alexey,
>>
>> You gave me an idea of further improvement. If one mupltiply the norm
>> with every term, there will be no large numbers at all. Please
>> consider the following patch for inclusion into GSL. It works with
>> practically arbitrary large l.
>>
> 
> Thanks for the patch.  I tried it and it solves the problem you found.
> I've committed the patch and extended it to include the error term, to
> avoid the error being underestimated.
>