[Developers] incomplete beta function is broken

[dave fournier]

On 12-06-27 10:13 AM, John Sibert wrote:


I should have mentioned that I attached the modified code for your perusal.

> I share Dave's concern about simply changing functions. Using capitals 
> to distinguish versions is one solution. Using a namespace is an 
> alternative, if vaguely annoying, solution. (I'm annoyed every time in 
> need to specify a namespace to use cout,) The namepasce Cephes is 
> already in use. It would make it easy to compare results with the 
> appropriate using namespace declaration.
>
> More generally, the lack of smoothness is probably a result of the 
> polynomial approximation. May functions use such approximations. Is 
> this a commonly encountered problem? Or does it mainly occur at the 
> extremes of the function's domain?
>
> John
>
> John Sibert
> Emeritus Researcher, SOEST
> University of Hawaii at Manoa
>
> Visit the ADMB project http://admb-project.org/
>
>
> On 06/27/2012 05:03 AM, dave fournier wrote:
>> As I noted earlier the cephes version of the beta (and probably  log 
>> gamma) functions
>> appear to lack sufficient smoothess for the random effects stuff.    
>> The solution is to go
>> back to the versions based on NR.  However rather than replacing them 
>> I think it would be
>> easier to simply add the old functions to the package with a slight 
>> name change.
>> Could use capitals as Betai, Gammln. Then both versions would be 
>> available and
>> experience would show which are best for various applications.  Any 
>> thoughts?
>> _______________________________________________
>> Developers mailing list
>> Developers at admb-project.org
>> http://lists.admb-project.org/mailman/listinfo/developers
>>
>