numerical evaluation of quotients of gamma functions



Yes that's also a good idea. But I think I'd need to write my own log-base-two-gamma-function? Or am I missing something?

--Barton

________________________________________
From: maxima-bounces at math.utexas.edu [maxima-bounces at math.utexas.edu] on behalf of Raymond Toy [toy.raymond at gmail.com]
Sent: Friday, November 01, 2013 17:04
To: maxima at math.utexas.edu
Subject: Re: [Maxima] numerical evaluation of quotients of gamma functions

>>>>> "Barton" == Barton Willis <willisb at unk.edu> writes:

    Barton> I need to numerically evaluate quotients of products of  gamma functions.  In order of preference,
    Barton> I would like to

    Barton> (1) avoid overflows, (2) have good accuracy,  and (3) be fast (much less important). Any
    Barton> suggestions?

Use base 2 logs?  This might be a little more accurate then base e
logs, but not any slower.  Base 2 since floats are base 2, so log(f*2^n)
can be represented in 2 parts very accurately: n + log(f).  Base e
logs don't have this simple separation, and for large values you lose
many bits to hold the integer part.

Ray

_______________________________________________
Maxima mailing list
Maxima at math.utexas.edu
http://www.math.utexas.edu/mailman/listinfo/maxima