Robert> I only need it to work for real arguments. But if that were an issue I could stick with Lanczos. Any given set of Lanczos parameters work equally well on the entire half-plane Real[z]>0, and with the same guaranteed error bound. This article describes it really well: http://www.haoli.org/nr/bookcpdf/c6-1.pdf which is from: William H. Press et al., Numerical recipes in C: the art of scientific computing (2nd edition). Cambridge University Press, (1992) ISBN 0521431085. On Mon, Jan 9, 2012 at 12:06, Warren Smith <warren.wds@gmail.com <http://gosper.org/webmail/src/compose.php?send_to=warren.wds%40gmail.com>> wrote:
Do you want it to work for complex or only for real argument?>> > I've been looking for a numeric Gamma method that gives 34 digits of> > precision suitable for quad-precision floating-point [2] although I would> > settle for 31 digits suitable for "double-double".> -- Robert Munafo -- mrob.com Follow me at: gplus.to/mrob - fb.com/mrob27 - twitter.com/mrob_27 -mrob27.wordpress.com - youtube.com/user/mrob143 - rilybot.blogspot.com
Just two b*ggerfactors, da*n near double precision on z>9: http://gosper.org/munafacmu16.png Probably could be double prec for z>10. (Manually Remezed coefficients aren't finished, but mu won't change much.) I'm guessing quad will take eight b*ggerfactors. --rwg