George Wo;tman wrote: > At 03:22 PM 10/31/2006, you wrote: > > >On the other hand, the excerpts from you and George which Steinar dug up > >suggest that you were indeed trying to fix the boundaries at the point where > >probability of roundoff error equalled the percentage time increase incurred > >by using the next size of FFT. > > No, Brian and I choose the boundary points such that a roundoff error would be > rare but not impossible. The definition of rare was rather arbitrary based on > extrapolation from experimental data.
I'm not convinced the risk of rounoff error fluctuates with exponent size in the way the average maximum error unsurprisingly does. > > Why "rare" instead of "optimal"? Well, I don't want to be the one to > tell someone > he missed a prime because his run had a not-unexpected round off error. Understandable, but it leaves little hope fot us humble double-checkers! Does my picture of the risk of roundoff error make sense? David Eddy _________________________________________________________________ Be one of the first to try Windows Live Mail. http://ideas.live.com/programpage.aspx?versionId=5d21c51a-b161-4314-9b0e-4911fb2b2e6d _______________________________________________ Prime mailing list [email protected] http://hogranch.com/mailman/listinfo/prime
