Stavros Macrakis wrote: > On Wed, Feb 11, 2009 at 6:20 AM, Robin Hankin <rk...@cam.ac.uk> wrote: > >>> library(Brobdingnag) >>> exp(1000)/(exp(1007)+5) >>> >> [1] NaN >> >> >>> as.numeric(exp(as.brob(1000))/(exp(as.brob(1007))+5)) >>> >> [1] 0.000911882 >> > > Though brob is certainly useful in many cases, it can't substitute for > thinking about numeric issues (roundoff, cancellation, overflow, > underflow) in general. > > For example: > > >> x<-40; log(exp(x)+1)-x >> > [1] 0 > >> x<-as.brob(40); log(exp(x)+1)-x >> > [1] -exp(-Inf) > > The correct answer is about 4e-18. Perhaps Ryacas or some other tool > gets this right,
bc gets it arbitrarily right: bc -l <<END scale=1000 l(e(40)+1)-40 END as does, e.g., mathematica (which appears to be using the *foss* gmp library): N[Log[Exp[40]+1]-40, 1000] > but in general I don't think it's wise to abdicate > responsibility to one's tools. interestingly, perl's bignum gets it slowly and actually wrong (or do i miss something?): perl -Mbignum=p,-50 -le 'print (log(exp(40)+1)-40' vQ ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
