If we are keeping score, R does too. :) I am thinking about how to get around it when I want to know whether a number is some prime to some power. Like in here like n=2^j. An obvious way for me in here is using modulo, but with high values of j, it won't work. Any suggestion?
On Sunday, 29 June 2014 21:42:01 UTC+2, Mauro wrote: > > > So a bug? > > Probably not a bug but an intrinsic limitation of floating point > numbers. > > Python does the same: > >>> from math import * > >>> log(32768)%log(2) > 0.6931471805599452 > > (Although Matlab gets it "right".) > > > On Saturday, 28 June 2014 13:59:22 UTC+2, Mauro wrote: > >> > >> I think that is a rounding issue. log(2)==0.693... Presumably it > >> can't quite fit log(2) 15x into log(32768), thus it returns a reminder > >> which is almost log(2): > >> > >> julia> rem(a,b)-log(2) > >> -1.1102230246251565e-16 > >> > >> which is around machine precision: > >> > >> julia> eps(1.0) > >> 2.220446049250313e-16 > >> > >> On Sat, 2014-06-28 at 10:40, [email protected] <javascript:> wrote: > >> > When using the rem() function, I found out a mysterious thing. (Maybe > >> only > >> > my ignorance, hard to tell.) The log(32768)/log(2) gives 15.0, but > >> > rem(log(32768),log(2)) gives 0.693... How so? > >> > >> > >
