[R] What precision is R

Prof Brian Ripley ripley at stats.ox.ac.uk
Tue Jul 6 07:50:48 CEST 2004

On Mon, 5 Jul 2004, Laura Holt wrote:

> What precision is R, please?  64 bit?

Your question lacks precision :)

R uses the underlying double precision floating-point arithmetic of the OS
it is running on.  On all currrent platforms that we know of, that is
IEC60566 arithmetic, with an (implicit) 53-bit mantissa.

So floating-point operations have double precision, usually with around 53
bits of accuracy.

> Where would I find that out typically, please?

Type ?.Machine, then print(.Machine).

One complication on machines with extended-precision registers (and since 
that includes the ix86 family, that is most of them) is that some 
calculations are done with a 64-bit mantissa and finally stored with a 
53-bit mantissa.  Those may turn out to be more accurate than your expect 
(which makes measuring accuracy difficult).

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

More information about the R-help mailing list