Re: Internal rounding errors

[Thomas Lumley <tlumley@u.washington.edu>]

On Tue, 3 Mar 2009 courses@zanybooks.com wrote:

> Using R (not S-Plus) yields the following:
> > data3=1:25+rnorm(25)
> > temp=data3-mean(data3)
> > sum(temp)
> [1] 1.154632e-14
And so?

> sum(1:25)*.Machine$double.eps
[1] 7.21645e-14

so the rounding error is about six times smaller than a bound based on machine 
precision, and is very close to mean(1:25)*sqrt(25)*.Machine$double.eps, which 
seems a reasonable expected value for the internal rounding error.

You weren't expecting zero, were you?

    -thomas

Thomas Lumley                   Assoc. Professor, Biostatistics
tlumley@u.washington.edu        University of Washington, Seattle