On Fri, Feb 15, 2002 at 07:09:42PM -0500,
Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>
> 3. The equation is numerically stable even when n is much smaller than
> N, because the only cancellation is in the term (n - f1) which we can
> compute exactly. A lot of the other equations I looked at depend on
> series like (1 - n/N)**i which are going to be really nasty when n/N
> is tiny.
You can work with the above for n << N by using power series. For n << N,
(1 - n/N)**i ~= 1 - in/N.