Re: Increasing precision in octave

[Rob Mahurin]

On Jun 20, 2008, at 10:09 AM, DimitryASuplatov wrote:
> Hello,
> I want to calculate the p-value with high precision.
> I mean that when I use normal_cdf function I get
>
> normal_cdf (real(24),m,s)
> ans =  1
>
> and "1" is not enough precision for me. I want to see something like
> 0.999999999999999999999888

normal_cdf is symmetric about the mean, so if you expect

        p = normal_cdf(x,m,s) = 1-ep

for some small ep, you could try

        ep = normal_cdf(m-x,m,s)

Mine underflows for m-x = 8s, at ep = 1e-16 (or so).  If you care  
about smaller differences than this you will need a different algorithm.

--
Rob Mahurin
Dept. of Physics & Astronomy
University of Tennessee         phone: 865 207 2594
Knoxville, TN 37996             email: address@hidden