Normalizing

Posted by Beetle B. on Fri 11 January 2019

We would like a unique way to represent \(x\). One approach is to pick the one which gives the smallest exponent possible (while still not smaller than \(e_{\textit{min}}\)).

This is called the normalized representation.

A normal number is one where \(1\le|m|<\beta\), or \(\beta^{p-1}\le|M|<\beta^{p}\).

If \(e=e_{\textit{min}}\), then \(x\) is said to be subnormal, or denormal. Then \(|m|<1\) or \(|M|\le\beta^{p-1}-1\)

The smallest possible denormal number is \(\alpha=\beta^{e_{\textit{min}}-p+1}\).

The smallest possible normal number is \(\beta^{e_{\textit{min}}}\)

The largest finite floating point number is:

\begin{equation*} \Omega=(\beta-\beta^{1-p})\beta^{e_{\textit{max}}} \end{equation*}