Re: How to merge required operations under a common name
On Wed, 27 Jan 2016 15:22:59 +0100, Paul Zimmermann wrote:
> ok, then integers (p/q with maximal denominator q=1) would be fine?
Almost. Technically you also need a bound on the numerator p, but that
can be enormous (e.g. 2**(2**64)) because it does not affect tightness.
You also need representations for Infinity, in order to support the
numeric functions of intervals such as width(), inf(), sup() -- since
unbounded intervals must be supported. So the type can be based on
integers, but must include flags or other means to denote +-oo.
I think Vincent Lefèvre worked out the minimum requirements for a
conforming number format: it must be finite, include +-oo, +-1, 0,
and be symmetric (for any x, -x must be included).
Michel.
---Sent: 2016-01-27 14:48:16 UTC