Re: Unbounded intervals
On 2012-04-25 11:00:26 -0400, Michel Hack wrote:
> There also is no largest representable number for a *program*, because
> even with a range-limited primitive type programs can use scaling to
> reach waaay beyond the range of the primitive type. And as I mentioned
> before, some number representations are essentially overflow-proof.
> The fact that no physical computer implements such a type as a machine
> primitive does not restrict what programs (or virtual machines) can do.
BTW, with MPFR, there was a suggestion to make the exponent field
an mpz_t (as an optional feature), i.e. an arbitrary precision
integer, so that overflow (and underflow) would no longer exist.
And this would still be compatible with the conventional Level 1
(unbounded intervals are still necessary for operations like
1 / [0,1] -- note that one cannot necessarily differentiate the
cases [0,1] as an exact interval and [0,1] as a superset of some
other closed interval that doesn't contain 0).
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)