Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Definition of intervals as subsets of R - the bad news



Ian McIntosh suggested:
>        Allow +Infinity and -Infinity as fully accepted values
>        in intervals.

This has been discussed at length already.  While it may solve some
problems, it raises others, most notably the need to accept NaN as
well (or NaI), because of Inf-Inf, 0*Inf, etc.   These exceptions
don't arise when Inf is only used as an (unreachable) bound.

I'm also a little annoyed that the red herring of 1e400 is brought
back into play, or unqualified constructs such as [1e300,1e300]
(which makes sense in DFP but not in the usual BFP formats).  If
literals like 1e400 or 1e300 are converted to Binary64-based intervals,
they become [Maxreal,Inf] and [rdown(1e300), rup(1e300)].  Treating them
as if converted first to Binary64 is (a) most likely an error, and (b)
dependent on the "current" rounding mode, which is not an IA concept.

Michel.
Sent: 2009-02-20 16:44:37 UTC