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

Re: [Fwd: Re: Neumaier-Pryce proposed decoration system (v03.2)]

Vincence Lefevre wrote:

On 2011-06-16 09:13:00 -0500, Nate Hayes wrote:

Personally, I wouldn't be sad to see P1788 amend its position on
Motion 3 such that IEEE 1788 is restricted to compact (closed and
bounded) intervals.

Then we only have to deal with issues of overflow.


Not just overflows (if you mean floating-point overflow). How would
you regard 1/[0,1], which is [1,+inf] at Level 1, for instance?


Sorry.  I meant at Level 2, only.

I think its fine to give the decorated result ([1,+Inf],D1) at Level 1. I
don't have any problem giving this result at Level 2, either, BTW.

My only thought was: I'm not really aware of a practical interval algorithm
that requires making the distinction between infinity and overflow at Level
2. So the result could just as well be ([1,+Overflow],D1) at Level 2, then;
and perhaps this could lead to a simpler standard (avoiding the complexities
mentioned by Dan Zuras). That's all.

Either way, I really like the method Ian McIntosh has presented to deal with
the distinction between overflow and infinity at Level 2.

Nate