I vote NO.
Implementations of interval arithmetic which conform to IEEE 754 already
exist and are well-studied. This is true even for Kaucher intervals. See,
e.g.,
E. Popova, "Extended Interval Arithmetic in IEEE Floating-Point
Environment," Interval Comp., 4, 1994, pp. 100-129.
E. Popova, "Interval Operations Involving NaNs," Reliable Computing 2.2,
1996, pp. 161-165.
E. Popova & C. Ullrich, "Generalizing BIAS Specification," ACM Press,
1998, pp. 207-214.
The goal of these models is to be 754-compliant. So they bring the
exceptional conditions of IEEE 754 into the interval arithmetic. I believe
classical and modal interval arithmetic can be implemented in floating-point
as an exception-free system (allowing for division by interval containing
zero, which is still undefined). However, this will likely require
deviations from IEEE 754.
If mid-rad intervals are to be considered, it will be significant advantage
to define new bit-patterns for them. This is because radius requires much
fewer bits of precision than midpoint, and also there are no unbounded
endpoints.
I would vote "YES" for a motion to adopt the electrical engineering terms
and concepts of IEEE 754 in our discussions. This includes denormalized
numbers, NaNs, binary fraction, exponent, rounding modes, etc. However, any
motion to encourage "compliance" or "independence" from IEEE 754 is, in my
view, putting the cart before the horse. So I don't see the current motion
(or Arnold's alternative) is necessary or useful.
Svetoslav provides a more accurate summary when he says:
"The idea of interval computations is revolutionary in scientific computing,
it already had an impact on the FP-standard and it should be expected to
have further impact."
Sincerely,
Nate Hayes
Sunfish Studio, LLC