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

Re: P1788: Our first formal motion: IR versus *IR



John:

I am very happy about your mail and I basically agree with it. But let me make the following remark:

IR certainly is the most natural notation for a set of real intervals somebody is using. I feel a little sad if we (by voting) restrict this simple notation to a set which is mathematically incomplete. With set inclusion as an order relation IR is an ordered set and it is most natural to take it as a complete lattice. The least element is the empty set and the greatest element is the set (-oo, +oo). The infimum is the intersection (which can be the empty set) and the supremum is the interval (or convex) hull. (It is an inf-sublattice of the power set PR.)

The past was a learning process for all of us. I am convinced that IR as a complete lattice is the view intervals are considered by R. E. Moore.

So whatever the StandardNotation for intervals will be I shall take the freedom to denote the set of closed (bounded and unbounded) real intervals by IR and I shall use an index notation for the bounds of intervals whenever it is convenient.

Best whishes
Ulrich Kulisch


John Pryce schrieb:
Arnold and all

On 16 Jan Arnold Neumaier wrote (response to Prof Kulisch, and circulated)
-----
More precisely, IR is the set of nonenpty, closed and bounded intervals
over R.

*IR is the set of bounded and unbounded intervals over R. So the empty set and the set R = (-oo, +oo) are elements of *IR.
Yes.

In my paper for the proceedings of the Dagstuhl Seminar (January 2008) I used the notation (IR) for this set avoiding the *. (This notoation could also be extended to (IR)^n.).
Yes. But it is unwise to change a long established notation; it will
not stick and only add confusion. Moore's intervals were bounded and
nonempty.
-----

But on 27 Oct last Arnold wrote to JDP:
-----
JDP: I may be wrong, but to me "Moore interval arithmetic" means one has only bounded intervals, and no empty set.

AN: I think his interval arithmetic was closed, and included things like
     4/[0,4]=[1,inf],
     1 - 8/[0,4] intersect [0,2] = empty.
At least, this is needed (for x^2+7 = 0, x in [0,2]) to make the
1D interval Newton method work, which was his invention.
Of course, his implementation could have no number inf; but the
arithmetic in his theory probably had.
-----

It seems to me Arnold came to two different conclusions at different times about what "Moore's model" was. Fair enough. I am probably at least as inconsistent.
My point is that we should use common sense in what "following" the standard notation paper means.

- Where it says "an interval, whether scalar or vector, is typeset in bold italic" - that's pretty basic typography, and if we all follow it, our papers will be more immediately readable both by established intervallers and by newcomers.

- Where it says IR denotes a certain set of intervals, and *IR, a certain larger set, let's be flexible. What one commonly needs is a notation for "the set of all allowed intervals in whatever model I am using". It is natural for this to be "the symbol for the underlying number system, with an I in front", so
   IR for an R-based system
and IR*, IC, etc. for other number systems.

So if some paper uses IR, while the notation paper says it should be *IR, that's fine by me as a reader, provided each paper clearly states what notation is being used.

More generally, I feel (as was the original intent of my motion) that we shouldn't make ANY amendments to the notation paper, for now. My motion asks you to take it for what it is, and to "follow" it flexibly.

Maybe I'll add to the Rationale a sentence on these lines. Unless two people feel strongly enough to propose, and second, a formal amendment in the next day or two, I ask that we go on to voting on the unchanged motion.

Best wishes

John P