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

Re: Motion P1788/M0013.04 - Comparisons - Overflow / Infinity



Dan Zuras wrote:
From: "Nate Hayes" <nh@xxxxxxxxxxxxxxxxx>
To: "John Pryce" <j.d.pryce@xxxxxxxxxxxx>,
 "stds-1788" <stds-1788@xxxxxxxxxxxxxxxxx>
Subject: Re: Motion P1788/M0013.04 - Comparisons - Overflow / Infinity
Date: Tue, 21 Sep 2010 08:19:45 -0500

John Pryce wrote:
>> . . .
>>
>> NO! You are falling into two traps at once.
>>
>> First, our interval model says intervals are subsets of the reals R; >> our >> notation says [1, Infinity] is just shorthand for {x in R | 1 <= x < >> oo}. >> Hence it is ONE set, not an infinite family of sets like [1, >> Overflow].
>> And standard set theory unambiguously says [1, Infinity] \subseteq [1,
>> Infinity] is true.

John is right.  There ARE conceptual traps here.


Here is where I stick my neck out a little and, I think, differ with Ian's
interpretation, too.

As John mentions, I also see [1,Infinity] as shorthand for
    { x in R | 1 <= x < oo }.
However, unlike Ian, I see [1,Overflow] as shorthand for
    { x in R | 1 <= x <= u },
where MAXREAL < u is some unknown real number.

Hmm.  I think John's approach is better here in that
there is only one unbounded or infinite endpoint &
that intervals are open at that endpoint.

So I suspect you cannot make that distinction
consistent mathematically.

But even supposing you could, we cannot make that
distinction computationally.

Ian has already given examples where we can.

So have you!  :-)




We have (& should have) but one infinite endpoint
to represent intervals whose bound cannot be
computed.  It make no difference whether we arrive
at that endpoint through overflow or some other
means.

And since we have only one such endpoint we can
have only one interpretation for an interval
possessing that endpoint.  That is that we know
no bound for that interval.


Hence, [1,Infinity] is a closed, unbounded interval, but [1,Overflow] is a
compact interval with a "really big" (but finite) upper bound.

Well, first, I think of [1,infinity] as a semi-open
interval.

Technically, [1,Infinity] is a closed, unbounded interval:

http://mathworld.wolfram.com/ClosedInterval.html





And, second, whether we know no bound because there
is no bound or we know no bound because we were unable
to compute & represent that bound amounts to the
same thing: we know no bound.

For us these are the same.


Then you just made an argument there should be no IsBounded decoration.






I don't claim to know (at least at this point) what [1,Overflow] \subseteq
[1,Overflow] means in this case. A bool_set result might be one possible
answer.

Nate

I have no comment on this last as it is outside
what I believe we must consider as our conceptual
framework.

If P1788 includes an IsBounded decoration, this question must be solved anyways.




John is also right historically: it has always
been hard to deal with the infinite.

I fully agree! :-)

THis is why I'm glad we have people like John and Arnold to make sure we stay within proper mathematical boundaries, however controversial some of the proposed suggestions or ideas might otherwise be.

Sincerely,

Nate