| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Hmmm ... I don't think [anything] / [-a,a] = NaN was what the proposers of that motion had in mind. This would seem to be connected to the interpretation of decorations, too. For instance, one interpretation would be 1 / [-1,1] = (-\infty,-1] \cup [1,\infty), where we only take those values in the denominator over which the floating point result is defined. How does that fit into the scheme of "intervals as subsets of real numbers? How does that fit into alternative decoration schemes? Baker On 05/25/2011 01:29 PM, N.M. Maclaren wrote:
On May 25 2011, Ralph Baker Kearfott wrote:Wasn't the "still unclear" addressed when motion 3 passed? That motion states that intervals are closed and connected sets of real numbers. Thus, 0 * (any interval) = 0 (and infinite bounds are not to be construed to mean infinity is included.) P.S. Other extended interval arithmetics define 0*\infty, and can also be used. However, we have decided to standardize with 0*interval = 0. (personal opinion): it is more important, at least in this case, that the programmer know how the system defines it, and that the system is the same across platforms, than what the actual definition is.The REALLY critical thing is that it is consistent. If you adopt that approach, then anything/0 (or an interval containing it) MUST be a NaN (and not (-inf,+inf)), and there must be no way to lose NaNs. Regards, Nick Maclaren.
-- --------------------------------------------------------------- Ralph Baker Kearfott, rbk@xxxxxxxxxxxxx (337) 482-5346 (fax) (337) 482-5270 (work) (337) 993-1827 (home) URL: http://interval.louisiana.edu/kearfott.html Department of Mathematics, University of Louisiana at Lafayette (Room 217 Maxim D. Doucet Hall, 1403 Johnston Street) Box 4-1010, Lafayette, LA 70504-1010, USA ---------------------------------------------------------------