A simple interval challenge
I thought that the different models of IA only differed in the case of
division by intervals containing zero (as endpoint or as interior point).
In the presented test there is only one division, and the divisor is a
sum of squares, one of which is strictly positive.
I suppose a compiler or interpreter that computes squares as products
of independent factors could have trouble...
I am aware of methods that compute products faster, but with slightly
wider results, than full comparison of cross-products, but again I did
not think that this was a modal issue.
I'm not saying that this reduces the challenge of finding the tightest
enclosure in the fewest number of operations, and that the techniques
used for some IA models might work better, but I am still surprised.
Michel.
Sent: 2008-11-26 14:29:42 UTC