Re: Please listen to Ulrich here...
Bill Walster wrote:
> Virtually all real engineering problems have measured inputs, the
> interval bounds to which must be non-degenerate.
Yes -- but that does not mean that the resulting intervals will ALL have
uncertain bounds. If a range is split into subranges, the splitting
points will be exact floating-point numbers, and in some cases they need
to be treated as exact through a transformation to preserve mathematical
consistency.
Also, non-linear transformations may sharpen intervals to the point where
they become single-ulp intervals -- though they would not become degenerate.
Finally, some engineering quantities are discrete, such as degrees of
freedom for example.
Michel.
---Sent: 2013-08-20 17:21:30 UTC