Re: General pow(xx,yy) function in Motion 10
Nate Hayes wrote:
> With the general power function, however, the width of the interval range
> zz will always be at least the distance between the two manifolds for any
> negative xx. This distance is quite large and will most likely be much
> larger than the user's tolerance criteria for the branch and bound algorithm.
What would be the meaning of an interval xx that includes relatively
large negative values, when raised to an uncertain real exponent?
It seems to me that the only sensible cases are those where xx is essentially
positive, but may include *small* (close to zero) negative values because of
earlier outward rounding, and we want to be able to continue computing with
this. Dan's pow(xx,yy) is the right tool for those cases.
(By "small" I mean that abs(xx.lo) is of the order of a few ulps of xx.hi)
Michel.
---Sent: 2009-12-07 04:11:55 UTC