Motion P1788/M0009.01_ExactDotProduct NO
I vote NO.
Rational : Following Arnold Neumaier's and George Corliss arguments, I
think requiring optimal enclosure is too strong. I would change to yes
if instead of the exact scalar product an accurately rounded scalar
product would be required. Indeed, this sounds enough for most
applications with only drawback that different implementation may lead
to slightly different results, thus preventing exact reproductibility.
However, I think that reproductibility is less important for interval
algorithms than for floating point algorithms, because in the first
case every answers, though possibly different, will be correct in the
sense that they enclose the exact result.
On Wed, Nov 18, 2009 at 10:03 AM, Christian Keil <c.keil@xxxxxxxxxxxxx> wrote:
> My vote on this Motion is
>
> NO.
>
> I'm following Arnold's reasoning here. My vote would change to yes if we
> recommend or require an accurately rounded scalar product.
>
> Christian
>
--
Dr. Alexandre Goldsztejn
CNRS - University of Nantes
Office : +33 2 51 12 58 37 Mobile : +33 6 78 04 94 87
Web: www.goldsztejn.com
Email: alexandre.goldsztejn@xxxxxxxxxxxxxx