Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: P-1788: Rough plan and request for volunteers

Am 09.10.2011 17:41, schrieb Ralph Baker Kearfott:

P-1788 working group:


In any case, here are the items, each of which seem to require
one or more motions.

Primary questions:
------- ---------

1. .....

     2. .....
     3. .....


4. Should the dot product mandated in Motion 9 (which passed) be
   faithful or accurate?
   ("Accurate" means the nearest floating point number to the
    exact result, whereas "faithful" means, I think, to within
    one or two ULPs of the exact result -- correct me if wrong.)
I think "exact" implies "accurate," but it was requested that this issue 
   be revisited.

   Note: Revol, Lefevre, and Rump have already volunteered to produce a
         position paper and motion on this, with me as acting
         chair, so Nathalie can actually participate in the
         motion.

------------------------------------------------------------------------
Let me comment on item 4.:
Motion 9 requests exact accumulation of products of two floating-point numbers (of finite floating-point numbers in the sense of IEEE 754. -oo and +oo are not considered to be floating-point numbers in P1788).
Immediately after the motion passed the authors (Ulrich Kulisch and Van Snyder) shortened and simplified the text and sent the new version to John Pryce and Juergen Wolff von Gudenberg. It should perhaps have been sent to all members of P1788. I attach it to this mail. 

For the SCAN-meeting at Lyon in 2010 I prepared a poster entitled:
"Fast and Exact Accumulation of Products". I attach this poster also.
The poster summarizes basic features of exact accumulation of products and it illustrates its realization. The basic idea of the method is not new. It can already be found on the very first computer by G. W. Leibniz, 1675.
The exact dot product has many applications in Numerical Analysis. See the references on the attachments. 

Best wishes
Ulrich

--
Karlsruher Institut für Technologie (KIT)
Institut für Angewandte und Numerische Mathematik (IANM2)
D-76128 Karlsruhe, Germany
Prof. Ulrich Kulisch

Telefon: +49 721 608-42680
Fax: +49 721 608-46679
E-Mail:ulrich.kulisch@xxxxxxx
www.kit.edu
www.math.kit.edu/ianm2/~kulisch/

KIT - Universität des Landes Baden-Württemberg und nationales Großforschungszentrum in der Helmholtz-Gemeinschaft

Attachment: DotProdP1788.pdf
Description: Adobe PDF document

Attachment: Poster22.pdf
Description: Adobe PDF document