Re: IEEE Standard 1788-2015

["Svetoslav Markov" <smarkov@xxxxxxxxxx>]

> I wonder whether they would be useful for the basic standard, at least
> not to solve the dependency problem.

Vincent,

In section 3.2 of Motion 12 (attached) we explain how
  the inner operations can be used for the "dependency
problem".

Roughly speaking,  one has the option to decide whether to
use standard or inner operations, providing some monotonicity
information is available (wrt  the direction of "tracing" the interval
arguments).

Svetoslav



On 28 Sep 2015 at 15:59, Vincent Lefevre wrote:

Date sent:      	Mon, 28 Sep 2015 15:59:20 +0200
From:           	Vincent Lefevre <vincent@xxxxxxxxxx>
To:             	Svetoslav Markov <smarkov@xxxxxxxxxx>
Copies to:      	Mehran Mazandarani <me.mazandarani@xxxxxxxxx>,
       	stds-1788 <stds-1788@xxxxxxxxxxxxxxxxx>,
       	"Kreinovich, Vladik" <vladik@xxxxxxxx>,
       	Andrzej Piegat <apiegat@xxxxxxxxxxxxx>
Subject:        	Re: IEEE Standard 1788-2015

> On 2015-09-27 10:50:10 +0300, Svetoslav Markov wrote:
> > In the standard there are interval operations that
> > allow you to obtain x-x=0, these are the cancel plus/minus
> > operations. However, the cancel +/-  are not "operations" in the
> > sense of algebra.
> 
> so that replacing x-y by the cancel minus form to get x-x=0 would be
> incorrect (let's make this clear).
> 
> The dependency problem is mainly a language issue, because the notion
> of "same source" can only be considered at this level initially. IMHO,
> the easiest way to formalize this would be from the computation DAG,
> but such information is not available at the IEEE 1788 level.
> 
> > I would suggest that the cancel plus/minus "operations" are included
> > in the basic standard.
> 
> I wonder whether they would be useful for the basic standard, at least
> not to solve the dependency problem.
>

Attachment: inner_add-sum-12.03.pdf
Description: Binary data