Re: Branch & bound for not everywhere defined constraints

[John Pryce <j.d.pryce@xxxxxxxxxxxx>]

Nate

On 9 Sep 2009, at 00:31, Nate Hayes wrote:
> > Nate, is such a scheme efficiently implementable? If so, it seems to
> > handle cases B1, B2 and B3 above without the tags on intervals, or
> > extra (explicit) output arguments of operations, required by
> > solutions D1, D2 above.
>
> Obviously, for case B2 we don't need to conern ourselves with the  
> flags. So
> the interesting cases are B1 and B3. IMHO, what you describe (or  
> something
> along these lines) is most likely required for B3. However, for B1,  
> why must
> the user bother allocating the extra memory for the flags? ...

I agree. I just meant that this mechanism seems able to handle each  
of cases B1 to B3 efficiently, though B3 is the one that needs it most.

Obviously, hardware could be built to make this "pFLAG" method  
efficient. But can it be done efficiently on typical existing  
architectures? Say, for
   sqrt(vector interval)
or more demanding
   (vector interval)/(vector interval)?

John