RE: YES on Motion P1788/0019.01
Two comments:
1) minor point: since one can always easily move from one form to another (modulo infinite endpoints and modulo accuracy) if a problem is NP-hard it is NP-hard no matter what the representation is. I think what Dan means is that some algorithms using mid-rad are more efficient; Dan, please clarify and explain
2) major point: what is happening is that in effect, we continue the discussion instead of stopping it. Maybe it is beneficial to stop the voting and go back to discussion if this will influence our votes?
3) serious point: looks like what Arnold is saying is that Motion 19 contradicts to Motion 16. What if both are accepted? Maybe we should modify Motion 19 so that it include appropriate modification of Motion 16?
-----Original Message-----
From: stds-1788@xxxxxxxx [mailto:stds-1788@xxxxxxxx] On Behalf Of Dan Zuras Intervals
I am moved by some papers on the subject to consider
that we must find a way to make mid-rads both efficient
& well characterized.
In particular, interval methods have a tendency to turn
floating-point tasks into combinatorial tasks. Thus,
one sometimes has to look at exponentially many endpoints
to find a tightest enclosure.
There are linear algebra problems for which Jacobian
methods exist that circumvent this problem. At least to
a good approximation & only for the mid-rad forms.
This means that some problems which are NP-hard for
inf-sup forms are polynomial for mid-rad forms.
It is this that gives merit to mid-rads for me.