[Coin-symphony] non-terminating problems

[Michael Hennebry]

On Fri, 21 Dec 2007, Ted Ralphs wrote:

> Yes, with integer programs, one must be very careful with tolerances.
> Occasionally, the LP solver will return solutions that are not within
> tolerances because of unscaled infeasibilities. A recent change to the
> code caused this situation not to be handled correctly. Release 5.1.7
> should fix this problem. Can you give it a try?

Not without considerable effort.
It's been a while since I did any integer programming.
I retain an interest.
Aware of the precision issues, I calculated how far
off I could be and still derive the right answer.
With this method no node was ever declared feasible.
Solutions always came from a rounding heuristic.

> Michael Hennebry wrote:
> > What I have done, many releases ago, with all-integer constraints,
> > is to calculate instance-specific tolerances.
> > The tolerances were selected to ensure that if I had an approximate
> > all-integer solution, rounding would not destroy feasibility.
> > That was why I needed all-integer data.
> > I could round, check whether it should be the new incumbent, and cut it off.

-- 
Michael   hennebry at web.cs.ndsu.NoDak.edu
"I AM DEATH, NOT TAXES.  *I* TURN UP ONLY ONCE."  --  Death