[Coin-discuss] Strange termination

[Kasper Bonne Rasmussen]

On Mon, 20 Jun 2005 js at jeannot.org wrote:

> > The status is still CPXMIP_INForUNBD (119) after the second solve.
> 
> That's strange. Do you have a public MPS file on which CPLEX exhibit this
> behavior ?

Inspired by your question i tried to save the problem as it was when
the solver returned CPXMIP_INForUNBD.

if(!si->isProvenOptimal())
	si->writeMps("problem");

and then start CPLEX manually and have it read and solve the problem. CPLEX
solved it with out complaining in 0.87 sec.

Since my problem is in fact solvable it must be something else that causes
the crash (maybe a memory-leak in the OSI?).

The algorithm I'm writing consists of a master lp-problem and a number of
mip-subproblems in a column generation setup. I solve the master problem
first, then all the subproblems. Based on their results i add more columns
to the master problem and resolve, then solve the subproblems again, and
so on until the solution of the master problem no longer improves.

That means that I'm asking CPLEX to solve an lp-problem then some
mip-problems and then resolve the lp-problem. Can I do that whit out 
saving the lp-solution somehow? Since I'm using the resolve feature and
not the initial solve every time, isn't CPLEX saving some information for
me? Maybe that information is getting mangled by the mip-problems?

One thing to note is that the algorithm fails while solving the
mip-problems and *not* the master problem, so I don't really know what to
think.

If anyone has any ideas, tested or not, please let me know.

Regards
/Kasper