Originally posted by: domokane

I have a 14 variable problem which is diagonal and positive definite but CPLEX says the problem is unbounded.

This is even simpler than the problem I listed in another question.

I attach the LP file.

Originally posted by: Laci Ladanyi

Thanks for the report. Unfortunately it looks like you have found a bug in cplex :-(. A workaround for now is to use the quadratic simplex method (tranopt in the interactive) or explicitly set the barrier algorithm to be used to "constant start" (set barrier algorithm 2).

Originally posted by: domokane

Can you please specify exactly what setting I need. 

If I set (in Python)

        alg = my_prob.parameters.qpmethod.values

        my_prob.parameters.qpmethod.set(alg.barrier)

it still fails with a Unbounded error.

It should work with one of the following:

alg = my_prob.parameters.qpmethod.values
my_prob.parameters.qpmethod.set(alg.dual)

or

alg = my_prob.parameters.barrier.algorithm.values
my_prob.parameters.barrier.algorithm.set(alg.infeas_constant)

These correspond to the suggestions above, respectively.

Originally posted by: domokane

Thanks. The second approach appears to work in all cases tried so far. The first does not.

Originally posted by: Laci Ladanyi

Could you post an instance for which the first case fails?

Originally posted by: EdKlotz

Just to follow up on this, the bug was fixed in CPLEX 12.6.1, released in December of 2014.