Incidentally, I’ve seen at least one report of a bounded problem that CPLEX declared to be unbounded (and not due to numerical instability) — further proof that the problem you are solving is not the problem you think you are solving.

]]>That’s theory.

In practice, all the issues you raise can happen. ]]>

In practice people use floating point MIP solvers. It seems that in this case the myth is true given these floating point numbers are rational numbers.

]]>