There was discussion around implementation of an algorithm to aid with debugging infeasible models / time-steps. Currently when a model stops because a linear program is infeasible it is often very difficult to identify why. A proposed algorithm could do the following:
On LP infeasibility:
Relax (remove) node capacity constraints (minimum flows are often a problem) on at a time.
For each one ask the solver whether the relaxed LP would be feasible (this should be quick compared to actually solving).
If the relaxed problem is feasible report to the user that the nodes related to this constraint could be the problem.
Questions:
Would the reporting be via logging or a more structured output?
Would this happen any time there is an infeasibility, or should it be optional? If optional what is the default?
There was discussion around implementation of an algorithm to aid with debugging infeasible models / time-steps. Currently when a model stops because a linear program is infeasible it is often very difficult to identify why. A proposed algorithm could do the following:
On LP infeasibility:
Questions: