Fix inversion of constraint bounds in conditional bounds presolve

[kaatish]

Updating the constraint bounds based on minimal and maximal activity can sometimes cause the update constraint lower bounds to be greater than constraint upper bounds even if the constraint is feasible within tolerance.

This PR fixes the constraint bounds to the same value if this happens.

[chris-maes]

pb.tolerances.absolute_tolerance is pretty large: 1e-4 if I remember correctly. I think ideally this would be around 1e-6. Maybe we should introduce a new tolerance here that we can remove later once we bring the absolute tolerance down to 1e-6.

[kaatish]

relative_tolerance default value is set to 1e-6. Do you want to just use that? If not, what should I name the new tolerance as?

[chris-maes]

Maybe we should add a tolerance called presolve_absolute_tolerance = 1e-6. We can remove it later when we set the absolute_tolerance = 1e-6.

[chris-maes]

Thanks for the fix @kaatish . Would you also be able to add a tolerance to check_var_bounds_sanity and check_constraint_bounds_sanity? We should not throw an error unless the lower bound is greater than the upper bound plus a tolerance.

We also should consider if we should declare the problem infeasible (rather than throwing an exception) if check_var_bounds_sanity or check_constraint_bounds_sanity are triggered on the original problem from the user.

[kaatish]

Thanks for the fix @kaatish . Would you also be able to add a tolerance to check_var_bounds_sanity and check_constraint_bounds_sanity? We should not throw an error unless the lower bound is greater than the upper bound plus a tolerance.

Same as above? Tolerance of 1e-6?

We also should consider if we should declare the problem infeasible (rather than throwing an exception) if check_var_bounds_sanity or check_constraint_bounds_sanity are triggered on the original problem from the user.

I can return a bool from check_bounds_sanity and return false from run_presolve if the check fails. What you are proposing is to add those checks (apart from the checks we already have) before we run trivial_presolve?

[chris-maes]

Thanks for the fix @kaatish . Would you also be able to add a tolerance to check_var_bounds_sanity and check_constraint_bounds_sanity? We should not throw an error unless the lower bound is greater than the upper bound plus a tolerance.

Same as above? Tolerance of 1e-6?

Yes, I would use 1e-6. You can refer to it as presolve_absolute_tolerance.

We also should consider if we should declare the problem infeasible (rather than throwing an exception) if check_var_bounds_sanity or check_constraint_bounds_sanity are triggered on the original problem from the user.

I can return a bool from check_bounds_sanity and return false from run_presolve if the check fails. What you are proposing is to add those checks (apart from the checks we already have) before we run trivial_presolve?

No need to add any additional checks. I'm just proposing we mark the problem as infeasible (rather than throwing an exception) if the check fails before presolve. During presolve if the bounds cross, we should already return Infeasible before calling this check.

[chris-maes]

LGTM

[kaatish]

/merge