Hi Manuel. Thanks for using OSQP!
When maximum iterations is hit, similar to the case when one gets solution inaccurate, OSQP internally multiplies
eps_prim_inf by a factor of 10. Then it checks the conditions for primal and dual infeasibility like stated in the paper. Hence there is a good chance that the obtained solution if feasible however this now depends on the inital setting of
Hope I didn’t get this wrong.
Yes this is correct. Feasibility, infeasibility and optimality are always up to the tolerances provided. OSQP reports the inaccurate statuses when the conditions are satisfied after relaxing the related tolerance by one order of magnitude.
l \le Ax \le u with using the standard settings of eps_dual_inf and eps_prim_inf results mostly in infeasible solutions that are not reported by OSQP.
Thanks for reporting it. Would it be possible to send us a minimal working example (MWE) for us to try?