Classical AC power flow methods can fail to converge when operating conditions become infeasible or when discrete controls, such as generator reactive limits, are enforced in ex-post schemes. This paper revisits a mixed complementarity formulation of AC power flow and extends it to jointly handle feasible and infeasible conditions through a relaxed optimization model, minimizing mismatches from the specified operating point while embedding generator reactive limits via complementarity constraints. The formulation introduces an additional degree of freedom by penalizing voltage setpoints, allowing the model to reconcile inconsistent data and exploit the flexibility of reactivelimit enforcement. It is implemented as an open-source module within the PowSyBl framework and solved using the Artelys Knitro interior-point algorithm. Numerical experiments on standard and large-scale systems show that the approach consistently restores feasibility, provides interpretable diagnostics, and replaces traditional outer-loop procedures for generator reactive limits by embedding their logic directly within the inner solution process.

