Title:A Globally Convergent Method for Computing B-stationary Points of Mathematical Programs with Equilibrium Constraints

Abstract:This paper introduces a method that globally converges to B-stationary points of mathematical programs with equilibrium constraints (MPECs) in a finite number of iterations. B-stationarity is necessary for optimality and means that no feasible first-order direction improves the objective. Given a feasible point of an MPEC, B-stationarity can be certified by solving a linear program with equilibrium constraints (LPEC) constructed at this point. The proposed method solves a sequence of LPECs, which either certify B-stationarity or provide an active-set estimate for the complementarity constraints, and nonlinear programs (NLPs) - referred to as branch NLPs (BNLPs) - obtained by fixing the active set in the MPEC. A BNLP is more regular than the MPEC, easier to solve, and with the correct active set, its solution coincides with the solution of the MPEC. The method has two phases: the first phase identifies a feasible BNLP or certifies local infeasibility, and the second phase solves a finite sequence of BNLPs until a B-stationary point of the MPEC is found. The paper provides a detailed convergence analysis and discusses implementation details. In addition, extensive numerical experiments and an open-source software implementation are provided. The experiments demonstrate that the proposed method is more robust and faster than relaxation-based methods, while also providing a certificate of B-stationarity without requiring the usual too-restrictive assumptions.