Stationarity and Convergence in Reduce-or-Retreat Minimization