Solving MINLPs to global optimality with FICO® Xpress global

We present the architecture and central parts of the FICO Xpress Global optimization solver. In particular, we focus on how we built a global solver for the general class of mixed-integer nonlinear optimization problems by combining and extending two existing components of the FICO Xpress Solver, namely the mixed-integer linear optimization solver and the successive linear approximation-based local solver for nonlinear optimization. We give a detailed report on the performance impact of each component: presolve, cutting, branching, heuristics, and parallelization.