Currently, few approaches are available for mixed-integer nonlinear robust optimization. Those that do exist typically either require restrictive assumptions on the problem structure or do not guarantee robust protection. In this work, we develop an algorithm for convex mixed-integer nonlinear robust optimization problems where a key feature is that the method does not rely on a specific structure of the inner worst-case (adversarial) problem and allows the latter to be non-convex. A major challenge of such a general nonlinear setting is ensuring robust protection, as this calls for a global solution of the non-convex adversarial problem. Our method is able to achieve this up to a tolerance, by requiring worst-case evaluations only up to a certain precision. For example, the necessary assumptions can be met by approximating a non-convex adversarial via piecewise relaxations and solving the resulting problem up to any requested error as a mixed-integer linear problem.In our approach, we model a robust optimization problem as a nonsmooth mixed-integer nonlinear problem and tackle it by an outer approximation method that requires only inexact function values and subgradients. To deal with the arising nonlinear subproblems, we render an adaptive bundle method applicable to this setting and extend it to generate cutting planes, which are valid up to a known precision. Relying on its convergence to approximate critical points, we prove, as a consequence, finite convergence of the outer approximation algorithm.As an application, we study the gas transport problem under uncertainties in demand and physical parameters on realistic instances and provide computational results demonstrating the efficiency of our method.

Currently, there are few theoretical or practical approaches available for general nonlinear robust optimization. Moreover, the approaches that do exist impose restrictive assumptions on the problem structure. We present an adaptive bundle method for nonlinear and nonconvex robust optimization problems with a suitable notion of inexactness in function values and subgradients. As the worst-case evaluation requires a global solution to the adversarial problem, it is a main challenge in a general nonconvex nonlinear setting. Moreover, computing elements of an ε-perturbation of the Clarke subdifferential in the [Formula: see text]-norm sense is in general prohibitive for this class of problems. In this article, instead of developing an entirely new bundle concept, we demonstrate how existing approaches, such as Noll’s bundle method for nonconvex minimization with inexact information [Noll D (2013) Bundle method for non-convex minimization with inexact subgradients and function values. Computational and Analytical Mathematics, Springer Proceedings Mathematics, vol. 50 (Springer, New York), 555–592.] can be modified to be able to cope with this situation. Extending the nonconvex bundle concept to the case of robust optimization in this way, we prove convergence under two assumptions: first, that the objective function is lower C1 and, second, that approximately optimal solutions to the adversarial maximization problem are available. The proposed method is, hence, applicable to a rather general setting of nonlinear robust optimization problems. In particular, we do not rely on a specific structure of the adversary’s constraints. The considered class of robust optimization problems covers the case that the worst-case adversary only needs to be evaluated up to a certain precision. One possibility to evaluate the worst case with the desired degree of precision is the use of techniques from mixed-integer linear programming. We investigate the procedure on some analytic examples. As applications, we study the gas transport problem under uncertainties in demand and in physical parameters that affect pressure losses in the pipes. Computational results for examples in large realistic gas network instances demonstrate the applicability as well as the efficiency of the method. Summary of Contribution: Nonlinear robust optimization is a relevant field of research as real-world optimization problems usually suffer from not precisely known parameters, for example, physical parameters that cannot be measured exactly. Currently, there are few theoretical or practical approaches available for general nonlinear robust optimization. Moreover, the methods that do exist impose restrictive assumptions on the problem structure. Writing nonlinear robust optimization tasks in minimax form, in principle, bundle methods can be used to solve the resulting nonsmooth problem. However, there are a number of difficulties to overcome. First, the inner adversarial problem needs to be solved to global optimality, which is a major challenge in a general nonconvex nonlinear setting. In order to cope with this, an adaptive solution approach, which allows for inexactness, is required. A second challenge is then that the computation of elements from an ε-neighborhood of the Clarke subdifferential is, in general, prohibitive. We show how an existing bundle concept by D. Noll for nonconvex problems with inexactness in function values and subgradients can be adapted to this situation. The resulting method only requires availability of approximate worst-case evaluations, and in particular, it does not rely on a specific structure of the adversarial constraints. To evaluate the worst case with the desired degree of precision, one possibility is the use of techniques from mixed-integer linear programming. In the course of the paper, we discuss convergence properties of the resulting method and demonstrate its efficiency by means of robust gas transport problems.

Pareto efficiency for robust linear programs was introduced by Iancu and Trichakis in [9]. We generalize their approach and theoretical results to robust optimization problems in Euclidean spaces with linear uncertainty. Additionally, we demonstrate the value of this approach in an exemplary manner in the area of robust semidefinite programming (SDP). In particular, we prove that computing a Pareto robustly optimal solution for a robust SDP is tractable and illustrate the benefit of such solutions at the example of the maximal eigenvalue problem. Furthermore, we modify the famous algorithm of Goemans and Williamson [8] in order to compute cuts for the robust max cut problem that yield an improved approximation guarantee in non-worst-case scenarios.

Pareto efficiency for robust linear programs was introduced by Iancu and Trichakis in [Manage Sci 60(1):130–147, 9]. We generalize their approach and theoretical results to robust optimization problems in Euclidean spaces with affine uncertainty. Additionally, we demonstrate the value of this approach in an exemplary manner in the area of robust semidefinite programming (SDP). In particular, we prove that computing a Pareto robustly optimal solution for a robust SDP is tractable and illustrate the benefit of such solutions at the example of the maximal eigenvalue problem. Furthermore, we modify the famous algorithm of Goemans and Williamson [Assoc Comput Mach 42(6):1115–1145, 8] in order to compute cuts for the robust max-cut problem that yield an improved approximation guarantee in non-worst-case scenarios.