Global Optimization of Nonlinear Mixed-Integer Programming

Global optimization of nonlinear mixed-integer programming (MINLP) addresses decision problems in which some variables are constrained to take integer values, while the objective function or constraints include nonlinear terms. Such problems arise in areas as diverse as process engineering, energy systems design, finance and logistics. The mixed-integer structure introduces combinatorial complexity, while nonlinearity often generates multiple local optima. Global solvers therefore combine techniques from integer programming and nonlinear programming to certify global optimality. Key algorithmic components include convex relaxations to bound the objective, branching schemes to explore discrete choices and reformulation methods to tighten relaxations. Outer‐approximation, spatial branch-and-bound and hybrid algorithms integrate these elements in order to strike a balance between computational tractability and solution quality. Advances in global MINLP have unlocked large-scale applications—for example, in optimal design of microgrids, water distribution networks and chemical process intensification—highlighting the global significance and practical utility of this class of problems.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent developments have focused on solver frameworks, reformulation strategies and transformation techniques. A major update to a prominent constraint integer programming framework has substantially enhanced its capacity to solve both convex and nonconvex MINLPs. The latest version introduces improved bound-tightening routines, extended constraint handlers for nonlinear expressions and more robust benchmarking protocols to compare global solvers under standardised conditions. In another strand, global solutions of nonconvex standard quadratic programs have been achieved via mixed-integer linear programming reformulations. Two novel formulations convert a quadratic form over the simplex into a linear model with binary variables and valid inequalities; computational tests show orders-of-magnitude speed-ups over classical global approaches. Finally, state-of-the-art surveys of transformation and linearisation techniques collate methods for handling products of variables, maximum/minimum operators, absolute values and square-root terms. By analysing piecewise approximations, Taylor expansions and novel variable‐substitution strategies, researchers have derived tighter linear relaxations that reduce solver effort and broaden the applicability of global MINLP methods.

Global Optimization of Nonlinear Mixed-Integer Programming publication trend

The graph below shows the total number of articles in global optimization of nonlinear mixed-integer programming across all publications each year (not limited to Nature Index journals).

Technical terms

Global optimization: The process of finding the absolute best solution of a given problem over all feasible choices, in contrast to local methods that may terminate at near-optimal points.

Mixed-Integer Nonlinear Programming (MINLP): An optimization framework combining integer decision variables with nonlinear relationships in the objective function or constraints.

Convex relaxation: A tractable approximation obtained by replacing nonconvex elements with convex counterparts to produce a bound on the original problem’s objective.

Branch-and-bound: A systematic search procedure that partitions the feasible region into subproblems, computes bounds for each, and prunes regions that cannot contain the global optimum.

Reformulation: The process of transforming an optimization problem into an equivalent or stronger form, often by introducing auxiliary variables or valid inequalities to improve solver performance.

References

1. An algorithmic framework for convex mixed integer nonlinear programs. Discrete Optimization (2008).
2. Global optimization of mixed-integer nonlinear programs with SCIP 8. Journal of Global Optimization (2023).
3. Global solutions of nonconvex standard quadratic programs via mixed integer linear programming reformulations. Journal of Global Optimization (2021).
4. Transformation and Linearization Techniques in Optimization: A State-of-the-Art Survey. Mathematics (2022).