A Chemist's guide to multi-objective optimization solvers for reaction optimization

Abstract

Recently, multi-objective optimization has garnered significant attention in the field of reaction optimization. Various multi-objective optimization solvers, such as MVMOO, EDBO+, Dragonfly, TSEMO, and EIM-EGO, have been developed and applied in real scenarios. However, the question of which solver to use persists, given that each problem is unique in terms of variables—be they continuous or categorical—and requires specific features, such as constraint handling and the capability for parallel evaluation. Although these solvers have been individually verified in real scenarios, a comparative analysis of their features and performance is lacking. This work focuses on assisting chemists in identifying the most suitable solver that best suits their problems, alongside a comparison of the different solvers' performances. For this purpose, the solvers were tested across 10 different chemical reaction-based in silico models, employing three metrics for performance comparison: hypervolume, modified generational distance, and worst attainment surface.