GENOA v3: A flexible framework for reduction and exploration of highly detailed chemical mechanisms

Abstract. Comprehensive atmospheric chemical mechanisms for volatile organic compound (VOC) oxidation contain thousands to millions of reactions and species, presenting major computational challenges for large-scale or long-term simulations. As mechanism complexity continues to increase, reduction strategies are required to enable their use in atmospheric modeling while preserving accuracy.

This paper presents the GENerator of Optimized Atmospheric chemical mechanisms (GENOA v3), a major advancement over earlier GENOA versions that enables scalable reduction of highly detailed mechanisms containing up to millions of reactions and species. GENOA v3 combines fast, strategy-driven threshold-based reduction (TBR) with simulation-based reduction (SBR) that explicitly controls accuracy. The framework is modular, graph-aware, and user-configurable, resulting in compact and chemically interpretable reduced mechanisms.

Applications to GECKO-A mechanisms for diverse VOC precursors across a range of scenarios show that TBR achieves mechanism size reductions of 20–90 % while preserving reasonable accuracy for metrics related to secondary organic aerosol (SOA) formation and gas-phase chemistry, with performance systematically dependent on precursor structure and chemical complexity across mechanisms. SBR achieves further reductions in mechanism size by several orders of magnitude; when trained with 15 % mean error constraints, SBR produces schemes within 0.02 % of the original size for preservation of SOA mass and 0.05 % when also considering gas-phase reactivity (e.g., OH, O₃, and NO₃).

These results demonstrate for the first time that GENOA v3 can reduce highly detailed chemical mechanisms while jointly preserving SOA mass and gas-phase reactivity, achieving substantial size reductions with reasonable accuracy across a wide range of scenarios. Continued application of GENOA v3 and growth of a user community will potentially support the development of libraries of reduced mechanisms and optimized reduction strategy sets tailored to specific modeling applications.