Scalability strategies for automated reaction mechanism generation

• Published in: Computers & chemical engineering 2019-12, Vol.131 (C), p.106578, Article 106578
• Main Authors: Jocher, Agnes, Vandewiele, Nick M., Han, Kehang, Liu, Mengjie, Gao, Connie W., Gillis, Ryan J., Green, William H.
• Format: Article
• Language: English
• Subjects: Chemical kinetics; Mechanism generation; Parallel computing; Scalability
• ISSN: 0098-1354; 1873-4375
• DOI: 10.1016/j.compchemeng.2019.106578

•Parallel computing for reaction generation and chemical data computation allows for new opportunities in detailed modeling of diverse real-world processes.•Improved load balancing with distributed reaction families for reaction generation.•RMG wall clock time reduces by at least 85% for isolated reaction generation and 95% for isolated chemical data computation. Detailed modeling of complex chemical processes, like pollutant formation during combustion events, remains challenging and often intractable due to tedious and error-prone manual mechanism generation strategies. Automated mechanism generation methods seek to solve these problems but are held back by prohibitive computational costs associated with generating larger reaction mechanisms. Consequently, automated mechanism generation software such as the Reaction Mechanism Generator (RMG) must find novel ways to explore reaction spaces and thus understand the complex systems that have resisted other analysis techniques. In this contribution, we propose three scalability strategies — code optimization, algorithm heuristics, and parallel computing — that are shown to considerably improve RMG's performance as measured by mechanism generation time for three representative simulations (oxidation, pyrolysis, and combustion). The improvements create new opportunities for the detailed modeling of diverse real-world processes. [Display omitted]