Machine Learning Interatomic Potential for Atmospheric Chemistry

Aerosols consist of solid or liquid particulate matter suspended in the atmosphere, varying in chemical composition and dimension. They play crucial roles in Earth’s climate system by affecting radiative forcing, cloud formation, and air quality, for instance, thus impacting both the environment and human health. Organic compounds, in particular, can largely contribute to the initial stage of particle aggregation[1, 2]. Computational chemistry methods have been fundamental to elucidating the reactions and processes involving aerosol particles in Earth’s atmosphere[3]. Nevertheless, these tools are limited by the size of the systems under investigation due to computational expenses, demanding faster and cheaper alternatives for large-scale modeling. Here, we present a scheme for a machine learning interatomic potential (MLIP) trained on the atmospherically relevant organic molecules derived from the GeckoQ dataset[1]. This model can be utilized for molecular dynamics simulations, providing results at the same level of accuracy as DFT, besides being capable of expediting the exploration of conformational chemical space. In addition, our MLIP will be trained to predict the saturation vapor pressure (a measure of a molecule’s volatility) of atmospheric molecules instead of only energies and forces. This particular property is central in atmospheric chemistry since organic molecules with low saturation vapor pressure tend to participate in particle formation processes. We anticipate the devised interatomic potential can supplant conventional quantum chemistry methods in further studies in aerosol chemistry. One of the most promising applications is the investigation of larger systems, such as accretion products (a class of large, low-volatility organic compounds resulting from chemical reactions) and clusters. Understanding the role of these products is essential in atmospheric chemistry as they are considered paramount to particle formation.

This work was supported by the VILMA (Virtual Laboratory for Molecular-Level Atmospheric Transformations) Center of Excellence, funded by the Academy of Finland under grant 13346377.

[1] Vitus Besel et al. “Atomic structures, conformers and thermodynamic properties of 32k atmospheric molecules”. In: Scientific Data 10.1 (July 2023). issn: 2052-4463. doi: 10.1038/s41597-023-02366-x. url: http://dx.doi.org/10.1038/s41597-023-02366-x.

[2] Veli-Matti Kerminen et al. “Atmospheric new particle formation and growth: review of field observations”. In: Environmental Research Letters 13.10 (Sept. 2018), p. 103003. issn: 1748-9326. doi: 10.1088/1748-9326/aadf3c. url: http://dx.doi.org/10.1088/1748-9326/aadf3c.

[3] Jonas Elm et al. “Quantum chemical modeling of organic enhanced atmospheric nucleation: A critical review”. In: WIREs Computational Molecular Science 13.5 (May 2023). issn: 1759-0884. doi: 10.1002/wcms.1662. url: http://dx.doi.org/10.1002/wcms.1662.