Secondary organic aerosols derived from intermediate-volatility n-alkanes adopt low-viscous phase state
Integrating tgBoost into large-scale chemical kinetics software to resolve contrasting experimental and simulation results on aerosol viscosity.
Secondary organic aerosol (SOA) derived from n-alkanes, as emitted from vehicles and volatile chemical products, is a major component of anthropogenic particulate matter. Yet the chemical composition and phase state remain poorly understood—and thus poorly constrained in aerosol models.
The Challenge
Existing simulations of n-alkane SOA produced results that starkly contrasted with laboratory experiments, particularly regarding the phase state of the aerosol. The disconnect pointed to missing physics in how models represented molecular composition and its influence on viscosity.
The complexity is compounded by scale: simulating alkane oxidation requires generating millions of chemical reactions via rule-based algorithms and solving stiff ordinary differential equations for the resulting kinetic systems.
Our Approach
We coupled our tgBoost machine learning method—originally developed for glass transition temperature prediction—to a FORTRAN-based chemical mechanism generator. This integration enabled viscosity predictions at the scale of explicit gas-phase chemistry modeling.
The combined framework brings together:
- Explicit chemistry modeling: Rule-based generation of millions of oxidation reactions
- Kinetic solvers: Stiff ODE integration for chemical kinetics
- Machine learning: tgBoost predictions of glass transition temperature and viscosity from molecular embeddings
Results
The ML-enhanced simulations reproduced experimental observations, resolving the prior experiment–model discrepancy. Our analysis reveals a complex interplay between molecular composition and SOA viscosity:
- Higher carbon number: SOA consists mostly of less functionalized first-generation products → lower viscosity
- Lower carbon number: SOA contains more functionalized multigenerational products → higher viscosity
Both regimes correspond to low-viscous semi-solid or liquid states—not the highly viscous glassy states sometimes assumed.
Conclusions
The results indicate few kinetic limitations of mass accommodation in SOA formation, supporting the application of equilibrium partitioning for simulating n-alkane SOA in large-scale atmospheric models.
This study opens a new avenue for SOA process analysis by demonstrating that machine learning can bridge the gap between detailed chemistry and bulk aerosol properties.
Published in Atmospheric Chemistry and Physics Read the full paper →