Soot, the black carbonaceous particulate matter that billows from engines, industrial burners, and aircraft turbines, remains one of the most stubborn problems in combustion science. It damages human health, warms the atmosphere, and fouls machinery, yet predicting exactly how much of it a given flame will produce has long required computationally punishing simulations. Now, a team of chemical engineers at the University of Melbourne has unveiled a modeling framework that promises to change that calculus. In a study published in the journal Aerosol Research, Arash Fakharnezhad, Joseph D. Berry, and Eirini Goudeli describe a coupled computational fluid dynamics and monodisperse particle dynamics model that tracks soot formation in a laboratory flame with an accuracy approaching that of far more expensive detailed simulations, while sidestepping the need for elaborate reaction kinetic mechanisms altogether.
The difficulty stems from the sheer breadth of scales involved in soot formation. At the molecular level, small hydrocarbon radicals a few angstroms across react to build polycyclic aromatic hydrocarbons, the well-established precursors of soot. These molecular clusters evolve through physical or chemical nucleation into soot nuclei smaller than two nanometers, which then grow by surface reactions into primary particles. Within microseconds to milliseconds, those primary particles coagulate into fractal-like aggregates exceeding one hundred nanometers. Meanwhile, the reactive flow inside a combustor spans dimensions on the order of meters, with particle residence times measured in seconds. Capturing all of this in a single simulation has traditionally forced researchers to choose between fidelity and feasibility.
The most accurate traditional approach relies on sectional population balance models, which divide the soot particle population into many size bins and solve transport equations for each one. These models deliver excellent agreement with experiments but carry a computational cost that grows with the number of sections, rendering them impractical for multidimensional simulations of realistic engine geometries. The method of moments offers a cheaper alternative by tracking only a few statistical properties of the distribution, but it depends on closure assumptions that lose accuracy when the true particle size distribution deviates from the assumed shape. Semi-empirical correlations, such as the widely used Moss-Brookes model, are fast and simple but were calibrated for specific flame conditions and do not generalize reliably across different fuels and operating regimes.
The Melbourne team’s framework threads this needle by adopting a monodisperse particle dynamics formulation, which represents the soot population at any location by a single average particle size. Rather than resolving the full spread of sizes, the model tracks only three quantities: the total particle number density, the total carbon molar concentration contained in soot, and a surface area reconstructed from those two using a morphological relationship governed by a surface fractal dimension. Although this simplification can introduce deviations of up to roughly thirty percent relative to fully polydisperse treatments, the soot population rapidly approaches a self-preserving size distribution with an asymptotic collision frequency, so the average aggregate characteristics remain reasonably reliable at a fraction of the computational cost.
The truly novel ingredient, however, is where the model gets its chemistry. Instead of deriving nucleation and surface growth rates from detailed gas-phase kinetic mechanisms, which require calculating the concentrations of many polycyclic aromatic hydrocarbon species through large reaction networks, the researchers plugged in rate expressions extracted directly from reactive molecular dynamics simulations. The nucleation rate was obtained from simulations of isothermal acetylene pyrolysis at temperatures of 1200 to 1800 kelvin and high pressures of 189 to 568 atmospheres, monitoring the emergence of incipient soot clusters across a range of initial acetylene concentrations. The surface growth rate was similarly derived from simulations of incipient soot nanoparticles growing in acetylene pyrolysis, yielding an Arrhenius-type temperature dependence. Because these lumped rates depend only on acetylene concentration and temperature, they demand no prior knowledge of specific reaction intermediates or precursor species.
To benchmark the framework, the team simulated a premixed ethylene burner-stabilized stagnation flame, a well-characterized configuration featuring a porous plug burner, nitrogen shielding, and a cooled stagnation plate. The computational domain, reduced to a two-dimensional axisymmetric slice, was discretized into 46,400 quadrilateral cells with resolution down to about fifty micrometers near the inlet and plate. Combustion chemistry was handled with the reduced DRM22 mechanism comprising 22 species and 104 reactions, which the authors first validated against experimental species profiles in a comparable ethylene flame, confirming accurate predictions of oxygen and ethylene consumption and of water, carbon monoxide, hydrogen, and carbon dioxide formation. The predicted centerline temperature profile, peaking near 1700 kelvin a few millimeters above the burner, matched measurements closely.
The validation against soot measurements revealed a nuanced picture. In the baseline configuration, combining the semi-empirical Moss-Brookes nucleation rate with the classical hydrogen abstraction-carbon addition surface growth mechanism, the model predicted a soot number density within about six percent of a detailed sectional model from the literature and captured the measured soot volume fraction to within an order of magnitude. Notably, swapping in the molecular dynamics-derived nucleation rate while retaining the classical surface growth treatment improved the volume fraction prediction by roughly fifty-seven percent, bringing it within four percent of the sectional model’s result. Among all the reduced models tested, this molecular dynamics-informed nucleation variant reproduced the measured soot mobility diameters most accurately, particularly in the post-flame region beyond six millimeters above the burner.
The study also exposed instructive failures. Models employing the molecular dynamics-derived surface growth rate underpredicted soot loading by about two orders of magnitude, because the effective surface growth constant they carry is approximately three times lower than that of the hydrogen abstraction-carbon addition correlation, throttling carbon accumulation on particle surfaces. Conversely, the standard Moss-Brookes model as implemented in commercial software overpredicted the soot volume fraction by nearly three orders of magnitude, owing to an excessive surface growth contribution and an assumption that particles coagulate only by full coalescence into spheres, ignoring the fractal aggregates that real soot forms. Near the burner, at a plate separation of four millimeters, all reduced models overpredicted measurements by two to three orders of magnitude, a discrepancy the authors attribute partly to simplified nucleation chemistry and partly to sub-two-nanometer soot nuclei that standard condensation particle counters cannot detect.
For the authors, the significance lies less in any single number than in the demonstration that molecular dynamics-derived rate constants can be systematically assessed and deployed inside a CFD-coupled soot model, something they describe as the first such systematic evaluation. The framework captures soot volume fraction and particle size with accuracy comparable to detailed sectional modeling in the post-nucleation flame region, at dramatically lower computational cost, and without relying on reaction kinetic modeling of the precursor chemistry. That combination makes it a candidate tool for design-oriented simulations of engines, industrial burners, and aerosol synthesis reactors, where engineers need fast, trustworthy estimates of particulate emissions. The team cautions that the molecular dynamics rate constants were derived within a specific temperature and pressure window, and that additional benchmarking against other flame configurations will be necessary before the framework can be applied more broadly. Still, as regulators tighten limits on black carbon and the aviation industry confronts its soot footprint, a model that trades exhaustive chemistry for molecular-level insight at a fraction of the cost may prove exactly what practical combustion design has been waiting for.
Subject of Research: Multiscale simulation of soot particle nucleation, surface growth, and coagulation in flames using molecular dynamics-informed particle dynamics coupled with computational fluid dynamics
Article Title: Soot growth by monodisperse particle dynamics coupled with computational fluid dynamics
Article References: Fakharnezhad, A., Berry, J. D., & Goudeli, E. (2026). Soot growth by monodisperse particle dynamics coupled with computational fluid dynamics. Aerosol Research, 4(2), 279-291. https://doi.org/10.5194/ar-4-279-2026
Image Credits: AI Generated
Keywords: soot, combustion, computational fluid dynamics, molecular dynamics, particle dynamics, aerosol research, nucleation, surface growth, coagulation, ethylene flame, population balance, emissions modeling
News Source: Russell Cooper. (October 10, 2026). Molecular Dynamics Meets Flame Simulation to Predict Soot Growth Faster and More Accurately. Scienmag.
