Stokeslet-Driven Boundary Modeling for Efficient Drag Prediction in Rarefied Gas Flows

Modeling the behavior of microscopic particles suspended in gases is essential in many aspects of life from the targeted delivery of inhalable nanoparticles in drug therapy, the control of airborne contaminants in semiconductor fabrication, to the behavior of nanoplastics in the atmosphere. Understanding how these particles move through rarefied gases is critical. At the heart of these questions lies a dsimple force: drag. Yet, despite its apparent simplicity, accurately calculating the drag force acting on micro- and nano-sized particles in non-continuum gas environments remains an unsolved challenge in many practical settings. This complexity arises because the size of these particles is comparable to the mean free path of surrounding gas molecules, thrusting the flow into what’s known as the transition regime, where neither classical fluid dynamics nor free-molecular theory alone is sufficient. For particles smaller than a micron, especially those falling in the 10 nm to 1 μm range, the Knudsen number (Kn)—a measure of flow rarefaction—tends to sit between 0.1 and 10. This is the transition regime: a zone where conventional continuum-based solvers, such as the Navier-Stokes equations with slip conditions, start to lose accuracy, but full kinetic models, like the Boltzmann equation, become computationally formidable. The go-to numerical method for simulating such systems is Direct Simulation Monte Carlo (DSMC), a particle-based approach capable of resolving the Boltzmann equation. However, DSMC is extremely sensitive to the size of the computational domain. To get an accurate measure of drag, the simulation domain must be made enormous to ensure that the outer boundaries are far enough from the particle to not reflect artificial confinement effects—an approach that quickly becomes prohibitively expensive, especially for particles of irregular geometry. The computational burden scales cubically with the domain size, making it practically impossible to simulate large ensembles of particles or explore a wide parameter space.

New research paper published in Journal of Computational Physics  and led by Professor Matthew Borg from The University of Edinburgh and conducted by Giorgos Tatsios, Nikos Vasileiadis and Livio Gibelli alongside Professor Duncan Lockerby from the School of Engineering at University of Warwick in England, the researchers developed a novel far-field boundary condition based on the analytical Stokeslet solution to accurately capture flow disturbances caused by micro- and nanoscale particles in rarefied gas environments. This boundary condition significantly reduces the required simulation domain size in DSMC calculations, enabling accurate drag force prediction with far less computational effort. By incorporating both velocity and stress disturbances into the boundary, their approach maintains fidelity to the underlying physics even in compact domains. This development opens the door to efficient modeling of arbitrary-shaped particles in the transition regime without compromising accuracy.

The researchers designed simulations using the DSMC method, focusing on both spherical and non-spherical particles in the transition flow regime. The simulations were performed using the open-source solver dsmcFoam+, allowing them to compare conventional boundary conditions against their novel Stokeslet-corrected approach. In the first set of simulations, they modeled the flow around a stationary sphere subjected to a low-speed, rarefied gas flow. They selected Knudsen numbers ranging from 0.4 to 1.0—values that are typical of micro- and nanoscale particles in ambient air—while maintaining a fixed Mach number to ensure the flow remained in the low-Reynolds regime. The computational domain radius was varied relative to the particle radius to assess how domain size affected the accuracy of drag force predictions under both boundary conditions. The authors found when using conventional boundary conditions, drag predictions were significantly inflated unless the domain extended 80 times beyond the particle radius—an impractical demand for 3D DSMC simulations. In contrast, the Stokeslet-corrected boundary condition delivered highly accurate drag estimates with a domain size only five times larger than the particle, representing a computational cost reduction by a factor of about 4000. These results were not just theoretical improvements; they closely matched benchmark experimental data and previously validated numerical results, giving confidence that the method was both correct and practical. In the second phase, the team extended the simulations to a prolate spheroid, a more complex geometry often encountered in real-world particulate systems. Again, they compared the convergence of drag calculations as the domain size increased. The Stokeslet-corrected approach proved robust, demonstrating much faster convergence than the conventional setup. While the improvement in convergence rate wasn’t as extreme as with the sphere—likely due to higher-order flow effects not fully captured by the Stokeslet alone—it still allowed for substantial domain reduction. Importantly, the correction held even when the drag force itself was dynamically updated within the simulation loop, showing the method’s stability despite the stochastic nature of DSMC.

The implications of the new study reach far beyond a mere reduction in computational cost. What the researchers have done is redefine how we approach drag force prediction in flow regimes where traditional models falter and brute-force computation has long been the default. By introducing a boundary condition rooted in the analytical structure of low-Reynolds-number flow, they’ve offered the community a practical tool to simulate micro- and nanoscale particle dynamics with unprecedented efficiency. This has deep relevance for disciplines that depend on accurate modeling of particle behavior in rarefied gases—from aerosol medicine and nanotoxicology to contamination prevention in ultra-clean manufacturing environments. More profoundly, this work serves as a bridge between fluid theory and numerical methods. The use of the Stokeslet—a classical solution that’s long been known in low-speed hydrodynamics but rarely exploited in kinetic simulations—represents a conceptual shift. It shows that our computational boundaries need not be silent or neutral. On the contrary, when boundaries are informed by the physical structure of the disturbance field, they can actively guide the simulation toward truth with far less numerical labor. We think one of the most immediate consequences is accessibility. Researchers and engineers who previously avoided detailed DSMC simulations due to cost constraints now have a viable alternative. Tasks that once required access to national supercomputing resources can now be executed on modest clusters or even high-end workstations. This democratizes advanced rarefied flow modeling and enables broader exploration of particle behaviors in the transitional regime. Moreover, the method is generalizable. While the authors implemented and tested their boundary condition within a DSMC framework, the formulation is equally applicable to other solvers, including those based on the discrete velocity method or even hybrid approaches. In liquid-phase systems, it may influence future adaptations in molecular dynamics simulations involving colloids or biological particles.