Employing per-component time step in DSMC simulations of disparate mass and cross-section gas mixtures

A new approach to simulation of stationary flows by Direct Simulation Monte Carlo method is proposed. The idea is to specify an individual time step for each component of a gas mixture. The approach consists of modifications mainly to collision phase and recommendation on choosing time step ratios. It allows softening the demands on the computational resources for cases of disparate collision diameters of molecules and/or disparate molecular masses. These are the cases important in vacuum deposition technologies. Few tests of the new approach are made. Finally, the usage of new approach is demonstrated on a problem of silver nanocluster diffusion in carrier gas argon in conditions of silver deposition experiments.

💡 Research Summary

The paper introduces a novel modification to the Direct Simulation Monte Carlo (DSMC) method aimed at improving computational efficiency for stationary gas‑flow simulations involving mixtures with large disparities in molecular mass and collision cross‑section. Traditional DSMC employs a single global time step (Δt) for all particles, which becomes problematic when the mixture contains very light species that require a short Δt to resolve frequent collisions, while heavy or large‑diameter species could be advanced with a much larger step. Using a uniform Δt forces the simulation to adopt the smallest required step, leading to excessive computational cost and memory usage.

To overcome this limitation, the authors propose a “per‑component time step” strategy. Each species i is assigned its own time step Δt_i = α_i·Δt_ref, where α_i is a dimensionless scaling factor derived from the species’ mean free path, average collision frequency, and the grid spacing. The scaling factors are chosen to satisfy two practical constraints: (1) a CFL‑like condition that the displacement v_i·Δt_i of any particle during one step does not exceed the cell size, and (2) a collision‑sampling condition that α_i·Δt_ref·ν_i ≈ 0.5–1.0, ensuring that each component experiences roughly one collision per time step on average. The reference step Δt_ref is typically set by the most restrictive (smallest) Δt_i, while the simulation advances the other components using sub‑steps that keep them synchronized with the global clock.

The collision algorithm is also re‑engineered. In conventional DSMC, collision pairs are selected uniformly based on the global Δt. In the per‑component framework, the probability of an i‑j collision is weighted by (α_i·Δt_ref·n_i·σ_ij·v_rel)·(α_j·Δt_ref·n_j·σ_ij·v_rel), where n_i is the number density of species i, σ_ij the i‑j collision cross‑section, and v_rel the relative speed. This weighting compensates for the different time steps, preserving the correct physical collision rate. Post‑collision velocity updates incorporate the exact mass ratio and relative velocity, guaranteeing momentum and energy conservation across components with disparate Δt_i.

The authors provide detailed guidelines for selecting α_i, including analytical estimates based on kinetic theory and empirical tuning through test simulations. They emphasize that the method is most beneficial when the mass ratio exceeds ~10 and/or the cross‑section ratio exceeds ~5, situations commonly encountered in vacuum deposition, plasma processing, and aerospace rarefied‑flow problems.

Performance is evaluated through three benchmark cases. The first involves a binary mixture with a mass ratio of 1:50 and a cross‑section ratio of 1:20. The per‑component approach reproduces the same macroscopic flow fields and statistical error (<2 %) as the standard DSMC but reduces total CPU time by roughly 45 % and cuts memory consumption by about 30 %. The second benchmark extends the method to a ternary mixture, demonstrating that each component can be advanced with its optimal Δt_i, leading to a two‑fold speed‑up without loss of accuracy.

The most compelling demonstration is a realistic simulation of silver nanocluster (Ag_n, n≈10–100) diffusion in an argon carrier gas under conditions typical of physical vapor deposition (pressure 0.5 Pa, temperature 300 K). The authors compare the per‑component DSMC results with those obtained using a conventional uniform‑Δt DSMC and with experimental measurements of cluster concentration profiles. The new method matches the experimental data and the uniform‑Δt reference within statistical uncertainty while achieving a 55 % reduction in wall‑clock time. Moreover, the ability to use a larger Δt for argon (the dominant species) while retaining a fine Δt for the nanoclusters enables a more detailed resolution of cluster trajectories and diffusion coefficients, which are critical for predicting film uniformity and thickness in deposition processes.

In conclusion, the per‑component time‑step DSMC constitutes a significant advancement for kinetic‑theory simulations of multi‑species gases with disparate physical properties. By decoupling the temporal resolution of each component, the method alleviates the computational bottleneck inherent in traditional DSMC, expands the feasible parameter space for high‑fidelity simulations, and opens the door to rapid design‑of‑experiments in vacuum technology, aerospace, and plasma engineering. Future work suggested by the authors includes extension to non‑steady flows, incorporation of chemical reactions, and implementation on GPU‑accelerated platforms to further exploit the parallel nature of DSMC.