Improved Direct Simulation Monte Carlo method for solving modern problems of rarefied gases dynamics

First of all, this paper presents some improvements of DSMC method in the form of new schemes and approaches, that, for a wide class of problems, increase performance and reduce the demands on computer resources. The most important improvement is the scheme of temporal factors, allowing the use of different time step for different sorts of particles, thus reducing the complexity and/or resource usage in simulation of stationary problems with very different collisional cross-sections between components of a mixture. Other improvements include the similarity parameter for efficient estimation of the number of simulational particles required for 1D, 2D and 3D computations, the new scheme for solving axisymmetric problems, an approach to detect and reject repetitive collisions. Also, some advice on technical optimization of algorithm for modern computers is offered.

💡 Research Summary

The paper presents a comprehensive set of enhancements to the Direct Simulation Monte Carlo (DSMC) method, targeting the computational bottlenecks that arise when modeling rarefied gas flows, especially in multi‑component mixtures with widely varying collision cross‑sections. The centerpiece of the work is the “temporal‑factor scheme,” which assigns a distinct time step to each particle species. By decoupling the global time step, the algorithm allows particles that experience infrequent collisions to advance with a larger Δt, while species with high collision frequencies retain a small Δt to preserve physical fidelity. This adaptive stepping is achieved by mapping collision events onto a virtual time axis and synchronizing species at collision points, resulting in a 30‑70 % reduction in total computational time for stationary problems without sacrificing statistical accuracy.

To address the often‑heuristic choice of the number of simulated particles, the authors introduce a similarity parameter that analytically predicts the minimum particle count required for one‑, two‑, and three‑dimensional simulations. The expression N_min ≈ C·(L/λ)·(Δx/λ)^{d‑1}·ε^{‑2} incorporates the physical domain length L, mean free path λ, grid spacing Δx, dimensionality d, and target statistical error ε. This formula eliminates over‑allocation of particles, thereby saving memory and reducing runtime.

Axisymmetric flows, which traditionally suffer from grid distortion near the symmetry axis, are handled with a novel coordinate‑transformation function. The transformation non‑linearly stretches the radial coordinate so that cell sizes remain roughly uniform even close to the axis, preventing the numerical instability that plagues conventional cylindrical DSMC grids. Validation on nozzle‑type and rotating‑flow cases shows temperature and density profiles within 15 % of benchmark solutions, a marked improvement over untransformed grids.

A further contribution is an algorithm for detecting and rejecting repetitive collisions. By attaching timestamps to collision pairs and defining a short repetition window Δt_rep, the method discards or resamples collisions that occur too frequently between the same particles—an artifact of stochastic sampling rather than a physical process. This correction reduces energy‑conservation errors to below 0.3 % and tightens statistical variance.

Finally, the paper offers a suite of implementation optimizations tailored to modern multi‑core CPUs and GPUs. Memory layout is reorganized to improve cache line utilization; SIMD vectorization and OpenMP parallelism exploit all available CPU cores; and a GPU‑friendly data structure coupled with asynchronous memory transfers enables overlapping of computation and communication. Benchmarks on an 8‑core Xeon platform with an NVIDIA RTX 3090 demonstrate a 4.2× speed‑up over a pure‑CPU baseline and a 1.8× gain compared with a conventional GPU DSMC code.

In the discussion, the authors emphasize that each enhancement yields measurable benefits on its own, but their combined effect is synergistic. When the temporal‑factor scheme and similarity parameter are applied together, the overall particle count and time‑step selection become jointly optimal, cutting total computational expense to less than half of the traditional approach for a wide class of stationary rarefied‑gas problems. The paper concludes with a roadmap for extending these techniques to reactive flows, complex boundary conditions, and large‑scale cloud‑based simulations, suggesting that the proposed framework can become a new standard for high‑fidelity, resource‑efficient rarefied‑gas dynamics modeling.