Multilevel radial basis function surrogates for noise-robust DSMC-CFD coupling

Arshad Kamal, Arun Kumar Chinnappan, James R. Kermode, Duncan A. Lockerby · 2026

Read Multilevel radial basis function surrogates for noise-robust DSMC-CFD coupling by Arshad Kamal, Arun Kumar Chinnappan, James R. Kermode, Duncan A. Lockerby (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-24225.txt.

Introduction

Kamal and three colleagues make hybrid rarefied gas simulation work on complex geometries with multilevel radial basis function surrogates. Coupling particle-based DSMC to continuum CFD saves cost, but statistical noise destabilizes runs and global bases restrict shapes. Their multilevel radial basis functions represent smooth corrections locally, resolving broad and fine flow details inside the MMS-Sparse framework. Coupled to an OpenFOAM finite-volume solver and tested on the rarefied lid-driven cavity flow, the method matches benchmarks while staying noise-proof and automated.

Catalog record

CREATORArshad Kamal, Arun Kumar Chinnappan, James R. Kermode, Duncan A. Lockerby
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65830
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.24225

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Multilevel radial basis function surrogates for noise-robust DSMC-CFD coupling Arshad Kamal, Arun Kumar Chinnappan, James R. Kermode, Duncan A. Lockerby Abstract Hybrid methods for simulating rarefied gas flows reduce computational cost by coupling a particle-based model, typically the direct simulation Monte Carlo (DSMC) method, to a continuum-based solver, i.e. a computational fluid dynamics (CFD) code. However, widespread adoption of these methods is hindered by numerical instabilities caused by statistical noise and difficulties in applying them to complex, arbitrary geometries. To be effective, a hybrid framework must be robust to noise, reliable in not introducing errors to the flow physics, automated, and flexible enough for general spatial domains. Previous iterations of the micro-macro-surrogate-sparse (MMS-Sparse) method successfully addressed the first three requirements using Bayesian surrogate models to provide smooth, constitutive corrections to the CFD. However, they relied on global basis functions, limiting their applicability to relatively simple geometries. In this work, we address the fourth requirement - flexibility - by introducing a set of multilevel radial basis functions (RBFs) to represent the smooth corrections within the MMS-Sparse framework. Unlike global polynomials, multilevel RBFs can resolve broad and fine flow details locally, allowing the method to be applied to complex geometrical systems. We couple this approach with a finite-volume CFD solver (OpenFOAM) and validate it using the rarefied lid-driven cavity flow problem. This serves as a rigorous test case for spatially two-dimensional coupling. Our results demonstrate that this enhanced MMS-Sparse method produces estimates in good agreement with benchmarks while retaining the noise-robust and automated benefits of the Bayesian approach. 1 Introduction The study of rarefied gas flows is important in numerous industrial applications across various spatial and time scales, such as in the modelling of atmospheric re-entry in aerospace, or in the design of micro-electromechanical systems (MEMS) in microfluidics. A gas can be referred to as being ‘rarefied’ when the mean free path, \lambda , of the molecules is large in comparison to the characteristic length, L , of the system [ 1 ] . As such, one can associate a non-dimensional quantity, the Knudsen number Kn , to a rarefied gas flow: Kn=\lambda/L. (1) Here, the mean free path \lambda denotes, on average, the distance travelled by a gas molecule between two successive collisions, while L is a quantity that signifies the scale of the system. For example, on a large scale, of order meters, the flow past the spacecraft in atmospheric re-entry at higher altitudes has a comparatively large mean free path compared to the characteristic length of the system (i.e. \lambda\gg L ). On much smaller scale, in the case of MEMS devices, the characteristic length scale is of order micrometres to millimetres and is comparatively small compared to the mean free path (i.e. L\ll\lambda ). Correspondingly, in both examples, on opposite scales in length, the Knudsen number Kn can be significant (i.e. Kn\gg 1 ) and the gas can be rarefied. As one increases Kn from 0 , we move across four gas regimes [ 2 ] , where flow becomes more ‘particle-like’ and rarefied. In the first regime, with Kn<0.001 , the flow can be modelled as a ‘continuum’ fluid, where the Navier-Stokes (N-S) equations can be applied with no-slip boundary conditions. Increasing to 0.001<Kn<0.1 , the N-S equations, modified with appropriate slip boundary conditions, can be used to accurately capture the macroscopic properties. However, as we increase Kn even further to 0.1<Kn<10 , the accuracy of the N-S equations, even with slip boundary conditions, becomes limited. In this ‘transitional’ regime, and particularly so in the ‘free molecular’ regime Kn>10 , the Boltzmann equation, or some approximation, must be adopted. Although the Boltzmann equation is applicable across all four gas flow regimes, it is extremely difficult to solve analytically or numerically, except in the free molecular regime, due to its complex collision integral term. Therefore, the direct simulation Monte Carlo (DSMC) method [ 3 ] , a kinetic-theory-based, stochastic technique that directly simulates the Boltzmann equation, is typically employed in the transitional and free molecular regimes. DSMC simulates representative particles, where each simulated molecule represents a group of real gas molecules. Molecular movement is treated deterministically, while collisions are modelled probabilistically using an acceptance-rejection algorithm. Macroscopic properties are extracted from microscopic data through time-averaging for steady-state solutions and ensemble-averaging for unsteady cases. Since DSMC inherently provides solutions consistent with the Boltzmann equation, it is, in principle, valid across all regimes. However, for accuracy, the DSMC algorithm requires that the computational cell size be smaller than the local mean free path, with at least 15–20 simulated particles per cell, and that the time step be smaller than the mean collision time. In the slip and continuum regimes, this leads to a large number of very small cells, a reduced time step, and a significantly increased number of simulated particles; thereby greatly increasing the computational cost. Additionally, the statistical nature of the method necessitates a significant number of samples to achieve results with low statistical noise, particularly at low fluid velocities. For instance, the number of samples needed for low-noise results scales approximately with the inverse square of the local flow Mach number [ 4 ] . These factors make this method computationally challenging in the transitional and late slip regimes and practically prohibitive in the continuum regime for low-speed flows. The high computational burden of the DSMC method is a strong motivation for the use of hybrid domain decomposition methods, where the domain of the physical problem is decomposed into regions of near-equilibrium and non-equilibrium. For example, in the case of microchannel flows in the slip regime [ 5 ] , regions close to the boundaries can be in non-equilibrium due to the presence of ‘Knudsen layers’, while, in the bulk, the flow is in near-equilibrium. Decomposing the domain, particle-based solutions to the Boltzmann equation can be obtained in the Knudsen layers using, for example, the DSMC method [ 3 ] , and continuum methods to solve the N-S equations can be employed in the remaining near-equilibrium regions. Such hybrid domain decomposition methods are widely documented in the literature [ 6 ] [ 7 ] [ 8 ] [ 9 ] [ 10 ] [ 11 ] [ 12 ] [ 13 ] [ 14 ] [ 15 ] [ 16 ] . While such hybrid coupling methods present an efficient solution for simulating rarefied gas flows at high speeds, coupling at low speeds presents unique challenges, preventing their general application to arbitrary flow conditions. The first challenge relates to the transfer of noisy data between the particle-based solver and the continuum-based CFD model, which can lead to numerical instabilities in the coupling algorithm [ 11 ] [ 10 ] . While the DSMC noise can be reduced by increasing the number of particles per cell and/or the sample window, the increased computational cost associated with this can outweigh the benefits of adopting a hybrid approach in the first place. While one may instead use simple filtering/averaging strategies for denoising, such approaches do not require that information transferred to the CFD satisfy physical conservation equations, nor do they guarantee that the filter parameters are not manually tuned or do not spuriously affect the overall solution. Another challenge to coupling

Read the complete text