Implementation and Extension of the Variance-Reduced BGK Method in PICLas
L. Teichröb, F. Garmirian, M. Pfeiffer · 2026
Read Implementation and Extension of the Variance-Reduced BGK Method in PICLas by L. Teichröb, F. Garmirian, M. Pfeiffer (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-2606-25813.txt.
Introduction
Simulating thin gas flows particle by particle drowns weak signals in statistical noise. The slower the flow, the worse the drowning, and the cost explodes. L Teichröb, F Garmirian, and M Pfeiffer advance a variance reduced BGK scheme in 2026 and implement it in PICLas, an open source gas kinetics framework. The method tracks only the deviation from equilibrium, which is where the signal lives. This version adds modified flow estimators and collision operators for stability. Two collision models are shown, the Shakhov model and an ellipsoidal statistical one.
New features join them: adaptive equilibria, variable particle weights, and axisymmetric domains. Benchmarks span synthetic tests and one, two, and axisymmetric dimensions. Against plain BGK runs the new code agrees exactly, as it should. Against an analytical solution for thermal transpiration in a microchannel, it shows the efficiency gains that low signal flows need.
Catalog record
| CREATOR | L. Teichröb, F. Garmirian, M. Pfeiffer |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.25813
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Implementation and Extension of the Variance-Reduced BGK Method in PICLas L. Teichröb, F. Garmirian, M. Pfeiffer Abstract Traditional particle-based kinetic methods, such as DSMC, suffer from prohibitive computational cost in low-signal flows, where the deviation from thermodynamic equilibrium is small and statistical noise overwhelms the signal of interest. The Variance-Reduced BGK-DSMC scheme is further advanced and implemented to support this class of flows in the open-source gas-kinetics framework PICLas . Modified versions of flow estimators and collision operators enhancing stability are developed. The Shakhov and Ellipsoidal Statistical models for BGK are demonstrated, along with entirely new features such as adaptive equilibria, variable particle weights and domain axisymmetry. The implementation is validated using synthetic benchmarks, 1D, 2D and axisymmetric simulations. Comparison of VRBGK to BGK simulations shows exact agreement of the models. A further comparison with an analytical solution of thermal transpiration in a microchannel showcases the low-signal efficiency of the method as well as newly proposed features. 1 Introduction Particle simulation methods are a widespread solution for the investigation of rarefied gas flows, with direct simulation Monte Carlo (DSMC) [ 8 ] as the standard tool for high altitude aerodynamics [ 19 , 32 , 25 ] or microscale gas dynamics [ 24 , 4 , 1 ] . However, the stochastic nature of the Monte Carlo method necessarily leads to non-physical statistical noise in the results, as illustrated in Figure 1 . The quantities of interest, such as flow velocity or temperature, are estimated by sampling the position and velocity of simulation particles, thus leading to a variance in the results that scales inversely with the number of samples. For slow rarefied flows, such as those occurring in microelectromechanical systems (MEMS), the signal-to-noise ratio can become vanishingly small. Maintaining a constant signal-to-noise ratio (SNR) requires the sample size to scale inversely with the signal magnitude \sqrt{N}\propto 1/\left|x\right| ; for small signal flows this quickly leads to prohibitive computational costs. Another class of methods solves rarefied gas problems deterministically by discretizing the Boltzmann equation, or its Bhatnagar-Gross-Krook (BGK) approximation [ 7 ] , in phase space. Such discrete velocity methods (DVM) [ 21 , 34 , 14 , 12 ] are free of statistical noise, but can require a fine discretization of velocity space to accurately capture the distribution function. In particular for three-dimensional cases, solving the differential equation on the fine six-dimensional space-velocity grid again leads to high computational costs. Furthermore, the adaptability of particle methods to boundary conditions and velocity distributions is a clear advantage over DVM. Therefore, several techniques have been developed to reduce the variance in DSMC simulations, for instance deviational formulations of the Boltzmann equation [ 6 ] or the usage of control variates [ 2 , 3 ] . Because these methods retain the restrictive requirements of DSMC in terms of spatial and temporal discretization, such noise-reduced particle methods have also been developed for the BGK [ 16 , 20 ] and the Fokker-Planck [ 13 , 23 ] collision models. Besides the variance reduction, these approaches solve approximations of the Boltzmann equation without having to resolve mean free paths or collision frequencies. The Information Preservation (IP) method, first proposed by J. Fan and C. Shen [ 9 ] , is suitable for the application to low-signal, pressure-driven Poiseuille flows, such as those found in Knudsen pumps [ 30 ] . Later, the method was successfully extended to also correctly model the effects arising from temperature gradients [ 29 , 31 ] , including thermal transpiration [ 35 ] . Figure 1 : Comparison of the velocity field of a noisy (left) and a variance-reduced (right) Monte Carlo lid-driven cavity simulation. Among these noise reduction techniques, the variance-reduced BGK (VRBGK) method has been selected for extension and application within the scope of this work, since its formulation allows for a relatively straightforward implementation into an existing BGK particle solver. This method employs a control variate approach first introduced by [ 3 ] as variance-reduced DSMC (VRDSMC) and adapted for the BGK model by [ 16 ] . Contrary to the deviational formulation used by [ 6 ] , the importance weights associated with particles remain positive, which is a desirable property for the inclusion of complex processes in the simulated gas. The VRDSMC class of methods provide good performance and an ideal foundation for inclusion of more advanced solver features, due to their greater simplicity both in theoretical correctness and implementation. In addition, the conditional probability generalization of VRDSMC offers a reliable path forward for the inclusion of new models. In this paper, we present a novel implementation of the BGK variation of VRDSMC into the open-source gas-kinetics framework PICLas [ 11 ] , with a focus on a rigorous definition of the importance weights and corresponding estimators to improve the overall stability and variance reduction. The new implementation is feature-rich and aims to make the VRBGK method accessible to researchers for direct application to practical problems. The implementation supports efficient 2D axisymmetric simulations, 3D, adaptive boundaries and numerous other quality-of-life features offered by building on top of a well-established solver. Additionally, the method is extended to support the Shakhov (SBGK) [ 27 ] and the ellipsoidal statistical (ESBGK) [ 15 ] collision models, which, unlike the classic BGK model, attain the correct Prandtl number in simulations. By extending the local equilibrium approach, simulations encompassing regions of differing densities and temperatures become accessible to VRBGK. Section 2 begins with the theory underlying control variates as well as a proof of the small-signal variance scaling to provide a foundation for discussing the changes we propose to the original method. This includes both a modification to the calculation of flow variables and to the relaxation operator, enhancing their stability. Practical questions of implementation, such as particle cloning, boundary-handling, and adaptive equilibria are documented in Section 3 . Validation of the method using a combination of synthetic benchmarks as well as practical simulations is presented in Section 4 and is followed by concluding remarks on difficulties and opportunities facing the class of VRDSMC methods. 2 Method Variance reduction in PICLas [ 22 ] is achieved using the BGK variation of the VRDSMC method, (VR) 2 -BGK [ 16 ] , with a number of important modifications that are detailed in the following section. “(VR) 2 -BGK” will be referred to as VRBGK for the remainder of this text. 2.1 Control Variate Theory For DSMC methods, a variance reduction method seeks to reduce the variance of the estimator of a velocity moment R(c)=c^{k} , k\in\mathbb{N}_{0} without increasing the required number of samples. The “control variate” method is well known and has been in use since the inception of Monte Carlo methods for neutron transport problems [ 18 ] . The control variate method works on the principle of finding a correlated estimate \bar{R}_{W} with analytically known mean \mathbb{E}\left[R_{W}\right] , allowing the common statistical noise to be removed. Taking \beta to be any real constant, \displaystyle\bar{R}_{VR} \displaystyle=\bar{R}-\beta\left(\bar{R}_{W}-\mathbb{E}\left[R_{W}\right]\right) (1) is an unbiased estimator of R that, for a suitable choice of \beta and correlation of the control variate, benefits from reduced variance. In the context of this work, due to the near-unity correlation corr(R,R_{W})\approx 1 that is observed, \beta=1 is