DSMC: Fast direct simulation Monte Carlo solver for the Boltzmann equation by Multi-Chain Markov Chain and multicore programming

This paper demonstrates the newly developed particle method in the Direct Simulation Monte Carlo (DSMC) framework, and we refer to it as the Maxwellian–DSMC (M–DSMC) method. The M–DSMC solver is utilized for various fluid flow problems with different length scales varied from continuum to transitional regime. The hypersonic flow of argon over a cylinder simulation is carried out using M–DSMC and the same simulation results are compared with the regular DSMC method. The present work shows that the M–DSMC results are quite a good match with the DSMC results. The computational cost is significantly reduced by using M–DSMC method and the same is compared with regular DSMC method.

The Boltzmann equation is the governing equation for rarefied gas dynamics, which is relevant to the aerodynamic design of rarefied hypersonic vehicles. Direct solution of the Boltzmann equation for realistic gas flows using either analytic or computational techniques is almost impossible. However, the Direct Simulation Monte Carlo (DSMC) method has been developed to provide realistic solutions for rarefied gas flows. Bird formally introduced this method hi 1969, and ever since several arguments have developed as to whether the DSMC method provides solutions of the Boltzmann equation. This paper demonstrates that the Boltzmann equation and the DSMC method have a common origin which rests on the fundamental tenets of kinetic theory. These fundamental tenets are mathematically condensed into the Boltzmann equation. On the other hand, the DSMC method is a numerical implementation of these same fundamental tenets. Thus, the Boltzmann equation and the DSMC method express the same idea but use a different language. The Boltzmann equation uses the language of mathematics whereas the DSMC method uses the language of physics. It is explained how both approaches are related and it is concluded that the DSMC method does indeed provide solutions of the Boltzmann equation. It is also shown that both approaches can be easily extended to a wider range of problems than the literature has recognised (such as to high-density gases and liquids). One important issue is how to set the range of the molecular potential used in deflection angle formulae. It is shown that, contrary to traditional expectations, within a macroscopic volume the range of the potential is always finite. Finally, the thorny issue of molecular chaos is examined and it is concluded that it is not as limiting as commonly feared.

The massively parallel computer Connection Machine is utilized to map an improved version of the direct simulation Monte Carlo (DSMC) method for solving flows with the Boltzmann equation. The kinetic theory is required for analyzing hypersonic aerospace applications, and the features and capabilities of the DSMC particle-simulation technique are discussed. The DSMC is shown to be inherently massively parallel and data parallel, and the algorithm is based on molecule movements, cross-referencing their locations, locating collisions within cells, and sampling macroscopic quantities in each cell. The serial DSMC code is compared to the present parallel DSMC code, and timing results show that the speedup of the parallel version is approximately linear. The correct physics can be resolved from the results of the complete DSMC method implemented on the connection machine using the data-parallel approach. 41 refs.

The macroscopic chemistry method for modelling non-equilibrium reacting gas flows with the direct simulation Monte Carlo (DSMC) method is developed and tested. In the macroscopic method, the calculation of chemical reactions is decoupled from the DSMC collision routine. The number of reaction events that must be performed in a cell is calculated with macroscopic rate expressions. These expressions use local macroscopic information such as kinetic temperatures and density. The macroscopic method is applied to a symmetrical diatomic gas. For each dissociation event, a single diatom is selected with a probability based on internal energy and is dissociated into two atoms. For each recombination event, two atoms are selected at random and replaced by a single diatom. To account for the dissociation energy, the thermal energies of all particles in the cell are adjusted. The macroscopic method differs from conventional collision-based DSMC chemistry procedures, where reactions are performed as an integral part of the collision routine. The most important advantage offered by the macroscopic method is that it can utilise reaction rates that are any function of the macroscopic flow conditions. It therefore allows DSMC chemistry calculations to be performed using rate expressions for which no conventional chemistry model may exist. Given the accuracy and flexibility of the macroscopic method, it has significant potential for modelling reacting non-equilibrium gas flows. The macroscopic method is tested by performing DSMC calculations and comparing the results to those obtained using conventional DSMC chemistry models and experimental data. The macroscopic method gives density profiles in good agreement with experimental data in the chemical relaxation region downstream of a strong shock. Within the shock where strongly non-equilibrium conditions prevail, the macroscopic method provides good agreement with a conventional chemistry model. For the flow over a blunt axisymmetric cylinder, which also exhibits strongly non-equilibrium conditions, the macroscopic method also gives reasonable agreement with conventional chemistry models. The ability of the macroscopic method to utilise any rate expression is demonstrated by using a two-temperature rate model that accounts for dissociation-vibration coupling effects that are important in non-equilibrium reacting flows. Relative to the case without dissociation-vibration coupling, the macroscopic method with the two-temperature model gives reduced dissociation rates in vibrationally cold flows, as expected. Also, for the blunt cylinder flow, the two-temperature model gives reduced surface heat fluxes, as expected. The macroscopic method is also tested with a number density dependent form of the equilibrium constant. For zero-dimensional chemical relaxation, the resulting relaxation histories are in good agreement with those provided by an exact Runge-Kutta solution of the relaxation behaviour. Reviews of basic DSMC procedures and conventional DSMC chemistry models are also given. A method for obtaining the variable hard sphere parameters for collisions between particles of different species is given. Borgnakke-Larsen schemes for modelling internal energy exchange are examined in detail. Both continuous rotational and quantised vibrational energy modes are considered. Detailed derivations of viscosity and collision rate expressions for the generalised hard sphere model of Hassan and Hash [Phys. Fluids 5, 738 (1993)] and the modified version of Macrossan and Lilley [J. Thermophys. Heat Transfer 17, 289 (2003)] are also given.

The Direct Simulation Monte Carlo (DSMC) method was widely used to simulate low density gas flows with large Knudsen numbers. However, DSMC encounters limitations in the regime of lower Knudsen numbers (Kn<0.05). In such cases, approaches from classical computational fluid dynamics (CFD) relying on the continuum assumption are preferred, offering accurate solutions at acceptable computational costs. In experiments aimed at imaging aerosolized nanoparticles in vacuo a wide range of Knudsen numbers occur, which motivated the present study on the analysis of the advantages and drawbacks of DSMC and CFD simulations of rarefied flows in terms of accuracy and computational effort. Furthermore, the potential of hybrid methods is evaluated. For this purpose, DSMC and CFD simulations of the flow inside a convergent–divergent nozzle (internal expanding flow) and the flow around a conical body (external shock generating flow) were carried out. CFD simulations utilize the software OpenFOAM and the DSMC solution is obtained using the software SPARTA. The results of these simulation techniques are evaluated by comparing them with experimental data (1), evaluating the time-to-solution (2) and the energy consumption (3), and assessing the feasibility of hybrid CFD-DSMC approaches (4).

Establishment of particle motion model and study of particle volume fraction distribution in the turbo air classifier based on DSMC

Rarefied flow conditions can have an impact on performance and radiation characterization at not only high-altitudes conditions but also for separated flow regions either behind blunt bodies or around divert jets at high continuum altitudes. Under rarefied flow conditions, standard continuum CFD methodologies fall short. More appropriate are Direct Simulation Monte Carlo (DSMC) techniques. However, DSMC techniques can quickly become intractable for flows that are either primarily rarefied and have small continuum regions, or, that are primarily continuum with regions of rarefied flow. The solution is to partition the flow into continuum and non-continuum regimes separated by an interface surface. Unfortunately, the interface surface can be quite complex leading to a substantial amount of ‘man-in-the-loop’ effort to obtain a surface that is both well-posed and compatible with the DSMC solver. An automated Arbitrary Breakdown Surface Generation procedure is developed to address this situation. This automated procedure not only reduces the amount of ‘man-in-the-loop’ effort but is also general enough to apply to a wide variety of continuum/non-continuum coupling problems. Particulates (from solid motors) are also of interest for high-altitude plume flowfields. Past efforts to simulate particles within the rarefied regime have decoupled the particulate simulation from the rarefied gas simulation. This inefficient process is eliminated by implementing particle integration directly into the DAC97 DSMC solver. Some generic straightback plume flow field simulations and divert jet studies serve to demonstrate the capabilities of these new procedures.

A hybrid particle scheme, which is based on the direct simulation Monte Carlo (DSMC) method and incorporates a low diffusion particle method for continuum flow simulation, is extended for improved efficiency and wider applicability. The hybrid scheme is intended for simulation of gas flows involving a wide range of Knudsen number regimes, where strong coupling is desired between calculations in rarefied and continuum flowfield regions. A series of modifications are presented for cell-based numerical weight and time step adaptation, diffusive transport of nonequilibrium internal energy modes, and decreased sensitivity to DSMC cell size. Proposed modifications are employed in a hybrid simulation of a Mach 20 flow around a cylinder with a global Knudsen number of 0.002, for which a factor of 2.6 efficiency increase relative to full DSMC computation is demonstrated. I. Introduction n a variety of gas flow problems, portions of the flowfield exhibit large translational nonequilibrium effects and require relatively expensive simulation techniques based on the Boltzmann equation, while other regions are within low Knudsen number (Kn) continuum flow regimes and can be prohibitively expensive to simulate using such Boltzmann equation techniques. These “multiscale” flow problems include flows around atmospheric entry vehicles and hypersonic aircraft, where local Kn values based on gradient length scales may be very small through much of the flowfield but become large within shock, forebody boundary layer and wake regions. Other examples include rocket exhaust plumes and fuel venting flows at high altitude, where the mean free path may differ by several orders of magnitude between nearfield and farfield regions. A. Simulation Methods for Multiscale Gas Flows Accurate simulations may be performed for some of these flows using a two step, one-way coupled approach. 1-3 In this type of approach, continuum and rarefied regions are considered independently and different simulation techniques are applied in each region. In general, however, effects of two-way coupling between rarefied and continuum regions must be considered, and a single simulation employing a coupled hybrid algorithm is required. In one popular type of hybrid algorithm, a computational fluid dynamics (CFD) technique based on the Navier-Stokes equations is used in low-Kn continuum regions, the direct simulation Monte Carlo (DSMC) method is employed in rarefied regions, and additional procedures are used for evaluation of continuum breakdown and two-way information transfer between CFD and DSMC calculations.

Introduction:A Direct Simulation Monte Carlo (DSMC) solver with a modified collisional routine is used to investigate an argon gas flow through a millimeter-scaled thruster nozzle with high-density ratios.Method:The limiter scheme, denoted as the constant selection limiter (CSL), limits the possible number of selected collisional pairs to a constant value in accordance with the present simulation particles in the cell.Results:Results of the CSL scheme are compared with the experimental and numerical results of a compressible Navier–Stokes solver and discussed in comparison with baseline DSMC simulations. The influence of collision limitation by the CSL is discussed on the stagnation pressure of the thruster and on thrust and specific impulse prediction. The application of the limiter scheme makes the prediction of stagnation pressure challenging in some cases.Discussion:In contrast, thrust and specific impulse are predicted well, and their study remains valid. Investigated mass flow rates are 0.178 mg/s ≤m.≤ 71.360 mg/s, and flow Knudsen numbers below Kn = 0.01 and over Kn = 10 are present. Near atmospheric conditions are reached inside the thruster, generating pressure ratios up to 3,741 along the nozzle. The computational performance of the scheme is also discussed, and speed-up factors up to 0.51 are achieved.

In the continuum flow regime, the Navier-Stokes (NS) equations are usually used for the description of gas dynamics. On the other hand, the Boltzmann equation is applied for the rarefied flow. These two equations are based on distinguishable modeling scales for flow physics. Fortunately, due to the scale separation, i.e., the hydrodynamic and kinetic ones, both the Navier-Stokes equations and the Boltzmann equation are applicable in their respective domains. However, in real science and engineering applications, they may not have such a distinctive scale separation. For example, around a hypersonic flying vehicle, the flow physics at different regions may correspond to different regimes, where the local Knudsen number can be changed significantly in several orders of magnitude. With a variation of flow physics, theoretically a continuous governing equation from the kinetic Boltzmann modeling to the hydrodynamic Navier-Stokes dynamics should be used for its efficient description. However, due to the difficulties of a direct modeling of flow physics in the scale between the kinetic and hydrodynamic ones, there is basically no reliable theory or valid governing equations to cover the whole transition regime, except resolving flow physics always down to the mean free path scale, such as the direct Boltzmann solver and the Direct Simulation Monte Carlo (DSMC) method. In fact, it is an unresolved problem about the exact scale for the validity of the NS equations, especially in the small Reynolds number cases. The computational fluid dynamics (CFD) is usually based on the numerical solution of partial differential equations (PDEs), and it targets on the recovering of the exact solution of the PDEs as mesh size and time step converging to zero. This methodology can be hardly applied to solve the multiple scale problem efficiently because there is no such a complete PDE for flow physics through a continuous variation of scales. For the non-equilibrium flow study, the direct modeling methods, such as DSMC, particle in cell, and smooth particle hydrodynamics, play a dominant role to incorporate the flow physics into the algorithm construction directly. It is fully legitimate to combine the modeling and computation together without going through the process of constructing PDEs. In other words, the CFD research is not only to obtain the numerical solution of governing equations but to model flow dynamics as well. This methodology leads to the unified gas-kinetic scheme (UGKS) for flow simulation in all flow regimes. Based on UGKS, the boundary for the validation of the Navier-Stokes equations can be quantitatively evaluated. The combination of modeling and computation provides a paradigm for the description of multiscale transport process.

In rarefied gas dynamics scattering kernels deserve special attention since they contain all the essential information about the effects of physical and chemical properties of the gas–solid surface interface on the gas scattering process. However, to study the impact of the gas–surface interactions on the large-scale behavior of fluid flows, these scattering kernels need to be integrated in larger-scale models like Direct Simulation Monte Carlo (DSMC). In this work, the Gaussian mixture (GM) model, an unsupervised machine learning approach, is utilized to establish a scattering kernel for monoatomic (Ar) and diatomic (H2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\hbox {H}_{2}$$\\end{document}) gases directly from Molecular Dynamics (MD) simulations data. The GM scattering kernel is coupled to a pure DSMC solver to study isothermal and non-isothermal rarefied gas flows in a system with two parallel walls. To fully examine the coupling mechanism between the GM scattering kernel and the DSMC approach, a one-to-one correspondence between MD and DSMC particles is considered here. Benchmarked by MD results, the performance of the GM-DSMC is assessed against the Cercignani–Lampis–Lord (CLL) kernel incorporated into DSMC simulation (CLL-DSMC). The comparison of various physical and stochastic parameters shows the better performance of the GM-DSMC approach. Especially for the diatomic system, the GM-DSMC outperforms the CLL-DSMC approach. The fundamental superiority of the GM-DSMC approach confirms its potential as a multi-scale simulation approach for accurately measuring flow field properties in systems with highly nonequilibrium conditions.

An attempt to use direct simulation Monte Carlo (DSMC) method for pulsed laser ablation at laser fluence as high as 17 J/cm2 is analyzed. Under these conditions, however, the evaporation flux is very high which results in a very high laser plume density. Therefore, both collision rate and collisional zone near the target are too large to effectively use a conventional DSMC technique whose possibilities are always limited by finite computer storage and available computation time. Nevertheless, the authors of the considered paper claim to propose a new approach to overcome the restrictions of DSMC. We found that the main idea of the approach is the plume reduction in space. However, according to the dimension theory, under this transformation one should preserve the Knudsen number, that defines the degree of rarefaction, and, hence, the number of cells to be used in DSMC. Therefore, the number density of the plume should be increased and particle collisions should be calculated with higher frequency than in the case without space reduction. In our opinion, this important point was neglected by the authors, the collision frequency was not increased, and the number of cells was not properly chosen. As a result, the proposed approach does not break the DSMC restrictions, and the results obtained do not correspond to the experiments with high evaporation flux.

This paper describes the basic study of the simulation of chemical vapor deposition (CVD) step coverage that is used to estimate the deposition of thin films in LSI production. Given the trend toward extremely high density in the semiconductor manufacturing process, such as that of LSIs, the deposition process of silicon thin film on the circuit pattern needs to be accurate. To optimize the deposition of thin film around the trench of the order of less than a micron, computer simulations of the step coverage and of the film growth rate in the process are very important. In this simulation reported in this paper, the thin film deposition process, whose reaction is modeled as silicon from silane (SiH4) and silylene (SiH2), is solved using the direct simulation Monte Carlo (DSMC) method. The DSMC method is usually adopted in this case, but, since 1) this method solves the Boltzmann equation with a huge amount of molecular collisions and free motions and 2) sticking probability of molecules is of the order of 10-5 [1], even a high performance computer needs much time for this calculation. Although the performance of supercomputers and other types of computers is improving, the DSMC method as conventionally applied seems to be reaching its limit in terms of practicality.

In this paper, an ellipsoidal statistical (ES) Bhatnagar–Gross–Krook (BGK)-type kinetic model with velocity-dependent collision frequency is proposed and further numerically tested for one-dimensional shock waves and planar Couette flow at steady state for hard sphere molecules. In this new kinetic model, a physically meaningful expression for the velocity-dependent collision frequency derived from the Boltzmann equation is used, while the important properties for a kinetic model are retained at the same time. This kinetic model can be simplified to the classical ES-BGK model and the BGK model with velocity-dependent collision frequency for suitable choices of parameters. The H theorem for this new kinetic model has so far been proven only for small Knudsen numbers. The numerical method used here for kinetic models is based on Mieussens’s discrete velocity model [L. Mieussens, J. Comput. Phys. 162, 429 (2000)]. Computational results from the kinetic models (including the BGK model, the ES-BGK model, the BGK model with velocity-dependent collision frequency, and this new kinetic model) are compared to results obtained from the direct simulation Monte Carlo (DSMC) method. It is found that results obtained from this new kinetic model lie in between results from the ES-BGK model and results from the BGK model with velocity-dependent collision frequency. For one-dimensional shock waves, results from this new kinetic model fit best with results from the DSMC, while for planar Couette flow, the classical ES-BGK model is suggested.

In every system, if Knudsen number is larger than 0.1, the Navier-Stokes equation will not be satisfied for investigation of flow patterns. In this condition, the Boltzmann equation, presented by Ludwig Boltzmann in 1872, can be useful. The conditions that this equation can be used were investigted by Cercignani in 1969. The most successful method for solving Boltzmann equation for a rarefied gas system is Direct Simulation Monte Carlo (DSMC) method. This method was suggested by Bird in 1974. The cylindrical Couette flow and occurrence of secondry flow (Taylor vortex flow) in a annular domin of two coaxial rotating cylinders is a classical problem in fluid mechanics. Because this type of gas flow can occur in many industrical types of equipment used in chemical industries, Chemical engineers are interested in this problem. In 2000, De and Marino studied the effect of Knudsen number on flow patterns and in 2006 the effect of temperature gradient between two cylinders was investigated by Yoshio and his co-workers. The aim of the present paper is investigation of understanding of the effect of different conditions of rotation of the cylinders on the vortex flow and flow patterns.