Elsevier

Journal of Computational Physics

Volume 298, 1 October 2015, Pages 602-621
Journal of Computational Physics

A fast spectral method for the Boltzmann equation for monatomic gas mixtures

https://doi.org/10.1016/j.jcp.2015.06.019Get rights and content

Abstract

Although the fast spectral method has been established for solving the Boltzmann equation for single-species monatomic gases, its extension to gas mixtures is not easy because of the non-unitary mass ratio between the different molecular species. The conventional spectral method can solve the Boltzmann collision operator for binary gas mixtures but with a computational cost of the order mr3N6, where mr is the mass ratio of the heavier to the lighter species, and N is the number of frequency nodes in each frequency direction. In this paper, we propose a fast spectral method for binary mixtures of monatomic gases that has a computational cost O(mrM2N4logN), where M2 is the number of discrete solid angles. The algorithm is validated by comparing numerical results with analytical Bobylev–Krook–Wu solutions for the spatially-homogeneous relaxation problem, for mr up to 36. In spatially-inhomogeneous problems, such as normal shock waves and planar Fourier/Couette flows, our results compare well with those of both the numerical kernel and the direct simulation Monte Carlo methods. As an application, a two-dimensional temperature-driven flow is investigated, for which other numerical methods find it difficult to resolve the flow field at large Knudsen numbers. The fast spectral method is accurate and effective in simulating highly rarefied gas flows, i.e. it captures the discontinuities and fine structures in the velocity distribution functions.

Introduction

There are a large number of applications for the Boltzmann equation (BE) in modeling rarefied gas flows frequently encountered in high-altitude aerodynamics, vacuum technologies, and microelectromechanical systems. While there is a great need for efficient and accurate methods for solving the BE, this is a challenge because the BE employs a molecular velocity distribution function (VDF) defined in a six-dimensional phase space to describe the system state, and the Boltzmann collision operator is highly complicated.

In 1912, Hilbert proposed obtaining approximate solutions to the BE by a series expansion in Knudsen number (Kn, defined as the ratio of the molecular mean free path to the characteristic flow length). The Hilbert expansion inspired Sone to develop an asymptotic analysis of the BE at small Kn, which helped reveal many interesting phenomena including the ‘ghost effect’ [1], [2]. In 1917, Chapman and Enskog (CE) obtained approximate solutions to the BE and calculated transport coefficients from first principles [3]. The first-order CE expansion recovers the Navier–Stokes equations, while the second- and third-order expansions lead to the Burnett and super-Burnett equations, respectively. Although the Burnett equations can be more accurate than the Navier–Stokes equations in some cases [4], they are rarely used nowadays due to their intrinsic instability to small-wavelength perturbations [5]. In 1949, by expanding the VDF into Gauss–Hermite polynomials, Grad proposed the moment method [6]. This has led to the development of regularized 13-moment [7] and 26-moment [8] equations, which have been successfully applied to rarefied gas flows up to Kn1 in some circumstances. For large Kn, a great number of moments are necessary but the convergence rate with increasing moment number is rather slow [9]. Moreover, multiple new boundary conditions are needed, so this approach it is not suitable to simulate highly rarefied gas flows.

For moderate and highly rarefied gas flows it is necessary to directly solve the BE. The direct simulation Monte Carlo (DSMC) method is the prevailing numerical technique for this [10]. In DSMC, the VDF is represented by a number of simulated particles that move in the computational domain and collide according to stochastic rules. The simulated particles automatically concentrate in regions where the VDF has large values, and discontinuities/fine structures in VDFs can be captured without difficulty. This method is efficient for hypersonic flows, however, its stochastic nature makes it not well suited for unsteady and low-speed flows. For instance, for a gas flow with a Mach number 0.001 (typically in micro-devices), about 108 independent samples are needed to reduce the error to 1% when there are 100 simulated particles in a cell [11]. Also, DSMC is an inefficient method for near-continuum flows, since the spatial and temporal steps must be smaller than the molecular mean free path and collision time, respectively. Progress has been made to ease these difficulties: to resolve small macroscopic quantities, the information preserving [12] and low variance [13], [14] DSMC methods have been proposed; to simulate near-continuum flows, hybrid continuum-particle approaches [15], [16], [17], [18], [19], [20], [21], time relaxed Monte Carlo method [22], [23], and particle–particle hybrid methods [24], [25] have been proposed. Recently, a new multiscale method has also been developed to simulate flows in geometries with high aspect ratio [26], [27], [28].

Deterministic numerical methods [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], based on the direct discretization of the BE in phase space, have distinct advantages in resolving small signals and low-speed flows as they are free of noise. Also, implicit time-marching can be employed to enlarge the time step [29], [30], [41], [42], [43]. Furthermore, compared to continuum/particle coupling, deterministic methods can be coupled more efficiently to macroscopic equations [44] such as the Navier–Stokes/moment equations and equations from the asymptotic kinetic analysis [2]. Among these deterministic solvers, however, only a few have been applied to spatially-inhomogeneous problems. The numerical kernel method [29], [30], in which the VDF is expanded into Laguerre polynomials for velocity components parallel to solid surfaces and quadratic finite-element functions for normal velocity component, provides accurate numerical results for many one-dimensional problems. As it allows non-uniform discretization of the normal velocity component, VDF discontinuities at large Kn are captured: for linearized Poiseuille and thermal transpiration flows, accurate results are produced up to Kn20 for hard-sphere molecules [45]. The conservative projection-interpolation method, in which uniform Cartesian velocity discretization is usually employed, and post-collision velocities are interpolated to closest neighbors, has been proposed to solve the BE with realistic intermolecular potentials in three-dimensional space [32], with a numerical error of the order Δv|ffM|v is the velocity grid step, f is the VDF, and fM is the equilibrium VDF) [44]. Since VDFs at post-collision velocities are obtained by polynomial interpolation, this method is accurate at small Kn where the VDF is usually smooth (in a recent paper [46] an accuracy of (Δv)2 has been demonstrated), but loses accuracy at large Kn where VDFs have discontinuities or steep variations. Discontinuous Galerkin methods, although able to capture VDF discontinuities, turn out to be feasible only for slightly rarefied subsonic gas flows [40] because of the high computational cost.

Recently, the fast spectral method (FSM) has been developed to numerically solve the BE for single-species monatomic gases [35], [47], [41], [42], [43]. It employs a Fourier–Galerkin discretization in velocity space, and handles binary collisions in the corresponding frequency space. The method is of spectral accuracy, and has a computational cost of O(M2N3logN), where N is the number of frequency nodes in each frequency direction and M2 is the number of discrete solid angles. Since it allows non-uniform velocity discretizations, and the number of frequency nodes can be far smaller than the number of velocity nodes [42], [43], the FSM is good at dealing with highly rarefied gas flows, where large numbers of velocity nodes are used to capture discontinuities/fine structures in the VDF: for linearized Poiseuille and thermal transpiration flows, accurate numerical results have been obtained even for Kn106 (note that even at such high Knudsen numbers, collision cannot be ignored). The FSM can be 50 times faster than the low variance DSMC method [42], and it is very efficient in dealing with linearized oscillatory flows [43] and Rayleigh–Brillouin scattering of light by rarefied gases [48], [49].

Deterministic numerical methods for the BE for mixtures of monatomic gases are even more scarce. Compared to the single-species BE, the non-unitary mass ratio between different molecular species poses an additional difficulty. Accurate numerical results have previously only been reported for hard-sphere molecules in some simple spatial configurations [50], [51], [52], [53], [54], [55]. For general cases, a multipoint conservative projection-interpolation method has been developed for an arbitrary ratio of molecular masses [56], but its accuracy is not known, especially at large Knudsen numbers. Recently, a spectral-Lagrangian method with a computational cost of O(mr3N6) has been proposed (where mr is the mass ratio of the heavier to the lighter species), and a normal shock wave in a binary mixture with a mass ratio of about 2 has been simulated [57]. As this method uses the same velocity discretization for each component, it cannot be applied to mixtures with large mass ratios, since the computational cost and storage are mr3 and mr3/2 times larger than that for the single-species BE, respectively. As an example, calculation of an argon–helium mixture with the mass ratio mr10 requires a 30-times increase in velocity grid points, and a 1000-times increase in computational time.

Because of the excellent performance of the FSM, in this paper we develop an efficient and accurate FSM for gas mixtures. The rest of this paper is organized as follows. In Section 2, the BE is introduced. The fast spectral approximation to the Boltzmann collision operator with general forms of the collision kernel is presented in Section 3, and its accuracy is assessed in Section 4 by comparing numerical results with analytical solutions for the spatially-homogeneous relaxation of pseudo-Maxwell molecules. In Section 5, an efficient algorithm is proposed to deal with gas mixtures of large molecular mass ratio. In Section 6, an iteration scheme is used to find the steady-state solutions of spatially-inhomogeneous problems, and various comparisons with results from the numerical kernel and DSMC methods are made for normal shocks and planar Fourier/Couette flows. A two-dimensional temperature-driven flow is also investigated. In Section 7, we conclude with a summary of our proposed numerical method, and outline future perspectives.

Section snippets

Boltzmann equation for a binary gas mixture

Consider a binary gas mixture of components A and B, where the mass of a molecule of component A is mA, and that of component B is mB. Without loss of generality, we assume mAmB. Let fA(t,x,v) and fB(t,x,v) be respectively the VDFs of the components A and B of the molecular velocity v=(v1,v2,v3) at spatial location x=(x1,x2,x3) and time t. In the absence of external forces, the BEs for a binary gas mixture of monatomic molecules take the following forms:fıt+vfıx=ȷ=A,BQıȷ(fı,fȷ)(ı=A,B).

Fast spectral method for the cross-collision operators

The approximation of the self-collision operator (with special forms of the collision kernel) by using FSM has been studied extensively [34], [36], [41], [42]. Here we consider the numerical approximation of cross-collision operators with general forms of the collision kernel.

In FSM, VDFs are periodized on the velocity domain. For simplicity, we assume the velocity domains of fA and fB are the same, say, DL=[L,L]3. The velocity bound L should be chosen as a compromise between the resolution

Accuracy analysis of the fast spectral method

We assess the accuracy of our newly-developed FSM by comparing its numerical results with the analytical Bobylev–Krook–Wu (BKW) solutions for the space-homogeneous relaxation of pseudo-Maxwell molecules when the collision kernel is Cıȷ=C0ıȷsin1(θ/2). For simplicity, we take kB=1, and set the equilibrium temperature to be 2. The exact BKW solutions can be written as follows [59], [60]:fBKWı(v,t)=nı(mı2πK)3/2exp(mı|v|22K)(13rpı+rpıKmı|v|2)(ı=A,B), wherepA=4nB5[2C0BBm0C0AB(53m0)],pB=4nA5[2C0

An efficient algorithm for large mass ratios

In Section 4 we used the same velocity discretization for both VDFs. This is acceptable for mass ratios less than 2, but it becomes computationally inefficient for large mass ratios, as the computer memory usage of FSM is proportional to mr3/2, while the computational cost is proportional to mr2. Here we propose a more efficient algorithm to deal with large-mass-ratio problems.

Before starting, we demonstrate how the cross-collision operators are calculated in Section 4 when using the same

Numerical results for spatially-inhomogeneous problems

It is convenient to use dimensionless variables. Here, we normalize the spatial location x by the characteristic flow length , the molecular velocity v by the thermal speed of the A-component vmA=2kBT0/mA at the reference temperature T0, the density n by the reference molecular number density n0, and the VDF by n0/(vmA)3. Then the BE (1) is recast to a dimensionless form, where the collision operator (2) with the collision kernel (6) is given byQıȷ(fı,fȷ)=1Kn1ıȷR3S2sinαıȷ+γıȷ1(θ2)cosγıȷ(θ

Conclusions

We have developed a fast spectral method to solve the Boltzmann equation for binary mixtures of monatomic gases, with general forms of the collision kernel. The method employs Fourier–Galerkin discretization in velocity space, and handles binary collisions in the corresponding frequency space. The method has spectral accuracy, and can deal with mixtures with large molecular mass ratios at a computational cost of O(mrM2N4logN), which is much smaller than the conventional spectral method's cost

Acknowledgements

The authors thank the reviewers of this paper for their valuable suggestions. This work is financially supported by the UK's Engineering and Physical Sciences Research Council (EPSRC) under grants EP/I036117/1 and EP/I011927/1.

References (61)

  • S.Y. Docherty et al.

    Multiscale simulation of heat transfer in a rarefied gas

    Int. J. Heat Fluid Flow

    (2014)
  • S. Tiwari et al.

    A particle–particle hybrid method for kinetic and continuum equations

    J. Comput. Phys.

    (2009)
  • M.H. Gorji et al.

    Fokker–Planck–DSMC algorithm for simulations of rarefied gas flows

    J. Comput. Phys.

    (2015)
  • M.K. Borg et al.

    A multiscale method for micro/nano flows of high aspect ratio

    J. Comput. Phys.

    (2013)
  • A. Patronis et al.

    Hybrid continuum-molecular modelling of multiscale internal gas flows

    J. Comput. Phys.

    (2013)
  • A. Patronis et al.

    Multiscale simulation of non-isothermal microchannel gas flows

    J. Comput. Phys.

    (2014)
  • I.M. Gamba et al.

    Spectral-Lagrangian methods for collisional models of non-equilibrium statistical states

    J. Comput. Phys.

    (2009)
  • A. Alekseenko et al.

    Deterministic solution of the spatially homogeneous Boltzmann equation using discontinuous Galerkin discretizations in the velocity space

    J. Comput. Phys.

    (2014)
  • G.P. Ghiroldi et al.

    A direct method for the Boltzmann equation based on a pseudo-spectral velocity space discretization

    J. Comput. Phys.

    (2014)
  • L. Wu et al.

    Deterministic numerical solutions of the Boltzmann equation using the fast spectral method

    J. Comput. Phys.

    (2013)
  • V.I. Kolobov et al.

    Unified solver for rarefied and continuum flows with adaptive mesh and algorithm refinement

    J. Comput. Phys.

    (2007)
  • S. Kosuge et al.

    Shock-wave structure for a binary gas mixture: finite-difference analysis of the Boltzmann equation for hard-sphere molecules

    Eur. J. Mech. B, Fluids

    (2001)
  • R.D.M. Garcia et al.

    The viscous-slip, diffusion-slip, and thermal-creep problems for a binary mixture of rigid spheres described by the linearized Boltzmann equation

    Eur. J. Mech. B, Fluids

    (2007)
  • R.D.M. Garcia et al.

    Couette flow of a binary mixture of rigid-sphere gases described by the linearized Boltzmann equation

    Eur. J. Mech. B, Fluids

    (2008)
  • R.D.M. Garcia et al.

    The linearized Boltzmann equation with Cercignani–Lampis boundary conditions: basic flow problems in a plane channel

    Eur. J. Mech. B, Fluids

    (2009)
  • Y. Sone

    Flows induced by temperature fields in a rarefied gas and their ghost effect on the behavior of a gas in the continuum limit

    Annu. Rev. Fluid Mech.

    (2000)
  • Y. Sone

    Kinetic Theory and Fluid Dynamics

    (2002)
  • S. Chapman et al.

    The Mathematical Theory of Non-uniform Gases

    (1970)
  • H. Grad

    On the kinetic theory of rarefied gases

    Commun. Pure Appl. Math.

    (1949)
  • H. Struchtrup

    Regularization of Grad's 13 moment equations: derivation and linear analysis

    Phys. Fluids

    (2003)
  • Cited by (64)

    • Hermite spectral method for multi-species Boltzmann equation

      2022, Journal of Computational Physics
      Citation Excerpt :

      For now, we have introduced the properties of the multi-species Boltzmann equation. Due to its high dimensionality and the complex collision terms, several numerical methods have been proposed, such as the stochastic DSMC method [2], the discrete velocity method [14,34], and the Fourier spectral method [22,38]. So far, the Hermite spectral method has been successfully utilized to solve the Boltzmann equation with single species and the plasma with collisions [21,27].

    • Projective and telescopic projective integration for non-linear kinetic mixtures

      2022, Journal of Computational Physics
      Citation Excerpt :

      It is of great interest to attempt the numerical solution of problems with extremely large mass ratios. Indeed, a simple mixture of Ar and He exhibits a mass ratio of 10, which can be a problem for some numerical methods, as noted in [61]. Their method can deal with mass ratios up to 35; however, ratios twice as large can easily be found in other scenarios, such as mixtures of H2 and Xe.

    View all citing articles on Scopus
    View full text