This article was downloaded by: [Erciyes University]

On: 10 January 2015, At: 13:34

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Journal of Computer

Mathematics
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/gcom20

Method of calculating the collision

Click for updates
integral and solution of the Boltzmann
kinetic equation for simple gases, gas
mixtures and gases with rotational
degrees of freedom
ab ab ab
Yu.A. Anikin , O.I. Dodulad , Yu.Yu. Kloss & F.G.
ac
Tcheremissine
a
Moscow Institute of Physics and Technology, Dolgoprudny, Russia
b
National Research Centre ‘Kurchatov Institute’, Moscow, Russia
c
Dorodnicyn Computing Centre of RAS, Moscow, Russia
Accepted author version posted online: 14 Apr 2014.Published
online: 28 May 2014.

To cite this article: Yu.A. Anikin, O.I. Dodulad, Yu.Yu. Kloss & F.G. Tcheremissine (2014): Method of
calculating the collision integral and solution of the Boltzmann kinetic equation for simple gases,
gas mixtures and gases with rotational degrees of freedom, International Journal of Computer
Mathematics, DOI: 10.1080/00207160.2014.909033

To link to this article: http://dx.doi.org/10.1080/00207160.2014.909033

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.
 This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [Erciyes University] at 13:34 10 January 2015
 International Journal of Computer Mathematics, 2014
http://dx.doi.org/10.1080/00207160.2014.909033

Method of calculating the collision integral and solution of the

Boltzmann kinetic equation for simple gases, gas mixtures and
gases with rotational degrees of freedom
Yu.A. Anikina,b , O.I. Dodulada,b∗ , Yu.Yu. Klossa,b and F.G. Tcheremissinea,c
Downloaded by [Erciyes University] at 13:34 10 January 2015

a Moscow Institute of Physics and Technology, Dolgoprudny, Russia; b National Research Centre
‘Kurchatov Institute’, Moscow, Russia; c Dorodnicyn Computing Centre of RAS, Moscow, Russia

(Received 16 August 2013; revised version received 10 November 2013; Second revision received 11 March 2014;
accepted 23 March 2014)

The conservative method of calculating the Boltzmann collision integral for simple gases, gas mixtures and
gases with rotational degrees of freedom of molecules is presented. In all cases the common approach based
on the projection technique of summing up the contributions in the collision integral is used. The method
is applied for solving the Boltzmann kinetic equation for two fundamental rarefied gas flow problems:
the heat transfer problem and the problem of shock wave structure. A comparison with experimental and
numerical data of other authors is reported. It is shown that the considered numerical method allows one
to solve the Boltzmann equation for real gases with high accuracy.

Keywords: Boltzmann equation; conservative method; gas mixture; rotational degrees of freedom; shock
wave; heat transfer

2010 AMS Subject Classifications: 65R20; 65Z05; 65C05

1. Introduction

Simulation of gas flows in microdevices and vacuum equipment, calculation of flows over aircraft
in the upper atmosphere, analysis of processes incide the shock waves, and solution of many
other problems, where the characteristic flow size is comparable with the mean free path of
gas molecules, require the application of kinetic theory methods and solution of the Boltzmann
equation.
The numerical solution of this equation is a complicated task due to its high dimensionality and
difficulties in calculating the multidimensional nonlinear collision integral. Today, the following
methods of the numerical solution of the Boltzmann equation or its replacement by more simple
approximate equations are known:
• Solution of model equations [8] where the complex collision integral is replaced with simple
relaxation forms. In this case, reliability of obtained results remains somewhat doubtful.
• Statistical simulation methods [9]. They are characterized by significant statistical noise that
makes it difficult to obtain highly accurate solutions.
• The numerical analysis method based on expansion of the distribution function in polynomials
and calculation of the collision integral using this approximation. The method was tested on

∗ Corresponding author. Email: dodulad@list.ru

© 2014 Taylor & Francis

 2 Yu.A. Anikin et al.

2 one-dimensional problems: shock wave structure problem [21,25] and heat transfer problem
[26]. The method is not conservative and is very intensive for calculations. Examples of solutions
of 2D or 3D problems are absent.
• Discrete-velocity methods that consider only those molecular collisions for which initial and
final velocities belong to the velocity grid nodes. These methods give solutions not affected by
statistical noise. It has been proved [27] that the solution converges to the solution of the exact
Boltzmann equation at higher discretization of the velocity space. However, the convergence
is slow, which calls for significant computing resources to obtain reliable results.
• Spectral methods. The basic approaches are described in [18,24]. An attempt to address the
spectral methods conservatism issue is presented in [14]. Earlier works consider only spatially
uniform problems, but in recent years the spectral method have been also applied to some
one-dimensional in physical space problems [15].
• Projection method [10]. This is a discrete ordinate method the key feature of which is the
Downloaded by [Erciyes University] at 13:34 10 January 2015

conservative projection method of computing the collision integral, which provides for strict
fulfillment of conservation laws, non-negative solution, and zero value of the collision integral
at the Maxwellian distribution function. The method has been used successfully for the cal-
culation of supersonic flows [31] and slow flows slightly deviating from the thermodynamic
equilibrium state [11].
The present work is devoted to the conservative projection method of calculating the Boltzmann
collision integral. The method is described in the following section. It is presented as applied to
a single-component monoatomic gas and then generalized for a gas mixture. A modification of
the conservative projection method, the multipoint projection method [12], which is efficient
for gas mixtures with considerable difference of molecular masses is described as well. For
diatomic molecules, the method of solving the generalized Boltzmann equation, which takes into
account the exchange of translational and rotational energy of molecules, is described. The method
uses pre-calculated differential cross-sections of molecular collisions with rotational degees of
freedom that make the calculations more realistic compared to those of [32], where a two-stage
model of inelastic collision and a probabilistic model of energy transitions between rotational and
translational modes [6] were applied.
The description of the projection method has been done earlier and its realiability has been
shown on a variety of rarefied gas flow problems, including 2D and 3D flows [2,3,13,19]. The
aim of this paper is to provide a description of the approach in a common way for all types of
gases and to demonstrate the accuracy of calculations. Two fundamental problems of rarefied gas
dynamics are analysed: heat transfer and shock wave structure.
The projection method is inplementated in a computer code which substitutes a problem-solving
environment (PSE) designed to simulate 1D, 2D and 3D problems. The struture of PSE is described
in Section 3.
Particular attention in the work is given to the application of realistic molecule interaction
potentials, which allowed to make a comparison between numerical results and experimental
data. This issue is discussed in Section 4.

2. Conservative projection method of calculating the collision integral

2.1 Projection method for a simple gas

The collision integral for a simple gas is written in the following form:

  
I(ξ) = f (ξ 1 )f (ξ  ) − f (ξ 1 )f (ξ) g σ (θ, g) d 2 n d 3 ξ 1 .
 International Journal of Computer Mathematics 3

Here, ξ, ξ 1 are pre-collision velocities of molecules, g = |g|, g = ξ 1 − ξ is the relative velocity

vector, ξ  = 0.5g − 0.5gn, ξ 1 = 0.5g + 0.5gn are post-collision velocities of the molecules, n
is the unit vector of the direction of scattering, σ (θ , g) is the scattering cross-section which, in
general case, depends on the normal angle cos θ = (n,g)/g and of the absolute value of relative
velocity g . For the hard-sphere model, the cross-section is a constant σ (θ, g) = 4π d 2 , where d is
the hard sphere diameter. For complex interaction potentials, the values σ (θ, g) are precomputed
and stored in an array. Below, to simulate realistic interaction of gas molecules the Lennard-Jones
potential is considered. More complex interactions, for example, ab initio potentilas used in [30]
can be applied as well.
When the Boltzmann equation is solved numerically, there is selected in velocity space domain
 of volume V where a grid of equally spaced velocity nodes  = {ξ γ }, γ = 1, . . . , N is built.
On the generated grid, the distribution function and collision integral are represented in the basis
of δ-functions in the following form:
Downloaded by [Erciyes University] at 13:34 10 January 2015

V  V 
N N
f (ξ) = fγ δ(ξ − ξ γ ), I(ξ) = Iγ δ(ξ − ξ γ ).
N γ =1 N γ =1

The symmetry property allows rearranging of the collision integral into the following form:

1  
Iγ = δ(ξ − ξ γ ) + δ(ξ 1 − ξ γ ) − δ(ξ  − ξ γ ) − δ(ξ 1 − ξ γ )
4
 

× f (ξ 1 )f (ξ ) − f (ξ 1 )f (ξ) gσ (θ, g)d 2 nd 3 ξ d 3 ξ 1 .

For the calculation of the symmetrized collision integral, a uniform eight-dimensional cubature
Korobov grid [20] in space {ξ αν ξ βν nν }, ν = 1 . . . Nν is used, whose nodes ξ αν ξ βν coincide with
the nodes of the velocity grid.
Since post-collision velocities ξ  , ξ 1 fall not in the grid nodes, the last two δ-functions in the
projector are replaced with a combination of projectors into two pairs of nodes (ξ λν , ξ λν +sν ) and
(ξ μν , ξ μν −sν ) nearest to ξ  and ξ 1 , respectively. Here, sν is a 3D vector of displacement to the
neighbouring grid node, each component of which can take values 0, 1, −1.
δ(ξ  − ξ γ ) = (1 − rν )δ(ξ λν − ξ γ ) + rν δ(ξ λν +sν − ξ γ ),
(1)
δ(ξ 1 − ξ γ ) = (1 − rν )δ(ξ μν − ξ γ ) + rν δ(ξ μν −sν − ξ γ ).
Such a replacement means that contributions to the collision integral in non-nodal points ξ 
and ξ 1 are distributed among pairs of the nearest grid nodes. The coefficient rν is found from the
energy conservation law
ξα2 + ξβ2 = (1 − r)(ξλ2 + ξμ2 ) + r(ξλ+s
2
+ ξμ−s
2
). (2)
The law of conservation of momentum is automatically fulfilled for the uniform grid by the
symmetric choice of projection nodes as described above.
The collision integral calculation formula takes the following form:

Nν
 
Iγ = Bν −(δγ ,αν + δγ ,βν ) + (1 − rν )(δγ ,λν + δγ ,μν ) + rν (δγ ,λν +s + δγ ,μν −s )
ν=1
 
× (f (ξ 1 )f (ξ ) − fαν fβν ) ξ αν − ξ βν  ,
 
(3)

1, α = β
where δα,β = is the Kronecker symbol, and fαν = f (ξ αν ), fβν = f (ξ βν ) are values
0, α  = β
of the distribution function in the grid nodes.
 4 Yu.A. Anikin et al.

 
Various interpolation types can be used to calculate f (ξ 1 ) f (ξ ), but the preferred one is the
polynomial interpolation
 
f (ξ 1 ) f (ξ ) = (fλν fμν )1−rν (fλν +sν fμν −sν )rν , (4)

which provides for zero value of the collison integral of the Maxwellian distribution function and
fulfillment of the Boltzmann’s H-theorem for the discretized equation [12].
At a time step τ , the Boltzmann equation is solved using the symmetric scheme of splitting
into the advection operator Tt/2 and collisional relaxation operator Ct
  
ft = Tt/2 Ct Tt/2 (f0 ) . (5)

The relaxation equation is integrated using a scheme that has correct asymptotics at t→ ∞.
Downloaded by [Erciyes University] at 13:34 10 January 2015

Write the equation in the integral form, having introduced intermediate time values tν = t0 +
τ ν/N1 and intermediate values of f γ ,ν , where f γ ,ν |ν=0 , . . . , f γ ,ν | ν=N1 are distribution functions at
times t0 ,…, t0 + τ , respectively

 t0 +τ N1 
 t0 +τ ν/N1
f γ ,N1
=f γ ,0
+ I dt = f
γ γ ,0
+ I γ dt.
t0 v=1 t0 +τ (ν−1)/N1

This formula can be rewritten as the ‘continuous computation’ scheme

f αν ,ν = f αν ,ν−1 − τ Bν ν gν , f λν ,ν = f λν ,ν−1 + (1 − rν )τ Bν ν gν , f λν +sν ,ν

= f λν +sν ,ν−1 + rν τ Bν ν gν ,
(6)
f βν ,ν = f βν ,ν−1 − τ Bν ν gν , f μν ,ν = f μν ,ν−1 + (1 − rν )τ Bν ν gν , f μν −sν ,ν
= f μν −sν ,ν−1 + rν τ Bν ν gν ,

where ν = (fλν fμν )1−rν (fλν +sν fμν −sν )rν − fαν fβν , Bν = π σ (θν gν )VN/N1 .
If the molecular interaction potential differs from the hard-sphere potential, collision cross-
sections σ (θ, g) are calculated in advance by the trajectory method and stored in the form of
grid dependence σ (θi , gj ). When the Boltzmann equation is being solved, the cross-section for
specified θ, g is restored by the interpolation of the grid dependence values σ (θi , gj ).
The grid dependence σ (θi , gj ) is obtained by simulation of a large number of scatter trajectories
for two interacting centres. To obtain the dependence for θ , the
impact parameter b is varied.
Suppose that a homogeneous stream of particles J = Ntraj π bmax 2
hits a scattering centre. Let
Ntraj be the number of trajectories participating in computations and bmax be the radius of truncation
of the interaction potential, which depends on the relative velocity g . The radius of truncation
is defined by the condition that the deviation angle becomes less than the angle resolved by the
velocity grid. By the definition of the differential cross-section, the number of particles scattred
into the solid angle 2π sin θ dθ is equal to dN = 2π σ (θ, g)Jsin θ dθ. This yields

 θi +θ / 2  θi +θ / 2 
1   1 1 dN 2
bmax
Ni
1
σ (θi , gj ) = σ θ , gj dθ = dθ = ,
θ θi −θ / 2 2πJθ θi −θ / 2 sin θ dθ 2Ntraj θ ν=i sin θν
(7)

where N

i is the number of all trajectories for which the deflection angle belongs to the interval
θi ± θ 2.
 International Journal of Computer Mathematics 5

The analysis of the process of trajectory calculation has showed that a higher accuracy, espe-
cially at small deviation angles, is achieved by means of direct numerical integration of the
equation of motion
  
m d2 r dU r − bmax
2 − b2 dr g
= −∇U(r) = − , r|t=0 = , |t=0 =
2 dt 2 dr r b dt 0

where U(r) is the centre–centre interaction potential. Calculation

of the trajectory stops when r

becomes larger than bmax . The direction of the velocity vector dr dt determines the deflection
angle.

2.2 Projection method for a gas mixture

Downloaded by [Erciyes University] at 13:34 10 January 2015

For a gas mixture, the Boltzmann equation takes the form

∂fi p      p 
p1 
+ · ∇fi =
 

fj (p1 )fi (p ) − fj (p1 )fi (p)  −  σ (θ, g) d 2 nd 3 p1 ,
∂t mi j
mi mj

where fi is the distribution function by momentum for molecules of type i, i = 1, . . . , N.

The projection method described above can also be used for calculation of the collision integral
for a gas mixture. The only requirement that shall be met for the sake of conservatism is the
change from the velocity distribution function to the momentum distribution function, and the
use of the same grid step in the momentum space pi = p, i = 1 . . . N for each component of
the gas mixture.
As in the case of a simple gas, each contribution into momentum after collisions is projected at
two nodes in the grid, which makes the method conservative in terms of energy. Conservativeness
in terms of momentum is achieved by the symmetric choice of the projection nodes in the grid.
The method works well at medium ratios of molecular masses of the mixture components (see
[3,29]), however, the equality of steps in momentum grids leads to the increased number of nodes in
the extra-fine grids for heavy components of the mixture. The number of nodes increases according
to relation Ni ∼ (mi /mmin )3/2 . As an example, calculation for the helium–argon mixture with the
mass ratio mHe /mAr ≈ 1/10 requires a 30-times increase of the grid nodes, which dramatically
rises the intensity of calculations.
To remove this shortcoming of the method, let us consider momentum grids with uniform steps
pi different for each component, and perform projection of the contributions independently for
each component of the mixture.
After presentation of the distribution function and collision integral on the basis of δ-functions,
reduction of the collision integral to the symmetrical form, and application of the Korobov cubature
grid {pαν pβν nν }, the expression for calculation of the collision integral takes the form


N 
Nν   

Ii (pγ ) = Bν δ(p1 − pγ ) + δ(p − pγ ) − δ(pβν − pγ ) − δ(pαν − pγ )
j=1 ν=1
 
 pαν pβ ν 
 

× (fj (p1 )fi (p ) − fi (pαν )fj (pβν ))  − , (8)
mi mj 

where δ(pαν − pγ ) = δαν ,γ , δ(pβν − pγ ) = δβν ,γ – Kronecker symbols, and fi (pαν ) = fi,αν ,
fi (pβν ) = fi,βν are values of the grid distribution function. As in the case of a simple gas, the
 6 Yu.A. Anikin et al.

Figure 1. Projection stencil.

following expressions have to be determined by means of the projection procedure:

Downloaded by [Erciyes University] at 13:34 10 January 2015

 
δ(pα ν − pγ ), δ(pβ  ν − pγ ), fi (p ), fj (p1 ).

To provide for conservation of mass, three components of momentum and energy, the minimal
number of projection nodes equals five. The projection stencil used 5 = {0, i, j, k, 5} is shown

in Figure 1, where sλν = (pλν − p )/p, sμν = (pμν − p1 )/p, and pλν , pμν are momenta on the

grid nearest to p , p1 . It should be noted that not all of five projection nodes can be chosen from
the vertices of the cube to which momentum pα ν belongs, for otherwise an inconsistent system
of algebraic equations for projection coefficients is obtained. The δ-functions are presented in the
form

δ(p − pγ ) = rλ,a δλν +sa ,γ , (9)
a∈

where sa is the vector of displacement of the ath stencil node, and projection coefficients are

1 2 1  
rλ,0 = 1 − p − q, rλ,5 = − (p − q), rλ,i = |sλx | + rλ,5 , rλ,j = sλy  + rλ,5 , rλ,k = |sλx | + rλ,5 ,
3 3 6
 
 
p = |sλx | + sλy + |sλz | , q = sλx
2
+ sλy
2
+ sλz
2
.

Direct substitution of these coefficients in the expressions below proves that the conservation
property holds
   
rλ,a = 1, rλ,a pλ+sa = p , rλ,a p2λ+sa = (p )2 .
a∈ a∈ a∈

The stencils can also be built on a larger number of neighbouring nodes, which additionally
conserve the tensor of the flux of momentum and the energy flux vector for each contribution
in collision integral [12]. Similarly, seven-point (cross) stencils were used in papers [7,23], and
applied to interpolate and project contributions on a single velocity grid for a simple gas.

Values of distribution functions fi (p ) and fj (p1 ) are found by interpolation of the nodal distri-
bution function. The interpolation used is similar to the polynomial interpolation (4) and posseses
the same property to annule the collision integral of the Maxwellian function


fi (p1 ) = fi (pλ+sa )rλ,a . (10)
a∈

Thus, the formula for calculating the collision integral takes the form (8), where undefined
expressions are calculated by the formulae (9) and (10).
 International Journal of Computer Mathematics 7

2.3 Projection method for a gas with rotational degrees of freedom

Diatomic gas whose molecules have rotational degrees of freedom is described at the kinetic level
by the generalized Boltzmann equation [6,32]

∂fi   
+ ξ · ∇fi = ∫(wijkl fl (ξ 1 )fk (ξ ) − fj (ξ 1 )fi (ξ))gσijkl (θ , g)d 2 nd 3 ξ 1 . (11)
∂t j,k,l

The equation is written for the distribution function, fi (ξ), of gas molecules at the ith rotational
quantum level. It represents an extension of the Wang Chang–Uhlenbeck equation for the case of
degeneracy of energy levels that take place for rotational degrees of freedom in the absence of a
magnetic field. The factor wijkl = qk ql /qi qj appears here due to the degeneracy of levels, with qi
being the degeneracy number.
Downloaded by [Erciyes University] at 13:34 10 January 2015

When the equation is solved numerically, the rotational quantum levels are replaced with the
nodes of discretization of the angular velocity of rotation. In general case, ‘numerical’ levels
will not coincide with the quantum levels. As demonstrated in [4], this approach introduces an
error of the order o(ω2 ), where ω is the step in the angular velocity grid. The angular velocity
corresponding to the ith level is equal to ωi = ω(i + 0.5), and the ‘degeneracy’ of the numerical
level equals qi = ωi , wijkl = ωk ωl /ωi ωj .
Similar to the cases of a simple gas and a gas mixture, due to the symmetry properties the
collision integral of the generalized Boltzmann equation can be reduced to the following sum:


Nν   

In (ξ γ ) = Bν δn,kν δ(ξ 1 − ξ γ ) + δn,lν δ(ξ − ξ γ ) − δn,iν δ(ξ αν − ξ γ ) − δn,jν δ(ξ βν − ξ γ )
ν=1
   
× fl (ξ 1 )fk (ξ ) − fi (ξ αν )fj (ξ βν ) ξ αν − ξ βν  .


Here, like in Equation (3) δ(ξ αν − ξ γ ) = δαν ,γ and δ(ξ βν − ξ γ ) = δβν ,γ are Kronecker sym-
bols, and fi (ξ αν ) = fi,αν and fj (ξ βν ) = fj,βν are the values of the grid distribution function,
Bν = π σijkl (θν , gν )VNNω4 /N1 . The Kronecker symbols δn,iν , δn,jν , δn,lν and δn,kν appear here because
the sums from Equation (11) are included in the cumulative sum over ν. In this case the integrating
grid is 12-dimensional {ξ αν , ξ βν , nν , iν , jν , kν , lν }.
 
The values of δ(ξ 1 − ξ γ ) and δ(ξ − ξ γ ) are determined from formulae (1), and those of
 
fl (ξ 1 )fk (ξ ) from formula (4). The conservation law similar to Equation (2) for determination
of the coefficient r is in this case is written as

mξα2 + Iωα2 + mξβ2 + Iωβ2 = (1 − r)(mξλ2 + Iωλ2 + mξμ2 + Iωμ2 )

+ r(mξλ+s
2
+ Iωλ+s
2
+ mξμ−s
2
+ Iωμ−s
2
).

The uniform relaxation equation is integrated according to the continuous computation scheme (6).
The calculation of differential collision cross-sections σijkl (θ , g) for a gas with rotational degrees
of freedom involves considerable difficulty. As in the case of cross-sections for a monatomic gas,
these cross-sections are pre-calculated, and the grid values are stored in data arrays. Here, the
scattering cross-sections depend on six variables, and their storage calls for a large memory size,
typically about 100 MB. The cross-sections are calculated by the classical trajectory method.
The simplest model of interaction between diatomic molecules is the rotator model. Each rotator
possesses two centres of mass and two centres of forces whose interaction is described by the
 8 Yu.A. Anikin et al.

Figure 2. Differential cross-section σijkl (θ , g) at i = 10, j = 5, θ = π/2, g = 5, ω = 3 as a function of l and k.

Lennard-Jones potential
 12  6 
σ σ
Uij = 4ε − ,
Downloaded by [Erciyes University] at 13:34 10 January 2015

rij rij
where rij is the separation distance between the chosen pair of interacting centres. The rotators’
moment of inertia is I = mra2 , where ra is the spacing between points in which the masses are
located.
Equations of interaction between two rotators have the form
⎛ ⎞ ⎛ ⎞
r V
⎜ r1 ⎟ ⎜ ω 1 × r1 ⎟
⎜ ⎟ ⎜ ⎟ F = F11 + F12 + F21 + F22 ,
⎜ r2 ⎟ ⎜ ⎟
dX
= G(X), X = ⎜ ⎟ , G = ⎜ω2 × r2 ⎟ , M 1 = r1 × (−F11 + F12 − F21 + F22 ), (12)
⎜V ⎟ ⎜ F/μ ⎟
dt ⎜ ⎟ ⎜ ⎟
⎝ω 1 ⎠ ⎝ M 1 /I ⎠ M 2 = r2 × (F11 + F12 − F21 − F22 ).
ω2 M 2 /I

Here, r is the vector connecting the molecule centres, ±r1,2 are relative mass position vectors, ω1,2
are angular velocities of rotation of molecules, μ is the reduced molecule mass, M 1,2 are momenta
of force, with values having subscripts 1 or 2 corresponding to the first and second molecule,
respectively, and Fij is the force of interaction between the ith centre of the first molecule and the
jth centre of the second molecule (i, j = 1, 2).
Equations (12) are solved by the Dormand–Prince method [28] representing the eighth-order
Runge–Kutta method. The calculations are done in dimensionless variables where σ , ε, m/2 are
taken as the units of length, energy and mass.
By modelling a huge number of trajectories the grid dependence σijkl (θm , gn ) is determined by
the formula similar to Equation (7). Figure 2 shows an example of the cross-section σijkl (θ, g) at
i = 10, j = 5, θ = π/2, g = 5, ω = 3 for molecules of nitrogen (N2 ) having σ = 3.31 A and
ε = 37.3 K

3. Software implementation

On the basis of the conservative projection method the PSE designed for simulation of rarefied
gas flows was developed. Originally the PSE allowed one to carry out calculations for a one-
component gas in computation areas which could be filled with a structured grid [19]. Then,
the PSE was improved, and it became possible to perform calculations of devices with complex
geometries by means of unstructured grids [2]. The part of the PSE, which is responsible for the
calculation of the collision integral, was extended to simulate gas mixture flows [3]. At present,
the integration of the generalized Boltzmann equation into the PSE is under development [4].
 International Journal of Computer Mathematics 9

Figure 3. Scheme of the PSE.

Downloaded by [Erciyes University] at 13:34 10 January 2015

The scheme of data flow from the input parameters provided by the user and the computational
grid generation to the results visualization by means of specialized packages is presented in
Figure 3. In the centre of the picture there is a solver that is a program that performs all necessary
calculations for solving the Boltzmann equation. The calculation is an iterative process of evolution
of the distribution function f (t, x, ξ). The gas macroparameters are obtained by integrating the
distribution function multiplied by the corresponding function of the velocity ξ. The formulae for
the calculation of gas density, velocity and temperature are as follows:

V  1V 
N N
n(t, x) = f (t, x, ξ γ ), v(t, x) = ξ f (t, x, ξ γ ), T (t, x)
N γ =1 n N γ =1 γ

m V 
N
= (ξ − v)2 f (t, x, ξ γ )
3kn N γ =1 γ

The extension of the formulae for gas mixtures and gases with rotational energy levels is evident.
Since the method of solving the Boltzmann equation involves the splitting scheme by physical
processes (5) where the advection of molecules and the collisions are calculated by turns, the
solver itself consists of separate parts responsible for the advection equation (left part of the
Boltzmann equation) and the calculation of the collision integral with solution of the relaxation
problem.

4. Examples of calculations

4.1 Heat transfer problem

The study of distribution of gas density, temperature and heat flux between two parallel plates
having different temperatures is a fundamental problem of rarefied gas dynamics. Numerous
works are devoted to this problem. In [26] the problem was solved for the hard-sphere potential
by computing the collision integral with the use of a polynomial approximation of the distribution
function.
Here, the problem of heat transfer is studied by the projection method described in Section 2.1.
The calculations are performed for a single-component monatomic gas. The cross-sections
σLJ (θ, g) had been calculated in advance. The Lennard-Jones potential was used whose param-
eters for argon are εAr = 124K, σ Ar = 3.418A. The molecular mean free path λ is found from
 10 Yu.A. Anikin et al.
Downloaded by [Erciyes University] at 13:34 10 January 2015

Figure 4. Gas densities between the plates for: (a) Kn = 0.0658, (b) Kn = 0.1395, (c) Kn = 0.1942, (d) Kn = 0.2992,
(e) Kn = 0.7582. Solid line – present results, dashed lined – results from [26], circles – experimental data [33].

Figure 5. Distribution function at x = 0 for Kn = 0.7582.

equality√of viscosities of Lennard-Jones and hard-sphere gases having the same mean free paths
λ = 1/ 2π σ 2 n(2,2) . For the above parameters the omega integral value at T = 300 K is equal
to (2,2) = 1.093 [17].
Heat transfer between plates with temperatures T± = T (1 ± T ), T = 0.14, is considered.
Diffuse-specular reflection boundary conditions with the accommodation coefficient α = 0.826
are set on the plate walls. The statement of the problem corresponds to experiments described in
[33]. The heat transfer is considered for the following Knudsen numbers (Kn = λ/d, where d is
a distance between the plates): Kn = 0.0658, 0.1395, 0.1942, 0.2992, 0.7582.
Results of the calculations are shown in Figure 4. The solid line corresponds to gas density
distribution obtained with the projection method, the dotted line is plotted for results taken from
[26] for the hard-sphere potential, and points show experimental data for argon from [33]. The
comparison shows a good agreement with the experimental data except for density in the vicinity
of the cold plate for Kn = 0.7582 (see Figure 4(e)). It is interesting to note that in this case the
velocity distribution function has a high gradient region on the line ξx = 0 shown in Figure 5. It
is apparent that the distribution is close to that composed by half-Maxwellians for temperatures
T± = T (1 ± T ) that occurs in the free-molecular flow.
Gas densities in Figure 4 are calculated with the following discretization parameters:
√ the number
of grid nodes in the spatial grid nx = 100 and the velocity grid step ξ = 0.4 kT /m. Stabilization
of the heat flux with an accuracy of 0.001% is chosen as a condition of convergence
√ of the iterative
process. The velocity grid nodes lie inside a sphere of the radius ξcut = 6 kT /m.
To determine the accuracy of the obtained results, a series of calculations by varying the grid
system are done. Accuracy of results is estimated by the value of the heat flux. Results that
 International Journal of Computer Mathematics 11

Figure 6. Heat flux versus velocity grid step for: (a) Kn = 0.0658, (b) Kn = 0.7582. Circles – calculated points, dashed
curve – quadratic approximation.
Downloaded by [Erciyes University] at 13:34 10 January 2015

Figure 7. Normalized heat flux q/qqfm versus Knudsen number: solid line – present results (HS), circles – numerical
results from [26], squares – experimental data [33].

Table 1. Normalized heat flux q/qfm for

various Knudsen numbers.

Kn Present (HS) From [26]

0.0658 0.2515 ± 0.0025 0.2538

0.1395 0.4049 ± 0.0040 0.4094
0.1942 0.4797 ± 0.0048 0.4843
0.2994 0.5759 ± 0.0058 0.5807
0.7582 0.7513 ± 0.0075 0.7558

vary are the following. Calculations performed for nx = 50, 100, 200 show heat flux values that
differ each√from other less than by 0.1%. Heat flux value obtained with enlarged velocity grid
ξcut = 7.5 kT /m differs also less than by 0.1%. The main error is introduced by the velocity grid
step. Figure 6 shows by circles the heat flux versus velocity grid step (à) for Kn = 0.0658 and
(b) for Kn = 0.7582. The dashed line corresponds to the quadratic approximation. Heat flux for
ξ = 0.4 differs from the approximated one for ξ → 0 less than by 0.5%. Summing up, one
can conclude that the accuracy of the applied discretization system is about 1%.
Figure 7 shows the normalized heat flux q/qfm , where qfm is the heat flux in the free-molecule
regime, versus Knudsen number. Solid line corresponds to the present results calculated for the
hard sphere gas, circles represent numerical results from [26] and squares show experimental data
from [33]. Present results with estimated errors and the results from [26] are given in Table 1.

4.2 Shock wave structure in inert gases

The determination of the shock wave structure is another classical problem of the kinetic theory.
It was the subject of a number of experimental works and numerical simulations. Sufficiently
reliable experimental data for monatomic gas Ar and diatomic gas N2 are published in [1]. There
are a number of publications based on the use of statistical simulation method direct simulation
Monte-Carlo [9,22].
 12 Yu.A. Anikin et al.

Figure 8. Reciprocal shock wave width δ for M = 3 in depends of: (a) spatial grid step hx , (b) velocity grid step ξ .
Circles – calculated points, dashed curve – quadratic approximation.

Solutions of the Boltzmann equation in the case of the hard-sphere potential for a single-
Downloaded by [Erciyes University] at 13:34 10 January 2015

component gas and a gas mixture are presented in [21,25], respectively, where the collision integral
is calculated using the same method of polynomial approximation of the distribution function as in
[26]. By the projection method for a single-component gas the problem was studied, for example
in [13], where results for a wide range of Mach numbers are given.
To determine grid parameters required for obtaining sufficiently accurate results we performed
a series of calculations in which the spatial grid step hx and the velocity grid step ξ are varied.
Velocity grid nodes are uniformly distributed. Shock wave for Mach number M = 3 in the hard-
sphere gas is considered. Results of calculations are shown in Figure 8. Figure 8(a) depicts
dependence of reciprocal shock wave width on spatial grid step hx , where reciprocal shock wave
width is determined by the formula

λ1 λ1 dn
δ= = .
l n2 − n1 dx max

√ n1 , n2 are gas densities ahead, in unperturbed area, and behind the shock wave front, λ1 =
Here
1/ 2π σ02 n1 is the molecules mean free path before the shock wave, σ0 is the hard sphere molecule
diameter. Locations of the obtained points are well approximated by a quadratic dependence
(dashed line), which agrees with the order of accuracy O(hx2 ) of the numerical scheme applied
for the advection equation. Figure 8(b) presents δ as a function of the velocity grid step. Here, we
again have a quadratic dependence, and the error is O(ξ 2 ). The shock width is computed using
the five-point central scheme for numerical differentiation with the step hdiff = 0.2λ1 .
One can see that the error by spatial grid step is less than 0.75% when the step became less or
equal to hx = 0.2, and the error by velocity grid step is less than 0.75% when ξ = 0.5. The total
error of the shock wave width is estimated as 1.5%. These discretization parameters are chosen
as basic grid steps in further calculations.
In the present work, the main part of the calculations is done for the realistic Lennard-Jones
potential. Figure 9 shows gas density profiles for the shock waves with Mach numbers M = 1.55
and M = 2.31. The points correspond to the experimental data [1], and the solid line is plotted
from calculations. Obtained reciprocal shock wave widths are equal to δ = 0.150 ± 0.002, δ =
0.247 ± 0.004, respectively. Parameters of the Lennard-Jones potential for argon are given in the
previous section.
Figures 10 and 11 present results of calculations for the gas mixture. Figure 10 compares our
data obtained with the multipoint modification of the projection method (see Section 2.2) with the
calculations presented in [21]. The hard-sphere potential was used. The case for mα /mβ = 1/2,
χ β = 0.1 is considered, where χ β is a fraction of the β-component.
A comparison of computations with experimental data of [16] is presented in Figure 11,
where the case of a mixture of helium with low molar fraction of xenon is considered. The
 International Journal of Computer Mathematics 13

Figure 9. Density in the shock wave in the monatomic gas, solid line – numerical results, ◦ – experimental data [1].
Downloaded by [Erciyes University] at 13:34 10 January 2015

Figure 10. Structure of the shock wave in gas mixture for M = 3, mα /mβ = 1/2, χ β = 0.1 in comparison with data
(◦) of [21].

Figure 11. Density in the shock wave for M = 3.89 in a mixture of 97% helium and 3% xenon, (◦) – experimental data
[16].

calculation was done for the following parameters of the Lennard-Jones potential: ε He = 10.2K,
εXe = 229K, σ He = 2.576A, σ Xe = 4.055A [17]. Parameters of interaction potential√between
different molecules were obtained on the basis of combination relations: ε He,Xe = ε Xe ε He ,
σ He,Xe = (σ He + σ Xe )/2. Values on the axis x are normalized by the mean free path before
the shock wave, which is defined by equality of the mixture viscosity to the viscosity of the
hard-sphere gas.

4.3 Shock wave structure in a diatomic ga

Figure 12 shows results of calculations for the diatomic gas performed through the solution of
the generalized Boltzmann equation by the method described in Section 2.3. The cross-sections
σijkl (θ, g) were obtained with the classical trajectory method. A comparison with experimental
data for nitrogen from [1] is presented. Reciprocal shock wave widths are δ = 0.145 ± 0.002,
and δ = 0.295 ± 0.004 for M = 1.53 and M = 2.4, respectively.
The comparisons show that for the monatomic gas, the application of the Lennard-Jones poten-
tial gives results close to the experimental data. For the gas mixture, a good agreement is observed
 14 Yu.A. Anikin et al.

Figure 12. Density in the shock wave in the diatomic gas compared with experimental data (◦) of [1].

in the density profile width (see Figure 11). The peak in the helium profile may be associated with
the experimental uncertainty. For the diatomic gas, the agreement between the data is slightly
Downloaded by [Erciyes University] at 13:34 10 January 2015

worse. Probably, the model of interaction between diatomic molecules should be improved by
introducing a dipole–dipole interaction and moving slightly apart the repulsion and attraction
centres [5].

5. Conclusion

The conservative method of calculating the collision integral for simple gases, gas mixtures and
gases with rotational degrees of freedom of molecules is presented. For discrete approximation of
the Boltzmann equation in velocity or momentum space, it uses the common approach based on
the projection method of summing up the contributions in the collision integral. For inert gases, the
method allows one to calculate the collision integral for arbitrary molecular potentials. Flows of
gas mixtures can be effciently calculated for any molecular mass ratios and any concentrations of
the mixture components. For diatomic gases, the methodology of pre-calculating the molecular
collision cross-sections with translational-rotational energy transitions was developed, which
allows one to calculate the collision integral without the use of approximate models of inelastic
collisons. The algorithm of collision integral calculation was optimized, which provided for an
increase in the computation speed. Some examples of calculations presented in the paper confirm
reliability and accuracy of the obtained results. Thus, the efficiency of the method for solving the
Boltzmann equation for a range of rarefied gas dynamic problems is demonstrated.

References

[1] H. Alsmeyer, Density profiles in argon and nitrogen shock waves measured by the absorption of an electron beam,
J. Fluid. Mech. 74 (1976), pp. 497–513.
[2] Yu.A. Anikin, E.P. Derbakova, O.I. Dodulad, Yu.Yu. Kloss, D.V. Martynov, O.A. Rogozin, P.V. Shuvalov, and F.G.
Tcheremissine, Computing of gas flows in micro- and nano-scale channels on the base of the Boltzmann Kinetic
equation, Proc. Comput. Sci. 1(1) (2010), pp. 735–744.
[3] Yu. A. Anikin, O.I. Dodulad, Yu.Yu. Kloss, D.V. Martynov, P.V. Shuvalov, and F.G. Tcheremissine, Development of
applied software for analysis of gas flows in vacuum devices, Vacuum 86(11) (2012), pp. 1770–1777.
[4] Yu.A. Anikin and O.I. Dodulad, Solution of a kinetic equation for diatomic gas with the use of differential scattering
cross sections computed by the method of classical trajectories, Comput. Math. Math. Phys. 53(7) (2013), pp. 1193–
1212.
[5] R.M. Berns and A. van der Avoird, N2-N2 interaction potential from ab initio calculations, with application to the
structure of (N2), J. Chem. Phys. 72 (1980), pp. 6107–6115.
[6] A.E. Beylich, An interlaced system for nitrogen gas, Technisch Hochcshule Report, Aachen, 2000.
[7] A. Beylich, Solving the kinetic equation for all Knudsen numbers, Phys. Fluids 12(2) (2000), pp. 444–465.
[8] P.L. Bhatnagar, E.P. Gross, and M.A. Krook, Model for collision processes in gases. I. small amplitude processes in
charged and neutral one-component systems, Phys. Rev. Online Arch. 94(3) (1954), pp. 511–525.
 International Journal of Computer Mathematics 15

[9] G.A. Bird, Molecular Gas Dynamics and the Direct Simulation of Gas Flows, 2nd ed., Oxford University Press,
New York, 1994.
[10] F.G. Cheremisin, Conservative method of calculating the Boltzmann collision integral, Dokl. Phys. 42 (1997),
pp. 607–610.
[11] F.G. Cheremisin, Solving the Boltzmann equation in the case of passing to the hydrodynamic flow regime, Dokl.
Phys. 45(8) (2000), pp. 401–404.
[12] O.I. Dodulad and F.G. Tcheremissine, Multipoint conservative projection method for computing the Boltzmann
collision integral for gas mixtures, 28th International Symposium on Rarefied Gas Dynamics, AIP Conf. Proc., Vol.
1501, 2012, pp. 302–309.
[13] O.I. Dodulad and F.G. Tcheremissine, Computation of a shock wave structure in monatomic gas with accuracy
control, Zh. Vychisl. Mat. Mat. Fiz. 53(6) (2013), pp. 1008–1026.
[14] I.M. Gamba and Sri H. Tharkabhushanam, Spectral-Lagrangian methods for collisional models of non-equilibrium
statistical states, J. Comput. Phys. 228(6) (2009), 2012–2036.
[15] I.M. Gamba and S.H. Tharkabhushanam, Shock and boundary structure formation by spectral-Lagrangian methods
for the inhomogeneous Boltzmann transport equation, J. Comput. Math. 28(4) (2010), pp. 430–460.
[16] A.S. Gmurczyk, M. Tarczynski, and Z.A. Walenta, Shock wave structure in the binary mixtures of gases with disparate
molecular masses, 11th International Symposium on Rarefied Gas Dynamics, Vol. 1, 1978, pp. 333–341.
Downloaded by [Erciyes University] at 13:34 10 January 2015

[17] J.O. Hirschfelder, C.F. Curtiss, and R.B. Bird, Molecular Theory of Gases and Liquids, University of Wisconsin,
Wiley, 1964.
[18] I. Ibragimov and S. Rjasanow, Numerical solution of the Boltzmann equation on the uniform grid, Computing 69(2)
(2002), pp. 163–186.
[19] N.I. Khokhlov, Yu.Yu. Kloss, B.A. Shurygin, and F.G. Tcheremissine, Simulation of the temperature driven micro
pump by solving the Boltzmann kinetic equation, Proceedings of the 26th International Symposium on Rarefied Gas
Dynamics. AIP Conf. Proc., Vol. 1084, 2008, pp. 1039–1044.
[20] N.M. Korobov, Approximate evaluation of multiple integrals, Dokl. Akad. Nauk SSSR. 124 (1959), pp. 1207–1210.
[21] S. Kosuge, K. Aoki, and S. Takata, Shock-wave structure for a binary gas mixture: finite-difference analysis of the
Boltzmann equation for hard-sphere molecules, Eur. J. Mech. B/Fluids 20(1) (2001), pp. 87–126.
[22] K. Koura, Monte Carlo direct simulation of rotational relaxation of diatomic molecules using classical trajectory
calculations: Nitrogen shock wave, Phys. Fluids 9(11) (1997), pp. 3542–3549.
[23] A.B. Morris, P.L. Varghese, and D.B. Goldstein, Monte Carlo solution of the Boltzmann equation via a discrete
velocity model, J. Comput. Phys. 230 (2011), pp. 1265–1280.
[24] C. Mouhot and L. Pareschi, Fast algorithms for computing the Boltzmann collision operator, Math. Comput. 75
(2006), pp. 1833–1852.
[25] T. Ohwada, Structure of normal shock waves: Direct numerical analysis of the Boltzmann equation for hard-sphere
molecules, Phys. Fluids, A Fluid Dyn. 5(1) (1993), pp. 217–234.
[26] T. Ohwada, Heat flow and temperature and density distributions in a rarefied gas between parallel plates with
different temperatures. Finite-difference analysis of the nonlinear Boltzmann equation for hard-sphere molecules,
Phys. Fluids 8 (1996), pp. 2153–2160.
[27] A. Palczewski, J. Schneider, and A.V. Bobylev, Consistency result for a discrete-velocity model of the Boltzmann
equation. l, SIAM J. Numer. Anal. 34(5) (1997), pp. 1865–1883.
[28] P.J. Prince and J.R. Dormand, High order embedded Runge-Kutta formulae, J. Comput. Appl. Math. 67(7) (1981),
pp. 67–75.
[29] A.A. Raines, Numerical solution of one-dimensional problems in binary gas mixture on the basis of the Boltzmann
Equation, AIP Conf. Proc. 663(1) (2003), pp. 67–76.
[30] F. Sharipov and J.L. Strapasson, Ab initio simulation of transport phenomena in rarefied gases, Phys. Rev. E 86
(2012), p. 031130.
[31] F.G. Tcheremissine, Solution to the Boltzmann kinetic equation for high-speed flows, Comput. Math. Math. Phys.
46(2) (2006), pp. 315–329.
[32] F.G. Tcheremissine, Method for solving the Boltzmann kinetic equation for polyatomic gases, Comput. Math. Math.
Phys. 52(2) (2012), pp. 252–268.
[33] W.P. Teagan and G.S. Springer, Heat-transfer and density-distribution measurements between parallel plates in the
transition regime, Phys. Fluids 11(497) (1968), pp. 497–506.