Downloaded from http://rsta.royalsocietypublishing.

org/ on May 26, 2016

Phil. Trans. R. Soc. A (2011) 369, 2228–2236

doi:10.1098/rsta.2011.0059

Higher order slip according to the linearized

Boltzmann equation with general
boundary conditions
BY SILVIA LORENZANI*
Dipartimento di Matematica, Politecnico di Milano, Milano 20133, Italy

In the present paper, we provide an analytical expression for the first- and second-
order velocity slip coefficients by means of a variational technique that applies to the
integrodifferential form of the Boltzmann equation based on the true linearized collision
operator and the Cercignani–Lampis scattering kernel of the gas–surface interaction. The
polynomial form of the Knudsen number obtained for the Poiseuille mass flow rate and the
values of the velocity slip coefficients are analysed in the frame of potential applications
of the lattice Boltzmann methods in simulations of microscale flows.
Keywords: variational method; microscale flows; slip coefficients

1. Introduction

The use of the Lattice Boltzmann (LB) method to describe flows in nano-
and microfluidic devices has been an area of intense research in recent years.
Since, historically, the LB method was derived as a Navier–Stokes solver,
criticism has been raised about the validity of applying the LB method to finite-
Knudsen-number flows. However, it is now generally accepted that the classical
hydrodynamical equations can supply realistic results well beyond the slip region,
provided that higher order boundary conditions are employed. Therefore, one
of the critical issues with the LB method is to specify suitable slip boundary
conditions to ensure microscale simulation accuracy. Very recently, Sbragaglia &
Succi [1] demonstrated how, in the presence of a non-zero slip coefficient, the LB
develops a physical slip flow component at the wall that can be matched exactly
to experimental data up to second order in the Knudsen number, provided the free
parameters appearing in the LB model are tuned in a correct way. Since the LB
equation has its roots in kinetic theory, in the current investigation, a variational
technique has been used to obtain asymptotic near-continuum solutions of the
linearized Boltzmann equation for hard-sphere molecules and general models
of boundary conditions, in order to provide analytical expressions for the first-
and second-order velocity slip coefficients that enable the mesoscopic approach
presented in Sbragaglia & Succi [1] to become predictive.

*[email protected].

One contribution of 26 to a Theme Issue ‘Discrete simulation of fluid dynamics: methods’.

2228 This journal is © 2011 The Royal Society

 Downloaded from http://rsta.royalsocietypublishing.org/ on May 26, 2016

Higher order slip 2229

2. Mathematical formulation

Let us consider two plates separated by a distance d and a gas flowing parallel to
them, in the z-direction, owing to a pressure gradient. Both boundaries, fixed at
x = ±d/2, are held at a constant temperature T0 . For small values of the pressure
gradient, it can be assumed that the velocity distribution of the flow is nearly
the same as that occurring in an equilibrium state. Under these conditions, the
Boltzmann equation can be linearized [2],
vh
kcz + cx = Lh, (2.1)
vx
where h(x, c) is the small perturbation on the basic equilibrium state, c is the
molecular velocity expressed in units of (2RT )1/2 (with R being the gas constant
and T being the temperature), k = (1/p)(vp/vz) = (1/r)(vr/vz) (with p and r
being the gas pressure and density, respectively) and Lh is the linearized collision
operator. In the current investigation, we analyse the Poiseuille problem on
the basis of the linearized collision operator for hard-sphere molecules in order
to obtain a better approximation of real-gas behaviour for isothermal flows.
Appropriate boundary conditions on the two plates must be supplied for the
Boltzmann equation (2.1) to be solved. Once the deviation h from the equilibrium
distribution is known, the bulk velocity of the gas can be written as
 +∞  +∞  +∞
q(x) = p− 2
3
e−c cz h(x, c) dc,
2
(2.2)
−∞ −∞ −∞

and the flow rate F (per unit time through unit thickness) as
 d/2
F =r q(x) dx. (2.3)
−d/2

To apply the variational formulation, we shall rewrite equation (2.1) in the

following symbolic form:
(D − L)h = S , (2.4)

where Dh = cx (vh/vx), S = −kcz . The boundary conditions to be matched with

equation (2.4) have the general expression [3]: h + = Kh − , where the explicit
form of the operator K depends on the scattering kernel used. Here, h ± are
the restrictions of the function h, defined on the boundary, to positive (negative)
values of cx . Most of the works in rarefied gas dynamics are based on the use
of the classical Maxwell gas–surface interaction law, characterized by a single
accommodation coefficient a, while in practice, every physical property has its
own accommodation coefficient. In order to include better physics, in the current
investigation, we will focus on the Cercignani–Lampis (CL) scattering kernel [3],
which is based on two different adjustable parameters. In this case, the boundary
conditions can be written as

+
h (−(d/2)sgncx , c) = RCL (−c → −c )h − (−(d/2)sgncx , c ) dc , (2.5)
cx <0

Phil. Trans. R. Soc. A (2011)

 Downloaded from http://rsta.royalsocietypublishing.org/ on May 26, 2016

2230 S. Lorenzani

where
 
 2cx [ct − (1 − at )c t ]2
RCL (c → c) = exp −
pat an (2 − at ) at (2 − at )
   √ 
[cx 2 + (1 − an )c  x 2 ] 2 1 − an 
× exp − Io cx c x , (2.6)
an an

with ct = (cy , cz ) being the two-dimensional vector of the tangential molecular

velocity and Io the modified Bessel function of first kind and zeroth order.
Physically, the parameters at and an appearing in equation (2.6) represent the
accommodation coefficient of the tangential momentum and the accommodation
of the kinetic energy owing to the velocity normal to the bounding walls,
respectively. The CL model recovers, as limiting cases, the specular reflection (for
at = an = 0) and the diffuse re-emission (for at = an = 1). Using the variational
principle described in Cercignani & Lorenzani [2], we introduce the following
functional J of the test function h̃:

J (h̃) = ((h̃, P(D h̃ − Lh̃))) − 2((PS , h̃)) + (h̃ + − K h̃ − , P h̃ − )B , (2.7)

where P is the parity operator in velocity space and ((, )), (, )B denote two scalar
products defined in Cercignani & Lorenzani [2]. The functional J (h̃) attains its
minimum value when h̃ = h(x, c) solves equation (2.4) with appropriate boundary
conditions. If we let h̃ = h, equation (2.7) gives: J (h) = −((PS , h)). Looking at the
definitions (2.2) and (2.3), it can be easily shown that the stationary value of J
is related to a quantity of physical interest, the flow rate of the gas, through
the relation: F = −r/kJ (h). Since the purpose of the present investigation is
to provide an analytical expression for the first- and second-order velocity slip
coefficients, it is sufficient to consider asymptotic results (near-continuum) for
mass flow rates. Therefore, the following simplified test function, rescaled by the
relative pressure gradient k and the length parameter q, has been used to evaluate
equation (2.7):
 
1
h̃(x, c) = 2cz A(x − 2xcx + 2cx ) + 2cz
2 2
B− , (2.8)
2

which is the same trial function that, for the Bhatnagar–Gross–Krook (BGK)
kinetic model, has been proved to produce very accurate results in the near-
continuum flow limit [2]. In equation (2.8), A and B are adjustable constants
to be varied in order to obtain the best value of J (h̃). Substituting h̃, given by
equation (2.8), into equation (2.7), we obtain the following polynomial of the
second order with respect to the constants A and B, which are to be determined:

 
√ −1 c11 2 c22 2 1 c22
J (h̃) = ( p) A + B + c12 AB − c1 A − c2 B + c2 − , (2.9)
2 2 2 4

Phil. Trans. R. Soc. A (2011)

 Downloaded from http://rsta.royalsocietypublishing.org/ on May 26, 2016

Higher order slip 2231

where the same symbol J (h̃) has been used to denote the dimensionless quantity
J (h̃)/(kq)2 . The coefficients in non-dimensional form are given by

4 √ 3 8 d3 ˆ 32 d4 √
c11 = − pd − J 1 + 4d2 − dJˆ 2 − at + pd3 at − 4d2 at
3 3p p 4
√ √ √
+ 12 pdat − 32at + 4 pdan − 16an − 4 pdat an + 16at an

+ 16d2 F0 an (1 − at ) + 16d2 F1 (1 − at )(1 − an ), (2.10)

√
c12 = −4 − d2 at + 2 pdat − 4at + 4an − 4at an , c22 = −4at , (2.11)
√
p 3 √ d2 √ √
c1 = d + 2 pd − at + pdat − 4at and c2 = 2 pd − 2at , (2.12)
6 2

where F0 = 0.196079 and F1 = 0.247679. In equations (2.10)–(2.12), d = d/q is

the rarefaction parameter (inverse Knudsen number) and the symbol Ji stands
for integral expressions defined by using the brackets [f, j],

 +∞  +∞  +∞
dc e−c f(c)Lj,
2
[f, j] = (2.13)
−∞ −∞ −∞

with Lj being the linearized Boltzmann collision operator.

√ For hard spheres
√ q is given by q = 2/(p s n) and the
3/2 2
of diameter s, the length parameter
mean free path l reads as l = 1/( 2ps2 n) (where n is the number density).
Therefore,

 2p p  +∞
1
dc1 e−c1 V (j1 + j − j1 − j),
2
Lj = √ de sin Q dQ · (2.14)
4 2p5/2 l 0 0 −∞

where j is a function of c, while j1 refers to c1 . V is the relative velocity |c − c1 |.

j ≡ j(c ) and j1 ≡ j(c1 ), where c and c1 are the velocities after collision of
two molecules with velocities c and c1 . The collision geometry, in conjunction
with the conservation laws, relates the velocities after collision to the velocities
before collision [2]. In equation (2.14), Q is the angle through which the relative
velocity has turned and e is the azimuthal angle the plane containing the relative
velocities before and after collision makes with a fixed reference plane. The
integrals J1 and J2 appearing in equation (2.10) are eight-fold integrals: J1 =
√
[cx cz , cx cz ] and J2 = −[cx 2 cz , cx 2 cz ], where Jˆ i = (2l/ p)Ji . The accuracy of our
variational analysis solution is mostly related to the precision in the computation
of such integrals, which have been numerically evaluated using a Monte Carlo
integration on an eight-dimensional space (see equations (2.13)–(2.14)) [2]. The
derivatives of J (h̃) with respect to A and B vanish in correspondence with the

Phil. Trans. R. Soc. A (2011)

 Downloaded from http://rsta.royalsocietypublishing.org/ on May 26, 2016

2232 S. Lorenzani

optimal values of these constants. The resulting expression for the minimum
of J (h̃) is
√ −1 2 −1
min J (h̃) = (8 p) · [c11 c22 − c12 ] · [8c12 c1 c2 − 4c12 c22
+ c12
2
· (c22 − 4c2 ) − c11 (c22 − 2c2 )2 ]. (2.15)

Thus, the computation of the optimal value of the functional J (h)

(equation (2.15)) will lead to an accurate estimate of the flow rate of the gas,
which in non-dimensional form is
F 2
Q(d) = = 2 J (h). (2.16)
−(r/2)kd 2 d

3. Near-continuum solution and slip coefficients

A notable advantage of the variational approach is that it permits one to write

down simple approximate equations, computed via kinetic theory, to be used
in classical hydrodynamic numerical tools in order to extend the continuum
description (that is significantly more efficient compared with molecular-based
approaches), even beyond the slip regime. In recent years, there has been
considerable success in the implementation of second-order slip boundary
conditions to extend the Navier–Stokes equations into the transition regime.
Assuming a second-order boundary condition at a flat wall, in the isothermal
case, the slip velocity is [2]
   2 
vq 2 v q
qs = A1 l − Ã2 l , (3.1)
vx w vx 2 w

where the gas-velocity gradients are evaluated

√ at the wall. Here, l is the mean
free path of the molecules, l = (m/P) p/2RT , where m is the gas viscosity,
P is the pressure, T is the absolute temperature and R is the universal gas
constant. In equation (3.1), A1 and Ã2 are the first- and second-order slip
coefficients, respectively. Recent experimental studies [2] have revealed large
discrepancies between the experimentally determined values of the second-order
slip coefficient and the theoretical values listed in the literature, most of which
include empirical parameters suitably a posteriori calibrated and, therefore, not
useful in a predictive sense. The lack of a well-founded value of Ã2 makes it
difficult to extend slip-flow predictions into the transition regime. The asymptotic
near-continuum solution for the Poiseuille mass flux obtained by means of
our variational technique can be used to predict slip coefficients. When the
linearized Boltzmann equation for hard-sphere molecules is considered and the
CL scattering kernel is used to describe the gas–wall interaction, equations (2.15)
and (2.16) give, in the limit d  1,

d s2
Q= + s1 + + ··· (3.2)
s0 d

Phil. Trans. R. Soc. A (2011)

 Downloaded from http://rsta.royalsocietypublishing.org/ on May 26, 2016

Higher order slip 2233

where



√ −1 96 √ √ 16
s0 = (4 p) · Jˆ 1 + 48 p , s1 = [4 pat A] · D − pat C ,
−1
(3.3)
p 9


√ 16 16
s2 = [4 pat A]−1 · E + pat C 2 − pat B − CD , (3.4)
9 9
32 ˆ 16 √ 128 ˆ 32 64
A= J1 + p, D = J 1 − p3/2 at + p3/2 , (3.5)
3p 3 3 3 3


128 ˆ √ √ √
B = A−1 · J 2 − 16 pat − 16 pan + 16 pat an , (3.6)
p
C = A−1 · [−16 − 4pat − 64F0 an (1 − at ) − 64F1 (1 − at )(1 − an )] (3.7)
64
and E = −64p + pat − 256pF0 an (1 − at ) − 256pF1 (1 − at )(1 − an ), (3.8)
3
with Jˆ 1 = −1.4180 and Jˆ 2 = 1.8909. Rewriting equation (3.2) in terms of
deviations from the continuum solution, one obtains



Q 2 4
S BLE
= = 1 + √ (s0 s1 ) Kn + (s0 s2 ) Kn 2 , (3.9)
(d/s0 ) p p
√
where the Knudsen number (Kn) is given by Kn = p/(2d). A comparison with
the solution of the Navier–Stokes equations obtained by using the boundary
condition (3.1),
S NS = 1 + 6A1 Kn + 12Ã2 Kn 2 , (3.10)
gives
s0 s1 s 0 s2
A1 = √ and A2 = Ã2 + O(1) = , (3.11)
3 p 3p
where the term O(1) indicates that a kinetic boundary layer, of the order
of Kn 2 , needs to be superposed to the Navier–Stokes component of the flow
field in order to capture the true solution of the Boltzmann equation [2].
A2 , evaluated starting from mass flow rate computations, can be considered a
sort of ‘effective’ second-order slip coefficient since it includes the contribution of
the Knudsen layer. Once the effective second-order slip coefficients are evaluated
by means of our variational computation of the Poiseuille flow rate, one can
filter out the Knudsen layer effects, following the theoretical considerations
presented below. In the BGK approximation, the Poiseuille flow rate can be split
as follows:  d/2

 d/2 2 
F 2 v q̂
= q(x) dx = q̂(x) + xl dx, (3.12)
r −d/2 −d/2 vx 2
where q̂(x) represents the Navier–Stokes component of the true flow field, while
the second term in the integral (3.12) accounts for the contribution of the Knudsen
layer. According to Cercignani & Lorenzani [2], the Knudsen layer contribution
to the flow rate is largely independent of the molecular model considered and,
therefore, also for a hard-sphere gas, x retains approximately the same value
it keeps for the BGK model equation, x  0.296. Thus, the second-order slip

Phil. Trans. R. Soc. A (2011)

 Downloaded from http://rsta.royalsocietypublishing.org/ on May 26, 2016

2234 S. Lorenzani

coefficient that should be used in the boundary conditions (3.1) is given by

Ã2 = A2 + x. The latter highlights that a successful second-order slip model is
one that does not agree with the Boltzmann equation solutions in the Knudsen
layers, in order to be able to predict within the same framework all the physical
quantities of interest: not only the flow rate but also the stress field, which is not
subject to a Knudsen layer correction.

4. Results and discussion

The first-order (A1 ) and second-order (A2 ) slip coefficients, obtained on the basis
of our variational solution of the linearized Boltzmann equation for hard-sphere
molecules, are given in tables 1 and 2, for several values of the accommodation
coefficients at and an . To test the reliability of our variational approach, we list
in table 1, the values of the first-order slip coefficient A1 computed by using
equation (3.11), along with the highly accurate numerical results reported by
Siewert [4]. We have rescaled the values of the viscous slip coefficient given by
Siewert [4] in order to use the same definitions of mean free path and units as
in the present investigation. As shown in the table, the agreement is fairly good
with the relative error lying within 1.2 per cent. This comparison reveals that
our variational datas are more accurate than those reported in McCormick [5],
where analytical equations for the first-order viscous slip have been obtained
by means of a variational principle applied to the linear Boltzmann equation
for a rigid-sphere gas and the CL scattering kernel. We have also included in
tables 1 and 2 results based on a variational solution of the linearized BGK model
equation in order to provide evidence that the BGK first-order slip coefficients
are similar to those determined by the solution of the Boltzmann equation for
hard-sphere molecules, while the second-order slip coefficients are found to be
significantly dependent on the interaction model. As stated in §2, the Maxwell
gas–surface interaction law is characterized by a single accommodation coefficient
a for every molecular property. Thus, one may expect that assuming the same
accommodation coefficients, the Maxwell kernel and the CL model will provide
results close to each other.
In table 3, a comparison of the slip coefficients, obtained for both kernels under
the condition a = at = an , is performed. One can see that the dependence of the
first-order slip coefficient on the scattering kernel used is weak, while the second-
order slip coefficient depends significantly on the gas–surface interaction model.
In spite of that, whatever model of boundary conditions is chosen, the estimates
we obtain for A2 , in the case of hard-sphere molecules, seem inconsistent with
most of the available theoretical models derived on heuristic or empirical grounds,
while they are very close to the values obtained in recent experiments (see [2] for a
detailed discussion). This means that our analytical expressions for the first- and
second-order velocity slip coefficients can be used to adjust the free parameters
appearing in the mesoscopic approach of the LB boundary conditions proposed
in Sbragaglia & Succi [1], in order to predict experimental data. In recent years,
the LB method has been recognized as one of the most promising approaches
for the simulation of microscale gas flows provided that boundary conditions
derived from continuous kinetic theory are employed. Two basis types of kinetic
boundary condition have been proposed: one is the discrete Maxwellian boundary

Phil. Trans. R. Soc. A (2011)

 Downloaded from http://rsta.royalsocietypublishing.org/ on May 26, 2016

Higher order slip 2235

Table 1. The first-order slip coefficient A1 for the CL boundary conditions. Comparison between our
results (variational method applied to the Boltzmann equation for a hard-sphere gas (var.(HS)))
and Siewert’s results [4] for a rigid-sphere gas. The outputs based on a variational solution of the
BGK model equation are also included (variational method applied to the BGK kinetic model of
the Boltzmann equation (var.(BGK))).

at an = 0 an = 0.25 an = 0.5 an = 0.75 an = 1

0.25 [4] 7.1947 7.1635 7.1385 7.1170 7.0979

var.(HS) 7.1090 7.0839 7.0588 7.0337 7.0086
var.(BGK) 7.2346 7.2100 7.1854 7.1607 7.1361
0.50 [4] 3.1696 3.1498 3.1336 3.1194 3.1067
var.(HS) 3.1502 3.1335 3.1168 3.1000 3.0833
var.(BGK) 3.2020 3.1855 3.1691 3.1527 3.1363
0.75 [4] 1.8069 1.7975 1.7896 1.7826 1.7763
var.(HS) 1.8085 1.8001 1.7917 1.7834 1.7750
var.(BGK) 1.8360 1.8277 1.8195 1.8113 1.8031
1.00 [4] 1.1113 1.1113 1.1113 1.1113 1.1113
var.(HS) 1.1209 1.1209 1.1209 1.1209 1.1209
var.(BGK) 1.1366 1.1366 1.1366 1.1366 1.1366

Table 2. The second-order slip coefficient A2 for the CL boundary conditions as given by our
variational results for a rigid-sphere gas (var.(HS)). The outputs based on a variational solution of
the BGK model equation are also included (var.(BGK)).

at an = 0 an = 0.25 an = 0.5 an = 0.75 an = 1

0.25 var.(HS) −1.2707 × 10−1 −4.2429 × 10−2 4.5998 × 10−2 1.3821 × 10−1 2.3420 × 10−1
var.(BGK) 3.2434 × 10−1 4.1084 × 10−1 5.0098 × 10−1 5.9477 × 10−1 6.9219 × 10−1
0.50 var.(HS) −1.3137 × 10−2 4.6217 × 10−2 1.0725 × 10−1 1.6997 × 10−1 2.3436 × 10−1
var.(BGK) 4.4069 × 10−1 5.0117 × 10−1 5.6327 × 10−1 6.2699 × 10−1 6.9233 × 10−1
0.75 var.(HS) 1.0745 × 10−1 1.3859 × 10−1 1.7015 × 10−1 2.0213 × 10−1 2.3453 × 10−1
var.(BGK) 5.6345 × 10−1 5.9510 × 10−1 6.2715 × 10−1 6.5961 × 10−1 6.9247 × 10−1
1.00 var.(HS) 2.3470 × 10−1 2.3470 × 10−1 2.3470 × 10−1 2.3470 × 10−1 2.3470 × 10−1
var.(BGK) 6.9261 × 10−1 6.9261 × 10−1 6.9261 × 10−1 6.9261 × 10−1 6.9261 × 10−1

condition, which is a straightforward discretization of Maxwell’s diffuse-specular-

reflection boundary condition in kinetic theory [6–8]; the other is the combination
boundary condition (CBC), which combines the no-slip bounce-back and the free-
slip specular-reflection schemes [1,8]. Recently, it was shown that both schemes
are virtually equivalent in principle [8] and that they correspond to a second-
order slip boundary condition [1,8]. In particular, the CBC approach provides
free tunable parameters to model slip flow phenomena. Therefore, comparing the
velocity slip coefficients obtained through the continuous Boltzmann equation
with those derived by using the LB method with the combination boundary

Phil. Trans. R. Soc. A (2011)

 Downloaded from http://rsta.royalsocietypublishing.org/ on May 26, 2016

2236 S. Lorenzani

Table 3. Comparison of variational results based on the Maxwell (M) and Cercignani–Lampis (CL)
kernels at a = at = an , for a rigid-sphere gas.

A1 A2
a M CL M CL

0.1 18.9108 18.8933 −1.8311 × 10−1 −1.5353 × 10−1

0.2 9.0832 9.0570 −1.4134 × 10−1 −7.7926 × 10−2
0.3 5.7981 5.7657 −9.8407 × 10−2 −8.6389 × 10−3
0.4 4.1485 4.1128 −5.4309 × 10−2 5.3366 × 10−2
0.5 3.1533 3.1168 −9.0487 × 10−3 1.0725 × 10−1
0.6 2.4851 2.4506 3.7375 × 10−2 1.5231 × 10−1
0.7 2.0039 1.9739 8.4961 × 10−2 1.8797 × 10−1
0.8 1.6394 1.6168 1.3371 × 10−1 2.1378 × 10−1
0.9 1.3529 1.3403 1.8362 × 10−1 2.2942 × 10−1
1.0 1.1209 1.1209 2.3470 × 10−1 2.3470 × 10−1

conditions, written in terms of adjustable parameters, one can fix the values

of these free parameters, in order to recover quantitative agreement up to second
order in the Knudsen number, and then use the LB method as a valuable tool
for predictive simulations.

References
1 Sbragaglia, M. & Succi, S. 2005 Analytical calculation of slip flow in lattice Boltzmann models
with kinetic boundary conditions. Phys. Fluids 17, 093602. (doi:10.1063/1.2044829)
2 Cercignani, C. & Lorenzani, S. 2010 Variational derivation of second-order slip coefficients
on the basis of the Boltzmann equation for hard-sphere molecules. Phys. Fluids 22, 062004.
(doi:10.1063/1.3435343)
3 Cercignani, C., Lampis, M. & Lorenzani, S. 2004 Plane Poiseuille flow with symmetric and
nonsymmetric gas–wall interactions. Transp. Theory Stat. Phys. 33, 545–561. (doi:10.1081/
TT-200053939)
4 Siewert, C. E. 2003 Viscous-slip, thermal-slip, and temperature-jump coefficients as defined by
the linearized Boltzmann equation and the Cercignani-Lampis boundary condition. Phys. Fluids
15, 1696–1701. (doi:10.1063/1.1567284)
5 McCormick, N. J. 2005 Gas-surface accommodation coefficients from viscous slip and
temperature jump coefficients. Phys. Fluids 17, 107104. (doi:10.1063/1.2111133)
6 Ansumali, S. & Karlin, I. V. 2002 Kinetic boundary conditions in the lattice Boltzmann method.
Phys. Rev. E 66, 026311. (doi:10.1103/PhysRevE.66.026311)
7 Ansumali, S., Karlin, I. V., Arcidiacono, S., Abbas, A. & Prasianakis, N. I. 2007 Hydrodynamics
beyond Navier–Stokes: exact solution to the lattice Boltzmann hierarchy. Phys. Rev. Lett. 98,
124502. (doi:10.1103/PhysRevLett.98.124502)
8 Guo, Z., Shi, B., Zhao, T. S. & Zheng, C. 2007 Discrete effects on boundary conditions for
the lattice Boltzmann equation in simulating microscale gas flows. Phys. Rev. E 76, 056704.
(doi:10.1103/PhysRevE.76.056704)

Phil. Trans. R. Soc. A (2011)