The Methods of Chapman-Enskog and Grad and

Applications
Gilberto Medeiros Kremer
Departamento de Fsica
Universidade Federal do Parana
81531-980 Curitiba, BRAZIL

Contents
1 Introduction 3

2 Boltzmann and Balance Equations 4

3 Chapman-Enskog Method 7

4 Application of Chapman-Enskog Method to Granular Gases 14

5 Grads Method of Moments 19

6 Chapman-Enskog-Grad Combined Method 23

7 Fourteen Moment Theory for Granular Gases 25

RTO-EN-AVT-194 9-1
9-2 RTO-EN-AVT-194
 1 INTRODUCTION

1 Introduction
We may say that one of the main objectives of the kinetic theory is to describe the macro-
scopic properties of gases like pressure, temperature, thermal conductivity, viscosity,
diusion, etc. from microscopic quantities that are associated with the molecules which
compose the gases like mass, velocity, kinetic energy, internal degrees of freedom and
interaction forces between the molecules.
The foundations of the modern kinetic theory of gases were established in 1867 by
James Clerk Maxwell (1831-1879) who proposed a general transport equation for arbi-
trary macroscopic quantities associated with mean values of microscopic quantities. This
equation of transport relates the time evolution of a macroscopic quantity with the motion
of the molecules, collision between the molecules and action of external forces. Although
the theory was valid for any molecular interaction potential, Maxwell could only deter-
mine the transport coecients of shear viscosity, thermal conductivity and diusion by
assuming that the interaction potential was derived from a repulsive force which was in-
versely proportional to the fth power of the relative distance between the molecules.
Nowadays this type of potential is known as Maxwellian potential. The kinetic theory of
gases gained a new impulse in 1872 with the work by Ludwig Eduard Boltzmann (1844-
1906), who proposed an integro-dierential equation the Boltzmann equation which
represents the evolution of the velocity distribution function in the phase space spanned
by the coordinates and velocities of the molecules. In the Boltzmann equation the tempo-
ral change of the distribution function has two terms, one of them is a drift term due to
the motion of the molecules while the other one is a collision term related to encounters
of the molecules. Based on this equation, Boltzmann proposed the so-called H function
which decreases with time or remains constant. The identication of this function as the
negative of the gas entropy gave a molecular interpretation of the increase of the entropy
for irreversible processes. Furthermore, Boltzmann in the same work presented a rigorous
deduction of the Maxwellian distribution function.
From the Boltzmann equation one could determine the velocity distribution function
hence the transport coecients of rareed gases, however this task was not so easy. Indeed,
it took almost forty years after the proposition of Boltzmann equation, for David Hilbert
(1862-1943) to show how one could get an approximate solution of the integro-dierential
equation from a power series expansion of a parameter which is proportional to the mean
free path. Further advances were due to Sydney Chapman (1888-1970) and David Enskog
(1884-1947) who in the years 1916 and 1917 calculated independently and by dierent
methods the transport coecients for gases whose molecules interact according to any kind
of spherically symmetric potential function. Another method derived from the Boltzmann
equation was proposed in 1949 by Harold Grad (1923-1986) who expanded the distribution
function in terms of tensorial Hermite polynomials and introduced balance equations
corresponding to higher order moments of the distribution function.
The aim of these notes is to discuss the methods of Chapman-Enskog and Grad with
an applications to granular gases.
(a) Cartesian notation for tensors is used with Latin indexes i, j, k, . . . which may range
from 1 to 3 denoting the three-dimensional system of spatial coordinates x, y, z;
3 Einsteins summation convention over repeated indexes is used, for example, Tij vj
(b)
j=1 Tij vj ;

RTO-EN-AVT-194 9-3
 2 BOLTZMANN AND BALANCE EQUATIONS

(c) Parentheses around the indexes denote the symmetric part of a tensor, brackets its
antisymmetric part while angular parentheses refer to its traceless symmetric part, for
example,
1 1 1
T(ij) = (Tij + Tji ), T[ij] = (Tij Tji ), Tij = T(ij) Trr ij .
2 2 3
Above, ij is Kroneckers symbol and the traceless tensor Tij is called tensor deviator.

2 Boltzmann and Balance Equations

We consider a monatomic rareed gas with N molecules enclosed in a recipient of volume
V , where a molecule is specied as a point in a six-dimensional phase space spanned by its
coordinates x = (x1 , x2 , x3 ) and velocity components c = (c1 , c2 , c3 ). The state of a gas is
characterized by the one-particle distribution function f (x, c, t) such that f (x, c, t)dx dc
gives at time t, the number of molecules in the volume element with position vectors
within the range x and x + dx and with velocity vectors within the range c and c + dc.
An elastic binary collision is characterized by the conservation laws of momentum and
kinetic energy, namely,
1 2 1 2 1 2 1 2
mc + mc1 = mc + mc1 , mc + mc1 = mc + mc1 , (1)
2 2 2 2
where (c, c1 ) refer to pre-collisional velocities and (c , c1 ) to post-collisional velocities of
two molecules. The subindex 1 is used in order to distinguish two identical molecules that
participate in the collision. The relative velocity is denoted by g = c1 c and the energy
conservation law implies that g = g .
The space-time evolution of the one-particle distribution function in the phase space
is governed by the Boltzmann equation, which in the absence of external forces, reads

f f
+ ci = (f1 f f1 f )g b db d dc1 . (2)
t xi
We note that it is a non-linear integro-dierential equation for f (x, c, t). The right-hand
side of the Boltzmann equation is related to the collisions of the molecules through the
product of two distribution functions. The relative motion of the molecules is characterized
by the impact parameter 0 b < and by the azimuthal angle 0 2 (see
Figure 1). Furthermore, the following abbreviations were introduced f f (x, c , t), f1
f (x, c1 , t), f f (x, c, t), f1 f (x, c1 , t).
The multiplication of the Boltzmann equation (2) by an arbitrary function
(x, c, t) and the integration of the resulting equation over all values of the velocity
components c leads to
[ ]

f dc + ci f dc + ci f dc = ( )f1 f g b db d dc1 dc
t xi t xi

1
= (1 + 1 )(f1 f f1 f )g b db d dc1 dc, (3)
4
where the two equalities in the right-hand side of the above equation where obtained from
the symmetry properties of the collision operator.

9-4 RTO-EN-AVT-194
 2 BOLTZMANN AND BALANCE EQUATIONS

Direct collision g = c 1 c

Collision cylinder
Restitution cylinder
d
g
db b
O
g

Collision plane
g

Figure 1: Geometry of a binary collision.

From the analysis of the last equality in the right-hand side of (3) we may infer that
it vanishes for any kind of distribution function f when 1 + = 1 + . A function
which fullls such condition is called a summational invariant. A summational invariant
is given by a linear combination of the conservative quantities: mass, linear momentum
and energy, and is expressed by (c) = A + B c + Dc2 where A and D are two scalar
functions and B a vectorial function, all of them being independent of c.
In kinetic theory a macroscopic state of a gas is characterized by quantities that are
dened in terms of the distribution function f (x, c, t). Firstly, based on the microscopic
quantities of the gas molecules as mass m, linear momentum mci and energy mc2 /2 we
dene the mass density , the momentum density vi and the total energy density u of
the gas, namely,

mc2
= mf (x, c, t) dc, vi = mci f (x, c, t) dc, u = f (x, c, t) dc. (4)
2
If we substitute the molecular velocity in (4)3 by the peculiar velocity Ci = ci vi such
that mCi f dc = 0 we obtain

1 2 1
u = v + , where = mC 2 f (x, c, t) dc. (5)
2 2
Hence, the total energy density of the gas is given by a sum of its kinetic energy density
v 2 /2 and its internal energy density .
We dene the moment of the distribution function of order N by

pi1 i2 ...iN = mCi1 Ci2 . . . CiN f (x, c, t) dc, (6)

which represents a symmetric tensor of order N with (N +1)(N +2)/2 distinct components.
The zeroth moment represents the mass density, the rst moment vanishes and the
second moment, known as the pressure tensor, reads

pij = mCi Cj f (x, c, t)dc. (7)

RTO-EN-AVT-194 9-5
 2 BOLTZMANN AND BALANCE EQUATIONS

We denote by pij the pressure deviator, which represents the traceless part of the pressure
tensor. It is given by

1 1 1
pij = pij prr ij , with p = prr = mC 2 f (x, c, t) dc, (8)
3 3 3
representing the hydrostatic pressure of the gas. The internal energy density , given
by (5)2 is related to the hydrostatic pressure by p = 2/3. The equation of state of an
ideal gas is given by p = nkT where k, n = /m and T denote the Boltzmann constant,
the particle number density and the temperature of the gas, respectively. Hence, we can
obtain the following expression for the temperature of a monatomic gas written in terms
of the distribution function:

p 2m 1
T = = = mC 2 f (x, c, t) dc. (9)
nk 3k 3nk
The heat ux vector is dened as the contracted third order moment

1 1
qi = pjji = mC 2 Ci f (x, c, t) dc, (10)
2 2
and the moments of higher order do not have specic proper names.
We obtain the balance equations for the moments of the distribution function from
the transfer equation (3) by choosing (x, c, t) equal to:
(i) Balance of mass density ( = m):
vi
+ = 0, (11)
t xi
(ii) Balance of momentum density ( = mci ):
vi
+ (vi vj + pij ) = 0, (12)
t xj
(iii) Balance of total energy density ( = mc2 /2):
[ ( 1 2 )] [ ( 1 2) ]
+ v + + v vi + qi + pij vj = 0, (13)
t 2 xi 2
(iv) Balance of N th order moment ( = mCi1 Ci2 . . . CiN ):
pi1 i2 ...iN N pi )k
+ (pi1 i2 ...iN k + pi1 i2 ...iN vk ) p(i1 i2 ...iN 1 N
t xk xk
vi )
+ N pk(i1 i2 ...iN 1 N = Pi1 i2 ...iN . (14)
xk
The above equation was obtained by eliminating the time derivative of the hydrodynamic
velocity vi by the use of (12). The parenthesis around the indexes indicate a sum over all
N ! permutations of these indexes divided by N !. Furthermore, the production term due
to the molecular collisions Pi1 i2 ...iN reads

Pi1 i2 ...iN = m(Ci1 Ci2 . . . CiN Ci1 Ci2 . . . CiN )f f1 g b db d dc dc1 . (15)

9-6 RTO-EN-AVT-194
 3 CHAPMAN-ENSKOG METHOD

If we subtract from the balance equation of the total energy density (13) the mo-
mentum balance equation (12) multiplied by the bulk velocity vi , we obtain the balance
equation for the internal energy density, namely,

vi
+ (vi + qi ) + pij = 0. (16)
t xi xj

3 Chapman-Enskog Method
First let us introduce the dimensionless variables

t = t/tc , x = x/Lc , c = c/c, g = g/g, b = b/d, = /,

f = f /fc ,
(17)
where
d denote the molecular diameter, c = 8kT /m the mean thermal velocity, and g =
2c the mean relative velocity. Furthermore, the characteristic value for the distribution
function fc is equal to n/c3 where n represents the particle number density of the gas and
tc and Lc are a representative time and length that characterize the uid ow. By taking
into account the dimensionless variables (17), the Boltzmann equation (2) can be written
as
f f

1
Sr + ci = (f1 f f1 f )g b db d dc1 . (18)
t xi Kn
In the above equation we have introduced the Strouhal number Sr and the Knudsen
number Kn which are dened by

Lc l
Sr = , Kn = , (19)
c tc Lc

where l = 1/ 2d2 n is the mean free path.
The Knudsen number is related to the degree of rarefaction of a gas. When Kn 1
the molecular collisions are very important, the distribution function is determined by the
collision term of the Boltzmann equation and the gas is described by a continuum regime.
Otherwise, when Kn 1 the molecular collisions become negligible, the distribution
function is determined via a collisionless Boltzmann equation and the regime of the gas
is known as free molecular ow.
In the method of Chapman-Enskog the distribution function f is expanded as

f =f (0)
+ f (1) 2 (2)
+ f + = r f (r) , (20)
r=0

i.e., in power series of a parameter which is of order of the Knudsen number. The
distribution functions f (0) , f (1) and f (2) represent the rst, second and third approximation
to the distribution function, and so on. The parameter can be set later equal to unity,
so that the proper dimensions of the Boltzmann equation are restored. Moreover, the
approximations must satisfy the constraints

f (r) dc = 0, r 1, (21)

RTO-EN-AVT-194 9-7
 3 CHAPMAN-ENSKOG METHOD

where represents the summational invariants m, mci (or mCi ) and mc2 /2 (or mC 2 /2).
If we insert the series expansion of the distribution function (20) into the denitions
of the pressure tensor and of the heat ux vector we obtain

1 2 (0) r (r) (r)
pij = ij mC f dc + mCi Cj f dc = pij + r pij , (22)
3 r=1 r=1

m 2 r (r)

(r)
qi = C Ci f dc = r qi . (23)
2 r=1 r=1

The parameter is introduced into the collision term of the Boltzmann equation (2),
yielding
f 1
Df + Ci = Q(f, f ), (24)
xi
where D = /t + vi /xi denotes the material time derivative. In (24) the collision term
was expressed in terms of an integral for a bilinear quantity

1
Q(F, G) = (F1 G + F G1 F1 G F G1 )g b db d dc1 . (25)
2
The material time derivative, like the distribution function, is also expanded in power
series as


D = D0 + D1 + D2 + =
2
r Dr , (26)
r=1

and the insertion of the expansions (22), (23) and (26) into (11), (12) and (16) lead to
the following decomposition of the balance equations
vi
D0 + = 0, Dr = 0 ( r 1), (27)
xi
(r)
p pij
D0 vi + = 0, Dr vi + = 0 ( r 1), (28)
xi xj
(r)
3 vi 3 q (r) vi
nkD0 T + p = 0, nkDr T + i + pij = 0 ( r 1). (29)
2 xi 2 xi xj

We can now obtain the integral equations for the approximations of the distribution
function by the inserting the expansions (20) and (26) into the Boltzmann equation (24)
and equating equal powers of . Hence, it follows
( )
Q f (0) , f (0) = 0, (30)
( ) f (0)
2Q f (0) , f (1) = D0 f (0) + Ci , (31)
xi
( ) ( ) f (1)
2Q f (0) , f (2) + Q f (1) , f (1) = D0 f (1) + D1 f (0) + Ci , (32)
xi

and so on. The above equations represent the three rst integral equations for f (0) , f (1)
and f (2) , respectively.

9-8 RTO-EN-AVT-194
 3 CHAPMAN-ENSKOG METHOD

For the determination of the rst approximation to the distribution function f (0) from
(0)
the integral equation (30) we note that its solution is given by f (0) f1 = f (0) f1 or
(0)

(0)
ln f (0) + ln f1 = ln f (0) + ln f1 . Hence, ln f (0) is a summation invariant so that it must
(0)

be represented by ln f (0) = A + B C + DC 2 . If we insert this representation for f (0) into

the denitions of the mass density (4)1 , hydrodynamic velocity (4)2 and temperature (9)
we obtain the Maxwellian distribution function, namely,
( ) 32
m
eC
(0) 2
f =n where = . (33)
2kT
For the determination of the second approximation f (1) we write f (1) = f (0) and the
integral equation (31) reduces to
{ ( ) }
( (0) (0) ) 1 5 T vi
2Q f , f = f (0)
C
2
Ci + 2Ci Cj
T 2 xi xj

= I[] = f (0) f1 (1 + 1 ) g b db d dc1 ,
(0)
(34)

by eliminating the material time derivatives through the use of (27)1 , (28)1 and (29)1 .
The general solution of the integral equation (34) is given by a sum of the solution
of the homogeneous integral equation plus a particular solution of it, i.e., = h + p .
The solution of the homogeneous integral equation I[h ] = 0 is a summational invariant
h = 1 +r2 Cr +3 C 2 , where 1 and 3 are scalar functions, while r2 is a vector function,
all of them do not depend on the peculiar velocity C.
The particular solution of the integral equation is expressed as a linear combination
of the temperature gradient and of the velocity gradient deviator, namely,
Ai T vi
p = 2Bij . (35)
T xi xj
where Ai and Bij denote a vector function and a tensor function of (C, , T ), respectively.
It is easy to show that the representations of the vector and tensor functions in terms
of the peculiar velocity C are given by Ai = A Ci and Bij = BCi Cj , where the scalar
functions A and B depend only on (C 2 , , T ).
Hence, the solution of the non-homogeneous integral equation (34) is given by
Ci T vi
= A 2BCi Cj + 1 + r2 Cr + 3 C 2 . (36)
T xi xj
The second approximation f (1) = f (0) must satisfy the constraints (21) which imply
the following restrictions
[ ]
3kT 5kT A T
1 + 3 = 0, 1 + 3 = 0, mC 2
+ j f (0) dc = 0. (37)
2
m m T xj
From the two rst equations we may infer that 1 = 3 = 0, whereas the last equation
implies that j2 must be proportional to T /xj . By writing the proportionality factor of
j2 as /T , the constraint (37)3 reduces to

C 2 Af (0) dc = 0, (38)

RTO-EN-AVT-194 9-9
 3 CHAPMAN-ENSKOG METHOD

with A = A being a new scalar function of (C 2 , , T ).

Hence, from the above results we may write the distribution function (20) up to the
second approximation and with = 1 as
{ }
A T vi
(0)
f = f (1 + ) = f (0)
1 Ci 2BCi Cj . (39)
T xi xj
Now the scalar coecients A and B in the distribution function (39) can be obtained
as solutions of the integral equations that follows from (34), namely,
[ ]
5
f (0)
C
2
Ci = I[ACi ], f (0) Ci Cj = I[BCi Cj ]. (40)
2
In order to solve the integral equations (40) the scalar coecients A(C 2 , , T ) and
B(C 2 , , T ) are expanded in series of Sonine polynomials. The Sonine polynomials of
order n and index (l + 1/2) in the variable C 2 = mC 2 /2kT are dened by

(n)

n
(n + l + 3/2)
Sl+1/2 (C 2 ) = (C 2 )k , (41)
k=0
k!(n k)!(k + l + 3/2)

and obey the orthogonality conditions

1 (n + l + 3/2)
eC C 2l+2 Sl+1/2 (C 2 )Sl+1/2 (C 2 ) l+3/2 dC =
2 (n) (p)
np , (42)
0 2 n!
where (n) denotes the gamma function. The rst three Sonine polynomials read
(0) (1) 3
Sl+1/2 (C 2 ) = 1, Sl+1/2 (C 2 ) = l + C 2 , (43)
2
1 ( 5 )( 3 ) ( 5 ) 1
(2)
Sl+1/2 (C 2 ) = l+ l+ l+ C 2 + 2 C 4 . (44)
2 2 2 2 2
The integral equations (40) in terms of the Sonine polynomials become
(1) (0) [ ]
f (0) S3/2 (C 2 )Ci = I [ACi ] , f (0) S5/2 (C 2 )Ci Cj = I BCi Cj . (45)

Moreover, the expansions of the scalar coecients A(C 2 , , T ) and B(C 2 , , T ) in series of
Sonine polynomials are written as


(r)


(r)
A(C 2 , , T ) = a(r) S3/2 (C 2 ), B(C 2 , , T ) = b(r) S5/2 (C 2 ). (46)
r=0 r=0

In the above equations, the scalar coecients a(r) and b(r) are only functions of (, T ),
and from this point on, the dependence of the Sonine polynomials in the variable C 2 is
omitted.
The coecient A must satisfy the constraint (38), hence by using its representation
(46)1 , we get
( ) 32
3 n (0)
S3/2 S3/2 C 4 eC dC =
2 (0) (r) (r) (r) (0) (r) 2
0= C f a S3/2 dc = 4n a a ,
r=0
r=0 0 2
(47)

9 - 10 RTO-EN-AVT-194
 3 CHAPMAN-ENSKOG METHOD

due to (42) and (43)1 . From (47) it follows that a(0) = 0 and the scalar coecient A
reduces to

(r)
A(C , , T ) =
2
a(r) S3/2 . (48)
r=1

Now the insertion of the expressions (48) and (46)2 into their respective integral equations
(45), yields

[ ]
[ ]
(1) (r) (0) (r)
f (0) S3/2 Ci = a(r) I S3/2 Ci , f (0) S5/2 Ci Cj = b(r) I S5/2 Ci Cj . (49)
r=1 r=0

In order to determine the coecients a(r) e b(r) from (49) we proceeds as follows. First
(s)
we multiply the integral equation (49)1 by S3/2 Ci and integrate the resulting equation
over all values of C. By using (42) we obtain

15 n (1,r) (r,s) (s)

1 (r) (s)
= a , where (r,s)
= S3/2 Ci I[S3/2 Ci ]dC. (50)
4 2 s=1

In the above equation (1,r) represents Kroneckers symbol. Following the same method-
(s)
ology, we multiply the integral equation (49)2 by 2 S5/2 Ci Cj , integrate the resulting
equation and get

5 n (0,r) (r,s) (s)

(r) (s)
= b , where (r,s)
= S5/2 Ci Cj I[S5/2 Ci Cj ]dC. (51)
2 2 s=0

The integrals in (50)2 and (51)2 depend on the molecularinteraction potential and on the
elds (, T ). Moreover, from the condition I[]dc = I[]dc which is valid for all
arbitrary functions (x, c, t) and (x, c, t) we may infer that

(r,s) = (s,r) , (r,s) = (s,r) and (r,0) = (0,r) = 0. (52)

We conclude that (50)1 and (51)1 represent an innite system of algebraic equations
for the coecients a(n) and b(n) , respectively.
To obtain the constitutive equations for the pressure tensor and for the heat ux
vector we proceed as follows. First, we insert the non-equilibrium distribution function
(39) together with the representations of A and B given by (48) and (46)2 , respectively
into the denitions of the pressure tensor and heat ux vector and get


(0) (r) vp
pij = mCi Cj f dc = p ij 2mCi Cj S5/2 f (0) b(r) S5/2 Cp Cq dc, (53)
r=0
xq
[ ]
m 2 m 5 (0) (1) (r) T
qi = C Ci f dc = S3/2 S3/2 Ci f (0)
a(r) S3/2 Cp dc. (54)
2 2T 2 r=1
x p
[ ]
(0) (1)
In (54) we have used the relationship C 2 = 52 S3/2 S3/2 which follows from (43)2 .
Next, the substitution of the Maxwellian distribution function (33) into (53) and (54) and

RTO-EN-AVT-194 9 - 11
 3 CHAPMAN-ENSKOG METHOD

the subsequent integration of the resulting equations over all values of C, yield
vi (0)
pij = p ij 2 , where = b , (55)
xj 2
T 5k (1)
qi = , where = a , (56)
xi 4m
thanks to the orthogonality condition of the Sonine polynomials (42). Equations (55)1
and (56)1 are the mathematical expressions for the laws of Navier-Stokes and Fourier,
with and representing the coecients of shear viscosity and thermal conductivity,
respectively.
We note from (55)2 and (56)2 that the transport coecients of shear viscosity and
thermal conductivity are functions only of the coecients a(1) and b(0) , respectively. We
can obtain these coecients from the system of equations (50)1 and (51)1 , which can be
written as
A B
a(1) = lim nn , b(0) = lim nn , (57)
n Ann n Bnn

where
 15 n 
 (1,2)
. . . (1,n)   
 4 2   (1,1) . . . (1,n) 
 0 (2,2) . . . (2,n)    
  .. . ..  ,
.
Ann = det  . . . .  , Ann = det  . .. (58)
 .. .. .. ..   (n,1) 
   ... (n,n) 
 0 (n,2) . . . (n,n) 
 5n 
 (0,1)
. . . (0,n)   
 2 2   (0,0) . . . (0,n) 
 0 (1,1) . . . (1,n)   
   .. . . 
Bnn = det  . . . . , Bnn = det  . . . .
. . (59)
 .. .
. . . .
.   (n,0) 
   ... (n,n) 
 0 (n,1) . . . (n,n) 

The coecients a(1) and b(0) follow from (58) and (59) through a method of successive
approximations, where the approximation of order p for the coecients a(1) and b(0)
denoted by [a(1) ]p and [b(0) ]p is given by the sub-determinant of order p. The rst two
approximations for the coecients a(1) e b(0) are
15 n 1 5 n 1
[a(1) ]1 = , [b(0) ]1 = , (60)
4 2 (1,1) 2 2 (0,0)
15 n (2,2) 5 n (1,1)
[a(1) ]2 = , [b(0) ]2 = . (61)
4 2 (1,1) (2,2) (1,2) 2 2 2 (0,0) (1,1) (0,1) 2
The integrals (r,s) and (r,s) are given in terms of the collision integrals

( )
e 2r+3 1 cosl b db d,
(l,r) 2
= (62)
0 0

where = arccos(g g/g 2 ) is the scattering angle and = /2 g represents a dimen-
sionless relative velocity.
The determination of the integrals (r,s) and (r,s) is not an easy task. As an example,
let us compute the integral (0,0) . For its calculation a change of the integration variables

9 - 12 RTO-EN-AVT-194
 3 CHAPMAN-ENSKOG METHOD

is performed by replacing the velocities C and C1 by the relative velocity g = C1 C and

by the center of mass velocity G = (C1 + C)/2. The linear momentum conservation law
implies that G = G and the Jacobian of the transformation of the volume elements has
modulus equal to one so that dCdC1 = dgdG. In terms of the new variables the integral
reads
( )3
n2 2
e2G e 2 g
2
(0,0)
= Ci Cj I[Ci Cj ]dc =
2
( )
1 ( )
Gi Gj Gi gj + gi gj gi gj

gi gj g b db ddgdG. (63)
4

The integration of (63) with respect to G and leads to

( ) 32
n2 2 ( )
(0,0)
= e 2 g cos2 1 g 5 b db dg, (64)
4 2

If we introduce the dimensionless relative velocity = /2 g, the integration of the
above equation with respect to the angles of g leads to
( ) 52
2
(0,0)
= n2 (2,2) , (65)

where (2,2) denotes the collision integral dened by (62).

5 5
(r,s) (r,s)
By introducing the integrals = n22 (r,s) , and = n22 (r,s) , we can obtain
2 2

in the same way the following expressions in terms of the collision integrals (l,r)
77 (2,2)
(1,1) = 4(2,2) , (1,2) = 7(2,2) 2(2,3) , (2,2) = 7(2,3) + (2,4) , (66)
4
301 (2,2) 35
(0,0) = (1,1) , (0,1) = (1,2) , 7(2,3) +(2,4) = (2,2) + (1,1) .
(1,1) =

12 24
(67)
(0) (1)
As was pointed out the coecients b and a are determined through a method of
successive approximations, so are the transport coecients and . Hence, it follows
from (60), (61), (55)2 and (56)2 that the two rst approximations to the coecients of
shear viscosity and thermal conductivity read

5 mkT 1 75k mkT 1
[]1 = , []1 = , (68)
4 (0,0) 16m (1,1)
(1,1)
(2,2)
5 mkT 75k mkT
[]2 = , [] 2 = . (69)
4 (0,0) (1,1) (1,2) 2 16m (1,1) (2,2) (1,2) 2

Now from (68) and (69) we can build the ratios

( )1 ( )1
(1,2) 2 (0,1) 2
[]1 5 []2 []2
= cv , = 1 (1,1) (2,2)
> 1 (0,0) (1,1)
= . (70)
[]1 2 []1 []1

RTO-EN-AVT-194 9 - 13
 4 APPLICATION OF CHAPMAN-ENSKOG METHOD TO GRANULAR GASES

We may infer from (70)1 that the ratio of the rst approximations to the coecients of
thermal conductivity and shear viscosity of monatomic ideal gases is equal to 5cv /2 for
all types of spherically symmetrical molecular interaction potentials, where cv = 3k/2m is
the specic heat at constant volume. Furthermore, from the relationship (70)2 it follows
that []2 /[]2 > []1 /[]1 = 5cv /2, i.e., the ratio of the second approximations is larger
than the ratio of the rst ones.
For a hard-sphere potential the impact parameter is given by b = d cos(/2) with
0 , so that the collision integral (2,r) becomes
( )
(2,r) 2 2r+3 1 2 d2 2r + 4
= e d d = . (71)
0 3 6 2

Hence, it follows from (66) and (67) that

45 2 205 2
(1,1) = (0,0) = 4d2 , (1,2) = (0,1) = d2 , (2,2) = d, (1,1) = d , (72)
4 12
and the two rst approximations (68) and (69) to the coecients of shear viscosity and
thermal conductivity for the hard-sphere potential read

5 mkT []1 5 []2 205 []2 45
[]1 = 2 , = cv , = 1.014 851, = 1.022 727.(73)
16d []1 2 []1 202 []1 44

We note that the second approximations to the transport coecients is a small correction
to the rst ones, so is the ratio []2 /[]2 1.007 761[]1 /[]1 .
The fth approximations to the transport coecients for hard-sphere potential are

[]5 []5
1.016 027, 1.025 197.
[]1 []1

4 Application of Chapman-Enskog Method to Gran-

ular Gases
The mechanical energy of a gas is conserved when its molecules undergo elastic colli-
sions and, in this case, the gas relaxes towards an equilibrium state characterized by a
Maxwellian distribution function. However, the inelastic collisions of the gas molecules
transform the translational kinetic energy into heat and the mechanical energy lost im-
plies a temperature decay of the gas. In the literature gases whose molecules undergo
inelastic collisions are known as granular gases. The main premisses of the kinetic theory
of granular gases are: (a) only binary collisions of hard spherical molecules are taken into
account and (b) the energy lost from inelastic collisions is small.
Let us consider the encounter of two identical molecules of mass m diameter d pre-
collisional velocities (c, c1 ) and post-collisional velocities (c , c1 ). The momentum conser-
vation law mc+mc1 = mc +mc1 holds for inelastic collisions, but the inelastic encounters
are characterized by the relationship (g k) = (g k) which relates the pre- and post-
collisional relative velocities at collision. The parameter 0 1 here considered as
a constant refers to the normal restitution coecient and k is the unit vector directed

9 - 14 RTO-EN-AVT-194
 4 APPLICATION OF CHAPMAN-ENSKOG METHOD TO GRANULAR GASES

along the line which joins the molecules centers and pointing from center of the molecule
labeled by the index 1 to the center of the molecule without label. The component of the
velocity perpendicular to the collision vector k does not change in inelastic collisions, so
that k g = k g. From the momentum conservation law it follows the relationships
which give the post-collisional velocities in terms of the pre-collisional ones, as well as
relationships which connect the relative velocities and their modulus:
1+ 1+
c = c + (g k)k, c1 = c1 (g k)k, (74)
2 2
g = g (1 + )(g k)k, g 2 = g 2 (1 2 )(g k)2 . (75)

In the case of elastic collisions, = 1 and it follows the conservation of the kinetic energy
g = g.
A restitution collision with pre-collisional velocities (c , c1 ) that corresponds to the
post-collisional velocities (c, c1 ) are characterized by the equations
1+ 1+
c = c + (g k)k, c1 = c1 (g k)k, (76)
2 2
where k = k and (g k) = (g k).
For the determination of the Boltzmann equation we have to know the transformation
of the volume elements dc1 dc = |J|dc1 dc, where the modulus of the Jacobian is given
by |J| = 1/. Hence, it follows the relationship
1
(g k ) dc dc1 = (g k) dc dc1 , (77)
2
and the Boltzmann equation for granular gases reads
( )
f f 1
+ ci = f f f1 f d2 (g k) dk dc1 QI (f, f ). (78)
t xi 2 1
Above we have introduced the bilinear form
( )
1 1 1
QI (F, G) = F G + 2 F G1 F1 G F G1 d2 (g k) dk dc1 . (79)
2 2 1
The balance equations for the mass density , for the momentum density vi and
for the specic internal energy = 3kT /2m are obtained from the Boltzmann equation
(78) by multiplying it successively by m, mci and mC 2 /2 and integrating the resulting
equations. Hence, it follows
( )
vi pij 2 qi vi
D + = 0, Dvi + = 0, DT + + pij + T = 0. (80)
xi xj 3nk xi xj

We observe that the balance equation for the temperature (80)3 has the additional term
T which represents the energy loss due to inelastic collisions. The coecient , known
as cooling rate of the granular gas, is given by

d2 m(1 2 )
= f1 f (g k)3 dk dc1 dc. (81)
12nkT

RTO-EN-AVT-194 9 - 15
 4 APPLICATION OF CHAPMAN-ENSKOG METHOD TO GRANULAR GASES

It is easy to verify from the above equation that the cooling rate vanishes for elastic
collisions where = 1.
The methodology used to determine the distribution function by using the Chapman-
Enskog method for granular gases is dierent from the one applied to a monatomic gas
of elastic spherical molecules, since we have to take into account that there exists no
equilibrium state characterized by a Maxwellian distribution function. The rst dierence,
refers to the decomposition of the balance equations, since they are written as
vi
D0 n = 0, D1 n = n , (82)
xi
1 p
D0 vi = 0, D1 v i = . (83)
xi
2T vi
D0 T = T (0) , D1 T = T (1) . (84)
3 xi
The rst and the second approximation to the cooling rate in (84) read

d2 m(1 2 ) (0)
(0)
= f1 f (0) (g k)3 dk dc1 dc, (85)
12nkT

d2 m(1 2 ) ( (1) (0) (0)
)
(1)
= f1 f + f1 f (1) (g k)3 dk dc1 dc. (86)
12nkT
In terms of the parameter we write the Boltzmann equation for granular gases (78)
as
1 f 1
Df + Ci = QI (f, f ), (87)
xi
instead of the representation (24), indicating that the material time derivative and the
collision term are of same order, while the spatial gradients are of higher order.
The two rst integral equations for the determination of f (0) and f (1) are obtained by
inserting the expansions (20) and (26) into the Boltzmann equation (87) and equating the
equal powers of , yielding

f (0)
D0 f (0) = QI (f (0) , f (0) ), D1 f (0) + D0 f (1) + Ci = 2QI (f (1) , f (0) ). (88)
xi
In order to determine the rst approximation of the distribution function the integral
equation (88)1 is written as

f (0)
T (0) = QI (f (0) , f (0) ), (89)
T
through the elimination of the material time derivatives D0 by using (82)1 , (83)1 and
(84)1 . We note that the solution of the integral equation (89) for the distribution function
f (0) is not a Maxwellian, but we may write f (0) as a Maxwellian plus an expansion in
Sonine polynomials of the peculiar velocity, i.e.,
[ ] ( ) 23


eC ,
(0) (n) 2 2
f = fM 1 + an S 1 (C ) , where fM = n (90)
n=1
2

9 - 16 RTO-EN-AVT-194
 4 APPLICATION OF CHAPMAN-ENSKOG METHOD TO GRANULAR GASES

denotes the Maxwellian distribution function. The scalar coecients an do not depend
on the peculiar velocity C and we suppose that this series expansion converges for large
values of the peculiar velocity. We may approximate the representation of the distribution
function (90) as
[ ]
(1) (2)
f (0) = fM 1 + a1 S 1 (C 2 ) + a2 S 1 (C 2 ) , where (91)
2 2

(1) 3 (2) 15 5 2 1 2 4
S 1 (C 2 ) = C 2 , S 1 (C 2 ) = C + C . (92)
2 2 2 8 2 2
If we insert the representation (91) into the denition of the temperature (9) and integrate
the resulting equation we get that a1 = 0.
For the determination of the coecient a2 we proceed as follows. First the product of
the distribution functions is written as
( )3 ( ){ [ ( )
2
(0) (0) 2G2 + g2 15 5 g2
f f1 = n 2
e 1 + a2 2
2G +
4 2 2
( )] }
2 g4
+ 4 2 2
2G + G g + + 2(G g)2
, (93)
2 8

by neglecting all product of the coecient a2 and changing the velocity variables (C, C1 )
by the relative and the center of mass velocities (g, G). Next, we multiply the integral
equation (89) by an arbitrary function of the peculiar velocity (C 2 ) and integrate the
resulting equation over all values of c, yielding

(0) f
(0)
1 [
(C )T
2
dc = (C12 ) + (C 2 )
T 2
] (0)
(C12 ) (C 2 ) f1 f (0) d2 (g k) dk dc1 dc. (94)

If we choose (C 2 ) = 1 and (C 2 ) = C 2 in (94) the integration of the resulting equations

lead to identities. However, if we choose (C 2 ) = C 4 in (94) and use the relationship

(2 1)2
C14 + C 4 C14 C 4 = 2( + 1)2 (g k)2 (G k)2 + (g k)4
8
(2 1)
+(2 1)(g k)2 G2 + (g k)2 g 2 4( + 1)(g k)(G k)(G g), (95)
4
we obtain through the integration of the resulting equation that the coecient a2 has the
form
16(1 )(1 22 )
a2 = . (96)
81 17 + 302 (1 )
We may infer from the above equation that the coecient a2 vanishes for elastic collisions
where = 1.
From the knowledge of the coecient a2 , the distribution function (91) is written as
[ ]
(0) 16(1 )(1 22 ) (2) 2
f = fM 1 + S 1 (C ) . (97)
81 17 + 302 (1 ) 2

RTO-EN-AVT-194 9 - 17
 4 APPLICATION OF CHAPMAN-ENSKOG METHOD TO GRANULAR GASES

1
= 1.0
0.8 = 0.8

(0)
0.6 f

0.4

0.2

0
0 0.5 1 1.5 2
x

Figure 2: Left frame: dimensionless distribution function 3/2 f (0) /n 3/2 as a function of
x = C 2 for = 1 and 0.8. Right frame: cooling rate (0) /nd2 versus the temperature
and the normal restitution coecient.

If we substitute the distribution function (97) into the expression for the cooling rate
(85) and integrate the resulting equation we get that
[ ]
4 2 kT 3(1 )(1 22 )
(0) = nd (1 )
2
1+ . (98)
3 m 81 17 + 302 (1 )

In the left frame of Figure 2 it is represented the dimensionless distribution function

f /n 3/2 as a function of x = C 2 and . We may infer from this gure that the
3/2 (0)

peak of the curve decreases by decreasing the normal restitution coecient. In the right
frame of the Figure 2 it is plotted the cooling rate (0) /nd2 versus the temperature and
the normal restitution coecient. We note that the cooling rate increases by decreasing
the normal restitution coecient and by increasing the temperature. Furthermore, the
cooling rate is zero when = 1, i.e., for elastic collisions of the molecules.
The determination of the second approximation to the distribution function f (1) from
the integral equation (88)2 is more involved and the details of such kind of calculation will
not be given here. From this integral equation we may conclude that f (1) is a function
of the thermodynamic forces: deviator of the velocity gradient vi /xj , gradient of the
particle number density n/xi and gradient of temperature T /xi . Hence, we may
write the second approximation f (1) as
[ ]
(1) (1) 2 ln T (0) 2 vi (1) 2 ln n
f = fM 1 S 3 (C )Ci + 2 S 5 (C )Ci Cj + 3 S 3 (C )Ci , (99)
2 xi 2 xj 2 xi

where 1 , 2 and 3 are scalar coecients that depend on n, T and . If we solve the
integral equation (88)2 we get that the coecients of the distribution function f (1) are

9 - 18 RTO-EN-AVT-194
 5 GRADS METHOD OF MOMENTS

given by
( )
15 m 15 1 m 3
1 = 2 , 2 = 2 ,(100)
d n(9 + 7)( + 1) kT 2d n(13 )( + 1) kT

300(1 ) m
3 = 2 . (101)
d n(9 + 7)( + 1)(19 3) kT

We obtain the constitutive equations for the pressure tensor and heat ux vector
through the substitution of the distribution function (99) into their denitions and by
integrating the resulting equations. Hence, it follows the laws of Navier-Stokes and Fourier

vi
pij = mCi Cj f dc = p ij 2 , (102)
xj

m 2 T ln n
qi = C Ci f dc = T . (103)
2 xi xi

Above, the coecients of shear viscosity , thermal conductivity and the one associated
with the gradient of particle number density read

15 mkT 75 k mkT
= , = 2 , (104)
2d2 (13 )( + 1) 2d (9 + 7)( + 1) m

750(1 ) k mkT
= 2 . (105)
d n(9 + 7)( + 1)(19 3) m

We note that the coecients are proportional to T so that they increase by increasing of
the temperature. With respect to the normal restitution coecient , we may infer that
the coecients also increase by decreasing the values of . Furthermore, by considering
elastic collisions where = 1 the coecients of shear viscosity and thermal conductivity
reduce to the rst approximations (73)1,2 , while the coecient vanishes.

5 Grads Method of Moments

Let us consider a thirteen scalar eld description of a rareed gas within Grads moment
method. In this case the elds are the mass density (x, t), the hydrodynamic velocity
vi (x, t), the pressure tensor pij (x, t) and the heat ux vector qi (x, t) and their balance
equations read

vi
+ = 0, (106)
t xi
vi (vi vj + pij )
+ = 0, (107)
t xj
pij (pijk + pij vk ) vj vi
+ + pki + pkj = Pij , (108)
t xk xk xk
qi (qij + qi vj ) vj vi pki pkj 1 prr pij
+ + pijk + qj = Qi . (109)
t xj xk xj xj 2 xj

RTO-EN-AVT-194 9 - 19
 5 GRADS METHOD OF MOMENTS

We note that the two rst equations above represent the balance equations of mass density
(11) and moment density (12), while the balance equation for the pressure tensor (108)
was obtained from (14) by taking N = 2. The balance equation for the heat ux vector
(109) follows from (14) by taking N = 3, contracting two indices and dividing the resulting
equation by 2. Moreover, we have introduced the following denitions in (108) and (109):

( )
pijk = mCi Cj Ck f dc, Pij = m Ci Cj Ci Cj f f1 d, (110)

m 2 m ( 2 )
qij = C Ci Cj f dc, Qi = C Ci C 2 Ci f f1 d. (111)
2 2
The tensors pijk and qij represent non-convective uxes for the pressure tensor and heat
ux vector, respectively, while Pij and Qi refer to their production terms. In the above
equations we have introduced the abbreviation d = g b db d dc1 dc.
If we analyze the system of partial dierential equations (106) (109) we may infer
that it cannot be considered a system of eld equations for the basic elds , vi , pij and
qi . In fact, to obtain a closed system of dierential equations, the non-convective uxes
pijk , qij and the production terms Pij , Qi must be expressed in terms of the basic elds,
and for this end we must express the distribution function in terms of the thirteen scalar
elds , vi , pij and qi .
In Grads moment method the distribution function is expanded in series of tensorial
Hermite polynomials Hi1 i2 ...iN (N = 0, 1, 2, . . . ) as follows
( )
1 1
f =f (0)
aH + ai Hi + aij Hij + + ai i ...i Hi i ...i + . . . , (112)
2! N! 1 2 N 1 2 N

where ai1 i2 ...iN (N = 0, 1, 2, . . . ) are tensorial coecients that depend on x and t. Fur-
thermore, the Maxwellian distribution function f (0) is written as
( ) 32 ( m ) 32
(0) C 2
f =n e =n (), (113)
kT

where () is the weight function:

1 2 /2 m
() = 3 e , with i = Ci . (114)
(2) 2 kT

On the basis of the weight function, the tensorial Hermite polynomials are written as

(1)N N
Hi1 i2 ...iN () = , (115)
i1 i2 . . . iN

being orthogonal with respect to (), i.e.,

()Hi1 i2 ...iN ()Hj1 j2 ...jM ()d = M N i1 j1 i2 j2 ...iN jN . (116)

In the above equation, i1 j1 i2 j2 ...iN jN represents the sum i1 j1 i2 j2 ...iN jN = i1 j1 i2 j2 . . . iN jN +

(all permutations of the j s indices) and M N is Kroneckers symbol.

9 - 20 RTO-EN-AVT-194
 5 GRADS METHOD OF MOMENTS

From (115) we can obtain the rst four tensorial Hermite polynomials, namely,

H() = 1, Hi () = i , Hij () = i j ij , (117)

Hijk () = i j k (i jk + j ik + k ij ). (118)

In terms of the tensorial Hermite polynomials we can write the thirteen moments of
the distribution function as
( ) 32 ( )2
kT kT
= m f Hd, 0= m f Hi d, (119)
m m
( ) 52
m kT
T = m f (Hii + 3H)d, (120)
3k m
( ) 52 ( ) ( )3
kT 1 m kT
pij = m f Hij Hrr ij d, qi = f (Hijj + 5Hi ) d.(121)
m 3 2 m
We recall that the temperature is given by T = 2mprr /3k.
The coecients ai1 i2 ...iN that appear in the distribution function (112) are determined
from the denition of the moments of the distribution function. Furthermore, for a thirteen
moment theory the distribution function (112) is written as
( )
(0) 1 1
f =f aH + ai Hi + aij Hij + arri Hssi , (122)
2 10
where a, ai , aij and arri represent thirteen scalar coecients to be determined. We note
that the term 3!1 aijk Hijk was decomposed according to
[ ]
1 1 1 1
aijk + (arri jk + arrj ik + arrk ij ) Hijk = aijk Hijk + arri Hssi , (123)
6 5 6 10

and the part associated with the third order tensor 16 aijk Hijk was not taken into account.
If we insert now the distribution function (122) into the denitions of the moments
(119) (121), integrate the resulting equations and use the orthogonality properties of
the tensorial Hermite polynomials (116), we obtain that

pij 2qi m
a = 1, ai = 0, arr = 0, aij = , arri = . (124)
p p kT
Hence, the distribution function for the thirteen moments (with = m/2kT ) becomes
{ [ ( )]}
2 2 4 5
f =f (0)
1+ pij Ci Cj + qi Ci C
2
, (125)
5 2
which is the so-called Grads distribution function.
In order to determine the non-convective uxes pijk and qij we insert Grads distribu-
tion function (125) into the denitions (110)1 and (111)1 and obtain by integrating the
resulting equations:
2 5p2 7p
pijk = (qi jk + qj ik + qk ij ) , qij = ij + pij . (126)
5 2 2

RTO-EN-AVT-194 9 - 21
 5 GRADS METHOD OF MOMENTS

For the calculation of the production terms Pij and Qi we change the integration
variables (c, c1 ) 7 (g, G) so that in these new variables (110)2 and (111)2 become

m[1 ( ) ( ) ]
Pij = gi gj gi gj Gi gj gj Gj (gi gi ) f f1 d, (127)
2 2
[ ( ) ]
m 1 1 1 2
Qi = Gr (gr gi gr gi ) G + g (gi gi ) Gi Gr (gr gr ) f f1 d, (128)
2
2 2 2 4
where the product of Grads distribution functions in a linearized theory reads
( )3 { [ ]
2
(2G2 + 2 g 2 ) 4 2 1
f f1 = n e 1+ pkl Gk Gl + gk gl (129)
4
2
[ ( ) ]}
8 5
+ qj 2Gj G2 + g 2 + gj Gk gk . (130)
5 4 2
The integration of (127) and (128), yields

pij 2qi 5 m 1
Pij = , Qi = , where r = , (131)
r 3r 16n kT (2,2)
represents the relaxation time of the pressure deviator.
Once we know the constitutive equations for the non-convective uxes (126) and pro-
duction terms (131) as functions of the thirteen basic elds , vi , pij and qi , a system of
linearized thirteen scalar eld equations can be obtained from (106) (109), namely
vi
D + = 0, (132)
xi
p pij
Dvi + + = 0, (133)
xi xj
3 qi vi vi
nkDT + +p + pij = 0, (134)
2 xi xi xj
4 qi vi pij
Dpij + + 2p = , (135)
5 xj xj r
p pik 5 k T 2qi
Dqi + + p = . (136)
xk 2 m xi 3r
If we restrict ourselves to a ve eld theory described by the elds of mass density,
momentum density and temperature, the corresponding balance equations are given by
(132), (133) and (134). In this case the pressure deviator and the heat ux vector are no
longer variables, just constitutive quantities. We may use the remaining eight equations
(135) and (136) in order to obtain the constitutive equations for the pressure deviator
and for the heat ux vector, by considering only the equilibrium values of the pressure
deviator pij = 0 and of the heat ux vector qi = 0 on the left-hand sides of (135) and
(136). Hence, only the underlined terms remain, and we obtain

vi 5 mkT 1
pij = 2 , where = pr = , (137)
xj 16 (2,2)

T 15k 75k mkT 1
qi = , where = pr = . (138)
xi 4m 64m (2,2)

9 - 22 RTO-EN-AVT-194
 6 CHAPMAN-ENSKOG-GRAD COMBINED METHOD

Equations (137) and (138) represent the laws of Navier-Stokes and Fourier, respectively,
and the coecients of shear viscosity and thermal conductivity correspond to the rst
approximation to these coecients found by applying the Chapman-Enskog method (see
(68)).

6 Chapman-Enskog-Grad Combined Method

We describe now another method to obtain the constitutive equations for the pressure
deviator and for the heat ux vector which combines the features of Chapman-Enskog
and Grads methods. In this method neither a solution of the integral equation is needed
as in the Chapman-Enskog method nor the eld equations for the moments of the
distribution function are used as in Grads method.
To begin with we observe that in the representation f = f (0) (1+) the deviation of the
Maxwellian distribution function in the Chapman-Enskog method which is a unknown
quantity can be written in terms of known quantities. These known quantities can be
chosen as the thirteen scalar elds of mass density, hydrodynamic velocity, temperature,
pressure deviator and heat ux vector which appear in Grads distribution function (125),
namely, [ ( )]
2 2 4 5
= pij Ci Cj + qi Ci C
2
. (139)
5 2
In the framework of the Chapman-Enskog-Grad combined method we insert the rep-
resentation (139) into the non-homogeneous integral equation of the Chapman-Enskog
method (34), so that it becomes an equation for the determination of the pressure devia-
tor pij and of the heat ux vector qi . This equation reads
{ ( ) }
1 5 T vk 2 2
f (0)
C 2
Ck + 2Ck Cl = pkl I[Ck Cl ]
T 2 xk xl
[( ) ]
8 2 5
+ qk I C
2
Ck . (140)
5 2

For the determination of the pressure deviator, we multiply (140) by Ci Cj and inte-
grate the resulting equation over all values of c, yielding

p vi 2 2
2 = pkl Ci Cj I[Ck Cl ]dc. (141)
m xj

To solve the above equation for the pressure deviator pkl , we introduce the integral

Iijkl = Ci Cj I[Ck Cl ]dc = A1 ij kl + A2 (ik jl + il jk ), (142)

which was written in terms of Kroneckers symbol with coecients A1 and A2 . If we

contract the integral Iijkl in two dierent manners, namely, Irrss = 9A1 + 6A2 and Irsrs =
3A1 + 12A2 we get that
3Irsrs Irrss
Iijkl = ki jl . (143)
15

RTO-EN-AVT-194 9 - 23
 6 CHAPMAN-ENSKOG-GRAD COMBINED METHOD

We use now the relationship (143) to write (141) as

p vi 2 2 2 2 (0,0)
2 = pij Cr Cs I[Cr Cs ]dc = pij , (144)
m xj 5 5

where (0,0) is given by (65). Hence, it follows from (144) Navier-Stokes law

vi 5 mkT 1
pij = 2 , where = , (145)
xj 16 (2,2)
is the coecient of shear viscosity.
We determine the heat ux vector by following the same methodology. We multiply
(140) by C 2 Ci and the integrate of the resulting equation over all values of c, resulting
[( ) ]
kp T 8 2 5 8 2 (1,1)
5 2 = qk C Ci I
2
C
2
Ck dc = qi , (146)
m xi 5 2 15
thanks to the following relationship
[( ) ]
5 Irr
Iik = C Ci I
2
C
2
Ck dc = ik . (147)
2 3
Hence, Fouriers law follows from (146)

T 75 k mkT 1
qi = , where = , (148)
xi 64 m (2,2)
is the coecient of thermal conductivity. Above, we have used the relationship (1,1) =
(0,0) which was obtained in the Chapman-Enskog method.
We can extended this method to obtain the successive approximations to the transport
(s) (s)
coecients, by introducing traceless second order tensors pij and vectors qi dened by

(s) (2s)!! (s) (s) 15(2s)!! (s)
pij = 15 mS 5 Ci Cj f dc, qi = mS 3 Ci f dc, (149)
(2s + 5)!! 2 4(2s + 3)!! 2

(0) (0)
where pij = pij represents the pressure deviator and qi = qi the heat ux vector,
(s) (s)
while the the higher-order moments pij (s
> 0) and qi (s> 1) do not have standard
designations.
(s) (s)
For the moments of the distribution function characterized by , vi , T , pij and qi
the distribution function reads
{ }
2
2

2 (s) (s) 8 (s) (s)
f = f (0) 1 + S 5 Ci Cj pij S 3 Ci qi , (150)
s=0 2 5 s=0 2

so that (140) becomes

{ }
2 2 (s) [ (s) ]

1 (1) T (0) vk
f (0)
S 3 Ck + 2S 5 Ck Cl = p I S 5 Ck Cl
T 2 xk 2 xl s=0 kl 2

8 2 (s) [ (s) ]

q I S 3 Ck . (151)
5 s=0 k 2

9 - 24 RTO-EN-AVT-194
 7 FOURTEEN MOMENT THEORY FOR GRANULAR GASES

(s)
The multiplication of (151) by S 5 Ci Cj and integration of the resulting equation
2
leads to
5 2 (0,r) vi
(s)
3
= (r,s) pij . (152)
2 m xj s=0

If we solve the above equation for the pressure deviator pij it follows the Navier-Stokes
law
vi 5 2 ( 1 )(0,0)
pij = 2 []p , where []p = (153)
xj 4 m 3

is the coecient of shear viscosity in the pth successive approximation. In (153)2 ( 1 )

(0,0)

represents the rst row and rst column of the inverse of the p p matrix (r,s) .
(s)
If we multiply (151) by S 3 Ci and integrate the resulting equation we get
2

75 n2 k (1,r) T
(s)
= (r,s) qi , (154)
16 xi s=0

which can be solved for the heat ux vector qi , yielding Fouriers law:

T 75 n2 k ( 1 )(1,1)
qi = []p , where []p = . (155)
xi 16

The coecient of thermal conductivity []p in the pth successive approximation is given
in terms of the element (1 )
(1,1)
of the inverse matrix.
The successive approximations of the transport coecients of shear viscosity (153)2
and thermal conductivity (155)2 are the same as those which follow from the Chapman-
Enskog method.

7 Fourteen Moment Theory for Granular Gases

We may characterize a macroscopic state of the granular gas by the fourteen elds of mass
density , hydrodynamic velocity vi , pressure tensor pij , heat ux vector qi and contracted
fourth moment piijj which are dened by

= mf dc, vi = mci f dc, pij = mCi Cj f dc, (156)

m 2
qi = C Ci f dc, piijj = mC 4 f dc. (157)
2

The balance equations for the fourteen elds are obtained from the Boltzmann equation
(78) by multiplying it successively by m, mci , mCi Cj , mC 2 Ci /2 and mC 4 and integrating

RTO-EN-AVT-194 9 - 25
 7 FOURTEEN MOMENT THEORY FOR GRANULAR GASES

the resulting equations. Hence, it follows

vi
+ = 0, (158)
t xi
vi
+ (vi vj + pij ) = 0, (159)
t xj
pij vj vi
+ (pijk + pij vk ) + pki + pkj = Pij , (160)
t xk xk xk
qi vj vi pki pkj 1 prr pij
+ (qij + qi vj ) + pijk + qj = Qi , (161)
t xj xk xj xj 2 xj
piijj vi 8 pik
+ (piijjk + piijj vj ) + 8qik qi = P. (162)
t xj xk xk
In the above equations the new moments of the distribution function are dened by

m 2
pijk = mCi Cj Ck f dc, qij = C Ci Cj f dc, piijjk = mC 4 Ck f dc, (163)
2
where qii = piijj /2. Furthermore, the production terms read
( )
1
Pij = m Ci1 Cj1 + Ci Cj Ci1 Cj1 Ci Cj f1 f d2 (g k) dk dc1 dc, (164)
2

1 m ( 2 1 2
)
Qi = C1 Ci + C Ci C1 Ci C Ci f1 f d2 (g k) dk dc1 dc,
2 1 2
(165)
2 2

1 ( )
P = m C14 + C 4 C14 C 4 f1 f d2 (g k) dk dc1 dc. (166)
2
We can decompose the balance equation for the pressure tensor (160) in its trace and
traceless parts by introducing the traceless tensors pij and pijk dened by
2
pij = pij + pij , pijk = pijk + (qi jk + qj ik + qk ij ) , (167)
5
where p = nkT is the hydrostatic pressure. Hence, we obtain from (160) the equations:
( )
T T 2 qi vi vi
+ vi + +p + pij + T = 0, (168)
t xi 3nk xi xi xj
pij ( ) vj vi 2 vr
+ pijk + pij vk + pki + pkj pkr ij
t xk xk xk 3 xk
4 qi vi
+ + 2p = Pij . (169)
5 xj xj

Equation (168) is the balance equation for the temperature with denoting the cooling
rate (81), while (169) is the balance equation for the pressure deviator.
We represent the non-equilibrium distribution function for the fourteen moments as
( )
f = fM a + ai Ci + aij Ci Cj + bi C 2 Ci + bC 4 , (170)

where the fourteen coecients a, ai , aij , bi and b do not depend on the peculiar velocity.

9 - 26 RTO-EN-AVT-194
 7 FOURTEEN MOMENT THEORY FOR GRANULAR GASES

At equilibrium the contracted fourth moment reads

( )2
(0) 4 kT
piijj = mC fM dc = 15 , (171)
m

and it is convenient to introduce a dimensionless non-equilibrium part of piijj , denoted by

and dened by

1 ( m )2 [ (0)
] 1 ( m )2
= piijj piijj = mC 4 (f fM )dc. (172)
15 kT 15 kT

The fourteen coecients a, ai , aij , bi , b are determined from the denition of the basic
elds (156), (157) and (172) together with the non-equilibrium distribution function (170),
yielding
{ ( ) ( ) }
2 2 8 2 5 15 5C 2
2 4
C
f = fM 1 + pij Ci Cj + qi Ci C 2 + + . (173)
5 2 8 2 2

We can now determine the constitutive equations for the moments of the distribution
function by inserting (173) into the expressions (163) and subsequent integration of the
resulting equations. Hence, it follows
( )2
kT 5 kT 7 kT
pijk = 0, piijjk = 28 qk , qij = [1 + ] ij + pij . (174)
m 2 m 2m

The constitutive equations for the cooling rate and for the production terms Pij , Qi , P
are obtained in the same manner and it follows from (81), (164), (165) and (166) through
integration that
[ ]
1 T 3 1 T
= (1 ) 1 +
2
, Pij = (1 + )(3 )pij , (175)
3 T0 16 5 T0

1 T
Qi = (1 + ) [49 33] qi , (176)
60 T0
( )2 { }
kT T 302 (1 ) + 271 207
P = (1 )(2 + 9)
2 2
1+ . (177)
m T0 (22 + 9)(1 ) 16

We have introduced in the above

equations a mean free time in terms of a reference
temperature T0 , namely, = kT0 /4nd2 . Furthermore, we have considered in the above
m

expressions only linear terms in . For the elastic case = 1 and the production terms
(175) (177) reduce to
( )2
4 T 8 T 8 kT T
= 0, Pij = pij , Qi = qi , P = . (178)
5 T0 15 T0 m T0

From the knowledge of the constitutive equations in terms of the basic elds, we get
by inserting (174) (177) into the balance equations (158), (159), (161), (162), (168)

RTO-EN-AVT-194 9 - 27
 7 FOURTEEN MOMENT THEORY FOR GRANULAR GASES

and (169) the following linearized system of fourteen eld equations for , vi , T , pij

qi and :
vi p pij
D + = 0, Dvi + + = 0, (179)
xi xi xj
( ) [ ]
2 qi vi T T 3
DT + +p + (1 ) 1 +
2
= 0, (180)
3nk xi xi 3 T0 16

4 qi vi 1 T
Dpij + + 2p = (1 + )(3 )pij , (181)
5 xj xj 5 T0
( )
kT pij 5(kT )2 n 5k 2 T n T
Dqi + + n + +
m xj 2m xi xi 2m xi

1 T
= (1 + ) [49 33] qi , (182)
60 T0
( )2 ( )2 {
kT 8kT qj kT T
15 D + = (1 + ) (1 22 )(1 )
m m xj m T0
}
[ ]
81 17 + 302 (1 ) . (183)
16

In (183) we have used (179) and (180) in order to eliminate the time derivatives of and
T and neglected products of .
Now we search for spatially homogeneous solutions of the fourteen eld equations
where the elds depend only on time. The eld equations (179) (183) in this case read
d dvi
= 0, = 0, (184)
dt dt
3 [ ]
dT T2 3
+ (1 ) 1 +
2
= 0, (185)
dt 3 16
dpij (1 + )(3 )
= T pij , (186)
dt 5
dqi (1 + )
= T [49 33] qi , (187)
dt 60
{ }
d 1 +
= T (1 2 )(1 ) [81 17 + 30 (1 )]
2 2
. (188)
dt 15 16

Above, the following dimensionless quantities were introduced: time t = t/ , temperature

T = T /T0 , pressure deviator pij and heat ux vector qi .
We may infer from (184) that the mass density and the velocity elds remain constant
in time, while (185) (188) compose a system of coupled dierential equations for the
determination of the temperature T , pressure deviator pij , heat ux vector qi and
fourth moment . Since this system of equations is non-linear, it was solved numerically
by considering the initial conditions T (0) = 1, pij (0) = 1, qi (0) = 1 and (0) = 1. The
left frame of Figure 3 shows the time decay of the temperature (solid line) in comparison
with Has law (dashed line) which is the solution of (185) when = constant (see (190)
below). We may conclude that the temperature decay T follows closely Has law and by

9 - 28 RTO-EN-AVT-194
 7 FOURTEEN MOMENT THEORY FOR GRANULAR GASES

increasing the restitution coecient the time decay of the temperature is less accentuated.
Furthermore, from the right frame of the Figure 3 we may observe that the pressure
deviator, the heat ux vector and the dimensionless fourth moment decay also with time
and the trend to equilibrium is more pronounced for the pressure deviator followed by
the heat ux vector and the dimensionless fourth moment. The curves in the right frame
were obtained for the restitution coecient = 0.75. By increasing the value of the
restitution coecient the time decay of the pressure deviator and dimensionless fourth
moment are more accentuated but the one of the heat ux vector is less pronounced, since
it is connected with the transport of energy which is directly aected by the inelasticity.

1 1
T*(t) (t)
0.8 Haffs law 0.8 p*<ij>(t)
q*i (t)
0.6 0.6

0.4 =0.95 0.4

=0.75

0.2 0.2
=0.75

0 0
0 10 20 30 40 50 0 5 10 15 20
t* t*

Figure 3: Left frame: time decay of the temperature. Right frame: time decay of the
pressure deviator, heat ux vector and fourth moment.

From the analysis of the dierential equations (186) we note that the pressure deviator
and the heat ux vector do not evolve with respect to time when the initial conditions for
these elds vanish. However, a vanishing initial condition for implies from (188) that it
could evolve with time. An interesting case is the one where the fourth moment remains
constant in time and which is expressed by the condition that the right-hand side of (188)
must vanish, i.e.,
16(1 )(1 22 )
= = a2 . (189)
302 (1 ) + 81 17
The above expression for is the same as that found for the coecient a2 which follows
from the Chapman-Enskog method (see (96)). We note that vanishes in the elastic case
where = 1.
If we substitute the expression of given by (189) into (186) and the integrate the
resulting equation, we get Has law
1
T (t) = { [ ] }2 . (190)
(12 ) 3(1)(122 )
1+ 6
1+ 8117+302 (1)
t

By considering the elastic case = 1, the temperature T remains constant in time

RTO-EN-AVT-194 9 - 29
 7 FOURTEEN MOMENT THEORY FOR GRANULAR GASES

and
pij (t) = pij (0)e 5 qi (t) = qi (0)e 15
4 8
T t T t
, , (191)
i.e, both decay exponentially with time, which is a well known result.
In the case of a ve eld theory described by the elds of mass density, velocity
and temperature, the pressure deviator and the heat ux vector are no longer elds but
constitutive quantities. The constitutive equations for these quantities are obtained from
the eight remaining eld equations (181) and (182) by inserting the equilibrium values
pij = 0 and qi = 0 on the left-hand side of these equations, yielding the laws of Navier-
Stokes and Fourier
vi T n
pij = 2 , qi = , (192)
xj xi xi
respectively. The transport coecients of shear viscosity , thermal conductivity and
the coecient are given by

5 mkT 1 75k mkT 1
= 2 , = 2 , (193)
4d (1 + )(3 ) 2d m (1 + )(49 33)

75kT mkT 16(1 )(1 22 )
= . (194)
2nd2 m (1 + )(49 33)[302 (1 ) + 81 17]

We note that if we consider a thirteen moment theory the heat ux vector is proportional
only to the temperature gradient and the dependence on the particle number gradient
does not show up. This fact can be understood by observing that this dependence comes
out from the underlined term in (182) which depends exclusively on the dimensionless
fourth moment. Furthermore, the transport coecients (193) and (194) are not the same
as those obtained from the Chapman-Enskog method (104) and (105), and it seems that
the equivalence of the two methods can be attained if we consider more moments of the
distribution function in Grads moment method.

9 - 30 RTO-EN-AVT-194
 REFERENCES

References
The description of the methods of Chapman-Enskog and Grad can be found in many
books, e.g.,

Cercignani, C. (1975). The Boltzmann equation and its applications. Springer, New
York.

Chapman, S. and Cowling, T. G. (1970). The mathematical theory of non- uniform gases
3rd edition. Cambridge University Press, Cambridge.

Grad, H. (1958). Principles of the kinetic theory of gases, in Handbuch der Physik Vol.
XII, pp. 205-294 ed. S. Flugge . Springer, Berlin.

Hirschfelder, J. O., Curtiss, C. F. and Bird, R. B. (1964) Molecular theory of gases and
liquids 2nd edition. John Wiley & Sons, New York.

Kogan, M. N. (1969). Rarefied gas dynamics. Plenum Press, New York.

Struchtrup H. (2005) Macroscopic transport equations for rarefied gas flows. Springer,
Berlin.

Waldmann, L. (1958). Transporterscheinungen in Gasen von mittlerem Druck, in Hand-

buch der Physik Vol. XII, pp. 295-514 ed. S. Flugge. Springer, Heidelberg.

Granular gases are treated in the book

Brilliantov, N. V. and Poschel, T. (2004). Kinetic theory of granular gases. Oxford

University Press, Oxford.

In these notes I followed the book

Kremer, G. M. (2010) An introduction to the Boltzmann equation and transport processes

in gases. Springer, Berlin.

The section on fourteen moment theory of granular gases was based on

Kremer, G. M. and Marques Jr. W. (2011). Fourteen moment theory for granular gases,
Kinetic and Related Models (in press).

RTO-EN-AVT-194 9 - 31
9 - 32 RTO-EN-AVT-194