Вы находитесь на странице: 1из 40

R M S T A DEL NUOVO CIMENTO

VOL. 18, N. 7

1995

Aerodynamical Applications of the Boltzmann Equation.


C. CERCIGNANI

Dipartimento di Matematica, Politecnico di Milano - Milano, Italy

1
6
7
8
18
24
26
31
33
34

1. -

1.
2.
3.
4.
5.
6.
7.
8.
9.
10.

Introduction.
Continuum vs. molecular models of a gas.
Rarefaction regimes.
Gas-surface interaction.
Polyatomic gases and mixtures.
Chemical reactions and thermal radiation.
Solving the Boltzmann equation. Analytical techniques.
Solving the Boltzmann equation. Numerical techniques.
Vortices and turbulence in a rarefied gas.
Concluding remarks.

Introduction.

This paper deals with an important application of the Boltzmann equation, i.e. the
study of upper-atmosphere flight, which occurs, e.g., in connection with the re-entry
of a space shuttle.
This is one of the fields of modern technology, where the concepts and tools
introduced by Boltzmann are essential. This would have pleased Boltzmann, who was
very much interested in technological advances (he also wrote a paper, in which he
correctly predicted the superiority of airplanes over dirigible airships) and is the
author of the following sentence written in 1902 [1], which nowadays may sound a bit
trivial:

However much science prides itself on the ideal character of its goa~ looking
down somewhat contemptuously on technology and practice, it cannot be denied that
it took its rise from a striving for satisfaction of purely practical needs. Besides, the
victorious campaign of contemporary natural science would never have been so
incomparably brillian~ had not science found in technologists such capable pioneers.
To this we may add another sentence, written in 1905 [1]:

That is why I do not regard technological achievements as unimportant


by-products of natural science but as logical proofs. Had we not attained these
practical achievements, we should not know how to infer. Only those inferences are
correct that lead to practical success.

C. CERCIGNANI

Before entering the main subject of the paper, we want to deal very briefly with
the objections which are frequently still voiced against the Boltzmann equation. This
will also serve the purpose of fLxing some notation.
We start by recalling this famous equation [2-6]:

(1.1)

af +

a/ + x . af

3t

~x

- I I (f 'f ,

- f f , ) B ( n . V , IVI)d~, (ln;

R 3 ~+

f = f(x, ~, t) is the distribution function of the molecules, which can be normalized


in various ways (as a number density, as a probability density, as a mass density).
t denotes time, x the position of a molecule, ~ the velocity of the same molecule,
X a body force acting on each molecule (such as, e.g., gravity). The right-hand side
contains a quadratic expression Q(f, f), given by
(1.2)

Q(f' f)= I I (f'f * -ff,)B(n.V, IVI)d ,d .


R3 S 2

Here B(n. V, IV I ) is a kernel containing the details of the molecular interaction,


f ' , f',, f . are the same as f, except for the fact that the argument ~ is replaced by ~',
~., ~ , , respectively, ~. being an integration variable (having the meaning of the
velocity of a molecule colliding with the molecule of velocity ~, whose path we are
following). ~' and ~. are the velocities of two molecules entering a collision that will
bring them to have velocities ~ and ~., whereas n is a unit vector giving the direction
of the straight line joining the centres of the molecules at the moment of impact and
describes a unit sphere S 2 . The relations between ~', ~ . , on the one hand, and ~ and
~ . , on the other hand, read as follows:
(1.3)

~'= ~ - n[(~ - ~,)-n],

~, = ~, + n[(~ - ~,).n].

We shall denote by G(f, f ) and fLf, respectively, the gain and loss parts of Q(f, f).
Here we have restricted ourselves to monatomic molecules; polyatomic gases will
be considered in sect. 5.
One important property of eq. (1.1) is that the collision integral vanishes when f is
a Maxwellian distribution M:
(1.4)

M = Q(2zRT)-~/2 exp [ - I ~ - v l2/(2RT)],

where the parameters Q, T, and v have the meaning of density, temperature and bulk
velocity of the gas.
If we multiply both sides of eq. (1.1) by log f and integrate with respect to ~, we
obtain

(1.5)

a:~v
a
-+ - - . J = ~f,
at
ax

AERODYNAMICAL APPLICATIONS OF T H E BOLTZMANN EQUATION

where
f

(1.6)

= / f log f d ~ ,
ha

(1.7)

J = f / i f log f d~,
ha

(1.8)

J" = ~ f log
ha

fQ(f, f) d~.

We know, however, that a famous inequality due to Boltzmann applies:


(1.9)

~f ~< 0

and

~f = 0

iff f is a Maxwellian.

Because of this inequality, eq. (1.5) plays an important role in the theory of the
Boltzmann equation. We illustrate the role of eq. (1.9) in the case of spacehomogeneous solutions. In this case the various quantities do not depend on x and
eq. (1.5) reduces to
(1.10)

8~V _ J" ~< O.


8t

This implies the so-called H-theorem (for the space-homogeneous case): ~ is a


decreasing quantity, unless f is a Maxwellian (in which case the time derivative of ~V
is zero). Since in this case the density Q, the momentum density Q$ and the internal
energy e are constant in time, we can build a Maxwellian M which has, at any time,
the same ~), ~ and e as any solution f corresponding to given initial data. Since
decreases unless f is a Maxwellian (i.e.f = M), it is tempting to conclude that f tends
to M when t - ~ oo. The temptation is strengthened when we realize that ~ is
bounded from below by ~ M , the value taken by the functional .~V when f = M. In fact
:~V is decreasing, its derivative is non-positive unless it takes the value ~VM; one feels
that ~ tends to :~VM! This conclusion is, however, unwarranted, without a more
detailed consideration of the source term ~f in eq. (1.10). This is shown in detail in
ref. [4].
If the state of the gas is not space-homogeneous, the situation becomes more
complicated. In this case it is convenient to introduce the quantity
(1.11)

H = f ~ dx,
~2

where ~2 is the space domain occupied by the gas (assumed here to be time
independent). Then eq. (1.5) implies

(1.12)

dH
dt

f J-,,do,
J

where n is the inward normal and da the measure on 8~2. Clearly, several situations
may arise. Among the most typical ones, we quote:
1) ~2 is a box with periodicity boundary conditions (flat toms). Then there is no

C. CERCIGNANI

boundary, dH/dt <. 0 and one can repeat about H what was said about ~ in the
space-homogeneous case. In particular, there is a natural (space-homogeneous)
Maxwellian associated with the total mass, momentum and energy (which are
conserved in this case).
2) ~9 is a compact domain with specular reflection. In this case the boundary
term also disappears because the integrand of J . n is odd on ~t9 and the situation is
similar to that in case 1). There might seem to be a difficulty for the choice of the
natural Maxwellian because momentum is not conserved, but a simple argument
shows that the total momentum must vanish when t--~ ~. Thus M is a non-drifting
Maxwellian, unless the domain has rotational symmetry about an axis. In the latter
case, the angular momentum about this axis is conserved and thus, in general, the
final state will be a (space-inhomogeneous) Maxwellian, with the gas rotating like a
solid body and being centrifuged away from the axis.
3) ~2 is the entire space. Then the asymptotic behaviour of the initial values at
oo is of paramount importance. If the gas is initially more concentrated at finite
distances from the origin, one physically expects and can mathematically prove (see
ref. [51]) that the gas escapes through infinity and the asymptotic state is a vacuum.
4) $2 is a compact domain but the boundary conditions on ~2 are different from
specular reflection (see sect. 4). Then the asymptotic state may be completely
different from a Maxwellian.
Boltzmann's H-theorem is of basic importance because it shows that his equation
has a basic feature of irreversibility: the quantities 9V (in the space-homogeneous
case) and H (in other cases where the gas does not exchange mass and energy with a
solid boundary) always decrease in time. This result seems to be in conflict with the
fact that the molecules constituting the gas follow the laws of classical mechanics
which are time-reversible. Accordingly, given a motion at t = to with velocities ~(1),
~(2), ~(m, we can always consider the motion with velocities -~(1), _ ~(2), _ ~(N)(and
the same positions as before) at t = to ; the backward evolution of the latter state will
be equal to the forward evolution of the original one. Therefore if dH/dt < 0 in the
first case, we shall have d H / d ( - t ) < 0 or dH/dt > 0 in the second case, which
contradicts Boltzmann's H-theorem.
This paradox is mentioned by Thomson in a short paper [7], but it is usually
attributed to Josef Loschmidt, who mentioned it briefly in the first [8] of four articles
devoted to the thermal equilibrium of a system of bodies subject to gravitational
forces.
Nowadays we are more prepared to discuss this kind of questions. In fact, we
remark that, when giving a justification of the Boltzmann equation [3, 4], we use the
laws of elastic collisions and the continuity of the probability density at the impact to
express the distribution functions corresponding to an after-collision state in terms of
the distribution functions corresponding to the state before the collision, rather than
the latter in terms of the former. It is obvious that the first way is the right one to
follow if the equations are to be used to predict the future from the past and not vice
versa; it is clear, however, that this choice introduced a connection with the everyday
concepts of past and future which are extraneous to molecular dynamics and are
based on our macroscopic experience. A striking consequence of our choice is that the
Boltzmann equation describes motions for which the quantity H (or 9V) has a
tendency to increase, while the opposite choice would have led to an equation having a

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

negative sign in front of the collision term, and hence describing only motions with
increasing H. We must remark that, in order to derive the Boltzmann equation, one
takes special (although highly probable) initial data; thus certain special data are
excluded. As the discussion in ref. [3, 4] shows, these excluded data correspond to a
state in which the molecular velocities of the molecules that are about to collide show
an unusual correlation. This situation can be simulated by studying the dynamics of
many interacting particles on a computer and leads to an evolution in which there is
an increasing H, as expected, while ,,randomly- chosen initial data invariably lead to
an evolution with decreasing H [3, 4]. In other words, the fact that H decreases is not
an intrinsic property of the dynamical system but a property of the level of
description.
It is not the place here to discuss the relation of the H-theorem with the notions of
past and future [2,9].
It is not, however, perhaps out of place to comment on a statement which is
frequently made, to the effect that no kind of irreversibility can follow by correct
mathematics from the analytical dynamics of a conservative system and hence some
assumption of kinetic theory must contradict analytical dynamics. It should be clear
that it is not a new assumption that is introduced, but the fact that we study
asymptotic properties of a conservative system in the so-called Boltzmann-Grad limit
(see below), under the assumption that the initial probability distributions are
restricted to a special (but highly probable) set [3, 4].
There is another objection which can be raised against the H-theorem when
presented as a rigorous consequence of the laws of dynamics. The starting point is a
theorem of Poincar~ [10] (the so-called ,,recurrence theorem,), which says that any
conservative system, whose possible states form a compact set in phase space, will
return arbitrarily close to its initial state, for almost any choice of the latter, provided
we wait long enough. This applies to a gas of hard-sphere molecules, enclosed in a
specularly reflecting box, because the set of the possible states S with a given energy
is compact and has a finite measure/x(S) (induced by the Lebesgue measure).
This theorem implies that our molecules can have, after a ,,recurrence time,,
positions and velocities so close to the initial ones that the one-particle distribution
function f would be practically the same; therefore, H should also be practically the
same, and if it decreased initially, then it must have increased at some later time. This
paradox goes under the name of Zermelo, who stated it in 1896 [11], but was actually
mentioned before in a short paper by Poincar~ [12]. The traditional answer to
Zermelo's paradox was given by Boltzmann himself[13]: the recurrence time is so
large that, practically speaking, one would never observe a significant portion of the
recurrence cycle. In fact, according to an estimate made by Boltzmann [13], the
recurrence time for a typical amount of gas is a huge number even if the estimated
age of the Universe is taken as time unit.
Nowadays we know that we can claim validity for the Boltzmann equation in the
Boltzmann-Grad limit only (when the number of molecules goes to infinity and the
sphere diameter a goes to zero in such a way that N a 2 is finite), we do not have
to worry about the recurrence paradox; in fact, the set S is no longer compact when
N --~ oo and the recurrence time is expected to go to infinity with N (at a much faster
rate).
The Boltzmann equation was written by Boltzmann in 1872 and has become a
practical tool for upper-atmosphere aerodynamics in the last twenty years. In order
to be useful for applications it must, of course, be matched with appropriate initial and

c. CERCIGNANI

boundary conditions. While the initial data are specified by assigning the initial
distribution, the boundary conditions are more complicated [2-4]. A discussion of this
aspect of rarefied-gas dynamics will be given, as indicated above, in sect. 4.
2. - C o n t i n u u m

vs.

molecular

models

of a gas.

In this section we shall explain why, when dealing with upper-atmosphere flight,
one may easily be led to questioning the validity of the Euler and Navier-Stokes
equations as an accurate description of the fluid dynamics of the atmosphere. In fact,
as we saw in the previous section, according to the kinetic theory of gases, the basic
description of the state of a gas is in terms of the distribution function, a function of
position x, velocity ~ and time t, which gives the probability density of finding a
molecule at position x with velocity ~ at time t. In terms of this non-negative function,
f = f ( x , ~, t), one can define the typical quantities used in macroscopic fluid dynamics,
such as density Q, bulk velocity v, stress components p~j, pressure p, temperature T,
heat flow q, in the following way:

Q(x, t): If(.,:, t)d ,

Qv(x,t):

p~/(x, t) = I ci cj f(x, ~, t) d~

f(x,

t)d+,
(c~ = ~i - v i ) ,

(2.1)

p(x, t) = QRT = (1/3) I [c[2f(x, ~, t ) d ~ ,


q(x, t) = (1/2) f clcl2f(x, ~, t)d~

(c= ~ - v ) .

(here the subscripts range from 1 to 3 and R is the gas constant, equal to k/m, k being
the Boltzmann constant and m the molecular mass).
In order to see that the Navier-Stokes equations become invalid on a sufficiently
small scale, it is sufficient to remark that the definition of the components of the
stress tensor in eq. (2.1) implies (because of the elementary inequality 21abl <~a 2 + b 2
and the fact that f cannot be negative):
(2.2)

IP~j I <~ (P~i + PjN)/2 ~< 3p/2.

In order to appreciate the practical consequences of this inequality, let us consider a


simple shear flow, such as plane Couette flow. In this case the significant component
of the stress in the Navier-Stokes equations is
(2.3)

P12 = -/~ 3u/~y,

where u = vl is the x-component of the velocity vector v and tt the viscosity


coefficient. Equation (2.2) then gives
(2.4)

I/~ ~u/~y I <~3p//2.

The significance of (2.4) is clear if we introduce the mean free path ~ related,
according to elementary kinetic theory, to /~ in the following way:
(2.5)

;t = ~(2RT) 1/2/p,

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

so that eq. (2.4) becomes


[Su/3y I ~< 3(2RT)1/2/,~..

(2.6)

In order to appreciate this simple result, we remark that, in the Earth's


atmosphere, the mean free path )~ is about 1 metre at about 130 km. More typical
rarefactions are met in the study of the flight of aircrafts transporting men or robots
for the planned exploration of other planets, such as Mars or Venus. In any case,
eq. (6) indicates that for rarefied gases and/or high speeds, one cannot rely on the
usual Navier-Stokes equations for a compressible fluid and must resort to kinetic
theory.
It is to be remarked that rarefied gases appear in many other areas of technology:
vacuum technology, micromachines, dynamics of aerosols, etc. All these problems
have in common the fact that the mean free path is not negligible with respect to some
other characteristic length.
3. - Rarefaction regimes.
In order to orient ourselves in a new field and chart the phenomena that can be
expected, it is very useful to consider the typical non-dimensional parameters
associated with the equations which rule these phenomena. Rarefied-gas dynamics is
no exception: it makes use of two basic non-dimensional numbers, the Knudsen
number
(3.1)

in

;/L,

where L is a characteristic length, such as the diameter of a pipe, the thickness of the
viscous boundary layer, the curvature radius of the nose of an airfoil, or t h e
wavelength of a sound wave, and the molecular speed ratio
(32)

S =

u/C,

where C is the thermal speed V2R-T.


Kn and S are related to the Mach and Reynolds numbers, Ma and Re, in the
following way:
(3.3)

Ma = V2V2V2VS;
~

Kn = ~ f ~ M a / R e .

This indicates that one might use the same parameters as in the usual continuum
theory, based on the Navier-Stokes equations. For rarefied gases, however, the
Knudsen number is more handy than the Reynolds number. Of course, one can
consider several Knudsen numbers, based on different characteristic lengths, exactly
as one does for the Reynolds number. Thus, in the flow past a body, there are two
important macroscopic lengths: the local radius of curvature and the thickness of the
boundary layer 5, and one can consider Knudsen numbers based on either length.
Usually the second one, (Kna = k/a), gives the most severe restriction to the use of
Navier-Stokes equations; when Kn~ > (say) 0.01, the presence of a thin layer near the
wall, of thickness - )~ (Knudsen layer), influences the viscous layer in a significant
way. This regime is called the slip regime because the gas slips upon the boundary
with a velocity us and, at the boundary, has a temperature different from the

C. CERCIGNANI

temperature of the boundary itself. The velocity slip is given by


(3.4)

us =

~(~u/Sn)w,

where ~ is the slip coefficient. A similar formula holds for the temperature jump.
Maxwell computed ~ to be approximately equal to (V~/2)s (for complete diffusion of
the molecules at the wall see below). Actually ~ is about 15% larger [2,3].
This and other regimes to be described below are met in high-altitude flight; in
particular, they are all met by a shuttle when returning to the Earth.
When the mean free path increases, one witnesses a thickening of the bow shock
wave which forms in front of a vehicle travelling at a supersonic speed in a gas. This
thickness is of the order of 6)[ and eventually the shock merges with the viscous layer;
that is why this regime is sometimes called the merged-layer regime. Another
frequently used name is transition regime.
When Kn is large (few collisions), phenomena related to gas-surface interaction
play an important role. One distinguishes between free-molecule and nearly freemolecule regimes. In the first case the collisions between molecules are completely
neglegible, while in the second they can be treated as a perturbation.
4. -

Gas-surface

interaction.

The problem of gas-surface interaction is basic in rarefied-gas dynamics, since it is


to this interaction that one can trace back the origin of the drag and lift exerted by the
gas on the body and the heat transfer between the gas and a solid boundary. This has
its mathematical counterpart in the fact that, if we want to describe a physical
situation where a gas flows past a solid body or is contained in a region bounded by
one or more solid bodies, the Boltzmann equation must be accompanied by boundary
conditions, which describe the above-mentioned interaction of the gas molecules with
the solid walls. Hence, in order to write down the correct boundary conditions for the
Boltzmann equation, we must possess information which stems from a discipline
which may be regarded as a bridge between the kinetic theory of gases and
solid-state physics.
The difficulties of a theoretical investigation are due, mainly, to our lack of
knowledge of the structure of surface layers of solid bodies and hence of the effective
interaction potential of the gas molecules with the wall. When a molecule impinges
upon a surface, it is adsorbed and may form chemical bonds, dissociate, become
ionized or displace surface molecules. Its interaction with the solid surface depends on
the surface finish, the cleanliness of the surface, its temperature, etc. It may also vary
with time because of outgassing from the surface. Preliminary heating of a surface
also promotes purification of the surface through emission of adsorbed molecules. In
general, adsorbed layers may be present; in this case, the interaction of a given
molecule with the surface may also depend on the distribution of molecules impinging
on a surface element.
The first observations of the interaction of gases with solid surfaces (apart from
early studies which arrived at the conclusion that the gas does not slip on the wall
under standard conditions) are due to Kundt and Warburg [14]. They noted that the
flow rates through tubes at very low pressures are appreciably higher than predicted
by the familiar Poiseuille formula and attributed that effect to slip at the boundary.
Maxwell [15] suggested that slip would be a consequence of kinetic theory and

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

computed the amount of slip for a particular model of the gas-surface interaction, still
very popular and known under his name. Smoluchowski [16] described temperature
jumps in a similar way. A more systematic effort started with Knudsen[17].
Full-scale research, however, only began about thirty years ago, under the impetus of
space flight and thanks to the developments in high-vacuum technology.
We shall try here to give an idea of the physical models and computations which
aim at simulating the complex phenomena to which we have briefly alluded, following
the surveys of Ku~Ser [18] and Cercignani [19].
The possible events, that a gas molecule hitting a solid (or liquid) surface may
experience, have been briefly described above. A more systematic classification may
read as follows:
1) Scattering.
specular reflection;
Elastic scattering (zero-phonon scattering) ~diffraction into discrete peaks;
/

[diffuse elastic scattering.


-phonon scattering;
Inelastic scattering [multi-phonon scattering.

Ione

Sputtering and other high-energy effects.


2) Adsorption.
mporary;
Adsorption at specific sites [permanent.

Ite

Mobile adsorption (surface diffusion).


Diffusion into the condensed phase.
3) Condensation (in case of identical composition of gas and condensed phase).
4) Reactive interaction.
Chemical reactions.
Ionization.
Elastic scattering is strong only if the surface behaves as it were almost rigid; this
may occur if both the energy of the gas molecule and the thermal energy of the wall
are comparable to or below kB TD (where kB is the Boltzmann constant and TD the
Debye temperature). In this case specular reflection can occur only if the equipotential surfaces are approximately flat on the molecular scale. Thermal motions in
the solid reduce elastic scattering, till, if the molecule kinetic energy is several times
in excess of kB TD, incident molecules become capable of ejecting atoms or whole
chunks of the solid (,~sputtering,).
Gas molecules with strong chemical affinity to the solid become adsorbed, i.e.
attached to the surface. Even in the absence of such affinity, attachment can occur at
sufficiently low temperatures, because of van der Waals forces. Although the
difference is mainly in the magnitude of the binding energy, one usually distinguishes
the two cases by means of the names chemisorption and physisorption. Any
adsorption is, in principle, temporary [20], but for strong binding and low surface

10

c. CERCIGNANI

temperatures the sitting time may well exceed the duration of any experiment or
even the age of the Universe; in this case one talks about permanent adsorption.
Heating strongly accelerates the desorption process. Strongly adsorbed molecules are
usually localized at particular adsorption sites, but even a face of an ideal crystal
offers several kinds of such sites. Polycrystals and amorphous solids are expected to
be far richer in this respect. The behaviour of the adsorbed substance very much
depends on the fraction of adsorption sites actually occupied. A complete layer looks
like a part of the crystal and has a similar if not the same symmetry. Above it a second
layer can sometimes be adsorbed.
The exchange of matter in condensation and evaporation is analogous to
adsorption and desorption, except for the identical composition of both phases. This
has the important consequence that the surface remains unchanged.
Chemical reactions may occur at the wall; e.g., the adsorbate can dissociate during
adsorption, or, if the gaseous phase is a mixture, the wall can catalyze reactions that
would not occur in the gas. Analogously one may have ionization of the adsorbate.
In general, a molecule striking a surface with a velocity ~' re-emerges from it with
a velocity ~ which is strictly determined only if the path of the molecule within the
wall can be computed exactly. This computation is very hard, because it depends upon
a great number of details, such as the locations and velocities of all the molecules of
the wall and an accurate knowledge of the interaction potential (see below). In fact a
sufficiently accurate calculation should be able to predict all the phenomena that we
have briefly described above.
Hence it is more convenient to think in terms of a probability density R(~'--~ $;
x, t; r) that a molecule striking the surface with velocity between ~' and ~' + d~' at
the point x and time t will re-emerge at practically the same point with velocity
between $ and ~ + d$ after a time interval v (adsorption or sitting time). If R is
known, then we can easily write down the boundary condition for the distribution
function f(x, ~, t). To simplify the discussion, the gas will be presently assumed to be
monatomic and the surface at rest.
The mass (or number, depending on normalization) of molecules emerging with
velocity between ~ and ~ + d~ from a surface element dA about x in the time interval
between t and t + dt is
(4.1)

d * ~ = f ( x , ~, t ) l ~ . n l d t d A d ~

(x E ~Y2, ~'n > 0),

where n is the unit vector normal to the surface ~t2 at x and directed from the wall
into the gas. Analogously, the probability that a molecule impinges upon the element
dA with velocity between ~' and ~ ' + d~' in the time interval between t - r and
t-r+dt
( r > 0 ) is
(4.2)

d * ~ ' = f ( x , ~', t - r ) l ~ ' . n ] d t d A d ~'

(x 9 ~Y2, ~'.n < 0).

If we multiply d'A//' by the probability of a scattering event from velocity ~' to a


velocity between ~ and ~ + d~ with an adsorption time between r and r + dr (i.e.
R(~'--~ ~; x, t; r)d~ dr) and integrate over all the possible values of ~' and r, we must
obtain d * ~ (here, for simplicity sake, we assume that each molecule re-emerges from
the surface element into which it entered, which is not so realistic when r is large):
cc

(4.3)

d*A'/=d~Idr
0

I
~<0

R(~'---~;x,t;v)d*.'r

(x 9

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

11

Equating the expressions in eqs. (4.1) and (4.3) and cancelling the common factor
dA d$ dt, we obtain
(4.4)

f(x, ,t)l 'nl = i f

R($'-o $; x, t; 7) f(x, ~', t - 7)[ ~' .nI d$'

o ~<o

(x 9 ~t2, ~.n > 0).


If ~ is an average adsorption time, ~ the average normal velocity with which the gas
molecules impinge upon the surface and n the number density of the gas, n ~ dA
molecules will impinge, per unit time, on a surface element of area dA and stay there
an average time ~ ; ff a0 is an effective range of the gas-surface interaction, each
molecule will occupy an area of the order of a~ and the total area occupied by adsorbed
molecules will be n~-~a~ dA, i.e. a fraction n ~ a ~ of the surface will be occupied. This
is, of course, just a rough order-of-magnitude argument, since the molecules may
penetrate somewhat into the solid and not necessarily remain at its surface.
If na~ ~ is not close to zero, the nature of the interaction of each incident molecule
depends on the total number and energy of the incident molecules. Under conditions
of extremely low density (as, for example, in the case of an orbiting satellite)
na~V~<< 1 and each incident molecule interacts with the surface independently of the
others. The same independence may show up in the other limiting case nao2~ ~ 1 (for
example, in chemisorption when ~ may be very large, or in a dense gas when n can be
extremely large); in this case the impinging molecule interacts directly with the
adsorbed layer rather than with the atoms of the solid (and the effective adsorption
time for that interaction can be much shorter than ~).
Whenever the effective adsorption time ~e and the number density are such that
na2o~e << 1, scattering events can safely be assumed instantaneous and statistically
independent and the kernel R(~'--) ~; x, t; 7) may be taken to be independent of the
distribution function f(x, ~, t); hence we can compute R(~'--) ~; x, t; 7) under the
assumption that one gas molecule of given velocity $' impinges upon the wall. We
shall always assume that we deal with the situation that we have just described; then
the kernel R ( $ ' - * $; x, t; 7) provides, a complete description of the surface scattering
law.
If, in addition ve is small compared to any characteristic time of interest in the
evolution off, we can let 7 = 0 in the argument o f f appearing in the right-hand side of
eq. (4.4), which then becomes:
(4.5)

f(x, ~,t)lv'n[ = IR($'---)~;x,t)f(x, ~',t)l~"nl~'

(xe~D, ~'n

>0),

where
ov

(4.6)

R($'---) ~; x, t) = I dTR($'--~ $; x, t; v).


o

Equation (4.5) is, in particular, valid for steady problems.


Although the idea of a scattering kernel had appeared before, it is only at the end
of the 1960s that a systematic study of the properties of this kernel appears [21, 22, 2].
In particular, the following properties were pointed out [21-26,2-4]:

12

C. CERCIGNANI

1) Non-negativeness, i.e. R cannot take negative values:

R(~'--+ ~; x, t; ~) >>-O

(4.7)
and, as a consequence,

R(~'--~ ~; x, t) >~O.

(4.8)

2) Normalization, if permanent adsorption is excluded; i.e. R, as a probability


density for the totality of events, must integrate to unity:
~c

(4.9)

fdr
0

f R(~'--+~;x,t;r)d~=l
~,, > 0

and, as a consequence,

f R(~'--+~;x,t)d~= l.

(4.10)

3) Reciprocity; this is a subtler property related to the fact that the microscopic
dynamics is time-reversible and the wall is assumed to be in a local equilibrium state,
not significantly disturbed by the impinging molecule. It reads as follows:
(4.11)

I~'.njMw(~')R(~'--~;x,t;r)=l~'nIMw(~)R(-~-*-~';x,t;~)

and, as a consequence,
(4.12)

I~'.nlMw(~')R(~'--~;x,t)=I~'n]Mw(~)R(-~-~-~';x,t).

Here Mw is a (non-drifting) Maxwellian distribution having the temperature of the


wall, which is uniquely identified apart from a factor.
We remark that the reciprocity normalization relations imply another property:
3') Preservation of equilibrium, i.e. the Maxwellian M~. must satisfy the
boundary condition (4.4):
OC

(4.13)

M w ( ~ ) l ~ ' n l = I d~ I R(~'--*~;x't;r)Mw(~')l~"nld~'
0

~<o

equivalent to
(4.14)

Mw(~)lv'nl = I R(~'o~;x't)Mw(~')l~"nld~"
~<0

In order to obtain eq. (4.13) it is sufficient to integrate eq. (4.11) with respect to ~' and
v, taking into account eq. (4.9) (with - ~ and - ~ ' in place of ~' and ~, respectively).
We remark that frequently, one assumes eq. (4.13) (or (4.14)), without mentioning
eq. (4.11) (or (4.12)); although this is enough for many purposes, reciprocity is very
important when constructing mathematical models, because it places a strong
restriction on the possible choices. A detailed discussion of the physical conditions
under which reciprocity holds has been given by B~irwinkel and Schippers [27].
The scattering kernel is a fundamental concept in gas-surface interaction, by
means of which other quantities should be defined. Frequently its use is avoided by

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

13

using the so-called accommodation coefficients (a.c.'s), with the consequence of lack of
clarity, misinterpretation of experiments, bad definitions of terms and misunderstanding of concepts. The basic information on gas-surface interaction, which should
in principle be obtained from a detailed calculation based on a physical model as
discussed in the next section, is summarized in a scattering kernel. The further
reduction to a small set of a.c.'s can be advocated for practical aims, provided this use
is firmly related to the scattering kernel, as indicated, e.g., in two previously quoted
books [3, 4].
The concepts of scattering kernel and a.c. are very useful schemes, where to fit the
information arising from a study of the complicated physical phenomena associated
with the interaction of the molecules of a gas with the atoms of a wall. This study is by
necessity difficult, since we are dealing with a many-body problem. Any theoretical
model leading to more than purely formal results turns out to be either a crude one or
a rough approximation, with the exception of special situations. The support that may
come from experimental data is sometimes deceptive, because these data are scanty.
A systematic review of existing models and approximations prior to 1976 may be
found in the book by Goodman and Wachman [28]. Here we only sketch a few
attempts and indicate a few recent developments.
An early attempt is due to Baule [29], who considered a model of the solid in the
form of an array of smooth hard spheres, exercising independent thermal motions;
later Logan et al. [30] replaced the spheres with hard cubes. Few examples are simple
enough to yield to a closed-form solution even in the case of a single-scattering event.
The scattering kernel for this model has been evaluated by B~irwinkel and
Schmidt[31]. We remark that the kernels obtained from single-scattering events
must be cut off in order to discard the molecules still moving inwards after the
interaction. Hence they are not normalized and violate reciprocity, because of the
arbitrariness in their way of dealing with multiple-scattering events.
Some of the difficulties are avoided by the soft-cube model, proposed by Logan
and Keck [32]. In this model each cube carries a rather fiat potential well, with
the purpose of explaining scattering events at energies of the order of the thermal
energy without having to deal with multiple scattering, which is now insignificant.
The typical lobular patterns in the angular distributions of scattered molecules, which
are seen in experiments, are well reproduced [28]. The model also explains adsorption [32], but does not account for tangential momentum exchange.
A realistic picture can only be obtained by taking an entire piece of solid wall and
simulating the dynamics on a computer[33,28], through the so-called trajectory
analysis. Recent examples of this kind of calculations will be briefly discussed.
At higher energies the impinging molecule begins to feel the surface waviness,
which results into a modified scattering pattern (,,structure scattering,,). A
conspicuous pattern shows two peaks near certain limiting angles (the so-called
,,rainbow effect,, [28]).
Models capable of dealing with multiple scattering were proposed by this
author [25, 3] but have received little if any attention. They are based on the idea of
using a transport equation for the molecule inside the solid, which is regarded as a
half-space. Then one has to solve the following equation:

(4.15)

~ . - 3P
- ~ + X ( x ) . 3P
~ = LP ,

14

C. CERCIGNANI

where P is the probability density of finding the molecule in the half-space which
simulates the wall (say xl < 0) at position x with velocity ~. X is the many-body force
exerted on the molecule by the solid atoms and L a linear operator (of the Boltzmann
and/or Fokker-Planck type) which describes the shorter-range interactions. This type
of equation is not so easy to deal with, except in artificial but interesting cases; in fact
by simulating the many-body force by a rigid surface and using collision frequencies
(in the Boltzmann-like term) or diffusion coefficients (in the Fokker-Planck term)
proportional to vl, it is possible to recover the Maxwell model and the CercignaniLampis model (to be discussed in the next section) for R(~'--~ ~). In fact, once
eq. (4.15) is solved with the boundary condition
(4.16)

P = ~(~ - ~'),

xl = 0,

~1 = ~.n < 0,

then R(~'--~ ~) is immediately recovered by

I 'nl
I "nl

R(~'-~ ~) - - - [ P ( x ,

(4.17)

~)]xl ~0.

The idea of using a statistical description of the motion of a molecule inside the wall
appears in a current approach known as the three-dimensional generalized Langevin
model [34, 35]. Details of the gas-surface interaction are computed as in a trajectory
simulation, but the surface atoms are coupled to the lattice ones through two
additional terms in the equation of motion: one of them accounts for the dissipation in
the lattice and the second, a fluctuating force, accounts for the effects of lattice
vibrations on surface atoms. The lattice atoms, in turn, are harmonically coupled to
each other.
Quantum-mechanical models have also been developed. Some interesting results
concerning elastic scattering have been obtained in the framework of the semiclassical theory by Doll [36,37], Masel et al. [38, 39], Dion and Doll [40]. The results
appear reliable in the case of an incident molecule of high energy and a direction of
approach almost normal to the wall. Fully quantum-mechanical calculations have also
been performed starting with Tsuchida [41] and Cabrera et al. [42, 43]. Details on this
topic are given in the book by Goodman and Wachman [28] as well as in the paper by
Armand and Lapujoulade [44].
In view of the difficulty of computing the kernel R(~'---~ ~) from a physical model
of the wall, we shall presently discuss a different procedure, which is less physical
in nature. The idea is to construct a mathematical model in the form of a kernel
R ( ~ ' - ~ ~) which satisfies the basic physical requirements expressed by eqs. (4.8),
(4.10), (4.12) and is not otherwise restricted except by the condition of not being too
complicated.
A possible approach to the construction of such models is through the eigenvalue
equation
(4.18)

I R(~'--~ ~)Mw(~')~(~')l~"nl d~ ' = ,~Mw(~)l~'nlPt~f,


~>0

where P t is the operator which reflects the tangential components of $. Let the
eigenvalues ;tk's form a discrete set and let ~ be the corresponding eigenfunctions.

AERODYNAMICALAPPLICATIONSOF THE BOLTZMANNEQUATION

15

Then one can show [8] that


(4.19)

R($'--->$)=Mw(~)l~'nlk=O
~ 2k~bk(--$')~k($),

where
(4.20)

~k($) = P t [ ~ ( $ ) ] .

In order to construct a model, it is necessary to choose the eigenfunctions ~ and the


eigenvalues )~k. Of course, it is convenient to make the choice in such a way that the
series in eq. (4.19) can be evaluated in finite terms. One possibility would be to take
only a finite number of eigenvalues different from zero [2,3], but this has the
disadvantage that, generally speaking, the positivity requirement, eq. (4.8), cannot be
met. The simplest choice, then, is to take the first eigenvalue to be unity (with a
constant eigenfunction, as required by eq. (4.14)) and the others all equal to the same
value 1 - a(0 ~< a ~< 1). The kernel then turns out to be
(4.21)

R(~'--~ ~) = aMw(~)l$-n I + (1 - a)5($ - $ ' + 2 n ( ~ ' . n ) ) .

This is the kernel corresponding to Maxwell model, according to which a fraction


(1 - a) of molecules undergoes a specular reflection, while the remaining fraction a is
diffused with the Maxwellian distribution of the wall Mw. This is the only model for
the scattering kernel that appeared in the literature before the late 1960s. Since this
model was felt to be somehow inadequate to represent the gas-surface interaction,
Nocilla [45] proposed to assume that the molecules are re-emitted according to a
drifting Maxwellian with a temperature, which is, in general, different from the
temperature of the wall. While this model is useful as a tool to represent experimental
data and has been used in actual calculations, expecially in free molecular flow [46],
when interpreted at the light of later developments, it does not appear to be tenable,
unless its flexibility is severely reduced [5, 47]. While the idea of a model like Nocilla's
can be traced back to Knudsen [10], the full development of these ideas led to the
so-called Cercignani-Lampis (CL) model [23]. From the point of view taken here, this
model can be easily obtained by taking the eigenfunctions ~k to be products of
Hermite polynomials in the tangential components of $ times Laguerre polynomials
in the square of the normal component v, = v- n of the same vector. The reason of the
different treatment of the components will be clear if one thinks that the range of v~ is
[ 0, r162
) rather than ( - ~ , ~ ) and the weight factor in the natural scalar product is not
Mw(~) but rather ~ M w ( $ ) . The eigenvalues 2k are taken to be subsequent powers of
two parameters. Then the series in eq. (4.19) can be summed up by means of the
so-called Miller-Lebedev formula [23,3] to yield
(4.22)

R($' --> $) =

[anat(2 -- at) ]
p2

9fl~w~nexp -flw v ~ + ( 1 - a n ) v n

~Zn

-flw

I~t--(1--(2t)~ 12 ]

at(2-

at)

(
Io ~w

2(1- an )~/2Vnv" )
an

16

CERCIGNANI

where Io denotes the modified Bessel function of first kind and zeroth order defined by
2~

(4.23)

Io(Y) = (2z) -1 I exp [ycosq~] dq~.


0

The two parameters at and an have a simple meaning; the frrst one is the a.c. for the
tangential components of momentum, the second one the a.c. for v2 , hence for the part
of kinetic energy associated with the normal motion. This model became rather
popular because it was found by other methods as well: through an analogy with
Brownian motion by KugSer et al. [47], under a special mathematical assumption by
Cowling [48], through an analogy with the scattering of electromagnetic waves from a
surface by Williams[49]. Finally, it followed from the solution of the steady
Fokker-Planck equation describing a (somewhat artificial) physical model of the
wall [25], as was already mentioned. Also, the comparison with the data from beam
scattering experiments were quite encouraging [23, 50, 3]. In spite of this, it must be
clearly stated that there are no physical reasons why this model should be considered
better than others. In particular, we remark that any linear combination of scattering
kernels with positive coefficients adding to unity is again a kernel that satisfies all the
basic properties. Thus, from a kernel with two parameters, such as the CL model we
can construct a general model, containing an arbitrary function of those parameters.
We want to mention another method to produce scattering kernels in a simple
way. This method was described about 15 years ago [51, 3] but does not appear to have
been ever used. The starting point is any positive function K(~, ~'), defined for ~n >I 0
and ~', >I 0, symmetric in its arguments and such that
(4.24)

H(~') =

I K(~, ~')Mw(~)~n d~
~.>0

is not larger than unity for any ~'.n i> 0. Then we form the kernel
(1 - H ( - ~ ' ) ) ( 1 - / - / ( ~ ) )

(4.25)

R(~'--~ ~) = ~nMw(~)K(~, - ~ ' ) + ~nMw(~) J"(1


H(w))wnMw(w)dw

In fact in this way both reciprocity and normalization are clearly satisfied and
positivity follows from the assumption on H(~).
In order to illustrate this procedure, one might think of costructing a mode] chosen
to be as close as possible to Nocilla's model and still satisfy the requirements
expressed by eqs. (4.8), (4.10) and (4.12). This was attempted by the author[19],
starting from the assumption (to be corrected later, according to the method just
described) that the emerging distribution is a drifting Maxwellian, when the
impinging distribution is a delta-function. Then the candidate kernel is
(4.26)

R 0 ( ~ ' ~ ~ ) = A ' I~" I-l~n e x p [ - f i ' I ~ - u' [el,

where A ' = A ( - ~ ' ) , f i ' = f i ( - ~') and u ' = u ( - ~') can be chosen to be dependent on
~'. We satisfy reciprocity first; this requires
~n exp [-fiw [ ~'[2]A' exp [ - f i ' l $ - u'l z ] = I~" I exp [-flw 1~12)]Aexp[-fll~' + u12],

17

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

where A = A(~), fl = fl(~) and u = u($). From this equation we conclude that the only
resonable choice is u(~) = a~R($R = ~ -- 2n(v "v)) and fl(~) = b (a and b constants).
Then
(4.27)

flw [~' 12 + bl~ + a~'R 12 = flw I~12 + bl~' + a~R

and flw + ba2 = b or b = f l w ( 1 - a2) -1. Thus A ' =


(4.28)

12

[~'1 and

Ro(~'-~ ~) = c~, exp [-fiw(1 - a2) -1 I~ - a ~ 12],

where c is a constant.
The symmetric kernel K(~, $') associated with this choice is
(4.29)

K(~,~')=cexp[-flw(1-a2)-ll~+a~'R]2+flwl~]2].

If we now calculate H(~) as defined in eq. (4.21), we obtain a rather lengthy


expression[19]. Unfortunately, this function does not turn out to satisfy the
requirement to be not larger than unity [52], due to the behaviour at large speeds.
A damping factor may be introduced [52] and thus one obtains a somewhat clumsier
kernel, which satisfies all the requirements. Some results have been compared [52]
with the experimental data by Legge [53].
All these models contain pure diffusion according to a non-drifting Maxwellian as
a limiting case. The use of this model is justified for low-velocity flows over technical
surfaces, but is inaccurate for flows with orbital velocity. In fact, the elaboration of
the measurements of lift-to-drag ratio for the Shuttle Orbiter in the free molecular
regime [54] implies a significant departure from diffuse, fully equilibrated re-emission of molecules at the wall. The experimental data by Legge [53] seem to exclude
most of the known models, which indicates that more work is required in this area.
The traditional methods of testing a model for gas-surface interactions is to
compare with molecular-beam experiments and drag and heat transfer measurements
in free molecular flow. The first type of test is, in principle, the best one can think of,
but requires a lot of systematic work, which does not appear to have been done. Also
the information from molecular beam experiments can be overwhelming for certain
applications, such as drag, lift and heat transfer calculations, and one must think
carefully about how the amount of material arising from experiments should be
represented and used. Good recent surveys of experimental work and its comparison
with existing theories are due to Hurlbut [55, 56].
An early example of a partial success of trajectory calculations is also due to
Hurlbut [33]. Recently, there have been several calculations of this kind that shed
light on gas-surface interaction. In particular, Hulpke and Mann [57] and Tenner et
al. [58-60] conducted studies on the scattering of potassium ions from clean tungsten
surfaces and analysis of ion trajectories and found reasonable agreement with
experiment. Among the important result arising from these studies, one can quote
the following ones: a) trajectory analysis can yield simulations representing details of
the physical scattering process very accurately; b) three-dimensional analyses are
required, because two-dimensional analyses miss the lower-energy tail in the energy
distribution of scattered particles; c) the exact value of the parameters in the
interaction potential can be fixed only after a comparison with experiments is made.
Another example of trajectory modelling accompanying experimental data is
provided by the paper of Kolodney et al. [61], who considered mercury scattering
from magnesium oxide.

18

c. CERCIGNANI

With a few exceptions, scattering studies have used single-crystal surfaces as


targets. Hurlbut [55] made the important remark that, while the use of these
well-defined surfaces is essential for progress in surface science, one should, in order
to understand the interaction of the upper atmosphere with the surfaces of orbiting
shuttles, study scattering events and the attendant physics and chemistry on surfaces
with the appropriate composition.
An experimental validation of the CL model was attempted by Bellomo, Dankert,
Legge and Monaco [62]. According to these authors, the CL model does not reproduce
the experimental drag and the heat flow on a metal plate placed in a supersonic nearly
free molecular flow. The interpretation of the experimental results is, however, based
on a misinterpretation of the CL model; in fact, the above-mentioned authors try to fit
the experimental data for the drag and heat transfer coefficients vs. the plate
temperature by means of expressions obtained from the CL model under the
additional tacit assumption that the parameters in the model are temperature
independent. The only fLrm conclusion that one can draw from the experimental data
is that the recovery temperature does not depend on the angle of attack 0 (at least in
the range of the experiments, 45 ~ ~< 0 <~ 90~ this is known [63] to imply that the
accommodation coefficients of the tangential and normal kinetic energies must be
equal, under the envisaged experimental conditions.
With the recent progress achieved in the numerical solution of the Boltzmann
equation by means of more and more powerful computers, it is now possible to think
about more sophisticated comparisons in the transition regime. As pointed out by
Becker et al. [64], their measurements provide such an opportunity. This prompted
several authors [65-67] to compute the flow experimentally investigated by Becker et
al. [64]. Thus, e.g., the computations in ref. [66] lead to the conclusion that neither the
Maxwell model nor the CL model can reproduce the whole set of experimental results
with sufficient accuracy. In particular, temperature and density are well reproduced
but the velocity field is not. Similar conclusions were also drawn by Hurlbut [65].
Apparently, one needs a gas-surface interaction model that assigns perfect accommodation to temperatures but gives rather high slip velocities. New models such as
those to be built according to the method discussed above should be used for further
comparisons.
Most of the above models have been developed for monatomic gases but one
should mention an extension of the CL model to polyatomic molecules proposed by
Lord [68].
We remark that in the terminology used by physicists, frequently the term
,,scattering kernel, is sometimes called ,~reflection coefficient-[69].

5. - P o l y a t o m i c

gases and mixtures.

As is well known, air at room pressure and temperature is a mixture, its main
components being two diatomic gases, nitrogen and oxygen. This immediately calls
for an immediate change in our basic equation, eq. (1.1), which is only suitable for a
single monatomic gas.
A remarkable feature of aerodynamics at the molecular level is that the evolution
equations change in a significant way, when dealing with polyatomic rather than
monatomic gases. This is not the case when the gas is treated as a continuum, where,

19

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

at least at room conditions, only a few changes in the equations occur, the most
remarkable being the change of the ratio of specific heats y. Incidentally, we remark
that the value of ~, caused some problems from the very beginning of kinetic theory.
In fact, the value for a monatomic gas is easily found to be 5/3 (the first calculation
goes back to Clausius). Now, at that time, no monatomic gas was known, except
mercury vapour. Indeed the data for this gas existed thanks to the experiments of
Kundt and Warburg, but they were old and were soon forgotten. Maxwell assumed
the molecules to be solid bodies different from perfectly smooth spheres and found
the value 4/3. Neither value agreed with the ratio for the most common gases, like
nitrogen, oxygen, hydrogen, etc. and this caused a difficulty, till it was pointed out
that a diatomic molecule, modelled as two mass points at a fixed distance, and hence
having all the mass placed on one axis, had only five degrees of freedom, which led to
the value y = 7/5, in agreement with the experimental data. Later Rayleigh and
Ramsay discovered the first rare gases and the value 5/3 was found to apply to the
ratio of their specific heats. More problems arise when vibrational degrees of freedom
are considered, which certainly is the case at high temperatures.
We shall first deal with the case of mixtures of monatomic gases, then with the
case of a single monatomic gas; the case of a mixture of polyatomic gases easily
follows by combining these two cases. In the case of a mixture of monatomic gases,
the differences between the various species will be in the values of the masses and in
the law of interaction between molecules of different species; in the simplest case,
when the molecules are pictured as hard spheres, the second difference is represented
by unequal values of the molecular diameters. In the mathematical treatment, a first
diffrence will be in the fact that we shall need n distribution functions j~ (i =
= 1, 2, ..., n) if there are n species. The notation becomes complicated, but there is no
new idea, except, of course, for the fact that we must derive a system of n coupled
Boltzmann equations for the n distribution functions. The result is

(5.1)

O'--"t

O"--'X

----~ ~--'kZl=

f f (f'f;,

R3

ffJ+

-f~A,)B~(n'~,

IVl)dr

(in,

where Bik is computed from the interaction law between the i-th and k-th species,
while in the k-th term in the left-hand side, V = $ - $. is the relative velocity of the
particle of the i-th species (whose evolution we are following) with respect to a
particle of the k-th species (against which the former is colliding). The arguments $'
and ~. are computed, as before, from the laws of conservation of mass and energy in
an impact with the following result:

(5.2)

$'= ~ -

2/~ik n [ ( $ -

mi

~,).n],

~ , = $ , + 2/~ikn[($ - $ , ) - n ] ,

mk

where I~ik = mimkl(mi + ink) is the reduced mass [3].


The description of polyatomic gases may also be reduced to the case of a mixture of
gases, with the following modification. We remark that eq. (5.1) can be rewritten as

20

C. C E R C I G N A N I

follows [3]:
(5.3)

5J~ + ~. ~fi + X,.. ~3~ _


~t
~x
~

k=l

(~' f ; , - j ~ j ~ , ) W~k(~, ~, I~', ~ , ) d ~ , d~' d ~ , ,

R3 R3 R3

where now ~', ~,, ~, ~. are independent variables (i.e. they are not related by the
conservation laws) and

(5.4)

W,ik(~, ~, I ~ ' ~ )

=S~k(n'L

1~1)5(m~$, + m k ~ , -mi~'-mk~,)"
.,~(m~ 1~, Ie + . ~ I~, I~ - ~n~ I~' I~ - m~ I ~ 1~),

where

n = (~ - ~')/1~

(5.5)

- ~'1

and

B~k(n'~, IVI)

S~k (n- ~, I ~1 ) =

2n. V

( m~ + m k

)3 m i

mk 9

Conservation of momentum and energy is now taken care of by the delta-functions


appearing in eq. (5.4)[3].
With a slight modification, eq. (5.3) can be extended to the case of a mixture in
which a collision can transform the two colliding molecules of species j, 1 into two
molecules of different species k, i (a very particular kind of chemical reaction). In this
case the relations between the velocities before and after the encounter are different
from the ones used so far, but we may still write a set of equations for the n species:

(5.6)

Dt

+~-

-~' + x , .
~x

= k,l,j=~
E

~-

J~J2* - ) ~ f k , ) W ~ ( ~ , ~, [~', ~ , ) d ~ , d~' d~,


(~'~'

R3 R3 R3

where W[j gives the probability density that a transition from velocities ~', ~, to
velocities ~, ~, takes place in a collision which transforms two molecules of species l, j,
respectively, into two molecules of species i, k, respectively. It is clear how the
previous model is included into the new one when the species do not occur.
A possible picture of a molecule of a polyatomic gas, suggested by quantum
mechanics, is as follows [70]. The molecule is a mechanical system, which differs from
a mass point by having a sequence of internal states, which can be identified by a
label, assuming integral values. In the simplest cases these states differ from each
other because the molecule has, besides kinetic energy, an internal energy taking
different values E~ in each of the different states. A collision between two molecules,
besides changing the velocities, can also change the internal states of the molecules
and, as a consequence, the internal energy enters in the energy balance. From the
viewpoint of writing evolution equations for the statistical behaviour of the system, it
is convenient to think of a single polyatomic gas as a mixture of different monatomic
gases. Each of these gases is formed by the molecules corresponding to a given
internal energy, and a collision changing the internal state of at least one molecule is

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

21

considered as a reactive collision of the kind considered above, w[J(~, ~. I~', ~ )


giving the probability density of a collision transforming two molecules with internal
states l, j, respectively, and velocities ~', ~ , respectively, into molecules with
internal states i, k, respectively, and velocities $, ~ . , respectively.
This model is amply sufficient to discuss aerodynamic applications. We want to
mention, however, that it requires non-degenerate levels of internal energy, if there
are, e.g., strong magnetic fields which can act on the internal variables such as
(typically) the spin of the molecules. In that case, if the molecule has spin s, the
distribution function f becomes a square matrix of order 2s 1 and the kinetic
equation reflects the fact that matrices in general do not commute and, as remarked
by Waldmann [71,72] and Snider [73], the collision term contains not simply the
cross-section but the scattering amplitude itself, which may not commute with f.
It is appropriate now to enquire why we started talking about quantum rather
than classical, mechanics. The main reason is not related to practice, but rather to
history. Classical models of polyatomic molecules are regarded with suspicion since
1887 when Lorentz found a mistake in the proof of the H theorem of Boltzmann [74]
for general polyatomic molecules. The question arises from the fact that when one
proves the H theorem for monatomic gas one usually does not explicitly underline
(because it is irrelevant in that case) that the velocities $' and ~. are n o t the velocities
into which a collision transforms the velocities ~ and $ . , but the velocities which are
transformed by a collision in the latter ones; this is conceptually very important, but
the lack of a detailed discussion does not lead to any inconvenience because the
expressions for ~. and ~' are invariant with respect to a change of sign of the unit
vector n, which permits an equivalence between velocity pairs that are carried into
the pair ~., ~ and those which originate from the latter pair, as a consequence of a
collision. The remarkable circumstance which we have just recalled is related to the
particular symmetry of a collision described by a central force, which allows to
associate to a collision [~., ~]--) [~., ~'] another collision, the so-called ,,inverse
collisiom, [ ~ , ~ ' ] - - ~ [ ~ . , ~], which differs from the former just because of the
transformation of the unit vector n into - n . When polyatomic molecules are dealt
with, the states before and after a collision require more than just the velocities of the
mass centres to be described (the angular velocity, e.g., if the molecule is pictured as a
solid body). Let us symbolically denote by [A, B] the state of the pair of molecules.
Then there is no guarantee that one can correlate with the collision [A, B] --->[A', B ' ]
an ,,inverse collisiom~ [A', B' ] --, [A, B], differing from the previous one just because
of the change of n into - n. Now in the original proof of the H theorem for polyatomic
molecules proposed by Boltzmann [74], the assumption was implicitly made that
there is always such collision. Lorentz remarked[75-77] that this is not true
in general. Boltzmann recognized his blunder and proposed another proof based on
the so-called ,,closed cycles of collisions~, [75-78]; the initial state [A, B] is reached
not through a single collision but through a sequence of collisions. This proof,
although called unobjectionable by Lorentz [75] and Boltzmann [77], never statisfied
anybody [79, 72].
For a while the matter was forgotten till a quantum-mechanical proof showed that
the required property followed from the unitarity of the S matrix [72]. A satisfactory
proof of the inequality required to prove the H theorem for a purely classical, but
completely general, model was given only about 15 years ago [80].
For aerodynamic applications all these aspects are not so relevant and, in fact, the
main problem is to find a sufficiently handy model for practical calculations.

22

c. CERCIGNANI

Lordi and Mates [81] studied the ,~two centres of repulsion- model and found that
a rather complicated numerical solution was required for a given set of impact
parameters. The lack of closed-form expressions makes the model impractical for
applications, where the numerical solution describing the impact should be repeated
many millions of times. Curtiss and Muckenfuss[82] developed the collision
mechanics of the so-called sphero-cylinder model, consisting of a smooth elastic
cylinder with two hemispherical ends. Whether or not two molecules collide depends
on more than one parameter; in addition, there are several <<chattering>> impacts in a
single-collision event. The loaded-sphere model had been already developed by
Jeans [83] in 1904 and was subsequently developed by Dahler and Sather [84] and
Sandler and Dahler [85]. Although it is spherical in geometry, the molecules rotate
about the centre of mass, which does not coincide with the centre of the sphere, with
the consequence that it has essentially the same disadvantages as the sphero-cylinder
model.
The only exact model which is amenable to explicit calculations is the perfectly
rough-sphere model, first suggested by Bryan [86] in 1894. The name is due to the fact
that the relative velocity at the point of contact of the two molecules is reversed by
the impact. This model has some obvious disadvantages. First, a glancing collision
may result in a large deflection; second, all collisions can produce a large interchange
of rotational and translational energy, with the consequence that the relaxation time
for rotational energy is unrealistically short; third, the number of internal degrees of
freedom is three, rather than two, which makes the model inappropriate for a
description of the main components of air, which are diatomic gases. One can
disregard the first disadvantage and put a remedy to the second by assuming that a
fraction of collisions follow the smooth-sphere rather that rough-sphere dynamics;
but there is obviously no escape from the third difficulty.
In practical calculations, one has learned, since long time, that one must
compromise between the faithful adherence to a microscopic model and the
computational time required to solve a concrete problem. This was true in the early
days of rarefied-gas dynamics (and may still be true nowadays when one tries to f'md
approximate closed-form solutions or spare computer time) even for monatomic
gases, as we shall discuss in sect. 7 (in connection with the so-called BGK model) and 8
(in connection with Monte Carlo simulation). As a matter of fact, when trying to solve
the Boltzmann equation one of the major shortcomings is the complicated structure of
the collision term; if to this problem, present even in the simplest case, one adds the
complication of the presence of internal degrees of freedom, any practical problem
becomes intractable, unless one is ready to accept the above-mentioned compromise.
Fortunately, when one is not interested in fine details, it is possible to obtain
reasonable results by replacing the collision integral by a phenomenological collision
model, e.g., a simpler expression which retains only the qualitative and average
properties of the true collision term. As computers become more and more powerful,
the amount of phenomenological simplification diminishes and the calculations may
more closely mimic the microscopic models.
For polyatomic gases, the basic new fact with respect to the monatomic ones
is that the total energy is redistributed between translational and internal degrees
of freedom at each collision. Those collisions for which this redistribution is neglegible
are called elastic, while the others are called inelastic. The simplest approach
would be to calculate the effect of collisions as a linear combination of totally

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

23

elastic and completely inelastic collisions, the second contribution being described
by a model analogous to the BGK model shortly described in sect. 7.
Let us now consider in more detail the case of a continuous internal-energy
variable. In this case, it is convenient to take the unit vector n in the centre-of-mass
system and use the internal energy E i and Ei, of the colliding molecules. As usual, the
values before a collision will be denoted by a prime. Equation (1.2) is replaced by
E

(5.7)

f dg, f E[:-2)/2dSi, f f(E,')(~

Q(f, f ) :

dSi'.

sr

E - Ei'

I (Ei',) ( ' - e)/2 d E i ' , ( f ' f ' , - f f , ) B ( E ;

n . n ' ; El', El',---~ El, E i , ) .

Here E = ml g 12/4 + Ei + Ei, is the total energy in the center-of-mass system which
is conserved in a collision. The kernel B satisfies the reciprocity relation
(5~)

[V[B(E;n'n';E~',Ei',-~Ei,

E~,)= [V'[B(E;n.n';E~,Ei,-~Ei',Ei',).

Here we follow a paper by Kug~er [87] and look for a one-parameter family of models,
assuming that the scattering is isotropic in the centre-of-mass system. The second
crucial assumption will be that the redistribution of energy among the various
degrees of freedom only depends upon the ratios of the various energies to the total
energy E, e i = E i / E , etc. This assumption is valid for collisions of rigid elastic bodies
and can be considered as a good approximation for steep repulsive potentials. It is
then possible to write B in the following form:
(5.9)

IV' [2 gtot(E )
]V[B(E; n . n ' ; Ei' , Ei',---->Ei, Ei, -- - - O(E[, E ; , ---> s
[V[ 4 z E n

~i* ] T)"

The denominator on the right takes care of normalization. Then the function 0
satisfies the following relations:
1

(5.10)

I e ( ' - 2>/2de I e:-e>/2 de. 0(el, e ; , ~ el, el. ; v ) = 1,


0

(5.11)

l-e
0

( 1 - 8i~ - E l , ) 0 ( c i ' , E l , - - )

Ei, E i , ; T) -~ ( 1 - E i -- E i , ) 0 ( E i , E i , - - > E l , E l , ; "g).

The dependence of atot on E makes it possible to adjust the model to the correct
dependence of the viscosity on temperature. The parameter v will be chosen in such a
way as to represent the degree of inelasticity of the collisions, r = 0 will correspond to
elastic collisions:
(5.12)

0(e[, E[,--'->Ei, Ei, , O)

e-(n-2)/2E.

(n-2)/2

6(E i

8[)6(e;

-- e i , )

r = oo will correspond to maximally inelastic collisions:


(5.13)

0(el, e' ---~ei, ei,; oo) = e - ( , - 2)/2e , ( ' - 2)/2 F(n + 2)


(r(n)) 2

(1 -

ei-

C. CERCIGNANI

24

A mixture of the two extreme cases will give the model first proposed by Borgnakke
and Larsen [88] in 1975:
(5.14)

0(ei', ej, ---) e i , e~,, r) =


P

= e x p [ - v ] 0 ( e ( , e~.--~ ei, el. ; 0) + (1 - e x p [ - r ] ) 0 ( e ( , e~. ~ c i , el.; ~ ) .


Kug~er [87] notices an analogy between this model and Maxwell model for gas-surface
interaction, as discussed in ref. [3], and introduces another model, called the thetamodel, which would correspond to the CL model in this analogy.
The Larsen-Borgnakke model has become a customary tool in numerical
simulations of polyatomic gases. It can be also applied to the vibrational modes
through either a classical procedure that assigns a continuously distributed
vibrational energy to each molecule, or through a quantum approach that assigns a
discrete vibrational level to each molecule. It would be out of place here to discuss
this point in more detail, for which we refer to the book of Bird [89]. We also refrain
from discussing the interesting recent developments [90,91] of an old idea of
Boltzmann [92] to interpret, in the frame of classical statistical mechanics, the
circumstance that at low temperatures the internal degrees of freedom appear to be
frozen, as due to the extremely long relaxation times for the energy transfer process.

6. - C h e m i c a l r e a c t i o n s a n d t h e r m a l r a d i a t i o n .

Chemical reactions are important in high-altitude flights because of the high


temperatures which develop near a vehicle flying at hypersonic speed. Up to 2000 K,
the composition of air can be considered the same as at standard conditions. Beyond
this temperature, Ne and 02 begin to react and form NO. At 2500 K dyatomic oxygen
begins to dissociate and form atomic oxygen O, till 02 completely disappears at about
4000 K. Nitrogen begins to dissociate at a slightly higher temperature (about 4250 K).
Thus NO disappears at about 5000 K. Ionization phenomena start at about 8500 K.
As we have implicitly remarked when we have written eq. (5.6), the kinetic theory
of gases is an ideal tool to deal with chemical reactions of a particular kind, i . e .
bimolecular reactions, which can be written schematically as
(6.1)

A + B ,-* C + D,

where A, B, C and D represent different molecular species. We already used the term
-molecule-, as usual in kinetic theory, to mean also atom (a monotomic molecule); in
this section we shall further enlarge the meaning of this term to include ions,
electrons and photons as well, when we have to deal with ionization reactions and
interaction with radiation.
As long as the reaction takes place in a single step with the presence of no other
species than the reactants, it is a well-known circumstance that the change of
concentration of a given species (A, say) in a space-homogeneous mixture can be
written as follows:
(6.2)

~dn---~ =
dt

kb

(T)n c n D - - kf(T)nAnB

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

25

We remark that, in chemistry, one uses the molar density in place of the number
density used here, the two being obviously related through Avogadro's number.
The rate coefficients kf and kb for the forward and backward (or reverse) reactions,
respectively, are functions of temperature and are usually written by a
semi-empirical argument, which generalizes the Arrhenius formula, in the form
(6.3)

k ( T ) = A T " exp

- -~

where A and y (= 0 in the Arrhenius equation) are constants, and Ea is the so-called
a c t i v a t i o n e n e r g y of the reaction. It is clear that these equations, though having a
flavour of kinetic theory, are essentially macroscopic and can be assumed to hold
when the distribution function is essentially Maxwellian.
In fact, the above reaction theory can be obtained by assuming that the
distribution functions are Maxwellian, the role of internal degrees of freedom may be
ignored and the r e a c t i o n cross-section vanishes if the translational energy Et in the
centre-of-mass system is less than E a and equals a constant aR if the energy is larger
than E a . A more accurate theory is obtained [93, 94, 89] by assuming that the ratio of
the reaction cross-section to the total cross-section is zero when the total collision
energy Ee (equal to the sum of Et and the total internal energy of the two colliding
molecules Ei) is less than Ea and proportional to the product of a power of Ee - E a and
a power of Ec. The exponents and the proportionality factor are essentialy dictated by
the number of internal degrees of freedom, the exponent of the temperature in the
diffusion coefficient of species A in species B and the empirical exponent ~] appearing
in eq. (6.3). This theory provides a microscopic reaction model that can reproduce the
conventional rate equations (6.2), (6.3) in the continuum limit. The model is, however,
as in the case of gas-surface interaction and models for polyatomic gases, largely
based on phenomenological considerations and mathematical tractability. The ideal
microscopic model would consist of complete tabulations of the differential crosssections as functions of the energy states and n. Some microscopic data, coming from
extensive quantum-mechanical computations, supported by experiments, are
available, but, unfortunately, not very much is known for reactions of engineering
interest. When comparisons can be made, the reaction cross-sections provided by the
phenomenological models is of the correct order of magnitude. This provides some
reasons of optimism about the validity of the results obtained with these models for
the highly non-equilibrium rarefied-gas flows.
Termolecular reactions provide some difficulty to kinetic theory, because the
Boltzmann equation essentially describes the effect of binary collisions. They are,
however, of essential importance in high-temperature air, where the reverse (or
backward) reaction of a dissociation one is a recombination reaction, which is
necessarily termolecular, as we shall presently explain. A typical dissociationrecombination reaction can be represented as
(6.4)

AB + T ~ A + B + T ,

where AB, A, B and T represent the dissociating molecule, the two molecules
produced by the dissociation and a third molecule (of any species), respectively. The
latter molecule, in the forward reaction, collides with AB and causes its dissociation.
This process is described by a binary collision and is an endothermic reaction,
requiring a certain amount of energy, the dissociation energy Eo.

26

C. CERCIGNANI

The recombination process is an exothermic reaction and it might seem that one
could dispense with the ,,third body, T and consider it as a bimolecular reaction
(6.5)

AB ~-- A + B.

However, one can easily see that the momentum and energy balance for a binary
collision cannot be simultaneously satisfied in the presence of energy release. The
molecule T is thus required to describe the recombination process.
In order to keep the binary-collision analysis, appropriate for a rarefied gas, we
must think of the recombination process as a sequence of two binary collisions. The
first of these forms an (unstable) orbiting pair, that is stabilized by the second
collision of this pair with T, as long as this collision occurs within a sufficiently small
elapsed time. One can then extend the previous theory, assuming that the activation
energy is zero and the cross-section acquires a factor proportional to the number
density of the species T.
Ionization reactions involve the electronic states and it is unlikely that a purely
classical theory will be successful in describing them, because of the selection rules.
Yet, one can use the phenomenological approach to provide at least an upper bound
for the reaction rates.
As mentioned above, one can, in principle, think of interaction with a radiation as
if it were a reaction involving photons as ,,molecules-. Here spontaneous emission
should also be taken into account. It becomes harder to develop phenomenological
models, because one should consider as many species as three excited levels for each
molecule.

7. - Solving the Boltzmann equation. Analytical techniques.

In the early sections of this paper we defined (in a qualitative fashion, to be sure)
how to attack the problems of rarefied-gas dynamics and what phenomena should be
looked for in their solutions. But time has come to say something about the way these
problems are solved: how does one handle the already complicated Boltzmann
equation with similarly complicated boundary conditions?
The history of approximate solutions goes back, after the first attempts of
Maxwell and Boltzmann, to Hilbert [95], Chapman [96] and Enskog [97]. As is well
known, they obtained solutions valid in the continuum limit, very useful to compute
the transport coefficients; there are standard monographs [98,99], which deal with
this part of the theory.
In 1949 Grad [100] devised a systematic method to deal with the solutions of the
Boltzmann equation, i.e. his famous 13-moment method. Although there is some
rationale in his approach and his equations give better results than the
Navier-Stokes equations for certain problems, it appears fair to say that they turned
out to be rather useless for the progress of rarefied-gas dynamics. The most notable
failure is related to the problem of shock structure, for which the 13-moment
equations fail to give a solution for a Mach number larger than 1.65.
An important early approximate solution of the shock wave problem was
Mott-Smith's solution [101], remarkable for its simplicity and resistance to any simple
improvement; we shall discuss it later.
All these methods paid little, if any, attention to the problem of boundary

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

27

conditions, which, as we have already remarked, is vital in any application to


upper-atmosphere flight.
The lesson taught by Mott-Smith's solution is that it is not so useful to look for
general methods with the aim of obtaining continuum-like equations, but one should
rather devise approximate methods for dealing with particular problems. This is the
rationale behind the use of models and the linearization techniques [2,3].
The linearized Boltzmann equation is obtained by assuming that the solution is a
small perturbation of a basic Maxwellian M0, and is useful for low Mach number flows
and, as such, is used in the applications of rarefied-gas dynamics to environmental
problems. It is also useful in order to deal with the kinetic layers and, as such, retains
its validity to compute the boundary conditions in the slip regime and hence can still
play an important role in the computations of re-entry problems. An example of this
kind of work is the evaluation of the slip coefficient and the temperature jump for
arbitrary models of gas-surface interaction, based on a variational principle for the
integro-differential form of the linearized Boltzmann equation [102,2,3].
In the area of environmental problems, understanding and control of the
formation, motion, reactions and evolution of particles of varying composition and
shapes, ranging from a diameter of the order of 0.001 ~m to 50 ~tm, as well as their
space-time distribution under gradients of concentration, pressure, temperature and
the action of radiation have grown in importance, because of the increasing awareness
of the local and global problems related to the emission of particles from electricpower plants, chemical plants, vehicles as well as of the role played by small particles
in the formation of fog and clouds, in the release of radioactivity from nuclear-reactor
accidents, and in the problems arising from the exhaust streams of aerosol reactors,
such as those used to produce optical fibers, catalysts, ceramics, silicon and carbon
whiskers.
When the mean free path is appreciable but still small with respect to size of the
particles, the study of the Knudsen layers to obtain the correct slip boundary
conditions is useful. When the Knudsen number becomes larger, the transfer
processes between an aerosol particle can be studied with the linearized Boltzmann
equation and hence variational techniques of the type hinted at above prove useful. In
fact already twenty years ago variational results [50] for the drag on a spherical
particle were obtained for the entire range of values of the Knudsen number based on
the radius of the particle. Similarly, if one considers the diffusion of a trace species
(water vapour in air) that may condense on a particle, the variational method can be
successfully applied, as indicated by Loyalka [103]. The heat transfer from a particle
was similarly studied by Cercignani and Pagani [104].
These problems have also been solved with numerical methods that produce
results in a very good agreement with the variational calculations. Other problems,
such as those related to the forces experienced by a particle in a thermal or
concentration gradients (thermophoresis and diffusiophoresis) have been treated by
numerical methods only [105].
Attempts to use the linearized Boltzmann equation in rarefied-gas dynamics
began in the late 1950s and early 1960s. One of the first fields to be explored was that
of the ,,simple flows,,, such as Couette and Poiseuille flows in tubes and between
plates; here it turned out that the equation to be solved is still formidable and various
approximation methods were proposed. Some of these were perturbation methods:
for large or small Knudsen numbers or about an equilibrium solution (Maxwellian).
The first two approaches gave useful results in the limiting regimes, while the third

28

C. CERCIGNANI

method led to studying the so-called linearized Boltzmann equation, which produced
predictions which are in a spectacular agreement with experiment and have shed
considerable light on the basic structure of transition flows, whenever non-linear
effects can be neglected [2, 3]. This gave confidence to further use of the Boltzmann
equation for practical flows. Other problems which were treated with the linearized
equation were the half-space problems which are basic in order to understand the
structure of Knudsen layers and to evaluate the slip and temperature jump
coefficients.
This explains the importance of the linearized equation for the hypersonic flows
met in the aerodynamics of space vehicles; in fact there is a large portion of upperatmosphere aerodynamics (vital to the dynamics of a re-entering body) for which the
Navier-Stokes equations may still be considered to be valid (0.01 < 1 / L < 0.1), except
in a thin layer near the body, having a thickness of the order of a mean free path
(Knudsen layer). This layer can be described by means of linearized equations and
this circumstance has led to a complete understanding of the phenomena occurring in
this situation. Actual calculations can be performed in an approximate analytical form
by means of variational techniques and also by numerical methods. This has led, in
particular, to confirming the existence of a minimum in the flow rate in the Poiseuille
flow of a rarefied gas as a function of the Knudsen number, discovered by Knudsen
about eighty years ago [2, 3], and to computing the drag upon a sphere at low subsonic
speeds with results in a very good agreement with the experimental data that
Millikan obtained as a pre-requisite to his celebrated oil-drop experiment to measure
the electron charge [106, 2, 3].
A particularly simple and useful technique to obtain appproximated but accurate
results from the linearized Boltzmann equation is provided by the variational
technique [102].
As is well known, a variational principle, which does not reduce to a triviality, i.e.
to something analogous to a least-square method, is based on a property of symmetry
of the operator appearing in the equation. For linear equations this property reduces
to the symmetry of a linear operator with respect to a suitable scalar product. In the
steady case, the linearized Boltzmann equation reads as follows:
~h

(7.1)

~. - -

~x

- Lh = go,

where h is the perturbation of the basic Maxwellian M0 in the distribution function


f = M0 (1 +h), go a source (which might arise from some inhomogeneity in M o ) and L the
linearized collision operator, related to the quadratic operator defined in eq. (1.2) by
(7.2)

2MoLh

= Q(M0(1 + h), M0(1 + h)) - Q(M0(1 - h), Mo(1 - h)).

For simplicity, we have omitted the body force term in eq. (7.1). This equation has
to be accompanied by the linearized version of the boundary conditions discussed in
sect. 4.
Now, it is easy to see [102, 3] that L is self-adjoint with respect to a scalar product
weighted with Mo, while the differential operator in eq. (7.1) (which we shall
henceforth abbreviate to D) is (with suitable boundary conditions) antisymmetric
with respect to the same scalar product. Thus the variational formulation of eq. (7.1) is
not immediate. There is, however, a property [102,3] that opens the way to such

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

29

formulation. Let P denote the operator which changes $ into - ~ (i.e. the parity
operator in velocity space). Then P commutes with L (at least for monatomic gases
and polyatomic gases with molecular interaction possessing spherical symmetry) and
P D is self-adjoint with respect to the above-mentined scalar product and we can
obtain a variational principle,/~e, we can fred a functional J(f~) whose first variation
vanishes if and only if f~ coincides with the solution h of eq. (7.1) satisfying the
appropriate boundary conditions. It is sufficient to replace the equation D h - L H =
= go by the equivalent equation P D h - P L H = Pgo. More work is, of course,
needed [102,3] to write down the principle in an appropriate way, when we want to
vary the functions on the boundary as well.
To have a variational principle at our disposal would be a matter of idle curiosity,
unless we knew that there is an important property associated with it. This is
provided by the circumstance that if we take a subclass of functions {f~} and try to
fred the one that best approximates the solution in the sense of the variational
principle, the above-mentioned functional J, evaluated at the approximate
stationarity point, differs from the value at the exact statioparity point by ~ if ~ is the
order of magnitude of the difference between the optimal h in the restricted class and
the solution of the problem. This paves the way to very accurate estimations of the
functional. The last step that remains to be done is to relate the value of J to a
physically important quantity, which will be thus computed with analogous accuracy.
This has been done for many problems including all those mentioned above.
A particularly interesting problem, which cannot be treated with the linearized
Boltzmann equation, has been already mentioned and is related to the structure of a
shock wave. The latter is not a discontinuity surface as in the theory of compressible
Euler equations, but a thin layer (having, usually, a thickness of the order of a few
mean free paths). In the case of a normal shock wave, one can imagine that it occurs in
the entire space without boundaries; finding the shock wave structure means solving
the Boltzmann equation when the solution (which depends on one space coordinate,
say x, and the three velocity components $~ (i = 1, 2, 3), but not on time and the other
two coordinates) tends to two different Maxwellians when x tends to + ~ and - ~ .
The two Maxwellians have the following shape:
(7.3)

fo2 = 0 ( 2 z r R T )-8/2 e x p [ - ] ~

- u 2 i[2 / ( 2 R T )],

where i is the unit vector of the x-axis and the superscripts _+ refer to down- and
up-stream states, respectively.
The shock wave problem for high values of the upstream Mach number provides
an example of a flow where very large local Knudsen numbers can occur at any
density, due to the sharp changes of the physical quantities in a very thin layer.
An early approach that was moderately successful in dealing with the shock
wave problem was the Mott-Smith method [101]. This method postulates that there
is a bimodal distribution, i.e. a linear combination of the two Maxwellians defined in
eq. (7.3):
(7.4)

f = vfi+ + (1 - v)fi- .

Here v = v ( x ) is a function that goes from 0 to 1 through the shock. Equation (7.4) is
easily shown to be compatible with the balance of mass, momentum and energy,
provided the constant values Q 2, u 2, T 2 satisfy a set of compatibility conditions,
which are nothing else than the Rankine-Hugoniot relations, familiar from the ideal

30

C. CERCIGNANI

fluid theory of shock waves. In order to determine v(x), several procedures have been
presented, none of which is very satisfactory, since they are essentially arbitrary.
Grad suggested [107, 3] that the problem of shock wave structure had a limit for the
upstream Mach number tending to infinity in the form of a delta-function plus a
relatively tame function. This matter has been revived by several authors.
The results obtained by this method for several physical quantities, including the
thickness of the shock, are considerably more accurate than the Navier-Stokes values
for other-than-low Mach numbers. In addition to the unsatisfactory status from a
mathematical point of view, the Mott-Smith approach suffers, however, the further
drawback of being restricted to the shock structure problem. One should also remark
that, while the values for the macroscopic quantities (density, bulk velocity, pressure)
are in a reasonable agreement with the most accurate numerical solutions of
the Boltzmann equation, this is not the case for the distribution function itself (see
sect. 8).
When trying to solve the Boltzmann equation for practical problems, one of the
major shortcomings is the complicated structure of the collision term, eq. (2). When
one is not interested in fine details, it is possible to obtain reasonable results by
replacing the collision integral by a so-called collision model, a simpler expression
J(f) which retains only the qualitative and average properties of the collision term
Q(f, f). The equation for the distribution function is then called a kinetic model or a
model equation.
The most well-known analytical (or semi-analytical) solutions in kinetic theory are
obtained through the simplest collision model, usually called the Bhatnagar, Gross
and Krook (BGK) model, although Welander proposed it independently at about the
same time as the above-mentioned authors [108, 109]. It reads as follows:
(7.5)

J(f) = u[q)($) - f ( ~ ) ] ,

where the collision frequency v is independent of g (but depends on the density Q and
the temperature T) and q~ denotes the local Maxwellian, i.e. the (unique) Maxwellian
having the same density, bulk velocity and temperature as f:
(7.6)

= O(2zRT) -3/2 exp [ - ] ~ - v]2/(2RT)].

Here 0, v, T are chosen is such a way that we have


(16)

f ~fa (~) r

d~ = I ~f. (~) f(~) d ~

(a = O, 1, 2, 3, 4),

where F0 = 1, Fi = ~i (i = 1, 2, 3) and ~4 = [g[2.


It should be remarked that the non-linearity of the BGK collision model, eq. (14), is
much worse than the non-linearity in Q(fi f); in fact the latter is simply quadratic in
f, while the former contains f in both the numerator and denominator of an
exponential, because v and T are functionals of f, defined by eq. (2.1).
The main advantage in the use of the BGK model is that for any given problem one
can deduce integral equations for Q, ~, T, which can be solved with moderate effort
on a computer. Another advantage of the BGK model is offered by its linearized
form.

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

31

The BGK model has the same basic properties as the Boltzmann collision integral,
but has some shortcomings. Some of them can be avoided by suitable modifications, at
the expense, however, of the simplicity of the model [2, 3].

8. - Solving the Boltzmann equation. Numerical techniques.

The kinetic models have been very useful in obtaining approximate solutions and
forming qualitative ideas on the solutions of practical problems, but, in general, do
not provide us with the detailed and precise answers to the sort of question that is
posed by the space engineer. Various numerical procedures exist which either
attempt to solve for f by conventional techniques of numerical analysis or efficiently
bypass the formalism of the integro-differential equation and simulate the physical
situation that the equation describes (Monte Carlo methods). Only recently proofs
have been given that these partly deterministic, partly stochastic games provide
solutions that converge (in a suitable sense) to solutions of the Boltzmann equation.
Numerical solutions of the Boltzmann equation based on finite-difference methods
meet with severe computational requirements due to the large number of
independent variables. In practice the only method that has been used for
inhomogeneous problems in more than one space dimension is the technique of
Hicks-Yen-Nordsiek [110,111], which is based on a Monte Carlo quadrature method
to evaluate the collision integral. This method was further developed by Aristov and
Tcheremissine[ll2,113] and has been recently applied to a few two-dimensional
flows [114,115].
An additional difficulty for these methods is the fact that chemically reacting and
thermally radiating flows (and even simpler flows of polyatomic gases) are hard to
describe with theoretical models having the same degree of accurateness as the
Boltzmann equation for monatomic non-reacting and non-radiating gases. These
considerations paved the way to the development of simulation methods, which
started with the work of Bird on the so-called direct-simulation Monte Carlo (DSMC)
method [116] and have become a powerful tool for practical calculations. There appear
to be very few limitations to the complexity of the flow fields that this approach can
deal with. Chemically reacting and ionized flows can be and have been analysed by
these methods.
In the DSMC method, the intermolecular collisions are considered on a
probabilistic rather than a deterministic basis. Furthermore, the real gas is modelled
by some thousands of simulated molecules on a computer. For each of them the space
coordinates and velocity components (as well as the variables describing the internal
state, if we deal with a polyatomic molecule) are stored in the memory and are
modified with time as the molecules are simultaneously followed through
representative collisions and boundary interactions in the simulated region of space.
The calculation is unsteady and the steady solutions are obtained as asymptotic limits
of unsteady solutions. The flow field is subdivided into cells, which are taken to be
small enough for the solution to be approximately constant through the cell. The time
variable is advanced in discrete steps of size At, small with respect to the mean free
time, i.e. the time between two subsequent collisions of a molecule. This permits a
separation of the inertial motion of the molecules from the collision process: one first
moves the molecules according to collision-free dynamics and subsequently the
velocities are modified according to the collisions occurring in each cell. The rate of

32

C. CERCIGNANI

occurrence of collisions is given by (hard spheres)

r=

(8.1)

(8.2)

r~j-

0
Nm

~'..rij,
f n'(gi-

where N is the number of particles in the sample and ~ is the velocity of the i-th
molecule (i = 1, ..., N). Each time a collision occurs, the velocities of a collision pair
are modified (and, as a consequence, r also varies). Let Tk be the length of the time
interval between the ( k - 1 ) - t h and the k-th collision in the time interval [0, At] (t = 0
is the time of the 0-th collision by definition). The Tk are chosen in this way: after the
(k - 1)-th collision, one samples a pair (i, j) on the basis of the probability distribution Pij = r~j/r (fixed velocity for each pair); then one takes Tk = 2N/[(N - 1)rij].
Then one samples a direction k of the post-collisional velocity ~ i - ~ i with the
probability distribution p(k) = r~i/r (i, j and I ~ - ~y I fLxed) and replaces ~ and ~5 by
~+ and ~], given by
1

(8.3)

i ; = ~ ( ~ + lj + k I ii - ~j I ),

(8.4)

~J+ = 2 ( ~ + ~j - k l ~ - ~j I).

The operation is repeated until ~] Tk exceeds AT.


k
Some variations of Bird's method have appeared, due to Koura[117],
Belotserkovskii and Yanitskii [118], Deshpande [119]. They differ from Bird's method
because of the method used to sample the time interval between two subsequent
collisions.
A different method was proposed by Nanbu [120]. One does not subdivide At and
works with the probability P~ that the i-th molecule collides in [0, At]. One has
N

(8.5)

P~ = At ~] r~j.
j=l

One starts with rij evaluated at t = 0 and samples a random number co in [0, 1].
According to whether co is smaller or larger than P~, the i-th molecule collides or does
not collide in [0, At]. If a collision occurs, one samples a collision partner of the i-th
molecule from the probability distribution
(8.6)

P)i) = rij / ~ rik .


I k=l

Then one samples a direction k of the post-collisional velocity ~ - ~ f with the


probability distribution p(k) = rii/r (fixed i, j and I~i+ - ~ I) and reptaces ~i by ~i+
given by eq. (6). Then the procedure is repeated for all the values of i.
The method of Nanbu was criticized by Koura [121], who asserted that momentum
and energy are not conserved by collisions, because one does not change the velocity
of the j-th molecule, when one changes the velocity of the i-th molecule. The criticism

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

33

is not well founded, however, because, as Nanbu pointed out [122], the Boltzmann
equation requires the overall conservation of momentum and energy of the system at
each space point, not the conservation of the same quantities at each single simulated
event. What is new in Nanbu's method is precisely the circumstance that it does not
try to simulate the N-body dynamics, but rather the description of the system given
by the Boltzmann equation. Nanbu's method is now well understood, from both a
physical and mathematical standpoint and has been rigorously proven to yield
approximations to solutions of the Boltzmann equation, provided the number of test
particles is sufficiently large. The relevant theorem reads as follows [123, 124]:

Theorem. If the Boltzmann equation with initial data fi has a smooth,


non-negative solution f(x, ~ ) 9 L l, then the solution j~ of Nanbu's method converges
weakly in L 1 to f, in the sense that, for any test function q~(x, ~ ) e L |
(8.7)
as N--> ~, Ax, At---)0.
Subsequently, a similar result was proved for Bird's method [125,126].
The computing task of a simulation method varies with the molecular model. For
models other than Maxwell's it is proportional to N for Bird's method, while it is
proportional to N 2 for Nanbu's method. Babovsky [123], however, found a procedure
to reduce the computing task of Nanbu's method and make it proportional to N. His
modification is based on the idea of subdividing the interval [0,1] into N equal
subintervals. If the random number oJ lies in the j-th segment, one calculates only
P i j = rij At and there is no collision if w < ( j / n ) - Pij, there is a collision with the j-th
molecule if w >I ( j / n ) - Pij. There is also a condition that Pij must satisfy, i.e.
Pij < l / N , but this is usually automatically verified, given the size of At. Application
of Nanbu's method in the form modified by Babovsky shows that the the computing
task is not only theoretically but also practically comparable to that of Bird's
method [127].
9. - V o r t i c e s a n d t u r b u l e n c e i n a r a r e f i e d g a s .

The Monte Carlo method discussed at the end of the previous section is not only a
practical tool for engineers, but also a good method for probing into uncovered areas
of the theory of the Boltzmann equation, such as stability of the solutions of t h i s
equation.
While the problems related to the instability of laminar flows and their transition
to turbulence have been studied for a long time in classical hydrodynamics, the
corresponding problems in kinetic theory have been paid attention to only recently.
This circumstance is clearly related to the extremely complex character of such
problems. It is clear, however, that the study of such problems might be of great
importance for the purpose of understanding fundamental phenomena of instability
and self-organization in molecular dynamics as well as computing hypersonic rarefied
flows. Typical examples in this area are B~nard's instability[128,129], Taylor
vortices [130, 131] and channel flow [132].
In the last few years, objections have been raised against the ability of the DSMC
technique to faithfully resolve vortex motion, due to the lack of accurate conservation

34

C. CERCIGNANI

of angular momentum in collisions. It seems that the first paper to cast doubts on this
ability is due to Meiburg[133], who applied both molecular dynamics and directsimulation Monte Carlo to the case of a rarefied flow past an impulsively started
inclined flatplate. A clear vortex structure was obtained in the wake region of the
molecular-dynamics calculation, but the wake in the directhsimulation Monte Carlo
was relatively devoid of structure. As pointed out by Bird [134], the approximations
associated with either method cannot be tested for the unsteady flow past a plate
because this problem places excessive demands on computer time. He therefore
introduced the forced vortex flow produced by a moving wall in a two-dimensional
cavity as an alternative test case and came to the conclusion that as long as the cell
size requirements are met, the non-conservation of angular momentum in collisions
does not appear to have a significant effect on the results of direct-simulation Monte
Carlo calculations. Bird [134] also examined the values of the parameters in Meiburg's
calculations [133] and showed that the density of the gas was too high to employ the
direct-simulation Monte Carlo method, because the mean free path was of the order of
the molecular diameter and the size of the cell is too large to analyse the vortical wake
structure. The issue of nomconservation of angular momentum was addressed by
Nanbu et al. [135], who showed that if the cell size is sufficiently small, the total
angular momentum of the molecules in a cell is almost conserved in the Monte Carlo
calculations.
The results obtained by Stefanov and Cercignani [128-132] (see also the book of
Bird [89]) have a preliminary character, but clearly indicate that the formation of such
vortex patterns arising from the above-mentioned instabilities are possible. The
calculations refer to Knudsen numbers of order 10 .2 and different choices of the other
parameters (including high values of the Mach number).
Encouraged by the positive result of these investigations, the same authors also
considered the fluctuations of the macroscopic quantities in a rarefied gas flowing in a
channel under the action of a constant external force in a direction parallel to the
walls (which are assumed to be at rest with the same temperature). Their final aim is
the study of the transition to turbulence by means of the Boltzmann equation, but
they remained far from achieving this goal, since their calculations are restricted to a
two-dimensional geometry. The main aim of their calculations was to investigate the
time evolution of both the macroscopic quantities and their fluctuations. The
numerical experiments refer to Knudsen numbers between 0.005 and 0.1 and three
different values of the body force. The analysis of the results indicates an increase in
the macroscopic fluctuation for Kn ~< 0.05 and certain magnitudes of the force. In
order to recognize the possible formation of vortex patterns and to estimate the
macroscopic fluctuations, a data analysis was performed. If one takes into account the
results of this analysis, he is tempted to conclude that he observes a transition from
laminar to two-dimensional small-scale turbulence.
A common statement in the books and papers dealing with turbulence is that the
scales of turbulence and molecular motions are widely separated and thus one may
well ask what is the purpose of calculations of the kind discussed above. To answer
this objection, one may remark that the common statement alluded to in the previous
sentence may always be true for a liquid; but what about a gas? Let us consider the
ratio between the mean free path l and the dissipative scale of turbulence /D:
(9.1)

1/lD = ( 1 / L ) ( R e ) 3/4 = Kn (Re) 3/4 = (Ma)3/4(Kn) 1/4 ,

AERODYNAMICAL APPLICATIONS OF THE BOLTZMANN EQUATION

35

where L is a macroscopic scale and the relation Re = Ma/Kn has been taken into
account. If we take Re = 104 for fully developed turbulence, we obtain that I and 1D are
of the same order for Kn = 10 .8 and Ma = 10. Thus, in hypersonic flow, for moderate
rarefaction, turbulence scale and mean free path are of the same order of
magnitude!
These estimates are, of course, to be taken with caution, since there might be
sizable factors ~,of order unity,. These considerations should, however, not be ignored
when considering high Mach number flows. This field is completely unexplored and is
certainly worth more attention.

10. - C o n c l u d i n g

remarks.

In this survey, the applications of kinetic theory to rarefied-gas dynamics have


been stressed. We have seen that kinetic theory has undergone a lot of progress since
1738, when Bernoulli first advanced the idea that gases are formed of elastic
molecules rushing hither and thither at large speeds, colliding and rebounding
according to the laws of elementary mechanics. In the second half of this century, the
theory has been for the first time considered to be a practical tool for certain problems
such as, e.g., the flow past a vehicle flying in the upper atmosphere.
The purely mathematical theory has been left in the background. It seems
appropriate, however, to point out, in this concluding section, that the Boltzmann
equation has originated many interesting mathematical researches. It is remarkable,
in fact, that the purely mathematical theory of this equation, concerned with such
basic questions as the existence and uniqueness of the solution for the initial-value
problem, had a start already in the 1930s (see ref. [2] for more details). A sudden
change in the mathematical theory occurred in the late 1980s, when DiPerna and
Lions [136] proved a global existence theorem for the initial-value problem with quite
general data, but several important problems, such as proving that the solution is
unique and conserves energy, are still open.
The extension of this result to initial-boundary-value problems was considered by
several authors. The case of an isothermal boundary was the first to be tackled [137].
In the case of non-isothermal data along the boundary, the initial-boundary-value
problem possesses data which are compatible, not with a Maxwellian, but rather with
one of those steady solutions, whose theory is still in its infancy [138]; thus one cannot
expect the solution to tend toward a Maxwellian when t---> oo as has been recently
shown for other kinds of boundary conditions [139-142]. The main difficulties in
tackling this problem seem to lie with large velocities. For this reason, a modified
Boltzmann equation was introduced [143], in which all the collisions, such that the
sum of the squares of two colliding molecules is larger than an assigned positive
constant, are cut off. The only place where the cut-off was used was, however, in the
entropy estimate and the need for the cut-off disappears when the temperature is
constant. Thus the above-mentioned paper[143] contains also a slightly different
proof of Hamdache's theorem [137], with an extension to more general boundary
conditions, to a more detailed study of the boundary behaviour, and for the full class
of collision operators of the DiPerna-Lions existence context.
In a recent paper [144], the previous results [143] were extended to the non-cutoff case at the price of introducing some restrictions on the scattering kernel.

36

C. CERCIGNANI

L a t e r the present author [145] showed that one of the conditions can be replaced
by a more natural one.
F o r applications to aerospace problems, the external case is of paramount
importance. This aspect had been treated [146,147] only in an L ~ perturbation framework, till the relevant tools for dealing with external problems for large data in an L 1
framework were provided [148].
F o r a more detailed survey of the mathematical theory of the Boltzmann equation
we refer to a recent book [4].
In a few recent papers [149-151], the author discussed inequalities which guarantee that both the gain and the loss t e r m in the collision integral of the Boltzmann
equation are in L ~ under a suitable truncation, which only depends upon the relative
speed and amounts to an acceptable assumption on the cross-section. These results
refer to solutions depending on just one space variable and guarantee that one can
dispense with the concept of renormalized solution used in the existence proof of
DiPerna and Lions. The key to this result is an inequality related to e n e r g y
conservation. The proofs can be extended to the case of a gas in a slab [152].

REFERENCES

[1] BOLTZMANNL., Theoretical Physics and Philosophical Problems, edited by B. Mc GUINNESS (D. Reidel Publishing Company, Dordrecht), 1974.
[2] CERCmNAm C., Mathematical Methods in Kinetic Theory, revised edition (Plenum Press,
New York, N.Y.) 1990.
[3] CERCIGNANI C., The Boltzmann Equation and its Applications (Springer, New York, N.Y.)
1988.
[4] CERCmNANI C., ILLNER R. and PULVIRENTIM., The Mathematical Theory of Dilute Gases
(Springer-Verlag, New York, N.Y.) 1994.
[5] KOGAN M. N., Rarefied Gas Dynamics (Plenum Press, New York, N.Y.) 1969.
[6] TRUESDELL C. and MUNCASTEa R. G., Fundamentals of Maxwell's Kinetic Theory of a
Simple Monatomic Gas (Academic Press, New York, N.Y.) 1980.
[7] THOMSONW. (LORD KELWN), Proc. R. Soc. Edinb., 8 (1874) 325.
[8] LOSCHMIDTJ., Wien. Ber., 73 (1876) 128.
[9] CERCIGNANI C., Alma Mater Studiorum, 2 (1989) 37.
[10] POINCARI~ H., Acta Math., 13 (1890) 1.
[11] ZERMELO E., Ann. Phys. (Leipzig), 54 (1896) 485.
[12] POINCAR~ H., Rev. Metaphys. Mor., 1 (1893) 534.
[13] BOLTZMANNL., Ann. Phys. (Leipzig), 57 (1896) 773.
[14] KUNDT A. and W~BUaG E., Pogg. Ann. Phys., 155 (1875) 337.
[15] MAXWELLJ. C., Philos. Trans. R. Soc., 170, Appendix (1879) 231.
[16] v. SMOLUCHOWSKIM. S., Philos. Mag., 46 (1898) 192.
[17] KNUDSEN M., The Kinetic Theory of Gases (Methuen, London) 1950.
[18] KUSCER I., in Fundamental Problems in Statistical Mechanics, IV, edited by E. G. D.
COHEN and W. FISZDON (Ossolineum, Warsaw) 1978.
[19] CERCmNANI C., in Proceedings of the Workshop on Hypersonic Flows for Reentry
Problems, Vol. I (INRIA, Antibes) 1990, p. 9.
[20] DE BOER J. H., The Dynamical Character of Adsorption (Clarendon Press, Oxford)
1953.
[21] CERCIGNANI C., in Transport Theory, edited by G. BIRKHOFF et al., Vol. I (SIAM-AMS,
Providence) 1968, p. 249.

AERODYNAMICALAPPLICATIONSOF THE BOLTZMANNEQUATION

37

[22] KU~ER I., in Transport Theory Conference (AEC Report 0R0-3588-1, Blacksburgh, Va.)
1969.
[23] CERCIGNANI C. and LAMPIS M., Transport Theory Stat. Phys., 1 (1971) 101.
[24] KU~ER I., Surf. Sci., 25 (1971) 225.
[25] CERCIGNANI C., Transport Theory Stat. Phys., 2 (1972) 27.
[26] SHEN S. F. and KUSCER I., in Rarefied Gas Dynamics, edited by D. DINI et al., Vol. I
(Editrice Tecnico-Scientifica, Pisa) 1974, p. 109.
[27] B~RWINKELK. and SCHIPPERSS., in Rarefied Gas Dynamics: Space-Related Studies, edited
by E. P. MUNTZ, D. P. WEAVER and D. H. CAMPBELL(AIAA~ Washington) 1989, p. 487.
[28] GOODMANF. O. and WACHMANH. Y., Dynamics of Gas-Surface Scattering (Academic Press,
New York, N.Y.) 1976.
[29] BAULE B., Ann. Phys. (Leipzig), 44 (1914) 145.
[30] LOGAN R. M. and STICKNEY R. E., J. Chem. Phys., 44 (1966) 195.
[31] B)~RWINKELK. and SCHMIDT H. J., in Rarefied Gas Dynamics, edited by S. S. FISCHER,
Vol. I (AIAA, New York, N.Y.) 1981, p. 150.
[32] LOGAN R. M. and KECK J. C., J. Chem. Phys., 49 (1968) 860.
[33] HURLBUTF. C., in Rarefied Gas Dynamics, edited by M. BECKER and M. FIEBIG, Vol. II,
Ax-E. 3-1 (DFVLR, Porz-Wahn) 1974.
[34] TULLY J. G., J. Chem. Phys., 73 (1980) 1975.
[35] TULLY J. G., Annu. Rev. Phys. Chem., 31 (1980) 319.
[36] DOLL J. D., Chem. Phys., 3 (1974) 257.
[37] DOLL J. D., J. Chem. Phys., 61 (1974) 954.
[38] MASEL R. I., MERRIL R. P. and MILLER W. H., Surf. Sci., 46 (1974) 681.
[39] MASEL R. I., MERRIL R. P. and MILLER W. n., J. Chem. Phys., 64 (1976) 45.
[40] DION D. R. and DOLL J. D., Surjf. Sci., 58 (1976) 415.
[41] TUTSCHIDAA., Surj(. Sci., 14 (1969) 375.
[42] CABRERA N., CELLI V. and MANSON J. R., Phys. Rev. Lett., 22 (1969) 346.
[43] CABRERA N., CELLI V., GOODMAN F. O. and MANSON J. R., Sur9(. Sci., 19 (1970) 67.
[44] ARMAND G. and LAPUJOULADEJ., in Rarefied Gas Dynamics, edited by R. CAMPARGUE,
Vol. II (CEA, Paris) 1979, p. 1329.
[45] NOCILLAS., in Rarefied Gas Dynamics, edited by L. TALBOT(Academic Press, New York,
N.Y.) 1961, p. 169.
[46] HURLBUT F. and SHERMAN F. S., Phys. Fluids, 11 (1968) 486.
[47] KUS(~ERI., MOZINAJ. and KRIZANICF., in Rarefied Gas Dynamics, edited by D. DINI et as
Vol. I (Editrice Tecnico-Scientifica, Pisa), 1974, p. 97.
[48] COWLINGT. G., J. Phys. D, 7 (1974) 781.
[49] WILLIAMS M. M. R., J. Phys. D, 4 (1971) 1315.
[50] CERCIGNANI C., in Rarefied Gas Dynamics, edited by D. DINI et al., Vol. I (Editrice
Tecnico-Scientifica, Pisa) 1974, p. 75.
[51] CERCIGNANI C., Theory and Application of the Boltzmann Equation (Scottish Academic
Press, Edinburgh; Elsevier, New York) 1975.
[52] CERCIGNANIC., LAMPISM. and LENTATIA., in Rarefied Gas Dynamics 19, Vol. 2 edited by
J. HARVEY and G. LORD (Oxford University Press, Oxford) 1995, p. 1190.
[53] LEGGE H., Heat transfer and forces on a LiF single-crystal disk in hypersonic flow
(DLR-IB 222-92 A 14 Report) 1992.
[54] BLANCHARDn. C., Paper No. ICAS 86-4.10.1, 15th ICAS Congress, London (September
1986).
[55] HURLBUTF. C., in Rarefied Gas Dynamics: Space-Related Studies, edited by E. P. MUNTZ,
D. P. WEAVER and D. H. CAMPBELL (AIAA, Washington) 1989, p. 419.
[56] HURLBUTF. C., in Thermophysical Aspects of Re-entry Flows, edited by J. N. Moss and
C. D. SCOTT (AIAA, New York, N.Y.) 1986.
[57] HULPKE E. and MANN K., Surf. Sci., 133 (1983) 171.
[58] TENSER A. D., GILLEN K. T., HORN T. C. M., LOS J. and KLEIN A. W., Phys. Rev.
Lett., 52 (1984) 2183.

38

C. CERCIGNANI

[59] TENNERA. D., SAXONR. P., GILLEN K. T., HARRISOND. E., HORN T. C. M. and KLEIN A. W.,
Surf. Sci., 172 (1986) 121.
[60] TENNERA. D., GILLENK. T., HORNW. C. M., LOS J. and KLEIN A. W., Surf. Sci., 172 (1986) 90.
[61] KOLODNEY E., AMIRAV A., ELBER R. and GERBER R. B., Chem. Phys. Lett., 113 (1985)
303.
[62] BELLOMO N., DANKERT C. and LEDGE H., in Rarefied Gas Dynamics, edited by O. M.
BELOTSERKOVSKII, M. N. KOGAN, S.S. KUTATELADZE and A. K. REBROV, Vol. I (Plenum
Press, New York, N.Y.) 1985, p. 421.
[63] CERCIGNANI C. and LAMPIS M., J. Appl. Math. Phys. (ZAMP), 27 (1976) 733.
[64] BECKER M., ROBBEN F. and CATTOLICAR., AIAA J., 12 (1974) 1247.
[65] HURLBUT F. C., AIAA Paper 87-1545 (1987).
[66] CERCIGNANI C. and FREZZOTTI A., in Rarefied Gas Dynamics: Theoretical and
Computational Techniques, edited by E. P. MUNTZ, D. P. WEAVER and D. H. CAMPBELL,
(AIAA, Washington) 1989, p. 552.
[67] HERMINA W. L., AIAA Paper 87-1547 (1987).
[68] R. G. LORD, Phys. Fluids A 3, (1991) 706.
[69] J. R. MANSON J. R. and CELLI V., Surf. Sci., 24, (1971) 495.
[70] WANG CHANGC. S. and UHLENBECK G. E., in Studies in Statistical Mechanics, edited by
J. DE BOER and G. E. UHLENBECK, Vol. II, Part c (North-Holland, Amsterdam) 1964.
[71] WALDMANNL., in Handbuch der Physik, edited by S. FLTJGGE,Vol. XII (Springer-Verlag,
Berlin) 1958.
[72] WALDMANNa., in The Boltzmann Equation. Theory and Application, edited by E. G. D.
COHEN and W. THIRRING(Springer-Verlag, Vienna) 1973, p. 107.
[73] SNIDER R. F., J. Chem. Phys., 32 (1960) 1051.
[74] BOLTZMANNL., Sitzungsb. Akad. WiNs., 66 (1872) 275.
[75] LORENTZ H. A., Sitzungsb. Akad. WiNs., 95 (1887) 115.
[76] BOLTZMANNL., Sitzungsb. Akad. WiNs., 95 (1887) 153.
[77] BOLTZMANNL., Vorlesungen i~ber Gastheorie (J. A. Barth, Leipzig) 1895-1898.
[78] TOLMAN R. C., The Principles of Statistical Mechanics (Clarendon Press, Oxford) 1938.
[79] UHLENBECKG. E., in The Boltzmann Equation. Theory and Application, edited by E. G. D.
COHEN and W. THIRRING (Springer-Verlag, Vienna) 1973, p. 107.
[80] CERCIGNANIC. and LAMPIS M., J. Stat. Phys., 26 (1981) 795.
[81] LORDI J. A. and MATES R. E., Phys. Fluids, 13 (1970) 291.
[82] CURTISS C. F. and MUCKENFUSS C., J. Chem. Phys., 29 (1958) 1257.
[83] JEANS J. H., Q. ,]. Pure Appl. Math., 25 (1970) 224.
[84] DAHLER J. S. and SATHER N. F., J. Chem. Phys., 38 (1962) 2363.
[85] SANDLER S. I. and DAHLER N. F., J. Chem. Phys., 47 (1967) 2621.
[86] BRYAN G. H., Rep. Br. Ass. Advant. Sci., (1894) 83.
[87] KUSCER I., in Operator Theory: Advances and Applications, Vol. 51 (Birkh~iuser Verlag,
Basel) 1991, p. 180.
[88] BORGNAKKEC. and LARSEN P. S., J. Comput. Phys., 18 (1975) 405.
[89] BIRD G. A., Molecular Gas Dynamics and the Direct Simulation of Gas Flows (Clarendon
Press, Oxford) 1994.
[90] BENETTIN G., GALGANI L. and GIORGILL1 A., Commun. Math. Phys., 113 (1987) 87.
[91] BENETTIN G., GALGANI L. and GIORGILLI n., Commun. Math. Phys., 121 (1989) 557.
[92] BOLTZMANNL., Nature, 51 (1895) 413.
[93] BIRD G. A., in Rarefied Gas Dynamics, edited by R. CAMPARGUE(CEA~ Paris) 1979, p. 365.
[94] BIRD G. A., in Rarefied Gas Dynamics, Part I (AIAA, Washington) 1981, p. 239.
[95] HILBERT D., Math. Ann., 72 (1912) 562.
[96] CHAPMAN S., Proc. R. Soc. London, Ser. A, 93 (1916-17) 1.
[97] ENSKOGD., Kinetische theorie der Vorggnge in mfissig verdi~nnten Gasen. L Allgemeiner
Teil (Almqvist and Wiksell, Uppsala) 1917.
[98] CHAPMAN S. and COWLING T. G., The Mathematical Theory of Non-Uniform Gases,
(Cambridge University Press, London) 1940.

AERODYNAMICALAPPLICATIONS OF THE BOLTZMANN EQUATION

39

[99] FERZIGER J. H. and gAPER H. G., Mathematical Theory of Transport Processes in Gases
(North-HoUand, Amsterdam) 1972.
[100] GRAD H., Commun. Pure Appl. Math., 2 (1949) 331.
[101] MOTT-SMITH H., Phys. Rev., 82 (1951) 885.
[102] CERCIGNANI C., J. Star. Phys., 1 (1969) 297.
[103] LOYALKA S. K., J. Chem. Phys., 58 (1973) 354.
[104] CERCIGNANIC. and PAGANIC. D., in Rarefied Gas Dynamics, edited by C. BRUNDIN,Vol. I
(Academic Press, New York, N.Y.) 1967, p. 555.
[lO5] SONEY. and AOKI K., in Rarefied Gas Dynamics, edited by R. CAMPARGUE(CEA, Paris)
1979, p. 1207.
[106] CERCIGNANI C., PAGAN1 C. D. and BASSANINI P., Phys. Fluids, 11 (1968) 1399.
[107] GRAD H. in Transport Theory, edited by R. BELLMAN et al. (American Mathematical
Society, Providence, R.I.) 1969, p. 269.
[108] BHATNAGAR P. L., E. P. GROSS and KROOKM., Phys. Rev., 94 (1954) 511.
[109] WELANDER P., Arkiv Fysik, 7 (1954) 507.
[110] NORDSIEKA. and HlCXS B., in Rarefied Gas Dynamics, edited by C. L. BRUNDIN,Vol. I I
(Academic Press, New York, N.Y.) 1967, p. 695.
[111] YEN S. M., HICKS B. and OSTEEN R. M., in Rarefied Gas Dynamics, edited by M. BECKER
and M. FIEBIG, Vol. I, A.12-1-A.12-10 (DFVLR-Press, Porz-Wahn) 1974.
[112] ARISTOV V. V. and TCHEREMISSINE F. G., U. S. S. R. Comput. Math. Math. Phys., 20
(1980) 208.
[113] TCHEREMISSINE F. G., U.S.S.R. Comput. Math. Math. Phys., 25 (1985) 156.
[114] TCHEREMISSINE F. G., in Rarefied Gas Dynamics: Theoretical and Computational
Techniques, edited by E. P. MUNTZ, D. P. WEAVER and D. H. CAMPBELL (AIAA,
Washington) 1989, p. 343.
[115] CERCIGNANI C. and FREZZOTTI A., in Rarefied Gas Dynamics: Theoretical and
Computational Techniques, edited by E. P. MUNTZ,D. P. WEAVER and D. H. CAMPBELL
(AIAA, Washington) 1989, p. 552.
[116] BIRD G. A., Phys. Fluids, 13 (1970) 2676.
[117] KOURA K., Phys. Fluids, 29 (1986) 3509.
[118] BELOTSERKOVSKIIO. i . and YANITSKIIV., Zhurn. Vyvh. Mat. Mat. Fiz., 15 (1975) 1195 (in
Russian).
[119] DESHPANDE S. M., Dept. Aero. Engng. Indian Inst. Science Rep., 78, FM4 (1978).
[120] NANBU K., J. Phys. Soc. Jpn., 49 (1980) 2042.
[121] KOURA K., J. Phys. Soc. Jpn., 50 (1981) 3829.
[122] NANRU K., J. Phys. Soc. Jpn., 50 (1981) 3831.
[123] BABOVSKY H., Europ. J. Mech. B/Fluids, 8 (1989) 41.
[124] BABOVSKY H. and ILLNER R., SIAM J. Numer. Anal., 26 (1989) 45.
[125] WAGNER W., J. Stat. Phys., 66 (1994) 1011.
[126] PULVIRENTI M., WAGNER W. and ZAVELANI M. B., Europ. J. Mech~ B, 7 (1994) 339.
[127] GROPENGIEfiER F., NEUNZERT N. and STRUCKMEIER J., ill Venice 1989: The State of Art in
Applied and Industrial Mathematics, edited by R. SPIGLER (Kluwer, Dordrecht) 1990.
[128] STEFANOV S. and CERCIGNANIC., Europ. J. Mech. B, 451 (1994) 543.
[129] CERCIGNANI C. and STEFANOV S., Transp. Theory Stat. Phys., 21 (1994) 371.
[130] STEFANOV S. and CERCIGNANIC., J. Fluid Mech., 256 (1993) 199.
[131] STEFANOV S. and CERCIGNANI C., ill Proceedings of the International Symposium on
Aerospace and Fluid Science (Institute of Fluid Science, Tohoku University, Sendai) 1994,
p. 490.
[132] STEFANOV S. and CERCIGNANIC., Europ. J. Mech. B, 13 (1994) 93.
[133] MEIBURG E., Phys. Fluids, 29 (1986) 3107.
[134] BIRD G. A., Phys. Fluids, 30 (1987) 364.
[135] NANBU K., WATANABE Y. and IGARASHIS., J. Phys. Soc. Jpn, 57 (1988) 2877.
[136] DI PERNA n. and LIONS P. L., Ann. Math., 130 (1989) 321.
[137] HAMDACHE K., Arch. Rat. Mech. Analysis, 119 (1994) 309.

40

c. CERCIGNANI

[138] ARKERYD L., CERCIGNANI C. and ILLNER R., Commun. Math. Phys., 142 (1991) 285.
[139] DESVILLETTES L., Arch. Rat. Mech. Anal., 110 (1990) 73.
[140] CERCIGNAm C., Rend. Mat. Appl., 10 (1990) 77.
[141] ARKERYD L., Studies Appl. Math., 87 (1994) 283.
[142] LIONS P. L., Cahiers de Mathdmatiques de la ddcision n ~ 9301 (CEREMADE, Paris)
1993.
[143] ARKERYD L., and CERCIGNANI C., Arch. Rat. Mech. Anal., 125 (1993) 271.
[144] ARKERYD L. and MASLOVA N., Preprint ISSN 03407-2809 (University of GSteborg,
G6teborg) 1994.
[145] CERCIGNANI C. in Proceedings of the Meeting in Honor of J. P. Guiraud, to appear,
1994.
[146] UKAI S., in Pattern and Waves. Qualitative Analysis of Nonlinear Differential Equations,
37 (1986).
[147] UKA] S. and ASANO K., Proc. Jpn. Acad., 56 (1990) 12.
[148] L~ONS P. L., Commun. Partial Diff. Equations, 19 (1994) 335.
[149] CERCIGNANI C., Appl. Math. Lett., 5 (1992) 59.
[150] CERCIGNANI C., Appl. Math. Lett., 8 (1995) 53.
[151] CERCIGNANI C., Appl. Math. Lett., 8 (1995) 95.
[152] CERCIGNANI C. and ILLNER Z., Arch. Rat. Mech. Anal., to appear 1995.

9 by Societfi Italiana di Fisica


Propriet~ letteraria riservata
Direttore responsabile: RENATO ANGELO RICCI
Questo fascicolo ~ stato realizzato in fotocomposizionedalla Monograf, Bologna
e stampato dalla tipografia Compositori,Bologna
nel mese di novembre 1995
su carta patinata ecologicachlorine-free
prodotta dalle Cartiere del Garda S.p.A., Riva del Garda (TN)

Questo periodico
iscritto
all'Unione Stampa
Periodica Italiana

Вам также может понравиться