You are on page 1of 11

PERGAMON International Journal of Engineering Science 36 (1998) 1313±1323

Bulk viscosity in the Navier±Stokes equations


George Emanuel *
School of Aerospace and Mechanical Engineering, The University of Oklahoma, Norman, OK 73019, USA
Received 23 January 1996; accepted 24 November 1997

Abstract

This article discusses the physics associated with the bulk viscosity coecient mb as it appears in the
compressible Navier±Stokes equations. Thus, the active rotational and vibrational modes of a
polyatomic molecule are discussed with emphasis on the importance of local thermodynamic
equilibrium. This condition is necessary if mb is to only depend on the thermodynamic state, as required
for the Navier±Stokes equations. A new perturbation formulation is provided for the absorption and
dispersion coecients and the entropy production that holds even when mb greatly exceeds the shear
viscosity, as is the case for gases such as CO2. A comparison for CO2 shows excellent agreement
between exact results and the low-frequency formula used for the absorption coecient. This agreement
stems from a large value for the Peclet number. # 1998 Elsevier Science Ltd. All rights reserved.

1. Introduction

Studies of sonic attenuation, from which the bulk viscosity can be determined, started in the
later part of the nineteenth century. As demonstrated by review articles [1±9], a number of
which list hundreds of references, it was a proli®c research topic. Over time, the experimental
technique improved with the use of crystals for generating and receiving the signal, better
instrumentation, and with careful gas handling techniques [10]. Most of this work, however,
both experimental and theoretical, is not pertinent to continuum ¯uid dynamics, which is the
focus of this article. Our interest is in gas ¯ows that are governed by the steady or unsteady,
compressible, Navier±Stokes (N±S) equations, which here includes the mass and energy
equations. In this context, thermodynamic and transport properties, including the bulk
viscosity, mb, are functions only of the thermodynamic state. Moreover, these parameters are
real, frequency independent, and a Kramers±Kronig relation [11] is not relevant. The condition
for mb to be real and frequency independent is discussed shortly.

* Tel.: +1-405-325-5011; fax: +1-405-325-1088.

0020-7225/98/$19.00 # 1998 Elsevier Science Ltd. All rights reserved.


PII: S 0 0 2 0 - 7 2 2 5 ( 9 8 ) 0 0 0 2 0 - 2
1314 G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323

For aerodynamics and the ¯uid dynamics of air, the terms containing mb in the
N±S equations are often inconsequential. Although mb for air is not zero, it has long
been conventional to take it as zero, i.e. Stokes hypothesis. For a laminar boundary layer,
Stokes hypothesis is appropriate, since mb introduces a negligible third-order e€ect [12].
Stokes hypothesis, however, cannot be used in the analysis of shock wave structure,
except for a monatomic gas. In addition, there are many gases for which the ratio a^ ˆ mb =m,
where m is the shear viscosity, is large compared to unity. This circumstance is the
primary subject of this article; it has been discussed for a laminar boundary layer [13],
turbulence [14], and is of interest for planetary atmosphere aerodynamics [15, 16]. For instance,
the Venus and Mars atmospheres are largely CO2, for which a^ is about 2100 [17]. Another
application occurs when a vapor exhibits anomalous behavior [18, 19], which happens near
the coexistence curve and when the speci®c heats are large. Large speci®c heats, however,
also result in a large value for a^ . Thus, modeling of the complex viscous behavior of
a retrograde [18] or Bethe±Zel'dovich±Thompson [19] ¯uid requires an accurate knowledge
of mb.
The bulk viscosity is determined by propagating a weak ultrasonic signal one-dimensionally
through a gas. As is customary, wave attenuation is assumed to be caused by heat conduction
and viscosity. For a polyatomic molecule, the bulk viscosity is ®nite because of the relaxation
of internal rotational and vibrational modes. Hence, there is energy transfer between these
modes and the translational mode. Rotational relaxation is generally quite rapid requiring only
a few molecular collisions. In this situation, there are sucient collisions to easily maintain
local thermodynamic equilibrium (LTE) with the translational mode. On the other hand,
vibrational relaxation may require several thousand collisions for equilibration. The relaxation
time, ti, of internal mode i is given as pti, where p is the pressure and where this combination
depends only on the temperature. For vibration, this dependence is provided by a Landau±
Teller type formula [20].
As Tisza [17] has pointed out, a frequency-independent bulk viscosity suces when oti<<1
for all active internal modes, where o is the angular frequency of the signal. In this situation,
there are sucient collisions to maintain LTE. At a suciently high frequency, of course, a
rate equation approach with a set of characteristic relaxation times is required. In other words,
if oti is not suciently small a complex valued, frequency dependent bulk viscosity is obtained.
In this circumstance, acoustic theory [21, 22] is utilized that is based on the Euler equations.
Consequently, a frequency dependent bulk viscosity depends on the relaxation processes but
not on the Reynolds or Prandtl numbers. For a continuum ¯ow, the N±S equations implicitly
assume LTE conditions for internal modes, and any bulk viscosity measurement should satisfy
Tisza's condition if the parameter is to be used in these equations. In an ultrasonic
measurement, o exceeds 20 kHz. This frequency is usually much larger than what is
encountered in an unsteady ¯ow, including a turbulent one. Hence, an ultrasonic mb
measurement, performed under LTE conditions, should hold for steady and most unsteady
N±S ¯ows. In passing, we note that at a high density the modi®ed Enskog theory has mb
proportional to the square of the density [23], even for a monatomic gas. This phenomenon is
not associated with the relaxation of internal modes and mb is generally much smaller than that
of a polyatomic molecule with one or more active vibrational modes. This high density aspect
is excluded from the discussion.
G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323 1315

Although the oti requirement is important, it apparently is often overlooked. For example,
Ref. [10] provides the absorption in CO2 over a temperature range and for a 53±147 kHz
frequency range. The value of pt for CO2±CO2 collisions [13] at 303 K is 6.8  10ÿ6 atm s.
Here, t is the relaxation time of the (0,11,0) vibrational state, which is the lowest lying n2
excited state. Since relaxation times are given as pt, we have the LTE requirement
o… pt†  p, …1†
which can be met by reducing o, increasing p, or both. An increase in temperature can also be
used to reduce pt. Alternatively, a sucient reduction in the temperature can be even more
e€ective if it results in some or all of the vibrational modes becoming inactive. With the above
CO2 pt value, the pressure should greatly exceed 1 atm at 147 kHz for the LTE condition to
apply. The authors of Ref. [10], however, do not state their experimental pressure. It appears
probable that condition (1) is frequently not met, not only in this reference, but in much of the
ultrasonic absorption literature.
In a review of energy transfer, Callear and Lambert [24] note that acoustic ``...methods
provide an accurate means of investigating translation±vibration and translation±rotation
transfer''. They also remark that the ultrasonic absorption and dispersion approach and
corresponding shock tube results are both in accord with theory. By theory, they mean a
relaxation process based on quantum mechanics. In the shock tube case, the shock is treated as
a discontinuity and rate equations for the relaxation process are numerically solved in
conjunction with the one-dimensional Euler equations. Thus, in a ¯ow where e€ects associated
with the shear viscosity are negligible, the energy transfer is a thermal e€ect that is accounted
for with rate, or master, equations. In an LTE viscous ¯ow, the energy transfer would be
represented by the bulk viscosity.
Prangsma et al. [25] ultrasonically determined mb for N2, CO, CH4, and CD4 between 77 and
300 K. The bulk viscosity is extracted from the conventional, low-frequency N±S model
discussed later. As a check, they also measured the acoustic attenuation for neon. This
measurement quite accurately veri®es Stokes hypothesis for a monatomic gas. In addition, the
measurements are free of excessive scatter. The method of data reduction and comparison with
a N±S based theory yields a self-consistent result for the bulk viscosity. The corresponding
rotational relaxation times are evaluated, since the vibrational modes are inactive at these low
temperatures. As a further consequence, t is quite small and LTE conditions prevail. These
measurements yield mb/m = 0.73 and pt = 7.9  10ÿ10 atm s for N2 at 293 K. Since O2 and N2
have similar values for their characteristic vibrational temperatures (2240 K for O2, 3350 K for
N2), air is expected to have a value for mb/m close to 0.7. This is con®rmed by the NTP value
of 2/3 provided by Truesdell [2]. With the above pt value, and inactive vibrational modes,
condition (1) is satis®ed when 7.9  10ÿ4o<<p, where o is in MHz and p in atm. Hence, even
in a turbulent ¯ow the LTE condition easily prevails.
When condition (1) holds, kinetic theory predicts [20, 26]
X
mb 0 … pt†j ,
j

where the summation is over all active rotational and vibrational modes. Only molecules
with one or more active vibrational modes have the possibility for a substantial value for
1316 G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323

the bulk viscosity, relative to the value for the shear viscosity. Meador et al. [27] have
questioned whether or not the bulk viscosity can be viewed as a relaxation parameter. Their
contention that mb is not properly a relaxation parameter runs counter to the discussion in
Refs. [24±26, 28].
The next section presents a systematic perturbation procedure for obtaining the attenuation
and dispersion coecients of a sinusoidal wave in a N±S ¯ow that adheres to LTE conditions.
A novel feature is the determination of these parameters by ®nding the root of a polynomial.
As is customary, a formula for mb is obtained in the low frequency limit. The derivation is free
of assumptions, such as a large Reynolds number or Stokes hypothesis. It is thus applicable to
gases with very large a^ values. We also resolve the question of the relevance of entropy
production. Results are presented for CO2. The ®nal section summarizes our ®ndings and
brie¯y discusses several gaps in our knowledge.

2. Analysis

2.1. Governing equations

A plane, acoustic, single frequency wave is considered that passes through a thermally and
calorically perfect gas. For the normalization, introduce
l*
x* ˆ l*x, t* ˆ t, u* ˆ a*0 u, r* ˆ r*0 r, p* ˆ p*0 p, T * ˆ T *0 T,
a*0
gR*a*0
~ ˆ m~ *0 m,
m* ~ k ˆ k*0 k, s_ *irr ˆ s_ irr ; …2†
l*
where a symbol without an asterisk is dimensionless, a zero subscript indicates the undisturbed
gas, a is the speed of sound [=(gRT)1/2], and the e€ective viscosity m~ is
ÿ 
~ ˆ 43 m* ‡ m*b ˆ m* 43 ‡ a^ :
m* …3†

In Eqs. (2), x, l*, t, u, r, g, R, T, k, and Çsirr are distance, characteristic length, time, ¯ow speed,
density, ratio of speci®c heats, gas constant, temperature, thermal conductivity, and the rate of
entropy production, respectively. The nondimensional N±S equations are [29]
@r @r @u
‡u ‡r ˆ 0, …4a†
@t @x @x
   
@u @u 1 @p 1 @ @u
r ‡u ˆÿ ‡ m~ , …4b†
@t @x g @x Re @x @x
     2  
@T @T g ÿ 1 @p @p gÿ1 @u 1 @ @T
r ‡u ˆ ‡u ‡ m~ ‡ k , …4c†
@t @x g @t @x Re @x Pe @x @x

p ˆ rT, …4d†
G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323 1317

where the Reynolds Re and Peclet Pe numbers are given by


r*0 a*0 l* ^
Re
Re ˆ ˆ , …5a†
m~ *0 4=3 ‡ a^
   
gR*m*0 ^ 4 ^ Re,
^
Pe ˆ Pr Re ˆ Re ˆ Pr ^
‡ a Re ˆ Pr …5b†
…g ÿ 1†k*0 3
^ and Pr
and Pr is the Prandtl number. The Re ^ numbers are conventional, i.e. they are based on
~
m, whereas Re and Pr are based on the e€ective viscosity m.

2.2. Wall conditions

The disturbance is caused by an oscillating surface that vibrates with an angular frequency
o*, i.e. x*w (t*) = l*w cos (o*t*). Now set
l*w o* a*0
eˆ , l* ˆ , …6†
a*0 o*
to obtain uw= ÿ e sin t for the wall speed. With typical values for l*w , o*, and a*0 , e is a small
constant. In an isentropic ¯ow, denoted by an i subscript, the wavelength l*i equals a*0 /o* and
the characteristic length is l*=l*i . When the ¯ow is viscous and diabatic, there is dispersion
and l* then slightly di€ers from l*i . The wall motion amplitude l*w generates a wave whose
length is approximately l*i . The length ratio, l*w /l*i , equals e, which is very small. Typically,
l*i is a few millimeters but corresponds to many tens-of-thousands of mean free paths.
Consequently, the maximum values of the ¯uid speed and temperature gradients are quite
modest. The Newtonian/Fourier linear approximations, utilized in the N±S equations,
therefore, holds.
There is a temperature or heat transfer condition at the vibrating surface. This condition
would be required if the gas motion adjacent to the surface is considered. The subsequent
analysis, however, is for a free wave at some distance from the wall and a surface condition is
unnecessary.

2.3. Perturbation solution

A ®rst-order perturbation procedure is introduced


u ˆ eu 0 …x,t†, r ˆ 1 ‡ er 0 …x,t†, p ˆ 1 ‡ ep 0 …x,t†, T ˆ 1 ‡ eT 0 …x,t†,
m~ ˆ 1 ‡ em~ 0 , k ˆ 1 ‡ ek 0 : …7†
These relations are substituted into Eqs. (4a)±(4d) and, after simpli®cation, yield
g 0 0 0 0 0
T txxxx ‡ T xxxx ÿ …g ‡ Pr†T ttxx ÿ PeT txx ‡PeT ttt ˆ 0, …8†
Re
1318 G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323

where the subscripts on T ' represent partial derivatives. The general solution for a right-
running wave is given by

T 0 …x,t† ˆ Ae…iÿb†xÿiat ˆ Aeÿbx ei…xÿat† , …9†

where A is a constant, i = (ÿ1)1/2, and a and b are real, positive, nondimensional


parameters that are to be determined. The a parameter represents dispersion, where a is
the actual wave speed normalized by a*0. The b parameter is the attenuation per isentropic
wavelength l*i .
If viscous forces and heat conduction are neglected, i.e. Re 4 1, Pe 4 1, then a = 1 and
b = 0. In this circumstance, the familiar isentropic result is obtained for a wave that
propagates at a speed a*0 without attenuation. This is also evident from Eq. (8), which reduces
to a second-order wave equation for T '.
Our objective is to determine a and b in terms of g, Pe, and Re. Toward this end, the
substitution of Eq. (9) into Eq. (8) results in

F ˆ ÿ…Pe ‡ gRe†…1 ÿ b2 †a2 ‡ 2‰Pe Re ‡ 2g…1 ÿ b2 †Šba ‡ Re…1 ÿ 6b2 ‡ b4 † ˆ 0, …10a†

G ˆPe Rea3 ÿ 2…Pe ‡ gRe†ba2 ÿ ‰Pe Re…1 ÿ b2 † ‡ g…1 ÿ 6b2 ‡ b4 †Ša


‡ 4Reb…1 ÿ b2 † ˆ 0: …10b†

A separable solution is computationally convenient. To simplify the algebra, de®ne

a ˆ 1 ÿ b2 , b ˆ 1 ÿ 6b2 ‡ b4 , c ˆ Pe Re, d ˆ Pe ‡ gRe, …11†

with the result

F ˆ ÿada2 ‡ 2…c ‡ 2ga†ba ‡ bRe ˆ 0, …12a†

G ˆ ca2 ‡ 2dba2 ÿ …ac ‡ gb†a ‡ 4bReb ˆ 0: …12b†

Since Eq. (12a) is quadratic in a, we obtain

a ˆ y‰1 ‡ …1 ‡ f†1=2 Š, …13†

where

…c ‡ 2ga†b ‰Pe Re ‡ 2g…1 ÿ b2 †Šb


yˆ ˆ , …14a†
ad …1 ÿ b2 †…Pe ‡ gRe†

abdRe …1 ÿ b2 †…1 ÿ 6b2 ‡ b4 †Re…Pe ‡ gRe†


fˆ ˆ …14b†
…c ‡ 2ga†2 b2 b2 ‰Pe Re ‡ 2g…1 ÿ b2 †Š2
G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323 1319

and Eq. (12b) becomes


 2     3  
2ga 2ga ad2 4 ad Re 2 4 2ga
32a 1 ‡ 1‡ ÿ 2 b ‡4 …1 ‡ b † ‡ d 1 ‡
c c c c4 c
 2   2 
ab gabd b2 Re 2a3 d b 2ga
 2
…6Re ÿ d† ÿ 3 …b ‡ 2a † ÿ 2 ÿ 2 ‡ …ac ‡ gb† 1 ‡ b2
c2 c c c c c
 
ad bRe 2
ÿ b 2 …ac ‡ gb† ÿ ˆ 0: …15†
c c
This is a sixteenth-degree polynomial in b that is independent of a and whose solution
only depends on g, Pe, and Re. In the isentropic limit, b equals zero. Hence, the physical
solution is the smallest, positive, real value for b, which is the root closest to bi=0. Once b is
established, Eq. (13) provides a. Subsequent computational results are based on Eqs. (13) and
(15).
The rate of entropy production is written as [29]
  2  2   
1 m~ @u 1 k @T e2 02 1 02
s_ irr ˆ ‡ ˆ ux ‡ Tx
p Re @x …g ÿ 1†Pr Re T @x Re …g ÿ 1†Pr
     
2 2 g…1 ÿ b2 † 2 1 1 2gb 2 1 1 ‡ b2
ˆe A ‡ aÿ ‡ eÿ2bx , …16†
…g ÿ 1†Pe Re gÿ1 Pe Re …g ÿ 1†Pe

where complex conjugates are used when squared terms, such as ux02 , are evaluated. Hence, Çsirr is
second order, although it depends on ux0 and Tx0 , which are ®rst-order quantities. Thus, Çsirr is
small but not zero, and has a simple dependence on a. The rate of entropy production,
however, may not be small for retrograde ¯uids, since these have very small values for (gÿ1).
The use of a relation based on an isentropic assumption, or the Euler equations, is thus not
consistent for evaluating a N±S parameter. The appearance of a second-order parameter in a
®rst-order analysis is not unusual. For instance, a comparable situation occurs in boundary
layer theory, where the transverse velocity component is of higher order than the longitudinal
component. Nevertheless, this component cannot be set equal to zero in ®rst-order boundary
layer theory.
Previous N±S based theories, of course, have used an energy equation. Quite often, however,
this equation has an isentropic form [2, 6, 9].
Equations (5a) and (5b), along with l*=l*i , allow the Peclet and Reynolds numbers to be
written as

g2 R* p*0
Pe ˆ , …17a†
…g ÿ 1†k*0 o*

g 1 p*0
Re ˆ : …17b†
m*0 4=3 ‡ a^ o*
1320 G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323

Both Pe and Re are inversely proportional to the frequency and, in turn, provide the frequency
dependence for a and b. Although these parameters are frequency dependent, within the
context of this theory, a^ is not frequency dependent.
Suppose the gas properties g, R*, k*0 , m*0 , p*0 , and a*0 are known. Measured values of a and b,
at a given frequency, then each provide a value for a^ , and hence for m*b . The separate
absorption and dispersion values for a^ should agree. Moreover, a^ should be independent of
frequency when the LTE condition prevails. Consequently, measurements at other
frequencies should yield the same a^ value. Separate a and b values for a^ and its independence
on frequency can be used to verify the theory. One diculty with this approach is that a is
very close to unity when condition (1) holds, and dispersion generally does not yield a separate
value for a^ .

2.4. Low frequency limit

For comparison purposes, assume Re >> 1 and that Pr is of order unity. Since Re inversely
depends on o, while Pr is independent of o, this corresponds to the low-frequency
approximation. The expansions are introduced [30]
 
a2 1
aˆ1‡ 2‡O , …18a†
Re Re4
 
b1 1
bˆ ‡O , …18b†
Re Re3

and a2 and b1 are evaluated from Eqs. (12a) and (12b), with the result

…Pr ‡ g ‡ 2† g ÿ Prb21
a2 ˆ b1 ‡ , …19a†
Pr 2Pr
1 gÿ1 1
b1 ˆ ‡ , …19b†
2 2 Pr

These formulas are in accord with those in Tsien and Schamberg [30] providing their m*b =0
assumption is introduced. None of the foregoing approximations or assumptions, however, are
invoked in Eqs. (13) and (15).
It is common practice to use the large Reynolds number approximations to obtain a value
for m*b . To derive the form, e.g. shown in Prangsma et al. [25], set

c*p
gˆ , …20a†
c*n
 
c*p m* 4 m*b
Pr ˆ ‡ : …20b†
k* 3 m*
G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323 1321

Since b is per unit wavelength, replace b1 with Rel*i b*; hence, Eq. (19b) becomes
   ÿ1
3 4 m*b 2
b1 ˆ Rel*i b* ˆ ‰r*a* b*Š m* ‡ …2pf*† , …21†
3 m*
where Re is given by Eq. (17b). This yields
   
b* 2p2 4 1 1
ˆ m* ‡ ÿ k* ‡ m*b , …22†
f*2 r*a*3 3 c*n c*p b

where quantities, such as a*, r*, . . ., are evaluated for the undisturbed gas. The low-frequency
formula used in the experimental determination of m*b is thereby obtained.

2.5. Numerical results

Calculations are performed for CO2 with T0=300 K, p0=10 atm, a^ =2100, and an angular
frequency ranging from 20 to 60 kHz. Svehla [31] provides the other transport properties. The
chosen p0 and o values satisfy condition (1). Equations (13) and (15) yield a and b values that
agree with those from Eqs. (18a)±(18b) and (19a)±(19b) to, at least, six signi®cant digits.
Moreover, both a values are unity, also to six digits, while the b values are of the order of
10ÿ4. As noted, a dispersion measurement cannot be used to evaluate mb when a is so close to
unity, while the attenuation per wavelength is small but measurable. Equation (22) is,
therefore, accurate for an LTE measurement when a^ >>1, even though the original derivation
utilizes assumptions that are no longer warranted. With a^ >>1, Re is greatly reduced, while Pr is
now quite large. What is important is that Eq. (22) holds when the Peclet number is large.

3. Concluding remarks

The importance of LTE for the transport of energy between the internal and translational
modes of a polyatomic molecule is stressed. This condition is required if the bulk viscosity is to
be a function of the thermodynamic state in the compressible Navier±Stokes equations. This is
generally the case for steady and unsteady (turbulent) ¯ows of air. Although a viscosity
coecient, the bulk viscosity is associated with energy transfer rather than momentum transfer.
A number of molecules, such as CO2, have very large values for the ratio of the bulk
viscosity to the shear viscosity, a^ . Nevertheless, the low-frequency formula, currently in use for
the bulk viscosity, still holds for these molecules. This formula depends on thermodynamic
properties, such as the speci®c heats and the speed of sound, the thermal conductivity and
shear viscosity, and the frequency and attenuation of the ultrasonic signal. Under LTE
conditions, the dispersion of the signal is too small to be useful.
There appears to be few, if any, ultrasonic measurements at temperatures well above room
temperature. For instance, a^ should start to increase for air above 600 K when vibrational
excitation of O2 occurs. At a somewhat higher temperature, the rate of increase should be
more rapid when vibrational excitation of N2 occurs. Measurements of the bulk viscosity for
1322 G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323

molecules of interest in dense gas ¯ows [18, 19] are needed. These measurements should also be
at elevated temperatures.
Unlike other thermodynamic and transport properties, there currently is only one
experimental method for measuring the bulk viscosity of a gas. A second method would
certainly be desirable. Such a method has been proposed and theoretically assessed [32]. The
approach should be applicable to polyatomics, such as CO2, SF6, and I2, as a dense gas. In
this approach, a shock tube experiment yields a shock that is many thousands of mean free
paths thick, and both the linear Newtonian assumption and the LTE condition hold. A
measurement of the shock thickness along with a numerical solution for the structure of the
shock wave then yields mb.

Acknowledgements

The author is indebted to Professor B.M. Argrow for his comments and to M. Ishmail for
the numerical computations.

References

[1] Herzfeld KF and Litovitz TA, Absorption and dispersion of ultrasonic waves. New York: Academic Press,
1959.
[2] Truesdell C. J Rat Mech Anal 1953;2:643.
[3] Hunt FV. J Acous Soc Am 1955;27:1019.
[4] Lick W. Adv Appl Mech 1967;10:1.
[5] Richards WT. Rev Mod Phys 1939;11:36.
[6] Markham JJ, Beyer RT, Lindsay RB. Rev Mod Phys 1951;23:353.
[7] Karim SM, Rosenhead L. Rev Mod Phys 1952;24:108.
[8] Lighthill MJ, In: Batchelor GK, Davies RM, editors. Survey in mech. Cambridge: University Press, 1956:250.
[9] Bhatia AB, Ultrasonic absorption. Oxford: Clarendon Press, 1967.
[10] Overbeck CJ, Kendall HC. J Acous Soc Am 1941;13:26.
[11] O'Donnell M, Janes ET, Miller JG. J Acous Soc Am 1981;69:696.
[12] Van Dyke M. In: Riddell FR, editor. Hypersonic ¯ow research. New York: Academic Press, 1962:37.
[13] Emanuel G. Phys Fluids 1992;A4:491.
[14] Orou JC, Johnson JA, III. Phys Fluids 1994;6:415.
[15] Candler GV, Reprint from AIAA 1990;90:1695.
[16] Park C, Howe JT, Ja€e RL, Candler GV, Reprint from AIAA 1991;91:0464.
[17] Tisza L. Phys Rev 1942;61:531.
[18] Thompson PA, Carofano GC, Kim Y-G. J Fluid Mech 1986;166:57.
[19] Cramer MS. In: Kluwick A, editor. Nonlinear waves in real ¯uids. New York: Springer, 1991:91.
[20] Vincenti WG, Kruger Jr. CH. Introduction to physical gas dynamics. New York: Wiley, 1965.
[21] Kneser HO. In: Mason WP, editor. Physical acoustics, vol. II, Part A. New York: Academic Press, 1965:133.
[22] Bauer H-J. In: Mason WP, editor. Physical acoustics, vol. II, Part A. New York: Academic Press, 1965:47.
[23] Hanley HJM, Cohen EGD. Physica 1976;83A:215.
[24] Callear AB, Lambert JD. In: Bamford CH, Tipper CFH, editors. Chemical kinetics, vol. 3. The formation and
decay of excited species. New York: Elsevier, 1969:182.
[25] Prangsma GJ, Alberga AH, Beenakker JJM. Physica 1973;64:278.
[26] Monchik L, Yun KS, Mason EA. J Chem Phys 1963;39:654.
G. Emanuel / International Journal of Engineering Science 36 (1998) 1313±1323 1323

[27] Meador WE, Miner GA, Townsend LW. Phys Fluids 1996;8:258.
[28] Emanuel G. Phys Fluids 1996;8:1994.
[29] Emanuel G, Analytical ¯uid dynamics. Boca Raton, Fl: CRC Press, 1994.
[30] Tsien H-S, Schamberg R. J Acous Soc Am 1946;18:334.
[31] Svehla RA, NASA Tech. Rep. TR-132, 1962.
[32] Emanuel G, Argrow BM. Phys Fluids 1994;6:3203.

You might also like