You are on page 1of 73

Physica A 349 (2005) 60132

NavierStokes revisited
Howard Brenner

Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge,


MA 02139-4307, USA
Received 27 August 2004
Communicated by J.V. Sergers
Available online 24 November 2004
Abstract
A revision of Newtons law of viscosity appearing in the role of the deviatoric stress tensor
in the NavierStokes equation is proposed for the case of compressible uids, both gaseous
and liquid. Explicitly, it is hypothesized that the velocity v appearing in the velocity gradient
term Vv in Newtons rheological law be changed from the uids mass-based velocity v
m
; the
latter being the velocity appearing in the continuity equation, to the uids volume velocity v
v
;
the latter being a stand-in for the uids volume current (volume ux density n
v
). A similar
v
m
v
v
alteration is proposed for the velocity v appearing in the no-slip tangential velocity
boundary condition at solid surfaces. These proposed revisions are based upon both
experiment and theory, including re-interpretation of the following three items: (i)
experimental near-continuum thermophoretic and other low Reynolds number phoretic
data for the movement of suspended particles in uids under the inuence of mass density
gradients Vr; caused either by temperature gradients in single-component uids undergoing
heat transfer or by species concentration gradients in inhomogeneous two-component
mixtures undergoing mass transfer; (ii) the hierarchical re-ordering of the Burnett terms
appearing in the ChapmanEnskog gas-kinetic theory perturbation expansion of the viscous
stress tensor from one of being based upon small Knudsen numbers to one of being based
upon small Mach numbers; (iii) Maxwells (1879) ubiquitous v
m
-based thermal creep or
thermal stress slip boundary condition used in nonisothermal gas-kinetic theory models,
recast in the form of a v
v
-based no-slip condition. The v
v
vs. v
m
dichotomy in the case of
compressible uids is shown to lead to a fundamental distinction between the uids tracer
velocity as recorded by monitoring the spatio-temporal trajectory of a small non-Brownian
ARTICLE IN PRESS
www.elsevier.com/locate/physa
0378-4371/$ - see front matter r 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.physa.2004.10.034

Tel.: +1 617 253 6687; fax: +1 617 258 8224.


E-mail address: hbrenner@mit.edu (H. Brenner).
particle deliberately introduced into the uid, and the uids optical or colorimetric
velocity as monitored, for example, by the introduction of a dye into the uid or by some
photochromic- or uorescence-based scheme in circumstances where the individual uid
molecules are themselves responsive to being probed by light. Explicitly, it is argued that the
uids tracer velocity, representing a strictly continuum nonmolecular notion, is v
v
; whereas its
colorimetric velocity, which measures the mean velocity of the molecules of which the uid is
composed, is v
m
:
r 2004 Elsevier B.V. All rights reserved.
PACS: 51.10.+y; 66.10.x; 66.20.+d; 66.60.+a
Keywords: NavierStokes; No-slip; Rheology; Thermophoresis; Korteweg stress
1. Introduction
1.1. Background
This is the rst in a projected two-part series of papers concerned with proposed
modications to the NavierStokesFourier equations of continuum uid mechanics
(hereafter referred to by the acronym NSF equations) for compressible uids, both
gases and liquids. The term compressibility as used here refers not to the usual
effects of pressure on uid density, but rather to the effects on density of temperature
and/or composition (the latter in multicomponent mixtures) in systems where
pressure effects on density are assumed to be small relative to these other effects, e.g.
liquids or essentially isobaric gases. The present paper, the rst in the series, focuses
on the seemingly less controversial aspects of the changes we propose, namely those
connected with: (i) purely viscous effects in uids associated with the form of the
constitutive equation for the deviatoric stress appearing in Newtons rheological
viscosity law; and (ii) the no-slip tangential velocity boundary condition imposed at
solid surfaces. Experimental and other data will be presented in support of the
hypothesized changes in these two items. All of these data pertain to phenomena
involving creeping ow situations, where inertial effects in the momentum equation
are negligible, as too are viscous dissipation effects appearing in the energy equation.
Moreover, the data presented here are, for all practical purposes, limited to ideal
gases, although the underlying theory to be developed appears to bear no such
limitation.
The second paper in the projected series focuses upon further modications in the
NSF equation set posed by the work of the late statistical mechanician Yuri L.
Klimontovich in the inertial and viscous dissipative terms (as briey discussed in
Section 7). Currently, no experimental data exists to support these further changes.
Moreover, these additional changes raise fundamental questions whose profundity
and controversial nature greatly exceed the comparable level of contentiousness
likely to be aroused by the changes to the NSF equations advocated herein. As
such, it seemed prudent to clearly separate the issues associated with Klimontovichs
work from those discussed in the present paper, by simply postponing a discussion of
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 61
their common features. Those readers whose appetites are likely to be whetted by the
deeper, even more fundamental, issues to be addressed in the second paper are
encouraged to study in advance the key works of Klimontovich, briey cited in
Section 7.
1.2. Newtons law of viscosity. Mass velocity v
m
vs. volume velocity v
v
The NavierStokes equations, valid for Newtonian uids (both gases and liquids),
has been a xture of continuum uid mechanics ever since 1845 following the
seemingly denitive work of Stokes [1] and others [2], who proposed the following
rheological constitutive expression for the uids deviatoric or viscous stress T:
T = 2mVv kIV
.
v ; (1.1)
in which the velocity v appearing therein is the mass velocity v
m
; namely the velocity
appearing in the continuity equation
qr=qt V
.
(rv
m
) = 0 : (1.2)
The overbar appearing in (1.1) denotes the symmetric and traceless portion of the
dyadic which it surmounts, so that with D any dyadic, D = (
1
2
)(D D
T
) (
1
3
)I(I: D);
in which I is the dyadic idemfactor.
Despite the essentially universal acceptance of the fact that [3]
v =
?
v
m
; (1.3)
we argue below that the velocity appearing in Newtons viscosity law should, in fact,
be the volume velocity, v
v
[4,5]:
v = v
v
: (1.4)
Our arguments are based upon a trio of interrelated arguments: (i) ascribing a
continuum interpretation to experimental thermophoretic particle velocity data in
the small Knudsen number gaseous regime, thereby challenging the contemporary
view of such thermophoretic motion as a noncontinuum phenomenon (Section 3);
(ii) re-scaling the hierarchical order of the Burnett thermal stress terms appearing in
the ChapmanEnskog Knudsen number expansion of gas kinetic theory [6,7],
thereby changing their status from noncontinuum to continuum-level stresses on a
par with the viscous terms appearing in the classical NavierStokes equations
(Section 4); (iii) re-interpreting Maxwells thermal creep slip condition imposed upon
v
m
into a no-slip condition imposed upon v
v
(Section 5). Individually and collectively
these lend credibility to Eq. (1.4).
The volume and mass velocities are not independent of one another, but are
related through the expression [4]
v
v
= v
m
j
v
(1.5)
in which j
v
represents the diffusive ux of volume. [Eq. (1.5) represents the
decomposition of the volume ux n
v
into respective convective and diffusive
contributions, n
m
^ v and j
v
; where ^ v = 1=r is the specic volume.] In the case of
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 62
single-component uids undergoing heat transfer the constitutive equation for j
v
is
given by the expression [4]
j
v
= aV ln r ; (1.6)
wherein a = k=r^ c
p
is the uids thermometric diffusivity, with k the thermal
conductivity and ^ c
p
the isobaric specic heat. Eq. (1.6) is also applicable to the case
of two-component mixtures undergoing isothermal mass transfer [4], but with a
replaced therein by the Ficks law binary diffusion coefcient D: The respective heat-
or mass-transfer cases to which (1.6) applies correspond to circumstances in which,
for a xed pressure, either r = r(T) or r = r(w); with T the temperature and where
w denotes either one of the two species weight fractions in the isothermal binary
mixture.
Obviously j
v
= 0 when the uid is incompressible, corresponding to the case where
r is uniform throughout the uid (for all time), and hence from Eq. (1.2) for which
V
.
v
m
= 0: In such circumstances the velocities v
m
and v
v
coalesce, whence the
modied Newtons viscosity law (1.1) (1.4) reverts to its conventional form, (1.1)
(1.3).
1.3. What velocity appears in the no-slip boundary condition?
Accompanying the v
m
vs. v
v
velocity issue with respect to the constitutive choice
for v appearing in the deviatoric Newtons law stress expression for T; Eq. (1.1), is a
comparable issue that arises in connection with the dynamical no-slip tangential
velocity boundary condition
I
s
.
(v U) = 0 on qV
s
(1.7)
imposed at a solid surface qV
s
; with U the velocity of the solid at a point lying
on its surface. In the above, with n a unit normal vector on the surface, I
s
= I nn is
the unit surface dyadic or surface projection operator. While v
m
is universally
regarded as being the appropriate velocity v to insert into (1.7) under ordinary
circumstances, we nevertheless present experimental data as well as theoretical
kinetic theory results dating back to Maxwell [8] in 1879 (see Section 5) that implicity
supports the fact that the velocity appearing in (1.7) should be v
v
rather than
v
m
: On the other hand, the purely kinematical no-penetration, normal-velocity
boundary condition,
n
.
(v
m
U) = 0 on qV
s
; (1.8)
retains its usual mass-based form.
1.4. The uids tracer velocity
Let the spatio-temporal curve x = x(x
0
; t) denote the trajectory through space of a
small (albeit non-Brownian) passive tracer particle entrained in the uid that passes
though the space-xed point x
0
at time t = 0 and is later observed to be present at
some other point x at time t: The current, purely kinematical, view of continuum
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 63
uid motion is that the uids tracer or Lagrangian velocity v
l
; dened by the relation
v
l
:=
qx
qt
_ _
x
0
(1.9)
and derived from the tracers trajectory, is identical to the uids mass velocity v
m
[911]. In opposition to the accepted view that v
l
?
=
v
m
; we will argue later (see Section
6), based upon the same experimental data and gas-kinetic theory result invoked in
support of the relation v = v
v
; that, in fact,
v
l
= v
v
: (1.10)
For the time being we do not proffer any general theoretical reasons as to why
seemingly incontrovertible theoretical notions underlying the classical hypothesis
that v = v
m
fail in the case of compressible uids. Rather, we simply regard our main
result, v = v
v
; as being purely empirical. Our results do, however, point up the
singular nature of so-called incompressible uids, a subject that has long attracted
the attention of theoreticians, and led, inter alia, to the development of the eld of
extended irreversible thermodynamics [12]. Indeed, our results will be seen to
impact on the general subject of irreversible thermodynamics [1316] as a whole,
with regard to both the presence of nonnegative quadratic terms in the local rate of
entropy production and the Onsager symmetry relations [17] for linear constitutive
force/ux relations.
2. Pre-constitutive transport equations
2.1. Generic physical laws
For simplicity in what follows we limit ourselves to single-component Newtonian
uids undergoing heat transfer. The basic equations governing transport in such
uids [18] consist of: (i) the continuity Eq. (1.2); (ii) the Cauchy linear momentum
equation (in the absence of external body forces),
r
D
m
v
m
Dt
= V
.
P; P = Ip T; (2.1)
and (iii) the energy equation,
r
D
m
^ e
Dt
= V
.
j
u
V
.
(P
.
v
m
); ^ e = ^ u v
2
m
=2 : (2.2)
In the above,
D
m
Dt
:=
q
qt
v
m
.
V (2.3)
denotes the material derivative, p is the thermodynamic pressure, P the pressure
tensor, and ^ e and ^ u are, respectively, the specic (i.e., per unit mass) total and
internal energies. Moreover, j
u
is the diffuse internal energy current (i.e., ux
density), whose constitutive form remains to be specied. No distinction need be
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 64
made between the diffusive total and internal energy uxes, j
e
and j
u
; since neither
kinetic nor potential energy (the latter present when time-dependent conservative
forces actsee Appendix A) can be transported diffusively, only convectively. While
we keep open the choice of velocity v appearing in the constitutive expression (1.1)
for T contributing to (2:1
2
); accounting for our use of the word pre-constitutive to
describe the above pair of Eqs. (2.1) and (2.2), no comparable freedom of choice is
possible for the other velocities appearing in these equations, all of which have been
chosen to be v
m
: These other velocities, three in number, refer to their respective roles
as: (i) the specic momentum density, say ^ m; the latter appearing in the guise of v
m
in
Eq. (2:1
1
) [19,20]; (ii) the kinetic energy, say v
2
k
=2; the latter appearing in the guise
of v
2
m
=2 in Eq. (2:2
2
); and (iii) the rate of working, say V
.
(P
.
v
w
) the latter
appearing in the guise of V
.
(P
.
v
m
) in Eq. (2:2
1
): As discussed in Appendix A, this
lack of freedom in possibly choosing one or more of these three velocities, namely
( ^ m; v
k
; v
w
); to be v
v
; say, rather than v
m
; is imposed by the requirement that the
continuity, momentum, and energy equations all remain invariant under choice of
reference frame.
While it is commonly assumed [18] that the diffusive internal energy current j
u
is
given in single-component uids by Fouriers law, namely
j
u
= q ; (2.4)
where
q = kVT ; (2.5)
we have argued elsewhere [4,21] that in the case of compressible uids the correct
expression should be given, rather, by the constitutive relation
j
u
= q pj
v
: (2.6)
To the extent that Eq. (2.6) is correct, the diffuse volume current j
v
transports not
only momentum diffusively, as implied by Eqs. (1.1) (1.4)(1.5), but, concurrently,
it also carries internal energy diffusively, as embodied in the j
v
term appearing in Eq.
(2.6), above any beyond the Fourier conduction term (2.5). In order to keep our
pre-constitutive options open for as long as possible, thereby eschewing a
particular constitutive choice for j
u
; i.e., Eq. (2.4) vs. (2.6), it proves convenient to
write the ux density j
u
appearing in the energy equation (2:2
1
) in the pre-
constitutively neutral form,
j
u
= q p(v v
m
) : (2.7)
In view of Eq. (1.5), this expression thus constitutively embodies both the
conventional and modied formulas for j
u
; namely (2.4) or (2.6), according as either
v = v
m
or v = v
v
:
It also proves convenient to reformulate the energy equation (2.2) in terms of
enthalpy, rather than internal energy. By denition, ^ u =
^
h p^ v in which
^
h is the
specic enthalpy and ^ v = 1=r is the specic volume. Application of the material
derivative operator (2.3) to this thermodynamic identity, together with use of both
the continuity equation (1.2) and the standard single-component thermodynamic
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 65
relation [22]
d
^
h = ^ c
p
dT ^ v T
q^ v
qT
_ _
p
_ _
dp ; (2.8)
eventually transforms Eq. (2.2) into the form
r^ c
p
qT
qt
v
m
.
VT
_ _
= V
.
(kVT) V
.
[p(v v
m
)]
q ln r
q ln T
_ _
p
D
m
p
Dt
2mVv: Vv
m
k(V
.
v)(V
.
v
m
) : (2:9)
In the classic case, where v = v
m
; this reduces to the standard equation [18, p. 589]
governing the spatio-temporal evolution of the temperature eld in single-
component uids.
Additionally, when written out explicitly, the constitutively neutral linear
momentum equation (2.1), applicable to both the conventional and unconventional
choices of v; becomes
r
qv
m
qt
v
m
.
Vv
m
_ _
= Vp V
.
(2mVv) V(kV
.
v) : (2.10)
In the subsequent context of analyzing the experimental phoretic data used to
support our hypothesis that v = v
v
; it will be seen [2325] that the dissipative
terms, other than V
.
(kVT); appearing in the energy equation (2.9) and the inertial
terms appearing in the momentum equation (2.10) prove to be negligible in the
present class of phoretic experiments, whence it will turn out that these equations
are, respectively, replaced by the much more tractable expressions,
r^ c
p
qT
qt
v
m
.
VT
_ _
- kV
2
T (2.11)
and
Vp mV
2
v kVV
.
v - 0 ; (2.12)
the latter constituting a creeping ow approximation (cf. [82]). Constant physical
properties have been assumed in the above pair of equations, except of course for the
density r:
2.1.1. Boundary and initial conditions
Irrespective of which of the two constitutive relations, v = v
m
or v = v
v
; is
ultimately identied as being the physically correct choice, the requisite velocity
boundary conditions (1.7) and (1.8) to be imposed upon the independent variables
(v
m
; p; r; T) appearing in Eqs. (2.9) and (2.10), or in the simplied phoretic forms
(2.11) and (2.12), are applicable for both choices of v:
2.1.2. Further simplications
For the class of phoretic and transpiration problems that we subsequently address
it will be assumed as a satisfactory approximation that the law of adiabatically
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 66
additive volumes [4] adequately describes the equation of state for the uid,
corresponding to the fact that (q^ v=qT)
p
is a constant at a xed pressure. (The latter
represents an obviously correct assumption in the case of single-component ideal
gases; moreover, in the case of liquids it is surely a valid approximation over
sufciently small temperature ranges). This assumption, in conjunction with the fact
that the viscous dissipative terms in the energy transport equation (2.9) are
negligible, leads to the conclusion [4] that the volume velocity is solenoidal,
V
.
v
v
= 0 ; (2.13)
despite the fact that V
.
v
m
a0 in the case of compressible uids. Eq. (2.13) is a purely
kinematical relation, and hence holds irrespective of whether the dynamical
constitutive choice made for v is correctly given by v
m
or v
v
[26,27].
2.1.3. The case v = v
v
As a consequence of (2.13) the bulk viscosity term disappears from Eq. (2.12)
when the constitutive equation v = v
v
applies (although this term will not generally
disappear in the classical case where v = v
m
unless k = 0). In such circumstances, Eq.
(2.12) simply becomes
Vp mV
2
v
v
- 0 : (2.14)
Note in this case that Eqs. (2.13) and (2.14) are now identical in appearance to the
classical creeping ow equations (cf. Ref. [82]) for incompressible uids, wherein v
m
appears in these equations in place of v
v
: When considered in conjunction with the
no-slip boundary condition (1.7) imposed upon v
v
; we see that the solutions (v
v
; p) of
such compressible ow problems may be simply obtained from those already existing
in the literature for conventional incompressible creeping ow solutions [28].
3. Thermophoretic motion. Experimental conrmation of the no-slip condition (1.7)
for v = v
v
The ultimate test of any physical theory lies in the agreement of its quantitative
predictions with experiment. In this context we advance the hypothesis that the
velocity v appearing in the pre-constitutive energy and momentum Eqs. (2.9) and
(2.10) as well as in the no-slip boundary condition (1.7) is correctly given by v = v
v
;
rather than by its traditional form v = v
m
: This hypothesis is based upon what will be
shown, inter alia, to be its accord with existing thermophoretic [23] and other [24,25]
phoretic data (the latter data being discussed in Section 7). Subsequently, in Section
6, we tentatively advance a reason for the success of this constitutive choice, derived
from the empirical observation, demonstrated after the fact, that v
v
; rather than v
m
;
constitutes the tracer (or Lagrangian) velocity v
l
of the uid continuum. However,
we neither require, nor do we proffer here, a rational explanation of the fact that
v = v
v
; rather, we simply regard the latter constitutive expression as representing a
purely empirical relation, one that when used in conjunction with the pre-
constitutive momentum and energy transport equations (2.1) and (2.2) and
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 67
boundary conditions of Sections 1 and 2, together with the continuity equation, ts
the all of the experimental facts to which it is addressed (including, of course, the
wealth of data existing for the incompressible case, for which circumstance v
v
and v
m
are synonymous).
Thermophoresis is a phenomenon whereby a generally heat-conducting, force-
and torque-free solid particle, typically spherical, suspended in a uid (usually a gas)
within which an externally imposed temperature gradient exists, is observed to move
from the hotter to the colder regions of the uid; that is, the particle moves against
the temperature gradient [2932]. Recognition of this phenomenon was apparently
rst recorded in the literature by Tyndall [33,34] in 1870. The rst successful
quantitative explanation of thermophoresis, limited to gases, was offered by Epstein
[35] in 1929, who, building upon Maxwells [8] explanation of the phenomenon of
thermal transpiration some 50 years earlier, also involving uid motion in gases
animated by a temperature gradient [cf. 24], attributed thermophoretic particle
motion in the so-called near-continuum range of Knudsen numbers (Kn51) to
small noncontinuum effects, resulting in Maxwell slip (thermal creep.) at the
particle surface (see Section 5). Maxwells argument ascribes this v
m
slip to the action
of thermal stresses existing in the gas proximate to the surface (see Section 4).
Despite many embellishments since Epsteins [35] original analysis, the explanation
of the thermophoretic movement of aerosol particles in gases for Kn51 remains,
today, universally accepted as a strictly noncontinuum phenomenon, since no such
motion, either of the particle or uid, is predicted by NSF equations when
considered in conjunction with the traditional no-slip boundary condition. However,
since the thermophoretic velocity of a macroscopic non-Brownian particle in the
near-continuum region is observed to be independent of its size, this noncontinuum
view of the phenomenon appears inconsistent to us. Explicitly, for a given gas
pressure, and hence a given mean-free path l; the non-Brownian particles
thermophoretic velocity U is observed to be independent of the Knudsen number,
Kn = l=a; based upon the sphere radius a: (In the latter, l = 1=nps
2
according to
elementary kinetic theory [18], with n the number density of molecules and s the
molecular collision diameter.) And since the Kn value is the sole determinant of
whether the observed phenomenon is, or is not, due to noncontinuum effects, such
particle size-and, hence, Knudsen number-independence appears inconsistent with
the latter possibility.
Epsteins [35] noncontinuum thermal creep interpretation of thermophoretic
motion in the small Knudsen number regime was based upon his use of the
traditional relation v = v
m
in the constitutive equation (1.1) for the deviatoric stress
T; together with his adoption of Maxwell slip [8] [cf. Eq. (5.2)] of the mass velocity v
m
at the particle surface in lieu of the traditional no-slip condition (1.7) (1.3). We
offer here an alternative, strictly continuum interpretation of thermophoresis based
upon use of the constitutive equation v = v
v
; in both (1.1) and (1.7), the latter
connoting no slip of the volume velocity v
v
at the sphere surface. As in the
accompanying thermophoresis paper [23; see also Appendix B], we address the
problem of a laterally unbounded single-component uid, either gas or liquid,
conned between two parallel walls separated by a distance L; with the hot wall
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 68
situated at the coordinate x = 0 maintained for all time at a temperature T
0
; and the
cold wall situated at x = L maintained at T
L
: A solid sphere of radius a (a=L51)
and thermal conductivity k
s
; present in the uid, is situated sufciently far from
either wall such that particle-wall hydrodynamic interaction effects can be neglected.
Gravity effects are also assumed negligible. The thermal boundary conditions, as
usual, entail continuity of both the temperature and the normal heat ux component
at the sphere surface. Using the mass, momentum, and internal energy equations (the
latter within both the uid and solid sphere) of Section 2, we seek to calculate the
velocity U of the force- and couple-free sphere for circumstances in which the uid
motion may be regarded as quasistatic. In what follows in the next two subsections,
this calculation is effected for both constitutive choices, v
m
and v
v
; of the velocity v:
Ultimately the respective predictions of U for the two cases are compared with
experiment.
3.1. The case v = v
m
Upon use of the traditional v
m
formulation of the NSF equations and no-slip
boundary condition imposed upon v
m
in (1.7) it readily follows [23], inter alia, from
the trio of mass, momentum, and energy equations, in conjuction with the appropriate
thermal boundary conditions, that v
m
= 0 and p = const: (\x); corresponding to the
absence of uid motion and concomitant sphere motion, U = 0: After all, the problem
is one of steady-state, convection-free, heat conduction through the uid as well as
through the spheres interior, governed in both phases by the classical heat conduction
equation, V
2
T = 0: As such, it is unsurprising that no force acts on the sphere that
would otherwise serve to animate it. It was this failure of the traditional continuum
NSF equations and boundary conditions to predict the observed thermophoretic
motion of the sphere that led Epstein [35], later followed by others (see the review in
Ref. [23]), to seek a noncontinuum slip-based explanation for this particle motion.
Apart from the nonuniform temperature eld T(x) characterizing this pure heat-
conduction problem, the only other interesting physical feature worthy of noteone
that would, perhaps, not normally come to mindis the fact that the undisturbed
temperature gradient, [VT[ = (T
0
T
L
)=L; existing in the quiescent isobaric uid
necessarily creates a corresponding density gradient, Vr; owing to the uids isobaric
equation of state, r = r(T): Nevertheless, according to conventional NSF theory,
the existence of this density gradient is predicted to be without physical effect as
regards steady-state momentum and energy transport. As will be seen, it is in this
respect, namely the absence of physical consequences stemming from the local thermal
expansion of the uid, that the traditional constitutive choice, v = v
m
; in Eqs. (1.1) and
(1.7) ultimately proves to be unsatisfactory in explaining the origin of the uid-
mechanical forces serving to animate the sphere.
3.2. The case v = v
v
With this alternative choice of consitutive equation, the governing equations
and boundary conditions outlined in Sections 1 and 2 now yield [23] nontrivial
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 69
physical results, wherein Ua0; stemming from the presence of a temperature-
induced density gradient. In effect, the departure from the traditional NSF result
described in the preceding paragraph, wherein U = 0; is a consequence of the
appearance of the density gradient in the constitutive equation (1.6) for the
diffusive volume ux, which, through Eqs. (1.4) and (1.5), serves to couple v
v
to Vr;
thereby ultimately linking the particles thermophoretic motion to the uids
thermal expansion. The mathematical details underlying the calculation of U
are set forth in a companion paper [23; but see also Appendix B], which incor-
porates several simplifying assumptions in order to reduce the algebraic effort
required to solve these now strongly coupled transport equations. In addi-
tion to assuming constant equilibrium and transport properties for ^ c
p
; k; k
s
and m; Eq. (2.13) was further assumed to be applicable. In circumstances
where the particle is nonconducting, corresponding to the case where k
s
= 0; it is
shown in Appendix B that the calculation of U can be effected trivially by the use of
Faxens law, without the need to literally solve the requisite boundary-value
problem!
In any event, the more detailed and general calculation [23] based upon the
constitutive relation v = v
v
; addressing the case where k
s
a0; eventually yields the
following expression for the thermophoretic velocity of the sphere:
U =
ab
1 (k
s
=2k)
VT (3.1)
in which b = (q ln ^ v=qT)
p
(q ln r=qT)
p
is the uids coefcient of thermal
expansion. Furthermore, VT is the temperature gradient in the neighborhood
of the sphere that would exist in the spheres absence from the uid. It is given by the
expression VT = ^ x(T
0
T
L
)=L; in which ^ x is a unit vector in the x-direction,
perpendicular to the walls. We note that the product, ab k(q^ v=qT)
p
=^ c
p
; appearing
in Eq. (3.1) is a temperature-independent constant, since each of the three thermal
transport and equilibrium properties appearing on the right-hand side of the product
have, individually, been supposed constant. According to its derivation, Eq. (3.1) is
equally applicable to all uids, whether gas or liquid.
3.3. Comparison of Eq. (3.1) with experimental data
3.3.1. Gases
In the case of nonpolar, generally polyatomic, ideal gases, the gass thermometric
diffusivity appearing in Eq. (3.1) can be expressed via the Eucken equation [18] as
a = (9 5g
1
)u=4; with u = m=r the uids kinematic viscosity and g = ^ c
p
=^ c
v
the
uids specic heat ratio. In addition, b = 1=T for ideal gases. Accordingly, in such
circumstances Eq. (3.1) becomes
U = C
/
s
1
1 (k
s
=2k)
uV ln T ; (3.2)
in which C
/
s
= C(9 5g
1
)=4 is an O(1) dimensionless coefcient. The values of g for
ideal monatomic and diatomic gases are, respectively, g =
5
3
and g =
7
5
; whence it
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 70
follows that:
C
/
s
=
1:5 (monatomic gases) ;
1:36 (diatomic gases) :
_
(3.3)
The Eucken relation entering into (3.3) is an approximate relation. More precisely,
with Pr = ^ c
p
m=k the uids Prandtl number, it follows that a = u=Pr; representing an
exact relation between a and u: Expressed more accurately, one thus has for ideal
gases that C
/
s
= 1=Pr:
Experimentally, in the case of gases, the thermophoretic velocity U of a spherical
particle relative to the walls (on whose surfaces the prescribed temperatures T
0
and
T
L
are time-independent constants) is given over the entire range of Knudsen
numbers, Kn = l=a; from the near continuum regime, Kn51; to the free-
molecule regime, Knb1; by the semi-empirical data correlation,
U =
C
s
(k=k
s
C
t
Kn)
(1 3C
m
Kn)(k=k
s
1=2 C
t
Kn)
uV ln T : (3.4)
This generally Knudsen number-dependent formula, due originally to Brock [36],
builds upon the prior work of Epstein [35] (see the correlations in Refs. [3640]), and
serves to organize a vast amount of experimental thermophoretic data for gases. The
dimensionless coefcients C
m
; and C
t
; are manifestations of noncontinuum
hydrodynamic isothermal velocity slip and thermal jump effects, respectively,
whereas C
s
is Maxwells, nonisothermal slip-velocity thermal creep coefcient. The
latter is assumed in Maxwells [8] model of the phenomenon (see Section 5) to
constitute a noncontinuum effect arising from the so-called thermal stresses. [In our
view, on which we will subsequently elaborate, this would appear to be an
inconsistency, since C
s
is invariably taken to be a constant, independent of Knudsen
number. This independence serves to distinguish its effects from those of the other
two terms appearing in (3.4) involving the pair of noncontinuum coefcients, C
m
and
C
t
; each of which, being multiplied by the Knudsen number, thus disappears from
consideration in the Kn = 0 continuum limit.] A compilation is available [39] of
best t experimental values for these three semi-empirical coefcients, correspond-
ing to the values C
s
= 1:17; C
m
= 1:14; and C
t
= 2:18:
In the free-molecule limit, Kn o; Eq. (3.4) shows that U 0: This was to be
expected intuitively, since the spacing between molecules is so large on average
compared with the size of the sphere that the molecules do not see the latter and
hence pass it by unimpeded, failing thereby to animate the sphere.
In the continuum limit, Kn 0; of interest to us in connection with the test of our
continuum hypothesis v = v
v
; Eq. (3.4) adopts the form
U = C
s
1
(1 k
s
=2k)
uV ln T : (3.5)
Following its original introduction in 1879 [8], the nature and magnitude of
Maxwells slip coefcient C
s
; the only one of the three noncontinuum coefcients
now contributing to U; has attracted the attention of a number of theoreticians and
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 71
experimentalists. As discussed in the accompanying paper [23], theoretical values of
C
s
have ranged from Maxwells [8] original estimate of
3
4
for Maxwell molecules to
Derjaguin et al.s [38] predicted value of
3
2
; with experimental values falling in the
range 0.891.17, the latter gure constituting the most recent best-t value [39,40].
These latter, experimentally based, values are assumed applicable to both
monatomic and polyatomic molecules, despite the fact, historically, the theories
dating back to Maxwell [8], and underlying the generic equation (3.4) were based
entirely upon, and therefore presumably limited to, monatomic Maxwell molecules.
Our theoretical formula (3.2) (3.3) for gaseous continua obviously accords well
with the experimental data correlation (3.5) for the near-continuum regime. This,
despite the fact that our formula is based upon purely continuum-mechanical no-slip
arguments, which suppose that v = v
v
: In contrast, Epsteins [35] theory, which
underlies Eq. (3.5), is based upon noncontinuum thermal creep arguments, while
supposing that v = v
m
; albeit with a Maxwell tangential slip velocity boundary
condition imposed upon v
m
(in contrast with the traditional case, where no v
m
slip is
normally assumed to occur). Subsequently, in Section 5, the agreement between these
two very different theoretical approaches, Epsteins and ours, is reconciled, where it
is pointed out that the attribution by Epstein and others of noncontinuum Knudsen-
based behavior to Maxwells slip condition appears to be unwarranted, a
misconception not actually due Maxwell himself, but rather to those who followed
him [4144].
3.3.2. Liquids
Our companion paper [23] also compares our theoretical v
v
-based equation (3.1)
with the experimental thermophoretic data of McNab and Meisen [45] for liquids,
the only such single-particle liquid-phase data known to us. Without repeating
what is said in Ref. [23], sufce it to say here that Eq. (3.1) accords satisfactorily with
these limited data. Clearly, liquids are incapable of displaying noncontinuum
behavior with respect to rationalizing the thermophoretic movement of macroscopic
(non-Brownian) particles. As such, there exists no possibility that the thermo-
phoretic behavior observed by McNab and Meisen in liquids can be explained by
other than a continuum mechanism [46]. Since, as we have pointed out, classical
continuum arguments, based on the assumption that v = v
m
; fail to predict any
thermophoretic motion whatsoever, the agreement of our nontraditional v = v
v
continuum model with the liquid-phase data of McNab and Meisen [45] would
appear to enhance the credibility of our nontraditional volume velocity hypothesis.
An alternative theory of thermophoresis in liquids has recently been put forward
by Semenov and Schimpf [47]; see also Refs. [135,136]. In common with our
approach, their expression for the thermopheretic velocity of a solute molecule
contains the same thermal expansion proportionality factor b[1 (k
s
=2k)]
1
VT as
appears in our Eq. (3.1) [see Eq. (8.9)]. Moreover, in agreement with our predictions
for liquids, their thermophoretic velocity too is independent of particle size.
However, in contrast with our analysis, where the solute particle plays a passive
role relative to the solvent (at least in the nonconducting case), while simply being
entrained in a uid already in (volumetric) motion owing to the temperature gradient
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 72
(see Section 6), in the SemenovSchimpf model the physicochemical properties of the
solute particle enter explicitly into their expression for U: Their results are principally
employed to calculate Soret and thermal diffusivites in polymer solutions, a topic
that we too have addressed elsewhere [48] based upon our Eq. (3.1), and wherein we
have compared our respective predictions. Further commentary on their model vs.
ours is offered in Section 8.
4. A modied Newtons law of viscosity. Theoretical conrmation of the deviatoric
stress relation for v = v
v
based upon the re-scaled Burnett equations
To the extent that the appropriate constitutive velocity v appearing in Eq. (1.1) for
the viscous stress is given by v
v
rather than by its more traditional value v
m
; Newtons
viscosity law (1.1) departs from its usually assumed form, resulting in the following
equation for the deviatoric stress [49]:
T = T
m
2mVj
v
; (4.1)
in which Eq. (1.5) has been used. Here,
T
m
= 2mVv
m
(4.2)
denotes the traditional form of Newtons rheological law. The revised relation (4.1)
plays a major role in the detailed thermophoretic calculations of Ref. [23] (see also
Appendix B for the nonconducting case) summarized in the preceding section. But in
those calculations the choice of the velocity v appeared not only in Newtons
viscosity relation (1.1), but also in the no-slip boundary condition (1.7). As such, it is
not inappropriate to ask: Is there any independent evidence of the correctness of the
revised rheological law (4.1) that does not rely on the solution of a particular
boundary-value problem, and hence on the validity of the nonstandard no-slip
boundary condition (1.7), such as entered into the thermophoretically based indirect
evidence provided in Section 3 for the viability of the constitutive relation v = v
v
?
The answer is Yes! Direct evidence does indeed exist, based upon the well-known
results of Burnett [7] in the kinetic theory of gases, as set forth in standard texts [6]
on the subject.
Extracting physically relevant results from the kinetic theory of gases involves
solving the Boltzmann equation for ideal, single-component, monatomic gases by
asymptotically expanding the solutions thereof for the deviatoric stress eld T and
heat ux vector q in powers of the Knudsen number for Kn51: These expansions
take the form [6]
T = T
0
Kn T
1
Kn
2
T
2
O(Kn
3
) ; (4.3)
with a similar expansion for q: The leading-order, O(Kn
0
) O(1); so-called
continuum term in this expansion corresponds to the Euler equation of ideal
uid theory, namely T
0
= 0 and q
0
= 0; according to which only convective
momentum and energy transport mechanisms exist. Equivalently, diffuse or
molecular transport mechanisms are absent. The next, O(Kn); term in the expansion,
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 73
the so-called near-continuum contribution to the stress tensor, yields Newtons
law of viscosity, namely Kn T
1
= T
m
as well as Fouriers law of heat conduction,
Kn q
1
= kVT: Proceeding further in the hierarchy, the so-called noncontinuum
O(Kn
2
) terms in the expansion of the deviatoric stress [50] correspond to a sequence
of generally nonlinear Burnett terms [6,7], representing the corrections to the
classical v
m
-based Newtonian deviatoric stress tensor (4.2) [5154]. Upon incorpor-
ating these Burnett terms into the expression for the deviatoric stress, it proves
convenient to write
T = T
m
T

O(Kn
3
) ; (4.4)
with T

Kn
2
T
2
; which is of O(Kn
2
); designated here as being the extra
deviatoric stress, above and beyond the Newtons law T
m
-level.
Kogan et al. [55] and Bobylev [56] (see also the recent work of Yariv and Brenner
[57], briey summarized near the end of the present section) each discuss the
hierarchical ordering of the six O(Kn
2
) Burnett terms contributing to T

(see [6, p.
286]) with respect to their relative orders-of-magnitude in relation to the near-
continuum O(Kn) Newtonian term, Eq. (4.2). The KoganBobylev discussion of the
pertinent scaling issues leads to the following conclusion: For Kn51; and for
circumstances in which both a characteristic Reynolds number Re LU=n and a
characteristic nondimensional temperature gradient L[VT=T[; each based upon
some characteristic length L of the system, are both of O(1); two terms among the six
appearing in the complete Burnett sequence of stresses [6,7] are, in fact, actually of
O(Kn); rather than of O(Kn
2
); and hence should properly be classied hierarchically
as belonging to the traditional NSF equation set. In drawing the distinction
between their modied view [5557] of the scaling of the Burnett terms and the more
traditional view [6,7] thereof, the velocity U appearing in their denition of Re used
in the scaling is taken to be the uid-mechanical velocity v
m
rather than the velocity
of sound, c:
These two re-scaled Burnett deviatoric stress terms are given explicitly by the
expression
T

=
m
2
rT
(K
1
VVT K
2
T
1
VTVT) ; (4.5)
wherein K
1
and K
2
are O(1) dimensionless constants whose respective values depend
upon the particular choice of intermolecular potential used in evaluating the
Boltzmann collision integral. As the two terms in Eq. (4.5) contributing to the
deviatoric stress both depend exclusively upon temperature gradients in the gas, they
are referred to in the literature as thermal stresses, a concept dating back to
Maxwell [8]. As discussed below, this pair of thermal stresses should not be regarded
as O(Kn
2
) noncontinuum terms, but rather as near-continuum O(Kn) terms in the
terminology of gas-kinetic theory, since they are of the same order as those
appearing in the NSF equation set. Of course, in uid-mechanical parlance, both
the Euler and NSF equations are regarded as strictly continuum-level linear
momentum equations! As such, in sorting out the issues, one must carefully
distinguish between the respective gas-kinetic and uid-mechanical notions of what
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 74
constitutes a continuum. This issue is further discussed in the next section in relation
to Maxwells thermal stress-induced, so-called slip, boundary condition.
For monatomic Maxwell molecules [6,7,58] the values of the two constants are,
respectively, K
1
= 3 and K
2
= 3d ln m=d ln T: Accordingly, Eq. (4.5) adopts the
form [59]
T

=
3m
rT
V(mVT):
Now, m = ru; in which u is the kinematic viscosity. Furthermore, for single-
component ideal gases the relation between density and temperature at constant
pressure is such that r = C=T; where C is a constant. Thereby, it readily follows that
for isobaric situations,
T

= 3mV(uV ln r):
However, for an ideal monatomic gas, one has from the Eucken relation [18] that
u = (2=3)a: Accordingly, with use of the constitutive equation (1.6) for j
v
applicable
to the present single-component heat-transfer case, the preceding equation for the
extra stress tensor becomes, exactly,
T

= 2mVj
v
: (4.6)
Upon inserting the latter into (4.4) and comparing the resulting expression with
Eq. (4.1), it is seen that the relation thereby obtained for T via gas-kinetic theory
arguments is identical to that obtained from our purely continuum-mechanical
relation (4.1), the latter simply having been hypothesized on the basis of the
agreement of the v = v
v
-based predictions derived therefrom with thermophoretic
(and other) experimental data. By any reasonable criterion this exact agreement
between two such disparate derivationsone based upon purely theoretical
molecular-level arguments, and the other upon purely experimental continuum-
level argumentsboth yielding exactly the same rheological constitutive expression
for the stress tensor, is so striking as to provide seemingly unequivocal corroboration
of the correctness of the proposed continuum-mechanical constitutive equation (1.1)
(1.4) as well as of the no-slip boundary condition (1.7) entering into the
calculation, certainly for gases. Concomitantly, on the gas-kinetic side of the ledger,
this agreement would appear to provide compelling evidence in support of the
Kogan/Bobylev [55,56] argument that the hierarchical perturbation order of the two
Burnett thermal stress terms (4.5) is such that they should indeed be regarded as
representing near-continuum O(Kn); but not noncontinuum O(Kn
2
); terms. That is,
in uid-mechanical terminology the pair of Burnett thermal stresses (4.5) should be
labeled as being strictly continuum-level terms, on a par with those in the NSF
equation set.
As subsequently discussed, this altered view of the classical gas-kinetic theory
Burnett hierarchy has important consequences in a variety of contexts, including a
re-interpretation of Maxwells slip condition (Section 5), currently regarded as a
manifestation of noncontinuum O(Kn
2
) effects. Equally striking is the fact thatto
the extent that our modied constitutive equation v = v
v
; together with Eqs. (1.5)
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 75
and (1.6), is accepted as being correct on the basis of experimental conrmation
such experiments may be regarded as also formally conrming key aspects of
Burnetts gas-kinetic theory calculations. On the other hand, a number of tantalizing
theoretical issues remain unresolved with respect to the foundations of gas-kinetic
theory. Why, for example, does it appear on the basis of the K
1
and K
2
values
appearing in Eq. (4.5) that it is only for Maxwell molecules that the constitutive
equation (1.6) holds for monatomic gases? And, more generally, how does the notion
of a volume velocity enter into kinetic theory?
In any event, and irrespective of the latter issue, a formal rationalization of the
retention of only the two thermal stress terms in the Burnett expansion, while
disregarding the other four terms appearing therein, is provided in Ref. [57]. That
rationalization is based upon the fact that classical ChapmanEnskog, small
Knudsen number expansion of the Boltzmann equation is not uniformly valid. This
derives from the fact that with v a characteristic velocity, the Knudsen number is
related to the Mach and Reynolds numbers, M = v=c and Re = Lv=u; respectively,
by the expression Kn = M=Re: The conventional scaling arguments used in
obtaining the ChapmanEnskog perturbation expansion for the Kn51 near-
continuum O(Kn) case are commonly understood as applying in the dual limits,
M = O(1) and Reb1; that is, where v = O(c) [6]. However, one could, alternatively,
achieve the limiting Kn51 near-continuum criterion by instead considering the dual
pair of inequalities, M51 and Re = O(1); corresponding to the case where v =
O(v
m
)5c: And it is precisely for those circumstances in which this dual combination
of limits (rather than the single limit Kn51) is met that one would expect a linearized
rheological law like Newtons law of viscosity to be valid. When each member of the
complete set of six Burnett terms is now individually rescaled according to these
altered criteria, one nds that the two thermal stress terms, designated by T

in Eq.
(4.5), indeed become of the same order as the classical viscous stress terms (4.2) with
respect to their hierarchical O(M) ordering in the perturbation expansion. It is this
formal small Mach number re-scaling [57] that singles out for special attention only
the two thermal stress terms among the six Burnett terms, the remaining ones all
being of higher order in Mach number. Explicitly, in place of Eq. (4.3) one would
now write
T = T
(0)
MT
(1)
M
2
T
(2)
O(M
3
); Re = O(1) ; (4.7)
valid for M51; with MT
(1)
given by Eq. (4.1).
Written out explicitly in the light of (1.5), Eq. (4.1) becomes
T = 2mVv
v
: (4.8)
This relation has, thus far in this section, been independently conrmed to apply
only to the case of single-component monatomic ideal gases. However, to the extent
that Eq. (1.1) (1.4) applies quite generally, all Newtonian uids [liquids and
polyatomic gases, both single- and two-component, the latter with a replaced in (1.6)
by the binary diffusivity D], would be expected to obey Eq. (4.8), at least in
circumstances where bulk viscosity effects are absent (the latter due either to the fact
that k is zero or that V
.
v
v
= 0) [60].
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 76
In summary of the results of this section, while the existence of thermal stresses as
embodied in Eq. (4.5) or any of its subsequent variants has, in the past, been
regarded as a noncontinuum, Knudsen-type phenomenon, this attribution does not
appear to be correct, as shown by the above analysis. By way of comparison, in the
case of solids, the comparable existence of thermoelastic stresses [61] in linearly
elastic media obeying Hookes law, and stemming from thermal expansion, has long
been recognized as a strictly continuum-level phenomenon. That it has taken so long
to recognize the existence of comparable continuum-level thermal stress effects in
uids, stemming from the same thermal expansion effects as exist in solids, is surely
attributable to the focus upon mass rather than volume in connection with
deformation in uid continua. In solids it is the geometry of continuous deformation
rather than the displacement of mass, viewed as a continuum, that commands ones
attention when addressing the fundamental issue of the movement of the respective
continua through space [62]. Dilatation, a feature common of both solids and uids,
as embodied in the role of b in governing the deviatoric stress, constitutes one aspect
of such deformation, the major one in the case of thermal expansion effects. In this
sense, the movement of volume [4], whether regarded as static as in the case of solids
or kinetic as in the case of uids, constitutes the common theme connecting the two
distinct types of continua. These come together in the notion of Korteweg stresses
[63], a strictly continuum-level concept discussed in Section 8, where it is the density
gradient or equivalent specic volume gradient, rather than the temperature
gradient, which serves to unify the notion of thermal stresses in solid and uid
continua.
5. Maxwells v
m
-slip boundary condition viewed alternatively as a no-slip condition
imposed upon v
v
Maxwells [8] well-known thermal creep slip-velocity boundary condition imposed
upon v
m
at a solid surface plays a fundamental role in the application of gas-kinetic
theory to actual physical problems. Its importance stems from the fact that the
Boltzmann equation, by itself, offers no insight into the boundary condition to be
imposed upon the tangential velocity at such surfaces [64, pp. 5256,98102]. As a
result, the use of Maxwells slip condition is ubiquitous in applications. It is relevant,
for example, in analyses of gaseous phoretic phenomena, including thermophoresis,
thermal transpiration, and diffusiophoresis (the latter when this slip condition is
extended from temperature to species concentration in isothermal binary systems
[25,65]), as well to analyses of other baroeffect phenomena [66]. This section
addresses, inter alia, the background behind Maxwells thermal creep condition,
including the approximations inherent in its original derivation. Our goal in doing so
is, in part, to demonstrate that the current view of Maxwell slip as a residual
manifestation of Burnett-level O(Kn
2
) noncontinuum behavior is a misinterpretation
of its true status in the asymptotic Knudsen number perturbation-expansion
hierarchy. Rather, we argue that his thermal creep condition is actually an O(Kn)
near-continuum term in the obfuscating language of statistical mechanics. On the
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 77
other hand, when expressed in the terminology of uid mechanics it constitutes a
strictly continuum-level term, on a par with the NSF equations (either the original
or modied versions thereof). Having thus relegated thermal creep to the role of an
ordinary continuum uid-mechanical boundary condition, we argue that what
appears to be Maxwell slip of the mass velocity v
m
corresponds exactly to no-
slip of the volume velocity v
v
; that is, Maxwells boundary condition is really a no-
slip uid-mechanical continuum-level boundary condition in disguise. In turn, this
recognition offers further quantitative corroboration of the viability of our
nontraditional constitutive relation v = v
v
; independently of that offered in Sections
3 and 4.
The review of Maxwell slip which follows is carried out in the context of our
proposed no-slip v = v
v
continuum boundary condition,
I
s
.
(v
v
U) = 0 on qV
s
; (5.1)
derived from Eqs. (1.7) (1.4) and used in our solution [23] of the thermophoretic
problem in the accompanying paper, as well as in the elementary Faxens law-based
version thereof derived in Appendix B. Explicitly, in what follows, it is shown in the
case of single-component heat-transfer problems in gases that the Maxwells
boundary condition [8],
I
s
.
(v
m
U) = C
s
uV
s
ln T on qV
s
; (5.2)
with V
s
= I
s
.
V the surface-gradient operator and C
s
Maxwells slip coefcient, is, in
fact, constitutively identical to our Eq. (5.1). Whereas Eq. (5.2) is currently regarded
in the literature as a noncontinuum O(Kn
2
) Burnett-level-derived relation, our Eq.
(5.1) is a strictly continuum relation in the language of uid mechanics. Indeed, the
fact that these relations prove to be constitutively identical in their common domain
of validity, namely ideal monatomic gases, adds credibility to the viability of our
fundamental relations, Eqs. (1.4)(1.6), as well as to that of our hypothesis regarding
the proper no-slip boundary condition to be imposed at solid surfaces, Eq. (5.1).
5.1. Chronology of the Maxwell slip condition
In an attempt to explain Crookess (1876) observation [67] of the rotation of a
windmill in a partially evacuated radiometer, Maxwell [8], in 1879, introduced the
hypothesis of a temperature-gradient-induced slip imposed upon v
m
at solidgas
boundaries [68]. From Maxwells own description [8] of the evolution of his thinking
about the problem, the derivation of his slip boundary condition (5.2) evolved in two
distinct stages and timelines. Our subsequent chronological review thereof serves to
clarify the issue of whether or not Maxwells boundary condition is indeed to be
viewed as an O(Kn
2
) noncontinuum effect owing to its Burnett-like origin, or as an
O(Kn) near-continuum condition (as we believe Maxwell himself clearly under-
stood).
In the rst stage of his celebrated paper, directed exclusively at qualitatively
rationalizing the observed behavior of Crookess radiometer as a strictly slip-based
phenomenon, Maxwell introduced the notion of thermal stresses existing in the
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 78
interior of the gas, molecularly far from any surfaces. It is useful here to recall that
the notion of stresses in a uid (gas) originally involved only viscous stresses, arising
from massvelocity gradients, as embodied in Newtons law of viscosity, Eq. (1.1)
(1.3). Indeed, one of the early triumphs of gas-kinetic theory was Maxwells
theoretical derivation in 1860 [69] of Newtons law of viscosity (accompanied by the
later, experimentally conrmed and surprising, conclusion that the viscosity of a gas
is independent of pressurealbeit a statement limited at the time to Maxwell
molecules). In his calculation, Maxwell used his own pre-Boltzmann molecular
collision model, involving so-called Maxwell molecules [58], the latter repelling one
another pairwise with a force inversely proportional to the fth power of their
separation distance, while devoid of an attractive component.
These thermal stresses, above and beyond the purely viscous stresses T
m
given by
Eq. (4.2), and arising from the massvelocity, were demonstrated by Maxwell to
stem from small temperature gradients existing within the gas. By adopting an
elementary perturbation scheme based upon the relative smallness of these gradients,
Maxwell [8] succeeded in deriving the following formula for the deviatoric thermal
stress T
/

in the gas, namely the stress above and beyond the Newtonian viscous
stress tensor (4.2):
T
/

=
3m
2
rT
VVT
1
2
IV
2
T O(T
1
[VT[
2
)
_ _
: (5.3)
Comparison of the above with the Burnett thermal stress tensor (4.5) for the case of
Maxwell molecules shows that Maxwells [8] calculation represented only an
approximation to Burnetts later [6,7], more accurate, calculations, which also
included nonlinear temperature gradient effects, not accounted for by Maxwell
owing to the latters assumption of smallness of the temperature gradient. It was
with the derivation of Eq. (5.3), valid in the interior of the gas, that the rst phase of
Maxwells eventual slip velocity calculation ended. That is, no attempt was made by
Maxwell at the time to use the knowledge embodied in Eq. (5.3) to estimate the
appropriate form of the tangential velocity boundary condition at a solid surface,
although he recognized that motion of the gas along a solid surface on which a
temperature gradient existed would somehow be affected by these thermal stresses.
The second stage of Maxwells derivation of his slip condition, Eq. (5.2), was
accomplished some months later. In an appendix added in May 1879 [70] to the
original version of his paper [8], and ending with the derivation of Eq. (5.2), Maxwell
writes: In the paper as sent in to the Royal Society [read on April 11, 1878], I made no
attempt to express the conditions which must be satised by a gas in contact with a solid
body, for I thought it very unlikely that any equations I could write down would be a
satisfactory representation of the actual conditions, especially as it is almost certain
that the stratum of gas nearest to a solid body is in a very different condition from the
rest of the gas. He then goes on to further state that: One of the referees, however,
pointed out that it was desirable to make the attempt.... I have therefore added the
following calculations, which are carried out to the same degree of approximation as
those for the interior of the gas. In his appendix, Maxwell then goes on to point out
that it was only after hearing Reynolds read his paper on thermal transpiration
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 79
[71,72] before the Royal Society that he ...began to reconsider the surface conditions
of a gas.... Maxwell then goes on to describe his scheme as being ...purely
statistical..., in addition to commenting that he is treating ...a gas which is not
highly raried [sic], and in which the space-variations within distances comparable to l
[essentially the mean-free path l; see below] are not very great.
The slip-velocity formula that Maxwell ultimately developed is given by the
following expression, in which x lies normal to the solid surface qV
s
; and v is the
tangential velocity component in the y-direction, lying in the tangent plane of the
surface:
v G
qv
qx

3
2
m
rT
q
2
T
qxqy
_ _

3
4
m
rT
qT
qy
= 0 on qV
s
: (5.4)
In the above
G =
2
3
2
f
1
_ _
l ; (5.5)
in which G is a coefcient of slipping, with l the mean-free path of the gas molecules,
and 1 f the fraction of molecules reected by the surface, f then being the fraction
absorbed. Maxwells equation (5.4) is based directly upon his approximate thermal
stress equation (5.3) [73]. It is only after setting G = 0 in this approximate Maxwell
molecule model that the classical Maxwell thermal creep formula (5.2) obtains,
wherein C
s
= 3=4: Given the scaling argument discussed in Section 4 demonstrating
that the Burnett thermal stress terms constitute a continuum-level phenomenon (in
the language of uid mechanics), it is apparent that Maxwells slip-velocity formula,
originating from his version (5.3) of these thermal stresses, must also be a continuum
result, unrelated to O(Kn
2
) noncontinuum phenomena.
To bring to fruition the process of demonstrating the equality of the Maxwell slip
formula (5.2) with our no-slip volume-velocity continuum formula (5.1), substitute
Eqs. (1.5) and (1.6) into (5.1), and invoke the Eucken formula [18] preceding Eq.
(3.2) so as to obtain the following formula, formally equivalent in physical content to
Eq. (5.1):
I
s
.
(v
m
U) = C
/
s
nV
s
ln T on qV
s
; (5.6)
in which C
/
s
is the coefcient dened following Eq. (3.2). Eq. (3.3) shows that C
/
s
=
3
2
for monatomic gases. Clearly, our continuum equation (5.6) is constitutively
identical to Maxwells slip formula (5.2). Maxwells value of
3
4
for C
s
differs from our
predicted value of
3
2
in Eq. (3.3) for monatomic gases. However, Maxwells result
applies only to Maxwell molecules, whereas our result bears no such limitation. And
it is known [6] that the Maxwells intermolecular force model does not generally yield
predictions that accord well with experiments. Moreover, Maxwells C
s
result also
depends to some extent on the details of the collision model adopted for the
reection of the molecules from the wall (cf. Ref. [139, pp. 367400]). As such, the
disparity in the respective values of the two slip coefcients is not regarded as
serious, especially given Talbot et al.s [39] best t value of C
s
= 1:17 which
straddles the two, together with the fact Maxwells thermal transpiration calculation
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 80
[8] based upon his C
s
=
3
4
value underpredicts the experimentally-observed
thermomolecular pressures by a factor of about 2 [24].
In comparing our proposed boundary condition with that posed by Maxwell it
needs to be stressed that Maxwells result has been shown to hold only for
monatomic gases. In contrast, our more general version,
I
s
.
(v
m
U) = aV
s
ln r on qV
s
; (5.7)
of Maxwells v
m
-based slip formula, representing the precursor of Eq. (5.6) without
invoking the Eucken relation, holds also for polyatomic gases [cf. Eq. (3.3)], and
presumably for liquids too, although the latter remains to be studied experimentally.
Indeed, as pointed out in Section 8 in connection with our discussion of Korteweg
stresses, in terms of our more generic formulation,
I
s
.
(v
m
U) = I
s
.
j
v
on qV
s
; (5.8)
of the slip boundary condition (5.1) [with j
v
given by Eq. (1.6) for single-component
gases and by its a = D two-component counterpart for isothermal binary mixtures],
it is apparent that the preceding relation should be equally applicable to the case of
binary diffusion (a fact which was subsequently conrmed by our analysis [25] of
experimental isothermal gaseous diffusiophoretic data in such systems). In any
event, it now appears clear that to refer to Maxwells boundary condition as a slip
condition is to obfuscate the fact that the physical boundary condition to be applied
along a solid surface, nonisothermal or not, is simply the standard continuum one of
no slip, albeit in disguise, since it is imposed upon v
v
rather than v
m
!
5.2. How can use of the wrong velocity eld (in conjunction with the correct boundary
conditions) lead to the correct global physical result?
Various nonisothermal experiments such as thermal transpiration and thermo-
phoresis have been rationalized [8,35] during the past 125 years by invoking
Maxwells boundary condition (5.2), which, as we have indicated, is the physically
correct no-slip boundary condition to be applied at solid surfaces, albeit formulated
in terms of v
m
rather than v
v
: However, from the perspective of v
v
being the correct
velocity to use for v in the generic pre-constitutive equations of Sections 1 and 2,
researchers solving these phoretic problems in the past have, according to our
interpretation, used the wrong momentum and energy transport equations, namely
those based upon the choice v = v
m
[74]. For consistency, these transport equations
should have added Maxwells thermal stress approximation, T
/

; Eq. (5.3), or, more


accurately, the Burnett thermal stress version thereof, Eq. (4.5), to the classical
viscous stress contribution T
m
so as to obtain the correct deviatoric stress T; Eq.
(4.4), to be used in the creeping ow relation, Vp V
.
T = 0: Moreover, these
equations should not have been based upon the assumption of incompressibility,
V
.
v
m
= 0; as was originally supposed by Maxwell [8] in his analysis of thermal
transpiration and later adopted by Epstein [35] in the latters analysis of
thermophoretic movement. Nevertheless, our more recent recalculations of the
physical forces (or, equivalently, the thermophoretic particle velocity U [23] and the
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 81
thermal transpiration thermomolecular pressure difference Dp [24] using our
modied theory, have yielded the same results obtained by these prior researchers
(at least to within a factor of 2 based upon Maxwells C
s
=
3
4
slip coefcient), results
which accord with experiment. But if these prior researchers indeed solved the wrong
set of transport equations (albeit subject to the correct boundary conditions), how
can it be that their nal results were correct, in the sense of agreeing with experiments
as well as with our results [23,24]? The answer will be seen to lie in the fact that their
results were fortuitous, valid only in circumstances where the temperature gradients
involved are small (when appropriately scaled), so that neither the nonlinear terms in
the Burnett thermal stress formula (4.5) nor the incompressibility assumption affect
the outcome of the calculation. However, in more general circumstances, where
nonlinear effects are retained in calculations of this genre, the scheme based upon the
incorrect constitutive formulation (1.3) will presumably lead to erroneous conclu-
sions relative to those based upon the constitutive relation (1.4).
Appendix C shows explicitly how and why these inherently wrong transport
equations can nevertheless lead fortuitously to the correct result, at least to leading
order in the externally imposed temperature gradient. A main point to note in this
class of thermal problems is that the convective uxes appearing in the momentum
and energy equations always prove to be negligible relative to the diffusive uxes
when the temperature gradients are sufciently small. The calculations presented in
Appendix C, applicable to both the thermophoretic [23] and thermal transpiration
[24] cases, are based upon the fact that the respective velocities v
m
and v
v
sought in
the two schemes are related by Eq. (1.5), in which the diffuse volume ux is given by
(1.6), namely j
v
= aV ln r or, equivalently, by j
v
= (k=^ c
p
)V^ v: Inasmuch as we have
supposed, based upon the thermal law of adiabatically additive volumes, that
(q^ v=qT)
p
= const; it follows that:
j
v
=
k
^ c
p
q^ v
qT
_ _
p
VT ; (5.9)
in which, as discussed, the multiplier of VT appearing therein is taken to be a
temperature- and pressure-independent constant.
5.3. Further comments on the Maxwell slip condition
5.3.1. Knudsen boundary layers
It is invariably assumed in the literature, inappropriately in our view [75], that
Maxwell arrived at his slip condition (5.2) by introducing noncontinuum arguments
based, in effect, upon the notion that a hypothetical Knudsen boundary layer or sub-
layer exists in proximity to a solid surface qV
s
: This notion likely stems from
Maxwells [8] remark: ...that the stratum of gas nearest to a solid body is in a very
different state than the rest of the gas. However, introducing a noncontinuum
O(Kn
2
) argument, while setting G = 0 in Eq. (5.4) in an attempt to rationalize
Maxwells slip condition lacks consistency, since the slip coefcient C
s
is independent
of the Knudsen number [in contrast to the limiting behavior displayed by the other
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 82
two noncontinuum slip effects appearing in Eq. (3.4)]. In particular, the terms
multiplied by G; characterized by a length term a; say, involve a derivative with
respect to x; the distance normal to the surface. Since G is proportional to l; the
operator Gq=qx is of O(l=a) O(Kn): This contribution, of course, vanishes in the
limit as Kn 0; whereas the Maxwell slip term in (5.4) does not. As such, the slip
manifested in Eq. (5.2) does not vanish in the Kn = 0 continuum uid-mechanical
limit, in disagreement with all contemporary views of the no-slip boundary condition
for uid continua, isothermal or not. Accordingly, it would appear that Maxwell slip
cannot be attributed to a hypothetical noncontinuum, Knudsen boundary layer.
Moreover, it appears clear from Maxwells discussion of the slip issue in
connection with Eqs. (5.4) and (5.5), that the stratum of gas in the neighborhood
of the surface, to which he refers as being physically different from that in the bulk, is
associated with the true slip coefcient G rather than with the thermal stress term in
his Eq. (5.4). Thus, setting G = 0 in the latter equation in order to obtain (5.2)
eliminates the possibility of Knudsen-based slip in the physical meaning of the word
slip. [In the contemporary literature, G is associated with what is now known as
velocity slip, a true noncontinuum temperature-gradient-independent phenomen-
on, corresponding to the appearance of the coefcient C
m
in Eq. (3.4)].
5.3.2. Maxwell slip violates the First law of thermodynamics
Last but not least it is pertinent to note that Maxwells slip condition (5.2), when
used in conjunction with the conventional incompressible NSF equation set, as
Maxwell has done in his treatment [8] of the thermal transpiration problem, violates
the First law of thermodynamics. This violation stems from the fact that were slip to
occur at a rigid immobile mass-impermeable surface, work would be performed by
the surroundings across this surface upon the uid conned within. But this is
impossible because under such circumstances no mechanism exists by which the
surroundings can do work on the uid! As such, this fundamental thermodynamic
fact furnishes yet another reason for rejecting the notion of the Maxwell boundary
condition (5.2) as representing physical state of slip; that is, while the boundary
condition itself is constitutively correct, it does not connote slip of the actual physical
velocity, v
v
: This point is further elaborated upon in Appendix D, where the First law
violation is demonstrated in a more general context transcending our phoretic
applications.
6. The uids tracer (Lagrangian) velocity v
l
: Rationalization of the constitutive
relation v = v
v
This section addresses what might appear supercially to be a kinematical rather
than dynamical issue pertaining to the fundamental notion of the velocity of the uid
at a point in space occupied by the continuum. Whereas it is trivial to understand the
notion of the velocity of a material object, particularly the velocity of a rigid body,
and especially if the body is envisioned simply as an effectively pointsize non-
Brownian object, the same cannot be said of the ease of understanding the concept of
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 83
uid velocity at a point in space. After all, a continuum does not physically exist as a
continuous material entity. Rather, it is an abstraction since the matter of which the
continuum is composed consists of discrete material entities, molecules with empty
spaces separating them. And these spaces, devoid of matter, can be quite vast in the
case of low-pressure gases. Thus, if a continuum does not exist physically, what is the
entity whose velocity we believe that we are measuring? While ones rst response
might be to answer that one is measuring the mean velocity of the molecules in the
neighborhood of the point of interest, this cannot really be so obvious. After all,
given that Maxwells notion of mobile molecules was not introduced into physics
until 1860, what in 1755, 100 years earlier, did Euler [76] believe was the physical
interpretation of the velocity that appeared in his continuity equation (1.2) which he
had then derived? Similarly, in 1845, what did Stokes [1] understand to be the
experimental realization of the velocity appearing in his formulation of Newtons
viscosity law, Eq. (1.1)?
While it might appear at rst that the issues being addressed are purely
philosophical, and hence without physical import, we believe that the reader will
ultimately think otherwise. For example, we raise the question of whether, even in a
single-species uid, the optical-colorimetric measurement of the spatio-temporal
motion through space of a small photochromically colored or dyed region of a
moving uid [11] will generally record the same velocity as would result from
comparably monitoring the motion of a tracer, a small (albeit non-Brownian)
solid particle deliberately introduced into the uid [9]. And if these two
measurements disagree, which, if either, should be understood as the velocity of
the uid at the spatial point in question? Though both experimental methods are in
wide use today, it is not obvious a priori that these two distinct schemes will
necessarily yield the same uid velocity under all circumstances. In this context,
imagine, for example, a (non-Brownian) particle placed into an ideal uid
undergoing a steady motion that is uniform far from the particle. According to
DAlemberts paradox [77] such a particle experiences no force, whence it will
presumably remain in place as the uid ows around it. Clearly, such a particle
would make a poor tracer of the undisturbed uniform uid motion. (I am grateful to
Dr. Ehud Yariv for this example.) On the other hand, were one to photochromically
color the molecules, so to speak, of such an ideal uid in some small uid region,
motion of the uid as dened by the movement of the resulting color through space
would presumably be observed (almost certainly, registering mass motion, v
m
).
While one might argue that no such inviscid uid exists, thus rendering the question
moot, this begs the fundamental question.
On the other hand, a rigid particle to which uid adheres [and hence on whose
surface the no-slip condition (1.7) prevails] has the potential to serve as a tracer of
the uid motion, provided that other qualications are met, including passivity of
the particle, an attribute which will subsequently be dened more precisely. In any
event, a truly critical set of experiments aimed at testing the hypothesis of equality
between the uids mass velocity v
m
(presumably the velocity being measured during
a colorimetric molecular-tagging velocimetry experiment [11]) and the uids tracer
velocity [9], say v
l
; appears to have not yet been performed, at least in circumstances
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 84
where a sensible difference between them might be expected based upon our modied
NSF equations (1.4) and (2.6). Nevertheless, on the basis of what is to follow,
these velocities would appear to be equal only in the case of incompressible uid
motions, V
.
v
m
= 0; wherein the density is uniform.
The instantaneous tracer or Lagrangian velocity v
l
(x; t) of a small [78] particle
(whose center of volume, say, is) situated at time t at some position x in space would
appear to be dened by Eq. (1.9), with the curve x = x(x
0
; t) constituting the
trajectory of the particle through space and time. One of the key words in this
denition of a tracer is the word particle. In continuum mechanics [79] this is
invariably interpreted as being a differentially sized material uid particle, a
purely hypothetical entity. However, since the continuum is composed of discrete
molecules, the notion of such a material particle is devoid of strict operational
signicance owing to the fact that individual molecules are free to enter and leave
such a body through its surface, despite the constancy of the material particles total
mass during its journey through space. As such, the denition (1.9) possesses an
unambiguous operational signicance only in the case of a real physical particle,
namely a small (albeit non-Brownian) rigid object deliberately introduced into the
uid and subsequently monitored in space and time in order to establish the
particles trajectory x = x(x
0
; t) and hence therefrom the velocity of the
undisturbed uid at a point of space. More will be said of this later. For the time
being, the important point that we wish to make here is to propose the tentative
kinematical hypothesis that the uids tracer velocity, as measured by the preceding
particle trajectory scheme, is identical to its volume velocity. Explicitly, as suggested
by Eq. (1.10), our hypothesis is that the uids tracer velocity v
l
is identical to the
uids volume velocity v
v
:
The case for the equality v
l
= v
v
can only be made experimentally, using real
particles, not hypothetical material uid particles. That is, it is not possible to prove
this relation purely theoretically, either by continuum-mechanical or statistical-
mechanical arguments applied only to the uid. Indeed, the very denition of v
l
;
requiring the deliberate introduction of a (real) foreign object into a hypothetical
uid continuum, implies the impossibility of a purely kinematical pointwise
continuum-mechanical denition of the undisturbed uids tracer velocity. As such,
Eq. (1.9) is not to be regarded as the denition of v
l
; but rather as representing the
mechanism by which the trajectory, x = x(x
0
; t); of a hypothetical uid particle x
0
moving through space can be determined once the tracer velocity eld v
l
(x; t) is
already known independently. In this paper, we pose but a single experiment in
support of the hypothesis (1.10), namely one using a thermophoretically animated
particle as the tracer. A subsequent paper [25] addresses a similar experiment based
upon the diffusiophoretic velocity of a particle, the analysis and interpretation of
which further supports the hypothesis. While the pertinent isolated-particle data
currently available in support of Eq. (1.10) are thus quite limited in terms of
quantity, it will be seen, however, that the purely kinematical equality (1.10)
possesses considerable intuitive appeal in a dynamical sense, in terms of providing a
rationale for the otherwise seemingly empirical constitutive relation, v = v
v
to be
employed in Eqs. (1.1) and (1.7). That is, if Eq. (1.10) is, in fact, true, then so also is
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 85
the resulting expression v = v
l
; derived from Eq. (1.4), the latter representing a
dynamical relation quantifying the diffuse transport of momentum through space in
connection with Eq. (1.1).
This brings us to the real motivation underlying the writing of this section, namely
that of offering a rational explanation for the seemingly empirical fact that the uid
continuums velocity v appears to be given generally by the uids volume velocity
v
v
; rather than by its mass velocity v
m
: As suggested above, we intend to do this by
introducing the additional notion of the tracer velocity of the uid v
l
(to which we
shall also refer as the uids Lagrangian velocity, accounting thereby for the
appended subscript), and subsequently arguing that the uids tracer velocity is
identical to its volume velocity, as set forth in Eq. (1.10). A tracer is a material object
that moves through space, but not through the uid. Indeed, the tracer moves with
the uid, being (by denition) conveyed piggy-back by the latter. The tracer thus
moves through space in a manner no different from that of an elementary mass
point moving though space. As such, repeating the closing comments of the
preceding paragraph, it is intuitively obvious as to why the tracer velocity v
l
should
qualify as the uids velocity. And since we intend to experimentally demonstrate
the viability of (1.10), albeit only to the data-limited extent currently possible, the
reason behind the seemingly empirical constitutive choice (1.4) for the uids velocity
will then be seen to be transparent.
6.1. Measurement of the uids tracer velocity
During the course of seeking to experimentally demonstrate the contention (1.10),
it is necessary at the outset to rst address in detail the question of how, at least in
principle, the uids tracer velocity eld v
l
[x(x
0
; t); t]; dened by Eq. (1.9), is to be
measured. By way of reviewing the generic background underlying the experimental
measurement of any eld variable (of which the uids velocity is but an example),
and as prelude to such an undertaking, we note that the operational or experimental
denition of any continuum eld variable at a point x in space involves inserting an
appropriate pointsize entity into the eld at that point, and subsequently performing
some measurement on that entity, such that the magnitude (and direction, if
relevant) of the entitys response quanties the strength (and, possibly, direction-
ality) of the undisturbed eld at that point. Of course, the entity needs to be of such a
nature as to not sensibly perturb the very eld that one is objectively attempting to
measure; hence, the requirement, inter alia, that the responsive entity be effectively
pointsize in its dimensions. But it must not be so small as to be able to interact
responsively with the individual sub-continuum (i.e., molecular) units of which the
hypothetical continuum is composed; otherwise, its response will not be to the
hypothetical continuum as a meaningful macroscopic physical entity in its own
right, but, rather, to the discrete microscopic sub-continuum molecular objects of
which the continuum is composed.
From a practical point of view, no such pointsize entities exist. Accordingly,
application of this scheme in practice, at least for steady uid motions [80], involves
the following sequential steps: (i) successively inserting, in the neighborhood of the
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 86
same specied point x of interest where one wishes to measure the eld, a series of
otherwise identical particles of ever-decreasing size; (ii) measuring and recording
their respective responses; (iii) subsequently extrapolating these measurements to the
limit of zero particle size; and (iv) repeating this sequence of steps with a series of
other geometrically, physically, or physicochemically different particles, so as to be
assured that the resulting, zero-size measurement obtained is independent of the
physical properties of the particle, and hence dependent only upon those of the
entraining uid itself. In circumstances where particles of different shape or different
physicochemical properties might yield different extrapolated measurements, one
must identify among those different particles that class or group of extrapolated
measurements characterized by having registered the same zero-size values of v
l
: For
example, in the case of the thermophoretic experiments described in Section 3, the
latter, perhaps seemingly abstruse point, becomes pertinent with respect to the
thermal conductivity k
s
of the tracer particle when measuring the uids tracer
velocity v
l
; cf. the subsequent remarks made in connection with Eqs. (6.1) and (6.2).
Thus, in the particular case where the uids tracer velocity eld v
l
(x; t) constitutes
the continuum eld of interest, one carries out the above-cited scheme (at least in
principle) by sequentially introducing a series of ever smaller rigid particles, say,
spherical in shape, and of diminishing radii a (all, however, sufciently large such as
to manifest no sensible Brownian motion) into the owing uid, with their respective
centers of volume initially situated in the neighborhood of the point x of interest.
Subsequent measurement of the vector displacement Dx of each such particle
through space during an appropriately small time interval Dt (using, say, streak
photography or some equivalent particle-tracer velocimetry measurement [9]), then
furnishes the particles radius-dependent vector velocity, U = U{a]; where U :=
lim
Dt0
Dx=Dt dx=dt at the point x: Extrapolation of these velocity data to zero
sphere radius then furnishes the pointsize tracer velocity, U{0] (qx=qt)
x
0
;
representing the uids tracer velocity v
l
with which the undisturbed uid is moving
at the point x in question. Repetition of this procedure at other points of the uid
serves to map out the entire uid tracer velocity eld. As earlier mentioned, it is
important that the uid adhere to the particle, so that the no-slip boundary
condition, imposed upon v
m
or v
v
; whichever is appropriate, is satised.
6.2. Tracer velocity in thermophoretically-animated ows
The experimentally conrmed equation (3.1) reveals, rather surprisingly, that a
non-Brownian thermally animated sphere moves (relative to the steady temperature
eld) with a velocity that is independent of its radius [81,82], at least in the effectively
zero Knudsen number continuum regime! An identical phenomenon is observed in
compilations of data derived from comparable (isothermal) diffusiophoretic
experiments [25,37,83], where the movement of suspended aerosol particles [relative
to the local mass velocity, cf. Eq. (7.1)] in binary gas mixtures arises from the
existence of species-concentration-induced density gradients Vr; with r = r(w): The
fundamental and profound signicance of this size-independence, occurring in all
phoretic experiments performed under effectively zero-Knudsen number circum-
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 87
stances, appears to have escaped the attention of previous investigators. This
size-independence, down to the limit of an effectively zero-size particle (a tracer),
will be seen to provide the experimental basis for our hypothesis regarding
the equality v
l
= v
v
of the tracer and volume velocities of the particle-free
uid itself.
In the limiting case, k
s
=k51; of an effectively insulated or nonconducting sphere,
Eq. (3.1) (see the derivation applicable to this nonconducting case in Appendix B)
governing the thermophoretic velocity of a particle reduces to the form
U = abVT : (6.1)
The experimental conditions under which this equation applies are precisely those
which qualify the experimentally conrmed thermophoretic particle velocity (6.1) to
be identied as the undisturbed uids tracer velocity, v
l
U{0]; namely the uids
intrinsic velocity through space in the absence of the particle. Specically, the
following requirements are met during the thermophoretic experiments underlying
(6.1): (i) owing to its applicability in the Kn = 0; continuum limit, Eq. (6.1) fullls the
requirement of representing the velocity of a non-Brownian particlethat is, a
particle sufciently large such as to be unresponsive to direct molecular-level
phenomena, and hence to be capable of responding only to continuum-level
phenomena; (ii) the, in-principle, extrapolation of a sequence of variable sphere-size
measurements to the hypothetical zero-size limit proves to be unnecessary in present
circumstances, since the experimental measurements conrming Eq. (6.1) reveal the
particle velocity to be independent of size, and hence not requiring size
extrapolation; (iii) the passivity requirement demanded of the tracer particle in
order that its measured velocity qualify as the uids tracer velocity v
l
is fullled by
the choice of a nonconducting particle (i.e., k
s
=k51), whereby, consistent with the
nature of this passivity requirement, Eq. (6.1) shows the particle velocity to depend
only upon the properties of the uid through which the particle moves,
independently of any physical attributes of the particle itself [84]; (iv) an additional
and important feature further supporting Eq. (6.1) as the (particle property-
independent) tracer velocity of the uid is the fact that in the zero particle
conductivity case Eq. (6.1) holds irrespective of particle shape (i.e., for nonspherical
particles), and independently of the orientation of such particles relative to the
direction of the temperature gradient, as is demonstrated to be the case in the
accompanying thermophoretic paper [23], as well as in Appendix B. And it is only in
this effectively zero conductivity case, k
s
=k51; that particle shape- and orientation-
independence obtains!
Given these circumstances together with the fact that v
l
U{0] we may write
v
l
= abVT in place of Eq. (6.1), since the experimentally conrmed particle
velocity (6.1) has now been established as being the tracer velocity of the gas relative
to the container walls. Inasmuch as the (undisturbed) temperature eld is also steady
relative to the container walls, this temperature eld is then, by denition, not being
convected relative to the constant-temperature walls. As such, since the velocity v
l
;
has been measured relative to these walls, this requires that v
m
= 0 with respect to
this same reference frame. Accordingly, one can infer from the experimentally
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 88
observed absence of wall effects on particle motion [effects that, if present, would
otherwise be manifested in Eqs. (3.4), (3.5) or (6.1) by the explicit appearance therein
of the ratio a=L of sphere radius a to wall separation distance L] that the walls play
no explicit hydrodynamic role in thermophoretic migration phenomena. Accord-
ingly, the relative motion of the undisturbed uid velocity eld with respect to the
temperature eld can be expressed objectively in the following reference-frame-
independent form [23]:
v
l
v
m
= abVT : (6.2)
According to this strictly uid-related, particle-free relation, the uids tracer and
mass velocities, v
l
and v
m
; are indeed different, at least in the presence of a
temperature gradient!
The strictly kinematical measurement of the motion of an entrained particle
manifested in Eq. (6.1) has nothing whatsover to do with the dynamical
circumstances (i.e., temperature gradients in the present thermophoretic case, or
molar species concentration- and molecular weight-gradients in the gaseous
diffusiophoretic case [25]) that animate the uids undisturbed motion v
l
existing
in the tracers absence. This is true whether the dynamical forces causing the pre-
existing uid motion v
l
into which the particle is subsequently introduced be caused
by temperature gradients, pressure gradients, buoyant gravitational forces (i.e.,
natural convection [85]), or otherwise. After all, the entrained particle plays no role
in creating the pre-existing undisturbed uid motion v
l
that exists prior to the
particles introduction into the uid. That the dynamical circumstances giving rise to
the entrained particles motion are kinematically irrelevant is evidenced by the fact
that a uid-mechanical experimentalist, concerned exclusively with kinematical
velocity measurements, need not have at her/his disposal a thermomometer,
manometer, gravitometer, or any other similar device in order to objectively
measure the movement of tracer particles, and, hence, to subsequently report his
ndings of the respective magnitude and direction of the undisturbed uid velocity
eld, as measured by the zero-size tracers response to that eld. In short, he/she
need have no idea whatsoever as to the physical cause(s) of the uid motion that he is
attempting to measure (see, however, the comments in Ref. [84] with regard to the
need to perform replicate experiments using tracers of varied materials and shapes).
Indeed, the less he knows about the origin of the uid motion upon which he is
reporting, the less biased will he be when interpreting his purely kinematical
observations, and, concomitantly, the more objectively will his results be viewed by
independent referees reviewing his work.
The preceding interpretation of the thermophoretic data leads to the seemingly
paradoxical conclusion that the (particle-free) uid is moving, v
l
a0; despite the
fact that the uid is apparently at rest, v
m
= 0: That is, the uid was already in
motion with velocity v
l
prior to the introduction into it of any foreign (tracer)
particle! How can this be? Resolution of this paradox lies in the fact that the vector
eld v
m
is a velocity in name only, but not in a true physical sense. In this context,
it is important to note that the sole Eulerian continuum ow eld parameter, other
than r; capable (at least in principle) of being directly measured by performing an
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 89
experiment at the space-xed point x; is not the velocity vector v
m
(x; t); but is, rather,
the mass ux density vector, n
m
(x; t): As earlier discussed in section 1 (see Ref. [5]),
the latter is dened such that the extensive quantity dS
.
n
m
constitutes the
instantaneous rate, at time t; at which mass ows across a space-xed directed
element of surface area dS centered at x: In contrast with n
m
; the indirectly derived
intensive vector eld, v
m
:= n
m
=r; is merely a dened quantity, essentially nothing
more than a mathematical symbol, to which one may, or may not, seek to assign a
physical interpretation. Velocity, on the other hand, viewed as a physical concept in
the classical Newtonian sense of referring to the movement of a real material object
through space, pertains to the motion of a pointsize object, initially situated at a
point in space, say, x
0
; at time t = 0; and subsequently traversing a denite
trajectory, x = x(x
0
; t); as discussed in connection with Eq. (1.9).
In contrast with v
m
; the uids tracer velocity v
l
clearly fullls the physical
notion of a velocity in the sense of an object of indisputable time-independent
integrity executing a deterministic trajectory through space, since the object with
which this velocity is experimentally identied is, in fact, a material entity,
namely the pointsize tracer, a foreign body deliberately introduced into the uid
in order to monitor the latters motion. As such, it is clearly not arbitrary to
identify the tracer velocity v
l
as being the velocity of the uid continuum.
The velocity v
l
would therefore appear to qualify as being the appropriate purely
kinematical uid velocity v referred to in textbooks in the context of uid
movement through space (although it is not the velocity appearing in the con-
tinuity equation, nor is it the uids specic momentum, both of which correspond
to v
m
).
6.3. Conrmation of the hypothesis embodied by Eq. (1.10)
Eqs. (1.5) and (1.6) combine in the present single-component heat-transfer case to
yield v
v
v
m
= aVln ^ v: In circumstances where the thermal law of adiabatically
additive volumes applies, whereupon viscous dissipative effects are necessarily
negligible, and in which the effects of pressure on specic volume are small compared
with those of temperature, so that ^ v = ^ v(T); the preceding expression in this
paragraph adopts the form
v
v
v
m
= abVT : (6.3)
Comparison of the latter with the experimentally conrmed tracer relation (6.2)
shows that the hypothesized equality v
l
= v
v
is indeed valid. As Eq. (6.2) has been
experimentally established as applicable to both gases and liquids (albeit with less
condence in its validity for liquids owing to the current dearth of appropriate
liquid-phase thermophoretic data), we conclude, tentatively, that the equality (1.10)
is general in scope. Admittedly, this has thus far been conrmed only for the
thermophoretic case. However, a similar analysis of experimental isothermal
diffusiophoretic data, coupled with a theory of that phenomenon based upon the
constitutive relation v = v
v
applicable to the binary mass-transfer case [25], further
supports (1.10) [see also Eq. (7.1)].
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 90
6.4. Further discussion of the notion of a tracer velocity
The notion of the tracer velocity, v
l
(x; t) v
l
(x
0
; t); of a uid (the latter functional-
dependence notation involving x
0
serving to emphasize the particle-like trajectory
aspect of the uids tracer velocity) is a strictly continuum concept. As such, no
direct gas-kinetic molecular interpretation [86] of v
l
appears possible. This is equally
true of the uids volume velocity, v
v
(x; t): Despite the fact that particle-tracer
velocimetry is a widely practiced and seemingly elementary experimental technique
for measuring uid motion, given our prior discussion the possible limitations of
tracer velocimetry appear not to have been scrutinized with the fundamental care
that they deserve, indeed require, in order for such measurements to be regarded as
accurately monitoring the undisturbed uid motion.
As a case in point, consider Eq. (3.1) or, more properly, the formally equivalent
expression
U = (v
m
)
0

ab
1 (k
s
=2k)
(VT)
0
; (6.4)
representing the velocity U of a non-Brownian force and torque-free spherical particle
(not necessarily a tracer, owing to the particles nonzero conductivity) immersed in a
nonisothermal uid whose undisturbed mass velocity and temperature elds are,
respectively, v
m
and T; the particle being supposed situated far from any boundaries.
The subscript 0 connotes evaluation at the point in the uid presently occupied by
(say, the center of volume of) the particle. In circumstances where the particles
conductivity is small compared with that of the uid (k
s
=k51); the particle can serve as
a tracer of the undisturbed uid motion, even in the presence of temperature gradients,
with the uids tracer velocity at the point 0 given by the expression U = v
m
abVT;
as in Eq. (6.2). On the other hand, in the opposite case, where k
s
=kb1; one nds that
U = v
m
; whence the thermophoretically animated particle no longer serves as a faithful
monitor of the undisturbed uids tracer motion, through space. Note that the particle
velocity U is size-independent in both cases, although in the highly conducting particle
case the particle velocity U will no longer be particle shape- and orientation-
independent [23], an important point; that is, in the highly conducting particle case the
formula U = v
m
would be valid only for a spherical particle. (Presumably, the
experimental observation of this departure from comparable results recorded with
other classes of particles would enable a thoughtful and rigorous experimentalist to
discover that this particular class of particles was not passive.) Additionally, it is
important to note that both limiting results hold only when the no-slip condition, Eq.
(5.1), imposed upon v
v
; obtains. The ideal, inviscid uid example discussed earlier
emphasizes the importance of the no-slip condition in tracer experiments designed to
monitor the intrinsic motion of the undisturbed, particle-free uid.
6.5. Concluding remarks
It is important to recognize that the experimentally observed difference in
velocities between v
l
and v
m
in Eq. (6.2) is not simply an artifact of the particular
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 91
class of phoretic thermal experiments that brought this velocity difference to light.
Stated explicitly, the distinction is intrinsic to the very nature of inhomogeneous
continuum uid mechanics itself, and arises from the existence of a density gradient,
Vr; rather than from a temperature gradient in the single-component case (or a
species concentration difference in the isothermal binary diffusion case). Explicitly,
from Eqs. (1.5) (1.6) and (1.10) the generic tracer/mass velocity difference is
v
l
v
m
= D
v
Vln r ; (6.5)
with D
v
equal to either a or D; thereby revealing the density gradient to be the
fundamental source of the velocity disparity for both thermophoretic and
diffusiophoretic particle motions. As such, the driving force animating thermo-
phoretic particle motion, for example, should not be attributed to the temperature
gradient; for, as is clear from Eq. (3.1), were the uids thermal expansivity b to be
zero, neither the uid nor the entrained particle would move, temperature gradients
notwithstanding. Thus, it is the existence of the density gradient Vr; rather than the
temperature gradient VT; to which the uids tracer motion should be attributed. In
this same context, the thermal stress T

(Section 4) should, more fundamentally, be


termed a Korteweg stress (as subsequently discussed in Section 8); for were b = 0 at
some intermediate point in a uid within which the temperature varied, as is possible
in the case of some liquids (e.g. water at 4

C), no thermal stress would presumably


exist at that point despite the presence there of a temperature gradient.
At least in the zero Knudsen number continuum uid regime, the latter
observation regarding expansivity belies the commonly held belief [43] that a freely
suspended particle moves through a gas because those molecules striking the particle
from its hotter side are more energetic, and hence possess greater momentum, than
those on its colder side, resulting thereby in the transfer of momentum from the gas
to the particle, thus causing movement of the particle through the suspending uid
towards the lower temperature region. (This argument may, however, possess some
measure of validity in the large Knudsen number, noncontinuum regime.) Rather,
the ultimate source of the particle motion resides in the fact that the uid itself is
moving (v
l
v
m
a0) relative to the space-xed temperature eld and, consequently,
relative to the space-xed walls. Since no force acts upon the unrestrained
thermophoretic particle, from the vantage point of an observer xed in the walls
(for which v
m
= 0), the particle is simply entrainedthat is, carried along piggy-
backby the moving uid (v
l
a0) [except, possibly, for the slip effect seen in Eq.
(3.1) due to particles nonzero thermal conductivity k
s
; causing the sphere to lag the
uid]. Indeed, were one to accept this discredited hot-side/cold-side argument,
when the sphere radius became vanishingly small, no particle sides would remain
to permit such a distinction, whence a very small particle would presumably not
move! Yet, based upon its size-independence, the theoretically conrmed expression
(3.1) shows, at least in the continuum limit, that the particle would indeed continue
to move through space relative to the temperature gradient, were its radius to
become vanishingly small.
Finally, as practical matter, we note that in circumstances where the uids
undisturbed mass velocity satises the inequality v
(0)
m

b j
(0)
v
= abVT
(0)

; Eq. (6.4)
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 92
shows that U - v
(0)
m
; in which case the experimentally measured particle velocity U
would closely approximate the uids tracer velocity, v
l
:
7. Further evidence in support of the v = v
v
hypothesis
Direct measurement of the uids volume velocity (namely the ux density of
volume n
v
[5]) at a point of a uid appears infeasible on purely theoretical grounds,
much less as a practical possibility. Accordingly, any evidence offered in support of
the fundamental hypothesis (1.4), as well as of the accompanying supplementary no-
slip boundary condition hypothesis (1.7) associated therewith, must necessarily take
an indirect form.
In this context, we note that the evidence offered thus far in support of the above two
interrelated hypotheses derives from experimental thermophoretic data for gases and, to
a much more limited extent, comparable data for liquids. Such corroboration involves
the observed agreement, discussed in Section 3, between the theoretically derived
equation (3.1) [23] and the experimentally based semi-empirical thermophoretic velocity
correlation (3.5) for gases, as well as with the raw liquid-phase velocity data of McNab
and Meisen [45]. Such agreement provides continuum-mechanical support for the
viability of the relation v = v
v
in the context of the no-slip condition (1.7). On the
molecular side of the ledger, our derivation in Section 4 of the nontraditional Newtonian
viscosity constitutive equation (4.8) from the Burnett solution [6,7] of the Boltzmann
equation (for monatomic Maxwell molecules) provides independent, molecularly based,
noncontinuum-mechanical evidence in support of the volume velocity hypothesis v = v
v
in the context of Eq. (1.1). This latter verication is accomplished independently of the
supplementary no-slip boundary condition (1.7), whose validity (or lack thereof) does
not enter into the question of the uids rheological equation of state for the deviatoric
stress T: Reciprocally, our demonstration in Section 5 of the fact that the experimentally
and theoretically based Maxwell slip velocity formula (5.2) for ideal monatomic gases
accords constitutively with our proposed no-slip volume velocity boundary condition
(5.1), offers additional independent evidence in support of our proposed nontraditional
no-slip boundary condition (1.7) (1.4).
What follows below is a summary of further, independent evidence offered in
afrmation of the hypothesis (1.4) (or, less stringently in several cases, at least of the
fact that vav
m
) and/or the proposed no-slip boundary condition (1.7)/(5.1) [87].
Each such evidentiary item is further documented in detail elsewhere. The evidence
offered below is of a tripartite nature, consisting of experimental, molecular, and
continuum-mechanical documentation. Each of these three distinct categories is
separately discussed below.
7.1. Additional experimental evidence
7.1.1. Diffusiophoresis
Diffusiophoretic particle motion [25], occurring with a velocity U (v
m
)
0
relative
to the uids undisturbed mass motion v
m
existing at the point presently occupied by
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 93
the center 0 of the particle, is the exact analog of the thermophoretic particle
motion discussed in Section 3 [see also Eq. (6.4)]; now, however, the density gradient
animating the particle, as embodied in the generic Korteweg stress [cf. Eq. (8.1)], is
caused by a binary species density gradient in an isothermal uid rather than by a
temperature gradient in a single-species uid. In circumstances where the effects of
pressure gradient effects on the uids density are small compared with those caused
by species concentration gradients, one obtains the following theoretical formula for
the diffusiophoretic velocity of a nite-size solid particle whose physicochemical
passivity does not permit diffusion of either species through its interior, nor
adsorption onto its surface [29]:
U v
m
= DVln r : (7.1)
The subscript 0 has been omitted for simplicity in the above; r( r
0
) is the density
of the undisturbed uid in the neighborhood of the particle, and D is the mixtures
Ficks law diffusivity. This formula, which is applicable to both gases and liquids,
was derived [25] on the basis of the pre-constitutive equations set forth in Sections 1
and 2, with the velocity v assumed given by the nontraditional constitutive equation
(1.4), together with the comparable nontraditional boundary condition (1.7).
Additionally, the energy equation (2.9) has been replaced by the elementary binary
species transport equation [18],
r
D
m
w
i
Dt
V
.
j
i
= 0 (i = 1; 2) ; (7.2)
with w
i
the mass fraction of species i; and in which the diffusive species current is
assumed given constitutively by Ficks law, j
i
= rDVw
i
; valid in the case of
isothermal isobaric systems [88]. Additionally, the law of additive volumes for
thermodynamically ideal mixtures, (q^ v=qw
1
)
p;T
= const, was assumed applicable,
together with the functional diffusivity dependence for gases demanded by gas-
kinetic theory [18], namely D = D
+
=r with D
+
a density-independent constant [89].
Eq. (7.1), based upon our nontraditional uid-mechanical hypothesis (1.4), is
shown elsewhere [25] to agree well with experimental diffusiophoretic data for binary
gas mixtures in the continuum regime. It also accords with other theoretical
approaches to the rationalization of diffusiophoretic particle motion based upon the
traditional NavierStokesFick formulation derived from the classical velocity
choice (1.3) (and the inconsistent assumption that the mixture is incompressible,
V
.
v
m
= 0). Moreover, in these latter approaches, Maxwells slip boundary condition
(5.2) was replaced by the so-called KramersKistemaker [65] concentration-slip
boundary condition for gases, the latter representing the (empirical) analog [90] of
Maxwells Eq. (5.2) for gases, with the temperature gradient replaced by a species-
gradient counterpart. Similar to the issues outlined in Section 5 and documented in
Appendix C, it is fortuitous that these NSF analyses (F here referring to Fick
rather than Fourier) furnish theoretical results that accord with experiment.
Eq. (7.1) shows the diffusiophoretic particle velocity to be independent of particle
size. Moreover, as a result of the particles supposed passivity, U is also independent
of particle shape and, hence, of the particles orientation relative to the space-xed
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 94
species concentration gradient [25]. Accordingly, in the sense of particle-tracking
velocimetry experiments, such a particle plays the role of a passive tracer of the
uids undisturbed tracer velocity v
l
in the sense that U v
l
: The latter conclusion
accords with the generic theoretical formula (6.5) upon there replacing the symbol D
v
by the binary diffusion coefcient D:
7.1.2. Thermal transpiration
The phenomenon of thermal transpiration, discovered experimentally by Osborne
Reynolds [71] in 1879, was, historically, the rst phoretic phenomenon to be
explained using Maxwells slip condition (5.2). Indeed, as discussed in Section 5, it
was to explain Reynolds observations that Maxwell created the celebrated
Appendix to his 1879 paper [8,70], wherein Eq. (5.2) was rst derived and
subsequently applied to Reynolds data.
The so-called thermolecular pressure associated with thermal transpiration
phenomena arises during steady-state heat transfer in horizontal, single-component,
gas-lled, capillary tubes (radius a; length L) of large aspect ratio ( = a=L51)
possessing insulated side-walls, and whose two closed heat-conducting ends are
maintained at different temperatures. According to the traditional NSF equations
based upon Eq. (1.3) and satisfying the traditional no-slip boundary condition (1.7)
(1.3) on the side-walls and ends, the pressure should be uniform throughout the
capillary tube owing to the predicted absence of mass motion, v
m
= 0; and
concomitant absence of pressure gradients based upon the traditional creeping ow
equation (2.12) with v = v
m
and V
.
v
m
= 0: Explicitly, the only transport
phenomenon predicted to occur in such circumstances on the basis of the NSF
equations is that of one-dimensional steady-state heat conduction between the hot
and cold reservoirs (respectively, maintained at temperatures T
h
and T
c
) situated at
the opposite ends of the capillary, with the resulting local temperature gradient VT
at each point of the uid accompanied by a corresponding local density gradient Vr
in the supposedly isobaric system. However, upon performing this experiment with a
gas, Reynolds [71,91] observed a pressure difference between the two ends of the
capillary, with the pressure being highest at the hotter end (p
h
4p
c
): The observed
thermomolecular pressure difference p
h
p
c
40 for gases has since been conrmed
by others (cf. the review in Ref. [24] encompassing a wide variety of gases). In fact,
thermal transpiration enjoys a number of current practical uses [92].
As the thermomolecular pressure difference was not explicable by conventional
NSF continuum uid-mechanical arguments, Maxwell attempted to the explain
the phenomenon on the basis of his slip hypothesis (5.2), and, in so doing,
concomitantly give quantitative credibility to his earlier qualitative speculations
regarding the workings of Crookess radiometer [67]. It was in this context that
Maxwell, in the Appendix to his 1879 paper [8], rst introduced his notion of thermal
slip along the insulated side walls of the capillary tube, still, however, regarding the
conventional incompressible NSF equations as correctly quantifying the uid
motion ensuing within the capillary tube interior, as in Section 5. Maxwells analysis
has since proved successful [24] in explaining the observed pressure difference for
(monatomic) gases [93], at least to within a factor of two. However, as in all other
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 95
applications which invoke Maxwell slip (e.g. thermophoresis and diffusiophoresis),
thermal transpiration is regarded as a noncontinuum O(Kn
2
) phenomenon despite
use of the conventional NSF continuum equations in its rationalization. Again,
this viewpoint appears to be inconsistent. In contrast, Bielenberg and Brenner [24],
using the nontraditional equation set and no-slip boundary condition of Sections 1
and 2 based upon Eq. (1.4), arrive in the case of gases at the same thermomolecular
pressure difference, p
h
p
c
= 8C
s
(m
2
=rTa
2
)(T
h
T
c
); as is given by Maxwells
original 1879 formula [8] for monatomic gases (albeit with our C
s
=
3
2
appearing in
place of Maxwells C
s
=
3
4
) in capillary tubes of large aspect ratio. In the preceding, a
is the capillary radius and rT - const. throughout the tube owing to the assumption
of a small temperature difference, T
h
T
c
; and a concomitantly sensibly uniform
pressure thus prevailing throughout the ideal-gas-lled capillary tube. A more
general formula than that given here, based upon our theory, and presumably also
valid for liquids, is given in Ref. [24].
That the two thermomolecular pressure formulas, ours and Maxwells, for the
thermal transpiration case should agree despite Maxwells apparently inconsistent
use of the traditional incompressible NSF equations can be explained in the same
manner as set forth generically for single-component gases in connection with Table
2 in Appendix C. While agreement thus obtains in this case between Maxwells
v = v
m
noncontinuum slip scheme and our nontraditional v = v
v
continuum scheme,
the use of the Maxwell slip condition in conjunction with the incompressible NSF
equations results in a violation of the rst law of thermodynamics in the present
thermal transpiration case, as earlier noted in Section 5 (see Appendix D). This
violation owes to the nonzero mechanical work predicted by Maxwells theory to
have been done on the conned uid by the surroundings [through the action of
the T
m
= 2mVv
m
viscous stresses existing at the wall in conjunction with the
Maxwells thermal creep condition (5.2) prevailing there], with the resulting
mechanical energy thereby supposedly introduced into the uid ultimately being
dissipated via viscous action occurring in the uids interior. In turn, as a
consequence of this dissipation, and for the steady-state circumstances characterizing
thermal transpiration, more internal energy would leave the capillary at its cold end
than entered at its hot end. Such behavior constitutes a clear-cut violation of basic
thermodynamic principles, given that no heat or work from the surroundings enters
the uid through the rigid immobile insulated side walls. In contrast, no such
contradiction arises in our scheme where no slip of the true uid velocity v
l
(or v
v
),
based upon Eqs. (1.4)/(1.10), occurs at the capillary walls!
Summarizing the conclusions of this subsection, the agreement of our unconven-
tional continuum theory with thermal transpiration experiments, as detailed in Ref.
[24], is offered as further evidence of the viability of Eq. (1.4). Conversely, the
resulting violation of thermodynamic laws implicit in Maxwells scheme is offered as
further evidence of the fact that the latters thermal-stress slip theory cannot be
regarded as a correct physical explanation of the phenomenon of thermal
transpiration, despite fortuitously furnishing the mathematically correct result for
the thermomolecular pressure difference. Our thermal transpiration theory,
emphasized above for gases, is presumably equally applicable to liquids [24], for
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 96
which physical state no other theory currently exists. Insofar as we are aware, no
liquid-phase experimental data bearing on thermomolecular pressure is yet available
to test this prediction. This poses an obvious challenge for experimentalists.
7.2. Additional molecular evidence
7.2.1. Nelsons stochastic mechanics [94]
Nelsons [95] stochastic-mechanical Brownian motion model of uid-physics is
based upon the It ^ o calculus [96]. Heretofore, Nelsons work has been employed
conceptually only in the context of posing an alternative deterministic interpretation
to the conventional probabilistic Copenhagen interpretation of the foundations of
quantum mechanics. However, when addressed instead to ordinary uid motion in
viscous continua, Nelsons model appears to implicitly afrm the purely kinematical
aspects of our theory. Explicitly, Nelsons relation between the uids respective
tracer and mass velocities, v
l
and v
m
; as derived from his model of uid motion is, in
our notation [97],
v
l
= v
m
DVln r : (7.3)
Nelsons purely kinematical Eq. (7.3) is equivalent to a composite of our Eqs.
(1.5)(1.6) and (1.10), independently of the dynamical aspects of our theory, the
latter as embodied in the use of either of Eqs. (1.3) or (1.4)] in Eq. (1.1) or (1.7).
Nelson designates DVln r as an osmotic velocity, with D presumably a self-
diffusion coefcient, the analog of Einsteins Brownian diffusion coefcient [98100]
appearing in the Ito calculus formula for single-component stochastic processes.
Nelsons version of Eq. (7.3) traces its quantum-mechanical roots back to
Schro dingers [101] search for a deterministic, nonprobabilistic foundation for
quantum mechanics, one based upon more classical Newtonian-like mechanical
concepts. Nelsons original Brownian motion-based theory [95], including its relation
to quantum mechanics [102,103], has since been elaborated by Garbaczewski et al.
[104,105], as well as by others [106]. From their common perspective the osmotic
term in Eq. (7.3) constitutes an extended version of the Einstein [107]Smoluchowski
[108] Brownian motion theory for a foreign (colloidal) particle dispersed in a solvent,
representing an essentially purely kinematical theory [109,110]. In Nelsons analysis
the material entity undergoing the diffusive process quantied by the coefcient D
effectively serves as a Brownian tracer of the undisturbed uid motion (7.3). The
existence of Nelsons osmotic velocity is attributed to what Garbaczewski [104] terms
a Brownian recoil effect of the solvent uid upon the solute tracer particle.
As regards the physical interpretation of Nelsons forward drift velocity v
l
;
Garbaczewski [104] states that (with x a generic point of the uid): If the diffusion
pertains to massive [point] particles, we have a natural physical interpretation of the
forward drift as the mean velocity of particles leaving x at time t [along sample
paths]. . . : In support of this view he goes on to further state that: The mean
position evaluated along sample paths of outgoing particles, a time Dt after they left x
at t, is x v
l
(x; t)Dt: Clearly, this interpretation of Nelsons forward drift velocity,
represented by the left-hand side of Eq. (7.3), accords with the requirements
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 97
demanded of a particle in order that the latter qualify as a tracer of the
undisturbed uid motion in the sense described earlier, namely necessitating that the
tracer be: (i) non-Brownian (i.e., massive relative to the masses of individual uid
molecules); (ii) pointsize in the particle-size-extrapolated limit; and (iii) inert or
passive.
In summary, the apparent agreement of Nelsons extended Brownian motion-
based kinematical formula (7.3) with our nontraditional kinematical view of the
uids tracer motion appears to add to the credibility of our volume velocity
hypothesis (1.4).
7.2.2. Klimontovichs statistical mechanics
The late Russian statistical-mechanician, Yuri L. Klimontovich (19242002) [111]
proposed a fundamental modication of the Boltzmann equation which involved
adding to the traditional collision integral pertaining to transport processes
occurring in velocity space an additional dissipative self-diffusion-type term
involving transport processes occurring in physical-space [112]. According to its
author, inclusion of the latter term results in a unied description of kinetic,
hydrodynamic, and diffusional processes, presumably valid for all Knudsen
numbers. The equations governing gas dynamics are then derived from the
preceding generalized kinetic equation without, however, invoking the standard
small Knudsen number perturbation theory arguments [6,7]. In the case where body
forces are absent and for the single-component uid case, the results obtained in this
manner by Klimontovich are identical to our modied Eqs. (4.1) for the deviatoric
stress and (2.6) for the internal energy ux, in which the quantity j
v
appearing in
these two equations is given by Eq. (1.6). Klimontovichs theory is based upon the
assumption that the uids thermometric diffusivity a; kinematic viscosity u; and self-
diffusivity D are all, equal, conditions not fully realized in practice. In
Klimontovichs work, the quantity that we have termed j
v
is identied by him as
constituting a self-diffusion ux rather than a volume ux.
As our continuum-mechanical results were hypothesized and compared with
existing data prior to becoming aware of Klimontovichs statistical-mechanical work
(which, incidentally, bears no obvious relation to the Maxwell-Burnett nonconti-
nuum notion of thermal stresses in gases) we regarded and continue to regard the
agreement between the two theories as offering conrmation of both theories. This is
especially true in view of the fact that Eq. (1.5) in the form v
m
= v
v
aVln r (our
notation) also appears naturally in Klimontovichs work, with v
m
implicitly
identied therein as being the uids mass velocity by virtue of its appearance in
the latters continuity equation (1.2) (again, our notation). Klimontovich uses the
symbol u for what we have here identied as the volume velocity v
v
; although no
comparable volume-related physical identication is assigned by him to u: While the
two modied theories agree exactly with regard to the common equations cited
above, they differ nevertheless in one important respect. Explicitly, in Klimonto-
vichs momentum and energy equations, the three velocities ( ^ m; v
k
; v
w
) upon which
we focused in Appendix A and identied therein as being v
m
; are, in fact, identied in
his work as being v
v
: This is amusing because in the original version of our present
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 98
theory we made exactly the same identication as did Klimontovich for these
three velocities. However, we subsequently had occasion to question this view,
owing to the fact that it appears to violate Newtonian mechanics, as per the
discussion of Appendix A. That is, Klimontovichs proposal in regard to the three
velocities appears to be in conict with the basic principles of classical mechanics.
This becomes especially clear upon including in the momentum equation body
forces (real and ctitious), as we have done in Appendix A. We confess to being
conicted about our altered choice of v
m
over v
v
; an issue which we plan to
review in the near future. In any event, the issue is moot at the present time
in the sense that, owing to the fact that these velocities prove to be negligible in
the creeping ow problems underlying the phoretic phenomena addressed, the
choice does not impact upon the experimental conrmation of the several hypo-
thesis that we have effected regarding the proposed modications of the NSF
equations.
7.3. Additional continuum-mechanical evidence
7.3.1. Fitts on diffuse internal energy ow [14]
Our modied theory of transport in uid continua centers directly on the twin
issues of the constitutive equation (1.4) and the related no-slip boundary condition
(1.7)/(5.1), both appearing supercially to pertain only to momentum transport.
However, a third transport-related issue enters indirectly into consideration, namely
that embodied in the pre-constitutive equation (2.6) involving the internal energy
current j
u
; whose explicit constitutive form is implicitly required in the basic internal
energy transport equation (2.9). In particular, the nontraditional precursor
thermodynamic work term pj
v
appearing therein, arising from the diffusive
transport of volume, is not present in any of the standard texts on transport
processes or irreversible thermodynamics, the sole exception (of which we are aware)
occurring in the book of Fitts [14]. In place of what we have denoted as pj
v
in Eq.
(2.6) there appears in Fittss expression for the internal energy ux density [see Fitts
Eqs. (3-21) and (3-25)] the term p

N
i=1
v
i
j
i
[113], with v
i
the particle specic volume of
species i: In isobaric isothermal multicomponent mixtures [4,13] one has that v
v
=

N
i=1
v
i
n
i
; with n
i
= n
m
w
i
j
i
the Eulerian ux density of species i: Inasmuch as
r =

N
i=1
w
i
v
i
; it thus follows from Eq. (1.5) in conjunction with the denition n
m
=
rv
m
of the mass velocity that
j
v
=

N
i=1
v
i
j
i
: (7.4)
Fitts extra energy ux term is clearly identical to our nontraditional pj
v
term,
although in his analysis this term arises only in the context of multicomponent uid
mixtures, being absent in the case of single-component uids. This contrasts with our
own work [4,21], where a diffusional volume ux contribution pj
v
adds to the
thermodynamic rate of working at a material surface, even in the case of pure uids
(for which j
v
= aVln r). Nevertheless, despite this issue of its domain of
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 99
applicability, we regard the analysis of Fitts (and, to a lesser extent, that of the
closely related work of de Groot and Mazur [113]) as offering an independent
measure of support for our nontraditional NSF theme.
7.3.2. Prior speculation in the literature regarding the equality v
l
= v
v
(for
multicomponent mixtures)
The evidence to be presented in this subsection cannot strictly be counted
as legally admissible in the same quantitative sense as in the examples offered
above. Nevertheless, the musings of other researchers whose work, described
below, relates peripherally to our nontraditional view, would appear to constitute
something more than mere hearsay. In this context, we refer here to remarks by
Onsager [114] and Haase [115117], independently made in connection with
their respective discussions of isothermal diffusion phenomena, and apparently
implying the validity of Eq. (1.10), namely v
l
= v
v
: (Of course, in the context of
their pronouncements on the latter equation, both authors already understood
that, generally, v
v
av
m
in multicomponent uids owing to species concentra-
tion-induced massdensity gradients. As such, it can be condently asserted that
each appreciated the fundamental fact that v
l
av
m
; one of the main items in
our theory.)
The hints found thus far in the literature pertaining to our proposed equality,
v
l
= v
v
; each involve the role of the volume velocity in multicomponent diffusion
problems. Thus, Onsager [114], upon recognizing the strictly relative, rather than
absolute, nature of the diffusional process occurring among the constituent species in
a mixture, refers to what he calls a hydrodynamic velocity in such problems. He
then proceeds to identify, albeit only implicitly, the latter velocity with what is now
called the volume velocity [4,1315], a name with which he was apparently
unfamiliar at the time, as would appear from the context of his remarks. Haase
[15,115], on the other hand, refers explicitly to the volume-average velocity v
v
in
multicomponent mixtures as being the convective velocity, presumably v
l
; clearly
distinguishing the latter from the mass-average velocity v
m
of the mixture, and
obviously understanding that the former, rather than the latter, represents the
physical velocity with which the uid moves through space. (That is, by the word
convection one understands the conveyance through space of an entity entrained
in a owing uid. Hence, the phrase convective velocity on the part of Haase
would appear to refer implicitly to the movement of a tracer entrained in the uid,
rather than to the movement of a hypothetical material particle, the latter moving
at the mass-average velocity v
m
:) The contexts of each of these two authors was
kinematic rather than dynamic, although Onsagers use of the word hydro-
dynamic suggests an intuitive understanding of the dynamical issues owing to the
mass ow engendered by the molecular diffusion processes. Interestingly, Maxwell
[118] himself refers to experimentally: . . .measuring [sic] the velocity by its volume
which passes through a unit area rather than by the distance traveled by a molecule in
unit of time, apparently regarding these as being equivalent, although he may have
intended these remarks to apply only in the uninteresting case of incompressible
uids, wherein v
l
= v
m
:
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 100
7.3.3. Streater et al. on the N S F equations
Streaters ongoing theoretical studies [119123] of single-component nonisother-
mal uids (termed by him compressible Navier Stokes with temperature, to which
topic he subsequently refers using the acronym C-N-S-T) raise questions similar to
those raised here, albeit from a very different physico-mathematical perspective,
namely statistical dynamics [123,124], a eld of inquiry closely related to the eld
of information dynamics. Grasselli and Streater [120], like us, argue that problems
exist with respect to the accepted form of the NavierStokes equations. This
important point is emphasized in their opening paragraph by the statement that: An
unusual feature of the system is that the Euler continuity equation acquires a diffusion
term. This bulk diffusion does not appear in the standard theory [124], but has arisen in
some other work [125]. It is likely that our equations are more stable and physically
more accurate than the usual Navier Stokes equations. While the actual form of
their diffusive term differs slightly from ours, this is likely accounted for by the fact
that the GrasselliStreater hopping rule [120], quantifying the process of
thermalization upon which their detailed calculations are based, is somewhat ad
hoc. This leads also to the apparent inability of these authors to explicitly relate their
thermalization parameter l directly to a physical property of the uid, such as the
uids thermometric diffusivity. Grasselli and Streaters [120] diffusive term spills
over into their momentum and energy equations. However, their ad hoc
thermalization model does not permit a direct comparison of their momentum
and energy transport equations with ours.
8. Discussion
This section addresses, inter alia, an apparent relation existing between the
MaxwellBurnett temperature-gradient stress and Kortewegs density-gradient
stress. Furthermore, we contrast our present body-force-free phoretic phenomena
with another class of phoretic motions animated by body forces appearing in the
linear momentum equation. The latter class, illustrated by electrophoresis, while
supercially appearing to involve slip of v
m
at the particle surface, actually
involves velocity jump across a thin boundary layer (the region in which the body
forces are sensible) accompanied by no slip of v
m
at the physical surface of the
particle. The important body-force distinction existing between these two classes of
particle locomotion is used to compare and contrast our model of thermophoresis in
liquids with an alternative theory of the phenomenon due to Semenov and Schimpf
[47]. Additionally, several potentially practical applications of the present theory are
discussed, extending beyond the realm of strictly phoretic phenomena.
8.1. Korteweg stresses
In 1901 Korteweg [63] introduced the notion of a density gradient-induced
mechanical stress tensor, above and beyond the standard Newtonian viscous stress
tensor, Eq. (4.2). After a long period of relative dormancy, the subject of Korteweg
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 101
stresses was resurrected in connection with interfacial phenomena. In the case of
immiscible uids this occurred in connection with the CahnHilliard stress tensor
[126,27], where the interface between the two bulk uid phases is regarded not
as a singular surface but rather as a diffuse interfacial region of effectively nite
thickness, whose local equilibrium and transport properties, particularly the
density r; vary continuously across the interfacial transition region. Similar diffuse
interface uses for the Korteweg stress were later adopted by Joseph and others
[26,127,128] in relation to the phenomenological possibility of a transient interfacial
tension arising between a pair of miscible uids (of different densities) brought
suddenly into intimate contact, thereby creating a fairly steep initial density
gradient interface between them. Since the two miscible bulk uids subsequently
undergo mutual diffusion, any such interfacial tension must necessarily be transient
in duration.
The phenomenological form proposed by Korteweg [63] for his (traceless and
symmetric) stress tensor is
T
K
= I[gV
2
r a(Vr)
2
] b(Vr)(Vr) dVVr ; (8.1)
where the four independent Korteweg coefcients (a; b; g; d) (his symbols) appearing
above are, in the absence of an appropriate molecular theory, to be determined
experimentally, including their algebraic signs. It is interesting to note that the
constitutive form of the above stress is identical to the extra stress given by Eq. (4.6)
jointly with Eq. (1.6) (generalized to be applicable to both liquids and gases, as well
as to single- and two-component mixtures, the latter upon replacing a by D
v
).
Indeed, upon identifying this T

extra stress with the Kortewegs T


K
stress, and
upon supposing D
v
to be a constant [129], the phenomenological coefcients
appearing in Eq. (8.1) are found to be given by the following expressions:
a =
2
3
D
v
m
r
2
; b = 2
D
v
m
r
2
; g =
2
3
D
v
m
r
; d = 2
D
v
m
r
: (8.2)
Given Maxwells prior thermal stress analysis as set forth in Eq. (5.3) and as
later amended by Burnett [cf. Eq. (4.5)], together with the fact that these thermal
stresses are, in reality, due to density gradients as discussed in Section 6, it would
appear in the case of single-species uids that the MaxwellBurnett thermal stress
and the Korteweg density-gradient stress have much in common. Although the
thermal stress analysis of Sections 4 and 5 was limited to gases, Kortewegs stress
appears to be bear no such restriction. As such, it is presumed by its users
[27,126128] to be applicable to liquids. In any event, both types of stresses represent
the extra stress, T

= 2mVj
v
; above and beyond the classical Newtonian viscous
stress tensor, T
m
= 2mVv
m
: It is interesting to note that despite their common
density-gradient ancestry, the Korteweg and MaxwellBurnett stresses are regarded
in the literature as being continuum and noncontinuum stresses, respectively,
without any rational reason being advanced for this distinction. Indeed, according
to the thermal stress arguments advanced in Section 4, such a distinction has no
basis in fact.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 102
8.2. Body-force-driven vs. body-force-free phoretic particle motions
The modied NavierStokes equation (2.1) (1.1) (1.4) proposed here in place
of its traditional formulation is
r
D
m
v
m
Dt
= Vp mV
2
v
v
; (8.3)
at least in circumstances where, for simplicity, the viscosity is taken to be a constant
and where the quasi-compressibility condition,
V
.
v
v
= 0 ; (8.4)
is further supposed applicable [130]. Additionally, the proposed no-slip boundary
condition to be used in the solution of transport problems is taken to be (5.1). In
arguing in favor of Eq. (8.3), we have restricted attention exclusively to
circumstances in which f = 0; and in which there is no externally imposed
undisturbed pressure gradient, Vp
(0)
= 0 (which would otherwise have resulted in
a forced-convective motion v
(0)
m
of the uid). As such, the body-force-free phoretic
uid motion which ensues is necessarily animated exclusively by massdensity
gradients, Vr: In this context it is crucial to note that our main motivation in
proposing (8.3) was not to explain thermophoretic [23] or diffusiophoretic [25]
motion; rather, it was just the opposite, namely to use knowledge of these phoretic
phenomena to demonstrate that our hypothesized modication (8.3) of the
NavierStokes equation as well as of the no-slip boundary condition, Eq. (5.1),
were consistent with these experimental data.
Acceptance of the modied theory as a working hypothesis enables one to make
predictions regarding phoretic and related phenomena (such as thermal diffusion
[48]) that is applicable to liquids, despite the fact that for all practical purposes only
gaseous experimental data were used in demonstrating the consistency of the theory.
We emphasize this fact because there does exist a body of liquid-phase thermal
diffusion data, including a comprehensive theory thereof due to Semenov and
Schimpf [47], briey discussed at the end of Section 3, and further discussed below,
that attempts to rationalize at least some of these experimental data [131]. Yet, as
will be seen, their liquid-phase thermophoresis/thermal diffusion theory does not
bear directly upon the validity of our modied uid-mechanical equations (8.3)
(5.1) because their work refers to thermophoretic particle motion animated by the
action of body forces (albeit differing sensibly from zero only in the immediate
neighborhood of the particles), whereas our thermophoretic theory is based upon
circumstances in which such body forces are everywhere precluded, including any
boundary-layer region proximate to the particle.
As a prelude to illustrating the fundamental differences existing between these
respective body-force-free and body-force-driven modes of thermophoretic motion,
below we contrast our analysis of body-force-free phoretic phenomena [2325] with
that of electrophoretic particle movement in ionic solutions, the latter driven by
electrical body forces. At the same time this electrophoretic focus enables us to draw
a sharp distinction between the purely mathematical Smoluchowski-type v
m
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 103
velocity jump condition existing across the hypothetical Debye ionic boundary
layer [cf. Eq. (8.7)], and its Maxwell-type v
m
slip counterpart (5.2) which
represents a true mass slip at the physical surface of the solid rather than a
mathematical velocity-jump across a thin boundary layer of the type associated with
the parameter G in Eq. (5.4).
8.2.1. Electrophoretic particle motion in incompressible liquids [132 134]
In contrast with Eqs. (8.3)(8.4), whose departure from the traditional
formulations thereof is based upon uid compressibility, explicitly involving the
fact that V
.
v
m
a0; the classical NavierStokes/continuity equation formulation for
incompressible uids takes the general form
r
D
m
v
m
Dt
= Vp mV
2
v
m
f ; (8.5)
together with
V
.
v
m
= 0 : (8.6)
In this incompressible, nonzero body-force context, consider the application of Eqs.
(8.5) and (8.6) to explain the electrophoretically animated movement with velocity U
of a solid particle through an unbounded uid in the limiting case of a uniform
electric eld E
o
at innity, explicitly for the case of an electrically nonconducting
particle possessing a uniform surface charge density characterized by the zeta
potential B; and in the Debye thin double-layer limit. The body-force density in such
problems is given generally by the expression f = r
e
E; where the electric charge
density r
e
and electric eld E are, respectively, given constitutively as r
e
= V
2
f
and E = Vf; with the dielectric permitivity of the uid and f the electric
potential. Outside of the double layer the electric eld is effectively uniform at its
value E
o
; so that f f
(0)
= E
o
.
x as [x[ o; whereas the potential satises
Laplaces equation, V
2
f = 0: The potential differs sensibly from its far-eld value
f
(0)
only in the immediate neighborhood of the particle surface qV
s
; where it satises
the boundary condition n
.
Vf = 0 appropriate to a nonconducting body. Moreover,
the HelmholtzSmoluchowski slip boundary condition,
v
m
U = MVf on qV
s
; (8.7)
with M = B=m the so-called electrophoretic mobility, reects the jump in the mass
velocity v
m
across the Debye boundary layer,
Upon neglecting inertial effects in (8.5), the joint solution of this system of
equations and boundary conditions for f and v
m
; implicitly satisfying the no-
penetration condition (1.8), leads eventually to the following expression for the
electrophoretic velocity of the body:
U = ME
o

B
m
Vf
(0)
; (8.8)
a result which holds independently of both the particles size and shape [132, p. 476,
133]. As discussed, for example, by Levich [132], the Smoluchowski massvelocity
jump boundary condition (8.7), substitutes mathematically for the electrical body
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 104
forces acting within the Debye layer in the immediate neighborhood of the particle,
being equivalent thereto in regard to its physical consequences.
We have elaborated on this classical electrophoretic problem because, super-
cially, it might otherwise appear to have much in common with our prior
thermophoretic [23] and diffusiophoretic [25] analyses of particle motion under the
inuence of mass density gradients. However, in contrast with the latter class of
phoretic motions leading to (8.3), such electrophoretic particle motion is, in fact,
ultimately caused by the action of electrical body forces, as embodied in f; rather
than by true massvelocity slip; that is, the apparent slip, being purely mathematical
in origin, is simply an artifact, rather than the root cause, of the particles
electrophoretic movement! Moreover, and perhaps most importantly, in contrast
with our nonconducting body-force-free [23,25] phoretic problems, electrophoretic
particle movement is not solely dependent upon the physical properties of the uid
(as embodied in and m); rather, it also depends upon the particles surface zeta
potential B; which is a joint uid solid property. Additionally, during
electrophoretic motion, the particle moves through the uid, rather than being
entrained in the already moving uid, as evidenced, inter alia, by the appearance of
the uids viscosity in the denominator of Eq. (8.8). Because of these fundamental
differences, body-force-animated electrophoretic motion is very different in its
physical origins from that of its body-force-free thermophoretic and diffusiophoretic
counterparts, which led to our proposed modication (8.3) of the NavierStokes
equation.
Despite these fundamental differences there nevertheless exists a striking analogy
between electrophoresis and thermophoresis, as discussed in Appendix E. When
exploited, this analogy enables many of the detailed electrophoretic solutions extant
in the literature to be directly and immediately applied to the resolution of their
thermophoretic boundary-value counterparts.
8.3. Thermophoretic particle motion in liquids. The work of Semenov and Schimpf
[47,135,136]
This subsection resumes the abbreviated discussion of the work of these authors
offered at the end of Section 3. Their model [135] is based upon the following general
arguments: When the temperature of a liquid (the solvent) varies along the surface of
a solid body, the number-density of solvent molecules will, owing to the dependence
of the solvent density upon temperature, be larger in the colder regions of the surface
than in its warmer regions. This gradient in molecular solvent number density creates
an osmotic pressure gradient along the surface, in turn giving rise to an interfacial
stress at the particle surface tending to balance this pressure gradient. These osmotic
forces are regarded as being short-range body forces f appearing in the traditional
incompressible creeping ow equations emanating from Eqs. (8.5) and (8.6), with
such forces being sensibly different from zero only in the immediate neighborhood of
the particle surface. There, similar to body-force-driven theories [3032] of phoretic
motions deriving from inhomogeneous surface adsorption phenomena, such forces
translate mathematically into an effective slip (strictly, velocity jump) of the
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 105
uids mass velocity v
m
near the surface of the particle (more precisely at a point
lying just outside of the adsorption boundary layer).
In physical terms, the situation is not unlike that resulting from the absorption of
surfactant molecules onto a surface, coupled with the fact that any variation in the
concentration of these surfactant molecules along the surface will produce an
interfacial stress [3032] acting in the tangent plane of the surface. Here, however,
the molecules of the solvent play the role of the surfactant molecules, with the
surface temperature gradient serving to produce solvent number-density concentra-
tion gradients along the surface. The precise details governing the adsorption of
these solvent molecules onto the surface depends upon the physicochemical
adsorption forces existing between the solvent molecules and those of the solid.
These intermolecular forces are quantied in the SemeneovSchimpf analysis by
Londonvan der Waals dipoledipole intrermolecular forces. Explicitly, the
following formula is presented for their thermophoretic particle velocity U [135]
[which we have re-written in a form deliberately reminiscent of Eq. (3.1)]:
U =
a
/
b
1 (k
s
=2k)
VT; a
/
=
ln 3
8
Ar
2
0
mv
0
: (8.9)
In the above, A is the Hamaker constant dening the interaction between the particle
and the uid, m is the uids viscosity, r
0
is the radius of a solvent molecule and v
0
is
solvent volume per molecule, i.e., v
0
= ^ vm=N; where ^ v = 1=r and m are, respectively,
the solvents specic volume and molecular mass, and N is Avogadros number. For
the liquids with which they are concerned the authors suggest as an approximation
that the molecular radius be estimated from the equation v
0
= 8r
3
0
; the latter valid for
an elementary cubic crystal packing model. They also suggest use of the
approximation A =

A
p
A
s
_
; where A
p
and A
s
are the respective Hamaker constants
for the particle and solvent. With these approximations, one would have that a
/
=
(ln 3)

A
p
A
s
_
=64mr
0
:
Despite their very different origins, Eqs. (8.9) and (3.1) obviously have much in
common including the appearance therein of the uids thermal expansivity b; and
the fact that the U is independent of the size of the thermophoretically animated
particle. However, we note the following major differences in the two models: (i) in
contrast with the passive role played by the particle in our analysis, at least in the
nonconducting case, the physicochemical properties of the particle play a critical role
in Eq. (8.9) as manifested by the presence therein of the particles Hamaker constant;
(ii) as evidenced by the appearance of the viscosity in the denominator of (8.9), the
particle moves through the uid in their model, whereas in our case it moves with the
uid. Of course there is also the fundamental issue of the presence and absence of
body forces as the animating uid-mechanical mechanism in the two models.
The work of these authors impacts upon the fundamental notion of the tracer
velocity of a uid, as discussed in Section 6. According to their theory, even in
nonconducting circumstances, where k
s
=k51; the particle will not be passive, in
the sense that its physicochemical properties enter into the particle velocity U in the
guise of the particles Hamaker constant. As such, the particle would not qualify in
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 106
the role of a tracer in the case of nonisothermal uids. More general versions of their
theory [47] suggest that this physicochemical dependence would, in the case of non-
spherical particles, also cause the particle velocity U to depend upon its orientation
relative to the direction of the imposed temperature gradient, further comprising its
role as tracer.
A variant of Eq. (8.9) [47] has been applied with some success to explain
LudwigSoret thermal diffusion effects in solutions of polymers dispersed in
nonpolar solvents, with the long-chain macromolecule viewed as a thermophoretic
particle. In creating their model these authors use a priori knowledge of the fact
that the LudwigSoret thermal diffusivities of polymer molecules are known to be
independent of chain length as well as of the presence of side branches [137]. By way
of comparison, our own model [48] of polymer thermal diffusion derived from (3.1)
demonstrates this independence, rather than assuming it a priori.
While it may appear that a conict exists between Eqs. (3.1) and (8.9) for the
thermophoretic velocity of a spherical particle, this is not necessarily the case.
Rather, it is possible that the two effects may be additive, with Eq. (3.1) referring to
that portion of the total particle velocity constituting motion with the uid, whereby
the particle is effectively entrained (except, perhaps, for the k
s
=k thermal
conductivity factor) in a uid which is already moving in the absence of
physicochemical body forces, and with Eq. (8.9) referring to that portion of the
particles motion taking place through the uid, stemming from the action of local
physicochemical body forces causing velocity slip. To explore this possibility, we
add the two equations in question to obtain
U = (1 )
ab
1 (k
s
=2k)
VT ; (8.10)
=
ln 3
8

A
p
A
s
_
r
2
0
mav
0
(8.11)
in which is a dimensionless parameter (not necessarily small) representing the
SemenovSchimpf correction to Eq. (3.1).
8.3.1. Liquids
Semenov and Schimpf provide data enabling to be calculated in the case of silica
particles in various liquid solvents. Notationally, with b
T
their proportionality
coefcient dened by the formula U = b
T
VT; Eq. (8.11) is equivalent to
= 1
k
s
2k
_ _
b
T
ab
: (8.12)
Explicitly, Semenov [135] summarizes estimated and experimentally available data
for the parameters b
T
; k
s
; k; b appearing above. Thermometric diffusivity data were
taken from Photothermal Spectroscopy Methods for Chemical Analysis, Wiley, 1996.
These data, together with the estimated value of obtained therefrom, are tabulated
below in Table 1.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 107
These data suggest that for liquids the SemenovSchimpf contribution provides a
relatively modest modication of the entrainment model (3.1), a quantitative
conclusion subject, however, to the caveat implicit in the uncertainty surrounding
use of the approximation A -

A
p
A
s
_
for the Hamaker constant. According to these
authors, their result alone, independently of the further contribution of (3.1),
appears to accord with experimental data for silica, at least as regards its order-of-
magnitude. However, it needs to be emphasized that the data on which this
conclusion is based are neither single-particle data, nor were the particles studied
sufciently large to exclude the possibility of Brownian motion. And, when both
factors are simultaneously present, one is no longer dealing with thermophoresis as a
phenomenon in its own right; rather, one is more likely dealing with Soret-effect
thermal diffusion phenomena, to which our single-particle, non-Brownian theory is
not directly applicable in the absence of further theoretical considerations [48]. In
any event, despite such reservations and caveats, Eq. (3.1) cannot be ruled out as
applicable to thermophoresis in liquids.
8.3.2. Gases
While Semenov and Schimpf make no claims that their analysis should apply
to gaseous continua, there appears to exist no reason in principle while gases
should be excluded from the domain of applicability of Eq. (8.9). Certainly, their
underlying notion of uid-generated osmotic pressure as constituting the driving
force for thermophoretic motion in uid continua would appear as applicable to
gases as it is to liquids. However, upon attempting to apply their equation to
gases, it becomes evident that major theoretical and experimental conicts exist
with Eq. (3.5) [in which C
s
= O(1)], the latter viewed as representing a summary of
experimental data for thermophoresis in gaseous continua. Even apart from
issues of differences in orders of magnitude, the experimental data support a model
in which the particles thermophoretic velocity increases with increasing gas viscosity
m and decreasing density r; whereas Eq. (8.9) would appear to predict exactly the
opposite trends. Moreover, whereas the experimental data (3.5) suggest that the
physicochemical properties of the particle should not be relevant to its thermo-
phoretic velocity (apart from particles thermal conductivity k
s
), the presence of
the particles Hamaker constant A
p
in the SemenovSchimpf formula suggests
otherwise.
ARTICLE IN PRESS
Table 1
SemenovSchimpf correction to Eq. (3.1) for the case of silica particles in various solvents
Solvent
b
T
(10
7
cm
2
s
1
K
1
) b(10
4
K
1
)
k
s
=k
a = k=r^ c
p
(10
3
cm
2
s
1
)

Water 0.19 2.1 2.4 1.43 0.14
Tetrahydrofuran 1.39 11 4.4 0.824 0.49
Acetonitrile 2.4 14 3.6 1.08 0.44
Cyclohexane 0.70 12 3.8 0.858 0.20
H. Brenner / Physica A 349 (2005) 60132 108
8.3.3. The need for experimental liquid-phase data
The preceding discussion reinforces the view expressed in Section 3 of the critical
need for single-component thermal gradient data in liquids. In subjecting these or
any other theories to test, such data need not be limited to thermophoretic of thermal
diffusion data. All such theories involve the notion of slip of one kind or another at
the boundary between a uid and a solid, whether such slip be of the Maxwell
thermal-creep or SemenovSchimpf velocity-jump types. Steady-state thermal
transpiration experiments [24] offer an interesting alternative to thermophoresis in
this respect, since the presence of slip is manifested macroscopically as a pressure
difference across a capillary tube or across a porous medium, whereas no such
thermolecular pressure difference would arise in the absence of slip. Experiments
of this nature are currently underway at Sandia National Laboratories under the
direction of Lisa A. Mondy.
8.4. Potential applications of the general theory
From a practical point of view, pragmatists could argue that our proposed generic
modications of the original NSF (or NSFF) equation set constitute but a
small correction to those equations in situations where density gradients exist. In
reference to the question of the general viability of this viewpoint, it is useful to rst
establish the circumstances under which the correction will be small. In particular,
the corrections do indeed prove to be small in those applications for which the ratio
:= [j
v
[= max [v
m
; v
v
[(0pp1) satises the inequality 51: Thus, as a practical
matter, for small density gradients it will often be true that [v
m
[ - [v
v
[; in which
situations the disparity between the mass- and volume- or tracer-velocities, v
v
= v
l
;
will indeed be small. Conversely, when = O(1) such corrections will no longer be
small, as occurs, for example, during thermophoretic and diffusiophoretic uid
motions. Other examples come readily come to mind, such as steady-state
thermal diffusion [48], where temperature- and species-gradients coexist. Closely
related to the zero massvelocity condition v
m
= 0 encountered during thermo-
phoretic motions, is the fact that in a number of applications, such as thermal
transpiration, the mass velocity does not vanish locally at each point of the uid, but
rather vanishes only in an integral sense,
_
A
v
m
; dA = 0; with A typically a cross-
sectional area. Obviously, in such circumstances the small-correction inequality
criterion 51 will again be violated, albeit this time in a global rather local pointwise
sense. In any event, it is obvious that effects stemming from the existence of a
diffusive volume ux, rather than always representing minor corrections to
conventional results, may, in fact, be relatively largeindeed, sometimes represent-
ing the dominant effect, as in the examples cited above. Microuidic applications,
involving very slow mass motions possibly driven by temperature gradients, come to
mind in this context.
Apart from practical engineering-type application of the types discussed above,
several other areas of potential interest come to mind. These include use of the
present continuum theory to rationally re-organize the hierarchical ordering of the
Burnett-type noncontinuum terms that arise when solving the Boltzmann equation
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 109
perturbatively at higher order [6,7,57,64,138,139] in the Knudsen or Mach
numbers. More rational theories of mixtures [140], involving elucidation of the
constitutive and transport properties of so-called interpenetrating continua,
appears to be another area ripe for exploitation. Transport phenomena in porous
media [141] also offer interesting challenges owing to the fact that the v
v
-based Eqs.
(2.13) and (2.14) together with (1.7) (1.4) are identical to their classical v
m
-based
creeping ow counterparts, often used in deriving Darcys law governing ow in
porous media.
8.5. Summary
The main results of the present work lie in our proposed modication (8.3) of the
NavierStokes equation and no-slip boundary condition (5.1) for the case of
compressible uid continua, where the mass density r is not uniform. In turn, these
lead to the prediction of a Korteweg-like stresses (8.1) in both liquids and gases
which, in turn, cause the phoretic entrainment of isolated particles toward
regions of high uid density, such that the velocity of the particle through space
proves to be independent of its size, shape, and physicochemical properties in
circumstances where the particle is passive. It is possible in a biological context
that such density-gradient-induced locomotion may play a role in chemotaxsis
or thermotaxis.
Note added in proof
Reconciliation of Maxwells slip coefcient C
s
= 3=4 for monatomic (Maxwellian)
molecules with our value of C
/
s
= 3=2
Apart from fundamental philosophical continuum vs noncontinuum differences
for gases, the only quantitative difference between the predictions of Maxwells
theory and our theory arises from the purely numerical factor of two stemming from
the different values ascribed to the O(1) Maxwell slip coefcient C
s
appearing in Eq.
(5.2). Whereas, at least in the case of monatomic gases, Maxwell [8] arrives
theoretically at the value C
s
= 3=4 (for Maxwell molecules) on the basis of molecular
arguments as per the discussion of section 5, our purely continuum theory gives
C
s
= 3=2 as set forth in Eqs. (5.6) (3.3) (although without restriction to Maxwell
molecules). In the case of monatomic molecules this difference can, however, be
reconciled, as discussed in what follows below.
Because the equations of gas dynamics are not valid in the Knudsen layer (whose
thickness is of the order of the mean-free path of the gas molecules) proximate to a
solid surface, a rigorous derivation of the continuum boundary conditions to be
applied at the solid-gas boundary involves the solution of the Boltzmann equation
inside of the Knudsen layer, and the subsequent matching of this inner solution with
the outer solution of the hydrodynamic equations outside of the Knudsen layer [139].
The boundary condition imposed on the inner problem adopts the form of a specied
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 110
law of molecular reection assumed to exist at the wall. The values of the outer elds
when extrapolated to the wall then provide the macroscopic boundary conditions
which, when enforced, provide the correct solution of the continuum hydrodynamic
equations outside of the Knudsen layer. The constitutive form thereby obtained via
this matching scheme for the boundary condition to be imposed upon the relative
tangential mass-velocity component v
m
accords with that of Maxwells Eq. (5.2),
modulo the explicit numerical value of the C
s
slip coefcient appearing therein,
which depends on the distribution of gas molecules reected from the wall
and, hence, upon the specic physicochemical natures of both the gas and solid [139,
p. 367].
Various derivations of C
s
exist based upon specic models of the gas and wall,
together with use of either the Boltzmann equation, the method of moments, or a
Monte Carlo scheme, as recently summarized by Sharipov and Kalempa [142].
Explicitly, Sharipov points out in Ref. [143] that: According to Refs. [144,145] the
thermal slip coefcient s
T
[Maxwells C
s
] varies from 0.75 up to 1.5, where the rst
value corresponds to the specular reection of molecules on [the] surface, while the
second value corresponds to the opposite situation, i.e. back reection. (With regard
to these values, Sharipov [143] is presumably referring here solely to the case of
monatomic gases.) Of course, the 3=4 value corresponds to Maxwells original 1879
model [8] based upon his implicit choice of an accommodation coefcient. On the
other hand, the 3=2 value accords with our result for the slip coefcient, derived from
the assumption that there is no slip of the volume velocity at solid surfaces (together
with use of specic heat data for monatomic gases). As such, acceptance of our
theory would imply that the volume velocity-based no-slip macroscopic boundary
condition is equivalent, at least in the case of monatomic gases, to back reection of
the gas molecules at the surface.
Acknowledgements
I am grateful to Dr. James R. Bielenberg of Los Alamos National Laboratories,
formerly a graduate student in the Chemical Engineering Department at MIT. He
shared with me the pleasure of performing the exact phoretic and thermodynamic
calculations, cited throughout this paper, permitting a comparison between
experiment and theoretical productions, and resulting in conrmation of the
nontraditional constitutive equation (1.4). I am also pleased to acknowledge many
useful hours engaged in pertinent conversations with Dr. Ehud (Udi) Yariv
of the Mechanical Engineering faculty at the Technion, formerly a postdoctoral
fellow in the Chemical Engineering Department at MIT. Equally enlightening
conversations were held with Aruna Mohan, currently a graduate student in
Chemical Engineering at MIT. She was instrumental in helping me to formulate the
proof offered in Appendix A. Finally, I am also grateful to Dr. Sangtae Kim of
Purdue University, formerly of Eli Lilly and Company, whose encouragement was
helpful in arranging nancial support through Lilly for the research embodied in the
present study.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 111
Appendix A. Invariance of the basic transport equations under change of reference
frame
From a constitutive viewpoint the respective forms we have adopted for the basic
mass, momentum, and energy transport equations, namely Eqs. (1.2), (2.1) and (2.2),
follow, inter alia, from the requirement that they remain invariant under translation
of the reference frame in which they are written. That they indeed possess this
property is formally demonstrated below.
In order to keep an open mind on the subject, we begin with the following pre-
constitutive forms of this trio of transport equations. By pre we refer not only to
the usual notion of constitutive equations for the various diffusive (i.e., molecular)
uxes such as Newtons law of viscosity or Fouriers law of heat conduction; rather,
the following equations are equally pre-constitutive with regard to the nonmolecular
constitutive expressions introduced for the specic momentum density ^ m appearing
below in the momentum equation, as well as for the kinetic and work velocities,
v
k
and v
w
; respectively, appearing below in the energy equation. For simplicity we
will suppose that each of the following transport equations are originally written, as
below, in an inertial reference frame, say x:
(i) Transport of mass : qr=qt V
.
(rv
m
) = 0 ; (A.1)
(ii) Transport of momentum : r
D
m
^ m
Dt
= V
.
P r
^
f ; (A.2)
(ii) Transport of energy : r
D
m
^ e
Dt
= V
.
j
e
V
.
(P
.
v
w
) : (A.3)
In these equations,
^
f is the specic body force density, assumed (for the time being)
to be conservative and hence expressed as the gradient of a time-independent
potential energy function
^
f(x); such that
^
f = V
^
f in which q
^
f=qt = 0: Additionally,
^ e = ^ u v
2
k
=2
^
f is the specic total energy; moreover, j
e
is the diffuse total energy
current which we suppose henceforth to be identical to the diffuse internal energy
current, j
u
on the assumption that neither kinetic nor potential energy can be
transported diffusively [146].
A.1. Coordinate transformations
Consider two systems of coordinates in space. One of these systems is an
inertial coordinate system x; which we regard as being at rest (i.e., xed in
space). The second system, say x
/
; moves relative to the former with a (generally
time-dependent) velocity U(t): Let the two coordinate systems coincide at time
t = 0: Then,
x x
/
=
_
t
0
U(t) dt : (A.4)
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 112
The respective velocities v and v
/
of an object as measured by the two observers are
obviously related by the expression
v v
/
= U(t) : (A.5)
With regard to the respective material derivatives in the two reference frames,
D
m
Dt
=
q
qt
_
x
v
m
.
V and
D
/
m
Dt
=
q
qt
_
x
/
v
/
m
.
V
/
;
in which V = q=qx)
t
and V
/
= q=qx
/
)
t
it is readily established since (qx=qx
/
)
t
= 0 and
(qx=qt)
x
/ = U(t) that
V
/
= V and
D
/
m
Dt

D
m
Dt
: (A.6)
On the constitutive supposition that all diffusive uxes and other equilibrium
physical quantities such as p and r remain invariant under the coordinate
transformation (so that, for example, P
/
= P), and since ^ m ^ m
/
= U(t); one nds
from the linear momentum equation (A.2) together with (A.6
2
) that
^
f
/
=
^
f
dU(t)
dt
: (A.7)
A.2. Energy equation
Consider the energy equation (A.3) irrespective of any constitutive choices made
for the various quantities appearing therein. Together with use of the linear
momentum equation (A.2) and the denition of the conservative force, in
conjunction with several of the above relations one obtains the following expression
upon rearrangement:
r
D
m
^ u
Dt
= V
.
j
u
r(v
m
v
w
)
.
^
f rv
w
.
D
m
^ m
Dt
r
D
m
Dt
1
2
v
2
k
_ _
P
T
: Vv
w
:
(A.8)
As an aside, while we have derived the term rv
m
.
^
f appearing above from time-
independent potential energy considerations, this same expression could have been
obtained more generally simply as a rate of working term (per unit volume), rv
m
.
^
f;
added to the right-hand side of Eq. (A.3) while excluding the potential, energy
contribution from its left-hand side. That the velocity by which
^
f is multiplied is v
m
rather than say v
w
(or even some other velocity) is then justied by reference to the
specic case where the corresponding work term is derived via potential energy
considerations of the type having led to Eq. (A.8).
Our goal in what follows is aimed at demonstrating on purely theoretical grounds
that the following equalities necessarily hold:
v
k
= v
w
= ^ m v
m
: (A.9)
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 113
We begin by examining the invariance of the internal energy density ^ u under the
coordinate transformation between the inertial and noninertial (i.e., accelerating)
frames. Explicitly, we write Eq. (A.8) in the primed (accelerating) system of
coordinates x
/
; thereby obtaining
r
D
/
m
^ u
Dt
= V
/
.
j
u
r(v
/
m
v
/
w
)
.
^
f
/
rv
/
w
.
D
/
m
^ m
/
Dt

1
2
r
D
/
m
Dt
(v
/
k
2
) P
T
: V
/
v
/
w
:
(A.10)
We have afxed primes to
^
f as well as to each of the velocities, v = (v
k
; v
m
; ^ m; v
w
);
appearing in the above. On the other hand we have suppressed the corresponding
addition of primes to the remaining quantities appearing in (A.10), namely
(r; ^ u; j
u
; P); based on the recognition that their special physical natures require
them to remain invariant under the coordinate transformation. Insert into (A.10)
the various transformations relating the primed variables to the unprimed ones,
including the fact that, in general, V
/
v
/
= Vv; and subtract the resulting expression
from (A.8). Upon rearrangement and division by r this yields the relation
(v
m
v
k
2v
w
)
.
dU
dt
U
.
D
m
Dt
(v
k
^ m) = 0 :
Inasmuchas U and dU=dt can each be chosen independently at any given instant of
time, such arbitrariness necessitates that
v
m
v
k
2v
w
= 0 and v
k
= ^ m : (A.11)
This is as far as one can go via internal energy invariance arguments alone.
In order to complete the theoretical proof of Eq. (A.9) we further suppose
that the internal energy transport equation (A.8) must not contain any kinetic
energy terms. This requires that v
w
.
D
m
^ m=Dt = D
m
=Dt(v
2
k
=2): Into this equa-
tion substitute Eq. (A.11
2
) so as to eliminate ^ m; and rearrange the resulting
expression to obtain (v
w
v
k
)
.
D
m
v
k
=Dt = 0: Equivalently, (v
w
v
k
)
.
D
m
^ m=Dt = 0:
Hence, with use of (A.2) we obtain (v
w
v
k
)
.
(V
.
P r
^
f) = 0: Since this latter
relation has to hold independently of the constitutive equations for either P
and
^
f; this clearly requires that v
w
= v
k
: Introduction of the latter into (A.11)
yields the further relation, v
m
= v
k
: Cumulatively, this completes the proof of
Eq. (A.9). Finally, elimination of ^ m; v
k
and v
w
from Eqs. (A.2) and (A.3) in
favor of v
m
furnishes the basic momentum and energy transport equations (2.1)
and (2.2).
Clearly, the pre-constitutive equations (2.1) and (2.2) hold independently of any of
the constitutive relations employed for the elds (r; ^ u; j
u
; p; T) and j
v
: Indeed, in the
context of gas-kinetic theory, Eqs. (2.1) and (2.2) apply even in circumstances where
the constitutive equations for the physical quantities appearing in the group cited
in the preceding sentence, such as T; include noncontinuum terms, as in the case
of (4.3).
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 114
Appendix B. Elementary calculation of the thermophoretic velocity of a nonconducting
sphere
Consider a velocity eld v satisfying the equations Vp mV
2
v = 0 and V
.
v = 0;
in which v is the eld appearing in the deviatoric stress tensor in Eq. (1.1). According
to Faxens theorem [85] for such incompressible creeping ows satisfying a zero-
vector velocity boundary condition, v = 0 on qV
s
; the hydrodynamic force exerted
by the uid on a solid sphere qV
s
of radius a translating with velocity U when
immersed in an undisturbed creeping ow, say {v
(0)
; p
(0)
]; from which the sphere is
absent, is given by the expression
F = 6pma (v
(0)
U)
a
2
6m
Vp
(0)
_ _
0
: (B.1)
The subscript zero appearing in the above connotes evaluation of the undisturbed
velocity and pressure elds at the center of the uid space presently occupied by the
(center of the) sphere. Accordingly, a force-free sphere will, in the absence of wall
effects, move quasistatically with a velocity
U = v
(0)
0

a
2
6m
(Vp
(0)
)
0
: (B.2)
While, in the past, Faxens law has only been applied to the case where v refers to the
usual mass velocity eld v
m
; from a purely mathematical view Faxens law may be
equally well applied to the volume velocity eld v
v
; since the latter satises the
volumevelocity-based creeping ow equations (2.13) and (2.14) and the vector
velocity boundary condition that v
v
U = 0 on qV
s
: As discussed in Ref. [28], the
latter condition is applicable only to the case where the sphere is nonconducting.
To determine the undisturbed volume-velocity elds {v
(0)
v
; p
(0)
] existing in the
absence of the sphere, we note that since v
(0)
m
= 0; it follows from Eq. (1.5) that
v
(0)
v
= j
(0)
v
: However, from Eq. (1.6) we nd upon rearrangement that, as in Section 3,
j
(0)
v
= abVT = const: so that
v
(0)
v
= abVT = const : (B.3)
This undisturbed, pure uid, sphere-free, volume-velocity eld obviously satises Eq.
(2.13) since V
2
T = 0; while from Eq. (2.14) we nd that Vp
(0)
= 0: Thus, Eq. (B.2)
becomes
U = abVT ; (B.4)
in exact agreement with the more detailed result cited in Eq. (3.1) for the
nonconducting-sphere case, k
s
=k = 0:
B.1. Nonspherical particles
While we have demonstrated that Eq. (B.4) applies to the case of nonconducting
spherical particles, it is equally applicable to nonconducting particles of arbitrary
shape and orientation. This follows from the fact that the generalization [82] of
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 115
Eq. (B.1) for an arbitrarily shaped particle is F = M
1
.
[(v
(0)
U) O(a=L)]
0
; where
M is the particles mobility dyadic, a is a characteristic particle size, and L is a
characteristic length appearing in the normalization, V
+
= LV; of the dimensionless
gradient operator appearing explicitly in the undisturbed nonuniform ow v
(0)
; so
that the O(a=L) term represents a wall effect. Accordingly, the velocity of such a
force-free body is U = v
(0)
0
O(a=L): With use of Eq. (B.3), and in the absence of
wall effects, one thus recovers Eq. (B.4). Thus, remarkably (a fact more formally
demonstrated in Ref. [23]), irrespective of size, shape, and (in the latter nonspherical
case) orientation relative to the undisturbed temperature gradient VT; all
nonconducting particles will move at the same velocity. Accordingly, provided that
one interprets uid motion physically as being the (undisturbed) uids volume
velocity rather than its mass velocity, Eq. (B.4) expressed more generally as
U = (v
(0)
v
)
0
O(a=L) (B.5)
simply states that any passive (i.e., nonconducting) no-slip particle is simply
entrained in the owing uid. Alternatively, with use of (1.5) this may be written in
the form
U (v
(0)
m
)
0
= (j
(0)
v
)
0
O(a=L) ; (B.6)
a result which holds even when v
(0)
m
a0: The relative motion represented by the left-
hand side of the above then constitiutes the phoretic velocity of the insulated
particle, as in Eq. (6.4) with k
s
=k = 0; or in Eq. (7.1).
Appendix C. Fortuitously-correct thermal solutions based upon Maxwell slip
The energy equation (2.9) pre-constitutively incorporates both possible choices for
v; namely Eqs. (1.3) and (1.4). In the case where v
m
is chosen, it can be shown that for
both the unsteady-state thermophoretic [23] and steady-state thermal transpiration
[24] cases that, to leading order in the appropriate small parameter, a[Vln T[51;
asymptotically characterizing the heat-transfer process, the energy equation reduces
to the form rv
m
.
VT = kV
2
T: This same equation is obtained using v = v
m
; albeit
only on the proviso that the uid is supposed incompressible, so that V
.
v
m
= 0
(see below), as it done in the literature for the classical thermophoretic [35] and
thermal transpiration [8] analyses. Now, in circumstances where a prescribed
temperature gradient, quantied by a characteristic magnitude [VT[; is the only
mechanism animating the uid motion v
m
(as in the thermophoretic and thermal
transpiration cases to be discussed), it is found in the leading-order linear
approximation that v
m
= O(a[Vln T[): As such, to this degree of approximation
the convective term, v
m
.
VT; in the energy equation is of O([VT[
2
); whereas in this
same linear approximation the diffusive term, kV
2
T; is of O([VT[): Consequently,
when [VT[ is small, convection may be neglected compared with conduction. In such
circumstances, the energy equation reduces simply to V
2
T = 0; which is applicable to
both cases, v = v
m
and v = v
v
: Similar arguments apply with respect to the
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 116
convective/inertial and diffusive/viscous terms in the momentum equation, so that,
only the viscous term remains relevant.
The following table shows the resulting energy, momentum, and mass transport
equations describing both the Maxwell-based v
m
scheme, Eq. (1.3), and the modied
v
v
scheme, Eq. (1.4).
In the Maxwell scheme the mass transport or continuity equation is taken to be
V
.
v
m
= 0 despite the fact that the uid density is not constant as a consequence of
the fact that the equation of state is of the form r = r(T; p): The scheme based upon
v
v
correctly takes the continuity equation to be V
.
(rv
m
) = 0: It is, however,
unnecessary to explicitly employ the latter equation in the calculations. Rather, as
the third modied equation we instead use the accurate volume transport equation
[cf. Eq. (2.13)], V
.
v
v
= 0; expressing the uids quasi-incompressibility [27].
The trio of Maxwell equations appearing in Table 2 is to be solved for v
m
using
Maxwells slip boundary condition (5.2), whereas the modied equations are to be
solved for v
v
; using the no-slip boundary condition (5.1). These two velocities are
related by Eq. (1.5), in which the required diffusional volume current j
v
; Eq. (1.6), is
given explicitly for the present heat-transfer case by Eq. (5.9). As such, as a
consequence of the fact that V
2
T = 0; it follows from the latter relation that V
.
j
v
=
0 and V
2
j
v
= 0: Accordingly, from Eq. (1.5) we nd that to the order of the present
approximation, V
.
v
m
= V
.
v
v
= 0 and V
2
v
m
= V
2
v
v
: Comparison of the latter
expression with that in Table 2, leads to the surprising conclusion that both the
Maxwell and modied schemes are each governed by exactly the same generic
equations, namely V
2
T = 0; V
.
v = 0; and Vp = mV
2
v: Nevertheless, their respective
solutions will not coincide, as each is to be solved subject to different boundary
conditions on qV
s
; namely I
s
.
(v
v
U) = 0 in the modied v
v
case and I
s
.
(v
m
U) =
C
s
vV
s
ln T in the Maxwell v
m
case, all other boundary conditions [such as the
impenetrability condition (1.8)] being the same for both. Note that despite the
velocities being different in the two cases, the respective pressure gradients Vp will
each be the same owing to the fact that V
2
v
m
= V
2
v
v
[147].
The calculation of some overall force F usually constitutes the main item of
interest in connection with the class of problems under discussion, either the force on
a thermophoretic particle [23] or the force on the walls of the capillary tube in the
thermal transpiration problem [24], each for the case of a prescribed temperature
difference at the two ends of the system conning the uid. The force on an
impermeable solid surface can be calculated from the expression F =
_
qV
s
dS
.
P;
with the pressure tensor given by Eq. (2.1
2
) in conjunction with (1.1). Since, as we
ARTICLE IN PRESS
Table 2
Energy, momentum, and continuity equations
Maxwell scheme, v = v
m
Modied scheme, v = v
v
Energy equation
V
2
T = 0 V
2
T = 0
Momentum equation
0 = Vp mV
2
v
m
0 = Vp mV
2
v
v
Continuity equation V
.
v
m
= 0 V
.
v
v
= 0
H. Brenner / Physica A 349 (2005) 60132 117
have indicated, the calculated pressure p will be the same for both the Maxwell and
modied schemes, one can form the difference, F
v
F
m
=
_
qV
s
dS
.
(T
v
T
m
);
between the respective forces. In view of Eq. (1.1) we nd upon using Eq. (1.5)
together with the relation V
.
j
v
= 0 that F
v
F
m
= 2m
_
qV
s
dS
.
Vj
v
: In conjunc-
tion with Eq. (5.9) considered jointly with the constancy of the multiplier of VT;
this yields
F
v
F
m
=
2mk
^ c
p
q^ v
qT
_ _
p
_
qV
s
dS
.
VVT :
The surface integral appearing in the latter relation can be converted into a volume
integral, and the relation V
2
T = 0 employed in the resulting expression to conclude
that F
v
= F
m
: This shows, at least for the present class of gaseous thermal problems,
that the Maxwell scheme does, in fact, furnish the same overall global results as the
modied scheme, despite the inconsistency of the set of Maxwell equations based
upon the constitutive relation, v = v
m
: However, such agreement, rather than
embodying some fundamental principle, is seen to be merely fortuitous, being limited
to a very special set of circumstances. More general circumstances, wherein, for
example, nonlinear terms are retained in the calculations, would almost certainly
point up a difference between the two. Note also that in this limited class of
problems, not only will the global forces be the same for both schemes, but so too
will be the respective local pressures p; densities r; temperatures T; and mass
velocities v
m
: As such, it is perhaps not surprising that the true, no-slip, nature of the
Maxwell slip condition has gone undiscovered for so long.
Appendix D. First-law violation occasioned by Maxwell slip in rigid, uid-lled,
immobile containers
Following up the closing remarks of Section 5, we demonstrate here that the
Maxwell slip condition (5.2) imposed upon v
m
violates the First law of
thermodynamics when used in conjunction with the following (single-component)
standard incompressible creeping ow equations used in the literature [8,35] to
analyze phoretic phenomena:
V
.
v
m
= 0 ; (D.1)
V
.
P
m
= 0; P
m
= Ip T
m
; (D.2)
T
m
= m[Vv
m
(Vv
m
)
T
] : (D.3)
According to the First law, the temporal rate at which the energy E (internal plus
kinetic) of a closed system increases is given by the expression dE=dt =
_
Q
_
W;
where
_
Q =
_
qV
s
dS
.
q is the rate of heat ow into the system and
_
W is the rate at
which the surroundings are doing work on the system. Here, qV
s
denotes the rigid,
solid, impermeable, and immobile boundaries of the apparatus conning the uid
within its interior.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 118
Based upon the above constitutive form (D:2
2
) for the pressure tensor P; the rate
of working is thus given in present circumstances by the expression
_
W =
_
qV
s
dS
.
P
m
.
v
m
(D.4)
for the case where the boundaries are immobile, so U = 0 in Eqs. (1.7) (1.3) and
(1.8). For deniteness we suppose that the container, within which some transient
heat transfer process is occurring, is suddenly insulated at time t = 0; so that
_
Q = 0
for all subsequent times t40: As no mechanism exists by which work can be
performed by the surroundings on the contents of the uid-lled container V
s
whose
immobile boundaries are qV
s
; this requires that
_
W = 0 (\t40) : (D.5)
This condition is consistent with the fact that the total energy E; internal plus
kinetic, of the isolated system will necessarily remain xed for all time follow-
ing placement of the insulation on the boundaries. As such, when the tran-
sients have decayed the uid will eventually attain a homogeneous equilibrium
state in which the pressure, temperature and density are each uniform throughout
the uid.
Under conventional non-Maxwellian no-slip conditions, one would have that
v
m
= 0 on qV
s
; whence Eq. (D.5) would be satised automatically irrespective
of the constitutive form of the stress tensor appearing in (D.4). However, in the
presence of initial temperature gradients imposed along qV
s
; the Maxwell slip
condition (5.2) generally obviates the possibility that v
m
= 0 on qV
s
; since the
integral (D.4) will not generally vanish owing to the generally nonzero tangential
mass velocity component, I
s
.
v
m
a0: As a consequence, in contrast with the classical
case (1.3) where v = v
m
; the velocity boundary condition alone no longer
automatically assures satisfaction of (D.5). It is possible, however, that the integral
may nevertheless vanish in some integral or other sense. To explore this possibility,
we convert (D.4) into a volume integral and use (D.2) in conjunction with (D.1)
to obtain
_
W =
_
V
s
2mVv
m
: Vv
m
dV : (D.6)
Obviously, the nonnegative nature of the above integrand at each point of the uid
precludes the possibility of Eq. (D.5) being satised. This demonstrates the
thermodynamic inconsistency of the set of mass-based velocity equations
(D.1)(D.3) when used in conjunction with the Maxwell slip condition (5.2). Viewed
alternatively, insistence upon the applicability of this equation set together with (5.2)
would violate the rst law of thermodynamics.
The escape from this dilemma involves replacing the set of equations (D.1)(D.3)
by our more physically appropriate set, namely with v
m
throughout replaced by v
v
[see Eqs. (2.13)(2.14) and (1.1), in which v = v
v
]. Thus, in place of (D.4), one would
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 119
now write
_
W =
_
qV
s
dS
.
P
v
.
v
m
; (D.7)
wherein
P
v
= Ip T
v
; T
v
= m[Vv
v
(Vv
v
)
T
] : (D.8)
It is conrmed in what follows that the state of motion of the conned uid is
correctly described by the fact that v
v
= 0 and p = const throughout the uid and
for all t40: This proposed solution, for which T
v
= 0 everywhere, clearly satises
Eqs. (2.14) and (2.13) as well as the no-slip boundary condition (5.1). Moreover, the
condition that the container be insulated requires that n
.
VT = 0 on qV
s
; from which
it follows from Eq. (5.9) that n
.
j
v
= 0 on qV
s
: In turn, from Eq. (1.5) this yields
n
.
v
v
= n
.
v
m
; thus assuring that the no-penetration condition (1.8) is satised. In
turn, it follows that dS
.
P
v
= 0 at each point on qV
s
: Satisfaction of Eq. (D.5) is thus
assured despite the slip of v
m
along qV
s
:
To complete the problem, it remains to demonstrate that the proposed solution,
v
v
= 0; is consistent with the continuity equation (1.2) and the energy equation
(2.11). In the course of the demonstration, we use these equations to determine the
remaining elds: r; T and v
m
: Consider rst the density eld. Substitution of (1.5)
and (1.6) together with use of v
v
= 0 yields
v
m
=
k
^ c
p
r
2
Vr : (D.9)
Introduction of the latter into the continuity equation (1.2) furnishes the following
equation governing the density eld r(x; t):
qr
qt
=
k
^ c
p
V
2
ln r : (D.10)
The insulation boundary condition n
.
VT = 0 on qV
s
together with the uids
equation of state furnishes the boundary condition n
.
Vr = 0 on qV
s
imposed upon
r: This same condition arises from the impermeability of the boundary, n
.
v
m
= 0 on
qV
s
; considered in conjunction with Eq. (D.9). The initial condition r(x; 0) imposed
upon r is obtained from the uids (assumed pressure-independent) equation of state
r = r(T) together with knowledge of the initial temperature distribution, T(x; 0)
within the uid. Solution of Eq. (D.10) subject to these boundary and initial
conditions serves to determine r(x; t): In turn, such knowledge immediately furnishes
v
m
(x; t) from Eq. (D.9), and T(x; t) from the uids equation of state, thereby
demonstrating the validity of the assumed vanishing volume velocity condition, v
v
=
0: As an aside, it should be noted that this latter condition is consistent with the
speculations of Onsager [114] and Haase [15,115], cited in Section 7, albeit in
connection with their isothermal binary diffusion problems rather than our present
similarly-structured single-component heat transfer problem.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 120
Appendix E. Mathematical analogy between electrophoretic and thermophoretic
motions
Upon identifying the pair of electrophoretic elds (v
f
; j
f
); dened below, with their
thermophoretic analogs (v
v
; j
v
); dened earlier in Eqs. (1.5) and (1.6), a complete
mathematical analogy will be seen to exist between steady-state electrophoretic and
thermophoretic particle motions in liquids for the respective nonconducting particle
cases, at least in the Debye thin double-layer limit. Explicitly, dene the following
electrophoretic quantities:
v
f
:=v
m
j
f
where j
f
:= M
f
Vf ; (E.1)
in which M
f
:=z=m is the electrophoretic mobility. Inasmuch as V
2
f = 0 outside of
the Debye layer it follows that
V
.
j
f
= 0 and V
2
j
f
= 0 : (E.2)
Equations (E.1) and (E.2) represent the analogs of the corresponding thermophoretic
quantities,
v
v
:=v
m
j
v
where j
v
:= M
v
VT ; (E.3)
in which M
v
:=ab: Since V
2
T = 0 in the nonconducting thermophoretic case (see
Table 2) it follows that
V
.
j
v
= 0 and V
2
j
v
= 0 : (E.4)
Owing to the respective analogs, Eqs. (E.2) and (E.4), one has for the respective
choices of f or v; both denoted below by the common symbol g; that
V
.
v
m
= V
.
v
g
= 0 and V
2
v
m
= V
2
v
g
= 0 : (E.5)
Both the electrophoretic boundary-value problem outlined in Section 8 (with the
subscript f implicitly appearing therein replaced throughout by g) and the
nonconducting thermophoretic boundary value problem outlined in Section 3 (with
the subscript v replaced therein by g) satisfy exactly the same set of equations:
Vp = mV
2
v
g
; V
.
v
g
= 0; V
.
j
g
= 0; v
g
= v
m
j
g
; (E.6)
and boundary conditions
n
.
j
g
= 0 on qV
s
; v
m
U = j
g
on qV
s
; v
g
j
(0)
g
=M
g
as [x[ o ; (E.7)
where j
(0)
g
= const refers to the undisturbed uniform value at innity.
The two classes of problems are thus seen to be mathematically identical, which is
why both lead to the same size-, shape-, and orientation-independent result for the
velocity of the respective phoretic particles, namely
U = j
(0)
g

M
f
Vf
(0)
;
M
v
VT
(0)
;
_
(E.8)
in the electrophoretic and thermophoretic cases. This analogy obtains despite the
very different physics underlying the sources of the respective particle motionsone
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 121
animated by externally-imposed electrical body forces, the other by an externally
imposed temperature gradient in the absence of body forces! [148].
While we have pointed out the analogy only for the elementary case of an isolated
particle in an effectively unbounded uid, the analogy can be shown to persist
irrespective of the presence of other (nonconducting) boundaries as, for example, in
cases where a plane wall or a circular cylinder bounds the particle-containing uid
externally. Existing solutions of the electrophoretic problems posed for such
congurations [149] thereby automatically provide solutions for their thermophore-
tic counterparts.
References
[1] G.G. Stokes, Trans. Cambridge Philos. Soc. 8 (1845) 287 (see also Mathematical and Physical
Papers, vol. 1, Cambridge University Press, Cambridge, 1901, p. 75).
[2] An account of pre-1845 work by others on the NavierStokes equations can be found in G.G.
Stokes, Report on recent researches in hydrodynamics, British Assoc. Advance. Sci., 1846, pp. 120.
Reprinted in Mathematical and Physical Papers, vol. 1, Cambridge University Press, Cambridge,
1901, p. 157. A concise history of the conceptual foundations of uid mechanics from the time of
Newtons Principia in 1687 up to the denitive work of Stokes in 1845, can be found in the following
articles: C. Truesdell, Am. Math. Monthly 60 (1953) 445;
O. Darrigol, Arch. Hist. Exact Sci. 56 (2002) 95.
[3] Here and throughout, a question mark surmounting an equality sign serves to suggest that the stated
equality is, at this point in the manuscript, an as yet unresolved issue, one which will, however, later
be resolved against the equality!.
[4] H. Brenner, Kinematics of volume transport, Physica A (2005), in press [doi:10.1016/
j.physa.2004.10.033].
[5] The volume velocity v
v
is, in fact, identical with the volume ux density or volume current n
v
[4],
dened such that with dS a directed element of surface area xed in space, the scalar dS
.
n
v
gives the
volume owing across dS per unit time. This is the analog of the fact that with dS
.
n
m
the ux of
mass across dS; in which n
m
is the mass current, the mass velocity, v
m
:=n
m
=r; represents the
convective portion of the volume ux and j
v
the diffusive portion.
[6] S. Chapman, T.G. Cowling, The Mathematical Theory of Non-Uniform Gases, third ed.,
Cambridge University Press, Cambridge, 1970.
[7] D. Burnett, Proc. London Math. Soc. 39 (1935) 385;
D. Burnett, Proc. London Math. Soc. 40 (1936) 382.
[8] J.C. Maxwell, Philos. Trans. Roy. Soc. (London) A 170 (1879) 231. Reprinted in: W.D. Niven (Ed.),
The Scientic Papers of James Clerk Maxwell, vol. 2, Cambridge University Press, Cambridge, 1890,
p. 681.
[9] R.J. Adrian, Ann. Rev. Fluid Mech. 23 (1991) 261;
R.J. Goldstein (Ed.), Fluid Mechanics Measurements, second ed., Taylor & Francis, Washington
DC, 1996;
Th. Dracos, Three-Dimensional Velocity and Vorticity Measuring and Image Analysis Techniques,
Kluwer, Dordrecht, 1996;
F.T.M. Nieuwstadt (Ed.), Flow visualization and image analysis, Fluid Mechanics and its
Application, vol. 14, Kluwer, Dordrecht, 1992;
M. Raffel, C. Willert, J. Kompenhans, Particle Image Velocimetry. A Practical Guide, Springer,
New York, 1998;
M. Stanislas, J. Kompenhans, J. Westerweel (Eds.), Particle Image Velocimetry, Kluwer, Dordrecht,
2000.
[10] Presumably, v
m
can be independently measured experimentally at a point of the uid by some
colorimetric method, involving the addition of dye to the uid or, even better, instead of adding a
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 122
foreign coloring agent (and thereby obfuscating the notion of a single-component uid) by
performing an optical experiment with a single-component uid whose molecules are photochromic
or uorescent. These latter techniques involve so-called molecular tagging velocimetry (MTV)
[11], as opposed to particle-image velocimetry (PIV) [9], which involves monitoring tracer
particles, namely foreign objects deliberately introduced into the uid.
[11] K.G. Roesner, Mol. Cryst. Liq. Cryst. 298 (1997) 243;
C.P. Gendrich, M.M. Koochesfahani, D.G. Nocera, Exp. Fluids 23 (1997) 361;
W.R. Lempert, in: A.J. Smits, T.T. Lim (Eds.), Flow Visualization: Techniques and Examples,
Imperial College Press, London, 2000;
P. Mavros, Trans. Inst. Chem. Eng. 79 (2001) 113;
S.J. Muller, Korea-Australia Rheol. J. 14 (2002) 93.
[12] I. Mu ller, T. Ruggeri, Extended Thermodynamics, Springer, New York, 1993;
K. Wilmanski, Thermomechanics of Continua, Springer, Berlin, 1998.
[13] S.R. de Groot, P. Mazur, Non-Equilibrium Thermodynamics, North-Holland, Amsterdam, 1962.
[14] D.D. Fitts, Nonequilibrium Thermodynamics, McGraw-Hill, New York, 1962.
[15] R. Haase, Thermodynamics of Irreversible Processes, Dover reprint, New York, 1990.
[16] G.D.C. Kuiken, Thermodynamics of Irreversible Processes: Applications to Diffusion and
Rheology, Wiley, New York, 1994.
[17] L. Onsager, Phys. Rev. 37 (1931) 405;
L. Onsager, Phys. Rev. 38 (1931) 2265;
H.B.G. Casimir, Rev. Mod. Phys. 17 (1945) 343;
I. Prigogine, Introduction to Thermodynamics of Irreversible Processes, second ed., Interscience,
New York, 1961.
[18] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena, second ed., Wiley, New York,
2002.
[19] With the apparent exception of a few authors (see, for example, Refs. [20]) it has not been clearly
recognized in the literature that a need exists for a formal proof that the specic momentum, say ^ m;
of a uid is equal to its mass velocity v
m
: Rather, as judged by accounts found in uid mechanics
textbooks, which implicitly assume it a priori without discussion, the constitutive relation in ^ m = v
m
is regarded as an identity.
[20] L.D. Landau, E.M. Lifshitz, Fluid Mechanics, second ed., Butterworth-Heinemann, Oxford, 1987,
p. 196;
P. Kosta dt, M. Liu, Phys. Rev. E 58 (1998) 5535.
[21] J.R. Bielenberg, H. Brenner, Continuum thermodynamics in the presence of the diffusive transport
of volume, Contin. Thermo. Mech., 2005, to be submitted.
[22] J.G. Kirkwood, I. Oppenheim, Chemical Thermodynamics, McGraw-Hill, New York, 1961.
[23] H. Brenner, J.R. Bielenberg, A continuum theory of phoretic phenomena: thermophoresis, Physica
A (2004) submitted.
[24] J.R. Bielenberg, H. Brenner, A continuum model of thermal transpiration, J. Fluid Mech., 2004,
submitted.
[25] J.R. Bielenberg, H. Brenner, A continuum theory of phoretic phenomena: diffusiophoresis, Phys.
Fluids, 2004, submitted.
[26] Indeed, at the hands of D.D. Joseph, his co-workers, and others (see Ref. [127] as well as the
extensive references cited in Ref. [4]), Eq. (2.13) is often used in applications to compressible uids,
at least in the case of isothermal binary diffusion problems, where our single-component
adiabatically additive volume law based on (q^ v=qT)
p
is replaced by its better known (cf. [4])
multicomponent species additive volume law counterpart based on (q^ v=qw
i
)
p;T
; where w
i
is the mass
fraction of species i: In the latter context, Eq. (2.13) is referred to as expressing a condition of quasi-
incompressibility [27] in circumstances where r is not constant throughout the uid.
[27] J. Lowengrub, L. Truskinovsky, Proc. Roy. Soc. (London) A 454 (1998) 2617.
[28] This elementary equivalence is true only in circumstances where the no-penetration boundary
condition (1.8) imposed upon v
m
at solid surfaces can be replaced by a comparable condition
imposed upon v
v
; for in such circumstances Eq. (1.7), in conjunction with the latter condition, then
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 123
leads to the single vector velocity boundary condition, v
v
= 0 on qV
s
: This no-penetration
equivalency will obviously obtain in circumstances where n
.
(v
m
v
v
) = 0 on qV
s
: Equivalently,
from Eq. (1.5) this necessitates that n
.
j
v
= 0 on qV
s
: From (1.6), this latter condition will prevail
whenever n
.
Vr = 0 on qV
s
or, equivalently, when n
.
V^ v = 0 on qV
s
: In the present single-
component case, and for the case where the law of adiabatically additive volumes prevails, this
requires that n
.
VT = 0 on qV
s
and, hence, from Eq. (2.5) that n
.
q = 0 on qV
s
: In turn, from
Eq. (2.7) this is equivalent to the condition that n
.
j
u
= 0 on qV
s
; which, because it is also true that in
these same circumstances that n
.
j
v
= 0 on qV
s
; leads to the observation that in such circumstances it
is immaterial whether j
u
is given constitutively by the classic expression (2.4) or by its nontraditional
counterpart (2.6). In summary, the complete vector velocity boundary condition, v
v
= 0 on qV
s
; will
obtain whenever no diffusive transport of internal energy occurs across the solid-uid interface,
corresponding to the insulation boundary condition, n
.
j
u
= 0 on qV
s
: For nonconducting cases,
Eqs. (2.13) and (2.14) together with the boundary conditions (1.7) and (1.8) are indistinguishable
from those governing v
m
in the classical creeping ow case.
[29] With regard to use of the term phoretic forces to describe particle motion in the presence of
gradients, Anderson [30] has inadvertantly sowed some degree confusion owing to his use of terms
like thermophoresis and diffusiophoresis, normally reserved for gases [2325], to describe
phenomena that are actually driven by surface-gradient forces in liquids [31,32], see also [132].
The latter category is typied by Marangoni forces resulting from interfacial tension gradient V
s
g;
caused by a surface temperature gradient V
s
T along the particle surface, owing to the functional
dependence of interfacial tension g upon T: The resulting Marangoni surface stress causes the
particle to move against the temperature gradient. However, the forces associated therewith give rise
to a particle velocity U generally dependent upon the size of the particle [32], whereas in non-
Brownian thermophoretic experiments [23] U is observed to be independent of particle size, ruling
out Marangoni forces as possibly responsible for the observed, size-independent, thermophoretic
movement.
[30] J.L. Anderson, Ann. Rev. Fluid Mech. 21 (1989) 61.
[31] D.A. Edwards, H. Brenner, D.T. Wasan, Interfacial Transport Processes and Rheology,
Butterworth-Heinemann, Boston, 1991.
[32] E. Ruckenstein, J. Colloid Interf. Sci. 83 (1981) 77;
T. Keyes, J. Stat. Phys. 33 (1983) 287;
V.G. Levich, V.S. Krylov, Ann. Rev. Fluid Mech. 1 (1969) 293.
[33] This occurred when Tyndall observed dust-free regions in proximity to heated surfaces and wires in a
chamber lled with dust-laden air; J. Tyndall, Proc. R. Inst. 6 (1870) 1. (For further historical
references, see Ref. [34]).
[34] F. Zheng, Adv. Colloid Interf. Sci. 97 (2002) 255.
[35] P.S. Epstein, Z. Phys. 54 (1929) 537.
[36] J.R. Brock, J. Colloid Sci. 17 (1962) 768;
G.M. Hidy, J.R. Brock, The Dynamics of Aerocolloidal Systems, Pergamon Press, Oxford,
1970.
[37] L. Waldmann, K.H. Schmitt, Thermophoresis and diffusiophoresis of aerosols, in: C.N. Davies
(Ed.), Aerosol Science, Academic Press, London, 1966, p. 137.
[38] B.V. Derjaguin, Yu.I. Yalamov, J. Colloid Sci. 20 (1965) 555;
B.V. Derjaguin, A.I. Storozhilova, Ya.I. Rabinovich, J. Colloid Interf. Sci. 21 (1966) 35;
B.V. Derjaguin, Yu.I. Yalamov, The theory of thermophoresis and diffusiophoresis of aerosol
particles and their experimental testing, in: G.M. Hidy, J.R. Brock (Eds.), Topics in Current Aerosol
Research, Pergamon Press, Oxford, 1972, p. 2;
B.V. Derjaguin, Ya.I. Rabinovich, A.I. Storozhilova, G.I. Shcherbina, J. Colloid Interf. Sci. 57
(1976) 451.
[39] L. Talbot, R.K. Cheng, R.W. Schefer, D.R. Willis, J. Fluid Mech. 101 (1980) 737;
L. Talbot, Thermophoresisa review, in: S.S. Fisher (Ed.), Rareed Gas Dynamics, Part 1, AIAA,
New York, 1981, p. 467.
[40] W. Oostra, J.C.M. Marijnissen, B. Scarlett, Space Forum 3 (1998) 251.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 124
[41] Knudsens work on noncontinuum effects in thermal transpiration ows did not even appear
until 1910 [42], whence it is unlikely that the concept of noncontinuum behavior would
have even arisen in Maxwells mind in 1879. [Indeed, the fact that Maxwell applied his slip
condition to the strictly continuum NSF equations supports our belief that he regarded
his so-called slip condition to be a continuum effect arising from the surface temperature gra-
dient. In this context it is noteworthy that the adherence of the uid to a solid surfaceso
widely accepted today in the case of continua, irrespective of whether or not the surface is
isothermalwould, in the case of nonisothermal continua, not likely to have been regarded as
sacrosanct during Maxwells era. After all, very little data pertinent to the issue existed at that
time.] Concomitantly, the standard explanation found in textbooks [43] to the effect that
the thermophoretic particle motion observed in gases is molecular (i.e., noncontinuum) in
origin, arising from more energetic particles striking the hotter side of the particle and overcom-
ing the opposing effects of the less energetic particles on the colder side, is untenable in the
continuum limit.
[42] M. Knudsen, Ann. Phys. (Leipzig) 31 (1910) 205;
M. Knudsen, Ann. Phys. (Leipzig) 33 (1910) 1435.
[43] E.H. Kennard, Kinetic Theory of Gases, McGraw-Hill, New York, 1938;
L.B. Loeb, The Kinetic Theory of Gases, Dover reprint, New York, 1961. For a discussion of
Maxwells role in explaining Crookess radiometer, see Ref. [44].
[44] S.G. Brush, The Kind of Motion that we call Heat, North-Holland, Amsterdam, 1976.
[45] G.S. McNab, A. Meisen, J. Colloid Interf. Sci. 44 (1973) 339.
[46] Others [32] have suggested that phoretic motion in liquids may actually be due to Marangoni-like
surface effects [31], wherein the surface is not passive, as in our model, but rather interacts
physicochemically with the uid. However, as discussed in Section 8 such particle motion requires
the action of body forces, which are absent as the animating mechanism underlying Eq. (3.1) for
liquids and (3.4) for gases.
[47] S. Semenov, M. Schimpf, Phys. Rev. E 69 (2004) 011201.
[48] J.R. Bielenberg, H. Brenner, A hydrodynamic/Brownian motion model of thermal diffusion in
liquids, Phys. Rev. E (2004), to be submitted.
[49] The absence of bulk viscosity effects in (4.1) derives from the fact that the volume velocity appearing
in Eq. (1.1) is assumed to obey Eq. (2.13), a conclusion consistent with the choice of the constitutive
equation (1.6) and valid, for example, in the case of ideal gases.
[50] To describe these as being the noncontinuum terms, without including the rst-order near-
continuum O(Kn) NSF terms in the appellation, is surely confusing, certainly to uid
mechanicians who regard the O(Kn) NSF terms, and not the O(Kn
0
) Euler terms, as the
equations of continuum uid mechanics; that is, owing to their apparent Knudsen number
dependence, the latter classical near-continuum rst-order NSF terms should, for consistency,
also be classied as noncontinuum terms, despite their being regarded by uid mechanicians as
strictly continuum-level terms.
[51] Even higher-order, O(Kn
3
); so-called super-Burnett terms [52] exist. For a contextual evaluation of
the Burnett, super-Burnett, and generally higher-order contributions to the linear momentum
equation, see Ref. [53].
[52] A.V. Bobylev, Sov. Phys. Doklady 27 (1982) 29;
F.J. Uribe, R.M. Velasco, L.S. Garcia-Colin, Phys. Rev. E 62 (2000) 5835;
M. Slemrod, Arch. Rational Mech. Anal. 150 (1999) 1.
[53] R.K. Agarwal, K.Y. Yun, R. Balakrishnan, Phys. Fluids 13 (2001) 3061.
[54] The comparable Burnett terms for the heat ux do not impact upon whether or not Eq. (2.6) is or is
not correct, since gas kinetic theory [6] draws no clear-cut distinction between the heat ux q and the
diffuse internal energy current j
u
:
[55] M.N. Kogan, V.S. Galkin, O.G. Fridlander, Sov. Phys. Usp. 19 (1976) 420;
M.N. Kogan, Ann. Rev. Fluid Mech. 5 (1973) 383;
M.N. Kogan, Progr. Aerospace Sci. 29 (1992) 271.
[56] A.V. Bobylev, J. Stat. Phys. 80 (1995) 1063.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 125
[57] E. Yariv. H. Brenner, A continuum alternative to the ghost effect of gas-kinetic theory, Phys.
Fluids, 2004, submitted.
[58] C. Truesdell, R.G. Muncaster, Fundamentals of Maxwells Kinetic Theory of a Simple Monatomic
Gas, Academic Press, New York, 1980.
[59] Although not required for the subsequent calculations, as an aside we note that m = (const:) T [6] for
Maxwell molecules, from which it follows that K
2
= 3 for such molecules.
[60] In the latter context, note that Eq. (4.8) is consistent with the fact that k is known [6] to be identically
zero for monatomic ideal gases owing to the assumed spherically symmetric nature of such
molecules.
[61] A.D. Kovalenko, Thermoelasticity, Wolters-Noordhoff, Groningen, 1969;
H. Parkus, Thermoelasticity, Springer, Wien, New York, 1976;
D. lesan, A. Scalia, Thermoelastic Deformations, Kluwer, Dordrecht, 1996;
G.A. Maugin, A. Berezovski, J. Thermal Stresses 22 (1999) 421;
N. Noda, R.B. Hetnarski, Y. Tanigawa, Thermal Streses, Taylor and Francis, London, 2002.
[62] Indeed, in the case of solids, the notion of a noncontinuum solid does not even appear to exist,
except perhaps in the case of granular materials, although fractures and dislocations, representing
isolated singularities, may exist within the solid.
[63] D.J. Korteweg, Arch. Ne erl. Sci. Exactes Naturelles II 6 (1901) 1.
[64] C. Cercignani, Mathematical Methods in Kinetic Theory, second ed., Plenum Press, New York,
1990.
[65] H.A. Kramers, J. Kistemaker, Physica 10 (1943) 699.
[66] D.A. Noever, Phys. Fluids A 2 (1990) 858;
D.A. Noever, Phys. Lett. A 144 (1990) 253;
D.A. Noever, Phys. Rev. Lett. 65 (1990) 1587;
D.A. Noever, Phys. Rev. A 45 (1992) 7302.
[67] W. Crookes, Philos. Trans. Roy. Soc. (London) 166 (1876) 325.
[68] A detailed and historical discussion of attempts to explain the principles underlying the windmill-like
rotation undergone by the rotor in Crookess radiometer based upon noncontinuum concepts is
given in Ref. [44].
[69] J.C. Maxwell, Philos. Mag. 19 (1860) 19;
J.C. Maxwell, Philos. Mag. 20 (1860) 21.
[70] Some of the historical context, chronology, and acrimony in the matter of priority surrounding the
competition between Maxwell and Osborne Reynolds [71] to use their respective thermal
transpiration models to explain the physical mechanism underlying the working of Crookess
radiometer [67] can be found in the biography by I. Tolstoy, James Clerk Maxwell, University of
Chicago Press, Chicago, 1981, pp. 150151, 166167; see also Ref. [44].
[71] O. Reynolds, Proc. Roy. Soc. London 38 (18791880) 300. This paper is only a preliminary abstract
of the lengthier paper published some time afterwards as O. Reynolds, Philos. Trans. Roy. Soc.
(London) 170 (1879) 727.
[72] The importance of understanding the mechanism behind Crookess radiometer [67] played a vital,
and under-appreciated, role in the history of gas-kinetic theory, in particular in regard to the
boundary conditions to be applied to the Boltzmann equation at solid surfaces. After all, an
important part of the verication of the validity of the Boltzmann equation necessarily lies in the
agreement of its predictions with experiment, for which circumstances the solution of boundary-
value problems (either imposed upon the Boltzmann equation itself or upon the coarser-scale
transport equations derived therefrom, such as the NSF equations) plays a pre-eminent role.
[73] For a modem version of the slip boundary condition involving G for gases, see F. Sharpov,
D. Kalempa, Phys. Fluids 15 (2003) 1800.
[74] Given the interpretation of Maxwell slip as a noncontinuum O(Kn
2
) effect owing to its origin in
connection with the Burnett terms, Epstein [35] and those who followed should, for mathematical
consistency as regards the hierarchical ordering of the Knudsen number terms appearing in their
transport equations, have then solved the corresponding noncontinuum O(Kn
2
)-level transport
equations, rather than the near-continuum O(Kn) NSF equations. At a minimum, this would have
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 126
resulted in adding the Maxwell thermal stress term (5.3) to the O(Kn) viscous Newtonian term (4.2)
appearing in the momentum equation. Additionally, because the gas is compressible owing to its
density varying with temperature, the continuity equation used by Epstein, namely V
.
v
m
= 0; is
valid only to O(Kn): At O(Kn
2
) another term should have appeared in his continuity equation in
order that the latter be correct. However, as discussed in Appendix C, owing to a fortuitous
combination of circumstances in the present class of phoretic thermal problems [23,24], these
additions do not affect the calculation of U:
[75] As discussed in connection with Eqs. (5.4) and (5.5), the notion of noncontinuum slip is associated
with the parameter G appearing therein, rather than with the last term of Eq. (5.4), which alone
governs Maxwells slip coefcient, C
s
: In the literature [73], G is associated with the notion of
velocity slip, a truly noncontinuum phenomenon occurring even in isothermal uids.
[76] L. Euler, Me m. Acad. Sci. Berlin 11 (1755) 274. Reproduced in: Leonhardi Euleri Opera Omnia.
Series II, vol. 12, Fu ssli, Zu rich, 1954, p. 54. Additional historical information can be found in the
Editors Introduction to the latter volume by C. Truesdell, Rational uid mechanics, 16871765,
pp. VIICXXV; see also L. Euler, Hist. Acad. Berlin 1755 (1757) 316361.
[77] G.K. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press, Cambridge,
1967.
[78] By small is meant the following: If a is the maximum linear dimension of the particle, it is required
that a[[Vv
l
[[=[v
l
[51; with the modulus bars denoting appropriate norms.
[79] C.C. Truesdell, R.A. Toupin, The classical eld theories, in: S. Flu gge (Ed.), Handbuch der Physik,
vol. IIII/1, Principles of Classical Mechanics and Field Theory, Springer, Berlin, 1960, p. 226;
C. Truesdell, W. Noll, The Nonlinear Field Theories of Mechanics, in: S. Flu gge (Ed.), Handbuch
der Physik, vol. III/3, Springer, Berlin, 1965;
W. Noll, R.A. Toupin, C.C. Wang, Continuum Theory of Inhomogeneities in Simple Bodies,
Springer, Berlin, 1968.
[80] Of course, in the case of unsteady ows, the necessity of performing repetitive experiments with
different size particles, all at the same instant of time, would, no doubt, pose a daunting challenge to
the experimentalist!.
[81] A perhaps equally remarkable fact about Eq. (3.2), applicable to gases, is that it reveals a totally
counter-intuitive uid-mechanical phenomenonnamely, the larger the viscosity of the gas the
faster does the particle move! This fact alone signals the extraordinarily unique nature of
thermophoretic motion, since viscosity generally retards rather than enhances relative particle
motion through uids, a fact well known to every low Reynolds number uid mechanician [82].
[82] J. Happel, H. Brenner, Low Reynolds Number Hydrodynamics, Prentice-Hall, Englewood Cliffs,
NJ, 1965.
[83] P. Goldsmith, F.G. May, Diffusiophoresis and thermophoresis in water vapour systems, in: C.N.
Davies (Ed.), Aerosol Science, Academic Press, London, 1966, p. 163 (see also Ref. [38]).
[84] In order for an investigator be able to objectively identify his velocity measurements as
representative of those of the uid itself, and not an artifact of the properties of the tracer
particle, he needs to assure himself that his experimental tracer particles do not possess any
physical attributes that, in the zero-size limit, would distinguish the particles velocity from
that of the uid itself. It was in order to fulll this requirement of passivity that only
(effectively) thermally insulated thermophoretic spheres were selected by us in order to identify
the velocity v
l
of the undisturbed uid. As revealed by Eq. (3.1), thermophoretically
animated spheres possessing a nonzero k
s
=k ratio move with a velocity that depends
signicantly upon the magnitude of this conductivity ratio, even in the limit of effectively
zero size. As such, (effectively) noninsulated particles may not serve as uid velocity tracers.
It is only to this extent that the experimental uid mechanician, in deciding upon the choice
of appropriate tracer particles with which to conduct his velocity experiments, would have
to contemplate the possible complicating effects of temperature gradients. Even were
he insufciently insightful to recognize a priori the need for insulated particles, were he to
next perform a sequence of replicate size-varying experiments using a series of particles
possessing different thermal conductivities (just as he might do with a series of particles of
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 127
different densities, so as to assure himself of their zero-size passivity), he would presumably
soon come to recognize that all low conductivity particles yielded identical extrapolated zero-size
velocities. Accordingly, he would presumably then reject all zero-size particle data obtained with
his high conductivity particles as failing to fulll the requirement of passivity (even were he
unable to identify thermal conductivity as the source of the observed differences in the zero-size
velocity measurements).
[85] Even were external forces such as gravity to act on the uid, enabling the particle to sediment
relative to the surrounding uid if its density differed from that of the uid, such relative motion
would vanish in the pointsize tracer-particle limit, thereby having no effect upon the ability of the
tracer particle to monitor the uid velocity that exists in its absence.
[86] By the phrase gas-kinetic molecular interpretation is meant that the property cannot be derived
directly simply by summing each of the three elemental extensive properties of the individual
molecules in some small domain of volume V (namely the mass m; kinetic energy mc
2
=2; and
momentum mc of the molecules, with c the molecular velocity) and subsequently dividing by the
volume of that domain in order to obtain the corresponding intensive volumetric pointwise mass,
kinetic energy, and momentum densities at a point of the continuum.
[87] The reason for separating these two items stems from the fact (noted in connection with Table 2
appearing in Appendix C) that it is possible under certain well-dened circumstances for the
traditional and modied NSF equation set to fortuitously yield identical results, both of which
accord with experiment, albeit on the proviso that the correct velocity boundary condition be used
(either that of no-slip imposed upon v
l
or the equivalent Maxwell slip condition imposed upon v
m
).
[88] The isothermal assumption is needed in order to avoid complications associated with thermal
diffusion species uxes, while the isobaric assumption is similarly required to avoid pressure
diffusion contributions to the species ux density j
i
[18].
[89] This has the effect of enabling the right-hand side of (7.1) [and, equivalently, that of Eq. (1.6) for the
isothermal, isobaric, binary diffusion case] to be re-written in the form j
v
= DV ln r = D
+
V^ v =
D
+
(q^ v=qw
1
)
p;T
Vw
1
= (q^ v=qw
1
)
p;T
j
1
:
[90] We use the word semi-empirical here because there does not appear to exist in the literature a
theoretical proof of the concentration-slip boundary condition, derived along the lines laid out by
Maxwell [8] in the thermal gradient case, wherein the concentration analog of Eq. (5.4) is derived
from the analog of the MaxwellBurnett thermal stress term (5.3). Rather, owing to this lack,
Kramers and Kistemaker [65] adopted their widely-used concentration-slip velocity condition on a
different basis, namely a molar rather than mass basis. Explicitly, we are not aware of the existence
in the literature of the Burnett extra stress concentration analog of Eq. (4.5), although if our theory
is correct it should be given by Eq. (4.6), in which j
v
= DV ln r (see Ref. [89]). According to our
theory, the generic no-slip boundary condition should be given by Eq. (5.8), where I
s
.
j
v
= DV
s
ln r
in the present binary mixture case.
[91] Reynolds [71] experiments were actually performed with porous plugs rather than with well-dened
capillary tubes.
[92] S.E. Vargo, E.P. Muntz, G.R. Shiett, W.C. Tang, J. Vac. Sci. Technol. A 17 (1999) 2308;
J.P. Hobson, D.B. Salzman, J. Vac. Sci. Technol. A 18 (2000) 1758;
F. Ochoa, C. Eastwood, P.D. Ronney, B. Dunn, Thermal transpiration based microscale propulsion
and power generation devices, seventh Intl. Microgravity Combustion Workshop, Cleveland, OH,
June 2003;
Y. Sone, K. Sato, Phys. Fluids 12 (2000) 1864.
[93] In regard to these experiments, note that according to Eq. (3.3) the slip coefcient is different for
monatomic and diatomic gases.
[94] H. Brenner, Molecular, Brownian motion conrmation of the tracer/mass velocity disparity, Phys.
Rev. E (under revision; originally submitted under the title Molecular, Brownian motion
conrmation of the Euler/Lagrange velocity disparity).
[95] E. Nelson, Phys. Rev. 150 (1966) 1079;
E. Nelson, Dynamical Theories of Brownian Motion, Princeton Univ. Press, Princeton,
1967;
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 128
E. Nelson, Connection between Brownian motion and quantum mechanics, in: H. Nelkowski, A.
Hermann, H. Poser, R. Schrader, R. Seiler (Eds.), Einstein Symposium, Berlin, Lecture Notes in
Physics, vol. 100, Springer, Berlin, 1979, p. 168.
[96] K. Ito , Stochastic Calculus, Springer Lecture Notes in Physics, vol. 39, Springer, New York, 1975,
p. 218.
[97] In probabilistic terminology, v
l
constitutes Nelsons drift velocity (more precisely, his forward
drift velocity, since a hypothetical mathematically-dened backwards drift velocity also appears in
Nelsons theory, a fact that need not concern us here). In his notation, Nelsons symbol v [no relation
to our v in either Eqs. (1.3) and (1.4)] is equivalent to our mass velocity v
m
; as is apparent from its
appearance in Nelsons continuity equation, analogous to our Eq. (1.2).
[98] In our interpretation of Nelsons work, D might better be termed the uids self-diffusion
coefcient, since his analysis appears limited to single-component uids within which
mass density gradients exist; that is, the symbol D appearing in (7.3) is regarded as being a
self-diffusion coefcient, intrinsic to the single-component uid itself, rather than arising from the
presence a foreign object, namely a colloidal particle, present in the uid. In this sense, D should be
regarded as an isotropic correlation coefcient, ID = (1=2)DxDx)=Dt; in which the position vector
x = x(x
0
; t) represents the statistical location at time t of a uid particle that at time t = 0 was
located at the position x
0
: The phrase uid particle here refers not to a material particle (which is
an extensive entity) but rather to a uid particle (an intensive entity) in the sense implicitly understood
in connection with Eq. (1.9), where the tracer uid eld, v
l
(x
0
; t); is regarding as describing the (mean)
tracer motion, the so-called forward motion, of such a hypothetical uid particle. Mathematically,
the symbol D is that appearing in the Markoff process stochastic relation [99] dx(t) = v
l
[x(t); t]dt

2D
_
dw(t); in which dx(t) x x
0
; with dw(t) a normalized Wiener process [100].
[99] P. Garbaczewski, Phys. Rev. E 57 (1998) 569.
[100] L. Arnold, Stochastic Differential Equations, Wiley-Interscience, New York, 1974.
[101] E. Schro dinger, Sitzungsber. Preuss. Akad. Wiss. Phys.-Math Klasse 1 (1931) 144;
J.C. Zambrini, Physica B,C 151 (1988) 327.
[102] E. Nelson, Quantum Fluctuations, Princeton University Press, Princeton, 1985.
[103] G. Bacciagaluppi, Founds. Phys. Lett. 12 (1999) 1.
[104] P. Garbaczewski, Phys. Lett. A 147 (1990) 168. With regard to the latters notion of Brownian
recoil, the work of Streater et al. cited later in Ref. [119] appears also to introduce a similar
Brownian recoil-like effect.
[105] P. Garbaczewski, Phys. Lett. A 143 (1990) 85;
P. Garbaczewski, Phys. Lett. A 172 (1993) 208;
P. Garbaczewski, Phys. Lett. A 178 (1993) 7;
P. Garbaczewski, J.P. Vigier, Phys. Rev. A 46 (1992) 4634;
P. Garbaczewski, J.P. Vigier, Phys. Lett. A 167 (1992) 445;
P. Blanchard, P. Garbaczewski, Phys. Rev. E 49 (1994) 3815. For later references see P.
Garbaczewski, Physica A 317 (2003) 449.
[106] K. Namsrai, Nonlocal Quantum Field Theory and Stochastic Quantum Mechanics, Kluwer,
Dordrecht, 1985;
J.C. Zambrini, Phys. Rev. A 33 (1986) 1532;
J.C. Zambrini, J. Math. Phys. 27 (1986) 2307;
J.C. Zambrini, Phys. Rev. A 35 (1987) 3631;
N.C. Petroni, F. Guerra, Found. Phys. 25 (1995) 297.
[107] A. Einstein, Ann. Phys. 17 (1905) 549; see also A. Einstein, in: R. Fu rth (Ed.), Investigations on the
Theory of the Brownian Movement, Dover reprint, New York, 1956.
[108] M.R. von Smoluchowski, Ann. Phys. 21 (1906) 756.
[109] The essentially kinematic analyses of Einstein and Smoluchowski preceded the later dynamical
theories of Brownian motion phenomena, such as those due to Langevin, Ornstein-Uhlenbeck,
Kramers, etc. (cf. Ref. [110]).
[110] S. Chandrasekhar, Rev. Mod. Phys. 15 (1943) 1; N. Wax (Ed.), Selected Papers on Noise and
Stochastic Processes, Dover reprint, New York, 1954.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 129
[111] Youri Lvovich Klimontovich, late Professor Emeritus in the Physics Department at Moscow State
University, died of cancer on November 27, 2002, approximately three months before I learned of
his contribution to the subject under discussion.
[112] Yu.L. Klimontovich, Statistical Theory of Open Systems, vol. 1: A Unied Approach to Kinetic
Descriptions of Processes in Active Systems, Kluwer, Dordrecht, 1995 (Chapters 13 and 14);
Yu.L. Klimontovich, Theor. Math. Phys. 92 (1992) 909;
Yu.L. Klimontovich, Theor. Math. Phys. 96 (1993) 1035.
[113] It is interesting to note that an identical term appears in the well-known book of de Groot
and Mazur [13], but only in the context of a class of applications involving what they
term discontinuous systems [cf. Eqs. (69) and (72) of their Chapter XV]. Indeed, they
explicitly identify the term j
v
appearing in our subsequent Eq. (7.4), which they term the
volume ow.
[114] L. Onsager, Ann. Trans. NY Acad. Sci. 46 (1945) 241. It is interesting to note in the present context
that Onsager comments as follows when discussing problems of pure multicomponent diffusion in
liquids, involving what appears to us to be thermodynamically ideal solutions: ..provided only that
the volume change due to mixing may be neglected, it is possible to arrange matters such that v = 0
everywhere (where Onsagers hydrodynamic velocity, v; is understood by us to be the volume
velocity). Moreover, he goes on later to further state that: Viscous ow is a relative motion of
adjacent portions of a liquid. Diffusion is a relative motion of its different constituents. Strictly
speaking, the two are inseparable; for the hydrodynamic velocity in a diffusing mixture is merely an
average determined by some arbitrary convention.
[115] As pointed out by Haase [15, p. 221] and others [116], experimentalists who measure molecular
diffusivities usually choose a volume- rather than mass-based reference frame (so as to avoid having
to explicitly address what Haase terms convective velocities, namely nonzero mass-average
velocities, v
m
): In this frame of reference it is supposed: (i) that the volume velocity vanishes
everywhere, v
v
= 0 (corresponding here to v
l
= 0), despite the fact that v
m
a0; and (ii) that the
diffusional process is unidirectional, However, to the best of our knowledge, it appears never to have
been pointed in this connection that the assumption of requiring that v
v
= 0 everywhere, including on
the boundary, is incompatible with the traditional no-slip tangential boundary condition, I
s
.
(v
m

U) = 0; imposed upon v
m
[117]. Equally, the KramersKistemaker [65] species concentration
boundary condition, analogous to (5.2), would also be violated unless it was true that, when
expressed in appropriate binary diffusion terminology, v
v
= v
m
C
s
uV
s
ln T; for, in that case,
Eq. (3.5) becomes identical with Eq. (5.1), corresponding to no slip of the volume velocity (and hence
of the Lagrangian velocity v
l
).
[116] E.L. Cussler, Diffusion, second ed., Cambridge University Press, Cambridge, 1997.
[117] J. Camacho, H. Brenner, Ind. Eng. Chem. Res. 34 (1995) 3326;
T.Y. Liao, D.D. Joseph, J. Fluid Mech. 342 (1997) 37;
P.S. Perera, R.F. Sekerka, Phys. Fluids 9 (1997) 376.
[118] J.C. Maxwell, Proc. Lond. Math. Soc. 3 (1870) 82. Reprinted in: W.D. Niven (Ed.), The Scientic
Papers of James Clerk Maxwell, vol. 2, Cambridge University Press, Cambridge, 1890, p. 208. This
paper discusses Maxwells view of the relation between the Eulerian and Lagrangian velocities of a
uid. While it might appear from Maxwells remarks that he is literally referring to a molecule, it
is clear from his Lagrangian pathline example that his paper addresses, as well as from his later use
of the word molecule, that he is actually referring to a particle of uid, what today would be
referred to as a material uid particle.
[119] R.F. Streater, J. Math. Phys. 38 (1997) 4570;
R.F. Streater, Rep. Math. Phys. 40 (1997) 557;
R.F. Streater, J. Stat. Phys. 88 (1997) 447.
[120] M.R. Grasselli, R.F. Streater, Rep. Math. Phys. 50 (2002) 13.
[121] R.F. Streater, Proc. Roy. Soc. (London) A 456 (2000) 205.
[122] R.F. Streater, Open Syst. Inform. Dynamics 10 (2003) 3.
[123] R.F. Streater, Statistical Dynamics, Imperial College Press, London, 1995.
[124] R. Balescu, Statistical Dynamics, Imperial College Press, London, 1997.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 130
[125] C. Beck, G. Roepstorff, Physica A 165 (1990) 270;
R. Dobrushin, Caricatures of hydrodynamics, in: B. Simon, A. Truman, I.M. Davies (Eds.),
Mathematical Physics, Adam Hilger, Bristol, 1989, p. 117;
Xing Xiu-San, Chinese Sci. Bull. 40 (2001) 447.
[126] J.W. Cahn, J.E. Hilliard, J. Chem. Phys. 28 (1958) 258;
U. Felderhoff, Physica 48 (1970) 541;
H.T. Davis, L.E. Scriven, Adv. Chem. Phys. 49 (1982) 357.
[127] D.D. Joseph, Eur. J. Mech. B/Fluids 9 (1990) 565;
P. Galdi, D.D. Joseph, L. Preziosi, S. Rioero, Eur. J. Mech. B/Fluids 10 (1991) 253;
D.D. Joseph, Y. Renardy, Fundamentals of Two-Fluid Dynamics, Part I, Springer, New York,
1992;
D.D. Joseph, A. Huang, H. Hu, Physica D 97 (1996) 104.
[128] J. Serrin, Q. Appl. Math. 41 (1983) 357;
J.E. Dunn, J. Serrin, Arch. Rat. Mech. Anal. 88 (1985) 95;
P. Pettijeans, T. Maxworthy, J. Fluid Mech. 326 (1996) 37;
C. Chen, E. Meiburg, J. Fluid Mech. 326 (1996) 37;
D.M. Anderson, G.B. McFadden, A.A. Wheeler, Ann. Rev. Fluid Mech. 30 (1998) 139;
J. Badur, J. Banaskiewicz, Archiv. Thermodynamics 19 (1998) 61;
C.Y. Chen, L. Wang, E. Meiburg, Phys. Fluids 13 (2001) 2447;
C.Y. Chen, E. Meiberg, Phys. Fluids 14 (2002) 2052;
J.A. Pojman, Y. Chekanov, J. Masere, V. Volpert, T. Dumont, H. Wilke, Effective interfacial
induced convection (EITIC) in miscible uids, AIAA 2001-0764.
[129] In the case of gases, kinetic theory [18] suggests that a better assumption would be D
v
= D
+
v
=r; where
D
+
v
is a constant, independent of density. This would result in different expressions for the four
Korteweg coefcients than those given in Eq. (8.2).
[130] In the interest of greater generality we could have added a body force per unit volume, say f; to the
right-hand side of (8.3). However, we have refrained from doing so in order to clarify subsequent
arguments regarding other classes of phoretic particle motion [30], which, in contrast with the thrust
of our work involving circumstances where f = 0; are driven by nonzero body forces.
[131] It is important to distinguish between the phenomenon of thermophoresis, which is a single-particle
theory applicable to non-Brownian particles, and that of thermal diffusion, which involves multiple
Brownian particles in a uid [48], collectively forming a second species, so that the solute and solvent
together constitute a nonisothermal binary mixture We bring this up because the SemenovSchimpf
interpretation of experimental data does not appear to strictly constitute a test of single-particle
thermophoresis.
[132] V.G. Levich, Physicochemical Hydrodynamics, Prentice-Hall, Englewood Cliffs, New Jersey, 1962.
[133] F.A. Morrison Jr., J. Colloid Inter. Sci. 34 (1970) 210.
[134] J.R. Melcher, Continuum Electromechanics, MIT Press, Cambridge, MA, 1981;
R.F. Probstein, Physicochemical Hydrodynamics, second ed., Wiley, New York, 1995.
[135] S.N. Semenov, Philos. Mag. 83 (2003) 2199.
[136] M.E. Schimpf, S.N. Semenov, J. Phys. Chem. B 105 (2001) 2285;
M.E. Schimpf, S.N. Semenov, J. Phys. Chem. B 104 (2000) 9935;
J.C. Giddings, P.M. Shiundu, S.N. Semenov, Colloid Interf. Sci. 176 (1995) 454.
[137] F. Brochard, P.-G. de Gennes, C.R. Hebd. Seances Acad. Sci. 293 (1981) 1025.
[138] J.O. Hirschfelder, C.F. Curtiss, R.B. Bird, Molecular Theory of Gases and Liquids, Wiley, New
York, 1954.
[139] M.N. Kogan, Rareed Gas Dynamics, Plenum Press, New York, 1969.
[140] R.J. Atkin, R.E. Craine, Quart. J. Mech. Appl. Math. 29 (1976) 209;
R.J. Atkin, R.E. Craine, J. Inst. Maths Appl. 17 (1976) 153;
C. Truesdell, Rational Thermodynamics, second ed., Springer, Berlin, 1984;
F. Dobran, Theory of Structured Multiphase Mixtures, Springer, Berlin, 1991;
C.R. Reid, F. Jafari, Int. J. Eng. Sci. 33 (1995) 411;
K.R. Rajagopal, L. Tao, Mechanics of Mixtures, World Scientic, Singapore, 1995;
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 131
D.A. Drew, S.L. Passman, Theory of Multicomponent Fluids, Springer, New York, 1998;
A.F. Mills, Int. J. Heat Mass Transfer 41 (1998) 1955;
D.S. Drumheller, Int. J. Eng. Sci. 38 (2000) 347.
[141] J.H. Cushman, The Physics of Fluids in Hierarchical Porous Media: Angstroms to Miles, Kluwer,
Dordrecht, 1997.
[142] F. Sharipov, D. Kalempa, Phys. Fluids 16 (2004) 759.
[143] F. Sharipov, Phys. Rev. E 69 (2004) 061201.
[144] F. Sharipov, Eur. J. Mech. B/Fluids 22 (2003) 133.
[145] F. Sharipov and V. Selenzev, J. Phys. Chem. Ref. Data 27 (1998) 657.
[146] In a generic context, the diffuse current j
c
; of some extensive property C is dened as the ux density
of the property over and above the corresponding convective contribution n
m
^
c thereto carried by the
mass current n
m
= rv
m
: Stated more explicitly, the total current n
c
of the extensive property under
discussion in a Eulerian space-xed reference frame is regarded as being of the form n
c
= n
m
^
c j
c
;
with
^
c is the amount of the property per unit mass, i.e., the specic density of the property C: The
latter density appears in the generic Eulerian transport equation qc=qt V
.
n
c
= p
c
in which c =
r
^
c and p
c
are, respectively, the amount of the property and temporal rate of production of the
property, both on a per unit volume basis. This generic Eulerian transport is formally equivalent to
the generic material derivative form, rD
m
^
c=Dt V
.
j
c
= p
c
:
[147] The latter fact shows, for example, why the thermomolecular pressure difference in the thermal
transpiration problem [24] is correctly given by Maxwells scheme despite the fact that Maxwells
transport equations are inappropriate.
[148] Note that in terms of the fundamental Newtonian stress issue (1.1), the extra deviatoric stress
[cf. (4.2)], T

g
= 2mVj
g
; makes no contribution to the present problems, just as was true in Appendix
C, owing to the fact that since V
2
j
g
= 0; it follows that V
.
T

g
= 0:
[149] E. Yariv, H. Brenner, Phys. Fluids 14 (2002) 3354;
E. Yariv, H. Brenner, J. Fluid Mech. 484 (2003) 85;
E. Yariv, H. Brenner, SIAM J. Appl. Math. 64 (2003) 423.
ARTICLE IN PRESS
H. Brenner / Physica A 349 (2005) 60132 132

You might also like