Download as pdf or txt
Download as pdf or txt
You are on page 1of 35

3B2v8:06a=w ðDec 5 2003Þ:51c Prod:Type:FTP ED:JyothiG

þ model
PHYSA : 9965 pp:1235ðcol:fig::NILÞ PAGN:anu SCAN:

ARTICLE IN PRESS

3
Physica A ] (]]]]) ]]]–]]]
www.elsevier.com/locate/physa
5

7
Fluid mechanics revisited
9
Howard Brenner
11
Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139-4307, USA
13 Received 12 November 2005; received in revised form 21 February 2006

15

17 Abstract

19 Öttinger’s recent nontraditional incorporation of fluctuations into the formulation of the friction matrix appearing in

F
the phenomenological GENERIC theory of nonequilibrium irreversible processes is shown to furnish transport equations
for single-component gases and liquids undergoing heat transfer which support the view that revisions to the

O
21
Navier–Stokes–Fourier (N–S–F) momentum/energy equation set are necessary, as empirically proposed by the author on
the basis of an experimentally supported theory of diffuse volume transport. The hypothesis that the conventional N–S–F
23
O
equations prevail without modification only in the case of ‘‘incompressible’’ fluids, where the density r of the fluid is
uniform throughout, serves to determine the new phenomenological parameter a0 appearing in the GENERIC friction
PR
25 matrix. In the case of ideal gases the consequences of this constitutive hypothesis are shown to yield results identical to
those derived theoretically by Öttinger on the basis of a ‘‘proper’’ coarse-graining of Boltzmann’s kinetic equation. A
27 major consequence of the present work is that the fluid’s specific momentum density v is equal to its volume velocity vv ,
rather than to its mass velocity vm , contrary to current views dating back 250 years to Euler. In the case of rarefied gases
D

29 the proposed modifications are also observed to agree with those resulting from Klimontovich’s molecularly based, albeit
ad hoc, self-diffusion addendum to Boltzmann’s collision integral. Despite the differences in their respective physical
TE

31 models—molecular vs. phenomenological—the role played by Klimontovich’s collisional addition to Boltzmann’s


equation in modifying the N–S–F equations is noted to constitute a molecular counterpart of Öttinger’s phenomenological
fluctuation addition to the GENERIC friction matrix. Together, these two theories collectively recognize the need to
33
EC

address multiple- rather than single-encounter collisions between a test molecule and its neighbors when formulating
physically satisfactory statistical–mechanical theories of irreversible transport processes in gases. Overall, the results of the
35 present work implicitly support the unorthodox view, implicit in the GENERIC scheme, that the translation of Newton’s
discrete mass-point molecular mechanics into continuum mechanics, the latter as embodied in the Cauchy linear
R

37 momentum equation of fluid mechanics, cannot be correctly effected independently of the laws of thermodynamics. While
Öttinger’s modification of GENERIC necessitates fundamental changes in the foundations of fluid mechanics in regard to
R

39 momentum transport, no basic changes are required in the foundations of linear irreversible thermodynamics (LIT)
beyond recognizing the need to add volume to the usual list of extensive physical properties undergoing transport in single-
O

41 species fluid continua, namely mass, momentum and energy. An alternative, nonGENERICally based approach to LIT,
derived from our findings, is outlined at the conclusion of the paper. Finally, our proposed modifications of both Cauchy’s
C

43 linear momentum equation and Newton’s rheological constitutive law for fluid-phase continua are noted to be mirrored by
N

45
U

47
Tel.: +1 617 253 6687; fax: +1 617 258 8224.
49 E-mail address: hbrenner@mit.edu.

51 0378-4371/$ - see front matter r 2006 Published by Elsevier B.V.


doi:10.1016/j.physa.2006.03.066
PHYSA : 9965
ARTICLE IN PRESS
2 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 counterparts in the literature for solid-phase continua dating back to the classical interdiffusion experiments of Kirkendall
and their subsequent interpretation by Darken in terms of diffuse volume transport.
3 r 2006 Published by Elsevier B.V.

Keywords: ’; ’; ’
5

7
1. Introduction
9
1.1. Preface
11
Previous contributions [1–3] relevant to the issues discussed here in have suggested, based primarily upon
13 conflicts noted between tracer- and hypothetical mass–velocity experiments, that revisions were needed to the
Navier–Stokes–Fourier (N–S–F) equations of viscous fluid mechanics—certainly in the case of single-
15 component fluids. Explicitly, in several key constitutive relations appearing in the N–S–F equation set it was
proposed, inter alia, that the fluid’s mass velocity vm appearing therein be replaced by the fluid’s volume
17 velocity vv [4]. The arguments underlying the proposed revisions were based, in part, upon our unorthodox
interpretation [5] of experimental data pertaining to thermophoretic particle motion in gases, when considered
19 in conjunction with Kogan et al.’s [6–9] and Bobylev’s [10] re-ordering of the Maxwell [11]—Burnett [12]

F
thermal stress contribution to the viscous stress tensor at small Knudsen numbers, Kn51. These revealed such

O
21 stresses to be of the same order in Kn as the classical Navier–Stokes stress contributions, at least for Reynolds
numbers of order one and Mach numbers small compared with unity. Prior to this reassignment, thermal
23
O
stresses were regarded as being of a noncontinuum nature in the Chapman–Enskog [13] hierarchical expansion
of solutions of the Boltzmann equation in powers of Kn. Our original proposal [1–3] emphasized gases. This
PR
25 owed to a lack at the time of convincing experimental evidence sufficient to demonstrate the applicability of
the theory to liquids. Subsequently, however, we were able to offer an interpretation of experimental data
27 involving thermal diffusion in binary liquid mixtures [14,15] that supported the applicability of our proposed
N–S–F revisions to liquids as well.
D

29 It proved unnecessary in our original N–S–F modification proposal to address the obvious question of a
further revision to these equations regarding the viability of Euler’s constitutive equation for the fluid’s specific
TE

31 momentum density v [16], including related kinetic energy density and ‘‘work-velocity’’ issues [3]. (The symbol
v refers here to the velocity appearing in both the local and convective inertial terms in the Navier–Stokes
33 equations.) That this was unnecessary owed to the smallness of the Reynolds numbers [17] encountered in the
EC

thermophoretic and thermal diffusion experiments cited above, as well as in the Knudsen number re-ordering
35 of the thermal stress contributions [6–10]. It was, however, pointed out in Appendix B of Ref. [3] that if one
accepts as being correct the standard balance equations for momentum and energy found in textbooks [18–21],
R

37 while concurrently adopting the unconventional constitutive equations proposed in Ref. [3] for the revised
forms of Fourier’s law of heat conduction and Newton’s law of viscosity, the specific momentum density v
R

39 would necessarily have to be equal to the fluid’s mass velocity vm , the latter being the velocity appearing in the
continuity equation (cf. Eq. (1)).
O

41 Although not explicitly pointed out in connection with the ‘‘proof’’ set forth in Ref. [3] of the apparent
C

viability of the standard Eulerian constitutive relation v ¼ vm for the fluid’s specific momentum density, it is
43 implicit in the concomitant use of the nonstandard form of Newton’s viscosity law (cf. the union of Eq. (44)
N

with (56)) (in which the gradient rvv of the fluid’s volume velocity [4] now appears in place of its standard
45 mass velocity counterpart rvm Þ that nonpositive-definite dissipative terms, namely the ‘‘mixed’’ terms 2Zrvm :
U

rvv and ZB ðr.vm Þðr.vv Þ [22], will appear in the Clausius–Duhem inequality (cf. Eq. (13)) for the local rate of
47 irreversible entropy production. Here, Z and ZB are the respective shear and bulk viscosities. The consequent
conflict between the apparent dictates of rational mechanics and irreversible thermodynamics manifested by
49 this lack of definitive positivity leads us to continue to regard the specific momentum density issue as an open
question despite the purported ‘‘proof’’ to the contrary offered in Ref. [3]. As a result of this fundamental
51 dichotomy between the respective predictions of continuum mechanics and the second law of thermo-
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 3

1 dynamics, the possibility must be entertained that, to quote Öttinger [23]: ‘‘Something is missing’’ from the
currently accepted energy and entropy transport equations of nonequilibrium irreversible thermodynamics
3 [19–21]; that is, in order to heal the breach engendered by this lack of positivity, additional terms, presumably
currently missing from these equations, need to be present in the constitutive expressions for the diffusive
5 fluxes of internal energy and entropy appearing therein.
In this context it is pertinent to note that prior to the emergence of this core conflict we had earlier [1]
7 opined, so to speak, in favor of thermodynamics over mechanics [24] in resolving apparent conflicts between
the two, by supposing that v ¼ vv rather than v ¼ vm so as to render the irreversible entropy production rate
9 nonnegative. However, substantive theoretical justification for that fundamental change in the constitutive
expression governing the specific momentum density appearing in the Cauchy linear momentum equation was
11 lacking at the time (although we nevertheless regarded our Lagrangian tracer velocity interpretation [1–3] of
thermophoretic particle motion as offering convincing experimental evidence in support of the possible change
13 in v from vm to vv Þ. This background sets the stage for the subsequent theoretical analysis, which purports to
resolve the momentum density issue in favor of the ‘‘mechanico-thermodynamic’’ relation v ¼ vv . This is
15 accomplished by adding fundamental fluctuational contributions [23] to the standard [19–21] energetic and
entropic balance equations of nonequilibrium irreversible thermodynamics, following a recent proposal to this
17 effect by Öttinger.
In his book [23], ‘‘Beyond Equilibrium Thermodynamics’’ (hereafter referred to as BET), Öttinger proposed
19 incorporating fluctuations into the friction matrix appearing in GENERIC theory. This latter nontraditional

F
feature, manifested by the appearance of a new phenomenological coefficient a0 therein, was not explicitly

O
21 included in earlier versions [25,26] of that theory (although neither was it explicitly ruled out). It is this
inclusion in the GENERIC friction matrix M of Öttinger’s previously ‘‘Something is missing’’ fluctuation
23
O
terms that resolves the conflict, described earlier, between continuum mechanics and thermodynamics, and
which will be seen to lead rationally to the relation v ¼ vv . Explicitly, Öttinger’s extension of GENERIC
PR
25 theory harmoniously unites our respective views of momentum density (while also confirming other elements
of our theory). A related paper by the present author [27] reinforces this unification of our view of momentum
27 transport with that of Öttinger by offering an independent argument, due to Klimontovich [28–30], with
respect to the viability of the constitutive relation v ¼ vv . In particular, Klimontovich’s unorthodox
D

29 modification of the collisional term in Boltzmann’s equation [13], involving the addition thereto of a physical-
space self-diffusion-like contribution, is counterpart to Ottinger’s fluctuational addition to GENERIC. Both
TE

31 additions, in turn, are formally equivalent to our addition of diffuse volume transport [1–4] to conventional
theories of transport phenomena [18].
33 The independent theories of Öttinger [23] and Klimontovich [28–30], which lead to identical results in their
EC

common domain of validity (namely, rarefied gases), were created to rectify what each regarded as the
35 fundamental failure of existing continuum-mechanical theories—especially those derived from the Boltzmann
equation [13]—to incorporate single-component fluctuations and, hence, the notion of Brownian motion into
R

37 their respective foundations [31]. (Further details with respect to this point of view are discussed in Sections 8.1
and 8.2.) However, the arguments underlying their proposed changes were purely philosophical, in the sense
R

39 that neither explicitly invoked experimental data, nor appealed to already existing theories to support their
claims with respect to the perceived incompleteness of current theories of irreversible thermodynamics and
O

41 Boltzmann-based gas-kinetic statistical mechanics. In that respect, our conclusions, if accepted as being
C

correct, offer experimental and theoretical evidence in support of both of their theories, whose common results
43 for rarefied gases coincide. Conversely, subject to this same caveat, their theories should be regarded as
N

supporting our earlier empirical proposal [1–3] for modifications in the N–S–F equation set, especially
45 including the specific momentum density constitutive equation v ¼ vv , as well as the energetic and entropic
U

consequences stemming therefrom.


47 Issues of entropy production and irreversibility are a matter of concern when attempting, as did Boltzmann
[13], to translate discrete Newtonian mass-point microscopic molecular mechanics into a rational macroscopic
49 continuum-mechanical analog thereof, namely the Cauchy linear momentum equation of fluid mechanics
[18–21]. However, according to the generally accepted view of this microscopic ! macroscopic transition
51 scheme [13], concern about its consistency with the notion of irreversibility, as embodied in the second law of
thermodynamics, currently surfaces only after the fact. This secondary ‘‘after the fact’’ role ascribed to the
PHYSA : 9965
ARTICLE IN PRESS
4 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 second law is evidenced by the subsequent appearance, only after Boltzmann’s transport equation was already
fully formulated, of his celebrated H-theorem, aimed at reconciling the consistency of his purely mechanically
3 derived reversible formula for calculating the molecular distribution function with the thermodynamic notion
of irreversibility. In short, current thinking holds that while the continuum version of Newtonian mechanics
5 impacts upon the foundations of irreversible thermodynamics, the reverse is not true; that is, the field of
irreversible thermodynamics is generally (although not universally [32]) believed to be without impact upon
7 ‘‘pre-constitutive’’ Newtonian dynamics-based continuum mechanics—explicitly upon the Cauchy linear
momentum equation prior to contemplating the form of the constitutive equation for the stress tensor
9 appearing therein. (Irreversible thermodynamics does, however, impact upon ‘‘post-constitutive’’ continuum
mechanics—for example, as a consequence of restrictions imposed by the Clausius–Duhem entropy
11 production inequality [19–21] upon the allowable class of constitutive expressions for the form ascribed to
the deviatoric portion of the stress tensor, the latter representing the diffuse momentum flux.)
13 As regards our ‘‘pre’’ and ‘‘post’’ terminology, we are referring here to the accepted view, dating back 250
years to Euler [33–35], that the identification v ¼ vm is not a constitutive relation but is, rather, a fundamental
15 physical truth emanating directly from the implicit a priori assumption that Newton’s laws of motion, known
to be valid for a discrete ponderable body of mass M, can unquestionably also be applied to a so-called
17 material fluid domain of mass M moving within a fluid continuum—at least in circumstances where M is
differential in magnitude. Indeed, we are aware of only a very few circumstances [36, p. 196 (but see also p. 28),
?
19 37] in which this apparent ‘‘fact,’’ namely that of the viability of the relation v ¼ vm , has been questioned.

F
However, because a material domain does not generally consist permanently of the same molecules, but only

O
21 of the same net amount of mass [38,39], discrete-body mechanics [40] based upon regarding individual
molecules as point masses subject to elementary action–reaction mechanical Newtonian laws cannot,
23
O
indiscriminately, be applied to such domains. Of course Euler [33], the father of rational fluid mechanics, was
unaware at the time of his foundational work in 1755 of the existence of molecules—either static, as in
PR
25 Dalton’s subsequent chemical theory, or mobile, as in Clausius–Maxwell–Boltzmann’s century-later gas-
kinetic theory. Hence, he was obviously unaware of the hidden constitutive assumption implicit in the relation
27 v ¼ vm . As such, a material domain must necessarily have been viewed by Euler as a continuous body of
permanent material integrity insofar as its underlying constitution was concerned. Today we know better.
D

29 Accordingly, as regards continuum mechanics, and in particular the correctness of the Cauchy linear
momentum equation, the relation v ¼ vm needs to be regarded as a tentative constitutive assumption, rather
TE

31 than as an identity, thereby subject to empirical (experimental) verification and, if possible, theoretical
justification.
33 Given these remarks, it remains to be established whether the seemingly unequivocal separation advocated
EC

in the literature between the statistical mechanics and thermodynamics of continuum models of large
35 multiparticle systems composed of materially identical particles is viable. GENERIC theory argues forcefully
against such separability. The current nontraditional version of GENERIC theory appearing in Öttinger’s
R

37 book [23] represents an attempt at a systematic synthesis of the two fields. It constitutes a broad general
theory, equally applicable to both single-component gases and liquids. In what follows, we point out that
R

39 Öttinger’s theory provides a rational basis for our proposed diffuse volume-based modifications [1–3] of the
N–S–F equations, including, most prominently, changes to the accepted constitutive formula for the specific
O

41 momentum [41].
C

43 1.2. Outline of the paper


N

45 Section 2 reviews the pertinent equations of Öttinger’s extended GENERIC theory [23] for the present
U

single-component fluid case, without invoking any constitutive relations for the various physical properties
47 appearing therein. These equations embody the balance equations for mass, linear momentum, energy and
entropy as posed by the GENERIC scheme. They differ in substance from the standard balance equations of
49 linear irreversible thermodynamics (LIT) found in textbooks [19–21], to which the former reduce in
circumstances where the difference v  vm between the fluid’s specific momentum density v and its mass
51 velocity vm vanishes. It is not that LIT is conceptually wrong, but rather that the present formulation of the
subject fails to recognize the role played by the diffuse transport of volume. Section 3 sets forth the general
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 5

1 constitutive equation emanating from GENERIC for the difference v  vm in velocities, while temporarily
leaving open issues pertaining to the constitutive equations for the heat flux vector and deviatoric stress
3 dyadic. Based solely upon our fundamental hypothesis that v ¼ vm if, and only if, the fluid is
‘‘incompressible,’’ i.e., r ¼ const. throughout the fluid, it is demonstrated that this is tantamount to requiring
5 that the velocity difference be of the constitutive form v  vm ¼ Kr ln r, where the coefficient K is
nonnegative-definite. Since this unorthodox momentum density constitutive result holds independently of the
7 constitutive equations governing the heat flux and stress tensor, it necessarily applies to any fluid continuum,
gaseous or liquid, irrespective of the fluid’s rheological and thermal properties. Section 4 discusses the
9 thermodynamic consequences stemming from this constitutive relation, including the fact that the diffusive
heat flux generally differs from the diffusive internal energy flux, and that the entropy flux is not simply equal
11 to the heat flux divided by the thermodynamic temperature T in single-component fluids. Both conclusions are
at odds with the current tenets of (linear) irreversible thermodynamics.
13 Section 5 uses molecularly derived results emanating from Burnett’s solution of the Boltzmann equation to
establish that the velocity-difference coefficient K is equal to the gas’s thermometric diffusivity a in the case of
15 rarefied gases. In turn, in Section 6, comparison of the preceding dynamical result, namely v ¼ vm þ ar ln r,
with the known, purely kinematical, formula [4] for the volume velocity of a liquid or gas, vv ¼ vm þ ar ln r,
17 leads to the conclusion that the fluid’s specific momentum density v is identical to its volume velocity vv . Since
the fluid’s volume velocity has already been shown elsewhere [2,3,15] to be equal to its tracer or Lagrangian
19 velocity, this equality serves, in turn, to demonstrate equality of the fluid’s specific momentum with its tracer

F
velocity, certainly in the case of rarefied gases. Section 7 offers justification for three key constitutive

O
21 assumptions which it was necessary for us to make in the course of evaluating the two undetermined
parameters, a0 and D0 , appearing in Ottinger’s nontraditional version of GENERIC [23]. Identifying these two
23
O
parameters was necessary in order to obtain definitive results, against which experimental data could be
compared. Section 8 broadly discusses the implications of the present theory for LIT as a whole, while
PR
25 stressing the fact that the consequences of the nontraditional GENERIC addition to the friction matrix
appearing therein are explicitly manifested in the Onsager force–flux LIT scheme for single-component fluids
27 by the addition thereto of a volume flux, above and beyond the usual heat and momentum fluxes. In turn, the
force–flux basis of LIT suggests a simple physical origin (i.e., one independent of the formal constitutive
D

29 friction matrix approach of GENERIC) for the apparently universal relation v ¼ vm þ Kr ln r, which is
believed to be applicable to all gases and liquids, with the value of the phenomenological coefficient K
TE

31 depending upon the particular physical application being addressed. This applicability extends to
multicomponent fluids [4], in which circumstance K ¼ D, with D the binary diffusion coefficient D for
33 isothermal two-component fluid mixtures. It is further pointed out (Section 8.4) that, based on the recent
EC

‘‘proper coarse-graining’’ work of Öttinger, an alternative may exist to our fundamental incompressibility
35 hypothesis in Section 3, although in the special, but important, case of ideal gases, the conclusions derived
therefrom are indistinguishable from those obtained from our hypothesis. Also noted (Section 8.5) is the fact
R

37 that issues of momentum transport and volume diffusion-induced stress, comparable to those issues identified
here for gases and liquids, also exist in the case of solids. Explicitly, based on experimental data dating back to
R

39 the classical interdiffusion experiments of Kirkendall and their subsequent interpretation by Darken, it is
pointed out that ‘‘two-velocity’’ modifications of the Cauchy linear momentum equation as well as the
O

41 Hooke’s law-type stress–strain constitutive equation appearing therein for solid solutions (alloys) have been
C

proposed in the literature. Just as in the case of fluid phases, these unorthodox dynamical solid-phase notions
43 are attributed to diffuse volume transport. Such transport in solids is ascribed largely to atom–vacancy
N

exchanges occurring within the lattice during the diffusional process, and, to a lesser extent, by comparable
45 atom–atom lattice-point interchanges when the respective molecular masses of the interdiffusing species differ.
U

Finally, Section 9 offers a summary and overview of the essence of our findings.
47

49 2. Review of Öttinger’s nontraditional GENERIC theory

51 GENERIC (general equation for the non-equilibrium reversible–irreversible coupling) theory, as set forth in
BET [23], offers a derivation of the following nontraditional trio of mass, linear momentum and energy
PHYSA : 9965
ARTICLE IN PRESS
6 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 transport equations for the case of single-component fluids (both gases and liquids) in the absence of external
body-force fields (Eqs. (2.78)–(2.80) of BET):
3 (i) Mass transport (continuity equation):
qr
5 þ r.ðrvm Þ ¼ 0. (1)
qt
7 (ii) Momentum transport (Cauchy linear momentum equation):
qðrvÞ
9 þ r.ðrvm vÞ ¼ r.P, (2)
qt
11 where
P ¼ Ip þ T (3)
13
is the pressure tensor (using the usual fluid-mechanical sign convention for stress, rather than the opposite
convention used in BET), with I the dyadic idemfactor and T the symmetric deviatoric or viscous stress tensor.
15
^ [16] by the
The ‘‘velocity’’ v appearing in Eq. (2) is defined in terms of the fluid’s specific momentum density m
^
relation v:¼m. (BET writes Eq. (2) in a form involving the momentum per unit volume, the latter identified
17
therein by the symbol M, equivalent to our rv.)
(iii) Energy transport:
19

F
qu
þ r.ðvm uÞ ¼ r.½jq  ðv  vm Þðra0  uÞ þ pu , (4)

O
21 qt
where u is the volumetric (i.e., per unit volume) internal energy density (designated by the symbol  in BET), jq
23 is the heat flux vector, and
O
PR
25 pu ¼ P : ðrvÞT (5)
is the temporal rate of production of internal energy per unit volume, wherein the superscript T denotes the
27 transpose operator. In addition, a0 is an unconstrained phenomenological parameter appearing in extended
GENERIC theory (for which the symbol a is used in BET). This parameter is to be determined constitutively
D

29 by some scheme, be it theory, simulation or otherwise, but ultimately subject to confirmation by experiment.
Determination of a0 , which will be seen to physically represent an Onsager coefficient serving to couple
TE

31 together the respective processes of diffuse internal energy and volume transport, is one of the central goals of
the present analysis.
33
EC

In addition to the preceding equations, we shall later require the nontraditional, fluctuation-based
GENERIC entropy transport equation, given in BET as Eq. (2.83), namely:
35 (iv) Entropy transport:
q 
R

37 qs j p þ u  ra0
þ r.ðvm sÞ ¼ r. þ ðv  vm Þ þ ps , (6)
qt T T
R

39 where
O

41 r2 jv  vm j2 1 1
ps ¼ þ jq .r þ T : ðrvÞT . (7)
C

D0 T T
43 Here, s and ps are, respectively, the volumetric entropy density and temporal rate of irreversible entropy
N

production.
45
U

It is important to note that neither the symbols pu and ps , respectively, appearing in Eqs. (4)–(5) and (6)–(7)
nor the respective volumetric production-rate interpretations that we have assigned to them appear in the
47 original BET equations cited. As such, justification for these interpretations is required. This is presented in
Section 7 at the conclusion of the analysis. For the time being we simply pursue the consequences of these
49 interpretations (cf. Eqs. (11) and (12)).
It will prove convenient in what follows to define the velocity difference vector:
51
j:¼v  vm . (8)
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 7

1 In the next section we will propose a constitutive equation for j. The system of equations described in
the preceding paragraphs represents the entire set of single-component GENERIC equations of which we
3 will avail ourselves in what follows. Of course, in order to understand the origin of these equations one
needs to master the basic reversible–irreversible model of transport processes underlying the GENERIC
5 scheme, as set forth in BET. This differs from the more common convective–diffusive model of such
phenomena [18].
7 Before proceeding to a discussion of the constitutive equations required to complete the governing set of
equations (1)–(7), it will prove convenient to re-express the preceding energy and entropy transport equations
9 in more conventional terminology. In terms of the general notation set out in Ref. [4] consider the transport of
any extensive physical property, with C, say, the amount of the property contained in a volume V. The
11 Eulerian balance equation governing transport of this extensive property is then
qc
13 þ r . n c ¼ pc , (9)
qt
15 where c ¼ C=V is the volumetric density of the property, nc is the Eulerian flux density of the property and pc
is the volumetric rate of production of the property.
17 The (total) flux density
nc ¼ vm c þ jc (10)
19

F
appearing above consists, respectively, of a convective portion vm c carried by the moving fluid mass and a
diffusive or nonconvective portion jc . In terms of this notation, the continuity equation retains its usual form

O
21
(1) since the law of conservation of mass together with the definition vm :¼nm =r of the mass velocity requires
that the diffuse mass flux density be zero: jm ¼ 0. On the other hand, the internal energy and entropy transport
23
equations adopt the respective forms O
PR
25 qu
þ r.ðvm uÞ ¼ r.ju þ pu (11)
qt
27
and
D

29 qs
þ r.ðvm sÞ ¼ r.js þ ps . (12)
qt
TE

31 In the latter equation the second law of thermodynamics requires satisfaction of the generic Clausius–Duhem
inequality [19–21],
33
EC

ps X0. (13)
35 Comparison of Eq. (11) with (4) furnishes the relation
ju ¼ jq  jðra0  uÞ. (14)
R

37
Similarly, comparison of Eq. (12) with (6) gives
R

39 1 q
js ¼ ½j þ jðp þ u  ra0 Þ. (15)
O

T
41
Elimination of jq between the latter pair of equations yields the expression
C

43 ju ¼ Tjs  pj. (16)


N

It is important to note that this relation is strictly a consequence of the principles of GENERIC theory,
45
U

independent of any constitutive relations (although more general nonlinear and nonlocal possibilities than (16)
exist within the GENERIC framework). However, its validity hinges critically upon the interpretations we
47 have assigned to the symbols pu and ps appearing in Eqs. (4)–(5) and (6)–(7), since it was those interpretations
which resulted in Eqs. (14) and (15), and which thereby led to Eq. (16).
49 The significance of Eq. (16) will subsequently be discussed in the context of the combined first and second
laws of thermodynamics,
51
dU ¼ T dS  p dV , (17)
PHYSA : 9965
ARTICLE IN PRESS
8 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 for a system of fixed mass. Explicitly, the obvious analogy existing between Eqs. (16) and (17) suggests that a
subscript v should be affixed to the symbol j in Eq. (16), with the resulting quantity jv identified as being the
3 diffusive flux density of volume [4]. The latter is a physically defined, purely kinematical quantity, related to
the fluid’s volume velocity vv discussed in the Introduction through the relation vv  vm ¼ jv . In turn,
5 comparison of the latter with Eq. (8) would then give rise immediately to the constitutive relation v ¼ vv , thus
identifying the fluid’s specific momentum density with its volume velocity. It is this fact, namely that j ¼ jv ,
7 which we propose to formally demonstrate, or at least make plausible, in what is to follow.
Before leaving this section, we note that all of the above BET transport equations are deliberately expressed
9 in a space-fixed, Eulerian, format rather than a (moving) ‘‘body-fixed’’ material format. This avoids the
ambiguous notion of so-called ‘‘material differentiation following the motion’’ of a material body, such
11 ambiguity arising from the issue of whether the fluid’s mass velocity vm , which is a (normalized) flux, is indeed
synonymous with the notion of the fluid’s Lagrangian or tracer ‘‘motion’’ through space [2,3]. The two classes
13 of description, Eulerian and material, are connected through the relation

15 qc Dm c^
þ r.ðvm cÞ ¼ r , (18)
qt Dt
17 ^ ¼ c=r is the specific density of the extensive property C, and
where c
Dm q
19 ¼ þ vm . r (19)
Dt qt

F
is the material derivative.

O
21

23 3. Constitutive relation for the specific momentum density v


O
PR
25 According to Öttinger’s fluctuation-based GENERIC scheme [23], the most general possible constitutive
equation for the velocity difference (8), applicable to both single-component gases and liquids, is (BET, Eq.
27 (2.82))
  
D0  p  1
D

0
29 j¼ 2 r  ðra  uÞr . (20)
r T T
TE

31 The nonnegative-definite diffusion-like phenomenological coefficient [23]


D0 X0 (21)
33
EC

appearing above quantifies the intensity of the fluid’s molecularly based fluctuations [42]. Apart from having
35 to satisfy the preceding inequality, D0 , like a0 , represents an otherwise unconstrained phenomenological
parameter, to be chosen in a manner such as to match experimental data and/or accommodate detailed
R

37 theories, molecular or otherwise, of the pertinent phenomena. Such phenomenological choices, of course,
include the trivial possibility that D0 ¼ 0, and hence j ¼ 0, corresponding as a consequence of Eq. (8) to Euler’s
R

39 conventional momentum density constitutive relation v ¼ vm .


Eq. (20) can be rearranged so as to express j in terms of rp and rr as follows: in place of the parameter a0 it
O

41 proves useful to introduce another parameter, g, the latter defined by the expression
C

ra0  u ¼ p  gT, (22)


43
N

so as to furnish the relation


45 D0
U

j¼ ðrp  grTÞ.
r2 T
47
But, from the single-component fluid equation of state relating T, p and r, we have that
49 rT ¼ ðqT=qpÞr rp þ ðqT=qrÞp rr, whence the above becomes
("   # )
D0 qT 1
51 j¼ 2 1g rp þ g r ln r ,
r T qp r b
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 9

1 where b is the coefficient of thermal expansion:


 
1 qr
3 b¼ . (23)
r qT p
5 In place of g it now proves convenient to instead introduce a related parameter K, defined by the expression

7 br2 T
g¼K , (24)
D0
9 whereupon it follows that
"   #
11 D0 qT
j ¼ 2  Kb rp þ Kr ln r. (25)
r T qp r
13
Thus far, no assumptions have been made beyond the basic nontraditional GENERIC transport relations set
forth in Section 2, including the general constitutive assumption for j embodied in Eq. (20). For later reference
15
we note that we have, in effect, merely replaced the original parameter a0 appearing in BET by the new
parameter K, the relationship between them being
17  
0 br2 T 1
a ¼ uþpK . (26)
19 D0 r

F
We now introduce what appears to us to be the physically reasonable hypothesis that when the fluid is

O
21
‘‘incompressible,’’ namely when r ¼ const, and only then, the two velocities, v and vm , appearing in the basic
BET transport equations of Section 2 should coalesce into a single entity, so that v ¼ vm . From Eq. (8) this is
23
equivalent to the relation O
PR
25 j ¼ 0 when r ¼ const ði.e., when rr ¼ 0Þ. (27)
The latter relation leads to the dual requirements that: (i) the bracketed pressure-gradient coefficient in Eq.
27 (25) needs to be identically zero; and (ii) the diffuse flux j must obey the constitutive relation
D

29 j ¼ Kr ln r. (28)
In the subsequent discussion of Section 7 it will be pointed out that the further, second law-based,
TE

31 Clausius–Duhem requirement, KX0, in the above relation (see Eq. (46)) is formally equivalent to requiring
that Fourier’s law (cf. Eq. (42)) be satisfied. This surprising linkage of Eq. (28), the latter involving a density
33 gradient, with Fourier’s law, the latter involving an independent temperature gradient, appears surprising.
EC

This connection will be seen (in Section 7) to arise from the fact that a0 represents a coupling coefficient which,
35 as a consequence of Onsager’s reciprocal theorem, cannot be arbitrarily chosen if one insists that Fourier’s law
of heat conduction must be satisfied under all circumstances (i.e., independently of the density gradient, and
R

37 hence of the pressure gradient in single-component systems).


By the chain rule for partial differentiation [43] in conjunction with definition (23), we have that
R

39 bðqT=qpÞr ¼ kT , where
 
O

41 1 qr
kT ¼ (29)
r qp T
C

43 is the coefficient of isothermal compressibility, assumed to be nonnegative for both liquids and gases:
N

45 kT X0. (30)
U

The latter inequality constitutes a requirement for stability of the fluid continuum (as discussed in Section 7.6;
47 cf. Eq. (101)).
With use of the preceding relations, the vanishing of the bracketed term in Eq. (25) necessitates that
49
D0 ¼ Kr2 TkT . (31)
51 This expression serves to relate D0 to the phenomenological coefficient K appearing in Eq. (28). In turn, from
Eq. (24), this makes
PHYSA : 9965
ARTICLE IN PRESS
10 H. Brenner / Physica A ] (]]]]) ]]]–]]]

 
1 b qp
g¼ ¼ , (32)
kT qT r
3
where we have again used the chain rule [43]. Upon inserting Eq. (31) into (26), subsequent use of Eq. (32)
5 furnishes the relation
"   #
0 1 qp
7 a ¼ uþpT . (33)
r qT r
9 The latter is, in effect, a constitutive expression serving to relate the GENERIC parameter a0 to the fluid’s
local equilibrium thermodynamic (i.e., nontransport) properties.
11 For the moment we leave open the choice of the phenomenological coefficient K appearing in the
constitutive equation (28) for the velocity difference j. This same coefficient serves to determine the parameter
13 D0 from Eq. (31). Hence, as a consequence of the hypothesis embodied in Eq. (28), of the original two
phenomenological parameters D0 and a0 appearing in the fundamental GENERIC constitutive equation (20),
15 only the parameter D0 remains yet to be determined, as presently manifested in the coefficient K. Of course, as
earlier mentioned, this includes the possibility that K ¼ 0, and hence D0 ¼ 0, corresponding to Euler’s
17 traditional momentum density relation, v ¼ vm .
It needs to be stressed that Eq. (33), expressing Öttinger’s GENERIC fluctuation parameter in terms of the
19 fundamental equilibrium properties of the fluid, derives directly from the constitutive assumption (27), whose

F
validity can be confirmed only by demonstrating that the physical results issuing therefrom accord either

O
21 directly with experiment or else with an accepted theory. As such, from this point on, the validity of all
subsequent relations in this paper derived indirectly from this assumption is to be regarded as tentative,
23
O
subject to verification. This cautionary emphasis is recapitulated and reviewed near the conclusion of the
paper, in Section 8.4. However, was Eq. (27) to be proved inconsistent with experiment, the subsequent effort
PR
25 expended in our paper would, nevertheless, not have been fruitless. Rather, all that would be required to
rectify the situation would be to simply carry a0 along as a free parameter, beginning with Eq. (20), and with a0
27 thus appearing explicitly in the subsequent LIT equations of Section 7. This issue is briefly discussed in Section
8.4.
D

29
TE

31 4. Consequences of the ‘‘incompressibility’’ hypothesis

33 4.1. Internal energy flux


EC

35 Introduction of Eq. (33) into (14) gives, for the internal energy flux,
"   #
R

37 q qp
ju ¼ j  j p  T . (34)
qT r
R

39
This expression can be re-formulated in more physical terms by noting from the first and second laws of
O

thermodynamics that for a single-component fluid [44]


41
   
C

qp
43 du^ ¼ c^v dT þ T  p d^v. (35)
qT v^
N

45 The caret atop a symbol denotes a specific (i.e., per unit mass) density, so that, for example, u^ ¼ u=r denotes
U

the specific internal energy and v^ ¼ 1=r the specific volume. Consequently, we have the thermodynamic
47 identity
     
qp qu^ qU
49 pT ¼  , (36)
qT v^ q^v T qV T;M
51 where, with M ¼ rV the mass contained in a volume V ¼ M v^, the extensive internal energy contained in V is
^ It follows that Eq. (34) is equivalent to the relation
U ¼ M u.
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 11

1 ju ¼ jq þ jðqU=qV ÞT;M . (37)


In the context of the possibility, discussed following Eq. (17), that j ¼ jv , the second term of Eq. (37) will, in
3
Section 7.2, be seen to possess a well-defined physical significance.
Eq. (36) shows the internal energy of an ideal gas to be independent of its volume, i.e.,
5
^ vÞT ¼ 0. In such circumstances it follows from Eq. (37) that
ðqU=qV ÞT;M  ðqu=q^
7 ju ¼ jq (ideal gases). (38)
This relation also holds for any single-component fluid, either gas or liquid, whose specific internal energy is a
9
function only of its temperature.
11
4.2. Entropy flux
13
From Eqs. (33) and (15) one obtains the following expression for the entropy flux:
15  
jq qp
js ¼ þ j . (39)
T qT v^
17
Alternatively, with use of the thermodynamic identity [44]
19      

F
qp q^s qS
¼  , (40)
qT v^ q^v T qV T;M

O
21
it follows that
23
js ¼
jq
þj
 
qS
.
O (41)
PR
25 T qV T;M
In the context of the possibility, discussed following Eq. (17), that j ¼ jv , the second term of Eq. (41) will later
27
be seen to possess a well-defined physical significance.
D

29
4.3. Entropy production
TE

31
In order for the present theory to be viable, the constitutive relations entering into the entropy production
33 rate (7) must be of such a nature that ps satisfies the Clausius–Duhem inequality (13). Following BET, we
EC

suppose that the heat flux vector jq appearing in Eq. (7) obeys Fourier’s law of heat conduction,
35
jq ¼ krT, (42)
R

37 with
R

kX0 (43)
39
the thermal conductivity. Indeed, rather than simply assuming the validity of (42), we will, in Section 7.5,
O

41 subsequently derive this equation as an immediate consequence of the second law of thermodynamics when
C

the latter is considered in conjunction with Onsager’s reciprocal relation. We also suppose that the deviatoric
43 stress T is given, for example, by the usual rheological constitutive equation for a viscous Newtonian fluid,
N

namely
45
U

T ¼ 2Zrv þ ZB Ir.v, (44)


47 where ZX0 and ZB X0 are the fluid’s respective shear and bulk viscosities. In as much as T represents the
diffuse momentum flux density, it appears appropriate to suppose that the symbol v appearing in the
49 preceding expression refers to the fluid’s specific momentum density [45] rather than to, say, its mass velocity
vm .
51 Upon introducing the above pair of constitutive equations jointly with Eqs. (28) and (31) into (7) we find
that
PHYSA : 9965
ARTICLE IN PRESS
12 H. Brenner / Physica A ] (]]]]) ]]]–]]]

 
1 1 K 2 k 2 2
ps ¼ .
ðrrÞ þ ðrTÞ þ 2Zrv : rv þ ZB ðr vÞ . (45)
T kT r2 T
3
Given the nonnegative-definite algebraic signs of kT , k, Z and ZB , Eq. (45) shows that the Clausius–Duhem
inequality (13) will be satisfied provided that
5
KX0. (46)
7 In view of Eq. (31), the latter inequality is equivalent to that requiring satisfaction of the previously stipulated
inequalities (21) and (30).
9 We have referred to the Newtonian rheological constitutive expression (44) as being ‘‘standard.’’ However,
this terminology is somewhat ambiguous since the velocity v appearing therein is often implicitly understood
11 in the literature to be the mass velocity vm appearing in the continuity equation (1), rather than representing
the specific momentum density v. (This implicit vm assumption stems from the fact that the symbol for velocity
13 first arises in courses in fluid mechanics in the context of deriving the continuity equation. Only later is this
mass-based symbol identified with the Lagrangian notion of the movement of an ‘‘object’’ through space in a
15 trajectory sense, the latter being precursive to the association of this symbol with Newtonian dynamics, and
hence its role as a momentum density.) With few exceptions [36, p. 196 (but see also p. 28), 37], the possibility
17 that a difference might exist between the fluid’s momentum density v and its mass velocity vm has not generally
been recognized. In a similar vein, referring to Fourier’s law (42) as ‘‘standard’’ is equally ambiguous, since in
19 the past it has not generally been recognized, certainly not in single-component systems [46], that a difference

F
might exist between the diffuse internal energy flux ju and the heat flux jq . Nevertheless, according to Eq. (34),

O
21 a difference does exist generally, except in the case of ideal gases, where Eq. (38) applies.

23
O
5. Confirmation of Eq. (28) and identification of the phenomenological coefficient K
PR
25 Apart from our acceptance of the nontraditional GENERIC formulation of irreversible thermodynamics
set forth in BET [23] as being physically correct, together with our subsequent constitutive assumptions
27 implicit in the production terms (5) and (7), the only additional assumption we have made thus far is explicitly
embodied in the hypothesis (27), namely the assumption that v ¼ vm when the fluid is ‘‘incompressible.’’ (Of
D

29 course, we have also supposed the applicability of Fourier’s law and Newton’s rheological law, Eqs. (42) and
TE

(44), although the latter equation is not critical, whereas the former is.) There remains only the task of relating
31 the phenomenological coefficient K appearing in Eq. (28) to the physical properties of the fluid, and of
subsequently confirming that the resulting expression for K satisfies inequality (46).
33
EC

In the absence of a microscopic theory of the pertinent phenomena this task is normally assigned to
experiment, namely that of empirically confirming the physical viability of the union of Eqs. (8) and (28), and,
35 concomitantly, establishing the functional dependence of K upon the system’s parameters. In the case of
liquids, for which no fully accepted theory yet exists, one has no recourse other than to revert to experiment in
R

37 pursuit of these goals. Fortunately, however, in the case of rarefied gases an accepted molecular theory already
R

exists, one that will be seen to suffice for these purposes. Explicitly, we can avail ourselves of Burnett’s
39 extension [12] to higher Knudsen numbers of the Chapman–Enskog scheme [13] for solving Boltzmann’s
O

transport equation in the rarefied gas regime.


41 Chapman and Enskog’s calculations [13] theoretically predict, inter alia, at least for rarefied gases, that the
C

Fourier and Newtonian rheological law constitutive relations (42) and (44), previously regarded as empirical
43 experimental laws valid for continua, are indeed applicable in the so-called ‘‘near-continuum,’’ OðKnÞ, region
N

of small Knudsen numbers, Kn51, with the Oð1Þ terms represented by the inviscid, ideal fluid Euler equations
45
U

[36]. Moreover, Chapman and Enskog’s theoretical perturbation scheme concomitantly furnishes the values of
the phenomenological coefficients k, Z and ZB appearing therein, at least for particular intermolecular collision
47 model choices (e.g., rigid–elastic spheres, Maxwell molecules, Lennard-Jones potentials, etc.). In what follows,
Burnett’s [12] Knudsen number extension of the Chapman–Enskog theory [13] will be seen to provide
49 confirmation of the constitutive equation (28) for j, at least for the case of monatomic Maxwell molecules
[13,47], while at the same time confirming the inequality (46) by showing that
51
K ¼ a, (47)
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 13

1 where
k
3 a¼ (48)
r^cp
5 is the thermometric diffusivity, in which c^p is the isobaric specific heat.
As heralded above, the proof to be offered of Eq. (28), wherein K is given by Eq. (47) (thus satisfying
7 inequality (46)), derives from Burnett’s extension of Chapman and Enskog’s expansion scheme for solving the
Boltzmann equation to OðKn2 Þ, namely beyond the so-called ‘‘near continuum,’’ OðKnÞ, N–S–F level. In that
9 context, Kogan et al. [6–9] and Bobylev [10] have shown that at small Mach numbers ðMa51Þ and for
Reynolds numbers of order unity ðRe ¼ Oð1ÞÞ, the so-called ‘‘thermal stress’’ terms appearing in Burnett’s
11 [12,13] expression for the deviatoric stress T, rather than being of OðKn2 Þ as originally supposed, are, in fact, of
the same OðKnÞ order as are the N–S–F equations themselves, at least in the case of so-called SNIF flows (slow
13 nonisothermal flows) [9,10]. The scaling argument used to rationalize this Knudsen number re-ordering
derives, in part, from the fact that Kn ¼ Ma=Re, so that a small Knudsen number can be achieved as indicated
15 above, rather than by the classically assumed circumstances [13] where Ma ¼ Oð1Þ and Reb1. The SNIF limit
[10] especially includes the limiting case where vm ¼ 0 throughout the fluid, such as would be encountered, for
17 example, in the elementary case of one-dimensional steady-state heat conduction through a gravity-free gas
confined between parallel, laterally unbounded, walls permanently maintained at different temperatures,
19 where (from a macroscopic viewpoint) the pressure is sensibly uniform throughout the fluid, except perhaps in

F
thin Knudsen boundary layers existing proximate to the wall [48]. In such circumstances, the Burnett thermal

O
21 stresses (to which we have referred elsewhere [3] as being the Maxwell–Burnett stresses [49]) are of the form
[13, p. 286]
23
T¼
Z2
ðK 1 rrT þ K 2 T 1 rTrTÞ, O (49)
PR
25 rT
where K 1 and K 2 are Oð1Þ nondimensional constants, whose respective numerical values depend slightly upon
27 the particular molecular collision model adopted [13].
For monatomic Maxwellian molecules the values of the two constants appearing above are, respectively,
D

29 K 1 ¼ 3 and K 2 ¼ 3 d ln Z=d ln T [13, pp. 288–289]. Accordingly, in that case Eq. (49) adopts the form
TE

3Z
31 T¼ rðZrTÞ.
rT
33 Now, Z ¼ ru, where u is the kinematic viscosity. Additionally, for a single-component ideal gas, the relation
EC

between density and temperature at constant pressure is such that rT ¼ const., whence it readily follows for
35 the present one-dimensional, steady-state, isobaric heat conduction case under consideration that
R

37 T ¼ 3Zrður ln rÞ.
2
However, for an ideal monatomic gas, the Prandtl number Pr ¼ u=a has the value [18]. Accordingly, the
R

3
39 preceding relation becomes
O

41 T ¼ 2Zrðar ln rÞ. (50)


C

In addition to the latter relation, it is also known theoretically in the case of ideal monatomic gases (which
43 includes Maxwell molecules) that the bulk viscosity is zero [13,47]. Hence, in addition to (50) it is also true in
N

present circumstances that


45
U

ZB ¼ 0. (51)
47 Upon comparing the union of Eqs. (50) and (51) with that of Eqs. (8) and (44) in the light of the fact that
vm ¼ 0 in present circumstances (thus making v ¼ j, and hence T ¼ 2ZrjÞ, it follows that
49
j ¼ ar ln r. (52)
51 Comparison of the latter with (28) serves to confirm the viability of hypothesis (27). While we have derived Eq.
(52) by considering only a rather restricted set of circumstances, namely, vm ¼ 0 (and rp ¼ 0Þ, the conclusion
PHYSA : 9965
ARTICLE IN PRESS
14 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 that K ¼ a bears no such restriction because Eq. (28) is true independently of such restrictions. In particular,
as discussed in the derivation of that equation, pressure gradients play no role in its validity.
3 While it is true that Eq. (47) has only been confirmed to hold for ideal monatomic gases composed of
Maxwell molecules, on the basis of Eqs. (8) and (28) this would nevertheless appear sufficient to conclude that,
5 in general, there does exist a fundamental difference between the gas’s specific momentum density v and its
mass velocity vm (provided, of course, that we accept the Boltzmann equation and the Burnett–Chapma-
7 n–Enskog Knudsen number expansion scheme as being valid, at least at small Reynolds numbers, where the
momentum density terms in the Navier–Stokes equations are small—a position accepted by most researchers).
9 More importantly, the inequality vavm depends critically upon the assumption that the present fluctuation-
based GENERIC equations [23] provide a proper physical foundation for irreversible (gas) transport
11 processes. As such, the experimental verification of Eq. (52) for gases (and liquids) would serve to provide a
rigorous test of the viability of Öttinger’s nontraditional GENERIC scheme [23]. In the next section we discuss
13 such tests. Before doing so, however, we briefly note below the existence of a second, equally unorthodox,
theory of irreversible processes for single-component ideal gases due to Klimontovich [28–30]—a theory based
15 directly upon molecular-level arguments rather than upon the present GENERIC macroscopic phenomen-
ological arguments—and which serves to independently confirm Eq. (52).
17 Explicitly, Klimontovich argues in favor of the relation v ¼ vm þ Dr ln r, analogous to Eqs. (8) and (28), in
which the symbols v and vm have the same physical significance as described above, and in which the symbol D
19 denotes Klimontovich’s physical-space single-component self-diffusion coefficient. Furthermore, he argues

F
that D is equal to the gas’s thermometric diffusivity a, thereby furnishing the same K ¼ a relation (47) as

O
21 derived earlier from Burnett’s Boltzmann equation-based molecular calculations. In fairness, however, it needs
to be emphasized that Klimontovich’s nontraditional scheme has not yet been subjected to the same searching
23
O
theoretical scrutiny as either Burnett’s original, more traditional, scheme or Kogan et al.’s [6–9] and Bobylev’s
[10] re-ordering of the thermal stress portion of the Burnett terms.
PR
25
6. Identification of the specific momentum density v as the volume velocity vv
27
Eqs. (8) and (52) or, more precisely, the similar-appearing pair, Eqs. (53) and (54), set forth below (wherein
D

29 the subscript ‘‘v’’ appears), have a pre-history which is completely independent of the present GENERIC
TE

scheme. In particular, in earlier work [4] we derived the purely kinematical relation
31
vv ¼ vm þ j v , (53)
33 where vv is the volume velocity (which is equivalent to the Eulerian flux density of volume nv —cf. Eq. (10)
EC

wherein c ¼ 1 for the case of volume) and jv is, by definition, the diffusive flux density of volume. The latter is
35 given constitutively for single-component gases or liquids undergoing heat transfer by the relation [4]
R

37 jv ¼ ar ln r. (54)
The pair of relations displayed above, each derived theoretically, have been confirmed indirectly by
R

39 comparison of the physical consequences stemming therefrom with experimental data for both gases and
O

liquids [2–5,14,50]—without, however, supposing the symbol vv appearing in Eq. (53) to be equal to the fluid’s
41 specific momentum density v (since inertial effects, wherein v would otherwise have proved pertinent, were
C

negligible in all of the experiments to which Eqs. (53) and (54) pertain [51]). Given that the respective right-
43 hand sides of Eqs. (52) and (54) are identical, it follows that
N

45 j ¼ jv . (55)
U

Comparison of Eq. (8) with (53) in the light of the latter identity reveals that
47
v ¼ vv . (56)
49 It has been argued [1–3,14,15] that vv is identical to the fluid’s ‘‘tracer’’ or ‘‘Lagrangian’’ velocity, say vl , the
latter representing the actual physical velocity of an object through space (as opposed to the fluid’s mass or
51 volume velocities, both of which are flux densities in disguise [2]). Thus, the physical essence of Eq. (56) is
embodied in the relation
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 15

1 v ¼ vl . (57)
In as much as momentum also involves the notion of an object moving through space, Eq. (57) constitutes one
3
of the key results of our analysis, namely the conclusion that the fluid’s specific momentum density is equal to
its Lagrangian velocity rather than to its mass velocity. And, to the extent that our hypothesis (27) is valid, it is
5
only in the case where the fluid is ‘‘incompressible’’ that the two results coincide.
7
7. Onsager-based irreversible, thermodynamic justification of our post-GENERIC assumptions
9
7.1. Prelude
11
The present paper has built upon Öttinger’s [23] incorporation of fluctuations into the GENERIC friction
13 matrix. This involved the introduction of two new parameters, a0 and D0 , into the scheme. Except for the
required algebraic sign of D0 as set forth in Eq. (21), both parameters are left constitutively undetermined in
15 Öttinger’s original BET development (see, however, Section 8.4). Accordingly, bringing his theory to fruition,
namely to the stage where its predictions can be compared with experiment, necessitates specifying constitutive
17 expressions for these two key phenomenological coefficients in terms of physically measurable properties. It is
this key step to which the present paper has largely been devoted. The critical assumptions that we made
19 enroute to the goal of establishing plausible constitutive relations for a0 and D0 involved our identification of

F
the internal energy and entropy production rates, pu and ps , indicated in Eqs. (5) and (7), respectively, together
with the adoption of the ‘‘incompressibility’’ hypothesis (27). These led to our constitutive determination of

O
21
the parameter a0 , Eq. (33), following which subsequent determination of the remaining parameter D0 (cf. Eqs.
23
O
(31) and (47)) was straightforward. Accordingly, we need focus here only on the three key hypotheses cited
above, leading to our eventual determination of a0 , as set forth in Eq. (33).
PR
25 The justification offered below for these constitutive assumptions is based, ultimately, upon the agreement
of the predictions of the resulting theory with experiment—which is as it should be. Because the theory is, at
27 this stage, limited to single-component fluids, the confirming experiments to which we refer are necessarily
limited to such systems. The key experimental laws in this connection are: (i) Fourier’s law of heat conduction,
D

29 Eq. (42), which refers to diffuse energy transport arising exclusively from a temperature gradient; (ii) the
Clausius–Duhem inequality, Eq. (13); (iii) Onsager’s reciprocal theorem, to be discussed, involving coupling
TE

31 between the independent fluxes appearing in the ‘‘force–flux’’ relations appearing therein; and (iv) Curie’s law,
which denies the possibility of coupling in linear isotropic systems between fluxes whose respective tensorial
33 orders differ by an odd integer. Together, these laws provide a test of the legitimacy of any theory of linear
EC

irreversible processes that purports to be generally applicable. Viewed alternatively, the a priori acceptance of
35 the general correctness of LIT provides a test of the internal consistency of our theory.
The critical importance of the coupling issue here resides in our claim that the diffuse flux of volume jv
R

37 represents an independent flux, over and above the traditional diffuse fluxes of internal energy ju (or heat jq Þ
and momentum T. Owing to Curie’s law, the diffuse second-rank tensor momentum flux is necessarily
R

39 uncoupled from the diffuse vector internal energy (or heat) flux. With diffuse volume accepted as representing
yet another independent vector flux, as is advocated here, it too would be uncoupled from the momentum flux.
O

41 On the other hand diffuse volume would be coupled to the internal energy flux, and thus subject to the
restrictions imposed by Onsager reciprocity, requiring equality of the cross coefficients in the flux vs. driving
C

43 force reciprocal relations.


N

In regard to coupling, the application of the Clausius–Duhem inequality to single-component fluids had, in
45 the past, where the issue of diffuse volume transport had not yet arisen, been essentially trivial. In those
U

conventional circumstances, standard LIT arguments [19–21] resulted in the simple conclusion that
47  
1 1
ps ¼ ju .r þ T : ðrvÞT , (58)
T T
49
there being no Onsager coupling of the internal energy flux to any other independent flux. As such, as a
51 consequence of Eqs. (42) and (44) together with the traditional assumption that ju  jq ¼ krT, the
inequality ps X0 was trivially satisfied (with v appearing therein identified as being the fluid’s mass velocity vm Þ.
PHYSA : 9965
ARTICLE IN PRESS
16 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 However, given the present conception of an independent diffuse volume flux jv , the latter dependent solely on
the density gradient rr as in Eq. (54), there then arises the possibility of coupling between the vector fluxes jq
3 and jv , subject to the restrictions imposed thereon by both Onsager reciprocity and the Cauchy–Duhem
inequality (13). It is the consequences stemming from these requirements that we examine below.
5 It is in this restricted coupling context that our trio of post-GENERIC constitutive assumptions embodied
in Eqs. (5), (7) and (27) receives theoretical justification. However, the issue is complicated by the fact that
7 there exists no unequivocal definition of the ‘‘heat flux’’ in the literature. This stems from the fact that since
heat is not an extensive property of a system, the notion of a heat flux is essentially ambiguous. Only extensive
9 physical properties such mass, internal energy, momentum, volume, entropy, electric charge, etc. can give rise
to fluxes, much less strictly diffusive (i.e., nonconvective) fluxes, such as is implied by the notion of heat
11 conduction. Stated explicitly, there exists no extensive quantity Q to which the symbol jq can be assigned as
representing its (diffusive) flux. This contrasts with the fact that extensive quantities like internal energy U can
13 ^ fluxes, with nm the mass flux density. This ambiguity in the
possess diffusive (ju Þ as well as convective (nm uÞ
interpretation of the heat flux permeates the literature of LIT.
15 One might believe that the problem is at least partially alleviated by invoking Fourier’s law, Eq. (42), in the
sense of the latter serving to ‘‘define’’ the heat flux jq as being that portion of the diffuse (internal) energy flux
17 which vanishes when rT ¼ 0. However, that would be analogous, for example, to defining the concept of a
force, say F, in the Newton–Euler point-mass rigid-body law, F ¼ ma [40], by regarding a as the purely
19 kinematically defined acceleration that it indeed is, and then simply defining the force as being the quantity

F
which vanishes when a ¼ 0 (with m the proportionality coefficient). The point here is that, objectively, a

O
21 relation only achieves acceptance as a physical law of nature when quantitative and independent definitions of
the respective variables appearing in that law (not including the phenomenological proportionality coefficient)
23
O
have already been set forth prior to proposing that the relation in question be elevated to the status of a bona
fide experimentally based law. Accordingly, returning to the Fourier law issue, the heat flux jq must be defined
PR
25 without any reference whatsoever to its possible relationship to a temperature gradient (cf. Eq. (63)), although
it is permissible to use the notion of the temperature itself in its definition. The Dufour effect [19, p. 274], said
27 to result in a ‘‘heat flow’’ in an isothermal mixture undergoing diffusion, is a case in point. The notion of a heat
flux in the absence of a temperature gradient strains credulity, as would surely have been true of Fourier. What
D

29 is almost certainly being referred to in this Dufour context is a diffuse internal energy flow.
It is with this lengthy preamble in mind that we now turn to the issue of justifying the trio of assumptions
TE

31 resulting in our constitutive expression for a0 , Eq. (33). In this context we begin by merely summarizing those
formulas which have been derived on the basis of those assumptions, and then using this information to
33 suggest a definition for the heat flux jq that meets the criteria that we have specified (at least in single-
EC

component systems). These criteria involve showing that Fourier’s law (42) is indeed satisfied by this definition
35 of jq , as too are the laws of Onsager and Clausius–Duhem. In this context it is interesting to note that insofar
as we are aware there exists no theoretical proof of Fourier’s law in the general case of continua [52] (i.e.,
R

37 involving both liquids and gases, the latter not necessarily rarefied). While one might believe that the
Boltzmann equation offers the basis for such a proof, at least in the case of single-component rarefied gases,
R

39 such a ‘‘proof’’ involves the implicit assumptions that: (i) the heat flux is identical to the (diffuse) internal
energy flux; and (ii) there exist no other independent vector fluxes, such as jv , to which the Boltzmann-based
O

41 heat flux vector might otherwise couple. However, as can be seen from Eq. (34), according to our present
arguments, jq and ju are the same only in the special case of ideal gases; moreover, even in that case a
C

43 Boltzmann-based proof cannot be accepted as complete without first addressing the coupling issue, and
N

subsequently showing that the Clausius–Duhem second law inequality, now possibly including such coupling,
45 is indeed satisfied. (Of course, such coupling issues had not even been recognized at the time that Boltzmann
U

and, later, Chapman and Enskog [13] did their foundational work.)
47
7.2. Definition of the heat flux jq
49
As a consequence of Eq. (55), Eq. (37) adopts the form
51
ju ¼ jq þ jv ðqU=qV ÞT;M . (59)
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 17

1 In a single-component isothermal system the quantity ðqU=qV ÞT;M would represent the internal energy per
unit volume, namely u. Since jv is the diffuse volume flux density, it follows that the term jv ðqU=qV ÞT;M
3 appearing above represents a nonconvective internal energy flux contribution that would exist in the absence of
a temperature gradient (thereby arising from a density gradient). As such, this term constitutes a diffuse
5 ‘‘isothermal’’ flux of internal energy accompanying the diffuse volume flow jv in a piggy-back mode. On the
other hand, the Fourier heat flux vector jq represents the nonconvective transport of internal energy across a
7 surface due exclusively to a temperature gradient. As such, it is quite reasonable to suppose that the sum of the
two terms appearing on the right-hand side of Eq. (59) should constitute the total nonconvective or diffuse
9 internal energy flux, ju .
Issues similar to those discussed above for energy transport also arise in the comparable entropy transport
11 case. In particular, in light of relation (55), Eq. (41) now becomes
 
jq qS
13 js ¼ þ jv . (60)
T qV T;M
15 This expression differs from the classical irreversible thermodynamic result [19–21] for single-component
fluids, namely js ¼ jq =T. The extra nontraditional term appearing above represents an ‘‘isothermal’’ flux of
17 entropy accompanying the diffusive volume flow. Note that in contrast with its corresponding energy flux
counterpart in Eq. (59), this unorthodox additional term does not vanish for ideal gases since, according to a
19 Maxwell relation, ðqS=qV ÞT;M ¼ ðqp=qTÞV ;M a0 in general.

F
Upon eliminating jq between Eqs. (59) and (60), subsequent use of Eq. (17) in the resulting expression yields

O
21 the following relation upon rearrangement:
ju ¼ Tjs  pjv . (61)
23
O
This result could also have been obtained directly from Eq. (16) upon invoking Eq. (55). This expression
PR
25 obviously constitutes the transport (i.e., flux) counterpart of the thermodynamic relation (17). While Eq. (61)
derives from nontraditional GENERIC theory, its validity is nevertheless intimately linked to our further
27 constitutive assumptions, as embodied in Eqs. (5) and (7) (although not Eq. (27)), from which Eqs. (59) and
(60) have evolved. Eq. (16) constitutes a powerful incentive to conclude that Eq. (55) must be correct in order
D

29 to fulfill the analogy with Eq. (17).


Eq. (59) or its equivalent entropic counterpart, Eq. (60), offers a definition of the heat flux in terms of well-
TE

31 defined physical properties. Moreover, this definition of jq is independent of any (subsequent) association with
the temperature gradient, such as in Fourier’s law, Eq. (42). A fundamental point here when proposing Eq.
33 (59) as the definition of the heat flux vector (at least in single-component systems) is that, rationally speaking,
EC

only extensive physical properties can possess a flux, as earlier noted. To repeat what was said there: since heat
35 is not an extensive property, in the sense that a system cannot be said to possess an ‘‘amount’’ of heat Q, the
concept of a heat flux cannot be a primitive concept in any theory of irreversible processes; rather, it must be a
R

37 defined quantity. And if heat flux cannot then serve as a primitive concept in nonequilibrium thermodynamics,
then heat itself cannot serve as a primitive concept in equilibrium thermodynamics. In effect, we are proposing
R

39 here that instead of defining the notion of internal energy U in terms of heat Q and work W—with W
O

representing a well-defined, strictly mechanical or electromechanical, concept (and thus able to serve as a
41 primitive concept)—we reverse the scheme by defining heat in terms of internal energy, with the latter now
C

serving as the basic primitive notion. In effect, we define the quantity dQ arising during an infinitesimal,
43 generally irreversible, process involving a change in thermodynamic state as
N

d=Q:¼dU  d=W , (62)


45
U

where dU is an exact differential. The inexact differential d=W is the (generally path-dependent) work done
47 during the process, the latter presumably obtainable via quantitative dynamical analyses embodying purely
mechanical or magneto-electromechanical concepts, even for inherently irreversible processes. Of course, in the
49 case of so-called reversible processes in single-component fluids, we have that d=W rev ¼ p dV .
Consistent with this philosophy of having internal energy rather than heat serve as the primitive concept in
51 equilibrium thermodynamics, with the heat transfer d=Q then defined in terms of the change dU in the latter
(together with the work d=W done), on the basis of Eq. (59), we instead propose the following definition of the
PHYSA : 9965
ARTICLE IN PRESS
18 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 heat flux in nonequilibrium thermodynamics:


jq :¼ju  jv ðqU=qV ÞT;M . (63)
3
Of course, this definition is understood to apply only in the case of single-component fluids. Its extension to
5 more general cases is obvious, and is briefly discussed later in Section 8.6. Use of the above in conjunction with
the fundamental proposition (61) and, hence, (60), all again valid only for the single-component case, will be
7 seen to lead to a systematic approach to the subject of nonequilibrium thermodynamics which, like
GENERIC theory itself, is not necessarily limited to linear processes. In effect, instead of regarding Eqs. (63)
9 and (61) as having been derived beginning with single-component GENERIC theory (including the
demonstration that v ¼ vm þ jv Þ, we propose elevating these equations to the level of fundamental postulates.
11 Together with Eq. (42)—the latter now regarded as an exact relation valid for all temperature gradients, not
necessarily small—the proposed scheme collectively serves to set forth the basic equations governing
13 irreversible thermodynamics. It needs to be stressed, however, that we are not advocating basic changes in the
fundamental structures of either GENERIC or LIT, both of whose foundations are sufficiently robust to,
15 respectively, accommodate therein the existence of newly recognized physical phenomena. Rather, in the case
of LIT, our proposed scheme simply entails adding volume to the present list of independent diffuse fluxes
17 appearing in that theory. Equivalently, in the case of GENERIC, following Öttinger [23], molecularly based
fluctuations are incorporated into the friction matrix appearing in that theory. In the final analysis these are
19 the only substantive additions advocated. After all, given that it was GENERIC that led to the precise set of

F
equations which we have proposed, and given the consistency of GENERIC with LIT, it could not seriously

O
21 be argued that any conflict existed between our proposal and either of these two structures. Our proposed re-
interpretation of the heat flux does not appear to impact on either of these basic structures, but rather only in
23
O
the constitutive manner in which they are to be applied. These issues will be discussed in a broader context in
subsequent papers, where, in a systematic, formal and axiomatic manner, we propose to go beyond the simple
PR
25 single-component systems studied here.
Inasmuch as the concept of heat, and hence of heat transfer, makes its initial entrée into the realm of
27 thermodynamics in connection with its role in the first law, logic would appear to demand that the definition
of the heat flux involve, at most, only first law of thermodynamic concepts. This would include the internal
D

29 energy U (as well as the nonthermodynamic notions of volume V and mass MÞ. Eq. (63) would appear,
superficially, to violate this concept since the thermodynamic absolute temperature T appears as one of the
TE

31 variables invoked in the partial derivative. And T is a strictly second law concept. This brings about the
recognition that the ‘‘temperature’’ which enters into the definition of the heat flux need not be formally
33 identified with the symbol T. Rather, temperature may, in a broader sense, be thought of as a strictly primitive
EC

empirical concept—as indeed it was viewed during the reign of the caloric theory of heat, prior to the
35 axiomatic work of Carnot, Joule, Kelvin and Clausius systematizing the foundations of thermodynamics. At
that pre-thermodynamic time, the notion of temperature was unrelated to the formal definition of the symbol
R

37 T appearing as the ‘‘integrating factor’’ in Clausius’s definition, dS:¼d=Qrev =T, of the entropy change
accompanying a reversible flow of heat (the latter heat flow a strictly first law concept); that is, following
R

39 Fourier and others of that pre-thermodynamic era, one may regard temperature as a primitive quantity,
represented, say, by the symbol y, in which case Eq. (63) would be then replaced by the expression
O

41
jq :¼ju  jv ðqU=qV Þy;M ,
C

43 ^ v; yÞ. Fourier’s law would


where, for a single-component fluid, one has the functional relationship U ¼ M uð^
N

then read
45
U

jq ¼ kry,
47 rather than being given by Eq. (42). In the context of the preceding discussion, it is illuminating to read that
portion of Fourier’s classical book, ‘‘The Analytical Theory of Heat,’’ concerned with attempting to explain
49 both ‘‘temperature’’ and the ‘‘communication of heat’’ between bodies in different thermal states.
Viewed alternatively, the single-component relation js ¼ jq =T appearing in textbooks (cf. [19, Eq. (20), p.
51 24])—which relation, incidentally, we believe to be incorrect on the basis of Eq. (60)—could not, even were it
?
to be correct, be viewed in reverse as the definition, jq :¼ Tjs , of the heat flux. The latter view would violate the
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 19

1 required sequential ordering of the first and second laws. Without entering into further details, this digression
suffices to identify the deep philosophical issues which surround both the definition of the heat flux and
3 Fourier’s law, especially when embedded in the context of Lebowitz’s oft-repeated view [52] that no formal
(statistical–mechanical) proof of Fourier’s law exists despite the almost 200 years that have elapsed since its
5 introduction into physics. Lastly, we note that temperature, y, as a physical concept, may, by analogy with
mechanics, be regarded literally as constituting a ‘‘potential’’ for the movement of (internal) energy through
7 space—an interpretation which is consistent with its appearance as a gradient in Fourier’s law.
We now resume the discussion that preceded the digression of the preceding two paragraphs. In order that
9 the program proposed herein, involving the definition of the heat flux, be deemed acceptable, it remains to
show that our definition is internally consistent, in the sense of satisfying the four basic criteria outlined above.
11 We pursue this agenda by first identifying the driving forces conjugate to the diffusive fluxes with respect to
Onsager’s reciprocity law. This is achieved by expressing the entropy production rate ps in terms of the
13 independent fluxes of the pertinent extensive properties involved in the analysis. It is important to note that
this identification does not involve the concept of the heat flux jq .
15
7.3. Identification of the driving forces conjugate to the diffusive fluxes
17
From Eqs. (12) and (19) the entropy production rate can be expressed as
19

F
Dm s^
ps ¼ r þ r .js . (64)
Dt

O
21
However, from Eq. (61),
23
1
js ¼
p
j þ j.
T u T v
O (65)
PR
25
Accordingly, Eq. (64) adopts the form
27   p
1
.
ps ¼ ju r þ jv .r þ D, (66)
T T
D

29
where we have defined
TE

31 Dm s^ 1 p
D¼r þ r.ju þ r.jv . (67)
Dt T T
33
EC

In the following paragraph we demonstrate that D ¼ T 1 T : rv.


35 The internal energy- and volume-production rate analogs of Eq. (64) are, respectively,
Dm u^
R

37 pu ¼ r þ r .ju (68)
Dt
R

39 and
Dm v^
O

41 pv ¼ r þ r.jv . (69)
Dt
C

The intensive form of the extensive combined first and second laws (Eq. (17)) for single-component systems is
43
du^ ¼ T d^s  p d^v, which we rewrite as
N

45 1 p
U

d^s ¼ du^ þ d^v. (70)


T T
47 We assume, as is also assumed in the case of irreversible thermodynamics [19–21], that the preceding equation
remains valid in the material form
49
1 p
Dm s^ ¼ Dm u^ þ Dm v^,
51 T T
allowing us to write
PHYSA : 9965
ARTICLE IN PRESS
20 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 Dm s^ 1 Dm u^ p Dm v^
¼ þ . (71)
Dt T Dt T Dt
3 Adoption of this relation is tantamount to supposing the local equilibrium postulate to be valid [19–21].
Hence, with use of Eqs. (68) and (69) the above becomes
5
Dm s^ 1 p
r ¼ ðpu  r.ju Þ þ ðpv  r.jv Þ.
7 Dt T T
Substitution of the above into (67) gives
9 1
D¼ ðpu þ ppv Þ. (72)
T
11
However, in the Eulerian form (9) of the volume-transport equation, where C ¼ V , we have that c ¼ 1.
13 Moreover, it follows from (10) and (53) that nv ¼ vv , whence we find that [4]
pv ¼ r.vv . (73)
15 In addition, from Eqs. (5) and (3),
17 pu ¼ pr.v þ T : rv. (74)
Consequently, Eq. (72) becomes
19

F
1
D¼ ½pr.ðv  vv Þ þ T : rv. (75)
T

O
21
Finally, then, with use of Eq. (56), we obtain
23

1
T
T : rv. O (76)
PR
25
Substitution of the latter into Eq. (66) makes
  p
27 1 1
ps ¼ T : rv þ ju r . þ jv .r . (77)
T T T
D

29
This expression for the entropy production rate possesses the classic ‘‘flux/driving force’’ summation-matrix
TE

format of LIT [19–21], namely


31
ps ¼ Sc ðj c X c Þ. (78)
33
EC

From (77), the driving forces X c conjugate to the fluxes j c of linear momentum, internal energy and volume
are, respectively,
35  
1 1 p
X M ¼  rv; X u ¼ r and X v ¼ r , (79)
R

37 T T T
R

where we have noted that jM ¼ T.


39 As a consequence of Curie’s law, the Clausius–Duhem inequality (13) applied to Eq. (77) requires separate
O

satisfaction of each of the following inequalities:


41
1
C

ps ðstressesÞ:¼ T : rvX0 (80)


43 T
N

and
45   p
U

1
ps ðvector fluxesÞ  Ps :¼ju .r þ jv .r X0. (81)
47 T T

49 7.4. Onsager reciprocity

51 The preceding identification of the respective conjugate driving force for each independent flux enables us to
explicitly address the restrictions imposed by Onsager’s reciprocal theorem upon our theory. By Curie’s
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 21

1 theorem, the diffuse momentum flux is uncoupled from those of internal energy and volume. Accordingly,
Onsager’s theory together with Eq. (77) requires that the following general constitutive relations apply to the
3 respective fluxes of internal energy and volume:
  p
1
5 ju ¼ Luu r þ Luv r , (82)
T T
7   p
1
jv ¼ Lvu r þ Lvv r . (83)
9 T T
This pair of relations may be written alternatively in the matrix form
11 ! " #( )
ju Luu Luv rð1=TÞ
13 ¼ , (84)
jv Lvu Lvv rðp=TÞ

15 wherein the square ½L matrix appearing in the preceding is both symmetric and nonnegative-definite in order
that inequality (81) be satisfied. Thus, we require satisfaction of the equality
17
Luv ¼ Lvu , (85)
19 as well as that of the trio of inequalities

F
Luu X0; Lvv X0 (86a,b)

O
21
and
23
Luu Lvv  Luv Lvu X0. O (86c)
PR
25 It proves convenient to re-express Eqs. (82) and (83) in terms of an alternative set of driving forces, namely
frð1=TÞ; rv^g, in place of the previous set, frð1=TÞ; rðp=TÞg. To do so, we note in the single-component case
27 that p ¼ pðT; v^Þ. Thus, in the identity
 p  rp  
1
D

29 r ¼ þ pr ,
T T T
TE

31 we have that
         
33 qp qp 2 qp 1 qp
rp ¼ rT þ rv^ ¼ T r þ rv^.
EC

qT v^ q^v T qT v^ T q^v T
35 Consequently, using Eq. (36), one obtains
p      
R

37 qU 1 1 qp
r ¼ r þ rv^. (87)
T qV T T T q^v T
R

39
Substitute the latter into Eqs. (82) and (83) so as to obtain
O

       
41 qU 1 1 qp
ju ¼ Luu  Luv r þ Luv rv^ (88)
C

qV T T T q^v T
43
N

and
45        
U

qU 1 1 qp
jv ¼ Lvu  Lvv r þ Lvv rv^. (89)
qV T T T q^v T
47

49 7.5. Fourier’s law

51 Introduction of Eq. (63) into (88) followed by the use of (89) furnishes the following expression for the heat
flux:
PHYSA : 9965
ARTICLE IN PRESS
22 H. Brenner / Physica A ] (]]]]) ]]]–]]]

"    2 #        
1 qU qU 1 qU 1 qp
q
j ¼ Luu  ðLuv þ Lvu Þ þ Lvv r þ Luv  Lvv rv^. (90)
qV T qV T T qV T T q^v T
3
Analogous to (27), we now wish to assign to the heat flux jq the universal property that
5
jq ¼ 0 in isothermal systems, i:e:, when rT ¼ 0, (91)
7 irrespective of the values of the other independent variables, say v^ (or, equivalently, pÞ and vm , entering into the
problem. In turn, this necessitates that the bracketed expression appearing in the last term of Eq. (90) vanish,
9 thus requiring that
 
qU
11 Luv ¼ Lvv ¼ Lvu , (92)
qV T
13 where we have also taken note of Eq. (85). Substitution of (92) into both (90) and (89) gives
"  2 #   "  2 #
15 q qU 1 1 qU
j ¼ Luu  Lvv r   2 Luu  Lvv rT (93)
qV T T T qV T
17 and
 
19 1 qp
jv ¼ Lvv rv^. (94)

F
T q^v T

O
21
Eq. (93) is, of course, Fourier’s law, namely Eq. (42), which we repeat here:
jq ¼ krT,
23
wherein kX0 as in Eq. (43). Introduction of (95) into (93) gives
O (95)
PR
25  2
qU
Luu ¼ Lvv þ kT 2 . (96)
27 qV T
It follows from this that if Lvv X0, as required by Eq. (86b), then, as a consequence of (43), it will also be true
D

29
that Luu X0, in accord with (86a).
TE

With use of the constitutive expressions (93) and (94) in Eq. (81) we find that
31 "  2 #  2    2
qU 1 1 qp
33 Ps ¼ Luu  Lvv r þ Lvv ðrv^Þ2 . (97)
qV T T T q^v T
EC

35 As such, Eq. (81) leads to the requirement that


 2
qU
R

37 Luu XLvv . (98)


qV T
R

39 From Eq. (96) it is seen that this condition is satisfied by the fact that the thermal conductivity is nonnegative.
O

Given the unequivocal statement embodied in Eq. (91) it is tempting to consider the possibility that
41 Fourier’s law, Eq. (95), may be valid under more general circumstances than would normally be expected,
C

namely the regime beyond the small temperature gradient case that would suffice to assure linearity of the flux/
43 driving force relation explicit in Fourier’s law. The issue of possible limitations, or lack thereof, on its realm of
N

applicability remains open as of this writing. Explicitly, we are unaware of any experimental data or theory
45 [52] that points to any limits.
U

47 7.6. Constitutive equation for the diffuse flux of volume


49 In as much as v^ ¼ 1=r, it follows that Eq. (94) is equivalent to the expression
51 1
jv ¼ Lvv r ln r, (99)
TkT
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 23

1 with kT is the coefficient of isothermal compressibility, defined in Eq. (29). In the light of Eq. (28) the above is
equivalent to the relation Lvv ¼ KTkT [53]. Given our identification of K with a in Eq. (47), together with the
3 subsequent argument that this relation applies equally to liquids, it follows that, in general,
Lvv ¼ aTkT , (100)
5
where the thermometric diffusivity a, Eq. (48), is a positive quantity. Consequently, the requirement that
7 Lvv X0 demands that the algebraic sign of kT be nonnegative, as already stipulated in connection with Eq. (30).
The required nonnegativity of the isothermal compressibility kT appearing in Eq. (100) is key to Öttinger’s
9 [23] central argument that his proposed extension of the GENERIC friction matrix is necessitated by the
existence of fluctuations. The intimate relationship of kT , especially its algebraic sign, to the theory of
11 fluctuations lies in its appearance in the theory of equilibrium fluctuations [54, p. 123],
hðdrÞ2 i kb TkT
13 ¼ , (101)
hri2 V
15 where kb is Boltzmann’s constant. The angular brackets denote thermal averaging. In the above, hri ¼ M=V is
the mean density, while hðdrÞ2 i is the mean-squared density fluctuation occurring within an open domain of
17 fixed volume V as a result of fluctuations in the instantaneous fluid mass M contained within that volume,
owing to the ability of individual molecules to freely enter and leave that domain through its surface. Given
19 the nonnegativity of all of the other parameters appearing in Eq. (101) it is obvious that the theory of

F
fluctuations requires satisfaction of the inequality kT 40. This inequality is also related to the purely
thermodynamic fact that stability of the fluid phase requires that ðq2 A=qV 2 ÞT;M X0, wherein A is the extensive

O
21
Helmholtz free energy. In view of the thermodynamic identity [54, p. 123] ðq2 A=qV 2 ÞT;M ¼ 1=V kT , it is
23
O
evident that stability demands that the isothermal compressibility be positive.
Fluctuations constitute a necessary ingredient when rationalizing the physical role played by Öttinger’s
PR
25 positive-definite material property coefficient D0 in relation to the diffusional contribution M diff to the
GENERIC friction matrix M (cf. Ref. [23, Eq. (2.77)]); that is, were the fluid to be truly ‘‘incompressible,’’ in
27 the sense that kT ¼ 0 identically, there would and could be no fluctuations in density, presumably requiring
that D0 ¼ 0 too. This is consistent with our basic hypothesis (27), according to which it is only in the case of
D

29 incompressible fluids that Euler’s specific momentum relation v ¼ vm holds.


Among other things, these remarks point up the (thermodynamically) singular nature of the notion of fluid
TE

31 incompressibility [55], a simplification lying at the heart of most contemporary fluid-mechanical applications,
especially in the case of liquids. Strict incompressibility corresponds to the case where ðqr=qpÞT ¼ 0, or,
33 equivalently, kT ¼ 0. From Eq. (101) such incompressibility rules out the possibility of fluctuations. At the
EC

same time, as evidenced by Eq. (99), this leads to an obvious singularity with regard to the existence of a
35 diffuse volume flux. If nothing else, this interplay between kT and jv shows clearly the intimate relationship of
fluctuations to the existence of a diffuse volume flux. In many practical situations this thermodynamic
R

37 incompressibility singularity, whether in the case of liquids or effectively isobaric gas transport processes, is
without appreciable effect on the accuracy of the fluid-mechanical predictions derived from solutions of the
R

39 classical Navier–Stokes and Fourier equations in this limit. In such circumstances the existence of the
singularity may be ignored with impunity. On the other hand, there exist a few key novel fundamental
O

41 phenomena—for example, thermophoresis [5] and thermal transpiration [50] in single-component gases, and
C

thermal diffusion-based Soret separation phenomena in multicomponent liquid mixtures [14,15]—whereby


43 ignoring the existence of this singularity would lead to fundamentally incorrect predictions, negating the very
N

existence of these physical phenomena.


45
U

8. Discussion
47
8.1. Brownian motion as the source of the deviation of the specific momentum from Euler’s constitutive
49 hypothesis

51 The issue of Brownian motion bears directly upon the fundamental question of whether or not the Cauchy
linear momentum equation (2) follows as an immediate consequence of Newton’s mechanics applied to mass-
PHYSA : 9965
ARTICLE IN PRESS
24 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 point molecular models, or whether second law of thermodynamic principles is needed in order to correctly
effect the transition from discrete to continuum mechanics. In the present single-component context the phrase
3 ‘‘Brownian motion’’ refers to the manifest consequences of multiple collisions occurring between a single
tagged molecule of the fluid and other molecules present therein. Explicitly, attention focuses on collisionally
5 induced changes occurring in the momentum and position of a tagged molecule over the long term. As regards
its role in transport phenomena, the emphasis here thus focuses upon the consequences of multiple collisions,
7 rather than upon the short-term consequences of single collisions (the latter as already currently embodied in
Boltzmann’s collisional integral in the particular case of rarefied gases).
9 The founders of gas-kinetic theory, including Clausius, Maxwell and Boltzmann, did not themselves identify
Brownian motion as a distinct macroscopic phenomenon arising from collisions of molecules [31], leaving it to
11 Einstein [56] and Smoluchowski [57] to later do so. This failure to incorporate the phenomenon into their
collisional model is, in our opinion, responsible for the presently held belief that Brownian motion is irrelevant
13 to the formulation of the subject of Boltzmann-based gas-kinetic theory, an attitude to which neither Öttinger
[23], Klimontovich [28–30], nor we, among others [58–60], subscribe. The collectively uniform attitude of this
15 latter group toward the issue can be gleaned from explicit remarks made by each on the role of Brownian
motion (respectively, focused on: fluctuations, self-diffusion and diffuse volume transport) in single-
17 component fluids during the course of translating Newton’s and Euler’s mass-point rigid-body mechanics into
continuum mechanics. This includes: (i) Öttinger’s ‘‘Something is missing’’ fluctuation addendum [23] to
19 earlier versions [25,26] of GENERIC theory; (ii) Klimontovich’s self-diffusion physical-space add-on [28–30]

F
to Boltzmann’s collision integral for the purpose of introducing irreversibility directly into mechanics; and (iii)

O
21 Brenner’s [4] recognition of the phenomenon of diffuse volume transport—the latter, like entropy, a statistical
rather than practical concept—over and above the previous, strictly convective, view that in single-component
23
O
fluids volume could be conveyed through space solely in the company of mass.
The seeming irrelevance of Brownian motion with regard to the foundations of gas-kinetic theory is
PR
25 rendered transparent by the obvious fact that in current rarefied gas theories [13,61] Brownian statistics do not
enter into the calculation of the singlet distribution function f ðx; p; tÞ (with x and p ¼ m dx=dt the respective
27 position and momentum vectors of a mass-point molecule of mass mÞ. Explicitly, the theory of Brownian
motion per se does not contribute directly to solving the Boltzmann equation, although the notion of
D

29 Brownian motion is implicit in the solutions thereof. This observation, in turn, indicates that the multi-
collision processes, which underlie the phenomenon of Brownian motion, play no role in the basic physics
TE

31 quantifying the macroscopic manifestation of molecular transport phenomena. Thus, philosophically


speaking, contemporary thinking argues that the notion of Brownian motion merely enriches the subject of
33 gas-kinetic theory without impacting directly upon its foundations. It is this short-term collisional perspective
EC

which Öttinger, Klimontovich and we challenge (see the discussion of Öttinger’s recent multicollisional model
35 in Section 8.4).
The failure of Brownian motion to impact upon the mechanical foundations of statistical mechanics—with
R

37 such motion viewed merely representing a completely predictable consequence thereof—should appear strange
to any unbiased observer unfamiliar with the apparently mechanically reversible treatment of the collisional
R

39 term in Boltzmann’s theory. That is, macroscopic experience teaches that collisions occurring among a
confined and isolated discrete collection of objects (‘‘molecules’’) separated by a vacuum are inherently
O

41 irreversible, ultimately causing such an isolated system to eventually come to a macroscopic state of rest via
C

‘‘friction’’, such as certainly occurs in granular gas models [62] lacking a continuous external supply of energy
43 (momentum). That this fate does not befall the molecules in Boltzmann’s collisional model must surely be
N

attributed to the phenomenon of Brownian motion, which should be viewed as the root cause making possible
45 the perpetual motion of (molecular) objects. This suggests that Brownian motion (namely self-diffusion and
U

fluctuations) should be regarded as an essential and heretofore overlooked contribution to kinetic theory,
47 rather than simply constituting a predictable consequence thereof. This attitude with regard to the role of
Brownian motion forms the basis of the implicit belief lying at the foundation of fluctuation-based GENERIC
49 theory [23] that the translation of point-mass Newtonian mechanics into continuum mechanics cannot be
correctly effected without explicitly incorporating entropy and its statistical-molecular foundations into the
51 translation scheme.
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 25

1 In this latter context it is interesting to contrast the rather different attitude displayed toward Brownian
motion when formulating the statistical foundations of quantum mechanics as a field theory. There, Nelson
3 [63] and others [64] have invoked fundamental ideas underlying the notion of Brownian motion in a ‘‘hidden-
variable,’’ Bohmian-like, attempt to show that quantum mechanics can be derived directly from little more
5 than Brownian motion concepts when combined with those of Hamilton–Jacobi dynamics.

7
8.2. Limitations of contemporary perturbation solutions of the Boltzmann equation for small Knudsen numbers
9
The agreement of rarefied gas solutions of the Boltzmann equation with the N–S–F equations in the so-
called near-continuum small Knudsen number regime, such as in the well-known perturbation solutions
11
thereof due to Chapman and Enskog [13], is often cited as convincing evidence of the success of these
perturbation schemes and, indeed, of the fundamental correctness of the Boltzmann equation itself. However,
13
such schemes are implicitly based upon the fact that rarefied gases obey the ideal gas law. But, according to the
theory advanced here, as embodied in the union of Eqs. (59) and (36), an ideal gas constitutes a highly singular
15
case owing to the fact that the contribution of the diffuse volume flux jv to the internal energy or heat flux
vanishes identically for such gases (although the corresponding diffuse volume contribution to the entropy flux
17
in Eq. (60) is not correspondingly singular for ideal gases). As such, the fact that our modified N–S–F
equations might appear to be in conflict with Chapman–Enskog theory needs to be placed in context. This
19
issue is already implicit in the role of the Maxwell–Burnett thermal stresses, discussed in connection with Eq.

F
(49).

O
21
In effect, it is the failure of theories of the Boltzmann equation [13] to unambiguously distinguish between
the heat flux jq and the diffuse internal energy flux ju , especially in single-component gases, that constitutes the
23
O
source of the problem. The problem is exacerbated in multicomponent gas mixtures, where, for example,
Chapman and Cowling [65] refer to ‘‘. . . the ordinary [my emphasis] flow of heat resulting from inequalities of
PR
25
temperature in the gas,’’ while concomitantly speaking of an additional heat flow due to diffusion (the Dufour
effect). Reciprocally, this heretofore unresolved heat flux ambiguity has resulted in the failure of gas-kinetic
27
theory to recognize the fundamental role played by the diffuse volume flux in properly interpreting the
hierarchical ordering of the sequential Knudsen number-based terms arising in perturbation solutions of the
D

29
Boltzmann equation for rarefied gases. This is not to state that the Boltzmann equation itself is in error, but
TE

rather that one must rigorously avoid supposing that the heat flux and diffuse internal energy flux are
31
synonymous if, at the same time, Fourier’s law, Eq. (42), is to be accepted as generally valid.
33
EC

8.3. Thermodynamic singularities in fluid mechanics


35
Thermodynamics and mechanics have traditionally been regarded as essentially separate and distinct fields
R

37 of inquiry, except in relation to the second law of thermodynamics, although GENERIC [23] as well as
‘‘extended thermodynamics’’ [66] among other nonequilibrium schemes represent attempts at their unification.
R

39 As regards the second law, the question of the irreversible nature of thermodynamics in contrast to the
seemingly reversible nature of Newtonian dynamics has long both intrigued and confounded physicists as well
O

41 as other scientists interested in the fundamentals of their disciplines, with attempts at the resolution of this
C

seeming paradox as a favorite philosophical topic.


43 As in the case of GENERIC [23], the present paper raises questions about whether continuum mechanics
N

and continuum thermodynamics can be truly separated into distinct branches of physics. Resolution of the
45 question leads, inter alia, to the surprising conclusion that the diffuse flux of volume renders continuum fluid
U

dynamics a branch of irreversible thermodynamics rather than of Newtonian dynamics (a conclusion which
47 will be evidenced more forcefully in subsequent installments in this series). The point to be made is that while
Newtonian mechanics is indeed applicable to molecules, and thus subject to the laws of dynamics, the Cauchy
49 linear momentum equation lacks a dynamical (i.e., molecular) basis owing to the presence therein of the stress
tensor, a strictly continuum concept. As such, the present series of papers will advance the view that the
51 Cauchy linear momentum equation is, in fact, an irreversible thermodynamic relation rather than a Newton’s
law-based dynamical relation. While the case for this unusual perspective may not seem wholly convincing in
PHYSA : 9965
ARTICLE IN PRESS
26 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 the present paper, it is believed that subsequent papers in the series will bring the issue home more pointedly
when we divorce the proposed diffuse volume-based addition to LIT from its present GENERIC ancestry.
3 The fundamental role played by thermodynamics in the present analysis of the Cauchy specific momentum
density v can be seen, for example, in the important role assigned in the present work to Onsager’s reciprocal
5 relation. Prior to this, to the best of the author’s knowledge, Onsager’s theorem had never been cited as being
relevant to any fluid-mechanical issue of a dynamical nature, especially that pertaining to the issue of
7 momentum.
The linkage here between fluid mechanics and thermodynamics is subtle, as can be seen most clearly by
9 considering the case where the fluid is isothermal. In that case, in particular in the present class of single-
component fluids being investigated, only the respective continuity and Cauchy linear momentum equations,
11 (1) and (2), enter the analysis. As such, only the purely ‘‘mechanical’’ variables p, r and vm would normally be
expected to arise in traditional views pertaining to this isothermal case. Accordingly, lacking the need for an
13 energy equation, whose presence would otherwise serve to couple fluid mechanics to thermodynamics in an
obvious way, one would not normally envision the existence of any common ground between these two fields
15 of study in the present isothermal instance, except perhaps in the seemingly minor context of the second law of
thermodynamics—the latter as embodied in the Clausius–Duhem inequality serving to demonstrate the
17 nonnegativity of the respective shear and bulk viscosity coefficients for rheologically Newtonian fluids (cf.
Eqs. (80) and (44)). Yet, despite this belief, these two fields remain inseparably linked in the isothermal case
19 through the diffuse volume flux jv , as given constitutively by Eq. (54). This flux enters the fluid-mechanical

F
portion of the analysis through the union of the Cauchy linear momentum equation (2) with Eqs. (56) and

O
21 (53). At the same time, the diffuse volume flux enters the thermodynamic aspect of the analysis through its
appearance in Eqs. (59)–(61). It is through this common presence, which transcends the issue of isothermality,
23
O
that these two fields are permanently linked despite the fluid being isothermal. The sole exception occurs in the
‘‘incompressible’’ fluid case, where the uniformity of the density results in the fact that jv ¼ 0.
PR
25 Here, however, were the isothermal fluid to be truly incompressible, the pressure would have to be uniform
throughout the fluid in accordance with the single-component equation of state, p ¼ pðT; rÞ, since T and r are
27 both constant. This, however, flies in the face of the fact that a pressure gradient rp normally exists in
isothermal flow situations involving incompressible fluids. Moreover, since only the pressure gradient, rather
D

29 than the pressure itself, appears in the equations of fluid mechanics, incompressible fluid mechanics, by itself,
can establish the prevailing pressure at a point of the fluid only to within an arbitrary additive, generally time-
TE

31 dependent, function. Obviously, one is dealing here with a highly singular situation [55]. This point clearly
comes to the fore in the person of the diffuse volume flux, as can be seen from the role played by jv when
33 addressing the Onsager coupling issue during the course of attempting to determine the heat flux jq (even when
EC

the latter is identically zero, as in the isothermal case).


35 The role of the diffuse flux of volume, especially in relation to the precise definition of the heat flux offered
here—thereby contributing to the clarification of this reversibility–irreversibility paradox—has remained
R

37 hidden until now. In retrospect, the reasons for this failure to recognize the existence of diffuse volume
transport, much less its major unifying role, are now obvious, although these reasons are different in the
R

39 respective gas and liquid cases. Though different in detail, the reason in both cases can be traced to the
singular nature of the respective (perturbation) approximations normally made in the literature of gases and
O

41 liquids.
C

In the case of gases, the Boltzmann equation, with the available solutions thereof largely focused on rarefied
43 gases (these obeying the ideal gas law), has, due to this focus, implicitly eliminated the need to clearly
N

distinguish between heat flow and (diffuse) energy flow. This can be seen from Eq. (63) where the distinction
45 disappears owing to the fact that the internal energy of an ideal gas is independent of its volume. As pointed
U

out in Section 8.1, we now recognize the latter condition as a singular limit, in the sense that in such
47 circumstances jv is no longer available to stimulate discussion of a possible distinction existing between jq and
ju . Yet a profound philosophical difference exists between the two, since, as earlier stressed, ju represents the
49 flux of an extensive physical property, namely the internal energy U, whereas there exists no extensive physical
property, namely heat Q, of which jq could rationally be called its flux (nor does there exist a volumetric
51 density, say cq , of heat). As such, the heat flux must be regarded as a slack variable, as in Eq. (63), namely the
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 27

1 residue remaining after subtracting from the diffuse internal energy flux all other sources that entrain internal
energy (e.g., diffuse volume) other than in the form of temperature (see Eq. (111) below).
3 In the cases of liquids it is the ambiguous notion of ‘‘incompressibility’’ (and the ensuing uncertainty in
regard to the notion of pressure) which is the source of the singularity. This is immediately apparent from Eq.
5 (99), where strict incompressibility would require that both r ln r and kT be zero, resulting in a Leibnitz-like
mathematical indeterminacy with regard to whether jv was, or was not, zero. This in turn reflects upon the
7 latter’s role in connection with Onsager’s reciprocal theorem and, hence, upon the same heat flow/diffuse
internal energy flow conundrum as cited above in the case of gases.
9

11
8.4. An alternative choice for Öttinger’s a0 coefficient
13
The basis for the LIT analysis of Section 7 derives largely from our identification of Öttinger’s GENERIC
15 fluctuation-based phenomenological coefficient a0 as being expressed in terms of the physical properties of the
fluid by Eq. (33). This relation, in turn, arose by applying our fundamental hypothesis (27) to Eq. (20), the
17 latter relating the velocity difference, v  vm  j, to a0 . Other seemingly plausible hypotheses might have led to
alternative expressions for a0 . Interestingly, one of these, due to Öttinger [67], leads in the special but
19 important case of ideal gases to a formula for a0 which is constitutively identical to our Eq. (28). Explicitly,

F
were it to have been supposed in Eq. (20) that a0 ¼ u=r, this would have led to the relation

O
21
D0  p 
j¼ r , (102)
r2 T
23
O
a formula derived theoretically by Öttinger [67] based on a ‘‘proper cross graining’’ of the Boltzmann
PR
25
equation. However, since p=T ¼ rR=M w for ideal gases (with R the universal gas constant and M w the
molecular weight) the above equation is seen to be constitutively identical to our Eq. (28), wherein
27
M wr
D

29 D0 ¼ K. (103)
R
TE

31 Thus, were the parameter K to be identified as being equal to the gas’s thermometric diffusivity a, as in Eq.
(47), the constitutive relation thereby obtained for the velocity difference j would have been exactly what we
33 obtained for jv based on our fundamental hypothesis (27) (upon bearing in mind that kT ¼ M w R=rT for ideal
EC

gases). Obviously, the agreement between Öttinger’s Eq. (102) and our Eq. (28) is, despite their very different
35 origins, a consequence of the fortuitous cancellation of the last two bracketed terms in Eq. (33) in the case of
ideal gases.
R

37 Experimental data involving thermal diffusion in liquids [14,15] support the hypothesis that Eq. (27), and
hence (28), including Eq. (33), is not limited to gases, but applies equally well to liquids. However, the limited
R

39 availability of critical data pertinent to the issue renders confirmation of the general applicability of these
relations to liquids somewhat tenuous. As yet, the possible applicability of Eq. (102) to liquids has not been
O

41 tested. This owes to the absence of experimental data sufficient to the task, e.g., data in which static pressure is
C

imposed externally on a liquid undergoing steady-state heat conduction. Even were such data available, lack
43 of independent knowledge as to the possible effect of pressure on the phenomenological coefficient D0 for
N

liquids would appear to render the interpretation of such data equivocal. As such, it is not yet possible to
45 distinguish among the two possibilities for a0 , namely Eq. (33) vs. a0 ¼ u=r. In this context it needs to be kept
U

in mind that were the incompressibility hypothesis (27) leading to Eq. (33) to prove wrong, the present analysis
47 would, because it is based on the general principles of GENERIC, nevertheless remain intact in broad outline,
although not in fine detail. For example, were the constitutive expression for a0 to be kept open throughout the
49 entire development, we would, more generally, in place of Eq. (63), propose the following definition for the
heat flux: jq :¼ju þ ða0 r  uÞj
^ v . Thus, were it to prove true that a0 ¼ u=r,
^ the following would obtain for the
51 proposed heat flux definition: jq :¼ju .This is, of course, the usually assumed constitutive relation (or definition)
of jq for single-component fluids.
PHYSA : 9965
ARTICLE IN PRESS
28 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 8.5. Solids

3 Our analysis has focused exclusively on transport processes occurring in fluids—gases as well as liquids.
However, the basic precepts of LIT and GENERIC apply to matter generally, irrespective of its macroscopic
5 physical state. As such, there is every reason to suppose that fundamental fluctuation-based issues similar to
those discussed above in the context of fluids also arise in the case of solids, particularly solid solutions
7 (alloys). This belief is supported by the work of Danielewski and his co-workers [68,69], who have repeatedly
expressed the view, as explicitly quantified in their analyses, that the Cauchy linear momentum equation
9 necessitates a two-velocity formulation when applied to solids undergoing interdiffusion, just as in the case of
fluids. Concomitantly, they point to the need to modify the classical model of strain-induced stress by adding
11 volume transport to simple strain-based displacement when formulating constitutive equations for stresses in
solids arising from diffusion (so-called diffusion-induced stresses). Explicitly, similar to the views expressed
13 here, Danielewski et al. [68,69] argue in favor of the existence of a second fundamental velocity, a so-called
‘‘drift’’ velocity, different from the mass velocity of the solid, and appearing (together with the mass velocity)
15 in the momentum conservation and stress constitutive equations for solids. Moreover, just as in our case, their
drift velocity arises as a direct consequence of volume transport, the latter occurring in solids during the course
17 of ‘‘atom–vacancy’’ exchanges within the solid lattice and/or via atom–atom exchanges when the diffusing
species possess different molar volumes [70]. (Readers not familiar with the exhaustively detailed terminology
19 employed in connection with atomic transport in solids may find it useful to refer to the IUPAC-recommended

F
publication: ‘‘Definition of terms for diffusion in the solid state’’ [71].)

O
21 Experimental justification for their nontraditional momentum transport model is based upon the widely
accepted Darken [72]–Kirkendall [73] notion of diffusion-induced stress [74–76] resulting from volume
23
O
transport accompanying interdiffusion in multispecies solids [77]. Being based, more or less, exclusively on
macroscopic experimental phenomenology, their volume-transport stress model, due to Stephenson [74], lacks
PR
25 the fundamental molecular foundation that we have earlier provided for fluids, at least for single-component
gases, and manifesting itself in the notion of temperature-induced stresses. Nevertheless, despite being less well
27 grounded theoretically than in the case of fluids, Danielewski et al.’s [68,69] two-velocity momentum transport
model for solids appears to be well-supported macroscopically. Indeed, there exists a vast body of literature
D

29 concerned with the role of ‘‘volume diffusion’’ [70] in rationalizing the phenomenon of diffusion-induced stress
[75,76], the latter manifested explicitly by the permanent deformation of solids noted at the conclusion of the
TE

31 diffusion process.

33 8.6. Preview of non-simple and multicomponent fluid systems


EC

35 Eqs. (59)–(61) were derived on the basis of Öttinger’s version [23] of GENERIC supplemented by several
key constitutive assumptions. However, once derived, this equation set is seen to collectively possess an
R

37 obvious physical interpretation in its own right, independent of its GENERIC origin. As such, these
fundamental equations, together with the implications that follow therefrom, stand or fall on their own
R

39 respective merits. Subsequent papers in this series will build upon generalizations stemming from the
O

foundations laid by this trio of equations.


41 Eqs. (59)–(61) are obviously valid only for simple, single-component systems. However, given their structure
C

and interrelationships it is not difficult to speculate on how this trio of equations might be generalized so as to
43 be applicable in more complex circumstances. Such speculations will be confirmed in subsequent papers
N

appearing in this series. These generalizations will be seen to be wholly independent of the GENERIC theory
45
U

[23] that spawned them. The material which follows immediately below is designed to provide a brief preview
of the scheme underlying these proposed generalizations.
47 The previous single-component case dealt with circumstances in which, for a fixed mass M, the
thermodynamic state of the system could be described exclusively in terms of the extensive variable set
49 ðU; S; V Þ. This allowed the combined first and second laws to be expressed entirely in terms of these three
variables as
51
dU ¼ T dS  p dV ðM ¼ const:Þ, (104)
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 29

1 wherein, in terms of these three variables,


   
qU qU
3 T¼ and p ¼  . (105a,b)
qS V ;M qV S;M
5 The simplest generalization of Eqs. (59)–(61) entails circumstances in which but a single additional extensive
variable, say C, is added to supplement the preceding set, so that, now, the pertinent extensive properties of
7 interest are ðU; S; V ; CÞ. In that case Eqs. (104) and (105) are to be, respectively, replaced by the following pair
of equations:
9
dU ¼ T dS  p dV þ mc dC ðM ¼ const:Þ (106)
11 and
     
qU qU qU
13 T¼ ; p¼ and mc :¼ . (107a,b,c)
qS V ;C;M qV S;C;M qC V ;S;M
15 The respective generalizations of Eqs. (59)–(61) are then, obviously,
   
qU qU
17 ju ¼ jq þ jv þ jc , (108)
qV T;C;M qC T;V ;M
19 "     #

F
1 q qS qS
js ¼ j þ jv þ jc (109)

O
21 T qV T;C;M qC T;V ;M

and
23
ju ¼ Tjs  pjv þ mc jc , O (110)
PR
25
where jc denotes the diffusive flux of the extensive property C (with the latter symbol denoting not only the
27 property itself, but also its amount). Obviously, the preceding can be generalized even further to the case
where there exists a multiplicity of additional extensive variables, say Ci ði ¼ 1; 2; . . . ; NÞ, resulting in the
D

29 extensive property set ðU; S; V ; C1 ; C2 ; . . . ; CN Þ.


The main point here is that the generalized definition of the heat flux, namely
TE

   
31 q qU qU
j :¼ju  jv  jc ðM ¼ const:Þ, (111)
qV T;C;M qC T;V ;M
33
EC

representing the extension of Eq. (63) to the present more general situation, is expected to result in the fact that
35 jq vanishes when the temperature T is uniform throughout the fluid, so that Fourier’s law, Eq. (42), will
continue to prevail irrespective of whether the other intensive variables, namely v^ and c^ (where c ^ ¼ C=MÞ,
R

37 vary throughout the fluid. It is this fact, among many, which will be demonstrated in subsequent
contributions.
R

39
9. Summary and commentary
O

41
C

9.1. Background, chronology and summary


43
N

According to the 250-year old view of Euler [33,34], the momentum velocity v appearing in the Cauchy
45 linear momentum equation (2) is equal to the fluid’s mass velocity vm , the latter being the velocity appearing in
U

the continuity equation (1). And Cauchy’s equation is precursor to the Navier–Stokes equations (as well as the
47 Fourier-based energy equation owing to the dissipative terms appearing therein arising from the stress tensor
in Cauchy’s equation). The goal of our paper was to demonstrate on theoretical grounds, using the basic
49 principles of LIT, especially as embodied in GENERIC [23], that Euler’s constitutive equation, v ¼ vm , is
incorrect in circumstances where density gradients exist in the fluid. This surprising result is implicit in the
51 unorthodox set of statistical–mechanically based continuum-level transport equations derived for single-
component ideal gases by the late Klimontovich [28–30] (see also [27]), although he never explicitly
PHYSA : 9965
ARTICLE IN PRESS
30 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 emphasized his disagreement with existing theory. A decade later, unaware of this prior work, the present
writer arrived, independently, at a set of equations [1–3] identical to those of Klimontovich by a very different
3 route, one based empirically on philosophical arguments stemming from the recognition that volume could be
transported purely diffusively in fluids [4]. The resulting unorthodox Cauchy momentum equation (2), in
5 which vavm in circumstances wherein the fluid density r is nonuniform (owing primarily to the presence of
temperature gradients), is implicitly supported, albeit somewhat tenuously, by purely macroscopic
7 experiments focused on the phenomenon of thermophoresis [5,14,15], where the issue of slip boundary
conditions at solid surfaces, dating back to Maxwell [11] in 1879, complicates the interpretation of the
9 experimental results. Whereas Klimontovich’s theory of fluid mechanics and heat transfer was limited in scope
to single-component (ideal) gases by the nature of the assumptions made in its derivation, the present writer’s
11 theory extended to liquids as well, including inhomogeneous ‘‘compressible’’ binary fluid mixtures embodying
density gradients, rra0, arising from spatial variations in composition.
13 Focusing solely on the single-component case common to both theories, the velocity difference was
predicted to be of the form v  vm :¼j, in which j was equivalent to the diffuse volume flux, j ¼ jv , with the latter
15 defined, physically, by the relation [4] jv ¼ nv  nm =r, in which nv and nm  rvm denote the respective Eulerian
fluxes of volume and mass through a surface element dS fixed in space (through which the fluid is flowing).
17 The diffuse volume flux was given constitutively by the expression [4] jv ¼ ar ln r, with a ¼ k=r^cp the
thermometric diffusivity, in which k is the thermal conductivity and c^p the isobaric specific heat. (In the work
19 of Klimontovich [27–30] the velocity disparity j was not identified in physical terms as being the diffuse volume

F
flux jv , although his constitutive formula for j is the same as that for our jv .) To the extent that the

O
21 Klimontovich/Brenner diffuse volume interpretation of the momentum–mass velocity difference proves to be
correct, their work provides a complete theory of single-component fluid mechanics and heat transfer, albeit
23
O
different from the orthodox Navier–Stokes–Fourier (N–S–F) versions accepted in the literature [18–21]. The
most stroking difference lies in the fact that a single velocity is no longer generally sufficient to characterize the
PR
25 kinematics, dynamics and energetics underlying irreversible thermodynamics. Accord between these respective
orthodox and unorthodox views appears to exist only for ‘‘incompressible’’ fluids, r ¼ const:
27 Euler’s proposed constitutive formula, v ¼ vm , for the specific momentum density predates, by about a
century, recognition of the existence of (mobile) molecules, as well as the codification of the first and second
D

29 laws of thermodynamics in the mid-1800s. Cauchy’s (1827) pre-molecular, pre-thermodynamic incorporation


of Euler’s relation into the linear momentum equation was predicated entirely on the basis of extending
TE

31 Newton’s laws of discrete rigid-body mechanics to fluid continua through the introduction of Cauchy’s stress
tensor (3), a continuum concept appearing in place of real, externally imposed, body forces exerted collectively
33 on the contents of a material domain. It is more or less universally believed that Boltzmann’s statistical
EC

mechanics has long since resolved any possible doubts in the matter of the constitutive equation for the
35 momentum density v in favor of Euler’s belief that it is the velocity vm of mass. However, objectively speaking,
the only extensive experimental justification of Boltzmann’s six-dimensional kinetic transport equation lies in
R

37 the apparent agreement of its small Knudsen number perturbation solutions [13] for rarefied gases with the
three-dimensional physical-space N–S–F hydrodynamic equations. But the most accurately executed and
R

39 extensive fluid mechanical experiments to date in support of the conventional form of the N–S–F equations
have involved incompressible and/or isothermal liquids, rather than the rarefied gases to which Boltzmann’s
O

41 theory applies.
C

Moreover, and perhaps equally importantly in view of Maxwell’s thermal creep slip condition [11] is the fact
43 that virtually all experiments have involved the use of no-slip boundary conditions imposed upon vm [78],
N

whereas the view we have advanced elsewhere [1–3] with regard to possible critical experiments [5,14,50]
45 involving fluids in which density gradients exist is that the no-slip boundary condition should be imposed,
U

instead, upon the volume vv , the latter being identical to the (total) volume flux nv . And it is only in the
47 uninteresting case of incompressible fluids that vv and vm are the same. Thus, despite the passage of many years
in which the N–S–F equations appear to have stood the test of time, and despite the virtually unanimous belief
49 that the Boltzmann equation unequivocally demonstrates the correctness of these equations (especially Euler’s
view that v ¼ vm Þ, the doubts raised here and elsewhere suggest the need for a careful reappraisal of the facts.
51 Öttinger, on becoming aware of these issues several years ago through reading the manuscript of a then, as
yet, unpublished version of Ref. [3] by the writer, recognized that ‘‘Something was missing’’ from earlier
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 31

1 versions of GENERIC [23], the latter but one of several competitive schemes [66] aimed, inter alia, at a
rational approach to the physics of non-equilibrium thermodynamics. What he identified as being ‘‘missing’’
3 from the basic notions of irreversible thermodynamics was the physical manifestation of molecularly based
fluctuations upon the N–S–F equations, more generally of the basic equations of transport processes [18–21].
5 He proposed to rectify this omission by incorporating into the friction matrix appearing in GENERIC a
rational model quantifying these fluctuations. This led him, in effect, to pose Eq. (20) as a general constitutive
7 relation for j based on the principles embodied in GENERIC. Two unknown parameters, D0 and a0 , appear in
this expression. Establishing the values of these two parameters, such as to render Öttinger’s fluctuation model
9 consistent with the work of Klimontovich and the present writer, formed the heart of the present paper.
In this context it must be borne in mind that the constitutive formula j ¼ jv ¼ ar ln r by which we
11 determined D0 and a0 in the present paper has not been independently verified by others, so that outstanding
issues remain. The key to reconciling Öttinger’s model with our formulas, namely that appearing in the
13 preceding sentence in conjunction with the definition jv :¼vv  vm of the diffusive volume current, lay in the
‘‘incompressibility hypothesis,’’ Eq. (27), the latter based on our belief that the N–S–F equations are likely to
15 be valid for fluids of uniform density (at least for single-component fluids). This belief derived from a number
of sources—experimental, theoretical and philosophical—as already outlined in our earlier papers [1–3], again
17 subject to the caveat of there yet being no independent verification of our notions. In any event, the hypothesis
embodied in Eq. (27) eventually led to the respective expressions for a0 in Eq. (33) and
19
D0 ¼ r2 TkT a (112)

F
(the latter derived from Eqs. (31) and (47)), with kT the isothermal compressibility, Eq. (29). By these means

O
21
we explicitly demonstrated the compatibility of our equations with Öttinger’s fluctuation model, without,
23
statistical–mechanical grounds, pointed to another possibility. O
however, excluding other possibilities. Indeed, as discussed in Section 8.4, Öttinger has, himself, on theoretical
PR
25 Were that all, the paper could effectively have concluded without the appearance of Section 7. However, the
exercise of establishing the indicated compatibility unearthed an important philosophical fact, one otherwise
27 hidden in the formal, strictly mathematical, aspects of our calculations. This refers to the fundamental reason
as to why seemingly minute fluctuations, which would normally be expected to average-out statistically,
D

29 should have such a profound effect as to set aside 250 years of unquestioned acceptance of Euler’s view?
Physically, the answer to this rhetorical question lies in the fact that the fluctuations of the molecules about
TE

31 their average positions, acting in concert with their inhomogeneous spatial distribution, create a macroscopic
bias [15]. In effect, the phenomenon is a consequence of the coupling of these two attributes, namely
33 fluctuations and inhomogeneities, the latter as manifested in the molecular number density gradient (which, in
EC

single-component systems, translates into a mass density gradient). Their union underlies the inseparable
35 coupling that exists between dynamics and thermodynamics, whose composite nature lies at the very heart of
GENERIC. Indeed, the analysis of Section 7 reveals the hidden role that Onsager’s celebrated coupling
R

37 theorem [79] plays in understanding why Euler’s view was not tenable once the molecular nature of mobile
fluid matter was recognized, since Onsager reciprocity theorem brings together the microscopic or molecular,
R

39 and the macroscopic or continuum. This fluctuation/inhomogeneity argument as the root cause of the
breakdown of Euler’s hypothesis is essentially physical in nature, and hence intuitive and informal. In what
O

41 follows, we show mathematically (and thus formally) why and how Onsager coupling undermines the
C

possibility of there being but a single velocity in fluid mechanics.


43
N

9.2. On the mathematical impossibility of Euler’s relation v ¼ vm being valid when viewed in the light of Onsager
45 coupling: thermodynamics vs. mechanics
U

47 From a strictly mathematical viewpoint the necessity for the presence of two velocities rather than one in the
Cauchy linear momentum equation (2) can be regarded as formally arising from the need to increase the
49 number of independent vector variables appearing therein in the formulation of the overall fluid-mechanical
problem in order to accommodate the additional restriction imposed by Onsager coupling (with the new
51 independent variable j, denoting the difference v  vm between these two velocity fields, as in Eq. (8)). This
restriction arose from the presence in the set of independent LIT-based ‘‘force–flux’’ Onsager relations, Eqs.
PHYSA : 9965
ARTICLE IN PRESS
32 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 (82) and (83), of an additional vector flux, namely the diffuse volume flux jv  j (given constitutively by the
relation jv ¼ ar ln r, with a the thermometric diffusivity), above and beyond the usual internal energy (or
3 heat) flux vector—the need for which had not been previously recognized. Were but a single velocity to have
appeared in Cauchy’s equation, the restriction imposed by the Onsager coupling relation (85) would have
5 resulted in the overall problem being over-determined owing to the resulting disparity between the number of
independent variables and the number of independent relations existing among them. As a further
7 consequence, the definition, jv :¼vv  vm , of the diffuse volume flux (with vv the volume velocity) then led to the
constitutive relation v ¼ vv for the momentum velocity or specific momentum density.
9 This general Onsager-based inequality vavm , impacting on Euler’s hypothesis, holds independently of the
correctness of the relation j ¼ jv , whose validity has not yet been unequivocally confirmed [67], as discussed in
11 Section 8.4. What appears to be certain, however, contrary to popular belief dating back to Euler, is that the
Cauchy linear momentum equation cannot be derived solely by dynamical arguments based simply upon
13 applying Newton’s laws of motion to a fixed mass of fluid viewed as a continuous body (a so-called material
domain) moving through space. Rather, because of fluctuations in the instantaneous contents of such a
15 domain, stemming from the molecular constitution of matter, non-equilibrium macroscopic thermodynamic
principles deriving from the classical statistical–mechanical work of Onsager necessarily enters into
17 consideration. This perspective serves to inseparably link together continuum mechanics and non-equilibrium
thermodynamics in a manner that has not previously been explicitly recognized in LIT [19–21], although
19 already implicit therein, as well as in the more broadly based structure of GENERIC [23].

F
In turn, this inseparable linkage of mechanics to thermodynamics, with the notion of mechanical work

O
21 common to both fields, led naturally in our paper to fundamental questions about the definition of heat—
questions which, in our view, had not previously been satisfactorily addressed, much less answered. Intimately
23
O
related thereto was the issue of the limits of applicability, if any, to the range of temperature gradients over
which Fourier’s law of heat conduction would be valid. Heretofore, the existence of definitive limits of
PR
25 applicability had been implied, despite the apparent lack of experimental data or rational theoretical argument
indicating any such limitation [52]. These issues, which might appear to border on the strictly philosophical,
27 will be discussed in subsequent papers aimed at attempting to extend the thermodynamic analysis of Section 7
to include inhomogeneous multicomponent fluid mixtures as well as more complex single-species fluids than
D

29 those discussed here?.


TE

31

33 Acknowledgments
EC

35 This work is the outgrowth of a dialog begun several years ago with Prof. Hans Christian Öttinger of the
Department of Materials, Institute of Polymers, of the Swiss Federal Institute of Technology (ETH) in Zürich,
R

37 following his receipt of a preliminary draft of the work cited in Ref. [3] questioning Euler’s constitutive
expression for the specific momentum. As clearly set forth in the ‘‘Something is missing’’ addendum to his
R

39 recent book [23], it was he who first recognized the commonality of our respective unorthodox views of the
current status of transport processes, while implicitly encouraging me to remain firm in my beliefs when faced
O

41 with the discouraging views of disbelieving referees. Our common views were recently brought jointly to
C

fruition during a collaborative visit to his Institute in June 2005. It was H.C.O.’s insight, as recorded in his
43 book [23], that provided the theoretical framework resulting in this paper, enabling the constitutive relation
N

for the specific momentum density embodied in Eq. (56) to be rationalized—a relation in whose correctness I
45 believed deeply intuitively at the outset [1] of my research on the role of diffuse volume transport [4] in fluid
U

mechanics and thermodynamics. Also sharing in the initial phases of the Clausius–Duhem/momentum density
47 issue was my former student and collaborator, Dr. ‘‘Jim’’ Bielenberg, who was a significant contributor to the
evolution of the thinking reflected in the present paper on issues of volume transport and the consistency
49 thereof with the second law.

51
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 33

1 References

3 [1] H. Brenner, Unsolved problems in fluid mechanics: on the historical misconception of fluid velocity as mass motion, rather than
volume motion, in: L.-S. Fan, M. Feinberg, G. Hulse, T.L. Sweeney, J.L. Zakin (Eds.), Unsolved Problems in Chemical Engineering,
Proceedings of the Ohio State University, Department of Chemical Engineering Centennial Symposium, April 24–25, 2003, pp.
5 31–39. This reference is available on-line as hhttp://www.che.eng.ohio-state.edu/centennial/brenner.pdfi, while the lecture on which
the reference is based can be seen on-line as a streaming video presentation at hhttp://www.chbmeng.ohio-state.edu/centennial/i.
7 [2] H. Brenner, Phys. Rev. E 70 (2004) 061201
H. Brenner, Inequality of the tracer and mass velocities of physicochemically-inhomogeneous fluid continua, J. Fluid Mech., 2006,
submitted for publication.
9
[3] H. Brenner, Physica A 349 (2005) 60.
[4] H. Brenner, Physica A 349 (2005) 10.
11 [5] H. Brenner, J.R. Bielenberg, Physica A 355 (2005) 251.
[6] M.N. Kogan, Ann. Rev. Fluid Mech. 5 (1973) 383.
13 [7] V.S. Galkin, M.N. Kogan, O.G. Fridlander, Izv. AN SSSR, Mekh. Zhidk. Gaza 5 (1970) 13;
M.N. Kogan, V.S. Galkin, O.G. Fridlander, Sov. Phys. Usp. 19 (1976) 420.
[8] M.N. Kogan, Non-Navier–Stokes gas dynamics and thermal-stress phenomena, in: Rarefied Gas Dynamics, vol. 15, 1986, p. 15.
15 [9] M.N. Kogan, Some solved and unsolved problems in kinetic theory, in: A.D. Ketsdever, E.P. Muntz (Eds.), Rarefied Gas Dynamics,
23rd International Symposium, American Institute of Physics, 2003.
17 [10] A.V. Bobylev, J. Stat. Phys. 80 (1995) 1063.
[11] J.C. Maxwell, Phil. Trans. R. Soc. London A 170 (1879) 231 reprinted in: W.D. Niven (Ed.), The Scientific Papers of James Clerk
19 Maxwell, vol. 2, Cambridge University Press, Cambridge, 1890, p. 681.

F
[12] D. Burnett, Proc. London Math. Soc. 39 (1935) 385;
D. Burnett, Proc. London Math. Soc. 40 (1936) 382.

O
21 [13] S. Chapman, T.G. Cowling, The Mathematical Theory of Non-Uniform Gases, third ed., Cambridge University Press, Cambridge,
1970.
23 [14] J.R. Bielenberg, H. Brenner, Physica A 356 (2005) 279.
[15] H. Brenner, Phys. Rev. E 72 (2005) 061201.
[16] In earlier publications [2,3] we used the symbol m
O
^ for the quantity here denoted by v.
PR
25 [17] J. Happel, H. Brenner, Low Reynolds Number Hydrodynamics, Prentice-Hall, Englewood Cliffs, NJ, 1965.
[18] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena, second ed., Wiley, New York, 2002.
27 [19] S.R. De Groot, P. Mazur, Non-Equilibrium Thermodynamics, North-Holland, Amsterdam, 1962.
[20] R. Haase, Thermodynamics of Irreversible Processes, Dover, New York, 1990 (reprint).
D

29 [21] G.D.C. Kuiken, Thermodynamics of Irreversible Processes: Applications to Diffusion and Rheology, Wiley, New York, 1994.
[22] The overbar operator is defined such that for any dyadic D, the dyadic D:¼ð12ÞðD þ DT Þ  ð13ÞI : D represents its symmetric traceless
TE

counterpart.
31 [23] H.C. Öttinger, Beyond Equilibrium Thermodynamics, Wiley, Hoboken, New Jersey, 2005 The apt phrase, ‘‘Something is missing,’’
which neatly encapsulates the momentum density problem, is taken verbatim from p. 61 of this reference.
33 [24] ‘‘If your theory is found to be against the second law of thermodynamics, I give you no hope; there is nothing for it [your theory] but
EC

to collapse in the deepest humiliation,’’ in A.S. Eddington, The Nature of the Physical World, Macmillan, New York, 1928, p. 74;
‘‘[Thermodynamics] is the only physical theory of universal content which I am convinced that within the framework of applicability
35 of its basic concepts will never be overthrown.’’ (A. Einstein), quoted in M.J. Klein, Science 157 (1967) 509.
[25] M. Grmela, H.C. Öttinger, Phys. Rev E 56 (1997) 6620.
R

37 [26] H.C. Öttinger, J. Non-Equilib. Thermodyn. 57 (1997) 386;


H.C. Öttinger, Phys. Rev. E 57 (1998) 1416;
R

39 H.C. Öttinger, J. Non-Equilib. Thermodyn. 27 (2002) 105.


[27] H. Brenner, originally submitted to Phys. Rev. E (March 2003) under the title ‘‘A molecular basis for the Euler/Lagrange velocity
O

disparity. The demise of the Navier–Stokes paradigm.’’ Revised version to be submitted to Physica A (2005) under the title: ‘‘On
41 Klimontovich’s proposed modification of the Boltzmann equation and its consequences for the Navier–Stokes–Fourier equations.’’
C

[28] Yu.L. Klimontovich, Theor. Math. Phys. 92 (1992) 909.


43 [29] Yu.L. Klimontovich, Theor. Math. Phys. 96 (1993) 1035.
N

[30] Yu.L. Klimontovich, Statistical Theory of Open Systems, Volume 1: A Unified Approach to Kinetic Descriptions of Processes in
Active Systems, Kluwer Academic Publishers, Dordrecht, 1995.
45
U

[31] Mazo, in the historical background to his book on Brownian motion (R.M. Mazo, Brownian Motion: Fluctuations, Dynamics and
Applications, Clarendon Press, Oxford, 2002) discussing the period between Robert Brown’s observations in 1829 and Einstein’s 1905
47 publication, states that (p. 3): ‘‘It is striking, however, that the founders and main developers of kinetic theory, Maxwell, Boltzmann
and Clausius, never published anything on Brownian motion.’’
[32] D. Straub, Alternative Mathematical Theory of Non-Equilibrium Phenomena, Academic Press, San Diego, 1997.
49
[33] L. Euler, Mém. Acad. Sci. Berlin 11 (1755) 274 (reproduced in Leonhardi Euleri Opera Omnia, Series II, vol. 12, p. 54 (Füssli, Zürich,
1954)). Additional historical information can be found in the ‘‘Editor’s Introduction’’ to the latter volume by C. Truesdell, ’’Rational
51
PHYSA : 9965
ARTICLE IN PRESS
34 H. Brenner / Physica A ] (]]]]) ]]]–]]]

1 fluid mechanics, 1687–1765,’’ pp. VII–CXXV;


L. Euler, Hist. Acad. Berlin 1755 (1757) 316.
3 [34] A concise history of the conceptual foundations of fluid mechanics, from the time of Newton’s Principia in 1687 up to the definitive
work of Stokes in 1845 [35], can be found in the following article: C. Truesdell, Amer. Math. Monthly 60 (1953) 445;
O. Darrigol, Arch. Hist. Exact Sci. 56 (2002) 95;
5 O. Darrigol, Worlds of Flow, Oxford University Press, Oxford, 2005.
[35] G.G. Stokes, Trans. Camb. Phil. Soc. 8 (1845) 287
7 reprinted in: Mathematical and Physical Papers, vol. 1, Cambridge University Press, Cambridge, 1901, p. 75
An account of pre-1845 work on the Navier–Stokes equations can be found in G.G. Stokes, British Assoc. Advance. Sci. (1846) 1;
reprinted in Mathematical and Physical Papers, vol. 1, p. 157, Cambridge University Press, Cambridge, 1901. This refers to earlier
9 works, as follows: C.L.M.H. Navier, Mém. Acad. R. Sci. Paris 6 (1823) 389; S.D. Poisson, J. École Polytech. Paris 13 (1831) 139; and
B. Saint-Venant, C. R. Acad. Sci. Paris 17 (1843) 1240.
11 [36] L.D. Landau, E.M. Lifshitz, Fluid Mechanics, second ed., Butterworth-Heinemann, Oxford, 1987.
[37] P. Kostädt, M. Liu, Phys. Rev. E 58 (1998) 5535.
13 [38] Recall that a material domain is a hypothetical continuous body of fluid, each of whose surface points moves with the local mass
velocity vm [39]. This fact in conjunction with the continuity equation assures that no net mass crosses the surface of such a body at
any point. However, this is a strictly macroscopic statement in the sense that individual molecules are nevertheless free to cross a
15 material surface in either direction without violating this no-flux condition, provided only that in some time-average sense there are
no large-scale variations in the number of molecules contained within the domain. Clearly, the material view has nothing to say about
17 the role of such fluctuations in regard to their possible impact upon Newton’s laws of motion, which apply strictly only to permanent
collections of molecules.
[39] J.C. Slattery, Momentum, Energy, and Mass Transfer in Continua, McGraw Hill, New York, 1972.
19 [40] H. Goldstein, Classical Mechanics, Addison-Wesley, Reading, MA, 1950, p. 4;

F
L.D. Landau, E.M. Lifshitz, Mechanics, Addison-Wesley, Reading, MA, 1960.

O
21 [41] However, neither his work nor that of Klimontovich [28–30] touches upon the volume velocity-based no-slip boundary condition
issue, which plays a prominent role in our unorthodox interpretation of the thermophoretic motion of macroscopic particles [1–3,5].
[42] E.M. Lifshitz, L.P. Pitaevskii, Statistical Physics, Part 2, Pergamon Press, Oxford, 1980, p. 369.
23
[43] According to the chain rule, for any function f ðx; y; zÞ ¼ 0, O
PR
25         1
qx qy qz qy qx
¼ 1 and ¼ ; etc.
qy z qz x qx y qx z qy z
27
.
D

29 [44] J.G. Kirkwood, I. Oppenheim, Chemical Thermodynamics, McGraw-Hill, New York, 1961.
[45] Explicitly, the vector v is the momentum density per unit mass appearing in the total Eulerian momentum current dyadic, say
TE

nM ¼ nm v  T, implicit in the Cauchy linear momentum (2) (rewritten as qðrvÞ=qt þ r.nM ¼ rpÞ, of which nm v and T are its
31
respective convective and diffusive portions (with nm :¼rvm the Eulerian mass current). However, only the total Eulerian flux of an
extensive property is physically objective, and not the separate convective and diffusive portions into which it is eventually
33 decomposed [4]. Armed with this fact, consider the particular case where the momentum density v is homogeneous, so that rv ¼ 0. In
EC

such circumstances there would no diffusive momentum flux only a convective portion, despite the fact that no restriction is imposed
35 upon vm , thus allowing the possibility that rvm a0. As such, it is evident that the velocity appearing in the constitutive equation (44)
for the diffusive momentum current must, in general, represent the fluid’s momentum density v rather than its mass velocity vm .
[46] However, as discussed in the most recent edition of Ref. [18, cf. Table 19.2-4], such differences have recently been recognized in multi-
R

37 component fluids.
[47] C. Truesdell, R.G. Muncaster, Fundamentals of Maxwell’s Kinetic Theory of a Simple Monatomic Gas, Academic Press, New York,
R

39 1980.
[48] D.W. Mackowski, D.H. Papadopoulos, D.E. Rosner, Phys. Fluids 11 (1999) 2106.
O

[49] The leading term of Eq. (49) was originally derived by Maxwell [11], including the value K 1 ¼ 3 for Maxwell molecules.
41
[50] J.R. Bielenberg, H. Brenner, A continuum model of thermal transpiration, J. Fluid Mech. 246 (2006) 1.
C

[51] Although not relevant to the present arguments, it should be noted that Eq. (54) was originally viewed [4] as being valid only in
43 circumstances where pressure effects on the fluid’s density were small compared with temperature effects. However, according to our
N

present GENERIC derivation, no such restriction appears to exist, either for gases or liquids. Furthermore, it should be noted that
45 while the traditional form of the diffuse internal energy/heat flux relation, namely ju ¼ jq , rather than the GENERIC form (34), was
U

used in the original derivation [4] of Eq. (54), the two forms coincide in the case of ideal gases since ju ¼ jq ¼ krT in that case, as
follows from Eq. (34). Accordingly, the derivation of (54) is consistent with that of (52), at least in the case of gases.
47 [52] F. Bonetto, J.J. Lebowitz, L. Rey-Bellet, Fourier’s law: a challenge to theorists, in: A. Fokas, A. Grigoryan, T. Kibble, B. Zegarlinski
(Eds.), Mathematical Physics 2000, Imperial College Press, London, 2000, p. 128.
49 [53] We note upon comparing the latter relation with Eq. (31) that D0 ¼ r2 Lvv . This relation is consistent with the requirement (21) that
D0 X0.
[54] P.G. Debenedetti, Metastable liquids, Princeton University Press, Princeton, NJ, 1996.
51
[55] S. Ansumali, I.V. Karlin, H.C. Öttinger, Phys. Rev. Lett. 94 (2005) 080602.
PHYSA : 9965
ARTICLE IN PRESS
H. Brenner / Physica A ] (]]]]) ]]]–]]] 35

1 [56] A. Einstein, Ann. Phys. 17 (1905) 549 translated in: R. Fürth, (Ed.), A. Einstein, Investigations on the Theory of the Brownian
Movement, Dover, New York, 1956 (reprint).
3 [57] M.R. von Smoluchowski, Ann. Phys. 21 (1906) 756.
[58] R.F. Streater, Proc. R. Soc. London A 456 (2000) 205.
[59] In this connection, see also R.F. Streater, Statistical Dynamics, Imperial College Press, London, 1995;
5 R.F. Streater, J. Math. Phys. 38 (1997) 4570;
R.F. Streater, Rep. Math. Phys. 40 (1997) 557;
7 R.F. Streater, J. Stat. Phys. 88 (1997) 447;
R.F. Streater, Open Syst. Inf. Dyn. 10 (2003) 3.
[60] M.R. Grasselli, R.F. Streater, Rep. Math. Phys. 50 (2002) 13.
9 [61] J.O. Hirschfelder, C.F. Curtiss, R.B. Bird, Molecular Theory of Gases and Liquids, Wiley, New York, 1954;
H. Grad, Rarified Gas Dynamics, Pergamon, London, 1960;
11 M.N. Kogan, Rarified Gas Dynamics, Plenum Press, New York, 1969;
C. Cercignani, Mathematical Methods in Kinetic Theory, second ed., Plenum Press, New York, 1990;
13 Y. Sone, Kinetic Theory and Fluid Dynamics, Birkhäuser, Boston, 2002.
[62] N.V. Brilliantov, T. Pöschel, Kinetic Theory of Granular Gases, Oxford University Press, Oxford, 2004;
J. Javier Brey, M.J. Ruiz-Montero, R. Garcia-Rojo, Phys. Rev. E 60 (1999) 7174.
15 [63] E. Nelson, Dynamical Theories of Brownian Motion, Princeton University Press, Princeton, NJ, 1967 see also the revised online
second edition (2001) of this book at hhttp://www.math.princeton.edu/nelson/books/bmotion.pdfi;
17 E. Nelson, Quantum Fluctuations, Princeton University Press Princeton, NJ, 1985 see also hhttp://www.math.princeton.edu/nelson/
books/Quantum_Fluctuations_1985.pdfi.
[64] P. Garbaczewski, J.P. Vigier, Phys. Rev. A 46 (1992) 4634.
19 [65] Indeed, their need for the elaborate footnote appearing on p. 143 of their book [13] to explain the unexpected appearance of a factor

F
of 52 in place of 32 appears to be a manifestation of this omission, since the existence of a volume velocity vv , and, hence, a nonzero

O
21 diffuse volume flux jv in inhomogeneous multicomponent fluid mixtures, is well known in the irreversible thermodynamics literature
[19,20].
[66] D. Jou, J. Casas-Vázquez, G. Lebon, Extended Irreversible Thermodynamics, third ed., Springer, Berlin, 2001;
23
O
Muller, T. Ruggeri, Rational Extended Thermodynamics, second ed., Springer, New York, 1997;
D. Jou, J. Casas-Vázquez, G. Lebon, Rep. Prog. Phys. 62 (1999) 1035 see also hhttp://telemaco.uab.es/eit/home/principal.phpi.
PR
25 [67] H.C. Öttinger, ‘‘Hydrodynamics from Boltzmann’s kinetic equation after proper coarse-graining,’’ Phys. Rev. Lett., 2005, submitted
for publication.
27 [68] M. Danielewski, Defect Diffusion Forum 95–98 (1993) 125;
M. Danielewski, Defect Diffusion Forum 95–98 (1993) 673;
D

M. Danielewski, Solid. State Phen. 41 (1995) 63.


29 [69] M. Danielewski, W. Krzyzanski, Phys. Status. Solidi. 145 (1994) 351;
M. Danielewski, B. Wierzba, J. Phase Equil. Diffusion 26 (2005) 573.
TE

31 [70] Y.C. Chen, Y.G. Zhang, C.Q. Chen, Mater. Sci. Eng. A 368 (2004) 1;
S.M. Schwarz, B.W. Kempshall, L.A. Giannuzzi, Acta Materialia 51 (2003) 2765.
[71] M. Kizilyalli, J. Corish, R. Metselaar, Pure Appl. Chem. 71 (1999) 1307.
33
EC

[72] L.S. Darken, Trans. AIME 175 (1948) 184.


[73] A.D. Smigelskas, E.O. Kirkendall, Trans. AIME 171 (1947) 130.
35 [74] G.B. Stephenson, Acta Metall. 36 (1988) 2663.
[75] D.L. Beke, I.A. Szabó, Z. Erdélyi, G. Opposits, Mater. Sci. Eng. A 387–389 (2004) 4.
R

37 [76] W.J. Boettinger, G.B. McFadden, S.R. Coriell, R.F. Sekerka, J.A. Warren, Acta Materialia 53 (2005) 1995.
[77] J. Philibert, Atomic Movements, Diffusion and Mass Transport in Solids. Translated from the French by S.J. Rothman, Les Editions
R

de Physique, Les Ulis, Paris, 1991;


39 A.R. Allnatt, A.B. Lidiard, Atomic Transport in Solids, Cambridge, 1993.
O

[78] E. Lauga, M.P. Brenner, H.A. Stone, Dummy, in: J. Foss, C. Tropea, A. Yarin (Eds.), Handbook of Experimental Fluid Dynamics,
41 Springer, New York, 2005 (Chapter 15);
C

C. Neto, D.R. Evans, E. Bonaccurso, H.-J. Butt, V.S.J. Craig, Rep. Prog. Phys. 68 (2005) 2859.
43 [79] L. Onsager, Phys. Rev. 37 (1931) 405;
N

L. Onsager, Phys. Rev. 38 (1931) 2265.


U

You might also like