Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

i An update to this article is included at the end

Pergamon Chemical Enoineerin 9 Science, Vol. 52, No. 15, pp. 2457 2469, 1997
~ 1997 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
PII: S0009-2509(97)00065-1 0009-2509/97 $17.00 + 0.00

Locally averaged equations of motion for


a mixture of identical spherical particles
and a Newtonian fluid
R. Jackson
Department of Chemical Engineering, Princeton University, Princeton, NJ 08544, U.S.A.

(Accepted 14 September 1996)

Abstract--The equations of motion of the particles and the interstitial fluid are averaged using
a 'soft' spatial averaging procedure, thus generating equations relating local average values of
the fluid pressure, the fluid velocity, and the velocity and angular velocity of the particles. Terms
whose form is not related to the above average variables are shown to be expressible as average
values of tensor-weighted integrals of the forces exerted on the surface of a particle by the fluid,
or by contact with other particles. Other undetermined terms relate to momentum transport by
velocity fluctuations about the average values. It is shown that the equations can be reduced to
closed forms, entirely in terms of the averaged variables, in the limiting case of Stokesian
particles at very low concentration. © 1997 Elsevier Science Ltd

Keywords: Suspensions; rheology; averaging.

INTRODUCTION Prosperetti (1994).] Some authors have used local


spatial averages of the variables in the system of
In the past I have worked with John Villadsen on interest, taken over regions small compared with the
various aspects of chemical reactor theory. This paper length scale of the problems considered. Others have
describes something a little different but I hope he averaged at each point over an ensemble of macro-
may still find it of some interest, since it originated scopically equivalent systems, relying on an assump-
from problems associated with fluidized-bed reactors. tion of ergodicity to relate the ensemble averages to
In principle the motion of a suspension of solid measurements on a single system. Each approach has
particles in a Newtonian fluid is completely deter- advantages and drawbacks, but in cases where they
mined by requiring the Navier-Stokes equations to be lead to disparate results this is probably an indication
satisfied at each point of the fluid, and equating the that averaging does not provide an adequate descrip-
rates of change of the linear and angular momentum tion of the physical behavior, and consequently
of each particle, respectively, to the resultant force and closure of the equations in terms of the averaged
the resultant torque exerted on it. In practice not only variables is unlikely to be successful.
is this a problem of impractical complexity, but its The process of averaging itself is purely formal and
solution would provide more detailed information leaves a number of terms indeterminate. The problem
than is usually needed. Indeed, a knowledge of the of closure is then to express these in terms of the
average values of the velocity of the fluid, the velo- averaged variables themselves. There has been a
cities and angular velocities of the particles, and the variety of approaches to this, ranging from appeals to
fluid pressure, over some appropriately small region general principles of continuum mechanics to the ad-
in the neighborhood of each point of the system of option of relations based on experimental evidence.
interest, is usually all that is required. The question Clearly the undetermined terms depend on details of
therefore arises whether equations can be found the motion at the scale of the particles, and in the
whose solutions determine these average variables present paper they will be expressed explicitly in terms
without needing to know details of the motion on of point properties of the fluid, and variables asso-
a finer scale. In general there is no guarantee of the ciated with individual particles, so that the informa-
existence of closed relations of this sort, involving tion at the 'microscopic' level which is needed to effect
only average values, but an extensive literature has closure can be discerned clearly. Though in general
grown up around the process of averaging. [See, for this simply illuminates the great difficulties that face
example, Anderson and Jackson (1967), Drew (1971), any rigorous attempt at closure, it will be shown that
Drew and Segel (1971), Hinch (1977), Nigmatulin the process can be carried through to completion in
(1979), Joseph and Lundgren (1990) and Zhang and one simple limiting case, namely a suspension of rigid,
2457
2458 R. Jackson

spherical, Stokesian particles in a Newtonian, incom- aoerage, (f)f of a point property f (e.g. pressure, or
pressible fluid, at the ‘Einstein’ limit of high dilution. a component of velocity) is defined by
The method used will be local space averaging, as in
the early work of Anderson and Jackson (1967), but
Zhang and Prosperetti (1994, 1995) have recently de-
rived very similar results using ensemble averaging.
44Wf(x) =
s VJ
fWs(lx - Yl)dvy. (3)

The fact that both methods are in agreement lends Expressions for the fluid phase average values of both
added confidence in the results. space and time derivatives of point variables are easily
The motion of a suspension can be viewed in two obtained [see, for example, Anderson and Jackson
ways. Workers in the fields of fluidization and (1967)]:
gas-particle transport have sought separate equations
of motion for each of the phases, while those inte-
rested primarily in the rheology of suspensions of
small particles have viewed the suspension as a whole
as an effective medium. Communication between
these groups has not always been satisfactory, and it is - C ~~~.0y)c(,.)u(lx - yI)ds, (4)
P
hoped that the present work will contribute to a rec-
onciliation between the two viewpoints which are, of
course, entirely equivalent.

LOCAL SPACE AVERAGES

For simplicity we restrict attention to a suspension +


P
1
c Js, f(y)nk(y)Uk(yb(IX - YI)ds,
of identical spherical particles in an incompressible
fluid. If n is the particle radius and L the shortest
(5)
significant macroscopic length scale associated with
the motion of interest it is assumed that L>>a. This Both here and in what follows the convention is
will be referred to as the condition of separation of adopted that repeated suffixes are summed over the
scales. Local space averages are then defined with the values one to three. In eqs (4) and (5) the integrals are
help of a weighting function g(r) which is a monotone taken over the surface sp of particle p and the summa-
decreasing function of the separation between a pair tion is over all particles. nk denotes the kth component
of points of space, denoted by r. The integral of g over of the unit normal to the particle surface, directed out
all space is normalized to unity: of the particle, and uk is the kth component of the
m point velocity. These relations are valid provided the
4n g(r) r2 dr = 1. (1) distance of x from the nearest point of the boundary
I0 of the system is considerably larger than the radius of
g will also be assumed to be differentiable as often as the weighting function. It is important to remember
required in our manipulations, and its ‘radius’, 1 is that the same restriction will apply to the validity of
defined by any result derived using eq. (4) or eq. (5).

1 n
g(r) r2 dr = g(r)? dr. Solid phase averages
s0 sI This term will be used to describe averages formed
in a manner exactly analogous to the fluid phase
Then, provided 1is chosen to satisfy a<<l<<L, averages
averages just described, except the domain of integra-
defined using g should not depend significantly on the
tion is that part of the system occupied by particles at
particular functional form of g or its radius. Several
the time in question, and is therefore the union of the
different types of average will now be defined and
interiors up of the separate particles. Thus the solids
some of their properties will be quoted.
volume fraction 4 is defined by

Fluid phase averages


The volume fraction E occupied by fluid is defined d(x) = 1 j- dlx - YI)du, (6)
at point x by P VP

and the solid phase average (f)” of any point prop-


E(X) = 1g(lx -
Jv,
yl)dl/, (2) ertyf of the solid material then follows as

where the integral is taken over the whole of the space


occupied by fluid in the system, denoted by V,. (Of
course this domain of integration changes with time
so E is a function oft, as well as x, but time dependence Expressions analogous to eqs. (4) and (5) relate aver-
will not be exhibited explicitly in formulas unless this ages of space and time derivatives to corresponding
is necessary to avoid ambiguity.) Then thejuid phase derivatives of averages.
Locally averaged equations of motion 2459
Overall (volume) and mass-weighted averages fluid-phase average with the solids-phase average, not
The overall, or volume average ( f ) of a point the particle-phase average.
propertyfis defined by Finally, with local space averages as defined above,
we should note that a second averaging process does
(f) (x) -- fvf(y)g(Ix - yl) dVr not leave the value of an average variable exactly
unchanged. However, the difference between an aver-
= e ( x ) ( f ) f ( x ) + qS(x)(f)S(x) (8) age variable and the result of subjecting it to a second
averaging is small of O(12/L 2) relative to the value of
and the mass-weighted average ( f ) " by the average variable and it can, therefore, be neglected
fi(f)" = pfs(f) f + p~(f)s (9) under our assumption of separation of scales.

where pf and p~ are the densities of the fluid and solid AVERAGED EQUATIONS OF MOTION
materials, respectively, and fi = epl + dpps. These The point continuity equation for the fluid is
averages are important in formulating equations for
OUk/&Xk= 0 and, setting f = Uk in eq. (4) and f = 1 in
the motion of the mixture as a whole. eq. (5) then adding the results, this yields the following
averaged continuity equation for the fluid phase:
Particle-phase averages
The solid-phase averages defined above invoke the ~e 0
values of point variables, such as a component of + ~f~Xk(e(Uk)I) = 0 (13)
velocity or a component of the stress tensor, within
each particle. However, since each particle is assumed and there is a corresponding solid-phase averaged
to be rigid, its motion is completely determined by the continuity equation, namely
translation of its center and rotation about the center &b
and, correspondingly, the distribution of stress within + ~ (¢<uk>S) = 0. (14)
the particle is not needed to determine the motion;
only the resultant force and the resultant moment are In terms of particle-phase averages, on the other hand,
relevant. Consequently, as pointed out by Anderson the continuity equation is a conservation equation for
and Jackson (1967), it is useful to introduce averages the number density of particles, and it is obtained
which depend only on properties of the particle as immediately on setting f - 1 in eq. (12), giving
a whole, such as the velocity of its center of mass or its
On 0
angular velocity. These will be referred to as particle- + ff~xk(n(uk)v) = 0. (lS)
phase averages to distinguish them from the solid-
phase averages defined above. The number n of par- The point momentum equation for the fluid has the
ticles, per unit volume, at position x and a particular familiar form
time is defined by
F~bli~@k]OGikoy = -- k + Pfgi
n(x) = ~ g ( l x - xPl) (10) Ps L-~ + (u~uk)
p
where a~, are the components of the stress tensor and
where x p is the position of the center of particle p at g~ those of the specific gravity force vector. It is aver-
the time in question. The particle phase average ( f ) P aged by multiplying both sides by g ( l x - yl) then
of any property fP of the particle as a whole is then integrating over V¢. By using eq. (5) with f - ui and
given by eq. (4) with f -= UiUkon the left-hand side of the result,
and eq. (4) with f = aik on the right-hand side, we
n(x)(f)V(x) = ~fVg(lx - xPl). (11) obtain
p

For this type of average there is no analog of eq. (4), I + 0 (~(u~uk)i)] =


since the averaged quantity fP is not defined point-
wise, but Anderson and Jackson (1967) have shown
that the analog of eq. (5) is -- ~ ~ aik(y)nk(y)g(lx -- yl)dsy + pfegi (16)
p ~s,
0 which is the averaged momentum equation for the
n(x) @f/&)P(x) = ~-~[n(x)(f)P(x)]
fluid phase.
When writing the momentum equation for a single
+~ ~fPu~g(I x - xPl) (12) particle we must take into account both traction for-
OXk p
ces exerted on its surface by the surrounding fluid and
where u~' denotes the kth component of the velocity of contact forces where it may be in contact with one or
the center of particle p. Particle-phase averages and more of its neighbors. Since the particles are assumed
solid-phase averages are not identical and relations incompressible contacts occur at points, and we de-
between them will be quoted below. It must be note by fPq the force exerted on particle p by particle
emphasized that eqs (8) and (9) express the overall q at their point of mutual contact. This takes a non
average and the mass average by combining the vanishing value only for a small number of particles
2460 R. Jackson
q which are in contact with particle p at the time in the term
question. Then the m o m e n t u m equation for particle
p is ~ ~ trik(y)nk(y)g(lx -- yDdsy (20)
p .]sp

p~vfi~ = .lsf,aik(y)nk(y)dsr + ~" flpq + p~vgl which is found on the right-hand side of the fluid-
q#p phase m o m e n t u m balance (16). In the former the
where v is the volume of the particle and u~' its center resultant forces due to the fluid tractions acting every-
of mass velocity. The first term on the right-hand side where on the surface of the particles are first cal-
of this, which represents the resultant force exerted on culated, then these are averaged using values of the
the particle p by the fluid, can be denoted by f/yp and in weighting function at the particle centers. In the latter,
the same way the second term, which represents the on the other hand, the traction forces at all elements
resultant force exerted on particle p at contacts with of fluid-solid interface are calculated, then these are
other particles, can be denoted byf~sp. The equation is averaged using values of the weighting function at the
averaged by multiplying both sides by g(Ix - xP]) and location of the surface elements. The former is what we
summing over p. Then, using eq. (12) we find that usually think of as the average fluid-particle interac-
tion force, so it must be related to the latter expres-

F
p~v L-~~? (n <u,> ~) + ~ (n<uiu~>~)] sion. This can be accomplished by expanding the
weighting function, which appears in the integrand of
eq. (20), as a Taylor series about the center of this
= n ( f / > " + n(f?>~ + p~ng~ (17) particle:
where the first and second terms on the right-hand
g(lx - Yl) = g(lx - x'l) @(Ix - x~l) (yj _ xf)
side are the particle-phase averages of the resultant ~Xj
forces exerted on the particles by the fluid and as
a result of solid solid contacts, respectively. 1 ~29([x -- xPD
-~ (yj--xP)(ym--X~) ....
To complete the set of m o m e n t u m equations we 2 c~xj~,x,.
need the angular m o m e n t u m balance for each par-
ticle, namely (21)

But on the surface sp, y - x p = an(y), where n is the


Bh~= ~,~If~(y~--x~)amk(y)nk(y)ds~ unit outward normal to the particle p, so the above
becomes

@(Ix - xPl)
+ ~ (~q - Xf)ffmq] g(lx - y]) = g(lx - xP[) - a nj(y)
qv~p d 63xj
where I = 2a2vpJ5 is the moment of inertia of the a 2 ~ag(lx - x'[)
particle about its center, yPq is the point of contact -~ nj(y) nm(y) . . . . (21')
2 Oxi?x,,
between particles p and q, e,z,, is the Kronecker per-
mutation symbol, and the two terms within the Using this expression for g in the integrals of eq. (20),
brackets represent the moments exerted on particle and reversing the order of differentiation with respect
p by the fluid traction forces and the solid-solid con- to x and integration over y, we then obtain
tact forces, respectively. The local average of this
equation, found in the same way as eq. (17), is ~' f alk(y)nk(y)g(lx -- y])ds, = n<f{>P(x)
p dsv

I[~(n(oJi)P)+~-~k(n(o)iUk)P) ] c3 1 c32
-- - - [n<sf>P(x)] -4- - - [n<Sfm>P(x)] ....
3xj 2 3xj~x~
= a@m~g('x--p xPl)[f~f,k(y)n,(y)nk(y)dsr (22)

where
d-q~pfPmqn~qI (18)
n <f{>'(x) = Y~g(I x - x'l) ~, t, ds
P
where n pq denotes the unit outward normal to the
surface of particle p at its point of contact with par- n<s~>V(x) = a ~ g ( l x -- xV[)fs tinjds (23)
ticle q. P v

It is important to recognize that the term


n<s~m>P(x) = a 2 2 g ( l x - xPl)fs tin~nmds
P p
n(f{) p = ~g(lx~ - xPl)Qaik(y)nk(y)ds,j~ (19)

which appears on the right-hand side of the particle- and ti = ~riknk is the ith component of the traction
phase m o m e n t u m balance (17) is not the same as exerted by the fluid on the surface of the particle.
Locally averaged equations of motion 2461
Using eq. (22) the second term on the right-hand such an explicit expression is available. Second, in
side of the fluid-phase averaged m o m e n t u m balance particular problems all or part of the first three terms
(16) is replaced by an infinite series of successively in eq. (22) may vanish identically, and in this case one
higher spatial derivatives of the terms defined in eq. of the later terms will become the leading contributor.
(23). However, using the assumption of separation of This presents no difficulty provided we recognize the
scales, it can be shown that this series can be trun- possibility, since the forms of all the terms are known,
cated with acceptably small error. To compare the and whatever terms are needed to deliver the leading
orders of magnitude of terms on the right-hand side of order contributions can be included. In other words,
eq. (22) replace the co-ordinates xl with dimensionless truncation of the series (22) is still possible, with
co-ordinates Yi = xi/L. Then successive terms contain neglect only of terms which are small of order a2/L 2
successively increasing powers of alL as factors. How- relative to those retained, though it may be necessary
ever, since t~ itself would be expected to be a complic- in certain problems to extend the series beyond the
ated function of position on the particle surface, third term.
containing terms that themselves depend on a, we Now turn attention to the second term on the
cannot conclude immediately that the orders of mag- right-hand side of the particle phase m o m e n t u m bal-
nitude of the terms in eq. (22) decrease correspond- ance (17), namely
ingly. To resolve this we may suppose t~ to be
expanded in surface harmonics on sp. Then only the n < ~ > " ( x ) = Y~g(Ix - x"l) Y. f," (24)
even harmonics contribute to the first, third, fifth, etc. P q~p
terms in eq. (23), while only the odd-order harmonics
and consider the double sum
contribute to the second, fourth, sixth, etc. terms.
Thus, if some term in the expression for t~ contributes
only to even terms in the harmonic expansion, then it ~ g(Jx - yPql)ff ~
P q#P
will contribute only to the odd lines of eq. (23), while
any term in the expression for t~ that contributes only where yPq is the position of the point of contact be-
to odd terms in the harmonic expansion will contrib- tween particles p and q. This double sum vanishes,
ute only to even lines of eq. (23). Of course, terms in since yPq -- yqP and fPq -- - fqP. Then expanding g, as
the expression for ti that contribute to both odd and before, in a Taylor series in the components of yPq
even terms in the harmonic expansion will make con- about the point x p, we obtain
tributions to all lines ofeq. (23). Thus, if a certain term
in the expression for t~ contributes to a given line of eq. 0 =~o(Ix-x'l) Zfy- a f-~-2 g ( [ x - x'l) Z f ~ PqPq
ni
p q~:p GXj p q#p
(23), we can conclude that this contribution will be
small of O(a2/L 2) compared to the contribution of the a2 ~2

same term to the second line above the line in question / . ~ JfVqnPqnpq
--~------7--'~g(lx - x p ~" x? i j m •
+ 2 GXjOX m p q~.p
but, in general, we have no basis for comparing the
magnitudes of contributions from adjacent lines, since where n pq is the unit outward normal to particle p at
they may come from different terms in t~, with differ- its point of contact with particle q. If we now define
ent dependence on a. the following quantities:
This argument suggests that all but the first two
terms on the right-hand side of eq. (22) can be ne- .<f,'>"(x) = Z o ( i x - x"l) ~ f,"q
glected, with error of O(a2/L 2) relative to the two p qv~p
terms retained. However, this is not quite correct,
since we are also interested in the overall m o m e n t u m n(s~>P(x) = a ~ g ( l x - xVl) ~ ffqny q (25)
balance for the mixture which, as we shall see, is found P q#P

by adding averaged m o m e n t u m balances for the two


phases. The term n (f/I>p appears in each of these, but n(S~m>P(x) = aZ~g(i x _ xPl) x7
/ , J i£pqnPqnpq
j m
with opposite signs, so it cancels on addition. Thus, in P q#P

the overall m o m e n t u m balance, the leading order


term containing contributions from odd harmonics of
t~ is the third term on the right-hand side of eq. (22), the above relation may be written
rather than the first. To avoid omitting the leading O
order contributions of certain terms to the overall n <~'>~(x) = 7xj In <s~j>~(x)]
m o m e n t u m balance we must, therefore, retain the first
three terms of the expansion (22). 1 dE
Two further comments on this truncation proced- - - [n(S~js>P(x)] + ... (26)
2 OXjOXm
ure are needed• First, it is not necessarily true that the
whole content of the first three terms in eq. (22) must Then, though the right-hand side of this represents an
be retained• When an explicit form for t~ is known it infinite series, following the same line of reasoning as
may turn out that certain groupings within these before it can be argued that all terms beyond those
terms can be omitted with error of O(a2/L 2) relative written out explicitly in eq. (26) can be omitted, with
to other groupings, but this cannot be known until error of O(aZ/L 2) relative to the terms retained.
2462 R. Jackson
Using eqs (22) and (26) in eqs (16) and (17), respec- the separate particles, giving
tively, the fluid-phase and particle-phase momentum
balances now take the following forms: ~ OaskV(IX
fvOa., g ..'x - yDdV = Jv,-~Yk ,, - yl)dV
fcq f ~q f 7
,~ ~" c~aik ,,
+ 2 . / - = - - g t l x - yl)dV. (32)
p dvp o y g

+ ~Xk(n(s£)P) 21 CgXkC?Xl
82
(n(s~y') + ps.egs (27) The integral over VI has already been evaluated in
deriving the fluid-phase momentum balance, and it is
and given by the first two terms on the right-hand side of
eq. (16) or, alternatively, by the right-hand side of
+ t?
psVE~(n(u~) p) ~Xk(n(UiUk)P)]=n(fif) p eq. (27) after deleting the gravitational term. In the
integral over vp we can introduce the momentum
8 1 02 equation for solid material, namely
+ ~Xk (n(s~k)p) 2 t?XkOXI(n(s~u)P) + p, nvgi (28)
~(Tik
- ps(ti~ - g3. (33)
In both of which terms are omitted which are, in @k
general, small of O(aE/L 2) compared with terms re- But within particle p we note that
tained.
Finally the particle-phase averaged angular mo- u/(y ) : Up + QabOgPa(yb-- X~)
mentum balance (18) can now be written, without
and hence, differentiating with respect to t
approximation, as
0fly) = ~if + o)fo)f(y, - xf) - t,o~ofl[(ys -- X~)
c~ + 8 p
I [ ~ ( n ( 6 9 s ) p) ~XkXk(rl(('OsUk))] + e.iabd)~(yb-- X~). (34)
Using eqs (33) and (34), and expanding g(Ix - Yl) in
= ~,ilmEn (sfl)P.+rl (sSms)P]. (29) the form (21) we then find that

Since the average angular velocity of the particles OGsk 9(I" x -- y l ) d g = psvg(lx - x'l)(fif - o3
fVp"~-yk
does not appear explicitly in eq. (27) or eq. (28) it may
seem that these are not coupled with the angular p~va2 0g(lx - xpl) p p
momentum equation. However, this is not the case {[-O,)s 09k -- o)P(J)P(~ik] -[- F.iak(f)Pa}
5 OXk
since, for example, the distribution of traction over the
particle surfaces will depend on the angular velocity, (35)
and hence this will influence quantities such as ( s ~ ) ' .
where the result
THE MIXTURE M O M E N T U M E Q U A T I O N ya 2
f r ( Y i - xf)(y i - x;)dg = T 3ij
This is the name given to the equation obtained by
taking a local average of the point momentum bal-
has been used.
ance
We can now sum over the particles and invoke the
[-c~ui c~(UsUk)] ~ri~ momentum and angular momentum balances for
PL-& + Oyk J = ~ + pg' (30) a single particle, quoted above, to evaluate the terms
0il' - g s ) and &,P. Then, after some manipulation,
over the whole neighborhood of a point x. Equation eq. (35) gives
(30) is satisfied everywhere, with p =py at points
occupied by fluid and p = p, at points occupied by
2 [ ~g(Ix - yl)dV = n(fif> p + n(fis) p
solid. Thus, averaging the left-hand side gives p Jv, cyk
cq s ~ s 1
Ps[~(~(u,) ) +~Xk(e(UsUk) )] ----[n(s~V' - n<s{,>" + n(s~,,>" - n ( s ~ , > " ]
2 t?Xk
t? u s psa 2 ~?
[nv((cOiO)k) p -- (~l(.O,)P6Sk)]. (36)
5 ~Xk
or Finally, eq. (26) can be used to expand the second
term on the right-hand side of this and the result can
~(fi(us)") + ~-XkXk(fi(UsUk) ). (31) be combined with the expression for

Averaging the first term on the right-hand side of (30) fVs-~-yk


~ffik gtrx
~' - yl)dV
is less simple. The integral over space can be expressed
as the sum of integrals over Vs. and V~ and the integral which forms the right-hand side of eq. (27) after delet-
over V~can be broken down as a sum of integrals over ing the gravitational term. This permits eq. (32) to be
Locally averaged equations of motion 2463
rewritten as balance (38), and these must be supplemented by the
particle-phase angular momentum balance (29). Not
~(Tik ,,
fv-~yk O(IX - y l ) d g all these are independent, of course; only two of the
first three are needed. We emphasize again that all
these equations neglect terms which are small of order
(a/L) 2 compared to terms included explicitly, but that
the higher-order terms are known and can be in-
Ps o2 p
+ ( ~ i ) p) + - i f " nv((tottot) 6ik - - (toiOk) p)
} cluded, if necessary, for problems where lower-order
terms may vanish identically.
The choice of which momentum balances to use is
1 a matter of convenience. Whenever the inertia asso-
O ~ ( n ( s £ l ) p + n(s~u)P). (37)
2 OXkOXt ciated with motion of particles relative to fluid is
important the overall momentum balance (38) takes
Using this, together with expression (31) for the left-
a very clumsy form. This is appreciated only when one
hand side, the volume averaged momentum balance
attempts to expand the momentum flux term on the
becomes
left-hand side in the usual form

~(fi(ui)m)'~kk(fi(UiUk)m)=~@~k[~(fTik)" ~<u,uk)" = f(u,>"(uk>" + ~<u',u'~)".


This generates a term that cancels the terms in the
n angular velocity averages on the right-hand side of
+ ~((s~()" + <d~)" + <s~y' + (sit)")
(38). However, as a result of the mean motion of each
phase relative to the mass average velocity, (U~U'k)m
p~a2 1 contains some complicated terms involving the rela-
tive velocity ( u ) I - ( u ) p and the average angular
1 02 velocity (to)P, as well as the averages of products of
--n((s£t) p + (Sfkl) p) + fig,. (38) deviations of fluid and particle velocities from their
20XkOXl
respective means. Thus, in cases where these inertial
Though not mentioned explicitly in the above ma- effects are important (e.g. 'two phase flow' problems,
nipulations, as in our earlier work, terms were omitted such as motion in bubbling fluidized beds) the indi-
which are small of O(a2/L 2) relative to those retained. vidual phase momentum balances (27) and (28) pres-
Though this overall momentum balance has been ent the problem more neatly than the overall
found by averaging the point condition of momentum momentum balance. On the other hand, for problems
balance through both phases, it can also be obtained where all inertial effects are negligible and any differ-
by an alternative route. If the particle phase averages ence ( u ) I - ( u ) p is not of primary interest, as in
appearing on the left-hand side of eq. (28) are replaced 'suspension rheology', the overall balance (38) takes
by solid-phase averages, using the easily derived rela- a simple form and is clearly an appropriate frame-
tions work.
(9 = nv + O(a2/L 2) (39)
T H E GENERAL CLOSURE PROBLEM
va 2
q~(u,) • = nv(u,y' - T e , , k -~-~ Zg(Ix - xPl)to~' The average momentum balances have been de-
OXk p rived by formal manipulation, together with some
+ O(a2/L 2) (40) ordering arguments, and they are by no means
a closed set of equations. Before the closure problem
and can be contemplated all averages of products appear-
va 2 (__ ing on the left-hand sides must be decomposed in the
dp(UiUk)~ = nV(UiUk)P -- --5 J pZg(IX -- X'[) following manner:
(u~ukY = <ui)S <uk) s + < u l u l /
x [to~tof - 6~fto~]
where ul denotes ui - (ui) c, the deviation of the point
+ ~,,j--~ Zo(Ix - xq)ui'~of velocity from the fluid-phase average. The second
Oxj p term can then be transposed to the right-hand side of
the equation, when it appears as a 'Reynolds' type
+ ~,~i~-Y~g(Ix - xq)ul'to~ of contribution to the effective stress tensor. A number
tTXj p
of terms of this type are generated from averages of
+ O(a2/L 2) (41) products, and these are listed below

then the result may be added to eq. (27), to give the (u~u'k) I, (u~u'k) p, (u~u'~)", (~o~tol) p, (colu'k) p .
same overall balance (38).
(42)
We have now derived three forms of local average
momentum balances, namely the fluid-phase balance Here it is important to note that symbols such as
(27), the particle-phase balance (28) and the mixture u} denote different things in the different terms; for
2464 R. Jackson
example, in the second term listed ul = u i - (uD p, CLOSURE TO ORDER ~b FOR A DILUTE SUSPENSION OF
while in the third term u~ = u i - ( u D ' . To achieve STOKESIANPARTICLES
closure of the equations in the most general case each
term in (42) must be expressed in terms of local The particles are assumed to be Stokesian, in the
average variables and their derivatives, and so must sense that inertia of the fluid can be neglected in
each of the following quantities which appear in the determining its motion relative to the particles. This
averaged equations: does not, necessarily, imply that fluid inertia is negli-
gible in determining the motion of the mixture relative
\ ik/ , ( s l u ) , (s~u) p. (43) to the boundaries of the system (here the relevant
length scale is L, so the Reynolds number of concern is
For rigid particles in a Newtonian fluid, (aik) y can be
much larger than that governing motion relative to
removed from this list since a general expression is
a particle), nor does it imply that inertia of the solid
easily found (Joseph and Lundgren, 1990). To obtain
particles can be neglected, even when considering
this first note that
their motion relative to the fluid. It is also assumed
that the particles are sufficiently widely spaced that
k'~yk + ~Yi) g(Ix - yD d V mutual interaction via the intervening fluid is not
fv/~u, ,~u~'~ ., significant in determining their motion. Then fluctu-
ations in the velocity of individual particles about
:fJ r, kSYk +ouk~y~y/;gtlx-
., yl)dV ( u ) p will not be induced by interactions, while fluctu-
ations that might be imposed by the initial conditions
since the rate of deformation vanishes at all points of the problem will decay in a time of the order of the
within the (rigid) particles. On the left-hand side of Stokes relaxation time. Thus, except possibly for
this relation we note that overall spatial averaging a short initial time interval, (U'iU'k) p can be neglected.
and spatial differentiation commute, while on the The fluctuations in fluid velocity about ( u ) I are con-
right-hand side we introduce the following relation: sequences of constraining the flow to satisfy the
no-slip boundary condition on the surface of each
Obli ~Uk t~ik + P~ik particle. But these are precisely the velocity contribu-
t~yk t~yi fl tions whose associated inertia is judged negligible
when asserting that the particles are Stokesian, so the
valid for a Newtonian fluid. Then it follows that , t f
inertial term pfe(UiUk) , which contributes to the ef-
fective stress tensor in the fluid-phase averaged mo-
~(uD ,~(uD e I mentum balance, can be neglected to the same
•((o~D + { p / )
OXk ~X i ]~ approximation. Thus, we need not consider any con-
or
tributions of the 'Reynolds stress' type to the averaged
equations, except possibly in a short interval follow-
ing the initial conditions.
(aik) I = -- (P)Y61k + p_|C~(Ui)/ O(Uk)|\ (44)
\ c~xk + Oxi J At the low concentration considered flow in the
neighborhood of each particle can be determined by
which is the required closure for this term. solving the Stokes equations, with a no-slip condition
The general closure problem, which requires that at the surface of the translating and rotating particle
all the quantities (42) and (43) be expressed in terms of and a requirement that the velocity and pressure fields
local average variables and their derivatives, is clearly degenerate into specified far field forms at large dis-
formidable. It is somewhat reduced in scope in cases tances from the particle. The solution then generates
where the interaction between particles is dominated increments which must be added to the far fields to
entirely by fluid forces or entirely by contact forces. In give the complete solution for the flow and pressure in
s p
the former case (S~k) p and (Sikt) , and in the latter the neighborhood of the particle. At any point occu-
( s ~ ) p and (s~l) p can be neglected. A criterion for pied by fluid the value of a velocity component, or the
distinguishing these cases, based on a Stokes number, pressure, can then be regarded as the sum of a far field
has been proposed by Koch (1990). Nevertheless, value and incremental contributions resulting from
complete closure has not been achieved, at even mod- the requirement that no-slip conditions should be
erate particle concentrations, though it is possible to satisfied on each particle surface. The sum of the
come near to this when contact interactions dominate incremental contributions should, at most be propor-
using the kinetic theory of granular materials. For tional to the number of particles per unit volume; that
Stokesian particles at extremely low concentrations, is, to the volume fraction ~b. Correspondingly the
on the other hand, where the particles are so widely far-field variables, used in solving the Stokes equation
separated that interaction between them is negligible, for each particle, should differ from the local average
complete closure is possible. This is the case originally variables by terms of order ~b, at most. (It might be
addressed by Einstein (1906) with the more limited thought that this argument is wrong in the case of the
objective of determining the effective viscosity of velocity components since, as is well known, super-
a neutrally buoyant suspension, and we shall now position of the velocity perturbations due to all the
examine the closure problem at this limit. particles diverges if these perturbations are calculated
Locally averaged equations of motion 2465
for each particle as though it were present alone. In particle surface. The quantity VVVU multiplying Ita 2
reality, of course, each particle is not alone; the solu- is not written out explicitly, but it is a sum of products
tion must satisfy no slip conditions at the surfaces of of numerical factors of order one, third derivatives of
all the other particles and this gives rise to a screening the far-field velocity and an odd number of compo-
effect which prevents the divergence from occurring.) nents of the unit normal. Similarly, the quantity
Once the velocity and pressure fields have been found VVVVU is a sum of products of factors of order one,
in the neighborhood of a typical particle the traction fourth derivatives of the far-field velocity and an even
exerted by the fluid on each element of its surface can number of components of the unit normal, and this
be found, and hence the integrals appearing in (f/I)p, pattern continues for higher terms, represented by
(s~) p and (s~t) p can be calculated. dots. Now suppose UM is the macroscopic velocity
Following this program we consider a particle scale for the problem of interest, and hence the scale
whose center of mass velocity is up, which is rotating for the far field velocity. Then in terms of UM and the
with angular velocity toP, and express the velocity field macroscopic length scale L the table below shows the
at points distant from the particle (the far field, in the orders of magnitude of certain terms on the right-
above sense) as a multipole expansion about the hand side of eq. (40), together with the number of
center of the particle. Then, if U denotes the far field multiplicative factors n~ that they contain.
velocity and P the corresponding pressure field
Term Order of mag. No. of factors n~
U=Uo+D.x+toAx+K:xx+L:xxx+ .-.
5itDia rla It UM/L odd
P = Po + pfg.x + 2ItI:K.x + -.- (45)
#a [...] (ItUM/L) (a/L) even
Here x is the position vector relative to the particle Ita2VVVU (it UM/L) (a/L) 2 odd
center, g is the specific gravity force vector and Ita3VVVVU (it UM/L) (alL) 3 even

=
1/~Ui ~Uj~ +
The orders of magnitude of the first two terms on the
right-hand side of eq. (47) cannot be estimated in
terms of UM and L because the first depends on the
(46) slip velocity U - up, while the second depends on the
rotational slip to - toP.
Kijk 2 \ 8XjaXkJo' Lijkl 6 ~OXjOXk(~XI/]O From eq. (23), to find n(flf) v and n(S~k)P we must
evaluate the following integrals:
where the suffix zero indicates that the derivatives are
to be evaluated at the location of the center of the
particle. For this problem an expression for the trac- fsf'ds and f~finJ nkds
tion on the particle surface, r = a, is available in
the literature (Nadim and Stone, 1991; Leal, 1992), and only those terms in the expression (47) for ti which
namely contain products of an even number of components of
n make nonvanishing contributions. To find n(sf) p,
3it on the other hand, we must evaluate the integral
(ti)r=a = ~a (Ui -- u~) -1- 3itnb[Ciab d- 8kabnink] ((2)a -- toP)

35 5 s tinj ds
+ 5itDian a + Ita ~Tabinanb + -~eiabOcbnan c

--
3]
3z.n.ni + -~zi -- Poni - apfgknkni
and only terms in ti containing products of an odd
number of components of n make nonvanishing con-
tributions. Thus, the second and fourth lines from the
above table make nonvanishing contributions to
q- Ita2VVVU q- ItaaVVVVU + .-- (47) n(f/Y) v and n(sfijk) p, and those from the fourth line
are small of O(a2/L 2) compared to those from the
where
second, so they can be neglected. Similarly, the first
1 and third lines from the table make non-vanishing
~ijk = "~ (Kuk + Kkij -1- Kjki)
contributions to n (s/~) p, and that from the third line is
small of O(a2/L z) compared with that from the first,
1
-- -~'~((~ijKppk -[" (~ikKppj q-- (~jkKppi) so it can be neglected. By the same token the contribu-
tions of all higher terms, represented by dots in eq.
Ol,, = Ktpqeqp,, + Krapqeqpt (47), can be neglected.
With the above approximations, and using the well
"ti = gpp i known results
Note that the term in square brackets multiplying Ita
on the right-hand side of eq. (47) is a sum of products
fs ninjds = 4ha2
T t~ij'
of numerical factors of order one, second derivatives of (48)
the far-field velocity and an even number of compo- r ninjnknt ds = 4ha2
nents of the unit outward normal vector n at the ~,~ 15 (6Ufikl + (~ik(~lj"~- (~il(~kj)
2466 R. Jackson
it is straightforward to evaluate the required integrals, and
with the following results:
[3/~( u, I a2pf
f~ 3v# t3~U/ n(s~)" = ¢ (Y () - (uJ)61~ 5
rids = 6~zlza(Ui - u~) + ~ - c3x~ pyvgi (49)
2~ob< ~2 ( u o ) S ]
fs fin~ds = ~4ha2
; - { - Pobi~ + 3Pei~(Oga - ~o~) + 5/~Di~} × (gi6~k + g~'$ki + gk~JU)+ lxa t~Uk ~---7----~ ?
GXbCXcJ
+ O(~b2) (54)
(50)
The last term within the braces represents a sum of
f,f,njnkdS= v{-~a2(U,--U~)6.ik various second derivatives of the fluid-phase average
velocity arising from the contents of the square
brackets on the right-hand side of (51). These have not
-- ~(gi6jk + gj6ki + gk6ij) been written out with explicit coefficients because, as
we shall see, they are dominated by other terms in the
+ ~L -~,/jk~+ 5(~ubOb~+ ~kbO~) fluid-phase momentum balance and can be omitted.
Equations (44), (52)-(54) provide all that is needed
to close the fluid-phase average momentum balance
- ~(6Urk + ~ikZS)+ zi6~k • (51) (27), since the 'Reynolds' term pie (U~U'k)s can be ne-
glected, as shown above. Thus we find
Then, using the first of eq. (23) together with eq. (49)
0 s ~3 I
v" (6n#a,, Pf[~(e(Ui) )-}-~Xk(C'(Ui) (~lk)f)]
,,<Yi>" = vEg(l~-
p
x ~ 1--:-,
L - -
t ~ , - ur)

3 O2 U i PYgi];

or, replacing the far-field variables by the correspond-


ing fluid-phase averaged variables, with error of 0(40
+ 2 \ 8Xk + 8xi J

n(fS) p = vZg(Ix
v
-xPl)f9#(,/ui~,s
( 2a2 \ /
- uf) + 3 ' l-l/a<u# ~<u,>r~
/

+ 4 c~x2 PYg~ + O(dp) ~. (o9,,)v]} -9#q~2a


--ei,,k 2

But from our earlier discussion of repeated averaging [ a2


x ((ul) y - ( u i ) v) q 6 ax]. J +pygi
Y~ gIIx - x ~ l ) ( u b q x ~ l = n ( ( u b S ) ~
p 1 ~2 {3__~
= n(ui) f + O(12/L 2) 2 C~XkC~X~ ((ul)Y -- (bli)P)(~kJ

and consequently the above becomes a2


5 pf~(giOjk + gj6kl + gk(~ij)
9/~b 1- ,~ a2 t~2(Ui) f"]
n(fif) v = 5~-a~Lt~.,) - <u,>.) + 6 tgx] J
q-#a2Ca~~ ' ~ .
- PsOgi + 0((02). (52) GXbGXcJ
The terms in this are easily recognizable; they repres- Scrutiny of the right-hand side now reveals some
ent the Stokes drag force, the Faxen force, and the further simplifications. The terms involving compo-
buoyancy force, respectively. nents of g in the last braces generate a contribution
Similar considerations, using the second and third that is clearly small of O(aZ/L 2) relative to the term
of eqs (23) together with eqs (50) and (51), respectively Plg~, so they can be omitted. Similarly the final term in
lead to the last braces generates contributions proportional
to fourth derivatives of ( u ) I and containing a 2 as
a multiplicative factor. These are small of O(a2/L 2)
~x, ) relative to terms in the second derivatives of ( u ) y that
come from the first set of braces, so they can also be
omitted. Finally, the average continuity equation for
+ ~x, }J the fluid phase can be used to re-write the left-hand
+ o(,/, ~) (53) side in terms of the material derivative following the
Locally averaged equations of motion 2467
fluid. Thus, the equation reduces to effective stress tensor for the fluid phase. It consists of
two parts, one symmetric and one antisymmetric. The
antisymmetric part vanishes if rotational relaxation is
PSe Dt - c~xi OXk t~ CqXi j complete; in other words, if the average angular velo-
cities for fluid and particles are the same. We also note
that

3 "flfct(ui)Z ~(uk)Z']
<u> = (u): + ,~[(uy - <u):]
so it appears that the symmetric part could be re-
placed by
2a2 L. . . . - (ui)v)
a2 c~2~ui)S] c12
+~ ~ J-~x--~[-~ ((u~)s with error of O(~b2), unless the slip velocity is large of
order 1/~b. This clearly exhibits the familiar Einstein
- (ui)V)] + Psgi (55) correction to the effective viscosity.
It is important to note that eq. (55) includes modifi-
where cations to the momentum equation of the pure fluid
arising from the presence of the particles, and these
Ds ~ s ~
terms are correct to O(~b). However, eqs (56) and (57)
D~t = ctt + <Uk> ~kXk"
do not include O(~b) corrections to the equations of
This form of the fluid-phase momentum balance is motion of an isolated particle. To find these it would
correct to order q5 in the particle concentration and be necessary to take into account the interactions
omits terms that are small of order a2/L 2 relative to between the particles through the motion of the inter-
those retained. vening fluid.
Similar manipulations applied to the averaged par- Finally the mixture momentum balance (38) be-
ticle-phase momentum equation (28), and angular comes
momentum equation (29) lead to the following closed
forms for these equations: U m ~ m
~(~( ~) ) + T~x l~(u~uD )
Dv(ui) v 9#[ u. f a2 ~2(ui)f]
P~' D----~ 2a z ( ( ' ) - (u'>V) + 6 8x 2 J
- + /

+ (P, - Ps)O~ (56)


and
5 ~ @(u,/ a(uD I']
Dv(oi) v 15p[-1 O(uk)S (~oi) v] (57)
+ ~(<~j>"<~j>",~k -
where
J
02 1-3m%, -s
De ~ L-y-t~,~,~ - <u,y)] + ~0,. (58)
N = ~-t + <uk>PUx~"
Equation (56) calls for little comment; the forces act- As remarked earlier the appearance of this equation
ing on the particles are gravity, buoyancy, the Stokes does not reflect its true complexity. When the average
drag and the Faxen force. The right-hand side of (57) of the product of two components of u on the left-
is proportional to the difference between the angular hand side is expanded, in the usual way
velocity associated with the averaged fluid velocity (UiUk) m = (ui)m(Uk) m ~- (U~Utk) m
field, and the averaged angular velocity of the par-
ticles; in other words, to the rotational slip velocity. the second term is not negligible, even though both
Thus the mixture will relax, as expected, to a state in (U~U'k)s and (u'iu'k) p can be neglected, as argued
which the angular velocities associated with the aver- above. As a result, a term appears which cancels those
aged motions of both phases are the same. On the terms in eq. (58) involving the angular velocity com-
right-hand side of eq. (55) there is a term containing ponents, but unfortunately in general other complic-
V E [ - ( ~ ( ( u ) f - ( u ) P ) l and this is clearly small of ated quadratic terms appear, involving components of
O(a2/L 2) compared to the Stokes drag force. How- the slip velocity, the particle velocity, and the angular
ever, it should be retained for a reason discussed velocity of the particles. However, in the common
earlier; the drag force cancels from the overall average situation where [(o)vl is of the same order of magni-
momentum balance for the mixture, and this then tude as [1/2Vx(u)S], expanding the left-hand side of
remains as the dominant term in the slip velocity. The eq. (58) leads to the following form for the mixture
content of the braces in eq. (55) can be regarded as an momentum balance, neglecting terms of O(aE/L 2)
2468 R. Jackson
relative to those retained would be expected to introduce dissipative effects.
Alternatively, if the fluid-phase continuity equation
~(~(uD") + (~(uD"(uk)") (13) is replaced by its approximation to order zero in
q~, namely V. ( u ) : = 0, to match the order of approxi-
~(p)~r - - < f (~(u,) ~{u,~)) mation in eq. (56), the uniform suspension is then
predicted to be neutrally stable. This is the correct
result in the limit q~~ 0, but it gives no indication
p<.,>, o<u,>,) whether the suspension is stable or unstable for small,
+ 2 \ Oxk + 8xi J but nonvanishing values of 4~.

'°~PI~((uDIp - (u,)')((uD s - (u,,)")} CONCLUSION


a z J-3/~b,, ,1 (ui}P)] In the first part of this paper equations of motion
L-q--t,u,, - J + ;g,. (59) for the concentration of the particles and the local
average velocities of both particles and fluid have
The last term in the braces arises from the fact that been derived in a form that exhibits explicitly, in terms
differences between (u} m, (u} z and ( u ) v contribute to of microscopic variables at the scale of the individual
(U~U'k)', even though (U~U'k}y and (U~U'k}p are both particles, the quantities that must be related to the
negligible. It is an inertial contribution to the effective local average variables to achieve closure of the equa-
stress tensor and would therefore be expected to be tions. These quantities are tensor-weighted integrals
negligible in the present case, where the Stokes ap- of the traction forces exerted on the surface of a par-
proximation has been invoked for the relative motion ticle by the fluid, averaged over the particles in the
of fluid and particles. The term V2[3#~b((ui)y - neighborhood of a point in question, analogous
(u~)P)/4], which is not inertial in origin, has neverthe- quantities with the traction forces replaced by forces
less not appeared in discussions of the rheology of exerted at points of direct solid-solid contact, and
dilute suspensions since these have focused on cases local averages of products of components of fluctu-
where there is no difference between the average ations in the velocities of both fluid and particles
velocities of the fluid and the particles or, if such about their respective mean values. In general there is
a difference exists, it corresponds simply to a uniform little hope of finding explicit expressions for all these,
sedimentation. but they could possibly be extracted from results of
It was indicated earlier that the general mixture planned simulations of the local motion using, for
momentum balance (38) can be obtained by translat- instance, the method of 'Stokesian dynamics' when
ing the particle-phase momentum balance (28) into Reynolds numbers based on the particle diameter are
a solid-phase momentum balance, using eqs (39)-(41), small enough. Their determination might therefore
then adding it to the fluid-phase momentum balance pose a goal for such simulations, and a formulation in
(27). In the present case it might be thought that this terms of local volume averages, rather than ensemble
procedure would miss some terms of order 4) since, as averages, is then an advantage, since an ensemble of
noted above, the particle momentum balance (56) simulations (or an ergodic hypothesis to supplement
does not include O(qS) corrections. However, this is one simulation) is not needed to evaluate them.
not the case, since eq. (56) must be multiplied through The general results have been presented as separate
by ~b to give the form of eq. (28) applicable in the momentum balances for the fluid and the particle
present limit, so when added to the fluid-phase mo- phases [eqs. (27) and (28)], and an overall momentum
mentum balance it contributes terms of order ~b. Any balance for the mixture has also been derived [eq.
correction terms of O(~b) added to eq. (56) would (38)]. We emphasize again that only two of these three
contribute at order ~b2 to the mixture momentum momentum balances are independent and the most
balance. appropriate choice depends on the nature of the prob-
Though the order of approximation in eq. (56) is lems of interest. Thus, for suspension rheology, where
not inconsistent with that in the mixture balance (58) the difference between the average velocities of fluid
it does not match that of the averaged continuity and particles is not important, it is appropriate to
equations (13) and (15), which are correct to all orders think in terms of the suspension as a single entity,
of 0. This has significant consequences. For example, together with its associated stress tensor whose diver-
if attention is confined to problems in a single, vertical gence appears on the right-hand side of eq. (38). This
space dimension, eqs (13), (15) and (56) provide three tensor determines the resultant force exerted on solid
equations for ~b and the vertical components of ( n ) y boundaries and hence the result of rheometric
and ( u ) p. However, these equations are ill-posed. measurements. In many problems of two-phase flow,
A uniform suspension is predicted to be unstable on the other hand, such as the motion of fluidized
against harmonic disturbances propagating upward, suspensions, the difference between the average velo-
and the rate of growth increases without bound as the cities of the separate phases is of central importance
wave number tends to infinity. This divergence in the so attention naturally focuses on individual mo-
growth rate would presumably be eliminated if eq. mentum balances for these phases. Despite this differ-
(56) were supplemented by corrections of O (qS),which ence of approach it must be emphasized that any
Locally averaged equations of motion 2469
particular choice for the pair of independent mo- REFERENCES
mentum balances is equivalent to any other; there is Anderson, T. B. and Jackson, R. (1967) Fluid mechan-
no single 'right' way of viewing the system, and no ical description of fluidized beds. Equations of
conflict of principle between the points of view of motion. Ind. Enyng Chem. Fundam. 6, 527-539.
workers in two-phase flow and those in suspension Drew, D. A. (1971) Averaged field equations for two
rheology. phase media. Stud. Appl. Math. 50, 133-166.
In the second part of the paper it has been shown Drew, D. A. and Segel, L. A., (1971) Averaged equa-
that an explicit, closed form for the equations of tions for two phase flows. Stud, Appl. Math. 50,
205-231.
motion can be found at the 'Einstein' limit of a very
Einstein, A. (1906) Eine neue Bestimmung der
dilute suspension of Stokesian particles. It is also Molekuldimensionen. Ann. Phys. 19, 289.
possible to approach a complete closure for a suspen- Hinch, E. J. (1977) An averaged-equation approach to
sion of sufficiently dense solid particles in a gas, where particle interactions in a fluid suspension. J. Fluid
contact forces dominate fluid tractions in determining Mech. 83, 695-720.
the interaction between particles, by invoking results Joseph, D. D. and Lundgren, T. S, (1990) Ensemble
of the 'kinetic theory of granular materials'. average and mixture theory equations for incom-
pressible fluid-particle suspensions. Int. J. Multi-
phase Flow 16, 35-42.
Koch, D. L. (1990) Kinetic theory for a monodisperse
Acknowledgements gas solid suspension. Phys. Fluids A2, 1711 1723.
The author wishes to thank Trinity College and the De- Leal, L. G. (1992) Laminar Flow and Convective Trans-
partment of Applied Mathematics and Theoretical Physics, port Processes. Butterworth-Heinemann, London.
Cambridge University, for their hospitality during the period Nadim, A. and Stone, H. A. (1991) The motion of
when this work was initiated. In particular he acknowledges small particles and droplets in quadratic flows.
the contribution of Dr. E. J. Hinch, who asked the right Stud. Appl. Math. 85, 53-73.
question. Thanks are also due to Professor Andrea Pros- Nigmatulin, R. I. (1979) Spatial averaging in the
peretti for showing him the work referred to as 'Zhang and mechanics of heterogeneous and dispersed systems.
Prosperetti (1995)' prior to its publication. As mentioned Int. J. Multiphase Flow 5, 353 385.
earlier, this contains results which parallel many of those Zhang, D. Z. and Prosperetti, A. (1994) Averaged
presented here, though they are obtained in terms of en- equations for inviscid disperse two-phase flow. J.
semble averages, so the derivations differ. It is encouraging Fluid Mech. 267, 185-219.
that the two methods of averaging lead to results in substan- Zhang, D. Z. and Prosperetti, A. (1995) private com-
tial agreement. munication.
Update
Chemical Engineering Science
Volume 52, Issue 23, December 1997, Page 4433

DOI: https://doi.org/10.1016/S0009-2509(97)80859-7
Pergamon

Erratum
Jackson, R. (1997) Locally averaged equations of motion for a mixture
of identical spherical particles and a Newtonian fluid. Chem. Enyng Sci.
52, 2457-2469.

In eq. (45) for the ‘far field’ flow in the neighbourhood by


of a particle, the pressure field was derived from the
Df <Ui)’
velocity field using the Stokes equation, which PJ 7
neglects inertia. This is not correct. Though inertia
can be omitted in determining details of the flow
round a particle it cannot be neglected in the large
scale fluid motion. Thus, the pressure should have
been derived using the full Navier-Stokes equation.
Equation (45) should, therefore, be replaced by the
following form

ll=lJ,+D.x+wAx+K:xx+Lfxxx+ ...

(45)
.x+2&K.x+ .‘.
and

This change has consequences that propagate, in an D~(ui>P 9~ f? i2(n,)


= u (W - <tO) + T 7
obvious way, through subsequent equations, with “- Dt 2 I 1
the result that the final fluid and particle phase
D, (us>’
momentum equations, (55) and (56), must be replaced + (Ps - /Jr)% + LJf 7- (561

4433

You might also like