Sunteți pe pagina 1din 12

Nonlinear Analysis: Real World Applications 10 (2009) 11391150

www.elsevier.com/locate/nonrwa
On the OberbeckBoussinesq approximation for uids with pressure
dependent viscosities
K.R. Rajagopal
a,
, G. Saccomandi
b
, L. Vergori
c
a
Department of Mechanical Engineering, Texas A&M University, College Station, TX 77843-3123, USA
b
Dipartimento di Ingegneria Industriale, Universit` a di Perugia, Via G. Duranti, 06125, Italy
c
Dipartimento di Matematica, Universit` a del Salento, Strada Prov. Lecce-Arnesano, 73100 Lecce, Italy
Received 14 November 2007; accepted 7 December 2007
Abstract
We derive, by using a rigorous perturbative approach, the OberbeckBoussinesq approximation within the framework of one of
the simplest implicit constitutive theories: that of uids with pressure dependent viscosity. To illustrate an application we use the
equations derived here to study the ow of a uid in a channel with heated walls.
c 2007 Elsevier Ltd. All rights reserved.
Keywords: Pressure dependent viscosity; OberbeckBoussinesq approximation
1. Introduction
Few approximations in uid mechanics have proved as useful and successful in predicting observed phenomena as
the OberbeckBoussinesq approximation; the other approximation that stands out is Prandtls brainchild, the Bound-
ary Layer Approximation. The OberbeckBoussinesq approximation has implications for a wide variety of ows
within the context of astrophysical and geophysical uid dynamics. In the Boundary Layer Approximation, the ow
domain is partitioned into a region wherein the effects of viscosity are predominant and expressed by the Boundary
Layer Equations, while outside this ow boundary layer the uid is approximated as an ideal uid. The Boundary
Layer Approximation makes use of two main hypotheses, one that the ow has distinct characteristics in two different
domains and hence can be described in the two sub-domains by means of different equations, and a further approx-
imation that in the domain adjacent to the solid boundary the NavierStokes equations can be further simplied to
what is referred to as the Boundary Layer Equations. This approximation is obtained by a making several assumptions
concerning the magnitude of the velocity components in different directions, the boundary layer thickness, etc. Unlike
the Boundary Layer Approximation, the OberbeckBoussinesq approximation is more in keeping with a perturbation
of the governing equation by identifying a small non-dimensional parameter and retaining terms of like order. While
this is the popular wisdom concerning the approximation, this is not a true depiction of the state of affairs as this is not

Corresponding author. Tel.: +1 409 862 4552.


E-mail addresses: krajagopal@tamu.edu (K.R. Rajagopal), saccomandi@mec.dii.unipg.it (G. Saccomandi), luigi.vergori@unile.it
(L. Vergori).
1468-1218/$ - see front matter c 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.nonrwa.2007.12.003
1140 K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150
what is strictly carried out in order to obtain the OberbeckBoussinesq equations. These celebrated equations are not
obtained by a standard perturbation technique. Many arguments have been put forward to justify the inclusion of terms
that appear in the OberbeckBoussinesq equation, but most of these arguments do not pass muster as explained below.
The approximate equations that have been used, and continue to be used, with great success, were rst derived
by Oberbeck [12,13] and subsequently and independently derived by Boussinesq [2]. Oberbeck and Boussinesq were
interested in obtaining equations that would govern the ow of a classical linearly viscous uid which undergoes
isochoric motion in isothermal ows, but which could change its volume due to changes in temperature. This then
implies that det F is a constant in motion when the temperature is a constant, but the value of det F could vary with
temperature, F being the deformation gradient. If the motions are sufciently smooth, this then implies that div v
vanishes when temperature is a constant but changes when the temperature changes, v being the velocity of the uid.
Justications for the approximation due to Oberbeck and Boussinesq are too numerous to be listed and here we
mention just some of them. Important studies are due to Rayleigh [16], Jeffreys [9], Chandrasekhar [3], Spiegel
and Veronis [19], Mihaljan [11], Roberts [17], Roberts and Stewartson [18], Spiegel and Weiss [20], and Hills and
Roberts [7]. Not all of the above mentioned papers try to provide a rigorous justication for the approximation; some of
them do try to provide some sort of rationale for the approximation, but they are not convincing for reasons discussed
below. Recently, Rajagopal, Ruzika and Srinivasa [15] carried out an analysis in which they delineate the status of the
OberbeckBoussinesq approximation on the basis of certain non-dimensional numbers that they introduce. However,
their study implies that the approximation cannot be viewed as a proper perturbation in which terms of like order are
retained and in their derivation they show that the OberbeckBoussinesq equations result as a consequence of mixing
terms of different orders in a small parameter. They also provided higher order approximations to the problem. It might
yet be possible to develop a proper perturbation scheme wherein the OberbeckBoussinesq equations are obtained as
an approximation at a specic order of the perturbation; however at this juncture no such analysis is available.
We now discuss briey some of the attempts to justify the OberbeckBoussinesq approximation; a more detailed
critique of the various attempts can be found in [15]. Spiegel and Veronis [19] considered the motion of a compressible
uid and they introduced a small parameter related to the ratio of the variation in density in the absence of motion
and the spatial average value of the density and then carried out a perturbation analysis. Spiegel and Veronis [19] were
fully aware that their approximation did not retain terms of the same order in the perturbation. In fact, Spiegel and
Veronis [19] explicitly state In Eq. (19) we have retained the term g(

/
0
)k even though it contains as a factor,
and this is clearly unacceptable as they recognize. Another shortcoming of the approach of Spiegel and Veronis [19]
is that the layer of uid has to be sufciently thin while the physical applications, especially in astrophysics and
geophysics, require considerably thick layers.
A common problem with many of the justications for the OberbeckBoussinesq approximation stems from the
need to retain a term that is the product of the coefcient of thermal expansion and gravity. This product should be of
order one, while the coefcient of thermal expansion has to tend to zero. This leads to the untenable requirement
that gravity has to tend to innity. As we saw above, Spiegel and Veronis [19] explicitly retain a term at rst
order in which the small parameter that relates to the perturbation appears and is multiplied by the acceleration
due to gravity. Similarly, in the study by Mihaljan [11] which is often cited for giving a rigorous justication of the
OberbeckBoussinesq approximation, we encounter a similar difculty. Mihaljan [11] uses two small parameters for
perturbation and he carries out the perturbation analysis. Unfortunately, he does not recognize that when one of the
small parameters goes to zero it immediately forces the other small parameter to tend to innity. In effect he encounters
the same problem as was faced by Spiegel and Veronis [19], but under a different guise. Hills and Roberts [7] in their
study, much in keeping with [19], require that the product of the coefcient of thermal expansion and the acceleration
due to gravity be a constant while the coefcient of thermal expansion tends to zero, an impossibility if gravity is
nite. In fact, they recognize the problem with their approach and in fact state explicitly that As we shall see this last
requirement is essential, because otherwise buoyancy forces are lost. Here, the requirement that they refer to is that
the product of the coefcient of thermal expansion and the acceleration due to gravity be a constant as the coefcient
of thermal expansion tends to zero.
It is important to recognize that in these studies one knows the answer that one wants, as Hills and Roberts [7]
observe the need to retain the buoyancy term, and hence they resort to making assumptions which will yield the desired
result even though the assumptions are patently wrong and the authors themselves are aware of it. In fact, these eminent
scientists would not have made such assertions had they not recognized the validity of the OberbeckBoussinesq
approximation and the need to somehow provide some sort of justication for the same.
K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150 1141
Another attempt at providing a rationale for the OberbeckBoussinesq approximation is due to Gray and
Giorgini [6]. After providing a very clear discussion of the subtle issues that need to be taken into account in order
to obtain the approximation, they make certain ad hoc assumptions concerning the smallness of certain parameters
to arrive at the OberbeckBoussinesq approximation. Though the study does not provide a rigorous basis for the
approximation, their study is an interesting attempt at arriving at the same. Mention must also be made of several
studies by Zeytounian [21] but in our opinion none of them provide a proper basis for the OberbeckBoussinesq
approximation. We shall not discuss any of these works here as this study is not meant to be an exhaustive review
of the derivation of the OberbeckBoussinesq approximation. Here we are interested in developing an approximation
similar to that of the OberbeckBoussinesq approximation when the viscosity of the uid depends on the pressure.
Our study here is similar in its approach to the study by Rajagopal, Ruzika and Srinivasa [15] of the celebrated
OberbeckBoussinesq equations. However, since the viscosity, the specic heat, the thermal conductivity, etc., are
all functions of both the temperature and pressure, the analysis is much more complicated. After introducing the
appropriate non-dimensional parameters, we assume that the various physical quantities such as the viscosity, specic
heat and thermal conductivity are analytic in the temperature and pressure and we express the various relevant physical
quantities in a double series with respect to both the temperature and pressure. We then carry out a perturbation
analysis in terms of a small parameter that is the second Froude number. We then group the terms appropriately,
and obtain a generalization of the results established by Rajagopal, Ruzika and Srinivasa [15]. When the physical
quantities under consideration are not dependent on the pressure, we recover the results due to Rajagopal, Ruzika
and Srinivasa [15]. Finally, we consider the laminar ow of a uid whose viscosity is dependent on the pressure but
whose coefcient of thermal expansion, thermal conductivity and specic heat are constants. While it is true that all
the physical quantities do vary with pressure, the variation in the viscosity with pressure is far more dramatic than
the variation of the other quantities with pressure. For instance, while viscosity might change by a factor of ten to
the power of eight or more (see [1]), the density will vary by merely a few per cent (see [4,14] for details). The
other properties also undergo much more modest changes in their values than the viscosity and hence we feel that
this assumption is a reasonable rst approximation. On the basis of this approximation we study steady unidirectional
ows subject to a temperature eld to assess the effect of buoyancy on the ow when the viscosity depends upon the
pressure. We nd that in the case of the viscosity being constant we recover the classical result.
The equations established in this paper will have relevance to geophysical ows wherein the viscosity will change
with the depth of the uid. For such ows, the approximation established here should be used rather than the standard
OberbeckBoussinesq equation.
2. Governing equations
We recall below the balance of mass, linear momentum and energy, as well as the second law of thermodynamics
(in the form of the ClausiusDuhem inequality):
+ div v = 0, (1)
v = div T + b, (2)
e + div q = T D + r, (3)

r

div
_
q

_
, (4)
where is the density, v the velocity eld, T the Cauchy stress tensor, b the specic external body force eld, e
the specic internal energy, r the specic radiant heating, the temperature, q the heat ux vector, D the symmetric
velocity gradient and the specic entropy.
Introducing the specic Helmholtz free energy through
= e (5)
and combining (1) and (3)(5) we arrive at the inequality
(

+

) T D +
1

q 0 (6)
which, as we shall soon show, places restrictions on the constitutive equation for the uid.
1142 K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150
The assumption that the uid can sustain only isochoric motions in isothermal processes implies that
det F = f (),
by which, under the assumption that the deformation gradient F is differentiable with respect to time,
div v = ()

(7)
where
() =
1
f ()
d f
d
()
is the coefcient of thermal expansion. So, from (1) and (7) it follows that

= ()

(8)
and consequently
=
1

. (9)
Concerning the stress T we shall express it through the relation
T = pI +

T, (10)
where
p :=
1
3
tr T
is the mechanical pressure and the traceless tensor

T is the deviatoric stress. From now on we shall regard the
mechanical pressure p and the temperature as independent variables on which the material parameters of the uid
depend. Thus, as we are interested in the linearly viscous uid, the requirement of material frame independence leads
us to consider the following constitutive uid model:
e = e( p, ) + u( p, )tr D, = ( p, ) + h( p, )tr D, (11)
q = k( p, ),

T = 2( p, )
_
D
1
3
(tr D)I
_
(12)
where k is the heat conductivity and is the viscosity. Then, with the use of (7), (10) and (12), the ClausiusDuhem
inequality (6) leads to
(

+ + p)


p
p
tr D

(tr D) +
k

2
+ 2
_
D
2

1
3
(tr D)
2
_
0. (13)
Now, by using standard arguments in continuum mechanics we can show that

p,
p
=
tr D
= 0 (14)
and that the constitutive functions k and are non-negative. Finally, by (5), (9) and (14) we readily deduce that
e = e( p, ) (i.e. u( p, ) = 0 in (11)), = ( p, ) (i.e. h( p, ) = 0 in (11)),
e
p
=
p
=

, e

= c
p

p, (15)
where c
p
= c
p
( p, ) := (/)
p
is the specic heat at constant pressure. Thus, in view of (7) and (10), introducing
(12) and (15) into the equations of balance (2) and (3) gives
v = p +

3
(div v)
2
3
(div v) + 2D + v + b (16)
K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150 1143
and
c
p

p = k + k + 2
_
D
2

1
3
(tr D)
2
_
+ r. (17)
Eqs. (7), (8), (16) and (17) form the governing equations for the determination of the elds , v, and p. It is
interesting to note that (17) is the equation for the determination of p since

is determined from (7).
3. OberbeckBoussinesq approximation
In this section, following [15], we shall try to provide a rigorous mathematical justication for the
OberbeckBoussinesq approximation in the case in which the material parameters of the uid depend on pressure
and temperature.
Let us consider a layer of uid of thickness d, the top and the bottom surfaces of which are held at constant
temperatures T
2
and T
1
(T
1
> T
2
), respectively. In order to non-dimensionalize Eqs. (7), (8), (16) and (17) we choose
a convenient reference state (
0
, T
0
) and introduce the following scales:
x

=
x
d
, v

=
v
U
,

0
, t

=
U
d
t,
p

=
p
0

0
gd
, b

=
b
g
,

=
T
0
T
0
,

0
Ud

0
, c

p
=
T
0
gd
c
p
, k

=
T
0

0
gUd
2
k, r

=
d
U
3
r,
(18)
where
T
0
= T
1
T
2
, U =
_
gd
0
T
0
and g is the acceleration due to gravity,
0
and
0
are the density and the thermal expansion at the reference
temperature T
0
, respectively. Introducing (18) into (7), (8), (16) and (17) leads to (omitting all stars)

= F
2


, (19)
div v = F
2


, (20)
F
2
v = p +
F
2
3
(div v)
2
3
F
2
(div v) + 2F
2
D + F
2
v + b (21)
and
c
p

F
2

_
+
T
0
T
0
_
p = k + k + 2F
2

_
D
2

1
3
(tr D)
2
_
+ F
2
r, (22)
where
F =
U

gd
=
_

0
T
0
is the Froude number.
We now introduce the small parameter with respect to which we shall carry out our perturbation. Let
= F
2
=
U
2
gd
1
1
and
v =
+

n=0

n
v
n
, =
+

n=0

n
, p =
+

n=0

n
p
n
(23)
1
The non-dimensional parameter F
2
is known as the second Froude number.
1144 K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150
be the power series in of the physical quantities v, and p. From now on we shall assume that , c
p
, k and
are analytic functions and we shall limit our analysis to pressure and temperature departures from the reference state
(
0
, T
0
) for which we can write
() =
+

n=0
1
n!
d
n

d
n
(0)
n
, (24)
c
p
( p, ) =
+

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)
c
p
p
j
1

j
2
(0, 0) p
j
1

j
2
, (25)
k( p, ) =
+

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)
k
p
j
1

j
2
(0, 0) p
j
1

j
2
(26)
and
( p, ) =
+

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)

p
j
1

j
2
(0, 0) p
j
1

j
2
. (27)
Thus, from (19) and (24) we get
= exp
_

n=0
1
(n + 1)!
d
n

d
n
(0)
n+1
_
and hence
() = 1
_
+

n=0
1
(n + 1)!
d
n

d
n
(0)
n+1
_
+ o(), (28)
where o() represents the terms of order
n
with n 2.
Inserting (23)(28) into (20)(22) we obtain
+

n=0

n
div v
n
=
+

j =0
d
j

d
j
(0)
+

n=0

n
_

j
_

n
t
+ v
__
n
, (29)

_
1
+

j =0
1
( j + 1)!
d
j

d
j
(0)
+

m=0

m
(
j +1
)
m
+ o()
_

n=0

n
_
v
n
t
+ (v v)
n
_
=
+

n=0

n
p
n
+

3
+

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)

p
j
1

j
2
(0, 0)
+

n=0

n
_
p
j
1

j
2
(div v)
_
n

2
3
+

j
1
+j
2
=1
1
j
1
! j
2
!

( j
1
+j
2
)

p
j
1

j
2
(0, 0)
+

n=0

n
_
div v( p
j
1

j
2
)
_
n
+2
+

j
1
+j
2
=1
1
j
1
! j
2
!

( j
1
+j
2
)

p
j
1

j
2
(0, 0)
+

n=0

n
[D ( p
j
1

j
2
)]
n
+
+

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)

p
j
1

j
2
(0, 0)
+

n=0

n
_
p
j
1

j
2
v
_
n
+b
_
1
+

j =0
1
( j + 1)!
d
j

d
j
(0)
+

m=0

m
(
j +1
)
m
+ o()
_
, (30)
K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150 1145
_
1
+

j =0
1
( j + 1)!
d
j

d
j
(0)
+

m=0

m
(
j +1
)
m
+ o()
_

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)
c
p
p
j
1

j
2
(0, 0)
+

n=0

n
_
p
j
1

j
2
_

t
+ v
__
n

m=0
1
m!
d
m

d
(0)
+

n=0
_

m
_
+
T
0
T
0
__
p
t
+ v p
__
n
=
+

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)
k
p
j
1

j
2
(0, 0)
+

n=0

n
_
p
j
1

j
2

_
n
+
+

j
1
+j
2
=1
1
j
1
! j
2
!

( j
1
+j
2
)
k
p
j
1

j
2
(0, 0)
+

n=0

n
[ ( p
j
1

j
2
)]
n
+2
+

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)

p
j
1

j
2
(0, 0)
+

n=0

n
_
p
j
1

j
2
D
2
_
n

2
3

j
1
+j
2
=0
1
j
1
! j
2
!

( j
1
+j
2
)

p
j
1

j
2
(0, 0)
+

n=0

n
[ p
j
1

j
2
(tr D)
2
]
n
+r
_
1
+

j =0
1
( j + 1)!
d
j

d
j
(0)
+

m=0

m
(
j +1
)
m
+ o()
_
. (31)
We are now in a position to equate the like powers of and obtain a systematic hierarchy of equations. Collecting
the terms of O(1) in Eqs. (29)(31) we obtain
div v
0
= 0, (32)
p
0
+ b = 0 (33)
and
c
p
( p
0
,
0
)
_

0
t
+ v
0

0
_
= k( p
0
,
0
)
0
+ [k( p
0
,
0
)]
0
. (34)
We notice that the above equations are not sufcient to determine all the eld variables at O(1). Therefore, in order
to attain closure, we proceed to obtain the equations at O(). Setting
G(
0
) =
_

0
0
()d =
+

j =0
1
( j + 1)!
d
j

d
j
(0)
j +1
0
,
from (30) we obtain
v
0
t
+ v
0
v
0
= p
1
+
1
3
( p
0
,
0
)(div v
0
)
2
3
(div v
0
)[( p
0
,
0
)]
+2D
0
[( p
0
,
0
)] + ( p
0
,
0
)v
0
G(
0
)b
which, in the light of (32), becomes
v
0
t
+ v
0
v
0
= p
1
+ 2D
0
[( p
0
,
0
)] + ( p
0
,
0
)v
0
G(
0
)b. (35)
Now Eqs. (32)(35) form a closed system and it is interesting to remark that p
0
is the pressure due to the body forces
acting on the uid while p
1
is the pressure due to the thermal expansion of the uid.
1146 K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150
Finally, by means of (18) we re-dimensionalize Eqs. (32)(35) and obtain the equations governing the ows in a
uid layer at small second Froude number
_

_
p
0
+
0
b = 0

0
_
v
0
t
+ v
0
v
0
_
=
0
(T
1
T
2
)p
1
+ 2D
0
[( p
0
,
0
)] + ( p
0
,
0
)v
0

0
G(
0
)b
div v
0
= 0

0
c
p
( p
0
,
0
)
_

0
t
+ v
0

0
_
= k( p
0
,
0
)
0
+ [k( p
0
,
0
)]
0
,
(36)
where the function G is now dened as
G(
0
) =
_

0
T
0
()d.
It is easy to check that system (36) coincides with the classical OberbeckBoussinesq equations when , c
p
, k and
are assumed to be constant or to depend only on temperature.
4. Laminar ows
Let Oxyz be a cartesian frame of reference with unit vector elds i, j, k, respectively, k pointed vertically upward.
In this section we shall determine the laminar ows in a uid whose viscosity is an analytic function of pressure and
temperature whereas the coefcient of thermal expansion , the specic heat at constant pressure c
p
and the heat
conductivity k are assumed to be constant. Therefore, if gravity is the only force acting on the uid, the equations
which govern the motion we have derived in the previous section become
_

_
p +
0
gk = 0

0
v
t
+
0
v v = (T
1
T
2
)P + ( p, T)v + 2D ( p, T) +
0
g(T T
0
)k
div v = 0
T
t
+ v T = T
(37)
in
d
= R
2
(d/2, d/2). In (37)
0
is the density at the reference temperature T
0
= (T
1
+ T
2
)/2, = k/(
0
c
p
) is
the thermal diffusivity, g and p are, respectively, the acceleration and the pressure eld due to gravity, P is the pressure
due to the thermal expansion of the uid and by T we now denote the temperature eld. The boundary conditions that
we append to system (37) are
_
T(x, y, d/2, t ) = T
2
, T(x, y, d/2, t ) = T
1
p(x, y, 0, t ) = p
0
(38)
where p
0
is the reference pressure.
Now it is convenient to non-dimensionalize (37) according to the scales
x

=
x
d
, t

=

0

0
d
2
t, v

=

0
d

0
v,
p

=
p p
0

0
gd
, P

=
P

0
gd
,

0
,
T

=
T T
0
T
1
T
2
, R =
(T
1
T
2
)
0
gd
3

, Pr =

0

,
(39)
where
0
= ( p
0
, T
0
) is the viscosity at the reference state ( p
0
, T
0
), R and Pr are the Rayleigh and Prandtl numbers,
respectively. With this scaling (37) becomes (omitting all stars)
_

_
p + k = 0
v
t
+ v v =
R
Pr
P + ( p, T)u + 2D ( p, T) +
R
Pr
Tk
div v = 0
Pr(T
t
+ v T) = T
(40)
K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150 1147
in R
2
(1/2, 1/2). Then to determine steady ows of the type
v = v(z)i, T = T(z),
we have to solve the following system:
_

_
p
x
= p
y
= P
y
= 0
p
z
= 1

R
Pr
P
x
+
z
v
z
+ v
zz
= 0

R
Pr
P
z
+
x
v
z
+
R
Pr
T = 0
T
zz
= 0
(41)
with boundary conditions
_
_
_
v(1/2) = V
1
, v(1/2) = V
2
T(1/2) = 1/2, T(1/2) = 1/2
p(0) = 0.
(42)
It is easy to check that the boundary value problem (41) and (42) admits the solution
_

_
p = T = z
P =
z
2
2
+
Pr
R
A
0
x + P
0
v = V
1
+
_
z
1/2
A
0
+ c
( )
d,
(43)
where A
0
is the pressure gradient and
c =
_
V
2
V
1
A
0
_
1/2
1/2

( )
d
__
_
1/2
1/2
d
( )
_
1
.
In particular we consider an exponential dependence of viscosity on temperature and pressure proposed by
Laun [10],
=
0
exp[( p p
0
) (T T
0
)], (44)
where the non-negative numbers and are the pressure and temperature coefcients of viscosity. Obviously, for
= = 0, (44) yields the classical case with constant viscosity. According to (39) and (43)
1
, the dimensionless
viscosity (44) is given by
= exp(z) (45)
with = (T
1
T
2
)
0
gd, and hence
v =
_
A
0

2
(z + 1) +
k
1

_
exp(z) + k
2
, (46)
where
k
1
=
(V
2
V
1
)
2 sinh(/2)
+
A
0
2
coth(/2)
A
0

and
k
2
=
V
2
exp(/2) V
1
exp(/2)
2 sinh(/2)
+
A
0
2 sinh(/2)
.
We have therefore a one-parameter family of laminar ows, the pressure gradient A
0
being the variable parameter,
which includes two important special cases:
1148 K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150
Fig. 1. Normalized velocity proles of Couette ow for different non-positive values of the parameter .
for A
0
= 0, V
2
= V and V
1
= V, the Couette ow
v =
V
sinh(/2)
_
cosh(/2) exp(z)
_
; (47)
for A
0
= 0 and V
1
= V
2
= 0, the Poiseuille ow
v =
A
0
2 sinh(/2)
{1 [2z sinh(/2) + cosh(/2)] exp(z)}. (48)
Observe that each laminar ow (46) can be thought of as a linear combination of Couette and Poiseuille ows.
Moreover we remark that in the limit as 0, (47) and (48) give the Couette and the Poiseuille ows in a uid
whose viscosity is assumed to be constant (see for example [5] page 154):
v = 2Vz (Couette ow)
and
v =
A
0
2
_
z
2

1
4
_
(Poiseuille ow).
Finally we normalize (47) and (48) by dividing them by V, where, in the former case, V is the velocity of the upper
plate, and, in the latter,
V =
A
0
2 sinh(/2)
_
1
2

sinh(/2) exp
_
cosh(/2)
2 sinh(/2)
1
__
is the velocity at the surface
z =
1


cosh(/2)
2 sinh(/2)
.
Normalized velocity proles of Couette and Poiseuille ows are plotted for different values of the non-dimensional
parameter in Figs. 14. We observe that the normalized velocity proles for Couette ow are convex for negative
values of , that is when the dependency of viscosity on pressure is stronger than that on temperature. Moreover, for
such values of , viscosity is a decreasing function of the height z so that the uid layer having velocity oriented as the
velocity of the upper plate (i.e. as i) is thinner than that with velocity oriented as the velocity of the lower one (i.e. as
i). In contrast, for positive values of , that is when the dependency of viscosity on temperature is stronger than that
on pressure, the normalized velocity proles for Couette ow are concave and the uid layer with velocity oriented
as i is thicker than that having velocity oriented as i. In Fig. 5 we show how the thickness d
+
of the uid layer with
velocity oriented as the velocity of the upper plate depends on the parameter . In Poiseuille ow, for negative values
of the velocity proles attain their maximum at z
max
]0, 1/2[, and as decreases z
max
approaches z = 1/2 where
both pressure due to gravity and viscosity are minimum (see (43)
1
and (45)). For positive values of the velocity
proles attain their maximum at z
max
] 1/2, 0[, and, as shown in Fig. 6, as increases z
max
approaches z = 1/2
at which temperature is maximum whereas viscosity is minimum.
K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150 1149
Fig. 2. Normalized velocity proles for Couette ow for different non-negative values of the parameter .
Fig. 3. Normalized velocity proles for Poiseuille ow for different non-positive values of the parameter .
Fig. 4. Normalized velocity proles for Poiseuille ow for different non-negative values of the parameter .
Fig. 5. Thickness d
+
as function of . For negative values of , d
+
< 1/2 decreases as decreases and in the limit as tends to zero.
If = 0, in particular in the classical case = = 0, d
+
= 1/2. For positive values of , d
+
> 1/2 increases as increases and in the limit as
+ tends to 1.
1150 K.R. Rajagopal et al. / Nonlinear Analysis: Real World Applications 10 (2009) 11391150
Fig. 6. The point z
max
as function of . For negative values of , z
max
]0, 1/2[ increases as decreases and in the limit as tends to
1/2. If = 0, in particular in the classical case = = 0, z
max
= 0. For positive values of , z
max
] 1/2, 0[ decreases as increases and
in the limit as + tends to 1/2.
5. Concluding remarks
We have considered the derivation of the OberbeckBoussinesq approximation within the framework of one of the
simplest implicit constitutive theories: that of uids with pressure dependent viscosity. By using a rigorous perturbative
approach (introduced for the rst time in [15]) we have been able to derive the approximation without ad hoc
considerations. Moreover, we have studied within this framework some simple ows generalizing the results of Hron,
Malek and Rajagopal [8]. What we have found is the presence of asymmetric ow proles and for large values of an
interesting structure: like changes in convexities and layers of transitions. This in marked contrast with the classical
situation and, in some cases, reminiscent of the behavior of complex structured uids (such as inhomogeneous uids or
two-phase mixtures). Our ndings may be of interest not only in the framework of geophysical ow where very high
pressure gradients necessitate taking into account the viscosity dependence of pressure, but also in the application
of uid mechanics in nanodevices. Indeed, in this case also the pressure dependent viscosity cannot be neglected
because of the high pressure gradient that may develop. This effect may be another important feature in describing
ow in microchannels (at least for some ow regimes) besides the presence of slip velocity at the boundaries. For all
these reasons we think that uids with pressure dependent viscosity constitute an important part of uid mechanics
that is not to be ignored.
References
[1] S. Bair, P. Kottke, Pressureviscosity relationship for elastohydrodynamics, Tribology Trans. 46 (2003) 289295.
[2] J. Boussinesq, Th eorie Analytique de la Chaleur, Gauthier-Villars, 1903.
[3] S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability, Oxford Univ. Press, 1961.
[4] D. Dowson, G.R. Higginson, Elastohydrodynamic Lubrication, The Fundamentals of Roller and Gear Lubrication, Pergamon, 1966.
[5] P.G. Drazin, W.H. Ried, Hydrodynamic Stability, Cambridge University Press, 1981.
[6] G.G. Gray, A. Giorgini, The validity of the Boussinesq approximation for liquids and gases, Int. J. Heat Mass Transfer 19 (1976) 545551.
[7] R.N. Hills, P.H. Roberts, On the motion of a uid that is incompressible in a generalized sense and its relationship to the Boussinesq
Approximation, Stability Appl. Anal. Continua 1 (3) (1991) 205212.
[8] J. Hron, J. Malek, K.R. Rajagopal, Simple ows of uids with pressure-dependent viscosities, Proc. R. Soc. Lond. A 457 (2001) 16031622.
[9] H. Jeffreys, Proc. Camb. Phil. Soc. 26 (1930) 170.
[10] H.M. Laun, Pressure dependent viscosity and dissipative heating in capillary rheometry of polymer melts, Rheol. Acta 42 (2003) 295308.
[11] J.M. Mihaljan, Arigorous exposition of the Boussinesq approximations applicable to a thin layer of uid, Astrophys. J. 136 (1962) 11261133.
[12] A. Oberbeck, Uber die W armleitung der Fl ussigkeiten bei Ber ucksichtigung der Str omungen infolge von Temperaturdifferenzen, Ann. Phys.
Chem. 7 (1879) 271292.
[13] A. Oberbeck, Uber die Bewegungsercheinungen der Atmosphere, Sitz. Ber. K. Preuss. Akad. Miss. (1888) 383. and 1120.
[14] K.R. Rajagopal, On implicit constitutive theories, J. Fluid Mech. 550 (2006) 243249.
[15] K.R. Rajagopal, M. Ruzika, A.R. Srinivasa, On the OberbeckBoussinesq approximation, Math. Models Methods Appl. Sci. 6 (8) (1996)
11571167.
[16] Lord Rayleigh, On convective currents in a horizontal layer of uid when the higher temperature is on the under side, Philos. Mag. 32 (1916)
529546.
[17] P.H. Roberts, An Introduction to Magnetohydrodynamics, Longman, 1967.
[18] P.H. Roberts, K. Stewartson, Astron. Nachr. (1977) 311.
[19] P.H. Roberts, G. Veronis, On the Boussinesq approximation for a compressible uid, Astrophys. J. 131 (1960) 442447.
[20] E.A. Spiegel, N.O. Weiss, Magnetic buoyancy and the Boussinesq approximation, Geophys. Astrophys. Fluid Dynam. 22 (1982) 219234.
[21] R.K. Zeytounian, Joseph Boussinesq and his approximation: A contemporary view, C.R. M ecanique 331 (8) (2003) 575586.