You are on page 1of 11

International Journal of Non-Linear Mechanics 40 (2005) 821831

www.elsevier.com/locate/nlm
Boussinesq-type approximation for second-grade uids
Arianna Passerini
a
, Gudrun Thter
b,
a
Dipartimento di Matematica, Universit di Ferrara, Via N. Machiavelli 35, I-44100 Ferrara, Italy
b
Institut fr Angewandte Mathematik, Universitt Hannover, Welfengarten 1, D-30167 Hannover, Germany
Received 7 January 2004; accepted 22 July 2004
Abstract
We derive the equivalent of the OberbeckBoussinesq approximation for second-grade uids. We nd that two new non-
dimensional numbers (in comparison to the Newtonian model) have to be taken into account in order to arrive at the
approximation. In addition we deduce a system which does not neglect dissipative heating together with criteria which qualify
the system to be suitable.
2004 Elsevier Ltd. All rights reserved.
Keywords: OberbeckBoussinesq approximation; Second-grade uids; Onset of convection; Dissipative heating
1. Introduction
Many interesting phenomena involve thermally
driven ows. This concerns such diverse phenomena
as convection in stars, the ocean and the atmosphere,
and the production of pure semiconductors. A very
prominent and useful model for ow patterns of a
uid layer under the inuence of a weight force and
heating from below is the OberbeckBoussinesq ap-
proximation [1,2]. It stems from the end of the 19th
century and is the basis for most of what is known
today about processes where acoustic phenomena as
sound propagation can be neglected. There has been
a long discussion how to justify this approximation in

Corresponding author.
E-mail addresses: ari@ifeuniv.unife.it (A. Passerini),
thaeter@ifam.uni-hannover.de (G. Thter).
0020-7462/$ - see front matter 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijnonlinmec.2004.07.019
some appropriate physical frame and to delineate its
validityhere we refer only to some of the numerous
contributions [39].
In 1996 Rajagopal, R ui cka, and Srinivasa provided
a justication for the approximation within the full
thermodynamical theory for NavierStokesFourier
uids [8]. One crucial idea was to adapt a new
method of treating the constraint of mechanical in-
compressibility introduced by Hills and Roberts
[7]. Simultaneously, the procedure advocated in [8]
clearly xes the validity of the approximation in terms
of quantities which can be measured for concrete
applications.
Nonetheless, due to the rapid technological ad-
vances there are important modern applications for
which the OberbeckBoussinesq approximation does
not seem to be appropriate for a multitude of reasons
(see [10,11] and references therein). For instance, the
822 A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831
linear stressstrain relation of the Newtonian uid
model apparently misses main features of the ow
for solidliquid suspensions occurring as solutions of
macromolecules, blood, glaciers, molten plastic. On
the other hand neglecting the dissipative heating is in-
appropriate for a large class of processes and domains.
Of course, all mathematical theories of nature only
approximate it. They dene materials, which are math-
ematical idealisations of the materials encountered in
nature [12, p. 5f]. This situation leads to two notable
questions:
What would be better (and still simple enough) mod-
els for certain non-Newtonian uids?
What are measurable quantities which qualify mod-
els to be suitable or not?
For some non-standard situations these questions are
answered in [10,11]. In [13] the behavior of second-
grade uids in isochoric convective motion has been
studied. To this end rst an OberbeckBoussinesq-
like system for second-grade uids has been pro-
posed by simply replacing
1
the Newtonian model
by the corresponding non-linear extra stress tensor

T = 2jD + :
1
d

dt
D + :
2
D
2
, i.e. some extra non-
linear terms appear in the equation of motion. Ob-
viously, this choice seems to contradict the idea of
the OberbeckBoussinesq approximation as a rst
order or linear one, and moreover, it does not
seem to be the only one that is possible. Thus, the
convincing formulation of an OberbeckBoussinesq-
type approximation for second-grade uids in some
appropriate physical frame is an interesting open
problem.
It is the aim of our paper to discuss this issue
in the light of the approach developed in [8]. As a
matter of fact, in experiments with polymeric liquids
and other typical non-Newtonian uids the onset of
convection seems to be unaffected by the nature of
the uid. This is also an outcome of the analysis in
[13], where the chosen approach cuts the high fre-
quencies in space. A theoretical investigation seems
necessary.
There are some very severe conceptual difcul-
ties in reconciling statistical mechanics with classical
1
Notation at the beginning of the next section. Note, div v =
tr D = 0.
thermodynamics. Shortly after Boltzmann claimed
to have reduced Thermodynamics to the statis-
tics of molecules, Josef Loschmidt and Ernst Zer-
melo (among others) pointed out several paradoxes.
Maxwell himself (who was the rst to suggest that
thermodynamics might simply be the macroscopic
effect of the statistics of molecules) introduced his
infamous Demon, who continues to bedevil physicists
today.
Within this point of view truncations, which cut all
high frequencies, become not only unavoidable as a
tool to approach solutions in calculations, but even
suitable and in some sense theoretically justied, be-
cause a velocity eld in the mean free path between
two molecules loses all physical sense. It is necessary
to set a frame in which suitable truncations can be sin-
gled out and controlled. In this sense the procedure
given in [8] is particularly suitable, since it does not
lead to require that the layer is thin, which would be
unacceptable if we decide that high spatial frequencies
have to be cut.
In (24) we dene two non-dimensional numbers A
1
and A
2
, which depend on the material constants :
1
,
:
2
and on the conguration through the spatial dimen-
sion L and the mean density j
r
. Namely, A
i
:=
:
i
L
2
j
r
(i =1, 2). They control the effect of the non-Newtonian
part of the stress tensor in the following sense: if A
i
,
i = 1, 2, is small (of order c for some known small
c), then the term is dropped, if it is of order 1 it must
be kept. It is very interesting that our number A
1
is the
same expression (in case j
r
=1) as the one measuring
the non-Newtonian behavior found in [13] by com-
pletely different arguments and that already in [14,15]
the absorption number, which is the reciprocal to A
1
had been implemented.
2
We start our analysis with a preparative section. Af-
ter that we derive the OberbeckBoussinesq approx-
imation for second-grade uids in Section 3, while
Section 4 is devoted to dissipative heating and the cor-
responding modied temperature equation. Precisely,
2
The aim in [14] was to characterize how the term with factor
:
1
crucially inuences the diffusion of velocity and vorticity from a
boundary while [15] studies uniqueness in creeping ow of second-
grade uids. Note that under the thermodynamical restriction 2:
1
+
:
2
= 0 the numbers A
1
and A
2
are of the same order.
A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831 823
the new systems are (notation below):
div v = 0,
v
1
Re
v = div
_
A
1
d

dt
D + A
2
D
2
_
+
1
c
3
(b p) +0b,

0
1
Pr Re
0
=
_
0,
Di
_
(0 +O)v b +
2
Re
D
2
+ A
1

D D
_
.
2. Notation and frame
2.1. Notation
In what follows, boldfaced minuscules stand for
vectors and vector valued functions whereas boldfaced
capital letters represent tensor valued functions, i.e.
v =(v
1
, v
2
, v
3
) (row or column according to the con-
text), T = (T
kl
)
3
k,l=1
. We set T

:= (T
lk
)
3
k,l=1
.
The dot between two quantities denotes the corre-
sponding scalar product between vectors or tensors.
Besides, tr D :=

3
k=1
D
kk
, DD =: D
2
, (this is the
usual tensor product, the result of which is a ten-
sor), D
2
:= D D, and I is the identity. Note, that
tr (D
2
) = D D = D
2
.
For partial derivatives we use two types of the ab-
breviations: j
k
:=
j
jx
k
if k {1, 2, 3} and j

:=
j
j
else. As usual A := j
11
+j
22
+j
33
, := (j
1
, j
2
, j
3
),
and the superposed dot is the material time derivative.
Henceforth let v denote the velocity eld, 0 the
absolute temperature, p the pressure and j the density.
Then, L := v, T is the Cauchy stress tensor, b the
density of external body forces, e the specic internal
energy, and q the heat ux vector. All these quantities
depend on x R
3
and t >0. For a detailed physical
background we refer the reader to [12,16].
2.2. Problem
Consider an innite horizontal layer of viscous uid
inside a gravitational eld the underside of which is
heated while the upper side is xed and held at a con-
stant temperature. This yields a temperature difference
between the top and the bottom of the uid. If it is
big enough to rule out viscosity and conduction, con-
vection will occur. All related forces are expressed as
a nondimensional number called the Rayleigh number
which is the buoyant force divided by the product of
viscous drag and the rate of heat diffusion,
Ra :=
g:jL
3
j
. (1)
Here g is the acceleration due to gravity, : is the coef-
cient of thermal expansion, the temperature differ-
ence between hot and cold end, L the width (vertical
distance between the walls), the thermal diffusivity,
and j the dynamic viscosity of the uid.
Convection sets in when the Rayleigh number ex-
ceeds a certain critical value. If only conduction is
present the corresponding so-called hydrostatic solu-
tion (v
h
, p
h
, 0
h
) has a zero velocity eld while the
temperature decreases linearly from the value 0
b
at the
bottom to 0
t
at the top. Of course, this solution exists
for all values of the temperature gradient.
2.3. Second-grade uids
Dening the extra stress tensor as

T := T + pI,
and writing D for the symmetric part of the velocity
gradient, i.e. D :=
1
2
(v+v

), a second-grade uid
is characterized by the constitutive relation
3

T := 2jD +z(tr D)I +:


1
d

dt
D +:
2
D
2
, (2)
where
d

dt
is the lower convected derivative, namely
d

dt
D := j
t
D + (v )D + v

D + Dv, (3)
and j, z, :
1
and :
2
are material constants, possibly
satisfying the thermodynamical restrictions :
1
>0 and
2:
1
+:
2
=0. Though these restrictions are not neces-
sary to get the system we are looking for, they can be
discussed and found suitable in the light of our result.
Concerning the comparison between a second-order
approximation of a simple uid with fading memory,
3
Mostly the denition

T := jA
1
+z(tr A
1
)I +:
1
A
2
+:
2
A
2
1
with A
1
:= 2D and A
2
:=
d

dt
A
1
is applied. Here we prefer to
use D instead of the A
i
s. Our normal stress moduli :
i
and the z
then differ by the factor 2 or 4, respectively.
824 A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831
and RivlinEricksen model for uids of grade two
(satisfying those thermodynamical restrictions), we re-
fer the reader to [17]. We mention that in looking for
solutions of the corresponding system of partial dif-
ferential equations, the conditions necessary to model
the interaction of the uid with its boundary could
be more restrictive than for Newtonian uids. Since
the equations are of higher order, in some situations
it is necessary to add new boundary conditions to the
Navier-Stokes usual ones in order to obtain well-posed
problems, by means of uniqueness results. This prob-
lem is discussed in [1820].
2.4. General approximation scheme
Starting point: Balance laws (4) and the sec-
ond law of thermodynamics in the form of the
ClausiusDuhem inequality (5).
To close the system we need constitutive relations
for a couple of quantities. They shall obey some
physical principles:
principle of material frame indifference,
entropy principle,
thermodynamic stability.
The key is the generalized incompressibility. If in-
serted into the mass balance the relation directly
leads to an equation for the density.
As independent variables we choose v, 0 and the
mechanical pressure p.
The next step is to supply constitutive relations for
T and q.
Then we express the energy balance in terms of the
Helmholtz free energy.
Now one has to choose the appropriate constitutive
relation for

T. For the usual OberbeckBoussinesq
approximation it is linear (Newtonian), but here
a second-grade uid rheology is chosen (non-
Newtonian model).
At this point we arrive at a highly nonlinear system
in which no term is neglected. It governs the ow of
material which obeys the constitutive assumptions
which were introduced.
Nowwe start the approximation procedure by a non-
dimensionalization. The problem is to choose repre-
sentative values for velocity, time and pressure since
they are not obvious from the conguration.
The next non-trivial choice is the small parameter c
which shall serve as perturbation parameter. In ac-
cordance with the idea of the OberbeckBoussinesq
approximation it shall measure compressibility.
With this non-dimensionalization we arrive at a
system including dimensionless numbers as the
Rayleigh and Prandtl number.
Now we express all independent variables as
power series of c, insert these series into our non-
dimensional system and read off a rst approxi-
mation by balancing the rst c-levels of the series.
The approximation shall be a closed system for the
partial sums of a certain order of the quantities.
One precisely knows, that terms of the order c
k
for
some k N are neglected if working with this ap-
proximation in place of the full system.
3. Derivation of the Boussinesq-type
approximation
In short we state all the steps, though they overlap
with [8] (and [10,11]) to some extent. This is conve-
nient for the reader since we give clear guidance and
correct some misprints in [8]. Moreover, we think that
this provides a better idea of the frame in which our
approximations makes sense.
3.1. Constitutive assumptions
Let us rst quote the Balance laws for mass, linear
momentum and energy. Since we are not interested
in radiative heating, we drop the corresponding term
from the energy balance and use the system
4
j +jdiv v = 0,
j v div T =jb,
j e div q = T L. (4)
It is handy to apply the entropy inequality in the form
of the ClausiusDuhem inequality
j p + div
_
q
0
_
0. (5)
To deduce the OberbeckBoussinesq approximation
one has to augment (4) and (5) with constitutive as-
sumptions and constraints. Notably, the (mechanical)
4
If T is symmetric, then T L = T D.
A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831 825
incompressibility constraint implies several simpli-
cations to system (4). We make the following assump-
tions.
(1) The uid shall be incompressible in the gener-
alized sense [7], i.e.
div v =:

0. (6)
To nd a simple approximation scheme we set : to be
constant (though it could depend on the temperature as
well). Eqs. (6) and (4)
1
directly lead to the differential
equation
:

0 =
j
j
.
Integrating this equation one nds an expression for
j in terms of 0, namely
j =j
r
e
:(00
r
)
, (7)
where the constants j
r
and 0
r
correspond to some con-
venient reference state where density and temperature
are known.
(2) Mechanical pressure: For the stress T we set

T := T + pI with p :=
1
3
tr T. (8)
Due to this denition the trace of

Tis zero. Note, that p
in (8) is not the thermodynamic pressure which would
be xed by an equation of state but stands for the
mechanical or hydrostatic pressure and is connected to
the (hydrostatic) equilibrium state. Since the density
is already determined by (7) we treat the mechanical
pressure p as the third independent variable, which
together with the velocity v and the temperature 0 has
to be determined from the nal system of equations.
Namely, inserting (8) in (4)
2
and taking the divergence
one arrives at
div(j v) + div(p div

T) = div(jb),
j v +jdiv v +p div(div

T)
=jdiv b + j b.
In the last term on the right-hand side we replace b
via (4)
2
and use (6) to calculate div v =:

0 + L L

.
Finally we nd the following equation for p
p
1
j
j p =jdiv b
1
j
j div

T j:

0
jL L

+ div(div

T). (9)
(3) For the heat ux we suppose the Fourier law
q = 0 ( is const.). (10)
(4) Now we express e by introducing the specic
Helmholtz free energy through
T := e 0p. (11)
The Helmholtz free energy measures the amount of
energy one has to put in to establish the system while
the spontaneous energy transfer to the system from the
environment is taken into account. Using (11) we can
rewrite (5). First we calculate with (10)
div
q
0
=
_
|0|
2
0
2

1
0
0
_
.
From (11) we deduce p =
1
0
(

T e +

0p) and nd
that (5) transforms to

j
0
(

T e +

0p) +

0
_
|0|
2
0
0
_
0.
Multiplying this by 0 and replacing j e via (4)
3
by
T L +0 we gain
j(p

0 +

T) + T L +
|0|
2
0
0. (12)
Due to (7) we see that T = T(0) which means in
particular that

T =

0j
0
T. Moreover, (8) leads to T
L = p div v +

T L = p:

0 +

T L. Inserting this
into (12) we obtain
j
_
p +
:p
j
+ j
0
T
_

0 +

T L +
|0|
2
0
0.
The linearity of this inequality (which must hold for
all processes) in

0 implies that the coefcient of

0
must be zero, i.e. p = j
0
T :p/j. Together with
(11) we arrive at
e =T:0
p
j
0j
0
T.
Utilizing this e the energy balance (4)
3
transforms to
(j0j
00
T+:
2
0p)

0 :0 p 0 =

T L, (13)
and the ClausiusDuhem inequality has altered to

T L +
|0|
2
0
0. (14)
826 A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831
We set C(0) := j0j
00
T. After a lengthy calculation
with this denition and the Gibbs equation one sees
that C is the specic heat at constant volume.
(5) In contrast to [8] the extra stress shall obey (2).
Since we xed p as mechanical pressure we observe
that

T = 2jD
2
3
j(tr D)I +:
1
d

dt
D +:
2
D
2
=:

T
N
+:
1
d

dt
D +:
2
D
2
. (15)
Clearly, div

T
N
= j(v +
1
3
(div v)) and

T
N
L =
2j(D
2

1
3
|tr D|
2
). Employing our assumptions
(1)(5) system (4) nally transforms to
j =j
r
e
:(00
r
)
,
div v =:

0,
j v + p =j
_
v +
1
3
(tr D) I
_
+:
1
div
d

dt
D
+:
2
div(D
2
) +jb,
(C(0) :
2
0p)

0 :0 p 0
= 2j
_
D
2

1
3
|tr D|
2
_
+
_
:
1
d

dt
D +:
2
D
2
_
D. (16)
System (16) governs the ow of second-grade uids,
which are incompressible in the generalized sense and
obey the Fourier law.
3.2. Non-dimensionalization
Let U be the representative velocity, L, t
0
and
typical length, time and pressure; j
r
is the reference
density (used in (7)) and g the gravitational acceler-
ation or the acceleration due to another applied body
force (which is proportional to j), respectively. More-
over, we set
:= 0
b
0
t
, O :=
1
2
(0
b
+0
t
),
c
0
:= C(O). (17)
This means c
0
is the specic heat at the mean tem-
perature level. We introduce the non-dimensional
variables:
x :=
x
L
, t :=
t
t
0
, v :=
v
U
, j :=
j
j
r
,
p :=
p

, C :=
C
c
0
, b :=
b
g
, 0 :=
0

O. (18)
Note that C(0)=C((0+O)), 0
t
=
1
2
, 0
b
=
1
2
. We do
not choose U, L and t
0
independently but t
0
:= L/U.
Moreover we x 0
r
:= O and set

C(0) :=
1
c
0
C((0 +O)), v :=
j
j
r
,
Re :=
LU
v
. (19)
The quantity v is called kinematic viscosity while Re
is the Reynolds number. Then (16) in terms of the
nondimensional quantities becomes (we skip all bars
for convenience) after some calculations
j = e
:0
,
div v =:

0,
j v =
1
Re
_
v +
1
3
(div v)
_
+

j
r
U
2
p
+
1
L
2
j
r
_
:
1
div
d

dt
D +:
2
div(D
2
)
_
+
gL
U
2
jb,
(

Cc
0
UL :
2
UL(0 +O)p)

0
:UL(0 +O) p 0
=
2jU
2

_
D
2

1
3
|tr D|
2
_
+
U
3
L
_
:
1
d

dt
D +:
2
D
2
_
D. (20)
A priori there is no obvious representative veloc-
ity. Nonetheless, natural convection processes are
reected if the gravitational potential energy is trans-
formed into kinetic energy. In particular, this means
that in the body force term we observe the balance [6]
gL
U
2
: 1 motivating U :=
_
g:L. (21)
The perturbation parameter is thermal compressibility
c
3
:= : (22)
(unlike [8], where c
3
= Re :). To balance inertial
and viscous effects the Reynolds number should be of
A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831 827
order 1 compared to this c. We calculate for Eq. (20)
3
that
U
2
gL
= : = c
3
. Moreover, with (19) and (21) we
nd, that
L =
Re v
U
=
Re v
_
:gL
L =
1
c
3
_
Re
2
v
2
/g. (23)
We dene two new non-dimensional numbers
A
1
:=
:
1
L
2
j
r
, A
2
:=
:
2
L
2
j
r
. (24)
They can be calculated from the conguration and
the material properties. With (23) we see that A
i
=
c
2
:
i
g
2/3
j
1/3
r
/Re
4/3
j
4/3
. Consequently, if e.g. :
i
or g
are very big then they can be of order 1 together with
Re. Finally, we introduce
:=
j
r

3
_
Re
5
v
5
g
2
, Pr :=
c
0
v

, := j
r
gL.
The Prandtl number Pr, which measures viscous ver-
sus inertial forces, shall be of order 1 at least. Divid-
ing now (20)
4
by we calculate for the coefcients
in (20)
4
, that c
0
UL/ = Re Pr, and,
:=
1

:UL =
:

j
r
gLv Re
=
1

c
3
j
r
g
1
c
3
_
Re
2
v
2
/gv Re =c
2
,
1

:
2
UL =: =c
5
,
1

jU
2
=
1

vj
r
:gL =
1
Re
=
1
Re
c
2
,
:
i
U
3
L
= A
i
j
r
LU
3

= A
i
= A
i
c
2
.
Thus, Eqs. (20) become with our abbreviations
j = e
c
3
0
,
div v =c
3

0,
c
3
j v =
c
3
Re
_
v +
1
3
(div v)
_
p
+c
3
_
A
1
div
d

dt
D + A
2
div(D
2
)
_
+jb,
(Pr Re

C c
5
(0 +O)p)

0 c
2
(0 +O) p 0
= c
2
_
2
Re
_
D
2

|tr D|
2
3
_
+
_
A
1
d

dt
D + A
2
D
2
_
D
_
. (25)
3.3. Approximation
The quantities v, 0 and p are expanded in a power
series with respect to the perturbation parameter c in
the form
v =

k=0
c
k
v
k
, 0 =

k=0
c
k
0
k
, p =

k=0
c
k
p
k
. (26)
The boundary conditions are attached to the respective
zero-order term and consequently, the higher-order
terms have to fulll zero boundary conditions. In sys-
tem (25) all quantities are replaced by these power
series. It is assumed that all quantities are as smooth
as necessary. We analyze the rst c-powers (or c-
levels) to read off the simplest approximate system.
From equation (25)
2
it is obvious that div v
0
=div v
1
=
div v
2
=0. Hence, the term div v in (25)
3
will give no
contribution in our analysis at the levels c
0
c
5
. Anal-
ogously, in (25)
4
the terms c
5
(0 + O)p

0 and tr D
will not appear at c
0
c
4
. We use the Taylor expansion
j(0) = 1 c
3
0 + o(c
6
). (27)
Also, the density enters with its rst c-level only. In-
serting (26) and (27) the velocity equation becomes
c
3
_
1 c
3

k=0
c
k
0
k
+ o(c
6
)
_

k=0
c
k
_
j
t
v
k
+

l+m=k
(v
l
)v
m
_

c
3
Re

k=0
c
k
(v
k
+ o(c
3
)) +

k=0
c
k
p
k
c
3
_
_
A
1
div
d

dt
D
_

k=0
c
k
v
k
_
+ A
2
div
_
_
_
D
_

k=0
c
k
v
k
__
2
_
_
_
_
=
_
1 c
3

k=0
c
k
0
k
+ o(c
6
)
_
b.
Namely, at c
0
we note
p
0
= b.
828 A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831
Therefore the body forces must have a potential to en-
sure the solvability of this equation. Nevertheless, with
regard to applications this is no serious restriction. For
c
1
and c
2
we see p
1
=p
2
=0, which together with
the zero boundary conditions means p
1
=p
2
=0. Let
d

0
dt
D := D(j
t
v
0
) + (v
0
)D(v
0
) + v

0
D(v
0
)
+ D(v
0
)v
0
.
Finally, for c
3
and c
4
we observe
j
t
v
0
+ (v
0
)v
0

1
Re
v
0
A
1
div
d

0
dt
D
A
2
div(D(v
0
)
2
) + p
3
= 0
0
b,
j
t
v
1
+ (v
0
)v
1
+ (v
1
)v
0

1
Re
v
1
A
1
div
_
D(j
t
v
1
) +

i+k=1
((v
i
)D(v
k
)
+ v

i
D(v
k
) + D(v
i
)v

k
)
_
2A
2
div(D(v
0
)D(v
1
)) + p
4
= 0
1
b. (28)
Next we analyze (25)
4
.

C is smooth enough to use a
Taylor series, i.e.

C(0)

C(0
0
) + (0 0
0
)

C

(0
0
)
=

C(0
0
) +c

C

(0
0
)

k=1
c
k1
0
k
,
where

C

:= j
0

C and we obtain
Pr Re
_

C(0
0
) +c

C

(0
0
)

k=1
c
k1
0
k
+ o(c
2
)
_

k=0
c
k
_
j
t
0
k
+

l+m=k
v
l
0
m
__
c
2

k=0
c
k
0
k
+O
__

k=0
c
k

_
j
t
p
k
+

l+m=k
v
l
p
m
__

k=0
c
k
0
k
= c
2
2
Re
_
D
_

k=0
c
k
v
k
_
D
_

k=0
c
k
v
k
_
+ o(c
3
)
_
+ c
2
_
A
1
d

dt
D
_

k=0
c
k
v
k
_
+ A
2
_
D
_

k=0
c
k
v
k
__
2
_
_
D
_

k=0
c
k
v
k
_
.
Let

0
0
:= j
t
0
0
+ v
0
0
0
. We see for c
0
, that
Pr Re

C(0
0
)

0
0
0
0
= 0, (29)
and c
1
raises
Pr Re(

C(0
0
)(j
t
0
1
+ v
0
0
1
+ v
1
0
0
)
+

C

(0
0
)0
1

0
0
) 0
1
= 0. (30)
We observe that v
1
= 0, 0
1
= 0, p
4
= 0 fulll Eqs.
(28) and (30) and have zero boundary values. Due
to our smoothness assumptions these are the unique
solutions.
Finally, we set v := v
0
, 0 := 0
0
and p := p
0
+c
3
p
3
to obtain the OberbeckBoussinesq-type system
div v = 0,
v
1
Re
v div
_
A
1
d

dt
D + A
2
D
2
_
+
1
c
3
(p b) = 0b,
Pr Re

C(0)

0 0 = 0. (31)
Since div v = 0 one calculates that [13]
div
_
A
1
d

dt
D + A
2
D
2
_
=
1
2
A
1

v + A
1
div(v

D)
+ (2A
1
+ A
2
) div(D
2
).
(note that the last term vanishes under the thermody-
namical restriction 2:
1
+ :
2
= 0). Let us summarize
that (31) holds if c is small while Re, Pr, A
1
, and A
2
are of order 1 at least. From our procedure it is clear
that in (31) both v and 0 are precisely of order c and
A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831 829
p of order c
4
since v
1
= 0, 0
1
= 0, p
4
= 0. More
precisely, let (v
f
, 0
f
, p
f
) solve the full system (16),
and (v
a
, 0
a
, p
a
) approximation (31) then
v
f
= v
a
+ o(c
2
), 0
f
=0
a
+ o(c
2
) and
p
f
= p
a
+ o(c
5
).
Let us have a closer look at the velocity equation (31)
2
.
We see that the originally dubious assumption of the
OberbeckBoussinesq approximation: j is constant
in all terms except the body force term, where it is lin-
ear in 0 has transformed to the fact that we combine
two approximation levels. One level is the equilibrium
of body forces and p
0
, the other one is the adjust-
ment between velocity changes (expressed by the dif-
ferential operators on the left hand side) and the buoy-
ancy term. One can hide this fact and obscure the way
in which the different approximations for j enter that
equation by setting p :=
1
c
3
(p b).
4. Natural convection with dissipative heating
Already for a long time in the literature one has
found the supposition that for large domains or other
non-standard applications it is not suitable to neglect
a priori dissipative heating though the processes seem
otherwise near to the regime for which the Boussinesq
approximation works ne. The crucial idea is that the
... effect of viscous dissipation in natural convection
is appreciable when the induced kinetic energy be-
comes appreciable compared to the amount of heat
transferred. This occurs when either the equivalent
body force is large or when the convection region is
extensive. [21, p. 225]. To express this balance Os-
trach introduced the so-called Dissipation number Di
in 1957 [22]. It is the ratio of kinetic energy of the uid
to the heat transferred to the uid and has been used
in a lot of approaches since then (e.g. [10,11,2124]).
If we introduce
Di :=
:gj
r
L
c
0
c
2
= Di Re Pr
as a non-dimensional number of order 1 instead of
the balance of the terms differs from the previous
section in the temperature equation. Namely, instead
of (25)
4
we nd after dividing by Re Pr
(

C c
3
Di(0 +O)p)

0 Di(0 +O) p
1
Pr Re
0
=
2Di
Re
_
D
2

|tr D|
2
3
_
+ Di
_
A
1
d

dt
D + A
2
D
2
_
D. (32)
Here we insert the power series to examine (32). For
this we need the more precise expression

C(0) =

C(0
0
) +c

C

(0
0
)

k=1
c
k1
0
k
+
c
2
2

(0
0
)
_

k=1
c
k1
0
k
_
2
+ o(c
3
). (33)
In this way and also recalling div v
0
=div v
1
=div v
2
=
0, we nd
_
_
C(0
0
) +c

C

(0
0
)

k=1
c
k1
0
k
+
c
2
2

(0
0
)
_

k=1
c
k1
0
k
_
2
+ o(c
3
)
_
_

_
j
t
0
0
+ v
0
0
0
+ c
_
j
t
0
1
+

i+k=1
v
i
0
k
_
+ o(c
2
)
_
Di
_

k=0
c
k
0
k
+O
_

_
j
t
p
0
+ v
0
p
0
+ c
_
j
t
p
1
+

i+k=1
v
i
p
k
_
+ o(c
2
)
_

1
Pr Re

k=0
c
k
0
k
=
2Di
Re
_
D
_

k=0
c
k
v
k
_
D
_

k=0
c
k
v
k
_
+ o(c
3
)
_
+ Di
_
_
A
1
d

dt
D
_

k=0
c
k
v
k
_
830 A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831
+ A
2
_
D
_

k=0
c
k
v
k
__
2
_
_
D
_

k=0
c
k
v
k
_
.
Now we see at c
0
, that

C(0
0
)

0
0
Di(0
0
+O)(j
t
p
0
+v
0
p
0
)
1
Pr Re
0
0
=
2Di
Re
D(v
0
) D(v
0
)
+ Di
_
A
1
d

0
dt
D + A
2
(D(v
0
))
2
_
D(v
0
).
It is reasonable (though for the following not essential)
to assume that b does not depend on the time.
5
Then
the relation p
0
= b leads to
j
t
p
0
+ v
0
p
0
= v
0
b,
and c
1
furnishes

C(0
0
)(j
t
0
1
+ v
0
0
1
+ v
1
0
0
)
+

C

(0
0
)0
1

0
0
Di0
1
v
0
b
Di(0
0
+O)(j
t
p
1
+ v
1
p
0
+ v
0
p
1
)

1
Pr Re
0
1
=
4Di
Re
D(v
0
) D(v
1
)
+ Di
_
A
1
d

0
dt
D D(v
1
) + A
1
_
D(j
t
v
1
)
+

i+k=1
((v
i
)D(v
k
) + v

i
D(v
k
)
+ D(v
i
)v

k
)
_
D(v
0
) + A
2

i+k+l=1
(D(v
i
)D(v
k
)) D(v
l
)
_
. (34)
5
Since b =1 for some scalar 1 always j
t
p
0
+v
0
p
0
=
j
t
1+ v
0
1.
At c
2
we nally collect

C(0
0
)(j
t
0
2
+ v
0
0
2
+ v
1
0
1
+ v
2
0
0
)
+

C

(0
0
)(0
2

0
0
+0
1
(j
t
0
1
+ v
0
0
1
+ v
1
0
0
))
+
1
2

C

(0
0
)0
2
1

0
0
Di(0
2
v
0
b
0
1
(j
t
p
1
+ v
1
b + v
0
p
1
))
Di(0
0
+O)(j
t
p
2
+ v
0
p
2
+ v
1
p
1
+ v
2
b)
1
PrRe
0
2
=
2Di
Re
(2 D(v
0
) D(v
2
)
+ D(v
1
) D(v
1
)) + DiA
1
_
D(j
t
v
2
)
+

i+k=2
_
(v
i
)D(v
k
) + v

i
D(v
k
)
+D(v
i
)v

k
)
_
D(v
0
) + DiA
1
_
D(j
t
v
1
)
+

i+k=1
((v
i
)D(v
k
) + v

i
D(v
k
)
+D(v
i
)v

k
)
_
D(v
1
) + DiA
1
d

0
dt
D D(v
2
)
+ DiA
2

i+k+l=2
(D(v
i
)D(v
k
)) D(v
l
). (35)
Since the solutions are smooth we infer from the
unique solvability that
0
1
= 0, p
4
= 0
because these values fulll the equations and have
zero boundary values. Now we need the c
5
-level in the
velocity equation, which is
j
t
v
2
+

i+k=2
(v
i
)v
k

1
Re
v
2
A
1
div
_
D(j
t
v
2
) +

i+k=2
((v
i
)D(v
k
)
+ v

i
D(v
k
) + D(v
i
)v

k
)
_
A
2
div(D(v
1
)
D(v
1
) + 2D(v
0
)D(v
2
)) + p
5
= 0
2
b. (36)
Eqs. (35), (36) and the known values for v
1
, 0
1
and
p
2
lead to
v
2
= 0, 0
2
= 0, p
5
= 0.
A. Passerini, G. Thter / International Journal of Non-Linear Mechanics 40 (2005) 821831 831
Finally,
6
by setting v := v
0
, 0 := 0
0
, and p :=
p
0
+ c
3
p
3
we obtain the approximate system with
the modied temperature equation. Since div v = 0 it
further simplies since [13, Lemma 1]
_
A
1
d

dt
D + A
2
D
2
_
D = A
1

D D.
Consequently, nally we arrive at

0
1
Pr Re
0 = Di
_
(0 +O)v b
+
2
Re
D
2
+ A
1

D D
_
. (37)
As in the previous section our approximation uses
different levels in the temperature and the velocity
equation and div v = 0. However, in contrast to (31)
here in system (31)
1,2
, (37) all terms are more precise:
both v and 0 are correct of order c
2
and p of order c
5
,
respectively, since v
1
=v
2
=0, 0
1
=0
2
=0, p
4
=p
5
=0.
Thus, the difference to the solution of the full system
is of the order c
3
for v and 0, and c
6
for p, respectively.
Acknowledgements
The second author was supported by the C.N.R.I.
to come to Ferrara for work on the problem.
References
[1] A. Oberbeck, ber die Wrmeleitung der Flssigkeiten
bei der Bercksichtigung der Strmungen infolge von
Temperaturdifferenzen, Ann. Phys. Chem. 7 (1879) 271;
A. Oberbeck, ber die Bewegungserscheinungen der
Atmosphre, Sitzungs Ber. K. Preuss. Akad. Wiss. 383 (1888)
1120.
[2] J. Boussinesq, Thorie Analytique de la Chaleur, Gauthier-
Villars, Paris, 1903.
[3] L. Rayleigh, On convection currents in a horizontal layer
of uid, when the higher temperature is on the under side,
Philos. Mag. 32 (6) (1916) 529.
[4] E.A. Spiegel, G. Veronis, On the Boussinesq approximation
for a compressible uid, Astrophys. J. 131 (1960) 442447.
[5] J.M. Mihaljan, A rigorous exposition of the Boussinesq
approximations applicable to a thin layer of uid, Astrophys.
J. 136 (1962) 11261133.
6
One computes that the next c-levels yield non-trivial
contributions.
[6] A. Giorgini, D.D. Gray, 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, SAACM 1 (1991) 205212.
[8] K.R. Rajagopal, M. R ui cka, A.R. Srinivasa, On
the OberbeckBoussinesq Approximation, Math. Models
Methods Appl. Sci. 6 (1998) 1157.
[9] S. Gnieys, M. Massot, From NavierStokes equations to
the OberbeckBoussinesq approximation a unied approach,
2001, in press.
[10] Y. Kagei, M. R ui cka, G. Thter, Natural convection with
dissipative heating, Commun. Math. Phys. 214 (2000)
287313.
[11] G. Thter, Natural convection, dissipation & power-law
rheologymathematical models & results, Habilitation
Thesis, Hannover, 2003.
[12] C.A. Truesdell, A First Course in Rational Continuum
Mechanics, General Concepts, second ed., vol. 1, Academic
Press, San Diego, London, 1991.
[13] C. Ferrario, A. Passerini, G. Thter, Generalization of the
Lorenz model to the two-dimensional convection of second-
grade uids, Int. J. Non-linear Mech. 39 (2004) 581.
[14] C.A. Truesdell, The natural time of a viscoelastic uid: its
signicance and measurement, Phys. Fluids 7 (1964) 1134.
[15] R.L. Fosdic, K.R. Rajagopal, Uniqueness and drag for uids
of second-grade in steady motion, Int. J. Non-linear Mech.
13 (1978) 131.
[16] K.R. Rajagopal, C.A. Truesdell, An Introduction to the
Mechanics of Fluids, Birkhauser, Boston, 1999.
[17] J.E. Dunn, K.R. Rajagopal, On uids of the differential type,
Int. J. Eng. Sci. 33 (1995) 689729.
[18] K.R. Rajagopal, On boundary conditions for uids of
the differential type, in: A. Sequeira (Ed.), NavierStokes
Equations and Related Non-Linear Problems, Plenum Press,
New York, 1995, pp. 273278.
[19] A.S. Gupta, K.R. Rajagopal, An exact solution for the ow
of a non-Newtonian uid past an innite plate, Meccanica
19 (1984) 158160.
[20] P.N. Kaloni, K.R. Rajagopal, Some remarks on boundary
conditions for uids of the differential type, in: G.A.C.
Graham, S.K. Malik (Eds.), Continuum Mechanics and its
Application, Hemisphere Publication, Washington, DC, 1989,
pp. 935942.
[21] B. Gebhart, Effects of viscous dissipation in natural
convection, J. Fluid Mech. 14 (1962) 225232.
[22] S. Ostrach, Internal viscous ows with body forces, in:
H. Grtler (Ed.), Grenzschichtforschung, Springer, Berlin,
Gttingen, Heidelberg, 1958, pp. 185208.
[23] F.H. Busse, Fundamentals of thermal convection, in: W.R.
Peltier (Ed.), Mantle Convection, Plate Tectonics and Global
Dynamics, Gordon and Breach, New York, London, Paris,
1989.
[24] R. Perez Cordon, M.G. Velarde, On the (non-linear)
foundations of Boussinesq approximation applicable to a thin
layer of uid II: viscous dissipation and large cell gap effects,
J. Phys. 37 (1976) 178182.