Upper Bounds for Convection in an Internally Heated Fluid Layer

Lu Lu
University of Michigan

Introduction

Finding the bounds of certain global quantities is an important approach in the theoretical
study of turbulence. Particularly, maximum convective heat transport in Rayleigh-Benard
convection is a much-studied subject. The idea dates back to Malkus [1]. Following this idea,
a rigorous upper bound of convective heat transport in Rayleigh-Benard convection without
continuity constraint was rst derived by Howard [2], using a variational approach. In the
same paper [2], a single-wavenumber boundary layer approximation is used to study the
asymptotic solutions of the Euler-Lagrange equations. Busse [3] extended this asymptotic
technique and introduced multi--solutions, which has been proved fruitful in studying other
uid dynamics problems [4], [5], [6], and this approach is reviewed in [7]. A new approach to
derive the rigorous bounds on turbulent ow quantities, the background method (DeoringContantin approach), appeared in 1992 [8], and subsequently applied to Rayleigh-Benard
convection [9], [10]. In this project report, these two above-mentioned techniques are applied
to bound the minimum average temperature in a uid layer with internal heating.

Governing Equations
z

y
x

T2

T1

Figure 1: Geometry of convection with uniform internal heating.

The geometry is shown in Fig. (1). The setup is very similar to Rayleigh-Benard convection. A uid layer is conned between two parallel plates with a distance d. However,a
J
uniform volumetric heat ux H(with unit sm
3 ) is pumped into the uid layer. The upper
and lower plates are held at xed temperatures T0 and T1 respectively, and there is no restriction on the temperature dierence T = T0 T1 . With internal heating, the governing
equations are identical to Rayleigh-Benard convection except for the additional term in

176

the heat equation:

u

+ u u = p + 2 u + kgT,
t
T
+ u T = 2 T + ,
t
u = 0,

(1a)
(1b)
(1c)

with the boundary conditions

u|z=0,1 = 0,

T |z=0 = T0 ,

T |z=1 = T1 ,

(1d)

where
=

H
.
c

(1e)

By introducing d as the unit of time, d of length, d of velocity and gd

3 as the unit of
temperature, the governing equations are put into the non-dimensional form
!
"
u

(2a)
P r1
+ u u + p = 2 u + T k
t
T
(2b)
+ u T = 2 T + R
t

where
Pr =

R=

gd5
2

(2c)

R is the heat Rayleigh number, which is proportional to the internal heating rate. The
boundary conditions in non-dimensional form are
u|z=0,1 = 0,

T |z=0 = T0 ,

T |z=1 = 0,

(2d)

where T0 = gd
|T |, which shows that the non-dimensional T0 is equivalent to the role
of the Rayleigh number Ra. A negative non-dimensional temperature at the bottom plate
means the upper plate is hotter. If T |z=0 is 0, then two plates have the same temperature. The case when the bottom plate is hotter corresponds to a positive non-dimensional
temperature at the bottom plate.

Linear and Energy Stability

A static solution can be found easily:

1
1
T = Rz 2 + ( R + T0 )z T0
2
2

(3)

where T |z=0 = T0 . Fig (2) shows several possibilities of the of the static solutions. When
two plates have the same temperature, the prole is symmetric about z = 12 . The maximum
177

Static Solutions
0.2

0.1

T/R

0.1

0.2

0.3
T /R=0
0
T0/R=0.3
T /R=0.6

0.4

0.5

0.6

0.1

0.2

0.3

0.4

0.5
z

0.6

0.7

0.8

0.9

Figure 2: Static solutions of an internally heated uid layer.

temperature occurs at z = 12 . Thus temperature gradient for the lower half of the uid layer
is positive, corresponding to a stable stratication. While in the upper half, the uid layer
is unstably stratied. This prole provides the possibility, when the internal heating rate R
is big enough, that convection starts in the upper half of the uid layer. When T0 increases,
but not exceeding 12 R,the unstable stratication persists with the position of maximum
temperature shifting toward the upper plate. Eventually, when T0 12 R, the temperature
gradient is positive everywhere which suggests that the uid layer is linearly stable, a fact
that will be established in the analysis of linear stability bellow.
Let T = 12 Rz 2 + ( 12 R + T0 )z T0 + , where is the temperature disturbance. We can
write down the linearized equations governing the growth of the disturbances and u
! "
1 u

(4a)
Pr
+ p = 2 u + k
t
"
!

1
(4b)
= z T0 w + 2
t
2
u =0

(4c)

with the boundary conditions

u|z=0,1 = 0,

|z=0,1 = 0.

(4d)

Taking the z component of the curl curl of the u equation leads to

P r 1

2
w = 4 w + 2 ,
t
178

(5)

where 2 is the horizontal Laplacian, dened as

2 =

2
2
+
.
x2 y 2

(6)

Then we make the ansatz

w = est f (x, y)W (z),

= est f (x, y)(z),

(7a)

and
2 f (x, y) = a2 f (x, y).

(7b)

The exchange of stability holds in this problem ([11]). Hence to determine the critical R,
we can set s = 0:
$2
# 2
(8a)
D a2 W = a2 R
!
"
# 2
$
1
D a2 = z T0 W
(8b)
2
with the boundary conditions

W = DW = = 0 at z = 0, 1.

(8c)

In Rayleigh-Benard convection, the equations for linear stability turn out to be identical
to those of energy stability. Thus it is of interest to investigate the energy stability of the
internally heated uid layer. The equations for the disturbances u and are
!
"
1 u

(9a)
+ u u + p = 2 u + k,
Pr
t
"
!

1
(9b)
+ u = Rz R T0 w + 2 ,
t
2
u = 0.
(9c)
Using u equation (9a), equation (9b), we integrate over the whole layer to get
% &
&
%
1 d u2
= |u|2 + $w% ,
Pr
% 2&dt '!
&
" ( %
||2
1 d 2
1 T0
=
z
w
.
2 dt R
2
R
R
Introducing a balance parameter s and adding the above equations yield
) % & % &*
)
&*
" ( %
'!
2
%
&
u2
2
||
T
1
d1
0
s
+
= s |u|2
+ s w +
z
dt 2
Pr
R
2
R
R

(10a)
(10b)

(11)

Let

$ & $||2 %
& %#
|u|2 z 12 TR0 + s w + R
+
,
,
=s
#u2 $
#2 $
1
+
s

2
Pr
R
179

(12)

then if min 0 among the elds with

u = 0,

u|z=0,1 = 0,

|z=0,1 = 0,

the disturbances decay exponentially in time, which implies energy stability. To determine
the critical value of R, we need the Euler-Lagrange equations:
!
"
1
1 T0
2
+ p + u = 0
z
+ s k
(13a)
s u+
2
2
R
R
!
"
1 T0
1
2 +
z+
w + = 0
(13b)
2
2
R
u =0
(13c)
Setting = 0 and applying the ansatz (7) yields
!
"
1
1 T0
2
2 2
z
+ s a2 = 0
s (D a ) W
2
2
R
!
"
1
1 T0
2
2
(D a ) +
z
+ s RW = 0
2
2
R

(14a)
(14b)

with the boundary conditions

W = DW = = 0

at

z = 0, 1

(14c)

For each s, the critical RE (s) can be found by minimizing the eigenvalue R with respect to
the horizontal wave-number a. Then s can be chosen such that it gives a maximal
RE (T0 ) = max min R(a, s)
s

.
The linear stability and energy stability equations are solved numerically, using a Chebyshev spectral method. The results are shown in Fig. (3) and Fig. (4). The rst gure (with
the bottom plate colder) shows the convergence of the linear stability to the straight line
T0 = 12 R, which corresponds the temperature dierence beyond which the entire uid layer
is stably stratied. However, the energy stability increases with T0 but does not converge
to the line T0 = 12 R. In the case of a hotter bottom plate, both the linear stability and
energy stability lines converge to the critical Rayleigh number 1708 for Rayleigh-Benard
convection. This is expected since the uid layer is linearly unstable even without any internal heating in that case. In both gures, the critical R for energy stability is lower than
that of linear stability. This suggests the possibility of subcritical bifurcations.
As shown above, even though the bottom plate is colder and the lower half of the uid
is stably stratied, the system can still be linearly unstable. Once the convection starts, it
tries to lower the average temperature of the uid layer. Thus it is of interest to investigate
the scaling of the minimum average temperature. In the following two sections, this scaling
will be studied with the background method and the multi--solution approach.

180

Linear/Energy Stability (Bottom Plate Colder)

11

10

Linear Stability
Energy Stability
R=2T (Stable Stratification)

10

10

10

10

10

10

10

10

10

10

10

10

10

10

10

10

10

10

Figure 3: Linear and energy stability.

5

Linear/Energy Stability (Bottom Plate Hotter)

10

Linear Stability
Energy Stability
Ra=1708

10

10

10

10
T

Figure 4: Linear and energy stability (Hotter lower plate).

181

10

10

Background method

To apply the background method, rst we decompose the temperature eld T into a background prole (z) and a uctuating part (x, y, z, t):
T = (z) + (x, y, z, t).

(15)

The boundary conditions of T are contained in (z):

(0) = (1) = 0.

(16)

The velocity eld u still satises divergence-free and no-slip boundary conditions:
u = 0,

u|z=0,1 = 0.

(17)

Then the governing equations (2) become

!
"
1 u
+ T k,

Pr
(18a)
+ u u + p = 2 u + k
t

(18b)
+ u = 2 + %% + R w % .
t
-t
Dene < >= limt=1 1t 0 dt% Lx1Ly dxdydz to be the average over both space and time.
Then $u (18a)% yields
%
&
|u|2 = $w% ,
(19)
and $ (18b)%, $ (18b)% yield, respectively,
&
%
& %
&
%
w % = ||2 % % + R $% ,
&
%
&
% &
%
w % = % % + R $ % %2 .

The dierence of the above two identities is

&
%
& % &
%
R ($% $ %) = ||2 + 2 w % %2 .

(20)
(21)

(22)

Since $T % = $ % + $%, we have

&
%
&
% &
%
R $T % = ||2 + 2 w % + 2R $ % %2 .

The identity (23) can also be written as

&
%
0 = a |u|2 a $w% ,

(23)

(24)

where a is a positive number used as an optimization parameter to be adjusted to yield the

best prefactor. Adding equation (24) to equation (23) enables us to express the average
temperature as follows:
% &
(25)
R $T % = 2R $ % %2 + H,
182

where
H=

.%

&
%
& %
&/
||2 + (2 % a)w + a |u|2 .

(26)

We require the functional H to be semi positive denite among the elds u and satisfying
u = 0,

, u|z = 0, 1 = 0,

|z=0,1 = 0.

A bound can readily be obtained by applying the following background prole

)
a
0 z 1
2 z,
(z) = a(1)
1 < z 1
2 (1 z),

(27)

With this background prole, 2 % a vanishes in the interval 1 z 1 . Thus we only

need to estimate | $(2 % a)w% | = a $w% in the region 1 < z 1, and adjust and a
to make H semi positive denite. Following ([9]), an estimate is given by
1
0
&
& 2 %
&
% %
a 2 c %
2
2
.
(28)
|u| +
||
| (2 a)w |
4 4
4c

Then
!
!
"
"
.%
&
%
& % %
&/
&
&
a 2c %
a 2 %
2
2
2
|| + (2 a)w + a |u|
a
|u| + 1
||2 .
16
4c
(29)
The choice

a 2
a
= 1,
c=
64
4
makes the right hand side equal to zero, which ensures H 0. Then
% &
R $T % 2R $ % %2

(30)

(31)
aR
a2 1
(1 )
.
2
4
Minimizing the right hand side with respect to a gives the optimal choice of the parameter
a,
=

a = R.

(32)

4
.
=
3
R

(33)

And can be solved from equation (30),

Finally,

&
%2
$T % 2 $ %
R
1
= R(1 )
4
= R2/3 4R1/3 .

(34)

This shows that as R , $T % R2/3 , which will be veried in the next section by using
the multi- solution approach.
183

5
5.1

Howard-Busse Approach
Formulation of the Functional

The governing equations are repeated here again

!
"
1 u

+ u u + p = 2 u + T k,
Pr
t
T
+ u T = 2 T + R,
t

(35a)
(35b)

with
u|z=0,1 = 0,

T (0) = T0 ,

T (1) = 0.

(35c)

Assume the turbulence is statistically stationary and the velocity eld v and temperature T
can be decomposed into a time independent horizontal average and a meanless uctuating
part:
T = T + ,

= 0 with

u = 0,

= 0..

(36)

After taking horizontal average of (35b), we have

d2 T
dw
= 2 + R.
dz
d z

(37)

Integrate once,
w =

dT
+ Rz + c.
dz

(38)

The integration constant c is determined by integrating above equation over [0, 1]. This
yields
!
"
1
dT
(39)
= w $w% R z
+ T0
dz
2
With the decomposition (36), equation (35b) can be written as

d2 T
dw
dT
+w
+ u = 2 + 2 + R = 2 +
t
dz
d z
dz

(40)

where equation (37) has been used. Multiply both sides with and integrate over the bulk,
we have
(
'
%
&
dT
= ||2 .
(41)
w
dz

Together with equation (39), the above expression yields

'!
" (
& %
&
%
1
R
z
w = ||2 + (w < w >)2 + T0 $w%
2
184

(42)

Another power integral is derived by multiplying equation (35a) by u and integrating over
the bulk:
&
%
(43)
|u|2 = $w%
We can nd an expression of the average temperature by multiplying equation (39) by z
and integrate over [0, 1]:
" (
'!
1
1
1
(44)
w + R T0
$T % =
z
2
12
2

In summary, we have the following expressions:

&
%
|u|2 = $w%
'!
" (
& %
&
%
1
R
z
w = ||2 + (w $w%)2 + T0 $w%
2
'!
" (
1
1
1
$T % =
z
w + R T0
2
12
2
From equation (46), R can be expressed as
&
%
||2 + < (w $w%)2 > +T0 $w%
%#
$ &
R=
.
z 12 w

(45)
(46)
(47)

(48)

Substitute the above expression into equation (47),

%
& %
&
'!
" (
1
1 ||2 + (w $w%)2 + T0 $w% 1
%#
$ &
T0
$T % =
z
w +
2
12
2
z 12 w
2
3 (49)
&
%#
&
%
$
&
%
1
||2 + (w < w )2 > 12 z 2 w
1
$w%
%#
$ &
=
+ T0
.
%#
$ &2
2
12 z 12 w
12 z 1 w
2

Let

h(z) =

1
12 z
2

"

(50)

Notice that
$h% = 0,

&
h2 = 1.

This property of h(z) yields the following identity:

4#
$2 5 4 2 5
w h $hw% $w%
= w $w%2 $hw%2 .

(51)

(52)

Together with equation (45), the average temperature can be expressed as follows,
%
&
"
!

||2 + < (w < w >)2 > $hw%2

< w >
< 12T > =
+ T0
3
$hw%
$hw%
&
%
&
%
"
!
||2 |u|2
< (w h < hw > < w >)2 >
< w >
+
+ T0
3 .
=
< w >< hw >
< hw >
$hw%
(53)
185

Thus the variational problem (T0 = 0 case) can be formulated as

Given =< hw >, nd the minimum of the functional
&%
&
&
%
%
(w h $hw% $w%)2
||2 |u|2
+
F=
$hw% $w%
$hw%2

(54)

among the u, elds with

u = 0,

u|z=0,1 = 0,

where

w = u k,

h(z) =

|z=0,1 = 0,
!

1
12 z
2

"

(55)

Since the functional F is homogeneous in both w and , we can impose two normalization
conditions
% 2& % 2&
(56)
$hw% = 1,
w = .

5.2

Multi- Solution

We are seeking the minimum of the functional F as . This implies that w = h+$w%
(here and in the following discussion the normalization condtions (55) have been assumed.)
in most of the interval 0 < z < 1 , which makes the second term in the functional vanish
in this interval. Only near the boundary z = 0, 1 the boundary conditions prevent a close
appoach of w to h + $w%. And the contribution to
the %functional
& is thus from possible
2
boundary layers at z = 0, 1. Since h(1) + $w% = 3 + |u| > 0 (equation (45 and
denition
(50)), there must be a boundary layer at z = 1. At z = 0, h(0) + $w% =

3 + $w% is indenite. Thus the existence of a boundary layer at z = 0 depends on

whether h(0) + $w% is zero. Without loss of generality, we assume there are two boundary
layers at z = 0, 1 respectively, and make the ansatz
6
6
=
n n + n n ,
(57)
w=
wn n + wn n ,

where n and n satify

2 n = n 2 n .

2 n = n2 n ,

We introduce the following boundary layer variables:

)
n)
for 1 z = O(rn ),
pn w(
w=
sn w(
n1 ) for 1 z = O(rn1 )
)
n)
pn (
for 1 z = O(rn ),
=
n1 ) for 1 z = O(rn1 )
sn (
)
(n )
for z = O(rn ),
pn w
w =
) for z = O(rn1 )
sn w
(n1
)
for z = O(rn ),
pn (n )
=
) for z = O(rn1 )
sn (n1

(58)

(59)
(60)
(61)
(62)
(63)

186

where
n = zrn

n = (1 z)rn ,

(64)

The boudnary layer structure is such that in the interior

1 1 h + $w% ,
w
1 1 + w

(65)

and in the boundary layers

n1 n1 h1 + $w% ,
w
n n + w

n1
w
n n + w
n1
h0 + $w% ,

for n = 1, . . . , N 1,
(66)

where

h0 = h(0) = 3,

h1 = h(1) =

3.

With boundary layer approximations, the functional becomes:

1
FN =
$w%
+

N
6

N
6
1

2pn +rn

qn rn +2sn b2n

!7

)N
6

n% 2 dn +

n2 dn1 + b2
n

"
n 2 dn

"

"

n2 dn1
+ q1

*
4 5
4 59
b21 12 + b2
12
1

"
7
1
"" 2

+ 2
w
n dn

bn 0
0
1
*
! 7
"
7
N
6
#
%
&$

qn rn1 2sn b2n

w
n2 dn1 + b2
w
n2 dn1
12 > +b2
q1 b21 < w
+
12
n
1 w
2

1rN

!7

3rn 2pn qn

1
b2n

w
n%% 2 dn

(w
N N h1 < w >)2 dN +

N
(w
N

h0 < w >) dN

";

(67)

Balancing the exponents in the above exrepssion yields

rn =

1 4n
,
3 4n

qn =

2 4n
,
3 4n

sn = 0,

2pn =

4n
.
3 4n

(68)

Then we have
2

FN = 34N FN

187

(69)

where
FN

) N !7
"
7

6
1
"
n% 2 dn +
n 2 dn
=
$w%
0
0
1
*
!
" 8
7
7
N
9

2
n2 dn1 + b2
n2 dn1
b2n
+ b21 < 12 > +b2
n
1 < 1 >
0

)N ! 7
"
7

6 1
1
%% 2
"" 2

w
n dn + 2
w
n dn
b2n 0
bn 0
1
*
"
7
7
N !
6
$
#

w
n2 dn1 + b2
w
n2 dn1
12 > +b2
12 >
b2n
+ b21 < w
n
1 <w
2

:7

(w
N N h1 < w >)2 dN +

N
(w
N

h0 < w >)

dN

. (70)

Now the Euler-Lagrange equatins for the functional FN can be written down:
8
9
1 D (4)
rN rn

+
$w%

h
(71)
n
1
n n
n+1 n+1 = 0,
$w% b2n n
8
9
1
n h1 + < w > w
n n w
n+1 n+1 = 0,
Dw n%% + rN rn w
n = 1, . . . , N
$w%
(72)
8
9
2
bn+1
n+1 rN rn n+1 h1 + < w > w
n n w
n+1 n+1 = 0,
D w
(73)
$w%
8
9
b2n+1
n+1 h1 + $w% w
n n w
n+1 n+1 = 0,
Dw n+1 rN rn w
n = 1, . . . , N 1
$w%
(74)
And for w
1 , 1 ,
:
D 2
D Dw

1 1
(h $w% + 1) + rN (h + $w% w
1 1 w
1 1 )
b1 w
$w%
2 $w%2
!7
";
7
2

2

+h
(w
N N h1 $w%) dN +
(w
N N h0 $w%) dN
= 0,
0

(75)

:
Dw 2
D Dw
1
(h $w% + 1) + rN (h + $w% w
1 1 w
1 1 )
b1 1 w
$w%
2 $w%2
!7
";
7
2

2

+h
(w
N N h1 < w >) dN +
(w
N N h0 $w%) dN
= 0,
0

(76)

n , n .
The same set of equations are also satised by the starred quantities w
n , n , w
From equation (75) and (76), we have
12 = Dw 12 ,
D w
D w
2 = Dw 2 .
1

188

(77)
(78)

Adding these two identities yields

4
5
% 2
&
12 = Dw 12 + 12 .
1 + w
D w

(79)

Hence the normalization condition < w2 >=< 2 > implies

D = Dw = D.

(80)

This identity together with equation (75) and (76) yields

w
12 = 12 ,

b1 = b1 .

(81)

Equation (73) together with equation (74) gives

2
2
n+1
= Dw n+1
.
D w

. Therefore
Same identity holds for w
n+1
and n+1
2
2
2
2
w
n+1
= n+1
,w
n+1
= n+1

for n = 1, . . . , N 1.

(82)

Substitute the above identity back into equation (73), we have

D
,
< w >
D
= rn rN b2
,
n+1
< w >

n n w
n+1 n+1 = rn rN b2n+1
h1 + < w > w

n = 1, . . . , N 1,

(83)

n+1
n n w
n+1
h0 + < w > w

n = 1, . . . , N 1.

(84)

Then equation (71) and equation (72) become

1 (4)
w
b2n+1 n = 0,
b2n n
n%% + b2n+1 w
n = 0,

(85)
n = 1, . . . , N 1.

(86)

The above equations hold in the region where w

n n ,= h1 + $w%. When the equality holds,
then from equation (71) and equation (72) we can derive:
(4)
n n%%
n%% w
n 2w
n% 2
w
n
2 w
=

=
(h
+
$w%)
1
b2n
wn
w
n5

With the following change of variables,

1/3 2/3

= bn bn+1 n
1/3
= b1/3

bn+1 (h1 + $hw%)1/2 w

n
n

1/3 1/3
1/2
= bn b
n
(h1 + $hw%)

(87)

(88)

n+1

equations (85), (86) and (87) become:

(4)

= 0
%% +
=0

(4) "" 2

"2
=
5

189

(89)

Starred quantities satisfy the same equations with h1 replaced by h0 . This set of dierential
equations have been studied in [3], which gives the constant :
7
7
%%2
)d.

(1
(90)
3 =
d +
0

And the integral

7
7 %% 2
wn
4/3
2
d
+
b2n+1 w
n+1
dn = 3 (h1 + $w%) b1/3
bn+1 ,
n
n
2
b
0
0
n

n = 1, . . . , N 1. (91)

When n = N , the dierential equations for w

N and N are
D
(4)
w
(h1 + $w% w
N N )N = 0,
< w > b2N N
D
%% + (h1 + $w% w
N N )w
N = 0,
< w > b2N N
With the following change of variables,

8
91/3
1/3
1/3

(h
+
$w%)
N

=
b
n
1

#w$

8
91/6
1/3
D
= bn
(h1 + $hw%)1/3 #w$
w
N

8
9

1/6

2/3
D
= b1/3
N
n (h1 + $hw%)
#w$

(92)
(93)

(94)

equation (92) and (93) become

(4) (1 ) = 0,

(95)

%%

+ (1 ) = 0.

In ([2]) the following result is given:

7
7
%%2
=
d =
0

1
d =
4
%2

(96)

(1 )2 d.

Thus the following integrals can be expressed in :

!
"
7 # (4) $2
w

D 2/3 1/3
5/3
dN = (h1 + $w%)
bN ,
$w%
b2N
0
!
"
7
D 2/3 1/3
5/3
% 2

N dN = (h1 + $w%)
bN ,
$w%
0
!
"1/3
7 8
92
D
1/3
5/3
N N dN = 4 (h1 + $w%)
bN .
h1 + $hw% w
$w%
0

(97)

(98)
(99)
(100)

Putting the above integrals together, the functional FN can be expressed as

!
!
"
"
D 1/3 1/3
D 1/3 1/3
D2
5/3
5/3
bN + 4 (h0 + $w%)
bN
,
+ 4 (h1 + $w%)
FN =
$w%
$w%
$w%
(101)
190

and
*
0 4 11/3
11/3
b4n+1
bn+1
3
(h1 + $w%) +
(h0 + $w%)
D=
bn
bn
n=1
!
"
,
D 2/3 +
1/3
1/3
+
(h1 + $w%)5/3 bN + (h0 + $w%)5/3 bN
+ b21 $w% . (102)
$w%
N1
6

)0

Minimizing FN with respect to bn and bn yields

@
! "4/3
! "4/3 A
b2
b
D
= 0 2b1 $w% = (h1 + $w%)
+ (h0 + $w%) 2
,
b1
b1
b1
0
0
1
1
D
bn+1 4/3
bn 1/3
=0
=4
,
bn
bn
bn1
0 14/3
0 11/3
bn+1
b
D
=0
= 4 n
,
bn
bn
bn1
0
0
1
1
bN +1 4/3
bN 1/3
FN
=0
=4
,
bN
bN
bN1
0 14/3
0 11/3
bN+1
bN
D
=
0

=
4
,
bN
bN
bN1

(103)
(104)
(105)
(106)
(107)

where
! "4/3 !
"

(h1 + $w%) $w% 1/2

bN+1 =
,

D
! "4/3 !
"

(h0 + $w%) $w% 1/2

.
bN+1 =

D
From the above relations, bn can be solved
@!
A 1N
"14n
14
b
n
N
N
+1
4 4

(4b
)
bn+1 = 4n1
1
4N1

(108)
(109)

(110)

bn+1 has a similar form

bn+1 = 4n1

@!

bN +1
4N1

"14n

4n 4N

(4b1 )

1
14N

(111)

It is clear from the above expressions that bn ,= bn for n ,= 1 since bN (equation (108) is
dierent from bN (equation (109). b1 can be solved from equation (103) and the recursion
relation

b1 =

@
A* 14N
3
! "
4(14N )
4(14N )
34N
N
4(14 )

N
N
34
34
+ (h0 + $w%)
(h1 + $w%)
25/3 $w%

(112)

191

Putting all these together, the prefactor FN is a function of $w% only:

FN =

D 2 3 4N
< w > 1 4N

= (3 4N )(1 4N )2

4N4N
34N

3 4(14N )
3
! "
34
4(14N )

25/3

4 14N

324N

324N

N
N
(h1 + $w%) 2(14 ) + (h0 + $w%) 2(14 )

134N
$w% 4(14N )

Now the value of $w% can be determined by setting

for $w% is
32c

FN
#w$

(113)

34N

to zero. The resulting equation

( 1)x 2(1c) x x 1c + ( 1) = 0

(114a)

where

3+ < w >
x=
,
3 < w >

3 2c
,
1 3c

c = 4N .

(114b)

For general values of N , the above equation has to be solved numerically:

N = 1,
N = 2,
N = 3,

$w% = 0.4831

$w% = 0.9259
$w% = 1.0120

When N , the above equation can be solved exactly:

3 3
$w% =
= 1.039.
5

..
.

(115)

This
shows that there indeed is a boundary layer at z = 0 since all $w%s are less than
h0 = 3. Now we can write down the scaling of $T % as N :
1
$T % = F 2/3 = 10.2852/3 .
12

(116)

Recalling the identity (47) with T0 = 0:

" (
'!
1
1
w + R,
$T % =
z
2
12
we know that as
1
R.
12
192

(117)

This leads to the scaling of $T % with respect to R:

$T % 4.421R2/3 .

(118)

The proles of w
1 and 1 can be determined from the fact that in the interior of the interval
0 < z < 1,
w
1 1 h + $w% ,

In the case N , h = 2 3z

2 3
5 .

and

w
1 = 1 .

(119)

And then

HK
IK
KK
IK
3K
2
w
1 = JK2 3z
K,
K
5 K

HK
IK
KK
IK
3K
2
1 = JK2 3z
K
K
5 K

(120)

However, whether changes sign in 0 < z < 1 can not be inferred from the variational
problem since only the product of w and appears in the funcional F. Thus the possibility
of w changing its sign can not be excluded.

Conclusion

In this project, the scaling of the min $T % has been studied for an internally heated uid
layer with both background method and multi- solution approach. For the case when
two plates are held at the same temperature these two methods yield the same scaling:
$T % R2/3 . The prefactor given by the background method is about a quarter of that from
the other appoach. By adjusting the background eld we expect the prefactor to be closer
to that predicted by the multi- approach. However, The scaling of the minimum average
temperature when two plates are at dierent temperatures is not clear yet. It is part of our
future work to investicate the scaling in this case.

References
[1] W. Malkus, The heat transport and spectrum of thermal turbulence, Proc. Roy. Soc.
London 225, 196 (1954).
[2] L. Howard, Heat transport by turbulent convecton, J. Fluid Mech. 17, 405 (1963).
[3] F. Busse, On Howards upper bound for heat transport by turbulent convection, J.
Fluid Mech. 37, 457 (1969).
[4] F. Busse, Bounds for turbulent shear ow, J. Fluid Mech. 41, 219 (1970).
[5] F. Busse and D. Joseph, Bounds for heat transport in a porous layer, J. Fluid Mech.
54, 521 (1972).
[6] S.-K. Chan, Innite prandtl number turbulent convectoin, Stud. Appl. Maths 50, 13
(1971).
193

[7] F. Busse, The optimum theory of turbulence, Adv. Appl. Mech. 18, 77 (1978).
[8] C. Doering and P. Constantin, Energy dissipation in shear driven turbulence, Phys.
Rev. Lett. 69, 1648 (1992).
[9] C. Doering and P. Constantin, Variatoinal bounds on energy dissipation in incompressible ows: Iii. convection, Phys. Rev. E 53, 5957 (1996).
[10] P. Constantin and C. Doering, Innite Prandtl number convection, J. Statist. Phys.
94, 159 (1999).
[11] A. Pellew and R. Southwell, On maintained convective motion in a uid heated from
bellow, Proc. Roy. Soc. A 176, 312 (1940).

194