J . Fluid Mech. (1967), vol. 27, part 1, p p .

29-48 29
Printed in Great Britain

Steady free convection in a porous medium

heated from below
By J. W. ELDER
Department of Applied Mathematics and Theoretical Physics, Cambridge

(Received 2 March 1966)

This is an experimental and numerical study of steady free convection in a porous

medium, a system dominated by a single non-linear process, the advection of heat.
The paper presents results on three topics: (1) a system uniformly heated from
below, for which the flow is cellular, as in the analogous BBnard-Rayleigh flows,
(ii) the role of end-effects, and (iii) the role of mass discharge. Measurements of
heat transfer are used to establish further the validity of the numerical scheme
proposed by the author (1966a), while the other flows allow a more extensive
study of the numerical scheme under various boundary conditions. The results
are very satisfactory even though only moderately non-linear problems can be
treated at present.
The main new results are as follows. For the Rayleigh-type flow, above a
critical Rayleigh number of about 40, the heat transferred across the layer is
proportional to the square of the temperature difference across the layer and is
independent of the thermal conductivity of the medium or the depth of the layer.
This result is modified when the boundary-layer thickness is comparable to the
grain size of the medium. The investigation of end-effects reveals variations in
horizontal wave-number and a pronounced hysteresis and suggestsan alternative
explanation of some observations by Malkus (1954).

1. Introduction
The main aim of this paper is a study of the application of the numerical
method proposed by the author (Elder 1 9 6 6 ~ to ) free convection in a porous
medium. From the point of view of fluid mechanics, the principal interest in
convection in a porous medium is that the system is non-linear solely because of
the advection of heat. This is therefore one of the simplest non-linear elliptic
systems.
The paper is concerned with the flow in a homogeneous horizontal slab which
is heated from below. It is mainly in this respect that the flow differs from that of
the previous study (Elder 1966a) in which a uniform temperature difference was
maintained across a vertical slot filled with viscous fluid. I n that case the flow is
largely uni-cellular. The present case is more complex in that, as in the BBnard-
Rayleigh problem, there is no motion below the critical Rayleigh number, the
consequent motion is multi-cellular and is considerably affected by end-effects.
It is of interest to study these effects for a flow in which there is no advection
30 J. W. Elder
or diffusion of vorticity especially in so far as these flows show similar behaviour
to the convection of a viscous fluid, in particular one with large Prandtl
number.
The above point of view has not been the original motivation for this and other
studies. Ever since the original work of Darcy (1856)there have been numerous
studies by hydrologists, petroleum geologists, chemical engineers, geologists and
geophysicists of flow in a porous medium. Until quite recently most of these
studies have been for isothermal flow (see,for example, the review by Richardson
1961),but in the last twenty years or so geophysicists and others have begun to
consider the possibility that convection in a porous medium occurs within the
earth. Two principal areas of interest have so far emerged. The first and better
known is related to the problem of so-called geothermal or hot-spring areas. An
early study was that of Einarsson (1942). The possibility of free convection in
a porous medium uniformly heated from below, and the similarity to the Benard-
Rayleigh problem, was pointed out by Horton & Rogers (1945) and Lapwood
(1948).Wooding (1957 and following papers) has greatly extended these studies,
Elder (1958) and Schneider (1963) have performed laboratory experiments and
Donaldson (1962)has given some numerical calculations. The second, and much
more speculative, possibility is that the earth’s mantle behaves like a porous
medium. This idea has been used in a discussion of earthquake sources by Frank
(1965) and in a model of vulcanism by Elder (1966b). These geophysical studies
have recently been reviewed by the author (1965).
The problem is formulated in $ 2 . The laboratory measurements of heat
transfer are described in $ 3 and a similar numerical investigation is given in 5 4.
I n the previous study it was soon found that most of the difficulties arose in the
application of the boundary conditions. It therefore seemed desirable to study
flows produced by a variety of boundary conditions. This is largely the purpose
of § $ 5 and 6, an investigation of end effects and mass discharge; followed by
a discussion in $ 7.

2. Formulation of the problem

Consider the steady motion of a fluid of kinematic viscosity v, which saturates
a horizontal slab of homogeneous porous material of thickness H , horizontal
extent E , permeability k, and diffusivityt K,. Let all the walls of the slab be held
at temperature Toexcept for a centrally placed portion of the base of the slab of
width L held at temperature (To+ A T ). Making the Boussinesq approximation,
that density variations are significant only in their generation of buoyancy forces,
and that other fluid parameters are independent of temperature, it is readily seen
by inspection of the equations of motion that the problem is specified by: the
acceleration due to buoyancy ygAT, where y is the coefficient of cubical expan-
sion; klv; HIK,; E ; L ; and the acceleration due to gravity g. This last quantity
must be included for problems involving mass discharge. Hence since these six
t It is important to note that K, = K,/pc, where K , is the thermal conductivity of the
saturated medium and pc is the thermal capacity of the fluid. We use K , K for the thermal
diffusivity and thermal conductivity of the fluid.
 Free convection in a porous medium 31
parameters involve only the dimensions of length and time, four dimensionless
parameters are required to define the problem. A convenient set is

A kygATH/Kmv (Rayleigh number),

i
=
B = gHk/K,v (discharge number), (1)
e = E/H and 1 = L / H (aspect ratios).

The only point of novelty here is the parameter B, to be called the discharge
number, which is a measure of the ratio of the imposed pressure forces to the
viscous forces, and can arise in problems involving mass discharge.
The field variables can be conveniently made dimensionless by choosing u n i t s
of length, temperature, pressure, velocity

H , A T , PovKJk, KmIH,
where po is the density of the fluid at temperature To.
For steady motion the field
equations, simplified by the Boussinesq approximation, can be written with the
above units in dimensionless form (Wooding 1957)

v . q = 0,
O = -Vp+AOk-q,
9 .ve = VW,

where q is the velocity, p is the departure of the pressure from its value when
A = 0 , 8 is the temperature, and k is a unit vertical vector. We note that there is
no Prandtl number effect since there are no inertia forces.
I n two dimensions, erecting a Cartesian co-ordinate frame Oxyz so that motion
is in planes y = constant and z is measured vertically upwards, and introducing
a stream function $, we have
w = AO,, (3a)
VZ$ = w, (3 b )
v = a($,@, (3c)
v2e = v, (34
where 8 is the Jacobian operator and x is the horizontal co-ordinate. Note that
q = ( - $B, $), and the vorticity w = V x q = - jw, where j is the unit vector
parallel to the y-axis (into the paper in the diagrams shown here). The source
term V is the rate of generation of temperature and is produced solely by advec-
tion. In the discussion a($., 8) is called the advection. We note that vorticity is
generated by the horizontal gradient of the buoyancy force and that there is no
advection or diffusion of vorticity.
We consider two arrangements of the boundary conditions. The first is for
impermeable walls for which in the two-dimensional analysis: $ = 8 = 0 on the
walls except that 8 = 1 on 1x1 < +l, x = 0. The second is the same except that
we allow parts of the walls to be permeable. This is discussed more fully in 9 6.
32 J . W . Elder

3. Experimental study of heat transfer

3.1. Experimental method
Most of the experimental work was performed in containers filled with uniform
granular material. Most convenient were glass spheres of diameter 3,5,8,18mm.
However, in the largest apparatus with L = 35.6 em and E = 180cm expanded
plastic balls (styropor) of diameter 6mm, glued into a porous matrix, were used.
I n all these cases the apparatus was of circular section.
The upper surfaces of the apparatuses were hollow metal cavities through
which water was circulated from a thermostatic unit which controlled To to
better than 2 0.1 "C. The temperature difference across the apparatus was
measured with a thermocouple and a potentiometer, which could be read to a
precision of i- 0.01 "C. I n spite of this, absolute values of A are known only to
10 % largely owing to uncertainty in k .
All the apparatuses used here were heated electrically, by placing the heater
resistance in a bridge so that its resistance could always be noted. The current
supplied to the bridge was determined by measuring, with a potentiometer, the
voltage developed across a standard resistance. Thus the power input Q could
be measured t o better than 1 yo,but uncertainty in the small heat loss through
the base of the apparatus makes the error in Q about i-5 %.
A less extensive set of measurements were made in a Hele-Shaw cell. Hele-
Shaw (1898) showed that in a cavity of width b < H, L the motion is similar to
that in a porous medium. Some interesting studies using a Hele-Shaw cell have
been reported by Saffman & Taylor (1958) and Wooding (1960a). The present
experiments confirm this prediction for the case of non-isothermal flows-the
phenomena found here with the beds of spheres can be duplicated in a Hele-Shaw
cell. Only two-dimensional motions can be investigated, but it is a simple matter
to observe the motion with suspended aluminium particles.
Similar arguments to those used by Hele-Shaw can be applied to the equations
for non-isothermal flow (Wooding 1960a). Let U be the velocity scale and S the
smallest length scale of the motion, then three conditions are necessary:

(9 b/&< 1,
(ii) the Reynolds number Ub2/vS-g 1,
(iii) the Peclet number Ub2/K8@ 1.
These conditions ensure that there is negligible advection of vorticity and rapid
diffusion of vorticity and heat across the flow. For the experiments described
below departure from the Hele-Shaw approximation occurred when b/6 2 1.

3.2. Heat transfer for 1 $ 1-f

The quantity of first interest is the Nusselt number N , a dimensionless thermal
conductivity defined by
Q = N K , (heated area) ATIH, (4)
-t A brief description of the data of 3 3 has already been given in Elder (1965).
 Free convection in a porous medium 33
where Q is the power transferred across the slab. Figure 1 gives data obtained:
( a ) in a medium of glass spheres, ( b ) in a Hele-Shaw cell. The accuracy of the
laboratory measurements is of order k 10%. Only a small portion of the
laboratory data is presented here. However, it is worth remarking that no
100

50

20

N lo

1
10' 102 103 10'
A

10

20 50 100 200 500 1000

A
FIGURE 1. Heat-transfer characteristic N ( A ) , Nusselt number as a function of Rayleigh
number: ( a )granular material; 0,
8 mm spheres; 0 , 18 mm spheres; ( b ) Hele-Shaw cell.
3 Fluid Mech. 27
34 J . W . Elder
significant differences were found in N ( A )for media of sand, gravel, rashid rings,
peas and children's marbles (the spheresof diameter 18mm referred to in figure 1).
The present results for granular material are similar to those of Schneider (1963).

3.3. Heat transfer :quadratic region

The heat-transfer measurements for A < 5000 and 1 9 1 are shown in figure 1.
+ +
Below A 40, N 1 so that the heat transfer is entirely by conduction. Indeed,
the measurements here can be used to determine K,. At A = 40 there is an
abrupt change in N ( A ); these experiments give this initial value
A, = 404 lo%,

1000 I I I

N* 100

I I I
10
108 lo9 10'0 10"
A*
FIGURE 2. Heat-transfer characteristic N*(A*) in the viscous-flow region with Nusselt
number N* and Eayleigh number A* based on the fluid properties alone: + , 6 mm
styropor balls, L = H = 35.6 cm, diameter of slab 180 cm; 0, 3 mm styropor balls,
diameter of slab 16 cm; x , 5 mm glass spheres, L = H = 8.0 em, diamcter of slab, 16 cm.

consistent with the prediction of 4n2of Lapwood (1948).Above A, the points lie
reasonably close to
N = A140 & 10 yo. (6)
I n dimensional form (6) becomes
&/(heatedarea) = k p y g ( A T ) 2 / 4 0 , (7)
a quadratic relation between Q and A T . We note that Q is independent of H
and K,.
3.4. Heat transfer :viscous boundary-layer region
Above some value of the Rayleigh number, A > A , say, the experimental points
begin to fall below (6). Some of these measurements are also shown in figure 1.
This behaviour has also been noted by Schneider (1963). It is no longer possible
to correlate the data of N ( A )solely by the parameters (1). Clearly an additional
quantity is required. The initial departure from (6) closely follows N I XA t , as in
simple viscous convection, suggesting the use of A* E y g A T H 3 / m ,the viscous
 Free convection in a porous medium 35
flow Rayleigh number. This has been done in figure 2-where N", A* are now
defined in terms of the parameters of the fluid alone.
The initial departure from (6) now follows
N * / A * i = 0.26 & 0.02, (8)
remarkably close to the value 0.24 found by Silveston (1958)for viscous laminar
convection between horizontal planes. At much higher values of A* the curve
follows
N*/A** = 0.1 f 0.02 (9)
again reasonably close to the viscous flow value of 0.06 given by Jakob (1949).
The data for (9) were obtained with L = 35.6 em and 6 mm plastic balls as the
medium.
Assuming both (6), (8) valid at A , we have

A, + 20(H2/k)*/&, (10)
where 5 =.I,. Writing at A , the Nusselt number N, = H/26,, we find typically
in these experiments 8,lA + 0.2, where A is the diameter of the spheres. It is not
surprising that relations of the form (8) and (9) occur when the boundary-layer
thickness is somewhat smaller than the scale of the porous medium. As noted by
Wooding (1958)it is to be expected that equations (2) are no longer valid within
a distance of order A from a boundary-indeed this is implicit in the definition
of q.
Here the relations (8), (9) are merely an inconvenience in so far as they limit
the range of A accessible to a given apparatus. However, in the problem of con-
vection in the earth's mantle, assuming that the mantle behaves like a porous
medium, the change from porous to viscous flow may be of importance in
describing the region between the lower and upper mantle.

4. Numerical study of heat transfer

4.1. Numerical method
A suitable finite difference representation of (3) is readily found by well-known
methods (Fox 1962). The equations are solved as written in (3) in the order
(a,b, c, d , a, ...) until the solution has converged or the calculation is terminated
when the solution begins to diverge. The solution of the Poisson equations ( 3 b ) ,
( 3 4 is very straightforward since both @ ( = 0 ) and 0 are given on the boundary.
Leibmann's extrapolated method with alternating directions of scan is used.
These matters are discussed elsewhere (Elder 1966a).
Most of the solutions are for a mesh spacing of order 1/20for which the accuracy
is of order 1yo.Solutions are generally terminated when values over the mesh
change by less than 0.1 yo.

4.2. Convergence of the solutions

The convergence of the solutions is indicated in figure 3 (obtained during the
solutions for figures 4, 5). A parameter typical of the temperature field, the
Nusselt number, is plotted against the number of scans of (3) for various Rayleigh
3- 2
36 J . W . Elder
numbers. .It is seen that the rate of convergence is more rapid a t higher values of
A and that for A 2 80 there is some overshoot. At Rayleigh numbers near the
critical value of 47r2 (Lapwood 1948) the convergence is extremely slow. It is of
interest to comment on this observation.

i
I
8

.
-
3.
z
I
3

0 100
S

FIGURE 3. Convergcnce of the numerical solution for I 1. Nussclt number N as a function

of the numbor of scans s at various Raylcigh numbers A . Values scaled using N o = 1 and
the final value N,. Mesh 21 x 21.

As Garabedian (1956) has pointed out, the solution of an iterative procedure,

such as that used here, converges in a manner similar to that of the corresponding
time-dependent problem at large times. We may obtain a rather crude estimate
of the relation between the number of scans s and the time t by solving (3) with
the same initial conditions (0 and $ = 0 everywhere except for 0 = 1 on x = 0 )
a t A = 0. We find that the final solution is approached exponentially so that, for
example, if N is the Nusselt number, N, - N N e-slT and rcc d P . For the data of
figure 3 wit8hd = 1/20 we find 7 = 2.0 scans so that r = 0.005d-2 scans. From the
theory of heat conduction (Carslaw & Jaeger 1959, 53.4) we obtain a similar
result with a time constant l/7r2. Hence, from the data of figure 3 we can relate
the number of scans s to the dimensionless time t by

t N (20OdZ/7r2)s. (11)
We note the important result that as the critical Rayleigh number is
approached (from above) the time for the establishment of the steady motion is
considerably larger than the time to heat the slab solely by conduction. For
example, at A = 50, as shown in figure 3, the solution has converged by s E 80
 Free convection in a porous medium 37
when, from ( 1 I ) , t x 40/n2,
corresponding to a rate 40 times slower than that of
conduction alone. For this reason attempts to study the instability of a layer of
fluid suddenly heated from below, by the assumption that the temperature field
is quasi-steady, are bound to fail (Foster 1965).

4.3. Heat transfer :quadratic region

The Nusselt number has been evaluated for some of the numerical solutions.
These values are given in table 1. They show, within the numerical accuracy, for
A > 40 that approximately Ncc A . The values are in excellent agreement with
the laboratory values.

A l/d = 10 15 20 25 30 40 00 100Nm/A
45 - - -_ (1.09) - - 1.20 2.66
50 >1.01 1.19 1.26 1.30 - - 1.36 2.73
60 1.36 1.51 1.58 1.62 1.64 - 1.70 2.83
70 - - - 1.87 1.90 - 1.92 2.75
80 1.80 1.96 2.04 2.08 2.12 - 2.02 2.52
90 - - - 2,27 2.30 - 2-33 2.58
100 2.09 2.28 2.37 (2.42) 2.46 2.51 2.62 2.62
TABLE1. Numerical values of the Nusselt number N ( A , d ) in a square box with mesh
spacing d and Rayleigh number A. The values d = co are obtained assuming N , - N d2,
N

and are cstimatcd t o about _+ 2 yo. The values in brackets are for l/d = 24. (PM1-32.)

4.4. The flow jield for 1 9 1

When 19 1, the flow field has either no motion or is composed of cells. This
cellular regime is predicted by the linear theory and confirmed both in the
laboratory and numerically.
Figure 4 shows the temperature distribution? in a half-cell for 1 9 1 at various
Rayleigh numbers, obtained numerically. Below A + 40, $ = 0 and 0 = 1 - 2 .
The numerical solution is very stable; any disturbance decays rapidly. As
A increases beyond 40 the isotherms steepen till mushroom-like distributions are
produced. This feature has already been pointed out by Wooding (1957), Elder
(1958) and Donaldson (1962).
Figure 5 shows the stream function and the advection at A = 120. The form
of the stream function is not greatly different from its value near A, except that
it is a little twisted. For 1+00 with equations ( 2 ) ,it is readily seen that the phase
of the cellular motion is theoretically indeterminate. Both in the laboratory and
numerically it is found that if 1 % 1 but finite the phase is determined by the
motion in the ends of the slab (see $5 5.1, 5.3).

4.5. Discussion of the heat transfer data

A result of the form (6)is to be expected. When A 9 A, so that there is a boundary
layer on z = 0, we have by (4)that N N116where Sis a measure of the boundary-
layer thickness. But if W is a vertical velocity scale typical of the layer it follows
Not,e: in these figures the vertical scalc is 5/3of the horizontal scale.
38 J . W . Elder
from ( 2 c ) ,assuming S,, B Ox,, that W N 118. Finally, we note from the z-compo-
nent of ( 2 b )that W A . Hence we expect N A . This result, therefore, implies
N N

that there is a balance between buoyancy forces and viscous forces and that
while the heat enters the slab by conduction it is carried away from the immediate
vicinity of the heated surface by vertical advection which is dominant in the
outer portion of the boundary layer on the heated surface.

A=50 60 80 100 120

Temperature
FIGURE4. Temperature distribution in a half-cell calculated a t Rayleigh number
A = 50, 60, 80, 100, 120. Isotherms drawn at intervale O(0.2)l. Mesh 21 x 21.

5.707 18.14
Stream function Advection
FIGURE 5. Stream function and advection in a half-cell calculated a t Rayleigh number
A = 120. Contours drawn a t intervals of -1(0.2)1 of the maximum absolute value
(indicated below the figure). Mesh 31 x 21.

The consistency of both the experimental and the numerical data in the quad-
ratic region of the empirical expression (6) confirms the experimental procedure,
the validity of the Hele-Shaw cell approximation for non-isothermal flows and
the numerical method. We may therefore proceed using the numerical method with
some confidence. Our interest now will concentrate on the role of different
boundary conditions from those used above. While our immediate objective is
to test the numerical scheme under diverse boundary conditions the flows are of
considerable interest in themselves.
 Free convection in a porous medium 39

5. End-effects
In spite of the numerous studies of free convection, it is surprising to note that
there has been relatively little consideration of end-effects, viz. for the present
problem the peculiarities which arise when the aspect ratios 1 and e are finite.

~~

1.000- 5.230

1.000 5305

1000 5156
FIGURE 6. End-effects. Isotherms and streamlines in a slab of horizontal extent e = 10
calculated a t Rayleigh number A = 80, mesh spacing d = -& and various values of the
heater length I ; (a)and (b), I = 2; (c) and (d), 1 = 3; ( e ) and (f), I = 4; ( 9 ) and ( h )I = 6 .
Apart from the contour 0.03, the contours are at intervals -1(0*2)l of the maximum
value of the function.

There is a great variety of end-effects in free convection. Some of the author’s

experiments were reported at the 1963 British Theoretical Mechanics conference
at Liverpool when his attention was directed to somewhat similar phenomena in
Couette flow being studied by Coles (1965). To take an analogy from atomic
physics, the end-effects tend to remove a degeneracy of the system. Much work
remains in this subject but the present results are of some interest.
40 J . W . Elder

5.1. Numerical experiments o n a simple end-effect

The numerical solution for a flow in a slab of horizontal extent e = 10 for
1 = 2 , 3 , 4 , and 6 at A = 80 are shown in figure 6. Only the right-hand half ofthe
flow is shown.

N
3.5

3
rn \ 2
I I
4
I

G I
(b)

2.5

n
2

1.5

0.5

0
FIGURE 7. End-effects, role of heater length I on: ( a )wavelength h of inner Rayleigh cell;
( b ) Nusselt number N ; (c) maximum value of the streamfunction $e (of the end cells).
Parameters as in figure 6. (PM 47-65.)
 Free convection in u porous medium 41
For 1 = 2 , figures 6 ( a )and ( b ) ,there are two regions of opposite circulation.
These end cells give a fairly narrow column of heated fluid rising above the heated
region with a more diffuse return flow in the outer portion of the slab. We note
that the flow distant from the heater, near the vertical walls, is very weak. How-
ever, for l > 2 in addition to the end cells there are nearly square cells, occurring
in pairs of opposite circulation, above the central portion of the heated surface.
We shall call these Rayleigh cells. For 1 = 3, figures 6 ( c ) and ( d ) ,there is one pair
of Rayleigh cells. Incidentally, we note that the flow on x = 0 is downward.
Clearly the phase of the Rayleigh cells is set by their number and the phase of
the end cells, which is always as in figure 6 (b).I n figures 6 ( e ) and (f)with 1 = 4,
the motion is as for 1 = 3 except that the Rayleigh cells have a much larger
wavelength. For 1 = 6, figures 6(g) and (h),three pairs of Rayleigh cells appear.
The gross changes of the system as 1 is varied are summarized in figure 7 which
shows: the wavelength h of the inner pair of Rayleigh cells for the same arrange-
ment used in figure 6; the Nusselt number N ; and the maximum value of the
streamfunction $,, which in fact corresponds to that of the end cells.
The breaks in the curves correspond to the appearance of pairs of Rayleigh
cells at 1 = 2.0, 4.3, 5-4. The changes in wavelength are closely paralleled by
corresponding changes in N and $,.

5.2. Linear anulysis of the simple end-effect

A partial understanding of the above phenomena is possible with an analysis
similar to that used by Zierep (1961)for the correspondingproblem with a viscous
fluid.
As A+O, $ + O and V2B-+0, so that we write the temperature field as
O = p ( l - z ) ; w h e r e p = 1forIxl <@tobecalledregionlandP=Oforlxl > $ l ,
to be called region 2 . This is a rather crude representation of 8 but sufficient for
the moment. Consider a steady perturbation (Or, $') superimposed on this state
of rest. Retaining only first-order quantities
{V4+pA(a2/t3~2)}
(O', $') = 0, (12)
where (O',$') = 0 on z = 0 , l .
The field is assumed symmetrical about x = 0, so we refer only to the region
x 3 0. Suitable solutions for regions 1, 2 are
0; rrx sin nm,
= u cos m
8; = u cos amrrl e-n(X-41) sin nnz,
where n = 1,2,3, ..., u is an arbitrary constant and we have placed 0; = 0; on
x = 41. Note that, when 1 is an odd integer and m = 1,8; = 0. This feature arises
from the approximation to the original temperature field. These are solutions of
( 12) provided
A = n2(m2+n2)2/m2. (14)
A has a minimum value when m = n of 4r2n2.The first occasion when this is
possible is when n = 1. Hence only for A 2 4n2 is the above solution possible.
For 1 = 00, 4772 is the critical Rayleigh number A , (Lapwood 1948). For 1 finite
and A < 4n2 the complex roots for m must be used. Here we simply consider the
42 J. W . Elder
situation A = 4n2,m = n = 1. It is seen that square cells appear in 1x1 < $1 for
I 2 2; with I < 2 there is merely a single column of rising heated fluid. The end
cells decay exponentially in the lateral direction.
This analysis suggests that successive pairs of Rayleigh cells will appear when
1 = 2 , 4 , 6 , .... In the numerical calculations of $5.1 we found these values to be
1 = 2.0, 4.3,5.4 at A = 80. The flow at A = 80 cannot be expected to satisfy the
linear theory, yet the progression of events is moderately well described.
A very crude discussion of the variation of the Nusselt number as a function of
1 and e is now possible. Let us assume as in (6) that each Rayleigh cell transfers
an amount of heat corresponding to a Nusselt number of aA and each end cell an
amount a‘A, where a is a constant and a‘ is independent of A . Then, if we write
I = 2(c + r ) where r is an integer, r is the number of half Rayleigh cells and c the
length of the heated boundary accessible to an end cell. Hence the total Nusselt
number is 2(ca‘+ ra).We try various guesses for the form of a’. If a’ = const., the
curve N ( l / e )continuously rises between the breaks which occur when r increases
by unity, contrary to the data of figure 7 ( b ) .If a ’ c ~1/c the curve similarly falls.
A crude compromise is a’ = a/cJ, where we choose for want of a better assumption
a‘ = a when c = 1. This leads to a curve roughly similar to that of figure 7 ( b )but
with the breaks at r = 2 , 4 , 6 , .... The principal oversimplification of the above
discussion is its neglect of the interaction between the end cells and the Rayleigh
cells, clearly shown in the data of figures 6 and 7 and especially in the experiment
of the following section. Nevertheless, the general form of the relation #(I, e ) is
indicated. Finally, we note therefore that the rapid fall in N for 1 5 0.5 shown in
figure 7 ( b )is probably a finite difference effect,viz. that the heater is represented
by too few mesh points.
5.3. Hysteresis
One of the most interesting effects in the experiments with the Hele-Shaw cell
is found with 1 finite but in which the horizontal extent of the slab is very much
larger thanl. Figure 8, plate 1, isaphotographic sequence of the flow revealed with
suspended aluminium powder. Each of these flows has been left for sufficient time
for it to come into a steady state. Figure 8 ( a ) shows a regular group of nearly
square cells above the heated surface with weaker circulations in the end regions.
The motion is quite steady. I n figure 8 ( b )a t a somewhat higher Rayleigh number
the distribution and width of the cells is rather irregular. The motion is very
slowly but continually changing. We note the growth of the cells in the end
regions. This process continues in figures 8 (c), (d). I n figure 8 ( d ) the inner cells
are again nearly square and the motion is nearly as steady as in figure 8 (a).In
figures 8 ( e ) ,( f ) the process continues. It is seen that the progressive change in the
flow is dominated by the growth and encroachment of the cells in the end region.
In general there are an even number of Rayleigh cells and the motion is steadiest
when the Rayleigh cells are square.
The motion of figure S ( f ) persists to large Rayleigh numbers. On the other
hand,if the Rayleighnumberisnow very slowly reduced, the motion of figure 8 ( f )
persists to Rayleigh numbers about that of figure 8 (a).The first reappearance is
shown in figure S(g). Here we have a pronounced hysteresis. It follows that to
specify the thermodynamic state of the system the path by which that state was
 Free convection in a porous medium 43
achieved should also be specified. A somewhat similar hysteresis, although more
complex, has been observed by Coles (1965) for the flow between concentric
rotating cylinders.

5.4. A comment on an experiment by Malkus

It is possible that this phenomenon of the appearance of end-effectsand in parti-
cular the variation in horizontal wave-number provides an alternative explana-
tion to that given by Malkus (1954) for his measurements of ‘transitions’ in
convection in a viscous fluid.
Malkus has reported, in his experiment to determine the relation between the
heat transfer and the Rayleigh number for a horizontal layer of a viscous fluid,
that there were discontinuous changes in the slope of the curve. He has inter-
preted these changes as due to the progressive appearance of higher modes of the
linear stability analysis. That is, changes occur when the Rayleigh number is
sufficiently high for the vertical wave-number to have values n,2n,3n,....
It has not been realized hitherto that an alternative possibility could be the
availability of additional horizontal wave-numbers. Some experiments by the
author for the Rayleigh-BBnard problem also reveal pronounced end-effects.
I have not; however, been able to reproduce the rather small changes of slope in
N ( A ) found by Malkus. These experiments are to be reported.

6. Mass discharge
A problem of particular interest to the geophysicist for studies of both hydro-
thermal systems and penetrative convection in the upper mantle is the flow
when it is dominated by mass discharge. The laboratory experiments and their
relation to hydrothermal systems have already been described (Elder 1965); here
we shall discuss the numerical experiments.
Our particular interest, keeping in mind other possible applications of the
numerical procedure, is to show that the method used below is satisfactory in
dealing with the boundary conditions which arise with mass discharge in a
convective system.
Two extreme possibilities arise: forced and natural discharge. I n forced dis-
charge we consider that the discharge arises from the ambient pressure field,
e.g. due to variations of the water-table level. In natural discharge we consider
that the ambient variations of the pressure are negligible ;the discharge is driven
solely by the pressure field arising from the differential heating of the fluid.

6.1. Forced discharge

Figure 9 shows the results of a calculation with forced discharge. The vertical
velocity 1c., has been specified on z = 1, to be finite on 1x1 < 4 and elsewhere to
be zero, and the velocity has been required to be horizontal on the vertical ends
of the slab. The flow field for A = 0, obtained from Vz$ = 0, is also shown in
figure 9, together with the flow field when there is no discharge. The motion when
there is both discharge and convection can be regarded as a non-linear inter-
action of these two fields. The discharge aids the motion in the end regions so that
44 J . W . Elder

7574

972 1

FIGURE 9. Forced discharge, calculated distributions of temperature and stream function.

Discharge over zone of width 1. A = 100, I = 2 , e = 5 , 1 0 1 x 21 mesh. Only the right ba!f
of the flow is shown. ( a )and ( d ) no discharge; ( b ) and ( e ) discharge velocity 2.5; ( c ) and (f)
discharge velocity 5 - 0 ; ( 9 ) A = 0 with discharge. (POR 36.)
 Free convection in a porous medium 45
the ordinary cellular mode is annihilated with the discharge used here. An
important feature (Elder 1965) of the flow is that even at quite large discharge
rates some of the fluid is recirculated even though, when the discharge velocity
is 5 as in figure 6 ( c ) ,the problem is essentially one of forced convection.

6.2. Natural discharge

For a natural discharge we assume that the pressure p is maintained at zero on
z = 1. Physically this implies that any fluid which passes above z = 1is allowed
to run away and is lost t o the system. Here the return flow is also allowed to enter
only on z = 1. The calculations proceed thus. After each scan of (3) we solve
V2p = AB,, (15)

I
4127
FIGURE
10. Natural discharge, calculated isotherms (above) and stream function.
A = 50, E = 1 , e = 5, mesh 101 x 21. (POR 67.)

which follow from ( 2 a )and (2b). Since on x = 1 we requirep = 0 and 0 = 0, from

( 2 b ) the surface vertical velocity q+= = -pz. I n the calculation we therefore, at
the end of each scan, simply impose the stream function on z = 1 to be
rx

where in (16)p is the solution of (15). This has proved to be a satisfactory scheme
although somewhat slower than the simple scheme (3).
Figure 10 shows the temperature and stream function distributions for a
Rayleigh number A = 50. Recharge occurs over the bulk of the upper surface,
the heated fluid rises as a fairly thin column even at this rather low Rayleigh
number, and only about 30 yoof the fluid in the plume is recirculated. I n natural
46 J . W . Elder
systems with very large Rayleigh numbers we anticipate that only a very small
portion of fluid is recirculated.
Figure 11 shows two characteristic parameters of the flow. Figure 11( a )shows
the local Nusselt number N' = N ' ( x ) on z = 0 and x = 1. The pronounced dis-
continuity at the ends of the heated surface is simply a feature of our boundary
conditions. What is of considerable geophysical interest is that the heat flux on
z = 1 falls rapidly to zero outside the region of the discharge. Figure l l ( b ) shows
the discharge velocity. Discharge occurs over a region of width 1.4; elsewhere the
flow is into the slab. This inward flow is sufficient to reduce the surface heat flux
to zero.

10

5
N

-5 I I I I I -5
-2 -1 0 1 2 -2 -1 0 1 2
X X
(0) (b)
11. Natural discharge, parameters as in figure 10. ( a ) Heat flux N ' ;
FIGURE
( b ) velocity through the upper surface W .

I n the Hele-Shaw cell the natural discharge is simply arranged by having a

small gap in the lid (through which the coolant is circulated) leading into an
upper wide chamber full of fluid. Figure 12, plate 2 , is a photograph of such a flow.
It should be compared with figure 10. Even though it is a t a much higher Rayleigh
number, the gross features, including the recirculation, are similar to the
numerical solution.
The solution to the linearized problem, following Lapwood (1948), is obtained
simply by requiring n = +,4, ... in (13) so that p = 0 on z = 1. Motion is first
possible with n = 4and the corresponding critical Rayleigh number of n2.Other-
wise there is no novelty in this solution. However, at Rayleigh numbers much
greater than critical the motion is seen to be dominated by a boundary layer on
the heated surface and a plume rising above it. Wooding (1963) has given a most
revealing analysis of the flow in the plume when the flow is sufficiently dominated
by the discharge that there is minor recirculation of fluid; i.e. fluid particles make
only a single pass through the system. Since in the fluid surrounding the plume
p = 0 , if the plume is sufficiently thin, to the order of the boundary-layer approxi-
mation, p = 0 everywhere and hence, by ( 2 b ) , W = AB. This relation implies
a balance between viscous and buoyancy forces.
 Free convection in a porous medium 47

7. Concluding remarks
Some of the above flows are crudely analogous to the corresponding flows of
a viscous fluid. This is particularly the case when inertial effects, which arise
with a viscous fluid, are small, either because the Prandtl number is large or
because, near the critical Rayleigh number, the advection of vorticity is negli-
gible. There is the important difference, however, that flow in a porous medium
provides a realization of so-called free-free boundary conditions, whereas, while
frequently used in a preliminary analysis, these are rather artificial boundary
conditions for a viscous fluid.
Near the critical Rayleigh number, the flows are very similar, as pointed out
in the analyses of Horton & Rogers (1945) and Lapwood (1948). The essential
difference is that, while in both cases inertial effects are negligible, we must allow
for diffusion of vorticity in a viscous fluid. This merely changes the operator V4
in (12) to V6, an increase by a factor of n2(m2+n2). Consequently the critical
Rayleigh number is increased from 4n2to 27;rr4/4and the width of the cells by 4 2
(Chandrasekhar 1961).
Above the critical Rayleigh number, (6) suggests that N + A/A,, so that we
might anticipate a similar relation in a viscous fluid. Inspection of Silveston’s
(1958) measurements shows that indeed N* = A*/A: up to N* x 1-5, sug-
gesting that inertial effects are negligible here. Beyond this value, however, the
experimental curve falls rapidly below the linear relation. This is undoubtedly
due to the rapid increase in the advection of vorticity and the growth of viscous
boundary layers. Nevertheless, both flows are cellular and strongly influenced
by end-effects.A study of end-effectsfor free convection of a viscous fluid will be
reported elsewhere. It is sufficient here to note that as with flow in a porous
medium the phase of the cellular motion, the distribution of horizontal wave-
numbers, and hysteresis are dominated by the end cells.

The laboratory experiments referred t o in 5 3 were done at GeophysicsDivision,

Department of Scientific and Industrial Research, Wellington, New Zealand.
I am very grateful for the assistance of Mr W. J. P. MacDonald. The numerical
study was supported by National Science Foundation Grant GP-2414 and Office
of Naval Research Contract Nonr-2216 while I was at the Institute of Geophysics
and Planetary Physics, La Jolla. The numerical work was done on the CDC 3600
computer at the University of California, San Diego computing centre. The
manuscript was written while I was supported by a grant from the British
Admiralty.
REFERENCES
CARSLAW,H. S. & JAEGER, J. C. 1959 Conduction of Heat in Solids. Oxford: Clarendon
Press.
CHANDRASEKHAR, S. 1961 Hydrodynamic and Hydromagnetic Stability. Oxford : Clarendon
Press.
D. 1965 J. Fluid Mech. 21, 385.
COLES,
DARCY,H. P. G. 1856 Les Fontaines publiques de la ville de Dijon. Paris: Victor Dalmont.
DONALDSON, I. G. 1962 J . Geophys. Res. 67, 3449.
48 J . W . Elder
EINARSSON, T. 1942 Rit. Visind. 191. 26, 1.
ELDER,J. W. 1958 Ph.D. Thesis, Cambridge.
ELDER,J. W. 1965 Terrestrial Heat Flow, chap. 8. (Ed. W. H. K. Lee.) Amer. Geophys.
U . Monograph, no. 8.
ELDER,J. W. 1966a J . Fluid Mech. 24, 823.
ELDER,J. W. 1966b Bull. Volcan. 29 (in the Press).
FOSTER,T. D. 1965 Personal communication, t o be published.
Fox, L. 1!362 Numerical Solution of Ordinary and Partial Differential Equations. London :
Pergamon Press.
FRANK, F. C. 1965 Rev. Geophys. 3, 485.
GARABEDIAN, P. R. 1956 Math. Tables Aids Comput. 10, 183.
HELE-SHAW, H. S. J. 1898 Trans. Inst. Naw. Arch. 40, 21.
HORTON, C. W. & ROGERS, F. T. 194.5 J . Applied Phys. 16, 367. (See also 22, 1476.)
JAKOB, M. 1949 Heat Transfer. New York: Wiley.
LAPWOOD, E. R. 1948 Proc. Camb. Phil. SOC.44, 508.
MALKUS,W. V. R. 1954 Proc. Roy. Soc. A 225, 185.
RICHARDSON, J. G. 1961 Handbook of Fluid Dynamics, sec. 16 (ed. V. L. Streeter). New
York: McGraw-Hill.
SAFFMAN, P. G. & TAYLOR, G. I. 1958 Proc. Roy. Soc. A 245, 312.
SCHNEIDER, K. J. 1963 11th Int. Cong. of Refrigeration, paper 11-4, Munich.
SILVESTON, P. L. 1958 Forsch. Ing. 24, 29.
WOODING, R. A. 1957 J . Fluid Mech. 2, 273.
WOODING, R. A. 1958 J . Fluid Mech. 3, 582.
WOODING, R. A. 1959 Proc. Roy. Soc. A 252, 120.
WOODING, R. A. 1960a J . Fluid Mech. 7, 501.
WOODING, R. A. 1960b J . Fluid Mech. 9, 183.
WOODING, R. A. 1962 J . Fluid &tech. 13, 129.
WOODING, R. A. 1963 J . Fluid Meclz. 15, 527.
ZIEREP, J. 1961 Z A M M . 41, 114.
Journal of Fluid Mechanics, Vol. 27, part 1 Plate 1

FIGURE 8. End-effect in a thin cavity. Photographs of steady flows uith medicinal paraffin
in a Helc-Shaw cell of depth 2.0 cm, lcngth 30.0 cm, width 0 6 em; heated strip of length
10 cm. Values of lo-* A : ( a ) 1.02, ( b ) 1.72, (c) 2.24, (cl) 2.38, ( e ) 2.90, (f)3.97, (y) 1.10
(HSI: 1-9).
Jozcimd of Fluid Meclmnics, T'ol. 27, port 1 Plate 2

FIGGRE 12. Natural discharge. Photograph of steady flow in a Hcle-Shaw c ~ l with

l :
H = L = 10 cm, E = 50 em, b = 0.318 cm, at ltaylcigh number A = 420. Compare with
figure 10.

ELDER