Pearson 1958 On Convection Cells Induced by Surface Tension
Pearson 1958 On Convection Cells Induced by Surface Tension
Pearson 1958 On Convection Cells Induced by Surface Tension
SUMMARY
A mechanism is proposed by which cellular convective motion
of the type observed by H. BCnard, which hitherto has been
attributed to the action of buoyancy forces, can also be induced
by surface tension forces. Thus when a thin layer of fluid is
heated from below, the temperature gradient is such that small
variations in the surface temperature lead to surface tractions
which cause the fluid to flow and thereby tend to maintain the
original temperature variations. A small disturbance analysis,
analogous to that carried out by Rayleigh and others for unstable
density gradients, leads to a dimensionless number B which
expresses the ratio of surface tension to viscous forces, and which
must attain a certain minimum critical value for instability to occur.
T h e results obtained are then applied to the original cells described
by BCnard, and to the case of drying paint films. It is concluded
that surface tension forces are responsible for cellular motion in
many such cases where the criteria given in terms of buoyancy
forces would not allow of instability.
INTRODUCTION
Experiments have shown* that drying paint films often display steady
cellular circulatory flow of the same type as that observed in the case of
fluid layers heated from below. I n the latter case (that of the so-called
Benard cellst) the motion can usually be ascribed to the instability of the
density'gradient that would be present if the fluid were stationary. This
cannot be the mechanism causing the flow in the former case, since the
circulation is observed whether the free surface is made the underside or the
topside of the paint layer, that is, even if the gravity vector is effectively
reversed. Instead it will be shown that surface tension forces are sufficient to
cause instability and are probably responsible for many of the cellular patterns
that have been observed in cooling fluid layers with at least one free surface.
21
490
J. R. A. Pearson
A small disturbance analysis will now be carried out for the particular
49 f
J. R. A. Pearson
492
Q = rate of heat loss per unit area from the upper free surface (now a
function of x and x ) ,
T = temperature distribution in the liquid (a function of x, y and z ) .
T' = T - Tois thus the perturbation temperature (also supposed small),
and we write T i for its value at the surface.
The equations of motion and heat conduction become
($-vv2)V2v 0,
($ - K V ) V ~ T '
=
= flu.
(4)
It will be noted that the equation for v contains no buoyancy term and is
thus much simpler than in the case considered by Rayleigh. We also write*
S = &-uTL
(5)
where - u = (aS/aT). =Tos represents the rate of change of surface
tension with temperature, evaluated at temperature Tos,with So= S( T,,),
and
Q = Qo + qTk
(6)
where q = (aQ/aT),= ToS represents the rate of change with temperature
of the rate of loss of heat per unit area from the upper surface to its upper
environment. Qo is as defined before. u is a function of the liquid only,
while qis likely to be affected in a complicated way by the surface-environment
relations. The relations ( 5 ) and (6) for S and Q are taken to depend linearly
on T' because we are considering an infinitesimal disturbance theory and
therefore need only the first two terms in a Taylor expansion.
The boundary conditions on the velocity are
av
v=-=o,
aY
when y = 0,
v
when y
d, where
0,
azv
= uV: T ' ,
aY
p v 2-
v ; = -ax2
+a 2- .
a2
az2
The second relation in (7) follows directly from the continuity condition.
The necessity for the first of the relations (8), particularly in the case of
neutral stability, is explained by Jeffreys. The second relation (8) equates
the change in surface tension due to temperature variations across the
* T h e boundary conditions (including ( 5 ) and (6)) are of crucial importance;
by means of a suitable choice for these many physical phenomena may be very reasonably idealized. The aim in this account is not to provide an exhaustive description
of these phenomena and their relevant idealizations, but rather to provide a general
treatment that illustrates the fundamental surface tension mechanism and comprehends its many realizations. I t is shown that the use of but three parameters suffices
to describe the system.
493
surface to the shear stress experienced by the fluid at the surface, use being
made of the continuity condition. The effect of the surface tension on the
normal stress condition is ignored because its effects are small. (This same
assumption is made in the conventional thermal convection problem.)
It is the boundary conditions on the temperature that need most careful
explanation. It is worth considering in this respect the possible physical
circumstances leading to the thermal equilibrium expressed by (l), (2), in
order to understand the significance of the boundary conditions that will
be applied to the perturbation temperature. The supply of heat to the
bottom surface of the liquid may be from a material whose coefficient of
heat conduction is either large or small compared with that of the liquid.
If the boundary consists of a good conductor of large extent, then the
temperature TOBmay be taken as a given fixed constant, and will lead, in
the perturbed case, to the conducting condition T = 0 a t the lower
boundary. If however, the bottom boundary of the liquid consists of a
layer of a bad conductor itself in contact at its lower surface with a good
conductor (for example, glass on metal), then the temperature gradient
in the material bounding the liquid may be large compared with the
will, in the steady unperturbed
temperature gradient in the liquid itself; TOB
state, be a constant, though it will be determined by the thermal properties
both of the badly-conducting solid and of the liquid, and of their respective
thicknesses. A slight change in the bottom surface temperature will not
affect the rate of heat conduction through the bad conductor by any
appreciable amount. Hence the lower boundary condition on the
perturbation temperature will be approximately aT/ay = 0. This we shall
define, illogically perhaps, but consistently with previous authors, as the
insulating boundary condition. The two cases chosen above are limiting
cases of the general mixed boundary condition
494
J. R. A. Pearson
jump may be small or large compared with the drop in temperature across
the liquid layer, depending on the efficiency of the processes for removing
heat from the surface. Whatever the process, the equality
must hold at y = d, using the relation (6) and the reasons given to justify
the equality (2). If the temperature jump is large, then we expect q to be
small, and vice versa. The particular value to be ascribed to q will depend
on circumstances, and for this reason we shall retain it as a parameter in
the subsequent analysis. It is worth noting that if q -+ 00, then we approach
the ' conducting ' boundary condition, and surface tension effects will not
be present, whereas if q + 0, we approach the ' insulating ' boundary
condition, as defined above, valid for perturbation of a strictly steady-state
system.
A further application of this small-disturbance analysis to concentration
gradients in a layer composed of a mixture of two liquids will show-if
only by comparison with experiment-that predictions of instability are
relevant for quasi-steady systems. If convection cells can be set up in
a time small compared with that in which changes occur in the quasi-steady
thermal gradient p, then the application of strict ' insulating ' boundary
conditions leads to meaningful results.
If we now introduce dimensionless variables
[p-(D2-a2)]g
-f,
(16)
where D = d/dy and Pr = Y / K . The particular choice of the forms (12), (13)
for v , T' is well explained by Pellew & Southwell (1940) who show that
-theconstant a is merely indicative of a periodic structure in the (4, t)-plane.
The shape of the cells associated with the solution obtained is not specified
and a second-order theory would be required to select a particular cellular
structure. Christopherson (1940) has given the solution of (14) for
hexagonal cells.
495
a2Bg(l), g(1) = - L g ( l ) ;
(17)
and
are dimensionless constants. Physically speaking, the number B is a
dimensionless grouping of physical parameters expressing the relative
importance of surface tension forces (caused by temperature variation)
and of viscous forces, in any small disturbance responsible for both of
them. The number L can be suitably interpreted in terms of the arguments
given earlier with regard to relation (10).
As a last boundary condition, we have for the conducting case ( Y = 0
in (9117
g(0) = 0,
and for the insulating case (Y-l = 0 in (9)),
g(0) = 0.
The general boundary condition (9) would lead to the relation
(20)
(21)
g(0) = Mg(O),
* ()
(25)
t It must be made clear that the evaluation of these three parameters in any physi-
J . R. A. Pearson
496
for the conducting case, and
a cosh a - sinh a
q2 cash
4a sinh a
- $qzsinh aq -
a cosh a - sinh a
3
q sinh aq - -sinh aq 4a2sinh a
4u2
B=
(27)
B =
(28)
for the insulating case. These are plotted in figures 1 and 2 for the various
values of L. The curves obtained are neutral stability curves, and without
repeating the analysis for p # 0 it is clear which regions refer to growing
disturbances and which to damped disturbances. All the curves asymptote
to B = 8a2 for large values of c(, whilst all display a critical (minimum)
value of B corresponding to a particular value for a for which stable
disturbances are first possible. T h e case L = 0 for the insulating boundary
condition is particular in that the critical value of cx is zero, B = 48. Apart
from this special case, the shape of the neutral stability curves is very similar
in the conducting and insulating cases ; in general larger positive values
of L lead to greater stability. T h e values of L encountered in practice
would depend on the thickness of the film and for very thin films would
tend to zero. Detailed solutions for the more general boundary condition (9)
with finite non-zero values of Y are not given but it is not expected that
these would behave very differently from the two limiting cases L= 0,
and Y-l= 0.
I t is interesting to compare the dimensionless number B , defined by (18),
which is relevant for the surface tension mechanism, with the Rayleigh
number
497
L = 0, from figure 1, with the corresponding critical value Acrit= 571* (see
Jeffreys (1926), the case of one free ' insulating 't and one fixed conducting
I00
200
300
4001
100
200
300
400
B
Figure 2. Neutral stability curves, insulating case.
* This value is the best estimate that is available, and although it was calculated
using erroneous boundary conditions, it is expected to be close to the correct result.
Since we are using it for order of magnitude arguments, the discrepancy need not
trouble us unduly.
t ' Insulating ' as regards the perturbation temperature. It corresponds to a case
of uniform heat flow, as does our case L = 0.
498
J. R.A. Pearson
surface, and the amending paper Jeffreys (1928)). The former of these
leads, for any given fluid layer, to a critical thickness
d, =
These will be equal for a value d12 given by
COMPARISON
WITH OBSERVATION AND EXPERIMENT
The original observations of BCnard (1900, 1901) provide a very suitable
illustration of the instabilities considered above. For with spermaceti
he achieved cells in layers less than mm thick when the temperature
of the lower surface was maintained at 100C. Moreover, these cells
persisted even when the temperature of the lower surface approached 50" C.
Since spermaceti solidifies at rather above 50" C an upper value can be placed
on the temperature variation across the layer and can only have been a degree
or two, if that, in some cases. BCnard gives a value of 7.7 x
c.g.s. units
for y and, fortunately for our purposes, a value of nearly 5 x
c.g.s. units
for u. We can assume that the case of a lower rigid conducting and an upper
free ' insulating ' boundary applies approximately to his experiments since
the lower surface was metallic and the temperature difference between the
upper surface and its surroundings can be assumed large compared with
the drop in temperature across the liquid layer. It would require detailed
calculation to obtain an accurate estimate for L in this case, since the
mechanism for heat transfer involves both convection and radiation.
However, a rough calculation shows that L < l-whether it be lop3or 10-1
is scarcely important since figure 1 shows that the critical value B necessary
for instability only increases by a factor of 3 when L = 5-and we can
therefore use the approximation L = 0 as an initial approximation.
Even if we assume that K was as small as
and v as small as 10-2,
the value obtained for A is still less than 571. On the other hand, for
reasonable values of K (5 x
and v (5 x
a value of B in excess
of 80 is obtained even for a temperature drop of only 1O C across the layer.
Furthermore the ratio of cell size to layer thickness quoted by BCnard
499
leads to values of M lying between 2.1 and 2-5 which are in good agreement
with the analysis given in the previous section, where uCrit= 2.0. The
equivalent value of
from the analysis of Jeffreys is 3.5, which even
allowing for error is significantly different from the observed values.
Thus we see that the buoyancy mechanism has no chance of causing
convection cells, while the surface tension mechanism is almost certain
to do so, and that observations support this. (Going back to the
choice of L = 0, we see that this was not critical and hence that an exact
analysis of the heat transfer at the surface is not necessary to sustain the
argument.) An intimation that the instability theory based on buoyancy
forces would not account for all of BCnards results appears in a paper by
Volkovisky (1939).
The allied problem of a liquid cooling by evaporation may be treated
in a similar fashion. Although no strictly steady equilibrium state may
be presupposed, owing to the loss of fluid from the surface, the effect of
evaporation may be reasonably well represented by a given heat loss from
the surface, using a value for L that depends on the rate of evaporation
and the latent heat of vaporization. A quasi-steady value for p must be
assumed to hold.
The application of the analysis to the case of a mixture of two liquids,
one more volatile than the other, requires more explanation. Two factors
tending to instability will now be relevant. The first, that due to temperature
variations, has been treated already. The second is that due to relative
concentration variations. This may be treated separately using a suitable
interpretation of the analysis given above for temperature effects. For
if the temperature T is interpreted as the concentration C of the volatile
in the non-volatile liquid z2,
the constant k is interpreted as the
liquid 9,
in p2and the parameters u, q and y refer to
coefficient of diffusion of 9,
variations with concentration C, then the equations (3) and (4) are
appropriate equations for describing small disturbances due to concentration
variations in a layer whose equilibrium (or quasi-steady) concentration
gradient is given by p-for the moment we regard /3 as some unspecified
function of y. Previous work has shown-see for example Batchelor (1954)
-that the most unstable case, for any given concentration drop across the
layer, occurs when the gradient p is a constant*. Consequently, although
the instantaneous concentration across the layer for the case of surface
evaporation of 8, from a mixture of 2,and P2would never be a linear
function of y, we may safely use a value of
= Cm,#
to obtain a lower limit for stability. The boundary conditions on the
velocity, ( 7 ) and (8), would still apply; the relation (10) would also apply
with suitable interpretation ; but at the bottom surface the insulating
condition would now be relevant. Hence the result (28), displayed in figure 2,
can fairly reasonably be applied to describe the concentration effect. I n
* This has been verified analytically for the particular choice of 8: ,6 = C,/bd,
.d > y > d(1-b); ,6 = O , d ( l - b ) > y > 0 ; 0 < b < 1. Instabilityisgreatestforb = 1.
500
J . R. A. Pearson
general terms the temperature and concentration effects are additive and
for any given circumstances a joint criterion replacing either (27) or (28)
could be calculated. This will not be carried out here because of the
difficulty in prescribing the equilibrium concentration gradient.
A series of experiments carried out in this laboratory on thin layers
of mixtures of liquid paraffin and ether have shown convection cells of
polygonal type when evaporation of the ether takes place. Estimates of
the buoyancy forces involved showed that these could not be responsible
for the motion. On the other hand, the difference in surface tension
between liquid paraffin (32 c.g.s. units) and ether (19 c.g.s. units), combined
with their temperature variations, leads to surface tractions large enough
to maintain the motion. For example, a 5% concentration of ether
evaporating at room temperature (say, 15" C or less) showed cells down
to a layer thickness of 0.5 mm, this critical thickness being highly reproducible; even allowing for a very high rate of evaporation, and adding
the concentration and temperature effects on the density, a value of A
of the order of 10 or less is obtained. Similarly crude estimates of B for
the two effects suggest instability. The ratio of cell size to layer thickness
proved to be relatively constant over a fair range of layer thickness and
using the analysis of Christopherson (1940) a value of o! 2.0 was obtained,
which is in reasonable agreement with our analysis.
Similar experiments using a wide range of both volatile and non-volatile
liquids have been carried out by Dr Cousens. All his results are in excellent
qualitative agreement with the ideas described here. I n particular, by
choosing a volatile fluid with higher surface tension than the non-volatile
base, cellular motion was inhibited. The addition of a very small quantity
of silicone oil to various mixtures also inhibited cell formation by reducing
the surface tension overall.
N
The author wishes to thank Dr R. Cousens for pointing out the probable
role of surface tension and for helpful discussion of the observations quoted.
It is hoped to publish a joint paper on the specific application of the
foregoing theory to paint films. The author is also grateful for many useful
criticisms by the referees.
RFEERENCES
BATCHELOR,
G. I(. 1954 Quart. J . Roy. Met. SOC.80, 339.
BENARD,
H. 1900 Rev. Gen. des Sci. purrs et appl. 1261, 1309.
BONARD,
H. 1901 Ann. Chim. Phys. (7), 23, 6 2 .
CHRISTOPHERSON,
D. G. 1940 Quart. J. Math. 11, 63.
JEFFREYS,
H. 1926 Phil. Mag. (7), 2, 833.
JEFFREYS,
H. 1928 Proc. Roy. SOC.A, 118, 201.
LIN,C. C. 1955 Theory of Hydrodynamic Stability. Cambridge University Press..
Low, A. R. 1929 Proc. Roy. SOC.A, 125, 180.
PELLEW,
A. & SOUTHWELL,
R. V. 1940 Proc. Roy. SOC.A, 176, 312.
PRANDTL,
L. 1952 Fluid Dynamics. London: Blackie.
RAYLEIGH,
LORD1916 Scientific Papers, vol. 6, p . 432. Cambridge University Press.
VoLKovIsKy, V. 1939 Publ. Sci. Tech. Minist. de I'Air, no. 151.