You are on page 1of 6

A rising bubble in a tube

Herbert Levine and Yumin Yang

Citation: Phys. Fluids A 2, 542 (1990); doi: 10.1063/1.857754


View online: http://dx.doi.org/10.1063/1.857754
View Table of Contents: http://pof.aip.org/resource/1/PFADEB/v2/i4
Published by the American Institute of Physics.

Additional information on Phys. Fluids A


Journal Homepage: http://pof.aip.org/
Journal Information: http://pof.aip.org/about/about_the_journal
Top downloads: http://pof.aip.org/features/most_downloaded
Information for Authors: http://pof.aip.org/authors

Downloaded 14 May 2013 to 35.8.11.2. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://pof.aip.org/about/rights_and_permissions
A rising bubble in a tube
Herbert Levine and Yumin Yang
Department of Physics and Institute for Nonlinear Science, University of California, San Diego,
California 92093
(Received 29 August 1989; accepted 28 November 1989)
The rise of an infinitely long bubble in a vertical cylindrical tube full of inviscid fluid is
studied. It is demonstrated numerically that the rising velocity of the bubble is determined by
the surface tension through a solvability mechanism. The numerical results are in good
agreement with the experimental ones.

I. INTRODUCTION II. FORMULATION


The problem of a large bubble rising in a vertical cy- Before formulating the problem of a three-dimensional
lindrical tube has been studied both theoretically and ex- rising bubble, we would like to comment on the difference
perimentally for the past few decades. I ,2 Early theoretical between the two- and three-dimensional bubble problems.
treatments often ignored the role of surface tension. How- In the two-dimensional bubble problem, both the stream-
ever, experiments done by Zukoski3 for large bubbles ris- function and potential function satisfy the Laplace equa-
ing in tubes clearly demonstrated that the rising velocity of tion; furthermore, they form the imaginary and real parts
the bubble is completely determined by the surface tension of an analytic function. Thus the powerful tools of a com-
when the Reynolds number of the flow is greater than 200. plex analysis can be used in this situation. Most papers on
Clearly, we need a theory that includes surface tension to this problem mentioned in the Introduction use this
explain Zukoski's experimental results. technique. 4,5,7 For the three-dimensional problem, this
In recent years, several authors4--8 have worked on the technique no longer applies because the streamfunction
does not satisfy the Laplace equation [see (2)]. Our for-
analogous problem of a two-dimensional (2-D) bubble ris-
mulation in terms of the Green's function closely resem-
ing in a channel. They assumed the fluid is inviscid and
bles the one by Kessler and Levine. 8
studied the relation between the surface tension and the
We assume the fluid inside the tube is inviscid, con-
rising velocity. The numerical work in these papers tells us
stant density and irrotational, and we neglect the density
that the bubble rising velocity is indeed determined by the of air inside the bubble. Since we are only considering the
surface tension, and the small surface tension limit is a case of the tube in a vertical position, we can further as-
singular perturbation of the zero surface tension case (i.e., sume the flow is axisymmetric. Under the above assump-
taking the limit of surface tension approaching zero is not tions, we define the streamfunction t/J as follows 10:
the same as setting it to zero). The calculated rising veloc-
A
ity versus the surface tension curve for the two-dimen- v=VX{[ t/J(r,z)/r]4>}, (1)
sional bubble is very similar in shape7 to that of the exper-
imental curve by Zukoski. Furthermore, these calculations where we have used the standard notation in cylindrical
could be compared to the (limited) experimental data of coordinates. According to this definition, the velocity field
Maneri9 in a channel geometry. Nevertheless, it is neces- is automatically divergence free. The irrotational condition
sary to do the three-dimensional calculation in order to gives us the following equation for t/J:
quantitatively explain the 3-D experimental results.
In this paper we formulate the problem of an infinitely (2)
large bubble rising in a vertical tube. We choose our units
such that R=g=p=I, where R is the radius of the tube Having derived the equation for t/J, let us now look at
(for the two-dimensional bubble problem R corresponds to the boundary conditions that t/J should satisfy. Since we are
the half-width of the channel), g is the gravitational accel- only going to look for those solutions in which the bubble
eration, and p is the density of the inviscid fluid. In such rises steadily, we can go into a moving frame in which the
units, the dimensionless rising velocity u is just the Froude bubble surface is stationary and the flow at infinity is mov-
number of the flow, i.e., u = u*IJgR, where u* is the ing down at a constant speed. In this frame the flow is
rising velocity with the dimension; and the dimensionless stationary, i.e., the velocity field does not change with
surface tension coefficient r is the inverse EOtvos number, time. On the wall of the tube and the bubble surface, we
i.e., r = r*/(pgR 2 ), where r* is the dimensional surface have
tension coefficient. After formulating the problem, we will
(3)
first discuss the zero surface tension case. Then, we present
results and observations for nonzero surface tension. (4)

542 Phys. Fluids A 2 (4), April 1990 0899-8213/90/040542-05$02.00 @ 1990 American Institute of Physics 542

Downloaded 14 May 2013 to 35.8.11.2. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://pof.aip.org/about/rights_and_permissions
where we denoted the bubble surface to be z=1J(r). Ber- asymptotic behavior far downstream (see the Appendix).
noulli's law on the bubble surface for the stationary flow We then modify the bubble shape by adding A1J(r) to the
provides us with another boundary condition, initial guess. We discretize the interface with a fixed grid,

11
2?- II ot/!(r,z) 121
on + 1J(r) - rK(r) =0. (5)
which becomes an equal arclength grid far downstream.
This procedure replaces the integrodifferential equation
with a coupled set of nonlinear equations. 9•11
z=1/(r)

Here K is the mean curvature of the three-dimensional Since the Green's function decays exponentially with
paraboloid-shaped surface, which can be written as respect to z, the downstream behavior does not affect the
tip region. Therefore we can replace the upper limit of the
integral in (11) by a finite number, say around 6. All the
(6) integrations in our programs are done using the Simpson
rule.
where 1J' and 1J" denote d1J/dr and d 21J/d?, respectively. For the zero surface tension case, we roughly divide
Had the bubble surface been given, we would have
our initial guess curve into two parts. We then keep the
overspecified the boundary conditions. Since this is a free
downstream part fixed and vary the upper part; this is valid
boundary problem, we can use the extra boundary condi-
because the downstream shape is almost correct. We take
tion (5) to determine the bubble shape. We will use the
A1J(xj) on N points ofthe upper part of the initial guess as
Green's function method to get an integrodifferential for
independent variables. The original integrodifferential
the bubble shape 1J(r). We define the Green's function in equation (11) provides N equations to be satisfied (one for
this case to be
each point). We now have N variables and N equations,

r1~if? G(r,r' ,z,z')


10 1~
-?- or G(r,r',z,z') + ra? G(r,r' ,z,z')
and we use the standard Newton method to solve them.
In the case of nonzero surface tension, we have
adopted a slightly different numerical procedure, which we
= -l>(r - r')l>(z - z'), (7) now describe. We first break up the integrodifferential (11)
into two equations as follows:
G(r,r',z,z') Ir=1 =0. (8)
The solution for this Green's function problem is
G(r,r',z,z')
21 u? - Jo
(""
G(r,r',1J(r),1J(r'» v (s')ds'=O, (13)

21 v2 + 1J - rK = 0. (14)

We now can vary not only A1J but also Av. We next expand
where Xm is the mth root of the first-order Bessel function
both A1J and Av in terms of Chebyshev polynomials. We
J 1(x).
choose the 2n coefficients of the expansion (n for A1J and n
In terms of this Green's function, t/! can be written as for Av) to be the independent variables. We then choose n

t/!(r,z) = - 21 u? + I lot/!
G(r,r',z,z')"ii on' ds'. (10)
collocation points among the N discretized points on the
upper part of the initial guess; note that N determines the
accuracy of the Simpson rule approximation for the inte-
o
grals appearing in the above equations. Equations (13) and
Setting z=1J(r) and substituting (5) into (10), we obtain (14) at these n points will provide us with 2n equations to
the integrodifferential equation we want be solved by the Newton method.
The coefficients in the Chebyshev polynomial expan-
~ u? - Io"" G( r,r',1J(r),1J(r'»
sion of a well-behaved function will typically converge
very rapidly. This allows us to limit the value of n to a
small number (usually around 15); this assumption has
X [2rK(r') - 21J(r')] 112 ds' =0. (11 ) been checked by varying n with no significant change in
the results. Typical values for N used in our program are
For a physical meaningful solution of the above equation,
we must require about 125. As a final consistency requirement, we check
whether the converged solution also satisfies equations
(12) ( 13) and (14) at the other discretized points, and the
The integrodifferential equation (11) with the bound- results are satisfactory.
ary conditions (12) does not have any obvious analytical The boundary condition (12) is usually relaxed when
solutions. Therefore we proceed to solve it numerically. we solve the integrodifferential equation. We check the
boundary condition afterward to see whether it is satisfied.
If it is not satisfied, we then move around in the parameter
III. NUMERICAL PROCEDURES
space u and r to satisfy this boundary condition. Therefore
To solve this integrodifferential equation, we first make we expect to find curves r( u) in the r-u plane that corre-
an initial guess of the bubble shape, which has the right spond to actual solutions.

543 Phys. Fluids A. Vol. 2. No.4. April 1990 H. Levine and Y. Yang 543

Downloaded 14 May 2013 to 35.8.11.2. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://pof.aip.org/about/rights_and_permissions
We now make several remarks on the computation of
the Green's function (9). 6.00
( 1) When Iz - z' I is large enough, we can just sum
the series directly, since it converges almost geometrically. 5.00

(2) When both r and r' are small (which implies that 4.00 ~.
Iz - z' I is small), it is unwise to sum the original series
(9) directly, because it converges extremely slowly. In this
~ 3.00
\
case we use an integral expression for the Green's function, \~
2.00 .~

which is '"""
1 (00 K,(k)
1.00 ""
' ........
G(r,r',z,z') = -;,rr' Jo cosk(z-z') I,(k)
u
XI, (kr)I, (kr')dk
FIG. 1. The radius of curvature at the bubble tip versus the rising velocity
1 (r-r')2+(Z-Z')2) for the zero surface tension case.
+ 21T PQI/2 1 + 2rr' '

of N there will be a maximum u at which the slope con-


where I, and K, are the modified Bessel function of the first dition is satisfied. The actual limiting value u. must then
order, QI/2 is the Legendre function of the second kind of be found by extrapolating the results to zero lattice spac-
the! order. ing. Fortunately, this is not very difficult since the radius of
(3) When Iz - z'l is small but rand r' are not, we curvature increases extremely rapidly as we approach this
have to subtract the leading-order behavior of the terms in point.
the series to make it converge more rapidly. The leading-
order behavior can usually be summed analytically and B. Nonzero surface tension case
added back in later. In this section we will focus most of our attention on
(4) The Green's function (9) has a logarithmic sin-
the situation where the surface tension coefficient y is pos-
gularity when r-r' and z-z'. This singularity has to be
itive. We will also mention the numerical results when y
subtracted out when performing the integral in (11). The becomes negative.
logarithmic function In Is - s' I can be integrated explic-
For a positive y, we find that the integrodifferential
itly along the interface. Thus we can add back in the inte- equation (11) with the boundary condition (12) defines a
gral contribution from the singular part of the Green's
nonlinear eigenvalue problem. In the parameter space
function. We refer readers to the paper by Kessler and
formed by u and y, for a given velocity u, there exists a
Levine ll for more detailed information on the last two discrete set of y such that the boundary condition (12) is
remarks.
satisfied. The major result of this paper is presented in Fig.
2. For comparison, we also plotted the experimental curve
IV. NUMERICAL RESULTS by Zukoski on the same graph. The difference between the
experimental curve and the numerical curve is less than
A. Zero surface tension case ten percent. This can possibly be attributed to the numer-
In this case, the only parameter is the velocity u. We ical uncertainty; on the other hand, we do not know the
solve for the shape and discover that the slope at the tip is
always small (about 10- 5) for a wide range of rising ve-
locities u. This tells us that there is a continuous spectrum. 0.70
In other words, we cannot get a unique rising velocity from G
•...•••..
this model with zero surface tension. 0.60
];I a Theoretical Curve
In Fig. 1, we have plotted the radius of curvature at the 0.50
•••••••• El.••• Experimental Curve

bubble tip as a function of the rise velocity; as u increases, ... 'B......

0.40
the tip becomes sharper. At a critical value of u, the curve c- ..a.······· n ......
in Fig. 1 will approach zero. This signifies the end of the 0.30
continuum branch, and our calculations place this point at
0.20
u.==O.70. For comparison, we have used the same scheme
to calculate the bubble shape for a two-dimensional geom- 0.10

etry, obtaining u.==O.51. This number agrees with previ-


ous calculations. 4•6•7
The above calculation also points out the effect of dis- u
cretizing the curve. For any fixed number of points, the
FIG. 2. The relation between the surface tension and the bubble rising
radius of curvature of the tip is effectively bounded by the velocity; all data is at Re>200. in the regime where rise velocity is inde-
lattice spacing. As we try to find solutions, for each value pendent of Re.

544 Phys. Fluids A. Vol. 2. No.4. April 1990 H. Levine and Y. Yang 544

Downloaded 14 May 2013 to 35.8.11.2. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://pof.aip.org/about/rights_and_permissions
------------------------------------------------------
1.00
--------",
\\\
N 0.00 ................................................................................ II
N -0.90
......•........................•...... Theoretical Profile

~~t!.~~
\\ • Experimental Profile

-1.00 ~:!?3.~~ \ -1.40


-,=o.OOS6 I
\
1.00
r r

FIG. 3. Three-dimensional bubble shapes correspond to the three lowest FIG. 4. Comparison between the calculated bubble profile (u = 0.49,
modes when the rising velocity u = 0.14. r = 0.0059), and the corresponding experimental one (Fig. 9 in Davies
and Taylor2).

experimental uncertainty, as Zukoski did not provide error


estimates in his paper. 3 Thus, with these caveats, our nu- calculated bubble shape will have some disagreement with
merical result is in agreement with the experiment. When the experimental one far downstream. This is because our
the surface tension approaches zero, we see that the degen- model treats the tube wall as perfectly slippery, whereas in
eracy of the zero surface tension situation is broken since a real experiments, the wall has a finite viscosity. The small
unique rising velocity Uo is picked out by the curve in Fig. viscosity on the tube wall will create a boundary layer that
2. For the three-dimensional bubble our numerical results has large effects on the bubble profile far downstream.
indicate that Uo = 0.49; again we have recomputed the Since the Green's function decays exponentially in the ver-
two-dimensional case as a test, obtaining Uo = 0.32, in tical direction, the effect of downstream profile on the tip
agreement with earlier work. region is very small.
The curve plotted in Fig. 2 corresponds to the primary Let us now discuss the situation where the surface
mode of this nonlinear eigenvalue problem. There are also tension coefficient y is negative. Our numerical simulations
other curves in the u-y plane corresponding to higher indicate that there is a continuous spectrum for both two-
modes. All these curves are quite close to the u axis (i.e., and three-dimensional bubbles, i.e., for any given pair of
y is very small), which makes them hard to obtain numer- parameters u and y, a solution can be found that satisfies
ically. We have calculated a few points on the curve cor- the boundary condition (12). Although the negative sur-
responding to the secondary mode. We found that y de- face tension situation does not have any physical meaning,
creases when u increases, and the difference between the it might offer some mathematical insight for the integro-
maximum height of the bubble and the tip height, which is differential equation (11).
local minimum for the secondary mode (see Fig. 3), also
decreases as u increases. These two facts prompt us to v. CONCLUSION AND DISCUSSION
make the conjecture that all these curves corresponding to In this paper we have shown that the three-
higher modes will terminate at Uo as the surface tension dimensional bubble problem has the same qualitative, but
coefficient y approaches zero. different quantitative, behavior as the two-dimensional
In Fig. 3 we present the bubble shapes corresponding one. Our numerical results for the three-dimensional bub-
to the three lowest modes for a fixed u. We can clearly see bles are in good agreement with the experimental data.
that the bubble shape will have one more extremum for We now discuss some unanswered questions related to
each additional mode. This is similar to the eigenmodes of this subject. As of yet, we do not have an analytic under-
a linear eigenValue problem. Obviously, the only bubbles lying velocity selection mechanism. Also, the stability of
that are seen experimentally are ones without the extra the steady states we have found are worth investigating in
"wiggles." Therefore, there must be a reason why only the the future. Finally, the problem of a bubble rising in a
primary branch is relevant. Now, in this paper we have non vertical tube 3 requires a different approach, because in
confined ourselves to find the steady state solutions and this case the surface can no longer be described by a two-
there are no dynamics present. We hypothesize that if one dimensional curve.
investigates the stability of these steady state solutions,
then all the higher-order modes will tum out to be linearly
ACKNOWLEDGMENTS
unstable. This hypothesized behavior has been shown to
apply for the problem of a Saffman-Taylor finger in a We wish to thank David Kessler and Wouter Rappel
Rele-Shaw cell. 12 for some helpful discussions.
In Fig. 4 we compare the calculated bubble profile Yumin Yang was supported by DARPA-University
with the corresponding experimental one. The agreement Research Initiative, URI Contract No. NOOOI4-86-K-
is excellent in the tip region. It is not surprising that the 0758.

545 Phys. Fluids A, Vol. 2, No.4, April 1990 H. Levine and Y. Yang 545

Downloaded 14 May 2013 to 35.8.11.2. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://pof.aip.org/about/rights_and_permissions
APPENDIX: ASYMPTOTIC SHAPE The conservation of flux tells us that the leading-order
In this appendix we derive the asymptotic behavior for term for the bubble shape 11(E) is - u2/8~. The leading-
11(r) when r-- 1. Instead of expanding the integrodifferen-
order term for g\ (11) is -J - 211. Therefore we assume the
tial equation (11) directly, our approach here uses the following expansion forms for 11 ( E) and g \ ( 11 ) :
original differential equation and the boundary conditions.
Let us define a new variable ~= 1- r. We expand the (A7)
streamfunction in terms of ~ as follows:
00

tP(~,z) = -! u + g\ (z)S + g2(Z)S2 + ... , (AI) g\(11)=~-2n+ L bj~-11(-j). (A8)


j=O
where g\ and g2, etc., are arbitrary functions of z. Notice
that we have already used the boundary condition on the Substituting these two expansions into (A5) and (A6), we
wall. We next substitute the expansion (A1) into Eq. (2) find that the coefficients an and bn can be solved recur-
and expand everything in terms of powers of S. Setting the sively. In practice we used MACSYMA to obtain the follow-
coefficients of different orders of S to zero, we find that the ing results:
functions gj (i > 1) can be expressed in terms of g\. The
a\=I,
first few of them are
(A2)
a2=~ - 8(ylu 2),
g2(Z) = -!g\(z),
2
g3(Z) = -igi'(z), (A3) a3=! - 8(ylu ),

g4(Z) =ngi'(z). (A4) a4=f6 - 8( ylu 2),

We assume when S = E we are on the bubble surface, i.e., a5=~ - 8(ylu 2),
z = 11(€). Using the expression (AI) and the boundary
condition (4) we obtain a6=i4 - 8(ylu 2 ) + 80/3u4,
- ! u + g\ (11)E - ! g\ (11)~ - i gi' (11 )E3 + n gi' (11 )E4 a7=rr, - 8(ylu 2) - 16/3u4 - 384ylu 6.

+ "'=0. (A5)
We emphasize that the prime denotes the derivative with 10. T. Oumitrescu, Z. Angew. Math. Mech. 23,139 (1943).
respect to 11, not E. The boundary condition (5) will give 2R. M. Oavies and G. I. Taylor, Proc. R. Soc. London Ser. A 200, 375
us another equation as follows: (1950).
3E. E. Zukoski, J. Fluid Mech. 25, 821 (1966).
gf(11) - g\(11)gi'(11)~ + g;2(11)~ - !g\(11)gi'(11)E
3 4J. M. Vanden-Broeck, Phys. Fluids 27, 1090 (1984).
sJ. M. Vanden-Broeck, Phys. Fluids 27,2604 (1984).
+g;2(11)E3 + ... 6B. Couet, G. S. Stromolo, and A. E. Oukler, Phys. Fluids 29, 2367
(1986).
7B. Couet and G. S. Stromolo, J. Fluid Mech. 184, 1 (1987).
(A6) 80. A. Kessler and H. Levine, Phys. Rev. A 39, 4462 (1989).
9C. C. Maneri, Ph.D. thesis, Polytechnic Institute of Brooklyn, 1970.
We now have to find two functions, g\(11) and 11(E), such IOH. Lamb, Hydrodynamics (Dover, New York, 1932).
that conditions (A5) and (A6) are satisfied to all orders 110. A. Kessler and H. Levine, Phys. Rev. A 39, 3041 (1989).
12
of E. 0. A. Kessler and H. Levine, Phys. Fluids 30, 1246 (1987).

546 Phys. Fluids A, Vol. 2, No.4, April 1990 H. Levine and Y. Yang 546

Downloaded 14 May 2013 to 35.8.11.2. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://pof.aip.org/about/rights_and_permissions

You might also like